1/*
2 * Copyright (c) 2011 The NetBSD Foundation, Inc.
3 * All rights reserved.
4 *
5 * This code is derived from software contributed to The NetBSD Foundation
6 * by Coyote Point Systems, Inc.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
18 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
19 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
21 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 * POSSIBILITY OF SUCH DAMAGE.
28 */
29
30/*
31 * Reduces the resources demanded by TCP sessions in TIME_WAIT-state using
32 * methods called Vestigial Time-Wait (VTW) and Maximum Segment Lifetime
33 * Truncation (MSLT).
34 *
35 * MSLT and VTW were contributed by Coyote Point Systems, Inc.
36 *
37 * Even after a TCP session enters the TIME_WAIT state, its corresponding
38 * socket and protocol control blocks (PCBs) stick around until the TCP
39 * Maximum Segment Lifetime (MSL) expires. On a host whose workload
40 * necessarily creates and closes down many TCP sockets, the sockets & PCBs
41 * for TCP sessions in TIME_WAIT state amount to many megabytes of dead
42 * weight in RAM.
43 *
44 * Maximum Segment Lifetimes Truncation (MSLT) assigns each TCP session to
45 * a class based on the nearness of the peer. Corresponding to each class
46 * is an MSL, and a session uses the MSL of its class. The classes are
47 * loopback (local host equals remote host), local (local host and remote
48 * host are on the same link/subnet), and remote (local host and remote
49 * host communicate via one or more gateways). Classes corresponding to
50 * nearer peers have lower MSLs by default: 2 seconds for loopback, 10
51 * seconds for local, 60 seconds for remote. Loopback and local sessions
52 * expire more quickly when MSLT is used.
53 *
54 * Vestigial Time-Wait (VTW) replaces a TIME_WAIT session's PCB/socket
55 * dead weight with a compact representation of the session, called a
56 * "vestigial PCB". VTW data structures are designed to be very fast and
57 * memory-efficient: for fast insertion and lookup of vestigial PCBs,
58 * the PCBs are stored in a hash table that is designed to minimize the
59 * number of cacheline visits per lookup/insertion. The memory both
60 * for vestigial PCBs and for elements of the PCB hashtable come from
61 * fixed-size pools, and linked data structures exploit this to conserve
62 * memory by representing references with a narrow index/offset from the
63 * start of a pool instead of a pointer. When space for new vestigial PCBs
64 * runs out, VTW makes room by discarding old vestigial PCBs, oldest first.
65 * VTW cooperates with MSLT.
66 *
67 * It may help to think of VTW as a "FIN cache" by analogy to the SYN
68 * cache.
69 *
70 * A 2.8-GHz Pentium 4 running a test workload that creates TIME_WAIT
71 * sessions as fast as it can is approximately 17% idle when VTW is active
72 * versus 0% idle when VTW is inactive. It has 103 megabytes more free RAM
73 * when VTW is active (approximately 64k vestigial PCBs are created) than
74 * when it is inactive.
75 */
76
77#include <sys/cdefs.h>
78
79#ifdef _KERNEL_OPT
80#include "opt_ddb.h"
81#include "opt_inet.h"
82#include "opt_inet_csum.h"
83#include "opt_tcp_debug.h"
84#endif
85
86#include <sys/param.h>
87#include <sys/systm.h>
88#include <sys/kmem.h>
89#include <sys/mbuf.h>
90#include <sys/protosw.h>
91#include <sys/socket.h>
92#include <sys/socketvar.h>
93#include <sys/errno.h>
94#include <sys/syslog.h>
95#include <sys/pool.h>
96#include <sys/domain.h>
97#include <sys/kernel.h>
98#include <net/if.h>
99#include <net/if_types.h>
100
101#include <netinet/in.h>
102#include <netinet/in_systm.h>
103#include <netinet/ip.h>
104#include <netinet/in_pcb.h>
105#include <netinet/in_var.h>
106#include <netinet/ip_var.h>
107#include <netinet/in_offload.h>
108#include <netinet/ip6.h>
109#include <netinet6/ip6_var.h>
110#include <netinet6/in6_pcb.h>
111#include <netinet6/ip6_var.h>
112#include <netinet6/in6_var.h>
113#include <netinet/icmp6.h>
114#include <netinet6/nd6.h>
115
116#include <netinet/tcp.h>
117#include <netinet/tcp_fsm.h>
118#include <netinet/tcp_seq.h>
119#include <netinet/tcp_timer.h>
120#include <netinet/tcp_var.h>
121#include <netinet/tcp_private.h>
122#include <netinet/tcpip.h>
123
124#include <netinet/tcp_vtw.h>
125
126__KERNEL_RCSID(0, "$NetBSD: tcp_vtw.c,v 1.16 2016/07/28 07:54:31 martin Exp $");
127
128#define db_trace(__a, __b) do { } while (/*CONSTCOND*/0)
129
130static void vtw_debug_init(void);
131
132fatp_ctl_t fat_tcpv4;
133fatp_ctl_t fat_tcpv6;
134vtw_ctl_t vtw_tcpv4[VTW_NCLASS];
135vtw_ctl_t vtw_tcpv6[VTW_NCLASS];
136vtw_stats_t vtw_stats;
137
138/* We provide state for the lookup_ports iterator.
139 * As currently we are netlock-protected, there is one.
140 * If we were finer-grain, we would have one per CPU.
141 * I do not want to be in the business of alloc/free.
142 * The best alternate would be allocate on the caller's
143 * stack, but that would require them to know the struct,
144 * or at least the size.
145 * See how she goes.
146 */
147struct tcp_ports_iterator {
148 union {
149 struct in_addr v4;
150 struct in6_addr v6;
151 } addr;
152 u_int port;
153
154 uint32_t wild : 1;
155
156 vtw_ctl_t *ctl;
157 fatp_t *fp;
158
159 uint16_t slot_idx;
160 uint16_t ctl_idx;
161};
162
163static struct tcp_ports_iterator tcp_ports_iterator_v4;
164static struct tcp_ports_iterator tcp_ports_iterator_v6;
165
166static int vtw_age(vtw_ctl_t *, struct timeval *);
167
168/*!\brief allocate a fat pointer from a collection.
169 */
170static fatp_t *
171fatp_alloc(fatp_ctl_t *fat)
172{
173 fatp_t *fp = 0;
174
175 if (fat->nfree) {
176 fp = fat->free;
177 if (fp) {
178 fat->free = fatp_next(fat, fp);
179 --fat->nfree;
180 ++fat->nalloc;
181 fp->nxt = 0;
182
183 KASSERT(!fp->inuse);
184 }
185 }
186
187 return fp;
188}
189
190/*!\brief free a fat pointer.
191 */
192static void
193fatp_free(fatp_ctl_t *fat, fatp_t *fp)
194{
195 if (fp) {
196 KASSERT(!fp->inuse);
197 KASSERT(!fp->nxt);
198
199 fp->nxt = fatp_index(fat, fat->free);
200 fat->free = fp;
201
202 ++fat->nfree;
203 --fat->nalloc;
204 }
205}
206
207/*!\brief initialise a collection of fat pointers.
208 *
209 *\param n # hash buckets
210 *\param m total # fat pointers to allocate
211 *
212 * We allocate 2x as much, as we have two hashes: full and lport only.
213 */
214static void
215fatp_init(fatp_ctl_t *fat, uint32_t n, uint32_t m,
216 fatp_t *fat_base, fatp_t **fat_hash)
217{
218 fatp_t *fp;
219
220 KASSERT(n <= FATP_MAX / 2);
221
222 fat->hash = fat_hash;
223 fat->base = fat_base;
224
225 fat->port = &fat->hash[m];
226
227 fat->mask = m - 1; // ASSERT is power of 2 (m)
228 fat->lim = fat->base + 2*n - 1;
229 fat->nfree = 0;
230 fat->nalloc = 2*n;
231
232 /* Initialise the free list.
233 */
234 for (fp = fat->lim; fp >= fat->base; --fp) {
235 fatp_free(fat, fp);
236 }
237}
238
239/*
240 * The `xtra' is XORed into the tag stored.
241 */
242static uint32_t fatp_xtra[] = {
243 0x11111111,0x22222222,0x33333333,0x44444444,
244 0x55555555,0x66666666,0x77777777,0x88888888,
245 0x12121212,0x21212121,0x34343434,0x43434343,
246 0x56565656,0x65656565,0x78787878,0x87878787,
247 0x11221122,0x22112211,0x33443344,0x44334433,
248 0x55665566,0x66556655,0x77887788,0x88778877,
249 0x11112222,0x22221111,0x33334444,0x44443333,
250 0x55556666,0x66665555,0x77778888,0x88887777,
251};
252
253/*!\brief turn a {fatp_t*,slot} into an integral key.
254 *
255 * The key can be used to obtain the fatp_t, and the slot,
256 * as it directly encodes them.
257 */
258static inline uint32_t
259fatp_key(fatp_ctl_t *fat, fatp_t *fp, uint32_t slot)
260{
261 CTASSERT(CACHE_LINE_SIZE == 32 ||
262 CACHE_LINE_SIZE == 64 ||
263 CACHE_LINE_SIZE == 128);
264
265 switch (fatp_ntags()) {
266 case 7:
267 return (fatp_index(fat, fp) << 3) | slot;
268 case 15:
269 return (fatp_index(fat, fp) << 4) | slot;
270 case 31:
271 return (fatp_index(fat, fp) << 5) | slot;
272 default:
273 KASSERT(0 && "no support, for no good reason");
274 return ~0;
275 }
276}
277
278static inline uint32_t
279fatp_slot_from_key(fatp_ctl_t *fat, uint32_t key)
280{
281 CTASSERT(CACHE_LINE_SIZE == 32 ||
282 CACHE_LINE_SIZE == 64 ||
283 CACHE_LINE_SIZE == 128);
284
285 switch (fatp_ntags()) {
286 case 7:
287 return key & 7;
288 case 15:
289 return key & 15;
290 case 31:
291 return key & 31;
292 default:
293 KASSERT(0 && "no support, for no good reason");
294 return ~0;
295 }
296}
297
298static inline fatp_t *
299fatp_from_key(fatp_ctl_t *fat, uint32_t key)
300{
301 CTASSERT(CACHE_LINE_SIZE == 32 ||
302 CACHE_LINE_SIZE == 64 ||
303 CACHE_LINE_SIZE == 128);
304
305 switch (fatp_ntags()) {
306 case 7:
307 key >>= 3;
308 break;
309 case 15:
310 key >>= 4;
311 break;
312 case 31:
313 key >>= 5;
314 break;
315 default:
316 KASSERT(0 && "no support, for no good reason");
317 return 0;
318 }
319
320 return key ? fat->base + key - 1 : 0;
321}
322
323static inline uint32_t
324idx_encode(vtw_ctl_t *ctl, uint32_t idx)
325{
326 return (idx << ctl->idx_bits) | idx;
327}
328
329static inline uint32_t
330idx_decode(vtw_ctl_t *ctl, uint32_t bits)
331{
332 uint32_t idx = bits & ctl->idx_mask;
333
334 if (idx_encode(ctl, idx) == bits)
335 return idx;
336 else
337 return ~0;
338}
339
340/*!\brief insert index into fatp hash
341 *
342 *\param idx - index of element being placed in hash chain
343 *\param tag - 32-bit tag identifier
344 *
345 *\returns
346 * value which can be used to locate entry.
