1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/protosw.h>
52 #include <sys/random.h>
53
54 #include <vm/uma.h>
55
56 #include <net/route.h>
57 #include <net/if.h>
58
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #ifdef INET6
63 #include <netinet/ip6.h>
64 #endif
65 #include <netinet/in_pcb.h>
66 #ifdef INET6
67 #include <netinet6/in6_pcb.h>
68 #endif
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #ifdef INET6
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #endif
76 #include <netinet/ip_icmp.h>
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_fsm.h>
79 #include <netinet/tcp_seq.h>
80 #include <netinet/tcp_timer.h>
81 #include <netinet/tcp_var.h>
82 #ifdef INET6
83 #include <netinet6/tcp6_var.h>
84 #endif
85 #include <netinet/tcpip.h>
86 #ifdef TCPDEBUG
87 #include <netinet/tcp_debug.h>
88 #endif
89 #include <netinet6/ip6protosw.h>
90
91 #include <machine/in_cksum.h>
92
93 #include <security/mac/mac_framework.h>
94
95 static uma_zone_t tcptw_zone;
96 static int maxtcptw;
97
98 /*
99 * The timed wait queue contains references to each of the TCP sessions
100 * currently in the TIME_WAIT state. The queue pointers, including the
101 * queue pointers in each tcptw structure, are protected using the global
102 * tcbinfo lock, which must be held over queue iteration and modification.
103 */
104 static TAILQ_HEAD(, tcptw) twq_2msl;
105
106 static void tcp_tw_2msl_reset(struct tcptw *, int);
107 static void tcp_tw_2msl_stop(struct tcptw *);
108
109 static int
110 tcptw_auto_size(void)
111 {
112 int halfrange;
113
114 /*
115 * Max out at half the ephemeral port range so that TIME_WAIT
116 * sockets don't tie up too many ephemeral ports.
117 */
118 if (ipport_lastauto > ipport_firstauto)
119 halfrange = (ipport_lastauto - ipport_firstauto) / 2;
120 else
121 halfrange = (ipport_firstauto - ipport_lastauto) / 2;
122 /* Protect against goofy port ranges smaller than 32. */
123 return (imin(imax(halfrange, 32), maxsockets / 5));
124 }
125
126 static int
127 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
128 {
129 int error, new;
130
131 if (maxtcptw == 0)
132 new = tcptw_auto_size();
133 else
134 new = maxtcptw;
135 error = sysctl_handle_int(oidp, &new, 0, req);
136 if (error == 0 && req->newptr)
137 if (new >= 32) {
138 maxtcptw = new;
139 uma_zone_set_max(tcptw_zone, maxtcptw);
140 }
141 return (error);
142 }
143
144 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
145 &maxtcptw, 0, sysctl_maxtcptw, "IU",
146 "Maximum number of compressed TCP TIME_WAIT entries");
147
148 static int nolocaltimewait = 0;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
150 &nolocaltimewait, 0,
151 "Do not create compressed TCP TIME_WAIT entries for local connections");
152
153 void
154 tcp_tw_zone_change(void)
155 {
156
157 if (maxtcptw == 0)
158 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
159 }
160
161 void
162 tcp_tw_init(void)
163 {
164
165 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
166 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
167 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
168 if (maxtcptw == 0)
169 uma_zone_set_max(tcptw_zone, tcptw_auto_size());
170 else
171 uma_zone_set_max(tcptw_zone, maxtcptw);
172 TAILQ_INIT(&twq_2msl);
173 }
174
175 /*
176 * Move a TCP connection into TIME_WAIT state.
177 * tcbinfo is locked.
178 * inp is locked, and is unlocked before returning.
179 */
180 void
181 tcp_twstart(struct tcpcb *tp)
182 {
183 struct tcptw *tw;
184 struct inpcb *inp = tp->t_inpcb;
185 int acknow;
186 struct socket *so;
187
188 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_tw_2msl_reset(). */
189 INP_WLOCK_ASSERT(inp);
190
191 if (nolocaltimewait && in_localip(inp->inp_faddr)) {
192 tp = tcp_close(tp);
193 if (tp != NULL)
194 INP_WUNLOCK(inp);
195 return;
196 }
197
198 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
199 if (tw == NULL) {
200 tw = tcp_tw_2msl_scan(1);
201 if (tw == NULL) {
202 tp = tcp_close(tp);
203 if (tp != NULL)
204 INP_WUNLOCK(inp);
205 return;
206 }
207 }
208 tw->tw_inpcb = inp;
209
210 /*
211 * Recover last window size sent.
