xref: /freebsd/sys/netinet/tcp_timewait.c (revision 8847579c57d6aff2b3371c707dce7a2cee8389aa)
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  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_sack.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/mac.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/proc.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
57 
58 #include <vm/uma.h>
59 
60 #include <net/route.h>
61 #include <net/if.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69 #include <netinet/in_pcb.h>
70 #ifdef INET6
71 #include <netinet6/in6_pcb.h>
72 #endif
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #endif
80 #include <netinet/ip_icmp.h>
81 #include <netinet/tcp.h>
82 #include <netinet/tcp_fsm.h>
83 #include <netinet/tcp_seq.h>
84 #include <netinet/tcp_timer.h>
85 #include <netinet/tcp_var.h>
86 #ifdef INET6
87 #include <netinet6/tcp6_var.h>
88 #endif
89 #include <netinet/tcpip.h>
90 #ifdef TCPDEBUG
91 #include <netinet/tcp_debug.h>
92 #endif
93 #include <netinet6/ip6protosw.h>
94 
95 #ifdef IPSEC
96 #include <netinet6/ipsec.h>
97 #ifdef INET6
98 #include <netinet6/ipsec6.h>
99 #endif
100 #include <netkey/key.h>
101 #endif /*IPSEC*/
102 
103 #ifdef FAST_IPSEC
104 #include <netipsec/ipsec.h>
105 #include <netipsec/xform.h>
106 #ifdef INET6
107 #include <netipsec/ipsec6.h>
108 #endif
109 #include <netipsec/key.h>
110 #define	IPSEC
111 #endif /*FAST_IPSEC*/
112 
113 #include <machine/in_cksum.h>
114 #include <sys/md5.h>
115 
116 int	tcp_mssdflt = TCP_MSS;
117 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
118     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
119 
120 #ifdef INET6
121 int	tcp_v6mssdflt = TCP6_MSS;
122 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
123 	CTLFLAG_RW, &tcp_v6mssdflt , 0,
124 	"Default TCP Maximum Segment Size for IPv6");
125 #endif
126 
127 /*
128  * Minimum MSS we accept and use. This prevents DoS attacks where
129  * we are forced to a ridiculous low MSS like 20 and send hundreds
130  * of packets instead of one. The effect scales with the available
131  * bandwidth and quickly saturates the CPU and network interface
132  * with packet generation and sending. Set to zero to disable MINMSS
133  * checking. This setting prevents us from sending too small packets.
134  */
135 int	tcp_minmss = TCP_MINMSS;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
137     &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
138 /*
139  * Number of TCP segments per second we accept from remote host
140  * before we start to calculate average segment size. If average
141  * segment size drops below the minimum TCP MSS we assume a DoS
142  * attack and reset+drop the connection. Care has to be taken not to
143  * set this value too small to not kill interactive type connections
144  * (telnet, SSH) which send many small packets.
145  */
146 int     tcp_minmssoverload = TCP_MINMSSOVERLOAD;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
148     &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
149     "be under the MINMSS Size");
150 
151 #if 0
152 static int	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
153 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
154     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
155 #endif
156 
157 int	tcp_do_rfc1323 = 1;
158 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
159     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
160 
161 static int	tcp_tcbhashsize = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
163      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
164 
165 static int	do_tcpdrain = 1;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
167      "Enable tcp_drain routine for extra help when low on mbufs");
168 
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
170     &tcbinfo.ipi_count, 0, "Number of active PCBs");
171 
172 static int	icmp_may_rst = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
174     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
175 
176 static int	tcp_isn_reseed_interval = 0;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
178     &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
179 
180 static int	maxtcptw;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, maxtcptw, CTLFLAG_RDTUN,
182     &maxtcptw, 0, "Maximum number of compressed TCP TIME_WAIT entries");
183 
184 /*
185  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
186  * 1024 exists only for debugging.  A good production default would be
187  * something like 6100.
188  */
189 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
190     "TCP inflight data limiting");
191 
192 static int	tcp_inflight_enable = 1;
193 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
194     &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
195 
196 static int	tcp_inflight_debug = 0;
197 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
198     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
199 
200 static int	tcp_inflight_rttthresh;
201 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
202     &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
203     "RTT threshold below which inflight will deactivate itself");
204 
205 static int	tcp_inflight_min = 6144;
206 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
207     &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
208 
209 static int	tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
210 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
211     &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
212 
213 static int	tcp_inflight_stab = 20;
214 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
215     &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
216 
217 uma_zone_t sack_hole_zone;
218 
219 static struct inpcb *tcp_notify(struct inpcb *, int);
220 static void	tcp_isn_tick(void *);
221 
222 /*
223  * Target size of TCP PCB hash tables. Must be a power of two.
224  *
225  * Note that this can be overridden by the kernel environment
226  * variable net.inet.tcp.tcbhashsize
227  */
228 #ifndef TCBHASHSIZE
229 #define TCBHASHSIZE	512
230 #endif
231 
232 /*
233  * XXX
234  * Callouts should be moved into struct tcp directly.  They are currently
235  * separate because the tcpcb structure is exported to userland for sysctl
236  * parsing purposes, which do not know about callouts.
237  */
238 struct	tcpcb_mem {
239 	struct	tcpcb tcb;
240 	struct	callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
241 	struct	callout tcpcb_mem_2msl, tcpcb_mem_delack;
242 };
243 
244 static uma_zone_t tcpcb_zone;
245 static uma_zone_t tcptw_zone;
246 struct callout isn_callout;
247 static struct mtx isn_mtx;
248 
249 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
250 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
251 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
252 
253 /*
254  * TCP initialization.
255  */
256 static void
257 tcp_zone_change(void *tag)
258 {
259 
260 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
261 	uma_zone_set_max(tcpcb_zone, maxsockets);
262 	uma_zone_set_max(tcptw_zone, maxsockets / 5);
263 }
264 
265 void
266 tcp_init(void)
267 {
268 	int hashsize = TCBHASHSIZE;
269 
270 	tcp_delacktime = TCPTV_DELACK;
271 	tcp_keepinit = TCPTV_KEEP_INIT;
272 	tcp_keepidle = TCPTV_KEEP_IDLE;
273 	tcp_keepintvl = TCPTV_KEEPINTVL;
274 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
275 	tcp_msl = TCPTV_MSL;
276 	tcp_rexmit_min = TCPTV_MIN;
277 	tcp_rexmit_slop = TCPTV_CPU_VAR;
278 	tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
279 
280 	INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
281 	LIST_INIT(&tcb);
282 	tcbinfo.listhead = &tcb;
283 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
284 	if (!powerof2(hashsize)) {
285 		printf("WARNING: TCB hash size not a power of 2\n");
286 		hashsize = 512; /* safe default */
287 	}
288 	tcp_tcbhashsize = hashsize;
289 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
290 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
291 					&tcbinfo.porthashmask);
292 	tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
293 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
294 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
295 #ifdef INET6
296 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
297 #else /* INET6 */
298 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
299 #endif /* INET6 */
300 	if (max_protohdr < TCP_MINPROTOHDR)
301 		max_protohdr = TCP_MINPROTOHDR;
302 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
303 		panic("tcp_init");
304 #undef TCP_MINPROTOHDR
305 	/*
306 	 * These have to be type stable for the benefit of the timers.
307 	 */
308 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
309 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
310 	uma_zone_set_max(tcpcb_zone, maxsockets);
311 	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
312 	if (maxtcptw == 0)
313 		maxtcptw = maxsockets / 5;
314 	tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
315 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
316 	uma_zone_set_max(tcptw_zone, maxtcptw);
317 	tcp_timer_init();
318 	syncache_init();
319 	tcp_hc_init();
320 	tcp_reass_init();
321 	ISN_LOCK_INIT();
322 	callout_init(&isn_callout, CALLOUT_MPSAFE);
323 	tcp_isn_tick(NULL);
324 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
325 		SHUTDOWN_PRI_DEFAULT);
326 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
327 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
328 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
329 		EVENTHANDLER_PRI_ANY);
330 }
331 
332 void
333 tcp_fini(void *xtp)
334 {
335 
336 	callout_stop(&isn_callout);
337 }
338 
339 /*
340  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
341  * tcp_template used to store this data in mbufs, but we now recopy it out
342  * of the tcpcb each time to conserve mbufs.
343  */
344 void
345 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
346 {
347 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
348 
349 	INP_LOCK_ASSERT(inp);
350 
351 #ifdef INET6
352 	if ((inp->inp_vflag & INP_IPV6) != 0) {
353 		struct ip6_hdr *ip6;
354 
355 		ip6 = (struct ip6_hdr *)ip_ptr;
356 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
357 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
358 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
359 			(IPV6_VERSION & IPV6_VERSION_MASK);
360 		ip6->ip6_nxt = IPPROTO_TCP;
361 		ip6->ip6_plen = sizeof(struct tcphdr);
362 		ip6->ip6_src = inp->in6p_laddr;
363 		ip6->ip6_dst = inp->in6p_faddr;
364 	} else
365 #endif
366 	{
367 		struct ip *ip;
368 
369 		ip = (struct ip *)ip_ptr;
370 		ip->ip_v = IPVERSION;
371 		ip->ip_hl = 5;
372 		ip->ip_tos = inp->inp_ip_tos;
373 		ip->ip_len = 0;
374 		ip->ip_id = 0;
375 		ip->ip_off = 0;
376 		ip->ip_ttl = inp->inp_ip_ttl;
377 		ip->ip_sum = 0;
378 		ip->ip_p = IPPROTO_TCP;
379 		ip->ip_src = inp->inp_laddr;
380 		ip->ip_dst = inp->inp_faddr;
381 	}
382 	th->th_sport = inp->inp_lport;
383 	th->th_dport = inp->inp_fport;
384 	th->th_seq = 0;
385 	th->th_ack = 0;
386 	th->th_x2 = 0;
387 	th->th_off = 5;
388 	th->th_flags = 0;
389 	th->th_win = 0;
390 	th->th_urp = 0;
391 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
392 }
393 
394 /*
395  * Create template to be used to send tcp packets on a connection.