347 *
348 *\note
349 * we rely on the fact that there are unused high bits in the index
350 * for verification purposes on lookup.
351 */
352
353static inline uint32_t
354fatp_vtw_inshash(fatp_ctl_t *fat, uint32_t idx, uint32_t tag, int which,
355 void *dbg)
356{
357 fatp_t *fp;
358 fatp_t **hash = (which ? fat->port : fat->hash);
359 int i;
360
361 fp = hash[tag & fat->mask];
362
363 while (!fp || fatp_full(fp)) {
364 fatp_t *fq;
365
366 /* All entries are inuse at the top level.
367 * We allocate a spare, and push the top level
368 * down one. All entries in the fp we push down
369 * (think of a tape worm here) will be expelled sooner than
370 * any entries added subsequently to this hash bucket.
371 * This is a property of the time waits we are exploiting.
372 */
373
374 fq = fatp_alloc(fat);
375 if (!fq) {
376 vtw_age(fat->vtw, 0);
377 fp = hash[tag & fat->mask];
378 continue;
379 }
380
381 fq->inuse = 0;
382 fq->nxt = fatp_index(fat, fp);
383
384 hash[tag & fat->mask] = fq;
385
386 fp = fq;
387 }
388
389 KASSERT(!fatp_full(fp));
390
391 /* Fill highest index first. Lookup is lowest first.
392 */
393 for (i = fatp_ntags(); --i >= 0; ) {
394 if (!((1 << i) & fp->inuse)) {
395 break;
396 }
397 }
398
399 fp->inuse |= 1 << i;
400 fp->tag[i] = tag ^ idx_encode(fat->vtw, idx) ^ fatp_xtra[i];
401
402 db_trace(KTR_VTW
403 , (fp, "fat: inuse %5.5x tag[%x] %8.8x"
404 , fp->inuse
405 , i, fp->tag[i]));
406
407 return fatp_key(fat, fp, i);
408}
409
410static inline int
411vtw_alive(const vtw_t *vtw)
412{
413 return vtw->hashed && vtw->expire.tv_sec;
414}
415
416static inline uint32_t
417vtw_index_v4(vtw_ctl_t *ctl, vtw_v4_t *v4)
418{
419 if (ctl->base.v4 <= v4 && v4 <= ctl->lim.v4)
420 return v4 - ctl->base.v4;
421
422 KASSERT(0 && "vtw out of bounds");
423
424 return ~0;
425}
426
427static inline uint32_t
428vtw_index_v6(vtw_ctl_t *ctl, vtw_v6_t *v6)
429{
430 if (ctl->base.v6 <= v6 && v6 <= ctl->lim.v6)
431 return v6 - ctl->base.v6;
432
433 KASSERT(0 && "vtw out of bounds");
434
435 return ~0;
436}
437
438static inline uint32_t
439vtw_index(vtw_ctl_t *ctl, vtw_t *vtw)
440{
441 if (ctl->clidx)
442 ctl = ctl->ctl;
443
444 if (ctl->is_v4)
445 return vtw_index_v4(ctl, (vtw_v4_t *)vtw);
446
447 if (ctl->is_v6)
448 return vtw_index_v6(ctl, (vtw_v6_t *)vtw);
449
450 KASSERT(0 && "neither 4 nor 6. most curious.");
451
452 return ~0;
453}
454
455static inline vtw_t *
456vtw_from_index(vtw_ctl_t *ctl, uint32_t idx)
457{
458 if (ctl->clidx)
459 ctl = ctl->ctl;
460
461 /* See if the index looks like it might be an index.
462 * Bits on outside of the valid index bits is a give away.
463 */
464 idx = idx_decode(ctl, idx);
465
466 if (idx == ~0) {
467 return 0;
468 } else if (ctl->is_v4) {
469 vtw_v4_t *vtw = ctl->base.v4 + idx;
470
471 return (ctl->base.v4 <= vtw && vtw <= ctl->lim.v4)
472 ? &vtw->common : 0;
473 } else if (ctl->is_v6) {
474 vtw_v6_t *vtw = ctl->base.v6 + idx;
475
476 return (ctl->base.v6 <= vtw && vtw <= ctl->lim.v6)
477 ? &vtw->common : 0;
478 } else {
479 KASSERT(0 && "badness");
480 return 0;
481 }
482}
483
484/*!\brief return the next vtw after this one.
485 *
486 * Due to the differing sizes of the entries in differing
487 * arenas, we have to ensure we ++ the correct pointer type.
488 *
489 * Also handles wrap.
490 */
491static inline vtw_t *
492vtw_next(vtw_ctl_t *ctl, vtw_t *vtw)
493{
494 if (ctl->is_v4) {
495 vtw_v4_t *v4 = (void*)vtw;
496
497 vtw = &(++v4)->common;
498 } else {
499 vtw_v6_t *v6 = (void*)vtw;
500
501 vtw = &(++v6)->common;
502 }
503
504 if (vtw > ctl->lim.v)
505 vtw = ctl->base.v;
506
507 return vtw;
508}
509
510/*!\brief remove entry from FATP hash chains
511 */
512static inline void
513vtw_unhash(vtw_ctl_t *ctl, vtw_t *vtw)
514{
515 fatp_ctl_t *fat = ctl->fat;
516 fatp_t *fp;
517 uint32_t key = vtw->key;
518 uint32_t tag, slot, idx;
519 vtw_v4_t *v4 = (void*)vtw;
520 vtw_v6_t *v6 = (void*)vtw;
521
522 if (!vtw->hashed) {
523 KASSERT(0 && "unhashed");
524 return;
525 }
526
527 if (fat->vtw->is_v4) {
528 tag = v4_tag(v4->faddr, v4->fport, v4->laddr, v4->lport);
529 } else if (fat->vtw->is_v6) {
530 tag = v6_tag(&v6->faddr, v6->fport, &v6->laddr, v6->lport);
531 } else {
532 tag = 0;
533 KASSERT(0 && "not reached");
534 }
535
536 /* Remove from fat->hash[]
537 */
538 slot = fatp_slot_from_key(fat, key);
539 fp = fatp_from_key(fat, key);
540 idx = vtw_index(ctl, vtw);
541
542 db_trace(KTR_VTW
543 , (fp, "fat: del inuse %5.5x slot %x idx %x key %x tag %x"
544 , fp->inuse, slot, idx, key, tag));
545
546 KASSERT(fp->inuse & (1 << slot));
547 KASSERT(fp->tag[slot] == (tag ^ idx_encode(ctl, idx)
548 ^ fatp_xtra[slot]));
549
550 if ((fp->inuse & (1 << slot))
551 && fp->tag[slot] == (tag ^ idx_encode(ctl, idx)
552 ^ fatp_xtra[slot])) {
553 fp->inuse ^= 1 << slot;
554 fp->tag[slot] = 0;
555
556 /* When we delete entries, we do not compact. This is
557 * due to temporality. We add entries, and they
558 * (eventually) expire. Older entries will be further
559 * down the chain.
560 */
561 if (!fp->inuse) {
562 uint32_t hi = tag & fat->mask;
563 fatp_t *fq = 0;
564 fatp_t *fr = fat->hash[hi];
565
566 while (fr && fr != fp) {
567 fr = fatp_next(fat, fq = fr);
568 }
569
570 if (fr == fp) {
571 if (fq) {
572 fq->nxt = fp->nxt;
573 fp->nxt = 0;
574 fatp_free(fat, fp);
575 } else {
576 KASSERT(fat->hash[hi] == fp);
577
578 if (fp->nxt) {
579 fat->hash[hi]
580 = fatp_next(fat, fp);
581 fp->nxt = 0;
582 fatp_free(fat, fp);
583 } else {
584 /* retain for next use.
585 */
586 ;
587 }
588 }
589 } else {
590 fr = fat->hash[hi];
591
592 do {
593 db_trace(KTR_VTW
594 , (fr
595 , "fat:*del inuse %5.5x"
596 " nxt %x"
597 , fr->inuse, fr->nxt));
598
599 fr = fatp_next(fat, fq = fr);
600 } while (fr && fr != fp);
601
602 KASSERT(0 && "oops");
603 }
604 }
605 vtw->key ^= ~0;
606 }
607
608 if (fat->vtw->is_v4) {
609 tag = v4_port_tag(v4->lport);
610 } else if (fat->vtw->is_v6) {
611 tag = v6_port_tag(v6->lport);
612 }
613
614 /* Remove from fat->port[]
615 */
616 key = vtw->port_key;
617 slot = fatp_slot_from_key(fat, key);
618 fp = fatp_from_key(fat, key);
619 idx = vtw_index(ctl, vtw);
620
621 db_trace(KTR_VTW
622 , (fp, "fatport: del inuse %5.5x"
623 " slot %x idx %x key %x tag %x"
624 , fp->inuse, slot, idx, key, tag));
625
626 KASSERT(fp->inuse & (1 << slot));
627 KASSERT(fp->tag[slot] == (tag ^ idx_encode(ctl, idx)
628 ^ fatp_xtra[slot]));
629
630 if ((fp->inuse & (1 << slot))
631 && fp->tag[slot] == (tag ^ idx_encode(ctl, idx)
632 ^ fatp_xtra[slot])) {
633 fp->inuse ^= 1 << slot;
634 fp->tag[slot] = 0;
635
636 if (!fp->inuse) {
637 uint32_t hi = tag & fat->mask;
638 fatp_t *fq = 0;
639 fatp_t *fr = fat->port[hi];
640
641 while (fr && fr != fp) {
642 fr = fatp_next(fat, fq = fr);
643 }
644
645 if (fr == fp) {
646 if (fq) {
647 fq->nxt = fp->nxt;
648 fp->nxt = 0;
649 fatp_free(fat, fp);
650 } else {
651 KASSERT(fat->port[hi] == fp);
652
653 if (fp->nxt) {
654 fat->port[hi]
655 = fatp_next(fat, fp);
656 fp->nxt = 0;
657 fatp_free(fat, fp);
658 } else {
659 /* retain for next use.
