xref: /freebsd/sys/netinet/tcp_subr.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
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/malloc.h>
47 #include <sys/mbuf.h>
48 #ifdef INET6
49 #include <sys/domain.h>
50 #endif
51 #include <sys/priv.h>
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 #include <security/mac/mac_framework.h>
117 
118 int	tcp_mssdflt = TCP_MSS;
119 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
120     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
121 
122 #ifdef INET6
123 int	tcp_v6mssdflt = TCP6_MSS;
124 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
125 	CTLFLAG_RW, &tcp_v6mssdflt , 0,
126 	"Default TCP Maximum Segment Size for IPv6");
127 #endif
128 
129 /*
130  * Minimum MSS we accept and use. This prevents DoS attacks where
131  * we are forced to a ridiculous low MSS like 20 and send hundreds
132  * of packets instead of one. The effect scales with the available
133  * bandwidth and quickly saturates the CPU and network interface
134  * with packet generation and sending. Set to zero to disable MINMSS
135  * checking. This setting prevents us from sending too small packets.
136  */
137 int	tcp_minmss = TCP_MINMSS;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
139     &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
140 /*
141  * Number of TCP segments per second we accept from remote host
142  * before we start to calculate average segment size. If average
143  * segment size drops below the minimum TCP MSS we assume a DoS
144  * attack and reset+drop the connection. Care has to be taken not to
145  * set this value too small to not kill interactive type connections
146  * (telnet, SSH) which send many small packets.
147  */
148 int     tcp_minmssoverload = TCP_MINMSSOVERLOAD;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
150     &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
151     "be under the MINMSS Size");
152 
153 #if 0
154 static int	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
155 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
156     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
157 #endif
158 
159 int	tcp_do_rfc1323 = 1;
160 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
161     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
162 
163 static int	tcp_tcbhashsize = 0;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
165      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
166 
167 static int	do_tcpdrain = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
169      "Enable tcp_drain routine for extra help when low on mbufs");
170 
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
172     &tcbinfo.ipi_count, 0, "Number of active PCBs");
173 
174 static int	icmp_may_rst = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
176     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
177 
178 static int	tcp_isn_reseed_interval = 0;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
180     &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
181 
182 static uma_zone_t tcptw_zone;
183 static int	maxtcptw;
184 
185 static int
186 tcptw_auto_size(void)
187 {
188 	int halfrange;
189 
190 	/*
191 	 * Max out at half the ephemeral port range so that TIME_WAIT
192 	 * sockets don't tie up too many ephemeral ports.
193 	 */
194 	if (ipport_lastauto > ipport_firstauto)
195 		halfrange = (ipport_lastauto - ipport_firstauto) / 2;
196 	else
197 		halfrange = (ipport_firstauto - ipport_lastauto) / 2;
198 	/* Protect against goofy port ranges smaller than 32. */
199 	return (imin(imax(halfrange, 32), maxsockets / 5));
200 }
201 
202 static int
203 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
204 {
205 	int error, new;
206 
207 	if (maxtcptw == 0)
208 		new = tcptw_auto_size();
209 	else
210 		new = maxtcptw;
211 	error = sysctl_handle_int(oidp, &new, sizeof(int), req);
212 	if (error == 0 && req->newptr)
213 		if (new >= 32) {
214 			maxtcptw = new;
215 			uma_zone_set_max(tcptw_zone, maxtcptw);
216 		}
217 	return (error);
218 }
219 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
220     &maxtcptw, 0, sysctl_maxtcptw, "IU",
221     "Maximum number of compressed TCP TIME_WAIT entries");
222 
223 static int	nolocaltimewait = 0;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
225     &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries"
226 			 "for local connections");
227 
228 /*
229  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
230  * 1024 exists only for debugging.  A good production default would be
231  * something like 6100.
232  */
233 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
234     "TCP inflight data limiting");
235 
236 static int	tcp_inflight_enable = 1;
237 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
238     &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
239 
240 static int	tcp_inflight_debug = 0;
241 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
242     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
243 
244 static int	tcp_inflight_rttthresh;
245 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
246     &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
247     "RTT threshold below which inflight will deactivate itself");
248 
249 static int	tcp_inflight_min = 6144;
250 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
251     &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
252 
253 static int	tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
254 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
255     &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
256 
257 static int	tcp_inflight_stab = 20;
258 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
259     &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
260 
261 uma_zone_t sack_hole_zone;
262 
263 static struct inpcb *tcp_notify(struct inpcb *, int);
264 static void	tcp_isn_tick(void *);
265 
266 /*
267  * Target size of TCP PCB hash tables. Must be a power of two.
268  *
269  * Note that this can be overridden by the kernel environment
270  * variable net.inet.tcp.tcbhashsize
271  */
272 #ifndef TCBHASHSIZE
273 #define TCBHASHSIZE	512
274 #endif
275 
276 /*
277  * XXX
278  * Callouts should be moved into struct tcp directly.  They are currently
279  * separate because the tcpcb structure is exported to userland for sysctl
280  * parsing purposes, which do not know about callouts.