396  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
397  * use for this function is in keepalives, which use tcp_respond.
398  */
399 struct tcptemp *
400 tcpip_maketemplate(struct inpcb *inp)
401 {
402 	struct mbuf *m;
403 	struct tcptemp *n;
404 
405 	m = m_get(M_DONTWAIT, MT_DATA);
406 	if (m == NULL)
407 		return (0);
408 	m->m_len = sizeof(struct tcptemp);
409 	n = mtod(m, struct tcptemp *);
410 
411 	tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
412 	return (n);
413 }
414 
415 /*
416  * Send a single message to the TCP at address specified by
417  * the given TCP/IP header.  If m == NULL, then we make a copy
418  * of the tcpiphdr at ti and send directly to the addressed host.
419  * This is used to force keep alive messages out using the TCP
420  * template for a connection.  If flags are given then we send
421  * a message back to the TCP which originated the * segment ti,
422  * and discard the mbuf containing it and any other attached mbufs.
423  *
424  * In any case the ack and sequence number of the transmitted
425  * segment are as specified by the parameters.
426  *
427  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
428  */
429 void
430 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
431     register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
432 {
433 	register int tlen;
434 	int win = 0;
435 	struct ip *ip;
436 	struct tcphdr *nth;
437 #ifdef INET6
438 	struct ip6_hdr *ip6;
439 	int isipv6;
440 #endif /* INET6 */
441 	int ipflags = 0;
442 	struct inpcb *inp;
443 
444 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
445 
446 #ifdef INET6
447 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
448 	ip6 = ipgen;
449 #endif /* INET6 */
450 	ip = ipgen;
451 
452 	if (tp != NULL) {
453 		inp = tp->t_inpcb;
454 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
455 		INP_INFO_WLOCK_ASSERT(&tcbinfo);
456 		INP_LOCK_ASSERT(inp);
457 	} else
458 		inp = NULL;
459 
460 	if (tp != NULL) {
461 		if (!(flags & TH_RST)) {
462 			win = sbspace(&inp->inp_socket->so_rcv);
463 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
464 				win = (long)TCP_MAXWIN << tp->rcv_scale;
465 		}
466 	}
467 	if (m == NULL) {
468 		m = m_gethdr(M_DONTWAIT, MT_DATA);
469 		if (m == NULL)
470 			return;
471 		tlen = 0;
472 		m->m_data += max_linkhdr;
473 #ifdef INET6
474 		if (isipv6) {
475 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
476 			      sizeof(struct ip6_hdr));
477 			ip6 = mtod(m, struct ip6_hdr *);
478 			nth = (struct tcphdr *)(ip6 + 1);
479 		} else
480 #endif /* INET6 */
481 	      {
482 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
483 		ip = mtod(m, struct ip *);
484 		nth = (struct tcphdr *)(ip + 1);
485 	      }
486 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
487 		flags = TH_ACK;
488 	} else {
489 		m_freem(m->m_next);
490 		m->m_next = NULL;
491 		m->m_data = (caddr_t)ipgen;
492 		/* m_len is set later */
493 		tlen = 0;
494 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
495 #ifdef INET6
496 		if (isipv6) {
497 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
498 			nth = (struct tcphdr *)(ip6 + 1);
499 		} else
500 #endif /* INET6 */
501 	      {
502 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
503 		nth = (struct tcphdr *)(ip + 1);
504 	      }
505 		if (th != nth) {
506 			/*
507 			 * this is usually a case when an extension header
508 			 * exists between the IPv6 header and the
509 			 * TCP header.
510 			 */
511 			nth->th_sport = th->th_sport;
512 			nth->th_dport = th->th_dport;
513 		}
514 		xchg(nth->th_dport, nth->th_sport, n_short);
515 #undef xchg
516 	}
517 #ifdef INET6
518 	if (isipv6) {
519 		ip6->ip6_flow = 0;
520 		ip6->ip6_vfc = IPV6_VERSION;
521 		ip6->ip6_nxt = IPPROTO_TCP;
522 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
523 						tlen));
524 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
525 	} else
526 #endif
527 	{
528 		tlen += sizeof (struct tcpiphdr);
529 		ip->ip_len = tlen;
530 		ip->ip_ttl = ip_defttl;
531 		if (path_mtu_discovery)
532 			ip->ip_off |= IP_DF;
533 	}
534 	m->m_len = tlen;
535 	m->m_pkthdr.len = tlen;
536 	m->m_pkthdr.rcvif = NULL;
537 #ifdef MAC
538 	if (inp != NULL) {
539 		/*
540 		 * Packet is associated with a socket, so allow the
541 		 * label of the response to reflect the socket label.
542 		 */
543 		INP_LOCK_ASSERT(inp);
544 		mac_create_mbuf_from_inpcb(inp, m);
545 	} else {
546 		/*
547 		 * Packet is not associated with a socket, so possibly
548 		 * update the label in place.
549 		 */
550 		mac_reflect_mbuf_tcp(m);
551 	}
552 #endif
553 	nth->th_seq = htonl(seq);
554 	nth->th_ack = htonl(ack);
555 	nth->th_x2 = 0;
556 	nth->th_off = sizeof (struct tcphdr) >> 2;
557 	nth->th_flags = flags;
558 	if (tp != NULL)
559 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
560 	else
561 		nth->th_win = htons((u_short)win);
562 	nth->th_urp = 0;
563 #ifdef INET6
564 	if (isipv6) {
565 		nth->th_sum = 0;
566 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
567 					sizeof(struct ip6_hdr),
568 					tlen - sizeof(struct ip6_hdr));
569 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
570 		    NULL, NULL);
571 	} else
572 #endif /* INET6 */
573 	{
574 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
575 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
576 		m->m_pkthdr.csum_flags = CSUM_TCP;
577 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
578 	}
579 #ifdef TCPDEBUG
580 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
581 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
582 #endif
583 #ifdef INET6
584 	if (isipv6)
585 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
586 	else
587 #endif /* INET6 */
588 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
589 }
590 
591 /*
592  * Create a new TCP control block, making an
593  * empty reassembly queue and hooking it to the argument
594  * protocol control block.  The `inp' parameter must have
595  * come from the zone allocator set up in tcp_init().
596  */
597 struct tcpcb *
598 tcp_newtcpcb(struct inpcb *inp)
599 {
600 	struct tcpcb_mem *tm;
601 	struct tcpcb *tp;
602 #ifdef INET6
603 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
604 #endif /* INET6 */
605 
606 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
607 	if (tm == NULL)
608 		return (NULL);
609 	tp = &tm->tcb;
610 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
611 	tp->t_maxseg = tp->t_maxopd =
612 #ifdef INET6
613 		isipv6 ? tcp_v6mssdflt :
614 #endif /* INET6 */
615 		tcp_mssdflt;
616 
617 	/* Set up our timeouts. */
618 	callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
619 	callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
620 	callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
621 	callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
622 	callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
623 
624 	if (tcp_do_rfc1323)
625 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
626 	tp->sack_enable = tcp_do_sack;
627 	TAILQ_INIT(&tp->snd_holes);
628 	tp->t_inpcb = inp;	/* XXX */
629 	/*
630 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
631 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
632 	 * reasonable initial retransmit time.
633 	 */
634 	tp->t_srtt = TCPTV_SRTTBASE;
635 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
636 	tp->t_rttmin = tcp_rexmit_min;
637 	tp->t_rxtcur = TCPTV_RTOBASE;
638 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
639 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
640 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
641 	tp->t_rcvtime = ticks;
642 	tp->t_bw_rtttime = ticks;
643 	/*
644 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
645 	 * because the socket may be bound to an IPv6 wildcard address,
646 	 * which may match an IPv4-mapped IPv6 address.
647 	 */
648 	inp->inp_ip_ttl = ip_defttl;
649 	inp->inp_ppcb = tp;
650 	return (tp);		/* XXX */
651 }
652 
653 /*
654  * Drop a TCP connection, reporting
655  * the specified error.  If connection is synchronized,
656  * then send a RST to peer.
657  */
658 struct tcpcb *
659 tcp_drop(struct tcpcb *tp, int errno)
660 {
661 	struct socket *so = tp->t_inpcb->inp_socket;
662 
663 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
664 	INP_LOCK_ASSERT(tp->t_inpcb);
665 
666 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
667 		tp->t_state = TCPS_CLOSED;
668 		(void) tcp_output(tp);
669 		tcpstat.tcps_drops++;
670 	} else
671 		tcpstat.tcps_conndrops++;
672 	if (errno == ETIMEDOUT && tp->t_softerror)
673 		errno = tp->t_softerror;
674 	so->so_error = errno;
675 	return (tcp_close(tp));
676 }
677 
678 void
679 tcp_discardcb(struct tcpcb *tp)
680 {
681 	struct tseg_qent *q;
682 	struct inpcb *inp = tp->t_inpcb;
683 	struct socket *so = inp->inp_socket;
684 #ifdef INET6
685 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
686 #endif /* INET6 */
687 
688 	/*
689 	 * XXXRW: This is all very well and good, but actually, we might be
690 	 * discarding the tcpcb after the socket is gone, so we can't do
691 	 * this:
692 	KASSERT(so != NULL, ("tcp_discardcb: so == NULL"));
693 	 */
694 	INP_LOCK_ASSERT(inp);
695 
696 	/*
697 	 * Make sure that all of our timers are stopped before we
698 	 * delete the PCB.