660 */
661 ;
662 }
663 }
664 }
665 }
666 vtw->port_key ^= ~0;
667 }
668
669 vtw->hashed = 0;
670}
671
672/*!\brief remove entry from hash, possibly free.
673 */
674void
675vtw_del(vtw_ctl_t *ctl, vtw_t *vtw)
676{
677 KASSERT(mutex_owned(softnet_lock));
678
679 if (vtw->hashed) {
680 ++vtw_stats.del;
681 vtw_unhash(ctl, vtw);
682 }
683
684 /* We only delete the oldest entry.
685 */
686 if (vtw != ctl->oldest.v)
687 return;
688
689 --ctl->nalloc;
690 ++ctl->nfree;
691
692 vtw->expire.tv_sec = 0;
693 vtw->expire.tv_usec = ~0;
694
695 if (!ctl->nalloc)
696 ctl->oldest.v = 0;
697
698 ctl->oldest.v = vtw_next(ctl, vtw);
699}
700
701/*!\brief insert vestigial timewait in hash chain
702 */
703static void
704vtw_inshash_v4(vtw_ctl_t *ctl, vtw_t *vtw)
705{
706 uint32_t idx = vtw_index(ctl, vtw);
707 uint32_t tag;
708 vtw_v4_t *v4 = (void*)vtw;
709
710 KASSERT(mutex_owned(softnet_lock));
711 KASSERT(!vtw->hashed);
712 KASSERT(ctl->clidx == vtw->msl_class);
713
714 ++vtw_stats.ins;
715
716 tag = v4_tag(v4->faddr, v4->fport,
717 v4->laddr, v4->lport);
718
719 vtw->key = fatp_vtw_inshash(ctl->fat, idx, tag, 0, vtw);
720
721 db_trace(KTR_VTW, (ctl
722 , "vtw: ins %8.8x:%4.4x %8.8x:%4.4x"
723 " tag %8.8x key %8.8x"
724 , v4->faddr, v4->fport
725 , v4->laddr, v4->lport
726 , tag
727 , vtw->key));
728
729 tag = v4_port_tag(v4->lport);
730 vtw->port_key = fatp_vtw_inshash(ctl->fat, idx, tag, 1, vtw);
731
732 db_trace(KTR_VTW, (ctl, "vtw: ins %P - %4.4x tag %8.8x key %8.8x"
733 , v4->lport, v4->lport
734 , tag
735 , vtw->key));
736
737 vtw->hashed = 1;
738}
739
740/*!\brief insert vestigial timewait in hash chain
741 */
742static void
743vtw_inshash_v6(vtw_ctl_t *ctl, vtw_t *vtw)
744{
745 uint32_t idx = vtw_index(ctl, vtw);
746 uint32_t tag;
747 vtw_v6_t *v6 = (void*)vtw;
748
749 KASSERT(mutex_owned(softnet_lock));
750 KASSERT(!vtw->hashed);
751 KASSERT(ctl->clidx == vtw->msl_class);
752
753 ++vtw_stats.ins;
754
755 tag = v6_tag(&v6->faddr, v6->fport,
756 &v6->laddr, v6->lport);
757
758 vtw->key = fatp_vtw_inshash(ctl->fat, idx, tag, 0, vtw);
759
760 tag = v6_port_tag(v6->lport);
761 vtw->port_key = fatp_vtw_inshash(ctl->fat, idx, tag, 1, vtw);
762
763 db_trace(KTR_VTW, (ctl, "vtw: ins %P - %4.4x tag %8.8x key %8.8x"
764 , v6->lport, v6->lport
765 , tag
766 , vtw->key));
767
768 vtw->hashed = 1;
769}
770
771static vtw_t *
772vtw_lookup_hash_v4(vtw_ctl_t *ctl, uint32_t faddr, uint16_t fport
773 , uint32_t laddr, uint16_t lport
774 , int which)
775{
776 vtw_v4_t *v4;
777 vtw_t *vtw;
778 uint32_t tag;
779 fatp_t *fp;
780 int i;
781 uint32_t fatps = 0, probes = 0, losings = 0;
782
783 if (!ctl || !ctl->fat)
784 return 0;
785
786 ++vtw_stats.look[which];
787
788 if (which) {
789 tag = v4_port_tag(lport);
790 fp = ctl->fat->port[tag & ctl->fat->mask];
791 } else {
792 tag = v4_tag(faddr, fport, laddr, lport);
793 fp = ctl->fat->hash[tag & ctl->fat->mask];
794 }
795
796 while (fp && fp->inuse) {
797 uint32_t inuse = fp->inuse;
798
799 ++fatps;
800
801 for (i = 0; inuse && i < fatp_ntags(); ++i) {
802 uint32_t idx;
803
804 if (!(inuse & (1 << i)))
805 continue;
806
807 inuse ^= 1 << i;
808
809 ++probes;
810 ++vtw_stats.probe[which];
811
812 idx = fp->tag[i] ^ tag ^ fatp_xtra[i];
813 vtw = vtw_from_index(ctl, idx);
814
815 if (!vtw) {
816 /* Hopefully fast path.
817 */
818 db_trace(KTR_VTW
819 , (fp, "vtw: fast %A:%P %A:%P"
820 " idx %x tag %x"
821 , faddr, fport
822 , laddr, lport
823 , idx, tag));
824 continue;
825 }
826
827 v4 = (void*)vtw;
828
829 /* The de-referencing of vtw is what we want to avoid.
830 * Losing.
831 */
832 if (vtw_alive(vtw)
833 && ((which ? vtw->port_key : vtw->key)
834 == fatp_key(ctl->fat, fp, i))
835 && (which
836 || (v4->faddr == faddr && v4->laddr == laddr
837 && v4->fport == fport))
838 && v4->lport == lport) {
839 ++vtw_stats.hit[which];
840
841 db_trace(KTR_VTW
842 , (fp, "vtw: hit %8.8x:%4.4x"
843 " %8.8x:%4.4x idx %x key %x"
844 , faddr, fport
845 , laddr, lport
846 , idx_decode(ctl, idx), vtw->key));
847
848 KASSERT(vtw->hashed);
849
850 goto out;
851 }
852 ++vtw_stats.losing[which];
853 ++losings;
854
855 if (vtw_alive(vtw)) {
856 db_trace(KTR_VTW
857 , (fp, "vtw:!mis %8.8x:%4.4x"
858 " %8.8x:%4.4x key %x tag %x"
859 , faddr, fport
860 , laddr, lport
861 , fatp_key(ctl->fat, fp, i)
862 , v4_tag(faddr, fport
863 , laddr, lport)));
864 db_trace(KTR_VTW
865 , (vtw, "vtw:!mis %8.8x:%4.4x"
866 " %8.8x:%4.4x key %x tag %x"
867 , v4->faddr, v4->fport
868 , v4->laddr, v4->lport
869 , vtw->key
870 , v4_tag(v4->faddr, v4->fport
871 , v4->laddr, v4->lport)));
872
873 if (vtw->key == fatp_key(ctl->fat, fp, i)) {
874 db_trace(KTR_VTW
875 , (vtw, "vtw:!mis %8.8x:%4.4x"
876 " %8.8x:%4.4x key %x"
877 " which %x"
878 , v4->faddr, v4->fport
879 , v4->laddr, v4->lport
880 , vtw->key
881 , which));
882
883 } else {
884 db_trace(KTR_VTW
885 , (vtw
886 , "vtw:!mis"
887 " key %8.8x != %8.8x"
888 " idx %x i %x which %x"
889 , vtw->key
890 , fatp_key(ctl->fat, fp, i)
891 , idx_decode(ctl, idx)
892 , i
893 , which));
894 }
895 } else {
896 db_trace(KTR_VTW
897 , (fp
898 , "vtw:!mis free entry"
899 " idx %x vtw %p which %x"
900 , idx_decode(ctl, idx)
901 , vtw, which));
902 }
903 }
904
905 if (fp->nxt) {
906 fp = fatp_next(ctl->fat, fp);
907 } else {
908 break;
909 }
910 }
911 ++vtw_stats.miss[which];
912 vtw = 0;
913out:
914 if (fatps > vtw_stats.max_chain[which])
915 vtw_stats.max_chain[which] = fatps;
916 if (probes > vtw_stats.max_probe[which])
917 vtw_stats.max_probe[which] = probes;
918 if (losings > vtw_stats.max_loss[which])
919 vtw_stats.max_loss[which] = losings;
920
921 return vtw;
922}
923
924static vtw_t *
925vtw_lookup_hash_v6(vtw_ctl_t *ctl, const struct in6_addr *faddr, uint16_t fport
926 , const struct in6_addr *laddr, uint16_t lport
927 , int which)
928{
929 vtw_v6_t *v6;
930 vtw_t *vtw;
931 uint32_t tag;
932 fatp_t *fp;
933 int i;
934 uint32_t fatps = 0, probes = 0, losings = 0;
935
936 ++vtw_stats.look[which];
937
938 if (!ctl || !ctl->fat)
939 return 0;
940
941 if (which) {
942 tag = v6_port_tag(lport);
943 fp = ctl->fat->port[tag & ctl->fat->mask];
944 } else {
945 tag = v6_tag(faddr, fport, laddr, lport);
946 fp = ctl->fat->hash[tag & ctl->fat->mask];
947 }
948
949 while (fp && fp->inuse) {
950 uint32_t inuse = fp->inuse;
951
952 ++fatps;
953
954 for (i = 0; inuse && i < fatp_ntags(); ++i) {
955 uint32_t idx;
956
957 if (!(inuse & (1 << i)))
958 continue;
959
960 inuse ^= 1 << i;
961
962 ++probes;
963 ++vtw_stats.probe[which];
964
965 idx = fp->tag[i] ^ tag ^ fatp_xtra[i];
966 vtw = vtw_from_index(ctl, idx);
967
968 db_trace(KTR_VTW
969 , (fp, "probe: %2d %6A:%4.4x %6A:%4.4x idx %x"
970 , i
971 , db_store(faddr, sizeof (*faddr)), fport
972 , db_store(laddr, sizeof (*laddr)), lport
973 , idx_decode(ctl, idx)));
974
975 if (!vtw) {
976 /* Hopefully fast path.
