1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
32 */
33
34 #include <sys/cdefs.h>
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_ipsec.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/syslog.h>
52 #include <sys/protosw.h>
53 #include <sys/random.h>
54
55 #include <vm/uma.h>
56
57 #include <net/route.h>
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/vnet.h>
61
62 #include <netinet/in.h>
63 #include <netinet/in_kdtrace.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_icmp.h>
69 #include <netinet/ip_var.h>
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #include <netinet6/in6_pcb.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #endif
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 #ifdef INET6
90 #include <netinet6/ip6protosw.h>
91 #endif
92
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95
96 #include <netipsec/ipsec_support.h>
97
98 #include <machine/in_cksum.h>
99
100 #include <security/mac/mac_framework.h>
101
102 VNET_DEFINE_STATIC(uma_zone_t, tcptw_zone);
103 #define V_tcptw_zone VNET(tcptw_zone)
104 static int maxtcptw;
105
106 /*
107 * The timed wait queue contains references to each of the TCP sessions
108 * currently in the TIME_WAIT state. The queue pointers, including the
109 * queue pointers in each tcptw structure, are protected using the global
110 * timewait lock, which must be held over queue iteration and modification.
111 *
112 * Rules on tcptw usage:
113 * - a inpcb is always freed _after_ its tcptw
114 * - a tcptw relies on its inpcb reference counting for memory stability
115 * - a tcptw is dereferenceable only while its inpcb is locked
116 */
117 VNET_DEFINE_STATIC(TAILQ_HEAD(, tcptw), twq_2msl);
118 #define V_twq_2msl VNET(twq_2msl)
119
120 /* Global timewait lock */
121 VNET_DEFINE_STATIC(struct rwlock, tw_lock);
122 #define V_tw_lock VNET(tw_lock)
123
124 #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0)
125 #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw))
126 #define TW_RLOCK(tw) rw_rlock(&(tw))
127 #define TW_WLOCK(tw) rw_wlock(&(tw))
128 #define TW_RUNLOCK(tw) rw_runlock(&(tw))
129 #define TW_WUNLOCK(tw) rw_wunlock(&(tw))
130 #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED)
131 #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED)
132 #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED)
133 #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED)
134
135 static void tcp_tw_2msl_reset(struct tcptw *, int);
136 static void tcp_tw_2msl_stop(struct tcptw *, int);
137 static int tcp_twrespond(struct tcptw *, int);
138
139 static int
tcptw_auto_size(void)140 tcptw_auto_size(void)
141 {
142 int halfrange;
143
144 /*
145 * Max out at half the ephemeral port range so that TIME_WAIT
146 * sockets don't tie up too many ephemeral ports.
147 */
148 if (V_ipport_lastauto > V_ipport_firstauto)
149 halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
150 else
151 halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
152 /* Protect against goofy port ranges smaller than 32. */
153 return (imin(imax(halfrange, 32), maxsockets / 5));
154 }
155
156 static int
sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)157 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
158 {
159 int error, new;
160
161 if (maxtcptw == 0)
162 new = tcptw_auto_size();
163 else
164 new = maxtcptw;
165 error = sysctl_handle_int(oidp, &new, 0, req);
166 if (error == 0 && req->newptr)
167 if (new >= 32) {
168 maxtcptw = new;
169 uma_zone_set_max(V_tcptw_zone, maxtcptw);
170 }
171 return (error);
172 }
173
174 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw,
175 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
176 &maxtcptw, 0, sysctl_maxtcptw, "IU",
177 "Maximum number of compressed TCP TIME_WAIT entries");
178
179 VNET_DEFINE_STATIC(int, nolocaltimewait) = 0;
180 #define V_nolocaltimewait VNET(nolocaltimewait)
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW,
182 &VNET_NAME(nolocaltimewait), 0,
183 "Do not create compressed TCP TIME_WAIT entries for local connections");
184
185 void
tcp_tw_zone_change(void)186 tcp_tw_zone_change(void)
187 {
188
189 if (maxtcptw == 0)
190 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
191 }
192
193 void
tcp_tw_init(void)194 tcp_tw_init(void)
195 {
196
197 V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
199 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
200 if (maxtcptw == 0)
201 uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
202 else
203 uma_zone_set_max(V_tcptw_zone, maxtcptw);
204 TAILQ_INIT(&V_twq_2msl);
205 TW_LOCK_INIT(V_tw_lock, "tcptw");
206 }
207
208 #ifdef VIMAGE
209 void
tcp_tw_destroy(void)210 tcp_tw_destroy(void)
211 {
212 struct tcptw *tw;
213 struct epoch_tracker et;
214
215 NET_EPOCH_ENTER(et);
216 while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
217 tcp_twclose(tw, 0);
218 NET_EPOCH_EXIT(et);
219
220 TW_LOCK_DESTROY(V_tw_lock);
221 uma_zdestroy(V_tcptw_zone);
222 }
223 #endif
224
225 /*
226 * Move a TCP connection into TIME_WAIT state.