699 	 */
700 	callout_stop(tp->tt_rexmt);
701 	callout_stop(tp->tt_persist);
702 	callout_stop(tp->tt_keep);
703 	callout_stop(tp->tt_2msl);
704 	callout_stop(tp->tt_delack);
705 
706 	/*
707 	 * If we got enough samples through the srtt filter,
708 	 * save the rtt and rttvar in the routing entry.
709 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
710 	 * 4 samples is enough for the srtt filter to converge
711 	 * to within enough % of the correct value; fewer samples
712 	 * and we could save a bogus rtt. The danger is not high
713 	 * as tcp quickly recovers from everything.
714 	 * XXX: Works very well but needs some more statistics!
715 	 */
716 	if (tp->t_rttupdated >= 4) {
717 		struct hc_metrics_lite metrics;
718 		u_long ssthresh;
719 
720 		bzero(&metrics, sizeof(metrics));
721 		/*
722 		 * Update the ssthresh always when the conditions below
723 		 * are satisfied. This gives us better new start value
724 		 * for the congestion avoidance for new connections.
725 		 * ssthresh is only set if packet loss occured on a session.
726 		 */
727 		ssthresh = tp->snd_ssthresh;
728 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
729 			/*
730 			 * convert the limit from user data bytes to
731 			 * packets then to packet data bytes.
732 			 */
733 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
734 			if (ssthresh < 2)
735 				ssthresh = 2;
736 			ssthresh *= (u_long)(tp->t_maxseg +
737 #ifdef INET6
738 				      (isipv6 ? sizeof (struct ip6_hdr) +
739 					       sizeof (struct tcphdr) :
740 #endif
741 				       sizeof (struct tcpiphdr)
742 #ifdef INET6
743 				       )
744 #endif
745 				      );
746 		} else
747 			ssthresh = 0;
748 		metrics.rmx_ssthresh = ssthresh;
749 
750 		metrics.rmx_rtt = tp->t_srtt;
751 		metrics.rmx_rttvar = tp->t_rttvar;
752 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
753 		metrics.rmx_bandwidth = tp->snd_bandwidth;
754 		metrics.rmx_cwnd = tp->snd_cwnd;
755 		metrics.rmx_sendpipe = 0;
756 		metrics.rmx_recvpipe = 0;
757 
758 		tcp_hc_update(&inp->inp_inc, &metrics);
759 	}
760 
761 	/* free the reassembly queue, if any */
762 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
763 		LIST_REMOVE(q, tqe_q);
764 		m_freem(q->tqe_m);
765 		uma_zfree(tcp_reass_zone, q);
766 		tp->t_segqlen--;
767 		tcp_reass_qsize--;
768 	}
769 	tcp_free_sackholes(tp);
770 	inp->inp_ppcb = NULL;
771 	tp->t_inpcb = NULL;
772 	uma_zfree(tcpcb_zone, tp);
773 
774 	/*
775 	 * XXXRW: This seems a bit unclean.
776 	 */
777 	if (so != NULL)
778 		soisdisconnected(so);
779 }
780 
781 /*
782  * Attempt to close a TCP control block, marking it as dropped, and freeing
783  * the socket if we hold the only reference.
784  */
785 struct tcpcb *
786 tcp_close(struct tcpcb *tp)
787 {
788 	struct inpcb *inp = tp->t_inpcb;
789 	struct socket *so;
790 
791 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
792 	INP_LOCK_ASSERT(inp);
793 
794 	in_pcbdrop(inp);
795 	tcpstat.tcps_closed++;
796 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
797 	so = inp->inp_socket;
798 	soisdisconnected(so);
799 	if (inp->inp_vflag & INP_SOCKREF) {
800 		KASSERT(so->so_state & SS_PROTOREF,
801 		    ("tcp_close: !SS_PROTOREF"));
802 		inp->inp_vflag &= ~INP_SOCKREF;
803 		tcp_discardcb(tp);
804 #ifdef INET6
805 		if (inp->inp_vflag & INP_IPV6PROTO) {
806 			in6_pcbdetach(inp);
807 			in6_pcbfree(inp);
808 		} else {
809 #endif
810 			in_pcbdetach(inp);
811 			in_pcbfree(inp);
812 #ifdef INET6
813 		}
814 #endif
815 		ACCEPT_LOCK();
816 		SOCK_LOCK(so);
817 		so->so_state &= ~SS_PROTOREF;
818 		sofree(so);
819 		return (NULL);
820 	}
821 	return (tp);
822 }
823 
824 void
825 tcp_drain(void)
826 {
827 
828 	if (do_tcpdrain) {
829 		struct inpcb *inpb;
830 		struct tcpcb *tcpb;
831 		struct tseg_qent *te;
832 
833 	/*
834 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
835 	 * if there is one...
836 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
837 	 *      reassembly queue should be flushed, but in a situation
838 	 *	where we're really low on mbufs, this is potentially
839 	 *	usefull.
840 	 */
841 		INP_INFO_RLOCK(&tcbinfo);
842 		LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
843 			if (inpb->inp_vflag & INP_TIMEWAIT)
844 				continue;
845 			INP_LOCK(inpb);
846 			if ((tcpb = intotcpcb(inpb)) != NULL) {
847 				while ((te = LIST_FIRST(&tcpb->t_segq))
848 			            != NULL) {
849 					LIST_REMOVE(te, tqe_q);
850 					m_freem(te->tqe_m);
851 					uma_zfree(tcp_reass_zone, te);
852 					tcpb->t_segqlen--;
853 					tcp_reass_qsize--;
854 				}
855 				tcp_clean_sackreport(tcpb);
856 			}
857 			INP_UNLOCK(inpb);
858 		}
859 		INP_INFO_RUNLOCK(&tcbinfo);
860 	}
861 }
862 
863 /*
864  * Notify a tcp user of an asynchronous error;
865  * store error as soft error, but wake up user
866  * (for now, won't do anything until can select for soft error).
867  *
868  * Do not wake up user since there currently is no mechanism for
869  * reporting soft errors (yet - a kqueue filter may be added).
870  */
871 static struct inpcb *
872 tcp_notify(struct inpcb *inp, int error)
873 {
874 	struct tcpcb *tp;
875 
876 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
877 	INP_LOCK_ASSERT(inp);
878 
879 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
880 	    (inp->inp_vflag & INP_DROPPED))
881 		return (inp);
882 
883 	tp = intotcpcb(inp);
884 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
885 
886 	/*
887 	 * Ignore some errors if we are hooked up.
888 	 * If connection hasn't completed, has retransmitted several times,
889 	 * and receives a second error, give up now.  This is better
890 	 * than waiting a long time to establish a connection that
891 	 * can never complete.
892 	 */
893 	if (tp->t_state == TCPS_ESTABLISHED &&
894 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
895 	     error == EHOSTDOWN)) {
896 		return (inp);
897 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
898 	    tp->t_softerror) {
899 		tp = tcp_drop(tp, error);
900 		if (tp != NULL)
901 			return (inp);
902 		else
903 			return (NULL);
904 	} else {
905 		tp->t_softerror = error;
906 		return (inp);
907 	}
908 #if 0
909 	wakeup( &so->so_timeo);
910 	sorwakeup(so);
911 	sowwakeup(so);
912 #endif
913 }
914 
915 static int
916 tcp_pcblist(SYSCTL_HANDLER_ARGS)
917 {
918 	int error, i, n;
919 	struct inpcb *inp, **inp_list;
920 	inp_gen_t gencnt;
921 	struct xinpgen xig;
922 
923 	/*
924 	 * The process of preparing the TCB list is too time-consuming and
925 	 * resource-intensive to repeat twice on every request.
926 	 */
927 	if (req->oldptr == NULL) {
928 		n = tcbinfo.ipi_count;
929 		req->oldidx = 2 * (sizeof xig)
930 			+ (n + n/8) * sizeof(struct xtcpcb);
931 		return (0);
932 	}
933 
934 	if (req->newptr != NULL)
935 		return (EPERM);
936 
937 	/*
938 	 * OK, now we're committed to doing something.
939 	 */
940 	INP_INFO_RLOCK(&tcbinfo);
941 	gencnt = tcbinfo.ipi_gencnt;
942 	n = tcbinfo.ipi_count;
943 	INP_INFO_RUNLOCK(&tcbinfo);
944 
945 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
946 		+ n * sizeof(struct xtcpcb));
947 	if (error != 0)
948 		return (error);
949 
950 	xig.xig_len = sizeof xig;
951 	xig.xig_count = n;
952 	xig.xig_gen = gencnt;
953 	xig.xig_sogen = so_gencnt;
954 	error = SYSCTL_OUT(req, &xig, sizeof xig);
955 	if (error)
956 		return (error);
957 
958 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
959 	if (inp_list == NULL)
960 		return (ENOMEM);
961 
962 	INP_INFO_RLOCK(&tcbinfo);
963 	for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
964 	     inp = LIST_NEXT(inp, inp_list)) {
965 		INP_LOCK(inp);
966 		if (inp->inp_gencnt <= gencnt) {
967 			/*
968 			 * XXX: This use of cr_cansee(), introduced with
969 			 * TCP state changes, is not quite right, but for
970 			 * now, better than nothing.