212 */
213 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt),
214 ("tcp_twstart negative window: tp %p rcv_nxt %u rcv_adv %u", tp,
215 tp->rcv_adv, tp->rcv_nxt));
216 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
217
218 /*
219 * Set t_recent if timestamps are used on the connection.
220 */
221 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
222 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
223 tw->t_recent = tp->ts_recent;
224 tw->ts_offset = tp->ts_offset;
225 } else {
226 tw->t_recent = 0;
227 tw->ts_offset = 0;
228 }
229
230 tw->snd_nxt = tp->snd_nxt;
231 tw->rcv_nxt = tp->rcv_nxt;
232 tw->iss = tp->iss;
233 tw->irs = tp->irs;
234 tw->t_starttime = tp->t_starttime;
235 tw->tw_time = 0;
236
237 /* XXX
238 * If this code will
239 * be used for fin-wait-2 state also, then we may need
240 * a ts_recent from the last segment.
241 */
242 acknow = tp->t_flags & TF_ACKNOW;
243
244 /*
245 * First, discard tcpcb state, which includes stopping its timers and
246 * freeing it. tcp_discardcb() used to also release the inpcb, but
247 * that work is now done in the caller.
248 *
249 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
250 * and might not be needed here any longer.
251 */
252 tcp_discardcb(tp);
253 so = inp->inp_socket;
254 soisdisconnected(so);
255 tw->tw_cred = crhold(so->so_cred);
256 SOCK_LOCK(so);
257 tw->tw_so_options = so->so_options;
258 SOCK_UNLOCK(so);
259 if (acknow)
260 tcp_twrespond(tw, TH_ACK);
261 inp->inp_ppcb = tw;
262 inp->inp_flags |= INP_TIMEWAIT;
263 tcp_tw_2msl_reset(tw, 0);
264
265 /*
266 * If the inpcb owns the sole reference to the socket, then we can
267 * detach and free the socket as it is not needed in time wait.
268 */
269 if (inp->inp_flags & INP_SOCKREF) {
270 KASSERT(so->so_state & SS_PROTOREF,
271 ("tcp_twstart: !SS_PROTOREF"));
272 inp->inp_flags &= ~INP_SOCKREF;
273 INP_WUNLOCK(inp);
274 ACCEPT_LOCK();
275 SOCK_LOCK(so);
276 so->so_state &= ~SS_PROTOREF;
277 sofree(so);
278 } else
279 INP_WUNLOCK(inp);
280 }
281
282 #if 0
283 /*
284 * The appromixate rate of ISN increase of Microsoft TCP stacks;
285 * the actual rate is slightly higher due to the addition of
286 * random positive increments.
287 *
288 * Most other new OSes use semi-randomized ISN values, so we
289 * do not need to worry about them.
290 */
291 #define MS_ISN_BYTES_PER_SECOND 250000
292
293 /*
294 * Determine if the ISN we will generate has advanced beyond the last
295 * sequence number used by the previous connection. If so, indicate
296 * that it is safe to recycle this tw socket by returning 1.
297 */
298 int
299 tcp_twrecycleable(struct tcptw *tw)
300 {
301 tcp_seq new_iss = tw->iss;
302 tcp_seq new_irs = tw->irs;
303
304 INP_INFO_WLOCK_ASSERT(&tcbinfo);
305 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
306 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
307
308 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
309 return (1);
310 else
311 return (0);
312 }
313 #endif
314
315 /*
316 * Returns 1 if the TIME_WAIT state was killed and we should start over,
317 * looking for a pcb in the listen state. Returns 0 otherwise.
