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 | |
130 | static void vtw_debug_init(void); |
131 | |
132 | fatp_ctl_t fat_tcpv4; |
133 | fatp_ctl_t fat_tcpv6; |
134 | vtw_ctl_t vtw_tcpv4[VTW_NCLASS]; |
135 | vtw_ctl_t vtw_tcpv6[VTW_NCLASS]; |
136 | vtw_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 | */ |
147 | struct 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 | |
163 | static struct tcp_ports_iterator tcp_ports_iterator_v4; |
164 | static struct tcp_ports_iterator tcp_ports_iterator_v6; |
165 | |
166 | static int vtw_age(vtw_ctl_t *, struct timeval *); |
167 | |
168 | /*!\brief allocate a fat pointer from a collection. |
169 | */ |
170 | static fatp_t * |
171 | fatp_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 | */ |
192 | static void |
193 | fatp_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 | */ |
214 | static void |
215 | fatp_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 | */ |
242 | static 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 | */ |
258 | static inline uint32_t |
259 | fatp_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 | |
278 | static inline uint32_t |
279 | fatp_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 | |
298 | static inline fatp_t * |
299 | fatp_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 | |
323 | static inline uint32_t |
324 | idx_encode(vtw_ctl_t *ctl, uint32_t idx) |
325 | { |
326 | return (idx << ctl->idx_bits) | idx; |
327 | } |
328 | |
329 | static inline uint32_t |
330 | idx_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 | |
353 | static inline uint32_t |
354 | fatp_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 | |
410 | static inline int |
411 | vtw_alive(const vtw_t *vtw) |
412 | { |
413 | return vtw->hashed && vtw->expire.tv_sec; |
414 | } |
415 | |
416 | static inline uint32_t |
417 | vtw_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 | |
427 | static inline uint32_t |
428 | vtw_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 | |
438 | static inline uint32_t |
439 | vtw_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 | |
455 | static inline vtw_t * |
456 | vtw_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 | */ |
491 | static inline vtw_t * |
492 | vtw_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 | */ |
512 | static inline void |
513 | vtw_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 | */ |
674 | void |
675 | vtw_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 | */ |
703 | static void |
704 | vtw_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 | */ |
742 | static void |
743 | vtw_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 | |
771 | static vtw_t * |
772 | vtw_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; |
913 | out: |
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 | |
924 | static vtw_t * |
925 | vtw_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; |
1010 | out: |
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 | */ |
1023 | static vtw_t * |
1024 | vtw_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; |
1124 | out: |
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 | */ |
1137 | static vtw_t * |
1138 | vtw_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; |
1252 | out: |
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 | */ |
1278 | static void |
1279 | vtw_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 | */ |
1343 | static inline uint32_t |
1344 | class_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 | */ |
1359 | static inline uint32_t |
1360 | msl_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 | */ |
1374 | static inline vtw_t * |
1375 | vtw_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 | */ |
1483 | static int |
1484 | vtw_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 | |
1538 | static 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 | */ |
1543 | static void |
1544 | vtw_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 | */ |
1574 | static void * |
1575 | tcp_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 | */ |
1600 | static int |
1601 | vtw_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 | */ |
1635 | static int |
1636 | tcp_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 | |
1650 | static int |
1651 | tcp_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 | */ |
1673 | static void * |
1674 | tcp_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 | */ |
1699 | static int |
1700 | vtw_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 | |
1735 | static int |
1736 | tcp_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 | |
1750 | static int |
1751 | tcp_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 | |
1770 | static 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 | |
1779 | static bool |
1780 | vtw_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 | */ |
1806 | static int |
1807 | vtw_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 | */ |
1872 | static vtw_ctl_t * |
1873 | vtw_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 | */ |
1900 | int |
1901 | vtw_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 | */ |
2080 | static void |
2081 | vtw_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 = ©.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 | */ |
2128 | static void |
2129 | vtw_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 = ©.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 | */ |
2178 | void |
2179 | vtw_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 | |
2190 | int |
2191 | sysctl_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 | |
2220 | int |
2221 | vtw_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 | */ |
2250 | int |
2251 | vtw_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 | |
2312 | static int vtw_syscall = 0; |
2313 | |
2314 | static int |
2315 | vtw_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 | |
2370 | struct sys_vtw_args { |
2371 | syscallarg(const vtw_sysargs_t *) req; |
2372 | syscallarg(size_t) len; |
2373 | }; |
2374 | |
2375 | static int |
2376 | vtw_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 | |
2399 | static void |
2400 | vtw_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 | */ |
2447 | static void |
2448 | vtw_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 */ |
2504 | static void |
2505 | vtw_debug_init(void) |
2506 | { |
2507 | return; |
2508 | } |
2509 | #endif /* !VTW_DEBUG */ |
2510 | |