971 			 */
972 			if (inp->inp_vflag & INP_TIMEWAIT) {
973 				if (intotw(inp) != NULL)
974 					error = cr_cansee(req->td->td_ucred,
975 					    intotw(inp)->tw_cred);
976 				else
977 					error = EINVAL;	/* Skip this inp. */
978 			} else
979 				error = cr_canseesocket(req->td->td_ucred,
980 				    inp->inp_socket);
981 			if (error == 0)
982 				inp_list[i++] = inp;
983 		}
984 		INP_UNLOCK(inp);
985 	}
986 	INP_INFO_RUNLOCK(&tcbinfo);
987 	n = i;
988 
989 	error = 0;
990 	for (i = 0; i < n; i++) {
991 		inp = inp_list[i];
992 		if (inp->inp_gencnt <= gencnt) {
993 			struct xtcpcb xt;
994 			void *inp_ppcb;
995 
996 			bzero(&xt, sizeof(xt));
997 			xt.xt_len = sizeof xt;
998 			/* XXX should avoid extra copy */
999 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1000 			inp_ppcb = inp->inp_ppcb;
1001 			if (inp_ppcb == NULL)
1002 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1003 			else if (inp->inp_vflag & INP_TIMEWAIT) {
1004 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1005 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1006 			} else
1007 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1008 			if (inp->inp_socket != NULL)
1009 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1010 			else {
1011 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1012 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1013 			}
1014 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1015 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1016 		}
1017 	}
1018 	if (!error) {
1019 		/*
1020 		 * Give the user an updated idea of our state.
1021 		 * If the generation differs from what we told
1022 		 * her before, she knows that something happened
1023 		 * while we were processing this request, and it
1024 		 * might be necessary to retry.
1025 		 */
1026 		INP_INFO_RLOCK(&tcbinfo);
1027 		xig.xig_gen = tcbinfo.ipi_gencnt;
1028 		xig.xig_sogen = so_gencnt;
1029 		xig.xig_count = tcbinfo.ipi_count;
1030 		INP_INFO_RUNLOCK(&tcbinfo);
1031 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1032 	}
1033 	free(inp_list, M_TEMP);
1034 	return (error);
1035 }
1036 
1037 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1038 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1039 
1040 static int
1041 tcp_getcred(SYSCTL_HANDLER_ARGS)
1042 {
1043 	struct xucred xuc;
1044 	struct sockaddr_in addrs[2];
1045 	struct inpcb *inp;
1046 	int error;
1047 
1048 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1049 	if (error)
1050 		return (error);
1051 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1052 	if (error)
1053 		return (error);
1054 	INP_INFO_RLOCK(&tcbinfo);
1055 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1056 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1057 	if (inp == NULL) {
1058 		error = ENOENT;
1059 		goto outunlocked;
1060 	}
1061 	INP_LOCK(inp);
1062 	if (inp->inp_socket == NULL) {
1063 		error = ENOENT;
1064 		goto out;
1065 	}
1066 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1067 	if (error)
1068 		goto out;
1069 	cru2x(inp->inp_socket->so_cred, &xuc);
1070 out:
1071 	INP_UNLOCK(inp);
1072 outunlocked:
1073 	INP_INFO_RUNLOCK(&tcbinfo);
1074 	if (error == 0)
1075 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1076 	return (error);
1077 }
1078 
1079 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1080     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1081     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1082 
1083 #ifdef INET6
1084 static int
1085 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1086 {
1087 	struct xucred xuc;
1088 	struct sockaddr_in6 addrs[2];
1089 	struct inpcb *inp;
1090 	int error, mapped = 0;
1091 
1092 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1093 	if (error)
1094 		return (error);
1095 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1096 	if (error)
1097 		return (error);
1098 	if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1099 	    (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1100 		return (error);
1101 	}
1102 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1103 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1104 			mapped = 1;
1105 		else
1106 			return (EINVAL);
1107 	}
1108 
1109 	INP_INFO_RLOCK(&tcbinfo);
1110 	if (mapped == 1)
1111 		inp = in_pcblookup_hash(&tcbinfo,
1112 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1113 			addrs[1].sin6_port,
1114 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1115 			addrs[0].sin6_port,
1116 			0, NULL);
1117 	else
1118 		inp = in6_pcblookup_hash(&tcbinfo,
1119 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1120 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1121 	if (inp == NULL) {
1122 		error = ENOENT;
1123 		goto outunlocked;
1124 	}
1125 	INP_LOCK(inp);
1126 	if (inp->inp_socket == NULL) {
1127 		error = ENOENT;
1128 		goto out;
1129 	}
1130 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1131 	if (error)
1132 		goto out;
1133 	cru2x(inp->inp_socket->so_cred, &xuc);
1134 out:
1135 	INP_UNLOCK(inp);
1136 outunlocked:
1137 	INP_INFO_RUNLOCK(&tcbinfo);
1138 	if (error == 0)
1139 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1140 	return (error);
1141 }
1142 
1143 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1144     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1145     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1146 #endif
1147 
1148 
1149 void
1150 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1151 {
1152 	struct ip *ip = vip;
1153 	struct tcphdr *th;
1154 	struct in_addr faddr;
1155 	struct inpcb *inp;
1156 	struct tcpcb *tp;
1157 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1158 	struct icmp *icp;
1159 	struct in_conninfo inc;
1160 	tcp_seq icmp_tcp_seq;
1161 	int mtu;
1162 
1163 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1164 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1165 		return;
1166 
1167 	if (cmd == PRC_MSGSIZE)
1168 		notify = tcp_mtudisc;
1169 	else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1170 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1171 		notify = tcp_drop_syn_sent;
1172 	/*
1173 	 * Redirects don't need to be handled up here.
1174 	 */
1175 	else if (PRC_IS_REDIRECT(cmd))
1176 		return;
1177 	/*
1178 	 * Source quench is depreciated.
1179 	 */
1180 	else if (cmd == PRC_QUENCH)
1181 		return;
1182 	/*
1183 	 * Hostdead is ugly because it goes linearly through all PCBs.
1184 	 * XXX: We never get this from ICMP, otherwise it makes an
1185 	 * excellent DoS attack on machines with many connections.
1186 	 */
1187 	else if (cmd == PRC_HOSTDEAD)
1188 		ip = NULL;
1189 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1190 		return;
1191 	if (ip != NULL) {
1192 		icp = (struct icmp *)((caddr_t)ip
1193 				      - offsetof(struct icmp, icmp_ip));
1194 		th = (struct tcphdr *)((caddr_t)ip
1195 				       + (ip->ip_hl << 2));
1196 		INP_INFO_WLOCK(&tcbinfo);
1197 		inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1198 		    ip->ip_src, th->th_sport, 0, NULL);
1199 		if (inp != NULL)  {
1200 			INP_LOCK(inp);
1201 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1202 			    !(inp->inp_vflag & INP_DROPPED) &&
1203 			    !(inp->inp_socket == NULL)) {
1204 				icmp_tcp_seq = htonl(th->th_seq);
1205 				tp = intotcpcb(inp);
1206 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1207 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1208 					if (cmd == PRC_MSGSIZE) {
1209 					    /*
1210 					     * MTU discovery:
1211 					     * If we got a needfrag set the MTU
1212 					     * in the route to the suggested new
1213 					     * value (if given) and then notify.
1214 					     */
1215 					    bzero(&inc, sizeof(inc));
1216 					    inc.inc_flags = 0;	/* IPv4 */
1217 					    inc.inc_faddr = faddr;
1218 
1219 					    mtu = ntohs(icp->icmp_nextmtu);
1220 					    /*
1221 					     * If no alternative MTU was
1222 					     * proposed, try the next smaller
1223 					     * one.  ip->ip_len has already
1224 					     * been swapped in icmp_input().
1225 					     */
1226 					    if (!mtu)
1227 						mtu = ip_next_mtu(ip->ip_len,
1228 						 1);
1229 					    if (mtu < max(296, (tcp_minmss)
1230 						 + sizeof(struct tcpiphdr)))
1231 						mtu = 0;
1232 					    if (!mtu)
1233 						mtu = tcp_mssdflt
1234 						 + sizeof(struct tcpiphdr);
1235 					    /*
1236 					     * Only cache the the MTU if it
1237 					     * is smaller than the interface
1238 					     * or route MTU.  tcp_mtudisc()
1239 					     * will do right thing by itself.
1240 					     */
1241 					    if (mtu <= tcp_maxmtu(&inc))
1242 						tcp_hc_updatemtu(&inc, mtu);
1243 					}
1244 
1245 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1246 				}
1247 			}
1248 			if (inp != NULL)
1249 				INP_UNLOCK(inp);
1250 		} else {
1251 			inc.inc_fport = th->th_dport;
1252 			inc.inc_lport = th->th_sport;
1253 			inc.inc_faddr = faddr;
1254 			inc.inc_laddr = ip->ip_src;
1255 #ifdef INET6
1256 			inc.inc_isipv6 = 0;
1257 #endif
1258 			syncache_unreach(&inc, th);
1259 		}
1260 		INP_INFO_WUNLOCK(&tcbinfo);
1261 	} else
1262 		in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1263 }
1264 
1265 #ifdef INET6
1266 void
1267 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1268 {
1269 	struct tcphdr th;
1270 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1271 	struct ip6_hdr *ip6;
1272 	struct mbuf *m;
1273 	struct ip6ctlparam *ip6cp = NULL;
1274 	const struct sockaddr_in6 *sa6_src = NULL;
1275 	int off;
1276 	struct tcp_portonly {
1277 		u_int16_t th_sport;
1278 		u_int16_t th_dport;
1279 	} *thp;
1280 
1281 	if (sa->sa_family != AF_INET6 ||
1282 	    sa->sa_len != sizeof(struct sockaddr_in6))
1283 		return;
1284 
1285 	if (cmd == PRC_MSGSIZE)
1286 		notify = tcp_mtudisc;
1287 	else if (!PRC_IS_REDIRECT(cmd) &&
1288 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1289 		return;
1290 	/* Source quench is depreciated. */
1291 	else if (cmd == PRC_QUENCH)
1292 		return;
1293 
1294 	/* if the parameter is from icmp6, decode it. */
1295 	if (d != NULL) {
1296 		ip6cp = (struct ip6ctlparam *)d;
1297 		m = ip6cp->ip6c_m;
1298 		ip6 = ip6cp->ip6c_ip6;
1299 		off = ip6cp->ip6c_off;
1300 		sa6_src = ip6cp->ip6c_src;
1301 	} else {
1302 		m = NULL;
1303 		ip6 = NULL;
1304 		off = 0;	/* fool gcc */
1305 		sa6_src = &sa6_any;
1306 	}
1307 
1308 	if (ip6 != NULL) {
1309 		struct in_conninfo inc;
1310 		/*
1311 		 * XXX: We assume that when IPV6 is non NULL,
1312 		 * M and OFF are valid.