227 * tcbinfo is locked.
228 * inp is locked, and is unlocked before returning.
229 */
230 void
tcp_twstart(struct tcpcb * tp)231 tcp_twstart(struct tcpcb *tp)
232 {
233 struct tcptw twlocal, *tw;
234 struct inpcb *inp = tp->t_inpcb;
235 struct socket *so;
236 uint32_t recwin;
237 bool acknow, local;
238 #ifdef INET6
239 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
240 #endif
241
242 NET_EPOCH_ASSERT();
243 INP_WLOCK_ASSERT(inp);
244
245 /* A dropped inp should never transition to TIME_WAIT state. */
246 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
247 "(inp->inp_flags & INP_DROPPED) != 0"));
248
249 if (V_nolocaltimewait) {
250 #ifdef INET6
251 if (isipv6)
252 local = in6_localaddr(&inp->in6p_faddr);
253 else
254 #endif
255 #ifdef INET
256 local = in_localip(inp->inp_faddr);
257 #else
258 local = false;
259 #endif
260 } else
261 local = false;
262
263 /*
264 * For use only by DTrace. We do not reference the state
265 * after this point so modifying it in place is not a problem.
266 */
267 tcp_state_change(tp, TCPS_TIME_WAIT);
268
269 if (local)
270 tw = &twlocal;
271 else
272 tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
273 if (tw == NULL) {
274 /*
275 * Reached limit on total number of TIMEWAIT connections
276 * allowed. Remove a connection from TIMEWAIT queue in LRU
277 * fashion to make room for this connection.
278 *
279 * XXX: Check if it possible to always have enough room
280 * in advance based on guarantees provided by uma_zalloc().
281 */
282 tw = tcp_tw_2msl_scan(1);
283 if (tw == NULL) {
284 tp = tcp_close(tp);
285 if (tp != NULL)
286 INP_WUNLOCK(inp);
287 return;
288 }
289 }
290 /*
291 * For !local case the tcptw will hold a reference on its inpcb
292 * until tcp_twclose is called.
293 */
294 tw->tw_inpcb = inp;
295
296 /*
297 * Recover last window size sent.
298 */
299 so = inp->inp_socket;
300 recwin = lmin(lmax(sbspace(&so->so_rcv), 0),
301 (long)TCP_MAXWIN << tp->rcv_scale);
302 if (recwin < (so->so_rcv.sb_hiwat / 4) &&
303 recwin < tp->t_maxseg)
304 recwin = 0;
305 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
306 recwin < (tp->rcv_adv - tp->rcv_nxt))
307 recwin = (tp->rcv_adv - tp->rcv_nxt);
308 tw->last_win = (u_short)(recwin >> tp->rcv_scale);
309
310 /*
311 * Set t_recent if timestamps are used on the connection.
312 */
313 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
314 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
315 tw->t_recent = tp->ts_recent;
316 tw->ts_offset = tp->ts_offset;
317 } else {
318 tw->t_recent = 0;
319 tw->ts_offset = 0;
320 }
321
322 tw->snd_nxt = tp->snd_nxt;
323 tw->t_port = tp->t_port;
324 tw->rcv_nxt = tp->rcv_nxt;
325 tw->iss = tp->iss;
326 tw->irs = tp->irs;
327 tw->t_starttime = tp->t_starttime;
328 tw->tw_time = 0;
329 tw->tw_flags = tp->t_flags;
330
331 /* XXX
332 * If this code will
333 * be used for fin-wait-2 state also, then we may need
334 * a ts_recent from the last segment.