318 */
319 int
320 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
321 struct mbuf *m, int tlen)
322 {
323 struct tcptw *tw;
324 int thflags;
325 tcp_seq seq;
326
327 /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */
328 INP_INFO_WLOCK_ASSERT(&tcbinfo);
329 INP_WLOCK_ASSERT(inp);
330
331 /*
332 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
333 * still present. This is undesirable, but temporarily necessary
334 * until we work out how to handle inpcb's who's timewait state has
335 * been removed.
336 */
337 tw = intotw(inp);
338 if (tw == NULL)
339 goto drop;
340
341 thflags = th->th_flags;
342
343 /*
344 * NOTE: for FIN_WAIT_2 (to be added later),
345 * must validate sequence number before accepting RST
346 */
347
348 /*
349 * If the segment contains RST:
350 * Drop the segment - see Stevens, vol. 2, p. 964 and
351 * RFC 1337.
352 */
353 if (thflags & TH_RST)
354 goto drop;
355
356 #if 0
357 /* PAWS not needed at the moment */
358 /*
359 * RFC 1323 PAWS: If we have a timestamp reply on this segment
360 * and it's less than ts_recent, drop it.
361 */
362 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
363 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
364 if ((thflags & TH_ACK) == 0)
365 goto drop;
366 goto ack;
367 }
368 /*
369 * ts_recent is never updated because we never accept new segments.
370 */
371 #endif
372
373 /*
374 * If a new connection request is received
375 * while in TIME_WAIT, drop the old connection
376 * and start over if the sequence numbers
377 * are above the previous ones.
378 */
379 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
380 tcp_twclose(tw, 0);
381 return (1);
382 }
383
384 /*
385 * Drop the segment if it does not contain an ACK.
386 */
387 if ((thflags & TH_ACK) == 0)
388 goto drop;
389
390 /*
391 * Reset the 2MSL timer if this is a duplicate FIN.
392 */
393 if (thflags & TH_FIN) {
394 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
395 if (seq + 1 == tw->rcv_nxt)
396 tcp_tw_2msl_reset(tw, 1);
397 }
398
399 /*
400 * Acknowledge the segment if it has data or is not a duplicate ACK.
401 */
402 if (thflags != TH_ACK || tlen != 0 ||
403 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
404 tcp_twrespond(tw, TH_ACK);
405 drop:
406 INP_WUNLOCK(inp);
407 m_freem(m);
408 return (0);
409 }
410
411 void
412 tcp_twclose(struct tcptw *tw, int reuse)
413 {
414 struct socket *so;
415 struct inpcb *inp;
416
417 /*
418 * At this point, we are in one of two situations:
419 *
420 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
421 * all state.
422 *
423 * (2) We have a socket -- if we own a reference, release it and
424 * notify the socket layer.
425 */
426 inp = tw->tw_inpcb;
427 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
428 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
429 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_tw_2msl_stop(). */
430 INP_WLOCK_ASSERT(inp);
431
432 tw->tw_inpcb = NULL;
433 tcp_tw_2msl_stop(tw);
434 inp->inp_ppcb = NULL;
435 in_pcbdrop(inp);
436
437 so = inp->inp_socket;
438 if (so != NULL) {
439 /*
440 * If there's a socket, handle two cases: first, we own a
441 * strong reference, which we will now release, or we don't
442 * in which case another reference exists (XXXRW: think
443 * about this more), and we don't need to take action.
444 */
445 if (inp->inp_flags & INP_SOCKREF) {
446 inp->inp_flags &= ~INP_SOCKREF;
447 INP_WUNLOCK(inp);
448 ACCEPT_LOCK();
449 SOCK_LOCK(so);
450 KASSERT(so->so_state & SS_PROTOREF,
451 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
452 so->so_state &= ~SS_PROTOREF;
453 sofree(so);
454 } else {
455 /*
456 * If we don't own the only reference, the socket and
457 * inpcb need to be left around to be handled by
458 * tcp_usr_detach() later.