281  */
282 struct	tcpcb_mem {
283 	struct	tcpcb tcb;
284 	struct	callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
285 	struct	callout tcpcb_mem_2msl, tcpcb_mem_delack;
286 };
287 
288 static uma_zone_t tcpcb_zone;
289 struct callout isn_callout;
290 static struct mtx isn_mtx;
291 
292 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
293 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
294 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
295 
296 /*
297  * TCP initialization.
298  */
299 static void
300 tcp_zone_change(void *tag)
301 {
302 
303 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
304 	uma_zone_set_max(tcpcb_zone, maxsockets);
305 	if (maxtcptw == 0)
306 		uma_zone_set_max(tcptw_zone, tcptw_auto_size());
307 }
308 
309 static int
310 tcp_inpcb_init(void *mem, int size, int flags)
311 {
312 	struct inpcb *inp = mem;
313 
314 	INP_LOCK_INIT(inp, "inp", "tcpinp");
315 	return (0);
316 }
317 
318 void
319 tcp_init(void)
320 {
321 	int hashsize = TCBHASHSIZE;
322 
323 	tcp_delacktime = TCPTV_DELACK;
324 	tcp_keepinit = TCPTV_KEEP_INIT;
325 	tcp_keepidle = TCPTV_KEEP_IDLE;
326 	tcp_keepintvl = TCPTV_KEEPINTVL;
327 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
328 	tcp_msl = TCPTV_MSL;
329 	tcp_rexmit_min = TCPTV_MIN;
330 	tcp_rexmit_slop = TCPTV_CPU_VAR;
331 	tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
332 
333 	INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
334 	LIST_INIT(&tcb);
335 	tcbinfo.listhead = &tcb;
336 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
337 	if (!powerof2(hashsize)) {
338 		printf("WARNING: TCB hash size not a power of 2\n");
339 		hashsize = 512; /* safe default */
340 	}
341 	tcp_tcbhashsize = hashsize;
342 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
343 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
344 					&tcbinfo.porthashmask);
345 	tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
346 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
347 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
348 #ifdef INET6
349 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
350 #else /* INET6 */
351 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
352 #endif /* INET6 */
353 	if (max_protohdr < TCP_MINPROTOHDR)
354 		max_protohdr = TCP_MINPROTOHDR;
355 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
356 		panic("tcp_init");
357 #undef TCP_MINPROTOHDR
358 	/*
359 	 * These have to be type stable for the benefit of the timers.
360 	 */
361 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
362 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
363 	uma_zone_set_max(tcpcb_zone, maxsockets);
364 	tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
365 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
366 	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
367 	if (maxtcptw == 0)
368 		uma_zone_set_max(tcptw_zone, tcptw_auto_size());
369 	else
370 		uma_zone_set_max(tcptw_zone, maxtcptw);
371 	tcp_timer_init();
372 	syncache_init();
373 	tcp_hc_init();
374 	tcp_reass_init();
375 	ISN_LOCK_INIT();
376 	callout_init(&isn_callout, CALLOUT_MPSAFE);
377 	tcp_isn_tick(NULL);
378 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
379 		SHUTDOWN_PRI_DEFAULT);
380 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
381 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
382 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
383 		EVENTHANDLER_PRI_ANY);
384 }
385 
386 void
387 tcp_fini(void *xtp)
388 {
389 
390 	callout_stop(&isn_callout);
391 }
392 
393 /*
394  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
395  * tcp_template used to store this data in mbufs, but we now recopy it out
396  * of the tcpcb each time to conserve mbufs.
397  */
398 void
399 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
400 {
401 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
402 
403 	INP_LOCK_ASSERT(inp);
404 
405 #ifdef INET6
406 	if ((inp->inp_vflag & INP_IPV6) != 0) {
407 		struct ip6_hdr *ip6;
408 
409 		ip6 = (struct ip6_hdr *)ip_ptr;
410 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
411 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
412 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
413 			(IPV6_VERSION & IPV6_VERSION_MASK);
414 		ip6->ip6_nxt = IPPROTO_TCP;
415 		ip6->ip6_plen = sizeof(struct tcphdr);
416 		ip6->ip6_src = inp->in6p_laddr;
417 		ip6->ip6_dst = inp->in6p_faddr;
418 	} else
419 #endif
420 	{
421 		struct ip *ip;
422 
423 		ip = (struct ip *)ip_ptr;
424 		ip->ip_v = IPVERSION;
425 		ip->ip_hl = 5;
426 		ip->ip_tos = inp->inp_ip_tos;
427 		ip->ip_len = 0;
428 		ip->ip_id = 0;
429 		ip->ip_off = 0;
430 		ip->ip_ttl = inp->inp_ip_ttl;
431 		ip->ip_sum = 0;
432 		ip->ip_p = IPPROTO_TCP;
433 		ip->ip_src = inp->inp_laddr;
434 		ip->ip_dst = inp->inp_faddr;
435 	}
436 	th->th_sport = inp->inp_lport;
437 	th->th_dport = inp->inp_fport;
438 	th->th_seq = 0;
439 	th->th_ack = 0;
440 	th->th_x2 = 0;
441 	th->th_off = 5;
442 	th->th_flags = 0;
443 	th->th_win = 0;
444 	th->th_urp = 0;
445 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
446 }
447 
448 /*
449  * Create template to be used to send tcp packets on a connection.
450  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
451  * use for this function is in keepalives, which use tcp_respond.