335 */
336 acknow = tp->t_flags & TF_ACKNOW;
337
338 /*
339 * First, discard tcpcb state, which includes stopping its timers and
340 * freeing it. tcp_discardcb() used to also release the inpcb, but
341 * that work is now done in the caller.
342 *
343 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
344 * and might not be needed here any longer.
345 */
346 tcp_discardcb(tp);
347 soisdisconnected(so);
348 tw->tw_so_options = so->so_options;
349 inp->inp_flags |= INP_TIMEWAIT;
350 if (acknow)
351 tcp_twrespond(tw, TH_ACK);
352 if (local)
353 in_pcbdrop(inp);
354 else {
355 in_pcbref(inp); /* Reference from tw */
356 tw->tw_cred = crhold(so->so_cred);
357 inp->inp_ppcb = tw;
358 TCPSTATES_INC(TCPS_TIME_WAIT);
359 tcp_tw_2msl_reset(tw, 0);
360 }
361
362 /*
363 * If the inpcb owns the sole reference to the socket, then we can
364 * detach and free the socket as it is not needed in time wait.
365 */
366 if (inp->inp_flags & INP_SOCKREF) {
367 KASSERT(so->so_state & SS_PROTOREF,
368 ("tcp_twstart: !SS_PROTOREF"));
369 inp->inp_flags &= ~INP_SOCKREF;
370 INP_WUNLOCK(inp);
371 SOCK_LOCK(so);
372 so->so_state &= ~SS_PROTOREF;
373 sofree(so);
374 } else
375 INP_WUNLOCK(inp);
376 }
377
378 /*
379 * Returns 1 if the TIME_WAIT state was killed and we should start over,
380 * looking for a pcb in the listen state. Returns 0 otherwise.
381 * It be called with to == NULL only for pure SYN-segments.
382 */
383 int
tcp_twcheck(struct inpcb * inp,struct tcpopt * to,struct tcphdr * th,struct mbuf * m,int tlen)384 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
385 struct mbuf *m, int tlen)
386 {
387 struct tcptw *tw;
388 char *s;
389 int thflags;
390 tcp_seq seq;
391
392 NET_EPOCH_ASSERT();
393 INP_WLOCK_ASSERT(inp);
394
395 /*
396 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
397 * still present. This is undesirable, but temporarily necessary
398 * until we work out how to handle inpcb's who's timewait state has
399 * been removed.
400 */
401 tw = intotw(inp);
402 if (tw == NULL)
403 goto drop;
404
405 thflags = th->th_flags;
406 KASSERT(to != NULL || (thflags & (TH_SYN | TH_ACK)) == TH_SYN,
407 ("tcp_twcheck: called without options on a non-SYN segment"));
408
409 /*
410 * NOTE: for FIN_WAIT_2 (to be added later),
411 * must validate sequence number before accepting RST
412 */
413
414 /*
415 * If the segment contains RST:
416 * Drop the segment - see Stevens, vol. 2, p. 964 and
417 * RFC 1337.
418 */
419 if (thflags & TH_RST)
420 goto drop;
421
422 #if 0
423 /* PAWS not needed at the moment */
424 /*
425 * RFC 1323 PAWS: If we have a timestamp reply on this segment
426 * and it's less than ts_recent, drop it.
427 */
428 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
429 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
430 if ((thflags & TH_ACK) == 0)
431 goto drop;
432 goto ack;
433 }
434 /*
435 * ts_recent is never updated because we never accept new segments.
436 */
437 #endif
438
439 /* Honor the drop_synfin sysctl variable. */
440 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
441 if ((s = tcp_log_addrs(&inp->inp_inc, th, NULL, NULL))) {
442 log(LOG_DEBUG, "%s; %s: "
443 "SYN|FIN segment ignored (based on "
444 "sysctl setting)\n", s, __func__);
445 free(s, M_TCPLOG);
446 }
447 goto drop;
448 }
449
450 /*
451 * If a new connection request is received
452 * while in TIME_WAIT, drop the old connection
453 * and start over if the sequence numbers
454 * are above the previous ones.
455 * Allow UDP port number changes in this case.