977 */
978 continue;
979 }
980
981 v6 = (void*)vtw;
982
983 if (vtw_alive(vtw)
984 && ((which ? vtw->port_key : vtw->key)
985 == fatp_key(ctl->fat, fp, i))
986 && v6->lport == lport
987 && (which
988 || (v6->fport == fport
989 && !bcmp(&v6->faddr, faddr, sizeof (*faddr))
990 && !bcmp(&v6->laddr, laddr
991 , sizeof (*laddr))))) {
992 ++vtw_stats.hit[which];
993
994 KASSERT(vtw->hashed);
995 goto out;
996 } else {
997 ++vtw_stats.losing[which];
998 ++losings;
999 }
1000 }
1001
1002 if (fp->nxt) {
1003 fp = fatp_next(ctl->fat, fp);
1004 } else {
1005 break;
1006 }
1007 }
1008 ++vtw_stats.miss[which];
1009 vtw = 0;
1010out:
1011 if (fatps > vtw_stats.max_chain[which])
1012 vtw_stats.max_chain[which] = fatps;
1013 if (probes > vtw_stats.max_probe[which])
1014 vtw_stats.max_probe[which] = probes;
1015 if (losings > vtw_stats.max_loss[which])
1016 vtw_stats.max_loss[which] = losings;
1017
1018 return vtw;
1019}
1020
1021/*!\brief port iterator
1022 */
1023static vtw_t *
1024vtw_next_port_v4(struct tcp_ports_iterator *it)
1025{
1026 vtw_ctl_t *ctl = it->ctl;
1027 vtw_v4_t *v4;
1028 vtw_t *vtw;
1029 uint32_t tag;
1030 uint16_t lport = it->port;
1031 fatp_t *fp;
1032 int i;
1033 uint32_t fatps = 0, probes = 0, losings = 0;
1034
1035 tag = v4_port_tag(lport);
1036 if (!it->fp) {
1037 it->fp = ctl->fat->port[tag & ctl->fat->mask];
1038 it->slot_idx = 0;
1039 }
1040 fp = it->fp;
1041
1042 while (fp) {
1043 uint32_t inuse = fp->inuse;
1044
1045 ++fatps;
1046
1047 for (i = it->slot_idx; inuse && i < fatp_ntags(); ++i) {
1048 uint32_t idx;
1049
1050 if (!(inuse & (1 << i)))
1051 continue;
1052
1053 inuse &= ~0U << i;
1054
1055 if (i < it->slot_idx)
1056 continue;
1057
1058 ++vtw_stats.probe[1];
1059 ++probes;
1060
1061 idx = fp->tag[i] ^ tag ^ fatp_xtra[i];
1062 vtw = vtw_from_index(ctl, idx);
1063
1064 if (!vtw) {
1065 /* Hopefully fast path.
1066 */
1067 continue;
1068 }
1069
1070 v4 = (void*)vtw;
1071
1072 if (vtw_alive(vtw)
1073 && vtw->port_key == fatp_key(ctl->fat, fp, i)
1074 && v4->lport == lport) {
1075 ++vtw_stats.hit[1];
1076
1077 it->slot_idx = i + 1;
1078
1079 goto out;
1080 } else if (vtw_alive(vtw)) {
1081 ++vtw_stats.losing[1];
1082 ++losings;
1083
1084 db_trace(KTR_VTW
1085 , (vtw, "vtw:!mis"
1086 " port %8.8x:%4.4x %8.8x:%4.4x"
1087 " key %x port %x"
1088 , v4->faddr, v4->fport
1089 , v4->laddr, v4->lport
1090 , vtw->key
1091 , lport));
1092 } else {
1093 /* Really losing here. We are coming
1094 * up with references to free entries.
1095 * Might find it better to use
1096 * traditional, or need another
1097 * add-hockery. The other add-hockery
1098 * would be to pul more into into the
1099 * cache line to reject the false
1100 * hits.
1101 */
1102 ++vtw_stats.losing[1];
1103 ++losings;
1104 db_trace(KTR_VTW
1105 , (fp, "vtw:!mis port %x"
1106 " - free entry idx %x vtw %p"
1107 , lport
1108 , idx_decode(ctl, idx)
1109 , vtw));
1110 }
1111 }
1112
1113 if (fp->nxt) {
1114 it->fp = fp = fatp_next(ctl->fat, fp);
1115 it->slot_idx = 0;
1116 } else {
1117 it->fp = 0;
1118 break;
1119 }
1120 }
1121 ++vtw_stats.miss[1];
1122
1123 vtw = 0;
1124out:
1125 if (fatps > vtw_stats.max_chain[1])
1126 vtw_stats.max_chain[1] = fatps;
1127 if (probes > vtw_stats.max_probe[1])
1128 vtw_stats.max_probe[1] = probes;
1129 if (losings > vtw_stats.max_loss[1])
1130 vtw_stats.max_loss[1] = losings;
1131
1132 return vtw;
1133}
1134
1135/*!\brief port iterator
1136 */
1137static vtw_t *
1138vtw_next_port_v6(struct tcp_ports_iterator *it)
1139{
1140 vtw_ctl_t *ctl = it->ctl;
1141 vtw_v6_t *v6;
1142 vtw_t *vtw;
1143 uint32_t tag;
1144 uint16_t lport = it->port;
1145 fatp_t *fp;
1146 int i;
1147 uint32_t fatps = 0, probes = 0, losings = 0;
1148
1149 tag = v6_port_tag(lport);
1150 if (!it->fp) {
1151 it->fp = ctl->fat->port[tag & ctl->fat->mask];
1152 it->slot_idx = 0;
1153 }
1154 fp = it->fp;
1155
1156 while (fp) {
1157 uint32_t inuse = fp->inuse;
1158
1159 ++fatps;
1160
1161 for (i = it->slot_idx; inuse && i < fatp_ntags(); ++i) {
1162 uint32_t idx;
1163
1164 if (!(inuse & (1 << i)))
1165 continue;
1166
1167 inuse &= ~0U << i;
1168
1169 if (i < it->slot_idx)
1170 continue;
1171
1172 ++vtw_stats.probe[1];
1173 ++probes;
1174
1175 idx = fp->tag[i] ^ tag ^ fatp_xtra[i];
1176 vtw = vtw_from_index(ctl, idx);
1177
1178 if (!vtw) {
1179 /* Hopefully fast path.
1180 */
1181 continue;
1182 }
1183
1184 v6 = (void*)vtw;
1185
1186 db_trace(KTR_VTW
1187 , (vtw, "vtw: i %x idx %x fp->tag %x"
1188 " tag %x xtra %x"
1189 , i, idx_decode(ctl, idx)
1190 , fp->tag[i], tag, fatp_xtra[i]));
1191
1192 if (vtw_alive(vtw)
1193 && vtw->port_key == fatp_key(ctl->fat, fp, i)
1194 && v6->lport == lport) {
1195 ++vtw_stats.hit[1];
1196
1197 db_trace(KTR_VTW
1198 , (fp, "vtw: nxt port %P - %4.4x"
1199 " idx %x key %x"
1200 , lport, lport
1201 , idx_decode(ctl, idx), vtw->key));
1202
1203 it->slot_idx = i + 1;
1204 goto out;
1205 } else if (vtw_alive(vtw)) {
1206 ++vtw_stats.losing[1];
1207
1208 db_trace(KTR_VTW
1209 , (vtw, "vtw:!mis port %6A:%4.4x"
1210 " %6A:%4.4x key %x port %x"
1211 , db_store(&v6->faddr
1212 , sizeof (v6->faddr))
1213 , v6->fport
1214 , db_store(&v6->laddr
1215 , sizeof (v6->faddr))
1216 , v6->lport
1217 , vtw->key
1218 , lport));
1219 } else {
1220 /* Really losing here. We are coming
1221 * up with references to free entries.
1222 * Might find it better to use
1223 * traditional, or need another
1224 * add-hockery. The other add-hockery
1225 * would be to pul more into into the
1226 * cache line to reject the false
1227 * hits.
1228 */
1229 ++vtw_stats.losing[1];
1230 ++losings;
1231
1232 db_trace(KTR_VTW
1233 , (fp
1234 , "vtw:!mis port %x"
1235 " - free entry idx %x vtw %p"
1236 , lport, idx_decode(ctl, idx)
1237 , vtw));
1238 }
1239 }
1240
1241 if (fp->nxt) {
1242 it->fp = fp = fatp_next(ctl->fat, fp);
1243 it->slot_idx = 0;
1244 } else {
1245 it->fp = 0;
1246 break;
1247 }
1248 }
1249 ++vtw_stats.miss[1];
1250
1251 vtw = 0;
1252out:
1253 if (fatps > vtw_stats.max_chain[1])
1254 vtw_stats.max_chain[1] = fatps;
1255 if (probes > vtw_stats.max_probe[1])
1256 vtw_stats.max_probe[1] = probes;
1257 if (losings > vtw_stats.max_loss[1])
1258 vtw_stats.max_loss[1] = losings;
1259
1260 return vtw;
1261}
1262
1263/*!\brief initialise the VTW allocation arena
1264 *
1265 * There are 1+3 allocation classes:
1266 * 0 classless
1267 * {1,2,3} MSL-class based allocation
1268 *
1269 * The allocation arenas are all initialised. Classless gets all the
1270 * space. MSL-class based divides the arena, so that allocation
1271 * within a class can proceed without having to consider entries
1272 * (aka: cache lines) from different classes.
1273 *
1274 * Usually, we are completely classless or class-based, but there can be
1275 * transition periods, corresponding to dynamic adjustments in the config
1276 * by the operator.