452  */
453 struct tcptemp *
454 tcpip_maketemplate(struct inpcb *inp)
455 {
456 	struct mbuf *m;
457 	struct tcptemp *n;
458 
459 	m = m_get(M_DONTWAIT, MT_DATA);
460 	if (m == NULL)
461 		return (0);
462 	m->m_len = sizeof(struct tcptemp);
463 	n = mtod(m, struct tcptemp *);
464 
465 	tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
466 	return (n);
467 }
468 
469 /*
470  * Send a single message to the TCP at address specified by
471  * the given TCP/IP header.  If m == NULL, then we make a copy
472  * of the tcpiphdr at ti and send directly to the addressed host.
473  * This is used to force keep alive messages out using the TCP
474  * template for a connection.  If flags are given then we send
475  * a message back to the TCP which originated the * segment ti,
476  * and discard the mbuf containing it and any other attached mbufs.
477  *
478  * In any case the ack and sequence number of the transmitted
479  * segment are as specified by the parameters.
480  *
481  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
482  */
483 void
484 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th,
485     register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags)
486 {
487 	register int tlen;
488 	int win = 0;
489 	struct ip *ip;
490 	struct tcphdr *nth;
491 #ifdef INET6
492 	struct ip6_hdr *ip6;
493 	int isipv6;
494 #endif /* INET6 */
495 	int ipflags = 0;
496 	struct inpcb *inp;
497 
498 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
499 
500 #ifdef INET6
501 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
502 	ip6 = ipgen;
503 #endif /* INET6 */
504 	ip = ipgen;
505 
506 	if (tp != NULL) {
507 		inp = tp->t_inpcb;
508 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
509 		INP_INFO_WLOCK_ASSERT(&tcbinfo);
510 		INP_LOCK_ASSERT(inp);
511 	} else
512 		inp = NULL;
513 
514 	if (tp != NULL) {
515 		if (!(flags & TH_RST)) {
516 			win = sbspace(&inp->inp_socket->so_rcv);
517 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
518 				win = (long)TCP_MAXWIN << tp->rcv_scale;
519 		}
520 	}
521 	if (m == NULL) {
522 		m = m_gethdr(M_DONTWAIT, MT_DATA);
523 		if (m == NULL)
524 			return;
525 		tlen = 0;
526 		m->m_data += max_linkhdr;
527 #ifdef INET6
528 		if (isipv6) {
529 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
530 			      sizeof(struct ip6_hdr));
531 			ip6 = mtod(m, struct ip6_hdr *);
532 			nth = (struct tcphdr *)(ip6 + 1);
533 		} else
534 #endif /* INET6 */
535 	      {
536 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
537 		ip = mtod(m, struct ip *);
538 		nth = (struct tcphdr *)(ip + 1);
539 	      }
540 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
541 		flags = TH_ACK;
542 	} else {
543 		m_freem(m->m_next);
544 		m->m_next = NULL;
545 		m->m_data = (caddr_t)ipgen;
546 		/* m_len is set later */
547 		tlen = 0;
548 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
549 #ifdef INET6
550 		if (isipv6) {
551 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
552 			nth = (struct tcphdr *)(ip6 + 1);
553 		} else
554 #endif /* INET6 */
555 	      {
556 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
557 		nth = (struct tcphdr *)(ip + 1);
558 	      }
559 		if (th != nth) {
560 			/*
561 			 * this is usually a case when an extension header
562 			 * exists between the IPv6 header and the
563 			 * TCP header.
564 			 */
565 			nth->th_sport = th->th_sport;
566 			nth->th_dport = th->th_dport;
567 		}
568 		xchg(nth->th_dport, nth->th_sport, n_short);
569 #undef xchg
570 	}
571 #ifdef INET6
572 	if (isipv6) {
573 		ip6->ip6_flow = 0;
574 		ip6->ip6_vfc = IPV6_VERSION;
575 		ip6->ip6_nxt = IPPROTO_TCP;
576 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
577 						tlen));
578 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
579 	} else
580 #endif
581 	{
582 		tlen += sizeof (struct tcpiphdr);
583 		ip->ip_len = tlen;
584 		ip->ip_ttl = ip_defttl;
585 		if (path_mtu_discovery)
586 			ip->ip_off |= IP_DF;
587 	}
588 	m->m_len = tlen;
589 	m->m_pkthdr.len = tlen;
590 	m->m_pkthdr.rcvif = NULL;
591 #ifdef MAC
592 	if (inp != NULL) {
593 		/*
594 		 * Packet is associated with a socket, so allow the
595 		 * label of the response to reflect the socket label.
596 		 */
597 		INP_LOCK_ASSERT(inp);
598 		mac_create_mbuf_from_inpcb(inp, m);
599 	} else {
600 		/*
601 		 * Packet is not associated with a socket, so possibly
602 		 * update the label in place.