1313 		 */
1314 
1315 		/* check if we can safely examine src and dst ports */
1316 		if (m->m_pkthdr.len < off + sizeof(*thp))
1317 			return;
1318 
1319 		bzero(&th, sizeof(th));
1320 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1321 
1322 		in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1323 		    (struct sockaddr *)ip6cp->ip6c_src,
1324 		    th.th_sport, cmd, NULL, notify);
1325 
1326 		inc.inc_fport = th.th_dport;
1327 		inc.inc_lport = th.th_sport;
1328 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1329 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1330 		inc.inc_isipv6 = 1;
1331 		INP_INFO_WLOCK(&tcbinfo);
1332 		syncache_unreach(&inc, &th);
1333 		INP_INFO_WUNLOCK(&tcbinfo);
1334 	} else
1335 		in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1336 			      0, cmd, NULL, notify);
1337 }
1338 #endif /* INET6 */
1339 
1340 
1341 /*
1342  * Following is where TCP initial sequence number generation occurs.
1343  *
1344  * There are two places where we must use initial sequence numbers:
1345  * 1.  In SYN-ACK packets.
1346  * 2.  In SYN packets.
1347  *
1348  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1349  * tcp_syncache.c for details.
1350  *
1351  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1352  * depends on this property.  In addition, these ISNs should be
1353  * unguessable so as to prevent connection hijacking.  To satisfy
1354  * the requirements of this situation, the algorithm outlined in
1355  * RFC 1948 is used, with only small modifications.
1356  *
1357  * Implementation details:
1358  *
1359  * Time is based off the system timer, and is corrected so that it
1360  * increases by one megabyte per second.  This allows for proper
1361  * recycling on high speed LANs while still leaving over an hour
1362  * before rollover.
1363  *
1364  * As reading the *exact* system time is too expensive to be done
1365  * whenever setting up a TCP connection, we increment the time
1366  * offset in two ways.  First, a small random positive increment
1367  * is added to isn_offset for each connection that is set up.
1368  * Second, the function tcp_isn_tick fires once per clock tick
1369  * and increments isn_offset as necessary so that sequence numbers
1370  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1371  * random positive increments serve only to ensure that the same
1372  * exact sequence number is never sent out twice (as could otherwise
1373  * happen when a port is recycled in less than the system tick
1374  * interval.)
1375  *
1376  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1377  * between seeding of isn_secret.  This is normally set to zero,
1378  * as reseeding should not be necessary.
1379  *
1380  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1381  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1382  * general, this means holding an exclusive (write) lock.
1383  */
1384 
1385 #define ISN_BYTES_PER_SECOND 1048576
1386 #define ISN_STATIC_INCREMENT 4096
1387 #define ISN_RANDOM_INCREMENT (4096 - 1)
1388 
1389 static u_char isn_secret[32];
1390 static int isn_last_reseed;
1391 static u_int32_t isn_offset, isn_offset_old;
1392 static MD5_CTX isn_ctx;
1393 
1394 tcp_seq
1395 tcp_new_isn(struct tcpcb *tp)
1396 {
1397 	u_int32_t md5_buffer[4];
1398 	tcp_seq new_isn;
1399 
1400 	INP_LOCK_ASSERT(tp->t_inpcb);
1401 
1402 	ISN_LOCK();
1403 	/* Seed if this is the first use, reseed if requested. */
1404 	if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1405 	     (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1406 		< (u_int)ticks))) {
1407 		read_random(&isn_secret, sizeof(isn_secret));
1408 		isn_last_reseed = ticks;
1409 	}
1410 
1411 	/* Compute the md5 hash and return the ISN. */
1412 	MD5Init(&isn_ctx);
1413 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1414 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1415 #ifdef INET6
1416 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1417 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1418 			  sizeof(struct in6_addr));
1419 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1420 			  sizeof(struct in6_addr));
1421 	} else
1422 #endif
1423 	{
1424 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1425 			  sizeof(struct in_addr));
1426 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1427 			  sizeof(struct in_addr));
1428 	}
1429 	MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1430 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1431 	new_isn = (tcp_seq) md5_buffer[0];
1432 	isn_offset += ISN_STATIC_INCREMENT +
1433 		(arc4random() & ISN_RANDOM_INCREMENT);
1434 	new_isn += isn_offset;
1435 	ISN_UNLOCK();
1436 	return (new_isn);
1437 }
1438 
1439 /*
1440  * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1441  * to keep time flowing at a relatively constant rate.  If the random
1442  * increments have already pushed us past the projected offset, do nothing.
1443  */
1444 static void
1445 tcp_isn_tick(void *xtp)
1446 {
1447 	u_int32_t projected_offset;
1448 
1449 	ISN_LOCK();
1450 	projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1451 
1452 	if (projected_offset > isn_offset)
1453 		isn_offset = projected_offset;
1454 
1455 	isn_offset_old = isn_offset;
1456 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1457 	ISN_UNLOCK();
1458 }
1459 
1460 /*
1461  * When a specific ICMP unreachable message is received and the
1462  * connection state is SYN-SENT, drop the connection.  This behavior
1463  * is controlled by the icmp_may_rst sysctl.
1464  */
1465 struct inpcb *
1466 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1467 {
1468 	struct tcpcb *tp;
1469 
1470 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
1471 	INP_LOCK_ASSERT(inp);
1472 
1473 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1474 	    (inp->inp_vflag & INP_DROPPED))
1475 		return (inp);
1476 
1477 	tp = intotcpcb(inp);
1478 	if (tp->t_state != TCPS_SYN_SENT)
1479 		return (inp);
1480 
1481 	tp = tcp_drop(tp, errno);
1482 	if (tp != NULL)
1483 		return (inp);
1484 	else
1485 		return (NULL);
1486 }
1487 
1488 /*
1489  * When `need fragmentation' ICMP is received, update our idea of the MSS
1490  * based on the new value in the route.  Also nudge TCP to send something,
1491  * since we know the packet we just sent was dropped.
1492  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1493  */
1494 struct inpcb *
1495 tcp_mtudisc(struct inpcb *inp, int errno)
1496 {
1497 	struct tcpcb *tp;
1498 	struct socket *so = inp->inp_socket;
1499 	u_int maxmtu;
1500 	u_int romtu;
1501 	int mss;
1502 #ifdef INET6
1503 	int isipv6;
1504 #endif /* INET6 */
1505 
1506 	INP_LOCK_ASSERT(inp);
1507 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1508 	    (inp->inp_vflag & INP_DROPPED))
1509 		return (inp);
1510 
1511 	tp = intotcpcb(inp);
1512 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1513 
1514 #ifdef INET6
1515 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1516 #endif
1517 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1518 	romtu =
1519 #ifdef INET6
1520 	    isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1521 #endif /* INET6 */
1522 	    tcp_maxmtu(&inp->inp_inc);
1523 	if (!maxmtu)
1524 		maxmtu = romtu;
1525 	else
1526 		maxmtu = min(maxmtu, romtu);
1527 	if (!maxmtu) {
1528 		tp->t_maxopd = tp->t_maxseg =
1529 #ifdef INET6
1530 			isipv6 ? tcp_v6mssdflt :
1531 #endif /* INET6 */
1532 			tcp_mssdflt;
1533 		return (inp);
1534 	}
1535 	mss = maxmtu -
1536 #ifdef INET6
1537 		(isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1538 #endif /* INET6 */
1539 		 sizeof(struct tcpiphdr)
1540 #ifdef INET6
1541 		 )
1542 #endif /* INET6 */
1543 		;
1544 
1545 	/*
1546 	 * XXX - The above conditional probably violates the TCP
1547 	 * spec.  The problem is that, since we don't know the
1548 	 * other end's MSS, we are supposed to use a conservative
1549 	 * default.  But, if we do that, then MTU discovery will
1550 	 * never actually take place, because the conservative
1551 	 * default is much less than the MTUs typically seen
1552 	 * on the Internet today.  For the moment, we'll sweep
1553 	 * this under the carpet.
1554 	 *
1555 	 * The conservative default might not actually be a problem
1556 	 * if the only case this occurs is when sending an initial
1557 	 * SYN with options and data to a host we've never talked
1558 	 * to before.  Then, they will reply with an MSS value which
1559 	 * will get recorded and the new parameters should get
1560 	 * recomputed.  For Further Study.