1277 */
1278static void
1279vtw_init(fatp_ctl_t *fat, vtw_ctl_t *ctl, const uint32_t n, vtw_t *ctl_base_v)
1280{
1281 int class_n, i;
1282 vtw_t *base;
1283
1284 ctl->base.v = ctl_base_v;
1285
1286 if (ctl->is_v4) {
1287 ctl->lim.v4 = ctl->base.v4 + n - 1;
1288 ctl->alloc.v4 = ctl->base.v4;
1289 } else {
1290 ctl->lim.v6 = ctl->base.v6 + n - 1;
1291 ctl->alloc.v6 = ctl->base.v6;
1292 }
1293
1294 ctl->nfree = n;
1295 ctl->ctl = ctl;
1296
1297 ctl->idx_bits = 32;
1298 for (ctl->idx_mask = ~0; (ctl->idx_mask & (n-1)) == n-1; ) {
1299 ctl->idx_mask >>= 1;
1300 ctl->idx_bits -= 1;
1301 }
1302
1303 ctl->idx_mask <<= 1;
1304 ctl->idx_mask |= 1;
1305 ctl->idx_bits += 1;
1306
1307 ctl->fat = fat;
1308 fat->vtw = ctl;
1309
1310 /* Divide the resources equally amongst the classes.
1311 * This is not optimal, as the different classes
1312 * arrive and leave at different rates, but it is
1313 * the best I can do for now.
1314 */
1315 class_n = n / (VTW_NCLASS-1);
1316 base = ctl->base.v;
1317
1318 for (i = 1; i < VTW_NCLASS; ++i) {
1319 int j;
1320
1321 ctl[i] = ctl[0];
1322 ctl[i].clidx = i;
1323
1324 ctl[i].base.v = base;
1325 ctl[i].alloc = ctl[i].base;
1326
1327 for (j = 0; j < class_n - 1; ++j) {
1328 if (tcp_msl_enable)
1329 base->msl_class = i;
1330 base = vtw_next(ctl, base);
1331 }
1332
1333 ctl[i].lim.v = base;
1334 base = vtw_next(ctl, base);
1335 ctl[i].nfree = class_n;
1336 }
1337
1338 vtw_debug_init();
1339}
1340
1341/*!\brief map class to TCP MSL
1342 */
1343static inline uint32_t
1344class_to_msl(int msl_class)
1345{
1346 switch (msl_class) {
1347 case 0:
1348 case 1:
1349 return tcp_msl_remote ? tcp_msl_remote : (TCPTV_MSL >> 0);
1350 case 2:
1351 return tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
1352 default:
1353 return tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
1354 }
1355}
1356
1357/*!\brief map TCP MSL to class
1358 */
1359static inline uint32_t
1360msl_to_class(int msl)
1361{
1362 if (tcp_msl_enable) {
1363 if (msl <= (tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2)))
1364 return 1+2;
1365 if (msl <= (tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1)))
1366 return 1+1;
1367 return 1;
1368 }
1369 return 0;
1370}
1371
1372/*!\brief allocate a vtw entry
1373 */
1374static inline vtw_t *
1375vtw_alloc(vtw_ctl_t *ctl)
1376{
1377 vtw_t *vtw = 0;
1378 int stuck = 0;
1379 int avail = ctl ? (ctl->nalloc + ctl->nfree) : 0;
1380 int msl;
1381
1382 KASSERT(mutex_owned(softnet_lock));
1383
1384 /* If no resources, we will not get far.
1385 */
1386 if (!ctl || !ctl->base.v4 || avail <= 0)
1387 return 0;
1388
1389 /* Obtain a free one.
1390 */
1391 while (!ctl->nfree) {
1392 vtw_age(ctl, 0);
1393
1394 if (++stuck > avail) {
1395 /* When in transition between
1396 * schemes (classless, classed) we
1397 * can be stuck having to await the
1398 * expiration of cross-allocated entries.
1399 *
1400 * Returning zero means we will fall back to the
1401 * traditional TIME_WAIT handling, except in the
1402 * case of a re-shed, in which case we cannot
1403 * perform the reshecd, but will retain the extant
1404 * entry.
1405 */
1406 db_trace(KTR_VTW
1407 , (ctl, "vtw:!none free in class %x %x/%x"
1408 , ctl->clidx
1409 , ctl->nalloc, ctl->nfree));
1410
1411 return 0;
1412 }
1413 }
1414
1415 vtw = ctl->alloc.v;
1416
1417 if (vtw->msl_class != ctl->clidx) {
1418 /* Usurping rules:
1419 * 0 -> {1,2,3} or {1,2,3} -> 0
1420 */
1421 KASSERT(!vtw->msl_class || !ctl->clidx);
1422
1423 if (vtw->hashed || vtw->expire.tv_sec) {
1424 /* As this is owned by some other class,
1425 * we must wait for it to expire it.
1426 * This will only happen on class/classless
1427 * transitions, which are guaranteed to progress
1428 * to completion in small finite time, barring bugs.
1429 */
1430 db_trace(KTR_VTW
1431 , (ctl, "vtw:!%p class %x!=%x %x:%x%s"
1432 , vtw, vtw->msl_class, ctl->clidx
1433 , vtw->expire.tv_sec
1434 , vtw->expire.tv_usec
1435 , vtw->hashed ? " hashed" : ""));
1436
1437 return 0;
1438 }
1439
1440 db_trace(KTR_VTW
1441 , (ctl, "vtw:!%p usurped from %x to %x"
1442 , vtw, vtw->msl_class, ctl->clidx));
1443
1444 vtw->msl_class = ctl->clidx;
1445 }
1446
1447 if (vtw_alive(vtw)) {
1448 KASSERT(0 && "next free not free");
1449 return 0;
1450 }
1451
1452 /* Advance allocation poiter.
1453 */
1454 ctl->alloc.v = vtw_next(ctl, vtw);
1455
1456 --ctl->nfree;
1457 ++ctl->nalloc;
1458
1459 msl = (2 * class_to_msl(ctl->clidx) * 1000) / PR_SLOWHZ; // msec
1460
1461 /* mark expiration
1462 */
1463 getmicrouptime(&vtw->expire);
1464
1465 /* Move expiration into the future.
1466 */
1467 vtw->expire.tv_sec += msl / 1000;
1468 vtw->expire.tv_usec += 1000 * (msl % 1000);
1469
1470 while (vtw->expire.tv_usec >= 1000*1000) {
1471 vtw->expire.tv_usec -= 1000*1000;
1472 vtw->expire.tv_sec += 1;
1473 }
1474
1475 if (!ctl->oldest.v)
1476 ctl->oldest.v = vtw;
1477
1478 return vtw;
1479}
1480
1481/*!\brief expiration
1482 */
1483static int
1484vtw_age(vtw_ctl_t *ctl, struct timeval *_when)
1485{
1486 vtw_t *vtw;
1487 struct timeval then, *when = _when;
1488 int maxtries = 0;
1489
1490 if (!ctl->oldest.v) {
1491 KASSERT(!ctl->nalloc);
1492 return 0;
1493 }
1494
1495 for (vtw = ctl->oldest.v; vtw && ctl->nalloc; ) {
1496 if (++maxtries > ctl->nalloc)
1497 break;
1498
1499 if (vtw->msl_class != ctl->clidx) {
1500 db_trace(KTR_VTW
1501 , (vtw, "vtw:!age class mismatch %x != %x"
1502 , vtw->msl_class, ctl->clidx));
1503 /* XXXX
1504 * See if the appropriate action is to skip to the next.
1505 * XXXX
1506 */
1507 ctl->oldest.v = vtw = vtw_next(ctl, vtw);
1508 continue;
1509 }
1510 if (!when) {
1511 /* Latch oldest timeval if none specified.
1512 */
1513 then = vtw->expire;
1514 when = &then;
1515 }
1516
1517 if (!timercmp(&vtw->expire, when, <=))
1518 break;
1519
1520 db_trace(KTR_VTW
1521 , (vtw, "vtw: expire %x %8.8x:%8.8x %x/%x"
1522 , ctl->clidx
1523 , vtw->expire.tv_sec
1524 , vtw->expire.tv_usec
1525 , ctl->nalloc
1526 , ctl->nfree));
1527
1528 if (!_when)
1529 ++vtw_stats.kill;
1530
1531 vtw_del(ctl, vtw);
1532 vtw = ctl->oldest.v;
1533 }
1534
1535 return ctl->nalloc; // # remaining allocated
1536}
1537
1538static callout_t vtw_cs;
1539
1540/*!\brief notice the passage of time.
1541 * It seems to be getting faster. What happened to the year?
1542 */
1543static void
1544vtw_tick(void *arg)
1545{
1546 struct timeval now;
1547 int i, cnt = 0;
1548
1549 getmicrouptime(&now);
1550
1551 db_trace(KTR_VTW, (arg, "vtk: tick - now %8.8x:%8.8x"
1552 , now.tv_sec, now.tv_usec));
1553
1554 mutex_enter(softnet_lock);
1555
1556 for (i = 0; i < VTW_NCLASS; ++i) {
1557 cnt += vtw_age(&vtw_tcpv4[i], &now);
1558 cnt += vtw_age(&vtw_tcpv6[i], &now);
1559 }
1560
1561 /* Keep ticks coming while we need them.
1562 */
1563 if (cnt)
1564 callout_schedule(&vtw_cs, hz / 5);
1565 else {
1566 tcp_vtw_was_enabled = 0;
1567 tcbtable.vestige = 0;
1568 }
1569 mutex_exit(softnet_lock);
1570}
1571
1572/* in_pcblookup_ports assist for handling vestigial entries.
1573 */
1574static void *
1575tcp_init_ports_v4(struct in_addr addr, u_int port, int wild)
1576{
1577 struct tcp_ports_iterator *it = &tcp_ports_iterator_v4;
1578
1579 bzero(it, sizeof (*it));
1580
1581 /* Note: the reference to vtw_tcpv4[0] is fine.
1582 * We do not need per-class iteration. We just
1583 * need to get to the fat, and there is one
1584 * shared fat.
1585 */
1586 if (vtw_tcpv4[0].fat) {
1587 it->addr.v4 = addr;
1588 it->port = port;
1589 it->wild = !!wild;
1590 it->ctl = &vtw_tcpv4[0];
1591
1592 ++vtw_stats.look[1];
1593 }
1594
1595 return it;
1596}
1597
1598/*!\brief export an IPv4 vtw.