603 		 */
604 		mac_reflect_mbuf_tcp(m);
605 	}
606 #endif
607 	nth->th_seq = htonl(seq);
608 	nth->th_ack = htonl(ack);
609 	nth->th_x2 = 0;
610 	nth->th_off = sizeof (struct tcphdr) >> 2;
611 	nth->th_flags = flags;
612 	if (tp != NULL)
613 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
614 	else
615 		nth->th_win = htons((u_short)win);
616 	nth->th_urp = 0;
617 #ifdef INET6
618 	if (isipv6) {
619 		nth->th_sum = 0;
620 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
621 					sizeof(struct ip6_hdr),
622 					tlen - sizeof(struct ip6_hdr));
623 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
624 		    NULL, NULL);
625 	} else
626 #endif /* INET6 */
627 	{
628 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
629 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
630 		m->m_pkthdr.csum_flags = CSUM_TCP;
631 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
632 	}
633 #ifdef TCPDEBUG
634 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
635 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
636 #endif
637 #ifdef INET6
638 	if (isipv6)
639 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
640 	else
641 #endif /* INET6 */
642 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
643 }
644 
645 /*
646  * Create a new TCP control block, making an
647  * empty reassembly queue and hooking it to the argument
648  * protocol control block.  The `inp' parameter must have
649  * come from the zone allocator set up in tcp_init().
650  */
651 struct tcpcb *
652 tcp_newtcpcb(struct inpcb *inp)
653 {
654 	struct tcpcb_mem *tm;
655 	struct tcpcb *tp;
656 #ifdef INET6
657 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
658 #endif /* INET6 */
659 
660 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
661 	if (tm == NULL)
662 		return (NULL);
663 	tp = &tm->tcb;
664 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
665 	tp->t_maxseg = tp->t_maxopd =
666 #ifdef INET6
667 		isipv6 ? tcp_v6mssdflt :
668 #endif /* INET6 */
669 		tcp_mssdflt;
670 
671 	/* Set up our timeouts. */
672 	callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE);
673 	callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE);
674 	callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE);
675 	callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE);
676 	callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE);
677 
678 	if (tcp_do_rfc1323)
679 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
680 	tp->sack_enable = tcp_do_sack;
681 	TAILQ_INIT(&tp->snd_holes);
682 	tp->t_inpcb = inp;	/* XXX */
683 	/*
684 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
685 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
686 	 * reasonable initial retransmit time.
687 	 */
688 	tp->t_srtt = TCPTV_SRTTBASE;
689 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
690 	tp->t_rttmin = tcp_rexmit_min;
691 	tp->t_rxtcur = TCPTV_RTOBASE;
692 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
693 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
694 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
695 	tp->t_rcvtime = ticks;
696 	tp->t_bw_rtttime = ticks;
697 	/*
698 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
699 	 * because the socket may be bound to an IPv6 wildcard address,
700 	 * which may match an IPv4-mapped IPv6 address.
701 	 */
702 	inp->inp_ip_ttl = ip_defttl;
703 	inp->inp_ppcb = tp;
704 	return (tp);		/* XXX */
705 }
706 
707 /*
708  * Drop a TCP connection, reporting
709  * the specified error.  If connection is synchronized,
710  * then send a RST to peer.
711  */
712 struct tcpcb *
713 tcp_drop(struct tcpcb *tp, int errno)
714 {
715 	struct socket *so = tp->t_inpcb->inp_socket;
716 
717 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
718 	INP_LOCK_ASSERT(tp->t_inpcb);
719 
720 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
721 		tp->t_state = TCPS_CLOSED;
722 		(void) tcp_output(tp);
723 		tcpstat.tcps_drops++;
724 	} else
725 		tcpstat.tcps_conndrops++;
726 	if (errno == ETIMEDOUT && tp->t_softerror)
727 		errno = tp->t_softerror;
728 	so->so_error = errno;
729 	return (tcp_close(tp));
730 }
731 
732 void
733 tcp_discardcb(struct tcpcb *tp)
734 {
735 	struct tseg_qent *q;
736 	struct inpcb *inp = tp->t_inpcb;
737 	struct socket *so = inp->inp_socket;
738 #ifdef INET6
739 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
740 #endif /* INET6 */
741 
742 	INP_LOCK_ASSERT(inp);
743 
744 	/*
745 	 * Make sure that all of our timers are stopped before we
746 	 * delete the PCB.
747 	 */
748 	callout_stop(tp->tt_rexmt);
749 	callout_stop(tp->tt_persist);
750 	callout_stop(tp->tt_keep);
751 	callout_stop(tp->tt_2msl);
752 	callout_stop(tp->tt_delack);
753 
754 	/*
755 	 * If we got enough samples through the srtt filter,
756 	 * save the rtt and rttvar in the routing entry.
757 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
758 	 * 4 samples is enough for the srtt filter to converge
759 	 * to within enough % of the correct value; fewer samples
760 	 * and we could save a bogus rtt. The danger is not high
761 	 * as tcp quickly recovers from everything.
762 	 * XXX: Works very well but needs some more statistics!
763 	 */
764 	if (tp->t_rttupdated >= 4) {
765 		struct hc_metrics_lite metrics;
766 		u_long ssthresh;
767 
768 		bzero(&metrics, sizeof(metrics));
769 		/*
770 		 * Update the ssthresh always when the conditions below
771 		 * are satisfied. This gives us better new start value
772 		 * for the congestion avoidance for new connections.
773 		 * ssthresh is only set if packet loss occured on a session.
774 		 *
775 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
776 		 * being torn down.  Ideally this code would not use 'so'.