456 */
457 if (((thflags & (TH_SYN | TH_ACK)) == TH_SYN) &&
458 SEQ_GT(th->th_seq, tw->rcv_nxt)) {
459 tcp_twclose(tw, 0);
460 TCPSTAT_INC(tcps_tw_recycles);
461 return (1);
462 }
463
464 /*
465 * Send RST if UDP port numbers don't match
466 */
467 if (tw->t_port != m->m_pkthdr.tcp_tun_port) {
468 if (th->th_flags & TH_ACK) {
469 tcp_respond(NULL, mtod(m, void *), th, m,
470 (tcp_seq)0, th->th_ack, TH_RST);
471 } else {
472 if (th->th_flags & TH_SYN)
473 tlen++;
474 if (th->th_flags & TH_FIN)
475 tlen++;
476 tcp_respond(NULL, mtod(m, void *), th, m,
477 th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK);
478 }
479 INP_WUNLOCK(inp);
480 TCPSTAT_INC(tcps_tw_resets);
481 return (0);
482 }
483
484 /*
485 * Drop the segment if it does not contain an ACK.
486 */
487 if ((thflags & TH_ACK) == 0)
488 goto drop;
489
490 /*
491 * If timestamps were negotiated during SYN/ACK and a
492 * segment without a timestamp is received, silently drop
493 * the segment, unless the missing timestamps are tolerated.
494 * See section 3.2 of RFC 7323.
495 */
496 if (((to->to_flags & TOF_TS) == 0) && (tw->t_recent != 0) &&
497 (V_tcp_tolerate_missing_ts == 0)) {
498 goto drop;
499 }
500
501 /*
502 * Reset the 2MSL timer if this is a duplicate FIN.
503 */
504 if (thflags & TH_FIN) {
505 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
506 if (seq + 1 == tw->rcv_nxt)
507 tcp_tw_2msl_reset(tw, 1);
508 }
509
510 /*
511 * Acknowledge the segment if it has data or is not a duplicate ACK.
512 */
513 if (thflags != TH_ACK || tlen != 0 ||
514 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) {
515 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
516 tcp_twrespond(tw, TH_ACK);
517 TCPSTAT_INC(tcps_tw_responds);
518 goto dropnoprobe;
519 }
520 drop:
521 TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
522 dropnoprobe:
523 INP_WUNLOCK(inp);
524 m_freem(m);
525 return (0);
526 }
527
528 void
tcp_twclose(struct tcptw * tw,int reuse)529 tcp_twclose(struct tcptw *tw, int reuse)
530 {
531 struct socket *so;
532 struct inpcb *inp;
533
534 /*
535 * At this point, we are in one of two situations:
536 *
537 * (1) We have no socket, just an inpcb<->twtcp pair. We can free
538 * all state.
539 *
540 * (2) We have a socket -- if we own a reference, release it and
541 * notify the socket layer.
542 */
543 inp = tw->tw_inpcb;
544 KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
545 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
546 NET_EPOCH_ASSERT();
547 INP_WLOCK_ASSERT(inp);
548
549 tcp_tw_2msl_stop(tw, reuse);
550 inp->inp_ppcb = NULL;
551 in_pcbdrop(inp);
552
553 so = inp->inp_socket;
554 if (so != NULL) {
555 /*
556 * If there's a socket, handle two cases: first, we own a
557 * strong reference, which we will now release, or we don't
558 * in which case another reference exists (XXXRW: think
559 * about this more), and we don't need to take action.
560 */
561 if (inp->inp_flags & INP_SOCKREF) {
562 inp->inp_flags &= ~INP_SOCKREF;
563 INP_WUNLOCK(inp);
564 SOCK_LOCK(so);
565 KASSERT(so->so_state & SS_PROTOREF,
566 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
567 so->so_state &= ~SS_PROTOREF;
568 sofree(so);
569 } else {
570 /*
571 * If we don't own the only reference, the socket and
572 * inpcb need to be left around to be handled by
573 * tcp_usr_detach() later.
574 */
575 INP_WUNLOCK(inp);
576 }
577 } else {
578 /*
579 * The socket has been already cleaned-up for us, only free the
580 * inpcb.