459 */
460 INP_WUNLOCK(inp);
461 }
462 } else
463 in_pcbfree(inp);
464 tcpstat.tcps_closed++;
465 crfree(tw->tw_cred);
466 tw->tw_cred = NULL;
467 if (reuse)
468 return;
469 uma_zfree(tcptw_zone, tw);
470 }
471
472 int
473 tcp_twrespond(struct tcptw *tw, int flags)
474 {
475 struct inpcb *inp = tw->tw_inpcb;
476 struct tcphdr *th;
477 struct mbuf *m;
478 struct ip *ip = NULL;
479 u_int hdrlen, optlen;
480 int error;
481 struct tcpopt to;
482 #ifdef INET6
483 struct ip6_hdr *ip6 = NULL;
484 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
485 #endif
486
487 INP_WLOCK_ASSERT(inp);
488
489 m = m_gethdr(M_DONTWAIT, MT_DATA);
490 if (m == NULL)
491 return (ENOBUFS);
492 m->m_data += max_linkhdr;
493
494 #ifdef MAC
495 mac_create_mbuf_from_inpcb(inp, m);
496 #endif
497
498 #ifdef INET6
499 if (isipv6) {
500 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
501 ip6 = mtod(m, struct ip6_hdr *);
502 th = (struct tcphdr *)(ip6 + 1);
503 tcpip_fillheaders(inp, ip6, th);
504 } else
505 #endif
506 {
507 hdrlen = sizeof(struct tcpiphdr);
508 ip = mtod(m, struct ip *);
509 th = (struct tcphdr *)(ip + 1);
510 tcpip_fillheaders(inp, ip, th);
511 }
512 to.to_flags = 0;
513
514 /*
515 * Send a timestamp and echo-reply if both our side and our peer
516 * have sent timestamps in our SYN's and this is not a RST.
517 */
518 if (tw->t_recent && flags == TH_ACK) {
519 to.to_flags |= TOF_TS;
520 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
521 to.to_tsecr = tw->t_recent;
522 }
523 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
524
525 m->m_len = hdrlen + optlen;
526 m->m_pkthdr.len = m->m_len;
527
528 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
529
530 th->th_seq = htonl(tw->snd_nxt);
531 th->th_ack = htonl(tw->rcv_nxt);
532 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
533 th->th_flags = flags;
534 th->th_win = htons(tw->last_win);
535
536 #ifdef INET6
537 if (isipv6) {
538 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
539 sizeof(struct tcphdr) + optlen);
540 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
541 error = ip6_output(m, inp->in6p_outputopts, NULL,
542 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
543 } else
544 #endif
545 {
546 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
547 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
548 m->m_pkthdr.csum_flags = CSUM_TCP;
549 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
550 ip->ip_len = m->m_pkthdr.len;
551 if (path_mtu_discovery)
552 ip->ip_off |= IP_DF;
553 error = ip_output(m, inp->inp_options, NULL,
554 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
555 NULL, inp);
556 }
557 if (flags & TH_ACK)
558 tcpstat.tcps_sndacks++;
559 else
560 tcpstat.tcps_sndctrl++;
561 tcpstat.tcps_sndtotal++;
562 return (error);
563 }
564
565 static void
566 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
567 {
568
569 INP_INFO_WLOCK_ASSERT(&tcbinfo);
570 INP_WLOCK_ASSERT(tw->tw_inpcb);
571 if (rearm)
572 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
573 tw->tw_time = ticks + 2 * tcp_msl;
574 TAILQ_INSERT_TAIL(&twq_2msl, tw, tw_2msl);
575 }
576
577 static void
578 tcp_tw_2msl_stop(struct tcptw *tw)
579 {
580
581 INP_INFO_WLOCK_ASSERT(&tcbinfo);
582 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
583 }
584
585 struct tcptw *
586 tcp_tw_2msl_scan(int reuse)
587 {
588 struct tcptw *tw;
589
590 INP_INFO_WLOCK_ASSERT(&tcbinfo);
591 for (;;) {
592 tw = TAILQ_FIRST(&twq_2msl);
593 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0))
594 break;
595 INP_WLOCK(tw->tw_inpcb);
596 tcp_twclose(tw, reuse);
597 if (reuse)
598 return (tw);
599 }
600 return (NULL);
601 }
Cache object: 7247b8d7ef2754fbc861cad450d7a819
|