1561 	 */
1562 	if (tp->t_maxopd <= mss)
1563 		return (inp);
1564 	tp->t_maxopd = mss;
1565 
1566 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1567 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1568 		mss -= TCPOLEN_TSTAMP_APPA;
1569 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1570 	if (mss > MCLBYTES)
1571 		mss &= ~(MCLBYTES-1);
1572 #else
1573 	if (mss > MCLBYTES)
1574 		mss = mss / MCLBYTES * MCLBYTES;
1575 #endif
1576 	if (so->so_snd.sb_hiwat < mss)
1577 		mss = so->so_snd.sb_hiwat;
1578 
1579 	tp->t_maxseg = mss;
1580 
1581 	tcpstat.tcps_mturesent++;
1582 	tp->t_rtttime = 0;
1583 	tp->snd_nxt = tp->snd_una;
1584 	tcp_output(tp);
1585 	return (inp);
1586 }
1587 
1588 /*
1589  * Look-up the routing entry to the peer of this inpcb.  If no route
1590  * is found and it cannot be allocated, then return NULL.  This routine
1591  * is called by TCP routines that access the rmx structure and by tcp_mss
1592  * to get the interface MTU.
1593  */
1594 u_long
1595 tcp_maxmtu(struct in_conninfo *inc)
1596 {
1597 	struct route sro;
1598 	struct sockaddr_in *dst;
1599 	struct ifnet *ifp;
1600 	u_long maxmtu = 0;
1601 
1602 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1603 
1604 	bzero(&sro, sizeof(sro));
1605 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1606 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1607 		dst->sin_family = AF_INET;
1608 		dst->sin_len = sizeof(*dst);
1609 		dst->sin_addr = inc->inc_faddr;
1610 		rtalloc_ign(&sro, RTF_CLONING);
1611 	}
1612 	if (sro.ro_rt != NULL) {
1613 		ifp = sro.ro_rt->rt_ifp;
1614 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1615 			maxmtu = ifp->if_mtu;
1616 		else
1617 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1618 		RTFREE(sro.ro_rt);
1619 	}
1620 	return (maxmtu);
1621 }
1622 
1623 #ifdef INET6
1624 u_long
1625 tcp_maxmtu6(struct in_conninfo *inc)
1626 {
1627 	struct route_in6 sro6;
1628 	struct ifnet *ifp;
1629 	u_long maxmtu = 0;
1630 
1631 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1632 
1633 	bzero(&sro6, sizeof(sro6));
1634 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1635 		sro6.ro_dst.sin6_family = AF_INET6;
1636 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1637 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1638 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1639 	}
1640 	if (sro6.ro_rt != NULL) {
1641 		ifp = sro6.ro_rt->rt_ifp;
1642 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1643 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1644 		else
1645 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1646 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1647 		RTFREE(sro6.ro_rt);
1648 	}
1649 
1650 	return (maxmtu);
1651 }
1652 #endif /* INET6 */
1653 
1654 #ifdef IPSEC
1655 /* compute ESP/AH header size for TCP, including outer IP header. */
1656 size_t
1657 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1658 {
1659 	struct inpcb *inp;
1660 	struct mbuf *m;
1661 	size_t hdrsiz;
1662 	struct ip *ip;
1663 #ifdef INET6
1664 	struct ip6_hdr *ip6;
1665 #endif
1666 	struct tcphdr *th;
1667 
1668 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1669 		return (0);
1670 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1671 	if (!m)
1672 		return (0);
1673 
1674 #ifdef INET6
1675 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1676 		ip6 = mtod(m, struct ip6_hdr *);
1677 		th = (struct tcphdr *)(ip6 + 1);
1678 		m->m_pkthdr.len = m->m_len =
1679 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1680 		tcpip_fillheaders(inp, ip6, th);
1681 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1682 	} else
1683 #endif /* INET6 */
1684 	{
1685 		ip = mtod(m, struct ip *);
1686 		th = (struct tcphdr *)(ip + 1);
1687 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1688 		tcpip_fillheaders(inp, ip, th);
1689 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1690 	}
1691 
1692 	m_free(m);
1693 	return (hdrsiz);
1694 }
1695 #endif /*IPSEC*/
1696 
1697 /*
1698  * Move a TCP connection into TIME_WAIT state.
1699  *    tcbinfo is locked.
1700  *    inp is locked, and is unlocked before returning.
1701  */
1702 void
1703 tcp_twstart(struct tcpcb *tp)
1704 {
1705 	struct tcptw *tw;
1706 	struct inpcb *inp;
1707 	int tw_time, acknow;
1708 	struct socket *so;
1709 
1710 	INP_INFO_WLOCK_ASSERT(&tcbinfo);	/* tcp_timer_2msl_reset(). */
1711 	INP_LOCK_ASSERT(tp->t_inpcb);
1712 
1713 	tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1714 	if (tw == NULL) {
1715 		tw = tcp_timer_2msl_tw(1);
1716 		if (tw == NULL) {
1717 			tp = tcp_close(tp);
1718 			if (tp != NULL)
1719 				INP_UNLOCK(tp->t_inpcb);
1720 			return;
1721 		}
1722 	}
1723 	inp = tp->t_inpcb;
1724 	tw->tw_inpcb = inp;
1725 
1726 	/*
1727 	 * Recover last window size sent.
1728 	 */
1729 	tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1730 
1731 	/*
1732 	 * Set t_recent if timestamps are used on the connection.
1733 	 */
1734 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1735 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1736 		tw->t_recent = tp->ts_recent;
1737 	else
1738 		tw->t_recent = 0;
1739 
1740 	tw->snd_nxt = tp->snd_nxt;
1741 	tw->rcv_nxt = tp->rcv_nxt;
1742 	tw->iss     = tp->iss;
1743 	tw->irs     = tp->irs;
1744 	tw->t_starttime = tp->t_starttime;
1745 	tw->tw_time = 0;
1746 
1747 /* XXX
1748  * If this code will
1749  * be used for fin-wait-2 state also, then we may need
1750  * a ts_recent from the last segment.
1751  */
1752 	tw_time = 2 * tcp_msl;
1753 	acknow = tp->t_flags & TF_ACKNOW;
1754 
1755 	/*
1756 	 * First, discard tcpcb state, which includes stopping its timers and
1757 	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
1758 	 * that work is now done in the caller.
1759 	 */
1760 	tcp_discardcb(tp);
1761 	so = inp->inp_socket;
1762 	SOCK_LOCK(so);
1763 	tw->tw_cred = crhold(so->so_cred);
1764 	tw->tw_so_options = so->so_options;
1765 	SOCK_UNLOCK(so);
1766 	if (acknow)
1767 		tcp_twrespond(tw, TH_ACK);
1768 	inp->inp_ppcb = tw;
1769 	inp->inp_vflag |= INP_TIMEWAIT;
1770 	tcp_timer_2msl_reset(tw, tw_time);
1771 
1772 	/*
1773 	 * If the inpcb owns the sole reference to the socket, then we can
1774 	 * detach and free the socket as it is not needed in time wait.
1775 	 */
1776 	if (inp->inp_vflag & INP_SOCKREF) {
1777 		KASSERT(so->so_state & SS_PROTOREF,
1778 		    ("tcp_twstart: !SS_PROTOREF"));
1779 		inp->inp_vflag &= ~INP_SOCKREF;
1780 #ifdef INET6
1781 		if (inp->inp_vflag & INP_IPV6PROTO)
1782 			in6_pcbdetach(inp);
1783 		else
1784 #endif
1785 			in_pcbdetach(inp);
1786 		INP_UNLOCK(inp);
1787 		ACCEPT_LOCK();
1788 		SOCK_LOCK(so);
1789 		so->so_state &= ~SS_PROTOREF;
1790 		sofree(so);
1791 	} else
1792 		INP_UNLOCK(inp);
1793 }
1794 
1795 /*
1796  * The appromixate rate of ISN increase of Microsoft TCP stacks;
1797  * the actual rate is slightly higher due to the addition of
1798  * random positive increments.
1799  *
1800  * Most other new OSes use semi-randomized ISN values, so we
1801  * do not need to worry about them.
1802  */
1803 #define MS_ISN_BYTES_PER_SECOND		250000
1804 
1805 /*
1806  * Determine if the ISN we will generate has advanced beyond the last
1807  * sequence number used by the previous connection.  If so, indicate
1808  * that it is safe to recycle this tw socket by returning 1.
1809  *
1810  * XXXRW: This function should assert the inpcb lock as it does multiple
1811  * non-atomic reads from the tcptw, but is currently called without it from
1812  * in_pcb.c:in_pcblookup_local().
1813  */
1814 int
1815 tcp_twrecycleable(struct tcptw *tw)
1816 {
1817 	tcp_seq new_iss = tw->iss;
1818 	tcp_seq new_irs = tw->irs;
1819 
1820 	new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1821 	new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1822 
1823 	if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1824 		return (1);
1825 	else
1826 		return (0);
1827 }
1828 
1829 void
1830 tcp_twclose(struct tcptw *tw, int reuse)
1831 {
1832 	struct socket *so;
1833 	struct inpcb *inp;
1834 
1835 	/*
1836 	 * At this point, we are in one of two situations:
1837 	 *
1838 	 * (1) We have no socket, just an inpcb<->twtcp pair.  Release it all
1839 	 * after validating.
1840 	 *
1841 	 * (2) We have a socket, which we may or may now own the reference
1842 	 * for.  If we own the reference, release all the state after
1843 	 * validating.  If not, leave it for the socket close to clean up.
1844 	 */
1845 	inp = tw->tw_inpcb;
1846 	KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1847 	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1848 	INP_INFO_WLOCK_ASSERT(&tcbinfo);	/* tcp_timer_2msl_stop(). */
1849 	INP_LOCK_ASSERT(inp);
1850 
1851 	tw->tw_inpcb = NULL;
1852 	tcp_timer_2msl_stop(tw);
1853 	inp->inp_ppcb = NULL;
1854 	in_pcbdrop(inp);
1855 
1856 	so = inp->inp_socket;
1857 	if (so != NULL) {
1858 		if (inp->inp_vflag & INP_SOCKREF) {
1859 			/*
1860 			 * If a socket is present, and we own the only
1861 			 * reference, we need to tear down the socket and the
1862 			 * inpcb.