1599 */
1600static int
1601vtw_export_v4(vtw_ctl_t *ctl, vtw_t *vtw, vestigial_inpcb_t *res)
1602{
1603 vtw_v4_t *v4 = (void*)vtw;
1604
1605 bzero(res, sizeof (*res));
1606
1607 if (ctl && vtw) {
1608 if (!ctl->clidx && vtw->msl_class)
1609 ctl += vtw->msl_class;
1610 else
1611 KASSERT(ctl->clidx == vtw->msl_class);
1612
1613 res->valid = 1;
1614 res->v4 = 1;
1615
1616 res->faddr.v4.s_addr = v4->faddr;
1617 res->laddr.v4.s_addr = v4->laddr;
1618 res->fport = v4->fport;
1619 res->lport = v4->lport;
1620 res->vtw = vtw; // netlock held over call(s)
1621 res->ctl = ctl;
1622 res->reuse_addr = vtw->reuse_addr;
1623 res->reuse_port = vtw->reuse_port;
1624 res->snd_nxt = vtw->snd_nxt;
1625 res->rcv_nxt = vtw->rcv_nxt;
1626 res->rcv_wnd = vtw->rcv_wnd;
1627 res->uid = vtw->uid;
1628 }
1629
1630 return res->valid;
1631}
1632
1633/*!\brief return next port in the port iterator. yowza.
1634 */
1635static int
1636tcp_next_port_v4(void *arg, struct vestigial_inpcb *res)
1637{
1638 struct tcp_ports_iterator *it = arg;
1639 vtw_t *vtw = 0;
1640
1641 if (it->ctl)
1642 vtw = vtw_next_port_v4(it);
1643
1644 if (!vtw)
1645 it->ctl = 0;
1646
1647 return vtw_export_v4(it->ctl, vtw, res);
1648}
1649
1650static int
1651tcp_lookup_v4(struct in_addr faddr, uint16_t fport,
1652 struct in_addr laddr, uint16_t lport,
1653 struct vestigial_inpcb *res)
1654{
1655 vtw_t *vtw;
1656 vtw_ctl_t *ctl;
1657
1658
1659 db_trace(KTR_VTW
1660 , (res, "vtw: lookup %A:%P %A:%P"
1661 , faddr, fport
1662 , laddr, lport));
1663
1664 vtw = vtw_lookup_hash_v4((ctl = &vtw_tcpv4[0])
1665 , faddr.s_addr, fport
1666 , laddr.s_addr, lport, 0);
1667
1668 return vtw_export_v4(ctl, vtw, res);
1669}
1670
1671/* in_pcblookup_ports assist for handling vestigial entries.
1672 */
1673static void *
1674tcp_init_ports_v6(const struct in6_addr *addr, u_int port, int wild)
1675{
1676 struct tcp_ports_iterator *it = &tcp_ports_iterator_v6;
1677
1678 bzero(it, sizeof (*it));
1679
1680 /* Note: the reference to vtw_tcpv6[0] is fine.
1681 * We do not need per-class iteration. We just
1682 * need to get to the fat, and there is one
1683 * shared fat.
1684 */
1685 if (vtw_tcpv6[0].fat) {
1686 it->addr.v6 = *addr;
1687 it->port = port;
1688 it->wild = !!wild;
1689 it->ctl = &vtw_tcpv6[0];
1690
1691 ++vtw_stats.look[1];
1692 }
1693
1694 return it;
1695}
1696
1697/*!\brief export an IPv6 vtw.
1698 */
1699static int
1700vtw_export_v6(vtw_ctl_t *ctl, vtw_t *vtw, vestigial_inpcb_t *res)
1701{
1702 vtw_v6_t *v6 = (void*)vtw;
1703
1704 bzero(res, sizeof (*res));
1705
1706 if (ctl && vtw) {
1707 if (!ctl->clidx && vtw->msl_class)
1708 ctl += vtw->msl_class;
1709 else
1710 KASSERT(ctl->clidx == vtw->msl_class);
1711
1712 res->valid = 1;
1713 res->v4 = 0;
1714
1715 res->faddr.v6 = v6->faddr;
1716 res->laddr.v6 = v6->laddr;
1717 res->fport = v6->fport;
1718 res->lport = v6->lport;
1719 res->vtw = vtw; // netlock held over call(s)
1720 res->ctl = ctl;
1721
1722 res->v6only = vtw->v6only;
1723 res->reuse_addr = vtw->reuse_addr;
1724 res->reuse_port = vtw->reuse_port;
1725
1726 res->snd_nxt = vtw->snd_nxt;
1727 res->rcv_nxt = vtw->rcv_nxt;
1728 res->rcv_wnd = vtw->rcv_wnd;
1729 res->uid = vtw->uid;
1730 }
1731
1732 return res->valid;
1733}
1734
1735static int
1736tcp_next_port_v6(void *arg, struct vestigial_inpcb *res)
1737{
1738 struct tcp_ports_iterator *it = arg;
1739 vtw_t *vtw = 0;
1740
1741 if (it->ctl)
1742 vtw = vtw_next_port_v6(it);
1743
1744 if (!vtw)
1745 it->ctl = 0;
1746
1747 return vtw_export_v6(it->ctl, vtw, res);
1748}
1749
1750static int
1751tcp_lookup_v6(const struct in6_addr *faddr, uint16_t fport,
1752 const struct in6_addr *laddr, uint16_t lport,
1753 struct vestigial_inpcb *res)
1754{
1755 vtw_ctl_t *ctl;
1756 vtw_t *vtw;
1757
1758 db_trace(KTR_VTW
1759 , (res, "vtw: lookup %6A:%P %6A:%P"
1760 , db_store(faddr, sizeof (*faddr)), fport
1761 , db_store(laddr, sizeof (*laddr)), lport));
1762
1763 vtw = vtw_lookup_hash_v6((ctl = &vtw_tcpv6[0])
1764 , faddr, fport
1765 , laddr, lport, 0);
1766
1767 return vtw_export_v6(ctl, vtw, res);
1768}
1769
1770static vestigial_hooks_t tcp_hooks = {
1771 .init_ports4 = tcp_init_ports_v4,
1772 .next_port4 = tcp_next_port_v4,
1773 .lookup4 = tcp_lookup_v4,
1774 .init_ports6 = tcp_init_ports_v6,
1775 .next_port6 = tcp_next_port_v6,
1776 .lookup6 = tcp_lookup_v6,
1777};
1778
1779static bool
1780vtw_select(int af, fatp_ctl_t **fatp, vtw_ctl_t **ctlp)
1781{
1782 fatp_ctl_t *fat;
1783 vtw_ctl_t *ctl;
1784
1785 switch (af) {
1786 case AF_INET:
1787 fat = &fat_tcpv4;
1788 ctl = &vtw_tcpv4[0];
1789 break;
1790 case AF_INET6:
1791 fat = &fat_tcpv6;
1792 ctl = &vtw_tcpv6[0];
1793 break;
1794 default:
1795 return false;
1796 }
1797 if (fatp != NULL)
1798 *fatp = fat;
1799 if (ctlp != NULL)
1800 *ctlp = ctl;
1801 return true;
1802}
1803
1804/*!\brief initialize controlling instance
1805 */
1806static int
1807vtw_control_init(int af)
1808{
1809 fatp_ctl_t *fat;
1810 vtw_ctl_t *ctl;
1811 fatp_t *fat_base;
1812 fatp_t **fat_hash;
1813 vtw_t *ctl_base_v;
1814 uint32_t n, m;
1815 size_t sz;
1816
1817 KASSERT(powerof2(tcp_vtw_entries));
1818
1819 if (!vtw_select(af, &fat, &ctl))
1820 return EAFNOSUPPORT;
1821
1822 if (fat->hash != NULL) {
1823 KASSERT(fat->base != NULL && ctl->base.v != NULL);
1824 return 0;
1825 }
1826
1827 /* Allocate 10% more capacity in the fat pointers.
1828 * We should only need ~#hash additional based on
1829 * how they age, but TIME_WAIT assassination could cause
1830 * sparse fat pointer utilisation.
1831 */
1832 m = 512;
1833 n = 2*m + (11 * (tcp_vtw_entries / fatp_ntags())) / 10;
1834 sz = (ctl->is_v4 ? sizeof(vtw_v4_t) : sizeof(vtw_v6_t));
1835
1836 fat_hash = kmem_zalloc(2*m * sizeof(fatp_t *), KM_NOSLEEP);
1837
1838 if (fat_hash == NULL) {
1839 printf("%s: could not allocate %zu bytes for "
1840 "hash anchors", __func__, 2*m * sizeof(fatp_t *));
1841 return ENOMEM;
1842 }
1843
1844 fat_base = kmem_zalloc(2*n * sizeof(fatp_t), KM_NOSLEEP);
1845
1846 if (fat_base == NULL) {
1847 kmem_free(fat_hash, 2*m * sizeof (fatp_t *));
1848 printf("%s: could not allocate %zu bytes for "
1849 "fatp_t array", __func__, 2*n * sizeof(fatp_t));
1850 return ENOMEM;
1851 }
1852
1853 ctl_base_v = kmem_zalloc(tcp_vtw_entries * sz, KM_NOSLEEP);
1854
1855 if (ctl_base_v == NULL) {
1856 kmem_free(fat_hash, 2*m * sizeof (fatp_t *));
1857 kmem_free(fat_base, 2*n * sizeof(fatp_t));
1858 printf("%s: could not allocate %zu bytes for "
1859 "vtw_t array", __func__, tcp_vtw_entries * sz);
1860 return ENOMEM;
1861 }
1862
1863 fatp_init(fat, n, m, fat_base, fat_hash);
1864
1865 vtw_init(fat, ctl, tcp_vtw_entries, ctl_base_v);
1866
1867 return 0;
1868}
1869
1870/*!\brief select controlling instance
1871 */
1872static vtw_ctl_t *
1873vtw_control(int af, uint32_t msl)
1874{
1875 fatp_ctl_t *fat;
1876 vtw_ctl_t *ctl;
1877 int msl_class = msl_to_class(msl);
1878
1879 if (!vtw_select(af, &fat, &ctl))
1880 return NULL;
1881
1882 if (!fat->base || !ctl->base.v)
1883 return NULL;
1884
1885 if (!tcp_vtw_was_enabled) {
1886 /* This guarantees is timer ticks until we no longer need them.