777 		 */
778 		ssthresh = tp->snd_ssthresh;
779 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
780 			/*
781 			 * convert the limit from user data bytes to
782 			 * packets then to packet data bytes.
783 			 */
784 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
785 			if (ssthresh < 2)
786 				ssthresh = 2;
787 			ssthresh *= (u_long)(tp->t_maxseg +
788 #ifdef INET6
789 				      (isipv6 ? sizeof (struct ip6_hdr) +
790 					       sizeof (struct tcphdr) :
791 #endif
792 				       sizeof (struct tcpiphdr)
793 #ifdef INET6
794 				       )
795 #endif
796 				      );
797 		} else
798 			ssthresh = 0;
799 		metrics.rmx_ssthresh = ssthresh;
800 
801 		metrics.rmx_rtt = tp->t_srtt;
802 		metrics.rmx_rttvar = tp->t_rttvar;
803 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
804 		metrics.rmx_bandwidth = tp->snd_bandwidth;
805 		metrics.rmx_cwnd = tp->snd_cwnd;
806 		metrics.rmx_sendpipe = 0;
807 		metrics.rmx_recvpipe = 0;
808 
809 		tcp_hc_update(&inp->inp_inc, &metrics);
810 	}
811 
812 	/* free the reassembly queue, if any */
813 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
814 		LIST_REMOVE(q, tqe_q);
815 		m_freem(q->tqe_m);
816 		uma_zfree(tcp_reass_zone, q);
817 		tp->t_segqlen--;
818 		tcp_reass_qsize--;
819 	}
820 	tcp_free_sackholes(tp);
821 	inp->inp_ppcb = NULL;
822 	tp->t_inpcb = NULL;
823 	uma_zfree(tcpcb_zone, tp);
824 }
825 
826 /*
827  * Attempt to close a TCP control block, marking it as dropped, and freeing
828  * the socket if we hold the only reference.
829  */
830 struct tcpcb *
831 tcp_close(struct tcpcb *tp)
832 {
833 	struct inpcb *inp = tp->t_inpcb;
834 	struct socket *so;
835 
836 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
837 	INP_LOCK_ASSERT(inp);
838 
839 	in_pcbdrop(inp);
840 	tcpstat.tcps_closed++;
841 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
842 	so = inp->inp_socket;
843 	soisdisconnected(so);
844 	if (inp->inp_vflag & INP_SOCKREF) {
845 		KASSERT(so->so_state & SS_PROTOREF,
846 		    ("tcp_close: !SS_PROTOREF"));
847 		inp->inp_vflag &= ~INP_SOCKREF;
848 		INP_UNLOCK(inp);
849 		ACCEPT_LOCK();
850 		SOCK_LOCK(so);
851 		so->so_state &= ~SS_PROTOREF;
852 		sofree(so);
853 		return (NULL);
854 	}
855 	return (tp);
856 }
857 
858 void
859 tcp_drain(void)
860 {
861 
862 	if (do_tcpdrain) {
863 		struct inpcb *inpb;
864 		struct tcpcb *tcpb;
865 		struct tseg_qent *te;
866 
867 	/*
868 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
869 	 * if there is one...
870 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
871 	 *      reassembly queue should be flushed, but in a situation
872 	 *	where we're really low on mbufs, this is potentially
873 	 *	usefull.
874 	 */
875 		INP_INFO_RLOCK(&tcbinfo);
876 		LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
877 			if (inpb->inp_vflag & INP_TIMEWAIT)
878 				continue;
879 			INP_LOCK(inpb);
880 			if ((tcpb = intotcpcb(inpb)) != NULL) {
881 				while ((te = LIST_FIRST(&tcpb->t_segq))
882 			            != NULL) {
883 					LIST_REMOVE(te, tqe_q);
884 					m_freem(te->tqe_m);
885 					uma_zfree(tcp_reass_zone, te);
886 					tcpb->t_segqlen--;
887 					tcp_reass_qsize--;
888 				}
889 				tcp_clean_sackreport(tcpb);
890 			}
891 			INP_UNLOCK(inpb);
892 		}
893 		INP_INFO_RUNLOCK(&tcbinfo);
894 	}
895 }
896 
897 /*
898  * Notify a tcp user of an asynchronous error;
899  * store error as soft error, but wake up user
900  * (for now, won't do anything until can select for soft error).
901  *
902  * Do not wake up user since there currently is no mechanism for
903  * reporting soft errors (yet - a kqueue filter may be added).
904  */
905 static struct inpcb *
906 tcp_notify(struct inpcb *inp, int error)
907 {
908 	struct tcpcb *tp;
909 
910 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
911 	INP_LOCK_ASSERT(inp);
912 
913 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
914 	    (inp->inp_vflag & INP_DROPPED))
915 		return (inp);
916 
917 	tp = intotcpcb(inp);
918 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
919 
920 	/*
921 	 * Ignore some errors if we are hooked up.
922 	 * If connection hasn't completed, has retransmitted several times,
923 	 * and receives a second error, give up now.  This is better
924 	 * than waiting a long time to establish a connection that
925 	 * can never complete.