1863 			 */
1864 			inp->inp_vflag &= ~INP_SOCKREF;
1865 #ifdef INET6
1866 			if (inp->inp_vflag & INP_IPV6PROTO) {
1867 				in6_pcbdetach(inp);
1868 				in6_pcbfree(inp);
1869 			} else {
1870 				in_pcbdetach(inp);
1871 				in_pcbfree(inp);
1872 			}
1873 #endif
1874 			ACCEPT_LOCK();
1875 			SOCK_LOCK(so);
1876 			KASSERT(so->so_state & SS_PROTOREF,
1877 			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1878 			so->so_state &= ~SS_PROTOREF;
1879 			sofree(so);
1880 		} else {
1881 			/*
1882 			 * If we don't own the only reference, the socket and
1883 			 * inpcb need to be left around to be handled by
1884 			 * tcp_usr_detach() later.
1885 			 */
1886 			INP_UNLOCK(inp);
1887 		}
1888 	} else {
1889 #ifdef INET6
1890 		if (inp->inp_vflag & INP_IPV6PROTO)
1891 			in6_pcbfree(inp);
1892 		else
1893 #endif
1894 			in_pcbfree(inp);
1895 	}
1896 	tcpstat.tcps_closed++;
1897 	crfree(tw->tw_cred);
1898 	tw->tw_cred = NULL;
1899 	if (reuse)
1900 		return;
1901 	uma_zfree(tcptw_zone, tw);
1902 }
1903 
1904 int
1905 tcp_twrespond(struct tcptw *tw, int flags)
1906 {
1907 	struct inpcb *inp = tw->tw_inpcb;
1908 	struct tcphdr *th;
1909 	struct mbuf *m;
1910 	struct ip *ip = NULL;
1911 	u_int8_t *optp;
1912 	u_int hdrlen, optlen;
1913 	int error;
1914 #ifdef INET6
1915 	struct ip6_hdr *ip6 = NULL;
1916 	int isipv6 = inp->inp_inc.inc_isipv6;
1917 #endif
1918 
1919 	INP_LOCK_ASSERT(inp);
1920 
1921 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1922 	if (m == NULL)
1923 		return (ENOBUFS);
1924 	m->m_data += max_linkhdr;
1925 
1926 #ifdef MAC
1927 	mac_create_mbuf_from_inpcb(inp, m);
1928 #endif
1929 
1930 #ifdef INET6
1931 	if (isipv6) {
1932 		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1933 		ip6 = mtod(m, struct ip6_hdr *);
1934 		th = (struct tcphdr *)(ip6 + 1);
1935 		tcpip_fillheaders(inp, ip6, th);
1936 	} else
1937 #endif
1938 	{
1939 		hdrlen = sizeof(struct tcpiphdr);
1940 		ip = mtod(m, struct ip *);
1941 		th = (struct tcphdr *)(ip + 1);
1942 		tcpip_fillheaders(inp, ip, th);
1943 	}
1944 	optp = (u_int8_t *)(th + 1);
1945 
1946 	/*
1947 	 * Send a timestamp and echo-reply if both our side and our peer
1948 	 * have sent timestamps in our SYN's and this is not a RST.
1949 	 */
1950 	if (tw->t_recent && flags == TH_ACK) {
1951 		u_int32_t *lp = (u_int32_t *)optp;
1952 
1953 		/* Form timestamp option as shown in appendix A of RFC 1323. */
1954 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
1955 		*lp++ = htonl(ticks);
1956 		*lp   = htonl(tw->t_recent);
1957 		optp += TCPOLEN_TSTAMP_APPA;
1958 	}
1959 
1960 	optlen = optp - (u_int8_t *)(th + 1);
1961 
1962 	m->m_len = hdrlen + optlen;
1963 	m->m_pkthdr.len = m->m_len;
1964 
1965 	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1966 
1967 	th->th_seq = htonl(tw->snd_nxt);
1968 	th->th_ack = htonl(tw->rcv_nxt);
1969 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1970 	th->th_flags = flags;
1971 	th->th_win = htons(tw->last_win);
1972 
1973 #ifdef INET6
1974 	if (isipv6) {
1975 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1976 		    sizeof(struct tcphdr) + optlen);
1977 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1978 		error = ip6_output(m, inp->in6p_outputopts, NULL,
1979 		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1980 	} else
1981 #endif
1982 	{
1983 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1984 		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1985 		m->m_pkthdr.csum_flags = CSUM_TCP;
1986 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1987 		ip->ip_len = m->m_pkthdr.len;
1988 		if (path_mtu_discovery)
1989 			ip->ip_off |= IP_DF;
1990 		error = ip_output(m, inp->inp_options, NULL,
1991 		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1992 		    NULL, inp);
1993 	}
1994 	if (flags & TH_ACK)
1995 		tcpstat.tcps_sndacks++;
1996 	else
1997 		tcpstat.tcps_sndctrl++;
1998 	tcpstat.tcps_sndtotal++;
1999 	return (error);
2000 }
2001 
2002 /*
2003  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
2004  *
2005  * This code attempts to calculate the bandwidth-delay product as a
2006  * means of determining the optimal window size to maximize bandwidth,
2007  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
2008  * routers.  This code also does a fairly good job keeping RTTs in check
2009  * across slow links like modems.  We implement an algorithm which is very
2010  * similar (but not meant to be) TCP/Vegas.  The code operates on the
2011  * transmitter side of a TCP connection and so only effects the transmit
2012  * side of the connection.
2013  *
2014  * BACKGROUND:  TCP makes no provision for the management of buffer space
2015  * at the end points or at the intermediate routers and switches.  A TCP
2016  * stream, whether using NewReno or not, will eventually buffer as
2017  * many packets as it is able and the only reason this typically works is
2018  * due to the fairly small default buffers made available for a connection
2019  * (typicaly 16K or 32K).  As machines use larger windows and/or window
2020  * scaling it is now fairly easy for even a single TCP connection to blow-out
2021  * all available buffer space not only on the local interface, but on
2022  * intermediate routers and switches as well.  NewReno makes a misguided
2023  * attempt to 'solve' this problem by waiting for an actual failure to occur,
2024  * then backing off, then steadily increasing the window again until another
2025  * failure occurs, ad-infinitum.  This results in terrible oscillation that
2026  * is only made worse as network loads increase and the idea of intentionally
2027  * blowing out network buffers is, frankly, a terrible way to manage network
2028  * resources.
2029  *
2030  * It is far better to limit the transmit window prior to the failure
2031  * condition being achieved.  There are two general ways to do this:  First
2032  * you can 'scan' through different transmit window sizes and locate the
2033  * point where the RTT stops increasing, indicating that you have filled the
2034  * pipe, then scan backwards until you note that RTT stops decreasing, then
2035  * repeat ad-infinitum.  This method works in principle but has severe
2036  * implementation issues due to RTT variances, timer granularity, and
2037  * instability in the algorithm which can lead to many false positives and
2038  * create oscillations as well as interact badly with other TCP streams
2039  * implementing the same algorithm.
2040  *
2041  * The second method is to limit the window to the bandwidth delay product
2042  * of the link.  This is the method we implement.  RTT variances and our
2043  * own manipulation of the congestion window, bwnd, can potentially
2044  * destabilize the algorithm.  For this reason we have to stabilize the
2045  * elements used to calculate the window.  We do this by using the minimum
2046  * observed RTT, the long term average of the observed bandwidth, and
2047  * by adding two segments worth of slop.  It isn't perfect but it is able
2048  * to react to changing conditions and gives us a very stable basis on
2049  * which to extend the algorithm.
2050  */
2051 void
2052 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2053 {
2054 	u_long bw;
2055 	u_long bwnd;
2056 	int save_ticks;
2057 
2058 	INP_LOCK_ASSERT(tp->t_inpcb);
2059 
2060 	/*
2061 	 * If inflight_enable is disabled in the middle of a tcp connection,
2062 	 * make sure snd_bwnd is effectively disabled.
2063 	 */
2064 	if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2065 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2066 		tp->snd_bandwidth = 0;
2067 		return;
2068 	}
2069 
2070 	/*
2071 	 * Figure out the bandwidth.  Due to the tick granularity this
2072 	 * is a very rough number and it MUST be averaged over a fairly
2073 	 * long period of time.  XXX we need to take into account a link
2074 	 * that is not using all available bandwidth, but for now our
2075 	 * slop will ramp us up if this case occurs and the bandwidth later
2076 	 * increases.
2077 	 *
2078 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
2079 	 * effectively reset t_bw_rtttime for this case.
2080 	 */
2081 	save_ticks = ticks;
2082 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2083 		return;
2084 
2085 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2086 	    (save_ticks - tp->t_bw_rtttime);
2087 	tp->t_bw_rtttime = save_ticks;
2088 	tp->t_bw_rtseq = ack_seq;
2089 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2090 		return;
2091 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2092 
2093 	tp->snd_bandwidth = bw;
2094 
2095 	/*
2096 	 * Calculate the semi-static bandwidth delay product, plus two maximal
2097 	 * segments.  The additional slop puts us squarely in the sweet
2098 	 * spot and also handles the bandwidth run-up case and stabilization.
2099 	 * Without the slop we could be locking ourselves into a lower
2100 	 * bandwidth.