581 */
582 in_pcbfree(inp);
583 }
584 TCPSTAT_INC(tcps_closed);
585 }
586
587 static int
tcp_twrespond(struct tcptw * tw,int flags)588 tcp_twrespond(struct tcptw *tw, int flags)
589 {
590 struct inpcb *inp = tw->tw_inpcb;
591 #if defined(INET6) || defined(INET)
592 struct tcphdr *th = NULL;
593 #endif
594 struct mbuf *m;
595 #ifdef INET
596 struct ip *ip = NULL;
597 #endif
598 u_int hdrlen, optlen, ulen;
599 int error = 0; /* Keep compiler happy */
600 struct tcpopt to;
601 #ifdef INET6
602 struct ip6_hdr *ip6 = NULL;
603 int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
604 #endif
605 struct udphdr *udp = NULL;
606 hdrlen = 0; /* Keep compiler happy */
607
608 INP_WLOCK_ASSERT(inp);
609
610 m = m_gethdr(M_NOWAIT, MT_DATA);
611 if (m == NULL)
612 return (ENOBUFS);
613 m->m_data += max_linkhdr;
614
615 #ifdef MAC
616 mac_inpcb_create_mbuf(inp, m);
617 #endif
618
619 #ifdef INET6
620 if (isipv6) {
621 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
622 ip6 = mtod(m, struct ip6_hdr *);
623 if (tw->t_port) {
624 udp = (struct udphdr *)(ip6 + 1);
625 hdrlen += sizeof(struct udphdr);
626 udp->uh_sport = htons(V_tcp_udp_tunneling_port);
627 udp->uh_dport = tw->t_port;
628 ulen = (hdrlen - sizeof(struct ip6_hdr));
629 th = (struct tcphdr *)(udp + 1);
630 } else
631 th = (struct tcphdr *)(ip6 + 1);
632 tcpip_fillheaders(inp, tw->t_port, ip6, th);
633 }
634 #endif
635 #if defined(INET6) && defined(INET)
636 else
637 #endif
638 #ifdef INET
639 {
640 hdrlen = sizeof(struct tcpiphdr);
641 ip = mtod(m, struct ip *);
642 if (tw->t_port) {
643 udp = (struct udphdr *)(ip + 1);
644 hdrlen += sizeof(struct udphdr);
645 udp->uh_sport = htons(V_tcp_udp_tunneling_port);
646 udp->uh_dport = tw->t_port;
647 ulen = (hdrlen - sizeof(struct ip));
648 th = (struct tcphdr *)(udp + 1);
649 } else
650 th = (struct tcphdr *)(ip + 1);
651 tcpip_fillheaders(inp, tw->t_port, ip, th);
652 }
653 #endif
654 to.to_flags = 0;
655
656 /*
657 * Send a timestamp and echo-reply if both our side and our peer
658 * have sent timestamps in our SYN's and this is not a RST.
659 */
660 if (tw->t_recent && flags == TH_ACK) {
661 to.to_flags |= TOF_TS;
662 to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
663 to.to_tsecr = tw->t_recent;
664 }
665 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
666 if (tw->tw_flags & TF_SIGNATURE)
667 to.to_flags |= TOF_SIGNATURE;
668 #endif
669 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
670
671 if (udp) {
672 ulen += optlen;
673 udp->uh_ulen = htons(ulen);
674 }
675 m->m_len = hdrlen + optlen;
676 m->m_pkthdr.len = m->m_len;
677
678 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
679
680 th->th_seq = htonl(tw->snd_nxt);
681 th->th_ack = htonl(tw->rcv_nxt);
682 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
683 th->th_flags = flags;
684 th->th_win = htons(tw->last_win);
685
686 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
687 if (tw->tw_flags & TF_SIGNATURE) {
688 if (!TCPMD5_ENABLED() ||
689 TCPMD5_OUTPUT(m, th, to.to_signature) != 0)
690 return (-1);
691 }
692 #endif
693 #ifdef INET6
694 if (isipv6) {
695 if (tw->t_port) {
696 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
697 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
698 udp->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
699 th->th_sum = htons(0);
700 } else {
701 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
702 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
703 th->th_sum = in6_cksum_pseudo(ip6,
704 sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
705 }
706 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
707 TCP_PROBE5(send, NULL, NULL, ip6, NULL, th);