926 	 */
927 	if (tp->t_state == TCPS_ESTABLISHED &&
928 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
929 	     error == EHOSTDOWN)) {
930 		return (inp);
931 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
932 	    tp->t_softerror) {
933 		tp = tcp_drop(tp, error);
934 		if (tp != NULL)
935 			return (inp);
936 		else
937 			return (NULL);
938 	} else {
939 		tp->t_softerror = error;
940 		return (inp);
941 	}
942 #if 0
943 	wakeup( &so->so_timeo);
944 	sorwakeup(so);
945 	sowwakeup(so);
946 #endif
947 }
948 
949 static int
950 tcp_pcblist(SYSCTL_HANDLER_ARGS)
951 {
952 	int error, i, n;
953 	struct inpcb *inp, **inp_list;
954 	inp_gen_t gencnt;
955 	struct xinpgen xig;
956 
957 	/*
958 	 * The process of preparing the TCB list is too time-consuming and
959 	 * resource-intensive to repeat twice on every request.
960 	 */
961 	if (req->oldptr == NULL) {
962 		n = tcbinfo.ipi_count;
963 		req->oldidx = 2 * (sizeof xig)
964 			+ (n + n/8) * sizeof(struct xtcpcb);
965 		return (0);
966 	}
967 
968 	if (req->newptr != NULL)
969 		return (EPERM);
970 
971 	/*
972 	 * OK, now we're committed to doing something.
973 	 */
974 	INP_INFO_RLOCK(&tcbinfo);
975 	gencnt = tcbinfo.ipi_gencnt;
976 	n = tcbinfo.ipi_count;
977 	INP_INFO_RUNLOCK(&tcbinfo);
978 
979 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
980 		+ n * sizeof(struct xtcpcb));
981 	if (error != 0)
982 		return (error);
983 
984 	xig.xig_len = sizeof xig;
985 	xig.xig_count = n;
986 	xig.xig_gen = gencnt;
987 	xig.xig_sogen = so_gencnt;
988 	error = SYSCTL_OUT(req, &xig, sizeof xig);
989 	if (error)
990 		return (error);
991 
992 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
993 	if (inp_list == NULL)
994 		return (ENOMEM);
995 
996 	INP_INFO_RLOCK(&tcbinfo);
997 	for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
998 	     inp = LIST_NEXT(inp, inp_list)) {
999 		INP_LOCK(inp);
1000 		if (inp->inp_gencnt <= gencnt) {
1001 			/*
1002 			 * XXX: This use of cr_cansee(), introduced with
1003 			 * TCP state changes, is not quite right, but for
1004 			 * now, better than nothing.
1005 			 */
1006 			if (inp->inp_vflag & INP_TIMEWAIT) {
1007 				if (intotw(inp) != NULL)
1008 					error = cr_cansee(req->td->td_ucred,
1009 					    intotw(inp)->tw_cred);
1010 				else
1011 					error = EINVAL;	/* Skip this inp. */
1012 			} else
1013 				error = cr_canseesocket(req->td->td_ucred,
1014 				    inp->inp_socket);
1015 			if (error == 0)
1016 				inp_list[i++] = inp;
1017 		}
1018 		INP_UNLOCK(inp);
1019 	}
1020 	INP_INFO_RUNLOCK(&tcbinfo);
1021 	n = i;
1022 
1023 	error = 0;
1024 	for (i = 0; i < n; i++) {
1025 		inp = inp_list[i];
1026 		INP_LOCK(inp);
1027 		if (inp->inp_gencnt <= gencnt) {
1028 			struct xtcpcb xt;
1029 			void *inp_ppcb;
1030 
1031 			bzero(&xt, sizeof(xt));
1032 			xt.xt_len = sizeof xt;
1033 			/* XXX should avoid extra copy */
1034 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1035 			inp_ppcb = inp->inp_ppcb;
1036 			if (inp_ppcb == NULL)
1037 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1038 			else if (inp->inp_vflag & INP_TIMEWAIT) {
1039 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1040 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1041 			} else
1042 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1043 			if (inp->inp_socket != NULL)
1044 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1045 			else {
1046 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1047 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1048 			}
1049 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1050 			INP_UNLOCK(inp);
1051 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1052 		} else
1053 			INP_UNLOCK(inp);
1054 
1055 	}
1056 	if (!error) {
1057 		/*
1058 		 * Give the user an updated idea of our state.
1059 		 * If the generation differs from what we told
1060 		 * her before, she knows that something happened
1061 		 * while we were processing this request, and it
1062 		 * might be necessary to retry.