2101 	 *
2102 	 * Situations Handled:
2103 	 *	(1) Prevents over-queueing of packets on LANs, especially on
2104 	 *	    high speed LANs, allowing larger TCP buffers to be
2105 	 *	    specified, and also does a good job preventing
2106 	 *	    over-queueing of packets over choke points like modems
2107 	 *	    (at least for the transmit side).
2108 	 *
2109 	 *	(2) Is able to handle changing network loads (bandwidth
2110 	 *	    drops so bwnd drops, bandwidth increases so bwnd
2111 	 *	    increases).
2112 	 *
2113 	 *	(3) Theoretically should stabilize in the face of multiple
2114 	 *	    connections implementing the same algorithm (this may need
2115 	 *	    a little work).
2116 	 *
2117 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
2118 	 *	    be adjusted with a sysctl but typically only needs to be
2119 	 *	    on very slow connections.  A value no smaller then 5
2120 	 *	    should be used, but only reduce this default if you have
2121 	 *	    no other choice.
2122 	 */
2123 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
2124 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2125 #undef USERTT
2126 
2127 	if (tcp_inflight_debug > 0) {
2128 		static int ltime;
2129 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2130 			ltime = ticks;
2131 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2132 			    tp,
2133 			    bw,
2134 			    tp->t_rttbest,
2135 			    tp->t_srtt,
2136 			    bwnd
2137 			);
2138 		}
2139 	}
2140 	if ((long)bwnd < tcp_inflight_min)
2141 		bwnd = tcp_inflight_min;
2142 	if (bwnd > tcp_inflight_max)
2143 		bwnd = tcp_inflight_max;
2144 	if ((long)bwnd < tp->t_maxseg * 2)
2145 		bwnd = tp->t_maxseg * 2;
2146 	tp->snd_bwnd = bwnd;
2147 }
2148 
2149 #ifdef TCP_SIGNATURE
2150 /*
2151  * Callback function invoked by m_apply() to digest TCP segment data
2152  * contained within an mbuf chain.
2153  */
2154 static int
2155 tcp_signature_apply(void *fstate, void *data, u_int len)
2156 {
2157 
2158 	MD5Update(fstate, (u_char *)data, len);
2159 	return (0);
2160 }
2161 
2162 /*
2163  * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2164  *
2165  * Parameters:
2166  * m		pointer to head of mbuf chain
2167  * off0		offset to TCP header within the mbuf chain
2168  * len		length of TCP segment data, excluding options
2169  * optlen	length of TCP segment options
2170  * buf		pointer to storage for computed MD5 digest
2171  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2172  *
2173  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2174  * When called from tcp_input(), we can be sure that th_sum has been
2175  * zeroed out and verified already.
2176  *
2177  * This function is for IPv4 use only. Calling this function with an
2178  * IPv6 packet in the mbuf chain will yield undefined results.
2179  *
2180  * Return 0 if successful, otherwise return -1.
2181  *
2182  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2183  * search with the destination IP address, and a 'magic SPI' to be
2184  * determined by the application. This is hardcoded elsewhere to 1179
2185  * right now. Another branch of this code exists which uses the SPD to
2186  * specify per-application flows but it is unstable.
2187  */
2188 int
2189 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2190     u_char *buf, u_int direction)
2191 {
2192 	union sockaddr_union dst;
2193 	struct ippseudo ippseudo;
2194 	MD5_CTX ctx;
2195 	int doff;
2196 	struct ip *ip;
2197 	struct ipovly *ipovly;
2198 	struct secasvar *sav;
2199 	struct tcphdr *th;
2200 	u_short savecsum;
2201 
2202 	KASSERT(m != NULL, ("NULL mbuf chain"));
2203 	KASSERT(buf != NULL, ("NULL signature pointer"));
2204 
2205 	/* Extract the destination from the IP header in the mbuf. */
2206 	ip = mtod(m, struct ip *);
2207 	bzero(&dst, sizeof(union sockaddr_union));
2208 	dst.sa.sa_len = sizeof(struct sockaddr_in);
2209 	dst.sa.sa_family = AF_INET;
2210 	dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2211 	    ip->ip_src : ip->ip_dst;
2212 
2213 	/* Look up an SADB entry which matches the address of the peer. */
2214 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2215 	if (sav == NULL) {
2216 		printf("%s: SADB lookup failed for %s\n", __func__,
2217 		    inet_ntoa(dst.sin.sin_addr));
2218 		return (EINVAL);
2219 	}
2220 
2221 	MD5Init(&ctx);
2222 	ipovly = (struct ipovly *)ip;
2223 	th = (struct tcphdr *)((u_char *)ip + off0);
2224 	doff = off0 + sizeof(struct tcphdr) + optlen;
2225 
2226 	/*
2227 	 * Step 1: Update MD5 hash with IP pseudo-header.
2228 	 *
2229 	 * XXX The ippseudo header MUST be digested in network byte order,
2230 	 * or else we'll fail the regression test. Assume all fields we've
2231 	 * been doing arithmetic on have been in host byte order.
2232 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2233 	 * tcp_output(), the underlying ip_len member has not yet been set.
2234 	 */
2235 	ippseudo.ippseudo_src = ipovly->ih_src;
2236 	ippseudo.ippseudo_dst = ipovly->ih_dst;
2237 	ippseudo.ippseudo_pad = 0;
2238 	ippseudo.ippseudo_p = IPPROTO_TCP;
2239 	ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2240 	MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2241 
2242 	/*
2243 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2244 	 * The TCP checksum must be set to zero.
2245 	 */
2246 	savecsum = th->th_sum;
2247 	th->th_sum = 0;
2248 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2249 	th->th_sum = savecsum;
2250 
2251 	/*
2252 	 * Step 3: Update MD5 hash with TCP segment data.
2253 	 *         Use m_apply() to avoid an early m_pullup().
2254 	 */
2255 	if (len > 0)
2256 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2257 
2258 	/*
2259 	 * Step 4: Update MD5 hash with shared secret.
2260 	 */
2261 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2262 	MD5Final(buf, &ctx);
2263 
2264 	key_sa_recordxfer(sav, m);
2265 	KEY_FREESAV(&sav);
2266 	return (0);
2267 }
2268 #endif /* TCP_SIGNATURE */
2269 
2270 static int
2271 sysctl_drop(SYSCTL_HANDLER_ARGS)
2272 {
2273 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2274 	struct sockaddr_storage addrs[2];
2275 	struct inpcb *inp;
2276 	struct tcpcb *tp;
2277 	struct tcptw *tw;
2278 	struct sockaddr_in *fin, *lin;
2279 #ifdef INET6
2280 	struct sockaddr_in6 *fin6, *lin6;
2281 	struct in6_addr f6, l6;
2282 #endif
2283 	int error;
2284 
2285 	inp = NULL;
2286 	fin = lin = NULL;
2287 #ifdef INET6
2288 	fin6 = lin6 = NULL;
2289 #endif
2290 	error = 0;
2291 
2292 	if (req->oldptr != NULL || req->oldlen != 0)
2293 		return (EINVAL);
2294 	if (req->newptr == NULL)
2295 		return (EPERM);
2296 	if (req->newlen < sizeof(addrs))
2297 		return (ENOMEM);
2298 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2299 	if (error)
2300 		return (error);
2301 
2302 	switch (addrs[0].ss_family) {
2303 #ifdef INET6
2304 	case AF_INET6:
2305 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2306 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2307 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2308 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2309 			return (EINVAL);
2310 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2311 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2312 				return (EINVAL);
2313 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2314 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2315 			fin = (struct sockaddr_in *)&addrs[0];
2316 			lin = (struct sockaddr_in *)&addrs[1];
2317 			break;
2318 		}
2319 		error = sa6_embedscope(fin6, ip6_use_defzone);
2320 		if (error)
2321 			return (error);
2322 		error = sa6_embedscope(lin6, ip6_use_defzone);
2323 		if (error)
2324 			return (error);
2325 		break;
2326 #endif
2327 	case AF_INET:
2328 		fin = (struct sockaddr_in *)&addrs[0];
2329 		lin = (struct sockaddr_in *)&addrs[1];
2330 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2331 		    lin->sin_len != sizeof(struct sockaddr_in))
2332 			return (EINVAL);
2333 		break;
2334 	default:
2335 		return (EINVAL);
2336 	}
2337 	INP_INFO_WLOCK(&tcbinfo);
2338 	switch (addrs[0].ss_family) {
2339 #ifdef INET6
2340 	case AF_INET6:
2341 		inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2342 		    &l6, lin6->sin6_port, 0, NULL);
2343 		break;
2344 #endif
2345 	case AF_INET:
2346 		inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2347 		    lin->sin_addr, lin->sin_port, 0, NULL);
2348 		break;
2349 	}
2350 	if (inp != NULL) {
2351 		INP_LOCK(inp);
2352 		if (inp->inp_vflag & INP_TIMEWAIT) {
2353 			/*
2354 			 * XXXRW: There currently exists a state where an
2355 			 * inpcb is present, but its timewait state has been
2356 			 * discarded.  For now, don't allow dropping of this
2357 			 * type of inpcb.
2358 			 */
2359 			tw = intotw(inp);
2360 			if (tw != NULL)
2361 				tcp_twclose(tw, 0);
2362 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2363 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2364 			tp = intotcpcb(inp);
2365 			tcp_drop(tp, ECONNABORTED);
2366 		}
2367 		INP_UNLOCK(inp);
2368 	} else
2369 		error = ESRCH;
2370 	INP_INFO_WUNLOCK(&tcbinfo);
2371 	return (error);
2372 }
2373 
2374 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2375     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2376     0, sysctl_drop, "", "Drop TCP connection");
2377