1887 */
1888 tcp_vtw_was_enabled = 1;
1889
1890 callout_schedule(&vtw_cs, hz / 5);
1891
1892 tcbtable.vestige = &tcp_hooks;
1893 }
1894
1895 return ctl + msl_class;
1896}
1897
1898/*!\brief add TCP pcb to vestigial timewait
1899 */
1900int
1901vtw_add(int af, struct tcpcb *tp)
1902{
1903#ifdef VTW_DEBUG
1904 int enable;
1905#endif
1906 vtw_ctl_t *ctl;
1907 vtw_t *vtw;
1908
1909 KASSERT(mutex_owned(softnet_lock));
1910
1911 ctl = vtw_control(af, tp->t_msl);
1912 if (!ctl)
1913 return 0;
1914
1915#ifdef VTW_DEBUG
1916 enable = (af == AF_INET) ? tcp4_vtw_enable : tcp6_vtw_enable;
1917#endif
1918
1919 vtw = vtw_alloc(ctl);
1920
1921 if (vtw) {
1922 vtw->snd_nxt = tp->snd_nxt;
1923 vtw->rcv_nxt = tp->rcv_nxt;
1924
1925 switch (af) {
1926 case AF_INET: {
1927 struct inpcb *inp = tp->t_inpcb;
1928 vtw_v4_t *v4 = (void*)vtw;
1929
1930 v4->faddr = inp->inp_faddr.s_addr;
1931 v4->laddr = inp->inp_laddr.s_addr;
1932 v4->fport = inp->inp_fport;
1933 v4->lport = inp->inp_lport;
1934
1935 vtw->reuse_port = !!(inp->inp_socket->so_options
1936 & SO_REUSEPORT);
1937 vtw->reuse_addr = !!(inp->inp_socket->so_options
1938 & SO_REUSEADDR);
1939 vtw->v6only = 0;
1940 vtw->uid = inp->inp_socket->so_uidinfo->ui_uid;
1941
1942 vtw_inshash_v4(ctl, vtw);
1943
1944
1945#ifdef VTW_DEBUG
1946 /* Immediate lookup (connected and port) to
1947 * ensure at least that works!
1948 */
1949 if (enable & 4) {
1950 KASSERT(vtw_lookup_hash_v4
1951 (ctl
1952 , inp->inp_faddr.s_addr, inp->inp_fport
1953 , inp->inp_laddr.s_addr, inp->inp_lport
1954 , 0)
1955 == vtw);
1956 KASSERT(vtw_lookup_hash_v4
1957 (ctl
1958 , inp->inp_faddr.s_addr, inp->inp_fport
1959 , inp->inp_laddr.s_addr, inp->inp_lport
1960 , 1));
1961 }
1962 /* Immediate port iterator functionality check: not wild
1963 */
1964 if (enable & 8) {
1965 struct tcp_ports_iterator *it;
1966 struct vestigial_inpcb res;
1967 int cnt = 0;
1968
1969 it = tcp_init_ports_v4(inp->inp_laddr
1970 , inp->inp_lport, 0);
1971
1972 while (tcp_next_port_v4(it, &res)) {
1973 ++cnt;
1974 }
1975 KASSERT(cnt);
1976 }
1977 /* Immediate port iterator functionality check: wild
1978 */
1979 if (enable & 16) {
1980 struct tcp_ports_iterator *it;
1981 struct vestigial_inpcb res;
1982 struct in_addr any;
1983 int cnt = 0;
1984
1985 any.s_addr = htonl(INADDR_ANY);
1986
1987 it = tcp_init_ports_v4(any, inp->inp_lport, 1);
1988
1989 while (tcp_next_port_v4(it, &res)) {
1990 ++cnt;
1991 }
1992 KASSERT(cnt);
1993 }
1994#endif /* VTW_DEBUG */
1995 break;
1996 }
1997
1998 case AF_INET6: {
1999 struct in6pcb *inp = tp->t_in6pcb;
2000 vtw_v6_t *v6 = (void*)vtw;
2001
2002 v6->faddr = inp->in6p_faddr;
2003 v6->laddr = inp->in6p_laddr;
2004 v6->fport = inp->in6p_fport;
2005 v6->lport = inp->in6p_lport;
2006
2007 vtw->reuse_port = !!(inp->in6p_socket->so_options
2008 & SO_REUSEPORT);
2009 vtw->reuse_addr = !!(inp->in6p_socket->so_options
2010 & SO_REUSEADDR);
2011 vtw->v6only = !!(inp->in6p_flags
2012 & IN6P_IPV6_V6ONLY);
2013 vtw->uid = inp->in6p_socket->so_uidinfo->ui_uid;
2014
2015 vtw_inshash_v6(ctl, vtw);
2016#ifdef VTW_DEBUG
2017 /* Immediate lookup (connected and port) to
2018 * ensure at least that works!
2019 */
2020 if (enable & 4) {
2021 KASSERT(vtw_lookup_hash_v6(ctl
2022 , &inp->in6p_faddr, inp->in6p_fport
2023 , &inp->in6p_laddr, inp->in6p_lport
2024 , 0)
2025 == vtw);
2026 KASSERT(vtw_lookup_hash_v6
2027 (ctl
2028 , &inp->in6p_faddr, inp->in6p_fport
2029 , &inp->in6p_laddr, inp->in6p_lport
2030 , 1));
2031 }
2032 /* Immediate port iterator functionality check: not wild
2033 */
2034 if (enable & 8) {
2035 struct tcp_ports_iterator *it;
2036 struct vestigial_inpcb res;
2037 int cnt = 0;
2038
2039 it = tcp_init_ports_v6(&inp->in6p_laddr
2040 , inp->in6p_lport, 0);
2041
2042 while (tcp_next_port_v6(it, &res)) {
2043 ++cnt;
2044 }
2045 KASSERT(cnt);
2046 }
2047 /* Immediate port iterator functionality check: wild
2048 */
2049 if (enable & 16) {
2050 struct tcp_ports_iterator *it;
2051 struct vestigial_inpcb res;
2052 static struct in6_addr any = IN6ADDR_ANY_INIT;
2053 int cnt = 0;
2054
2055 it = tcp_init_ports_v6(&any
2056 , inp->in6p_lport, 1);
2057
2058 while (tcp_next_port_v6(it, &res)) {
2059 ++cnt;
2060 }
2061 KASSERT(cnt);
2062 }
2063#endif /* VTW_DEBUG */
2064 break;
2065 }
2066 }
2067
2068 tcp_canceltimers(tp);
2069 tp = tcp_close(tp);
2070 KASSERT(!tp);
2071
2072 return 1;
2073 }
2074
2075 return 0;
2076}
2077
2078/*!\brief restart timer for vestigial time-wait entry
2079 */
2080static void
2081vtw_restart_v4(vestigial_inpcb_t *vp)
2082{
2083 vtw_v4_t copy = *(vtw_v4_t*)vp->vtw;
2084 vtw_t *vtw;
2085 vtw_t *cp = &copy.common;
2086 vtw_ctl_t *ctl;
2087
2088 KASSERT(mutex_owned(softnet_lock));
2089
2090 db_trace(KTR_VTW
2091 , (vp->vtw, "vtw: restart %A:%P %A:%P"
2092 , vp->faddr.v4.s_addr, vp->fport
2093 , vp->laddr.v4.s_addr, vp->lport));
2094
2095 /* Class might have changed, so have a squiz.
2096 */
2097 ctl = vtw_control(AF_INET, class_to_msl(cp->msl_class));
2098 vtw = vtw_alloc(ctl);
2099
2100 if (vtw) {
2101 vtw_v4_t *v4 = (void*)vtw;
2102
2103 /* Safe now to unhash the old entry
2104 */
2105 vtw_del(vp->ctl, vp->vtw);
2106
2107 vtw->snd_nxt = cp->snd_nxt;
2108 vtw->rcv_nxt = cp->rcv_nxt;
2109
2110 v4->faddr = copy.faddr;
2111 v4->laddr = copy.laddr;
2112 v4->fport = copy.fport;
2113 v4->lport = copy.lport;
2114
2115 vtw->reuse_port = cp->reuse_port;
2116 vtw->reuse_addr = cp->reuse_addr;
2117 vtw->v6only = 0;
2118 vtw->uid = cp->uid;
2119
2120 vtw_inshash_v4(ctl, vtw);
2121 }
2122
2123 vp->valid = 0;
2124}
2125
2126/*!\brief restart timer for vestigial time-wait entry
2127 */
2128static void
2129vtw_restart_v6(vestigial_inpcb_t *vp)
2130{
2131 vtw_v6_t copy = *(vtw_v6_t*)vp->vtw;
2132 vtw_t *vtw;
2133 vtw_t *cp = &copy.common;
2134 vtw_ctl_t *ctl;
2135
2136 KASSERT(mutex_owned(softnet_lock));
2137
2138 db_trace(KTR_VTW
2139 , (vp->vtw, "vtw: restart %6A:%P %6A:%P"
2140 , db_store(&vp->faddr.v6, sizeof (vp->faddr.v6))
2141 , vp->fport
2142 , db_store(&vp->laddr.v6, sizeof (vp->laddr.v6))
2143 , vp->lport));
2144
2145 /* Class might have changed, so have a squiz.