1063 		 */
1064 		INP_INFO_RLOCK(&tcbinfo);
1065 		xig.xig_gen = tcbinfo.ipi_gencnt;
1066 		xig.xig_sogen = so_gencnt;
1067 		xig.xig_count = tcbinfo.ipi_count;
1068 		INP_INFO_RUNLOCK(&tcbinfo);
1069 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1070 	}
1071 	free(inp_list, M_TEMP);
1072 	return (error);
1073 }
1074 
1075 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1076 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1077 
1078 static int
1079 tcp_getcred(SYSCTL_HANDLER_ARGS)
1080 {
1081 	struct xucred xuc;
1082 	struct sockaddr_in addrs[2];
1083 	struct inpcb *inp;
1084 	int error;
1085 
1086 	error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED,
1087 	    SUSER_ALLOWJAIL);
1088 	if (error)
1089 		return (error);
1090 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1091 	if (error)
1092 		return (error);
1093 	INP_INFO_RLOCK(&tcbinfo);
1094 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1095 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1096 	if (inp == NULL) {
1097 		error = ENOENT;
1098 		goto outunlocked;
1099 	}
1100 	INP_LOCK(inp);
1101 	if (inp->inp_socket == NULL) {
1102 		error = ENOENT;
1103 		goto out;
1104 	}
1105 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1106 	if (error)
1107 		goto out;
1108 	cru2x(inp->inp_socket->so_cred, &xuc);
1109 out:
1110 	INP_UNLOCK(inp);
1111 outunlocked:
1112 	INP_INFO_RUNLOCK(&tcbinfo);
1113 	if (error == 0)
1114 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1115 	return (error);
1116 }
1117 
1118 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1119     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1120     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1121 
1122 #ifdef INET6
1123 static int
1124 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1125 {
1126 	struct xucred xuc;
1127 	struct sockaddr_in6 addrs[2];
1128 	struct inpcb *inp;
1129 	int error, mapped = 0;
1130 
1131 	error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED,
1132 	    SUSER_ALLOWJAIL);
1133 	if (error)
1134 		return (error);
1135 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1136 	if (error)
1137 		return (error);
1138 	if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1139 	    (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1140 		return (error);
1141 	}
1142 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1143 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1144 			mapped = 1;
1145 		else
1146 			return (EINVAL);
1147 	}
1148 
1149 	INP_INFO_RLOCK(&tcbinfo);
1150 	if (mapped == 1)
1151 		inp = in_pcblookup_hash(&tcbinfo,
1152 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1153 			addrs[1].sin6_port,
1154 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1155 			addrs[0].sin6_port,
1156 			0, NULL);
1157 	else
1158 		inp = in6_pcblookup_hash(&tcbinfo,
1159 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1160 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1161 	if (inp == NULL) {
1162 		error = ENOENT;
1163 		goto outunlocked;
1164 	}
1165 	INP_LOCK(inp);
1166 	if (inp->inp_socket == NULL) {
1167 		error = ENOENT;
1168 		goto out;
1169 	}
1170 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1171 	if (error)
1172 		goto out;
1173 	cru2x(inp->inp_socket->so_cred, &xuc);
1174 out:
1175 	INP_UNLOCK(inp);
1176 outunlocked:
1177 	INP_INFO_RUNLOCK(&tcbinfo);
1178 	if (error == 0)
1179 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1180 	return (error);
1181 }
1182 
1183 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1184     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1185     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1186 #endif
1187 
1188 
1189 void
1190 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1191 {
1192 	struct ip *ip = vip;
1193 	struct tcphdr *th;
1194 	struct in_addr faddr;
1195 	struct inpcb *inp;
1196 	struct tcpcb *tp;
1197 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1198 	struct icmp *icp;
1199 	struct in_conninfo inc;
1200 	tcp_seq icmp_tcp_seq;
1201 	int mtu;
1202 
1203 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1204 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1205 		return;
1206 
1207 	if (cmd == PRC_MSGSIZE)
1208 		notify = tcp_mtudisc;
1209 	else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1210 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1211 		notify = tcp_drop_syn_sent;
1212 	/*
1213 	 * Redirects don't need to be handled up here.
1214 	 */
1215 	else if (PRC_IS_REDIRECT(cmd))
1216 		return;
1217 	/*
1218 	 * Source quench is depreciated.
1219 	 */
1220 	else if (cmd == PRC_QUENCH)
1221 		return;
1222 	/*
1223 	 * Hostdead is ugly because it goes linearly through all PCBs.
1224 	 * XXX: We never get this from ICMP, otherwise it makes an
1225 	 * excellent DoS attack on machines with many connections.
1226 	 */
1227 	else if (cmd == PRC_HOSTDEAD)
1228 		ip = NULL;
1229 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1230 		return;
1231 	if (ip != NULL) {
1232 		icp = (struct icmp *)((caddr_t)ip
1233 				      - offsetof(struct icmp, icmp_ip));
1234 		th = (struct tcphdr *)((caddr_t)ip
1235 				       + (ip->ip_hl << 2));
1236 		INP_INFO_WLOCK(&tcbinfo);
1237 		inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1238 		    ip->ip_src, th->th_sport, 0, NULL);
1239 		if (inp != NULL)  {
1240 			INP_LOCK(inp);
1241 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1242 			    !(inp->inp_vflag & INP_DROPPED) &&
1243 			    !(inp->inp_socket == NULL)) {
1244 				icmp_tcp_seq = htonl(th->th_seq);
1245 				tp = intotcpcb(inp);
1246 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1247 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1248 					if (cmd == PRC_MSGSIZE) {
1249 					    /*
1250 					     * MTU discovery:
1251 					     * If we got a needfrag set the MTU
1252 					     * in the route to the suggested new
1253 					     * value (if given) and then notify.
1254 					     */
1255 					    bzero(&inc, sizeof(inc));
1256 					    inc.inc_flags = 0;	/* IPv4 */
1257 					    inc.inc_faddr = faddr;
1258 
1259 					    mtu = ntohs(icp->icmp_nextmtu);
1260 					    /*
1261 					     * If no alternative MTU was
1262 					     * proposed, try the next smaller
1263 					     * one.  ip->ip_len has already
1264 					     * been swapped in icmp_input().