708 error = ip6_output(m, inp->in6p_outputopts, NULL,
709 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
710 }
711 #endif
712 #if defined(INET6) && defined(INET)
713 else
714 #endif
715 #ifdef INET
716 {
717 if (tw->t_port) {
718 m->m_pkthdr.csum_flags = CSUM_UDP;
719 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
720 udp->uh_sum = in_pseudo(ip->ip_src.s_addr,
721 ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP));
722 th->th_sum = htons(0);
723 } else {
724 m->m_pkthdr.csum_flags = CSUM_TCP;
725 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
726 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
727 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
728 }
729 ip->ip_len = htons(m->m_pkthdr.len);
730 if (V_path_mtu_discovery)
731 ip->ip_off |= htons(IP_DF);
732 TCP_PROBE5(send, NULL, NULL, ip, NULL, th);
733 error = ip_output(m, inp->inp_options, NULL,
734 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
735 NULL, inp);
736 }
737 #endif
738 if (flags & TH_ACK)
739 TCPSTAT_INC(tcps_sndacks);
740 else
741 TCPSTAT_INC(tcps_sndctrl);
742 TCPSTAT_INC(tcps_sndtotal);
743 return (error);
744 }
745
746 static void
tcp_tw_2msl_reset(struct tcptw * tw,int rearm)747 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
748 {
749
750 NET_EPOCH_ASSERT();
751 INP_WLOCK_ASSERT(tw->tw_inpcb);
752
753 TW_WLOCK(V_tw_lock);
754 if (rearm)
755 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
756 tw->tw_time = ticks + 2 * V_tcp_msl;
757 TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
758 TW_WUNLOCK(V_tw_lock);
759 }
760
761 static void
tcp_tw_2msl_stop(struct tcptw * tw,int reuse)762 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
763 {
764 struct ucred *cred;
765 struct inpcb *inp;
766 int released __unused;
767
768 NET_EPOCH_ASSERT();
769
770 TW_WLOCK(V_tw_lock);
771 inp = tw->tw_inpcb;
772 tw->tw_inpcb = NULL;
773
774 TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
775 cred = tw->tw_cred;
776 tw->tw_cred = NULL;
777 TW_WUNLOCK(V_tw_lock);
778
779 if (cred != NULL)
780 crfree(cred);
781
782 released = in_pcbrele_wlocked(inp);
783 KASSERT(!released, ("%s: inp should not be released here", __func__));
784
785 if (!reuse)
786 uma_zfree(V_tcptw_zone, tw);
787 TCPSTATES_DEC(TCPS_TIME_WAIT);
788 }
789
790 struct tcptw *
tcp_tw_2msl_scan(int reuse)791 tcp_tw_2msl_scan(int reuse)
792 {
793 struct tcptw *tw;
794 struct inpcb *inp;
795
796 NET_EPOCH_ASSERT();
797
798 for (;;) {
799 TW_RLOCK(V_tw_lock);
800 tw = TAILQ_FIRST(&V_twq_2msl);
801 if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
802 TW_RUNLOCK(V_tw_lock);
803 break;
804 }
805 KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
806 __func__));
807
808 inp = tw->tw_inpcb;
809 in_pcbref(inp);
810 TW_RUNLOCK(V_tw_lock);
811
812 INP_WLOCK(inp);
813 tw = intotw(inp);
814 if (in_pcbrele_wlocked(inp)) {
815 if (__predict_true(tw == NULL)) {
816 continue;
817 } else {
818 /* This should not happen as in TIMEWAIT
819 * state the inp should not be destroyed
820 * before its tcptw. If INVARIANTS is
821 * defined panic.
822 */
823 #ifdef INVARIANTS
824 panic("%s: Panic before an infinite "
825 "loop: INP_TIMEWAIT && (INP_FREED "
826 "|| inp last reference) && tw != "
827 "NULL", __func__);
828 #else
829 log(LOG_ERR, "%s: Avoid an infinite "
830 "loop: INP_TIMEWAIT && (INP_FREED "
831 "|| inp last reference) && tw != "
832 "NULL", __func__);
833 #endif
834 break;
835 }
836 }
837
838 if (tw == NULL) {
839 /* tcp_twclose() has already been called */
840 INP_WUNLOCK(inp);
841 continue;
842 }
843
844 tcp_twclose(tw, reuse);
845 if (reuse)
846 return tw;
847 }
848
849 return NULL;
850 }
851