2146 */
2147 ctl = vtw_control(AF_INET6, class_to_msl(cp->msl_class));
2148 vtw = vtw_alloc(ctl);
2149
2150 if (vtw) {
2151 vtw_v6_t *v6 = (void*)vtw;
2152
2153 /* Safe now to unhash the old entry
2154 */
2155 vtw_del(vp->ctl, vp->vtw);
2156
2157 vtw->snd_nxt = cp->snd_nxt;
2158 vtw->rcv_nxt = cp->rcv_nxt;
2159
2160 v6->faddr = copy.faddr;
2161 v6->laddr = copy.laddr;
2162 v6->fport = copy.fport;
2163 v6->lport = copy.lport;
2164
2165 vtw->reuse_port = cp->reuse_port;
2166 vtw->reuse_addr = cp->reuse_addr;
2167 vtw->v6only = cp->v6only;
2168 vtw->uid = cp->uid;
2169
2170 vtw_inshash_v6(ctl, vtw);
2171 }
2172
2173 vp->valid = 0;
2174}
2175
2176/*!\brief restart timer for vestigial time-wait entry
2177 */
2178void
2179vtw_restart(vestigial_inpcb_t *vp)
2180{
2181 if (!vp || !vp->valid)
2182 return;
2183
2184 if (vp->v4)
2185 vtw_restart_v4(vp);
2186 else
2187 vtw_restart_v6(vp);
2188}
2189
2190int
2191sysctl_tcp_vtw_enable(SYSCTLFN_ARGS)
2192{
2193 int en, rc;
2194 struct sysctlnode node;
2195
2196 node = *rnode;
2197 en = *(int *)rnode->sysctl_data;
2198 node.sysctl_data = &en;
2199
2200 rc = sysctl_lookup(SYSCTLFN_CALL(&node));
2201 if (rc != 0 || newp == NULL)
2202 return rc;
2203
2204 if (rnode->sysctl_data != &tcp4_vtw_enable &&
2205 rnode->sysctl_data != &tcp6_vtw_enable)
2206 rc = ENOENT;
2207 else if ((en & 1) == 0)
2208 rc = 0;
2209 else if (rnode->sysctl_data == &tcp4_vtw_enable)
2210 rc = vtw_control_init(AF_INET);
2211 else /* rnode->sysctl_data == &tcp6_vtw_enable */
2212 rc = vtw_control_init(AF_INET6);
2213
2214 if (rc == 0)
2215 *(int *)rnode->sysctl_data = en;
2216
2217 return rc;
2218}
2219
2220int
2221vtw_earlyinit(void)
2222{
2223 int i, rc;
2224
2225 callout_init(&vtw_cs, 0);
2226 callout_setfunc(&vtw_cs, vtw_tick, 0);
2227
2228 for (i = 0; i < VTW_NCLASS; ++i) {
2229 vtw_tcpv4[i].is_v4 = 1;
2230 vtw_tcpv6[i].is_v6 = 1;
2231 }
2232
2233 if ((tcp4_vtw_enable & 1) != 0 &&
2234 (rc = vtw_control_init(AF_INET)) != 0)
2235 return rc;
2236
2237 if ((tcp6_vtw_enable & 1) != 0 &&
2238 (rc = vtw_control_init(AF_INET6)) != 0)
2239 return rc;
2240
2241 return 0;
2242}
2243
2244#ifdef VTW_DEBUG
2245#include <sys/syscallargs.h>
2246#include <sys/sysctl.h>
2247
2248/*!\brief add lalp, fafp entries for debug
2249 */
2250int
2251vtw_debug_add(int af, sin_either_t *la, sin_either_t *fa, int msl, int msl_class)
2252{
2253 vtw_ctl_t *ctl;
2254 vtw_t *vtw;
2255
2256 ctl = vtw_control(af, msl ? msl : class_to_msl(msl_class));
2257 if (!ctl)
2258 return 0;
2259
2260 vtw = vtw_alloc(ctl);
2261
2262 if (vtw) {
2263 vtw->snd_nxt = 0;
2264 vtw->rcv_nxt = 0;
2265
2266 switch (af) {
2267 case AF_INET: {
2268 vtw_v4_t *v4 = (void*)vtw;
2269
2270 v4->faddr = fa->sin_addr.v4.s_addr;
2271 v4->laddr = la->sin_addr.v4.s_addr;
2272 v4->fport = fa->sin_port;
2273 v4->lport = la->sin_port;
2274
2275 vtw->reuse_port = 1;
2276 vtw->reuse_addr = 1;
2277 vtw->v6only = 0;
2278 vtw->uid = 0;
2279
2280 vtw_inshash_v4(ctl, vtw);
2281 break;
2282 }
2283
2284 case AF_INET6: {
2285 vtw_v6_t *v6 = (void*)vtw;
2286
2287 v6->faddr = fa->sin_addr.v6;
2288 v6->laddr = la->sin_addr.v6;
2289
2290 v6->fport = fa->sin_port;
2291 v6->lport = la->sin_port;
2292
2293 vtw->reuse_port = 1;
2294 vtw->reuse_addr = 1;
2295 vtw->v6only = 0;
2296 vtw->uid = 0;
2297
2298 vtw_inshash_v6(ctl, vtw);
2299 break;
2300 }
2301
2302 default:
2303 break;
2304 }
2305
2306 return 1;
2307 }
2308
2309 return 0;
2310}
2311
2312static int vtw_syscall = 0;
2313
2314static int
2315vtw_debug_process(vtw_sysargs_t *ap)
2316{
2317 struct vestigial_inpcb vestige;
2318 int rc = 0;
2319
2320 mutex_enter(softnet_lock);
2321
2322 switch (ap->op) {
2323 case 0: // insert
2324 vtw_debug_add(ap->la.sin_family
2325 , &ap->la
2326 , &ap->fa
2327 , TCPTV_MSL
2328 , 0);
2329 break;
2330
2331 case 1: // lookup
2332 case 2: // restart
2333 switch (ap->la.sin_family) {
2334 case AF_INET:
2335 if (tcp_lookup_v4(ap->fa.sin_addr.v4, ap->fa.sin_port,
2336 ap->la.sin_addr.v4, ap->la.sin_port,
2337 &vestige)) {
2338 if (ap->op == 2) {
2339 vtw_restart(&vestige);
2340 }
2341 rc = 0;
2342 } else
2343 rc = ESRCH;
2344 break;
2345
2346 case AF_INET6:
2347 if (tcp_lookup_v6(&ap->fa.sin_addr.v6, ap->fa.sin_port,
2348 &ap->la.sin_addr.v6, ap->la.sin_port,
2349 &vestige)) {
2350 if (ap->op == 2) {
2351 vtw_restart(&vestige);
2352 }
2353 rc = 0;
2354 } else
2355 rc = ESRCH;
2356 break;
2357 default:
2358 rc = EINVAL;
2359 }
2360 break;
2361
2362 default:
2363 rc = EINVAL;
2364 }
2365
2366 mutex_exit(softnet_lock);
2367 return rc;
2368}
2369
2370struct sys_vtw_args {
2371 syscallarg(const vtw_sysargs_t *) req;
2372 syscallarg(size_t) len;
2373};
2374
2375static int
2376vtw_sys(struct lwp *l, const void *_, register_t *retval)
2377{
2378 const struct sys_vtw_args *uap = _;
2379 void *buf;
2380 int rc;
2381 size_t len = SCARG(uap, len);
2382
2383 if (len != sizeof (vtw_sysargs_t))
2384 return EINVAL;
2385
2386 buf = kmem_alloc(len, KM_SLEEP);
2387 if (!buf)
2388 return ENOMEM;
2389
2390 rc = copyin(SCARG(uap, req), buf, len);
2391 if (!rc) {
2392 rc = vtw_debug_process(buf);
2393 }
2394 kmem_free(buf, len);
2395
2396 return rc;
2397}
2398
2399static void
2400vtw_sanity_check(void)
2401{
2402 vtw_ctl_t *ctl;
2403 vtw_t *vtw;
2404 int i;
2405 int n;
2406
2407 for (i = 0; i < VTW_NCLASS; ++i) {
2408 ctl = &vtw_tcpv4[i];
2409
2410 if (!ctl->base.v || ctl->nalloc)
2411 continue;
2412
2413 for (n = 0, vtw = ctl->base.v; ; ) {
2414 ++n;
2415 vtw = vtw_next(ctl, vtw);
2416 if (vtw == ctl->base.v)
2417 break;
2418 }
2419 db_trace(KTR_VTW
2420 , (ctl, "sanity: class %x n %x nfree %x"
2421 , i, n, ctl->nfree));
2422
2423 KASSERT(n == ctl->nfree);
2424 }
2425
2426 for (i = 0; i < VTW_NCLASS; ++i) {
2427 ctl = &vtw_tcpv6[i];
2428
2429 if (!ctl->base.v || ctl->nalloc)
2430 continue;
2431
2432 for (n = 0, vtw = ctl->base.v; ; ) {
2433 ++n;
2434 vtw = vtw_next(ctl, vtw);
2435 if (vtw == ctl->base.v)
2436 break;
2437 }
2438 db_trace(KTR_VTW
2439 , (ctl, "sanity: class %x n %x nfree %x"
2440 , i, n, ctl->nfree));
2441 KASSERT(n == ctl->nfree);
2442 }
2443}
2444
2445/*!\brief Initialise debug support.
2446 */
2447static void
2448vtw_debug_init(void)
2449{
2450 int i;
2451
2452 vtw_sanity_check();
2453
2454 if (vtw_syscall)
2455 return;
2456
2457 for (i = 511; i; --i) {
2458 if (sysent[i].sy_call == sys_nosys) {
2459 sysent[i].sy_call = vtw_sys;
2460 sysent[i].sy_narg = 2;
2461 sysent[i].sy_argsize = sizeof (struct sys_vtw_args);
2462 sysent[i].sy_flags = 0;
2463
2464 vtw_syscall = i;
2465 break;
2466 }
2467 }
2468 if (i) {
2469 const struct sysctlnode *node;
2470 uint32_t flags;
2471
2472 flags = sysctl_root.sysctl_flags;
2473
2474 sysctl_root.sysctl_flags |= CTLFLAG_READWRITE;
2475 sysctl_root.sysctl_flags &= ~CTLFLAG_PERMANENT;
2476
2477 sysctl_createv(0, 0, 0, &node,
2478 CTLFLAG_PERMANENT, CTLTYPE_NODE,
2479 "koff",
2480 SYSCTL_DESCR("Kernel Obscure Feature Finder"),
2481 0, 0, 0, 0, CTL_CREATE, CTL_EOL);
2482
2483 if (!node) {
2484 sysctl_createv(0, 0, 0, &node,
2485 CTLFLAG_PERMANENT, CTLTYPE_NODE,
2486 "koffka",
2487 SYSCTL_DESCR("The Real(tm) Kernel"
2488 " Obscure Feature Finder"),
2489 0, 0, 0, 0, CTL_CREATE, CTL_EOL);
2490 }
2491 if (node) {
2492 sysctl_createv(0, 0, 0, 0,
2493 CTLFLAG_PERMANENT|CTLFLAG_READONLY,
2494 CTLTYPE_INT, "vtw_debug_syscall",
2495 SYSCTL_DESCR("vtw debug"
2496 " system call number"),
2497 0, 0, &vtw_syscall, 0, node->sysctl_num,
2498 CTL_CREATE, CTL_EOL);
2499 }
2500 sysctl_root.sysctl_flags = flags;
2501 }
2502}
2503#else /* !VTW_DEBUG */
2504static void
2505vtw_debug_init(void)
2506{
2507 return;
2508}
2509#endif /* !VTW_DEBUG */
2510