1265 					     */
1266 					    if (!mtu)
1267 						mtu = ip_next_mtu(ip->ip_len,
1268 						 1);
1269 					    if (mtu < max(296, (tcp_minmss)
1270 						 + sizeof(struct tcpiphdr)))
1271 						mtu = 0;
1272 					    if (!mtu)
1273 						mtu = tcp_mssdflt
1274 						 + sizeof(struct tcpiphdr);
1275 					    /*
1276 					     * Only cache the the MTU if it
1277 					     * is smaller than the interface
1278 					     * or route MTU.  tcp_mtudisc()
1279 					     * will do right thing by itself.
1280 					     */
1281 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1282 						tcp_hc_updatemtu(&inc, mtu);
1283 					}
1284 
1285 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1286 				}
1287 			}
1288 			if (inp != NULL)
1289 				INP_UNLOCK(inp);
1290 		} else {
1291 			inc.inc_fport = th->th_dport;
1292 			inc.inc_lport = th->th_sport;
1293 			inc.inc_faddr = faddr;
1294 			inc.inc_laddr = ip->ip_src;
1295 #ifdef INET6
1296 			inc.inc_isipv6 = 0;
1297 #endif
1298 			syncache_unreach(&inc, th);
1299 		}
1300 		INP_INFO_WUNLOCK(&tcbinfo);
1301 	} else
1302 		in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1303 }
1304 
1305 #ifdef INET6
1306 void
1307 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1308 {
1309 	struct tcphdr th;
1310 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1311 	struct ip6_hdr *ip6;
1312 	struct mbuf *m;
1313 	struct ip6ctlparam *ip6cp = NULL;
1314 	const struct sockaddr_in6 *sa6_src = NULL;
1315 	int off;
1316 	struct tcp_portonly {
1317 		u_int16_t th_sport;
1318 		u_int16_t th_dport;
1319 	} *thp;
1320 
1321 	if (sa->sa_family != AF_INET6 ||
1322 	    sa->sa_len != sizeof(struct sockaddr_in6))
1323 		return;
1324 
1325 	if (cmd == PRC_MSGSIZE)
1326 		notify = tcp_mtudisc;
1327 	else if (!PRC_IS_REDIRECT(cmd) &&
1328 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1329 		return;
1330 	/* Source quench is depreciated. */
1331 	else if (cmd == PRC_QUENCH)
1332 		return;
1333 
1334 	/* if the parameter is from icmp6, decode it. */
1335 	if (d != NULL) {
1336 		ip6cp = (struct ip6ctlparam *)d;
1337 		m = ip6cp->ip6c_m;
1338 		ip6 = ip6cp->ip6c_ip6;
1339 		off = ip6cp->ip6c_off;
1340 		sa6_src = ip6cp->ip6c_src;
1341 	} else {
1342 		m = NULL;
1343 		ip6 = NULL;
1344 		off = 0;	/* fool gcc */
1345 		sa6_src = &sa6_any;
1346 	}
1347 
1348 	if (ip6 != NULL) {
1349 		struct in_conninfo inc;
1350 		/*
1351 		 * XXX: We assume that when IPV6 is non NULL,
1352 		 * M and OFF are valid.
1353 		 */
1354 
1355 		/* check if we can safely examine src and dst ports */
1356 		if (m->m_pkthdr.len < off + sizeof(*thp))
1357 			return;
1358 
1359 		bzero(&th, sizeof(th));
1360 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1361 
1362 		in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1363 		    (struct sockaddr *)ip6cp->ip6c_src,
1364 		    th.th_sport, cmd, NULL, notify);
1365 
1366 		inc.inc_fport = th.th_dport;
1367 		inc.inc_lport = th.th_sport;
1368 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1369 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1370 		inc.inc_isipv6 = 1;
1371 		INP_INFO_WLOCK(&tcbinfo);
1372 		syncache_unreach(&inc, &th);
1373 		INP_INFO_WUNLOCK(&tcbinfo);
1374 	} else
1375 		in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1376 			      0, cmd, NULL, notify);
1377 }
1378 #endif /* INET6 */
1379 
1380 
1381 /*
1382  * Following is where TCP initial sequence number generation occurs.
1383  *
1384  * There are two places where we must use initial sequence numbers:
1385  * 1.  In SYN-ACK packets.
1386  * 2.  In SYN packets.
1387  *
1388  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1389  * tcp_syncache.c for details.
1390  *
1391  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1392  * depends on this property.  In addition, these ISNs should be
1393  * unguessable so as to prevent connection hijacking.  To satisfy
1394  * the requirements of this situation, the algorithm outlined in
1395  * RFC 1948 is used, with only small modifications.
1396  *
1397  * Implementation details:
1398  *
1399  * Time is based off the system timer, and is corrected so that it
1400  * increases by one megabyte per second.  This allows for proper
1401  * recycling on high speed LANs while still leaving over an hour
1402  * before rollover.
1403  *
1404  * As reading the *exact* system time is too expensive to be done
1405  * whenever setting up a TCP connection, we increment the time
1406  * offset in two ways.  First, a small random positive increment
1407  * is added to isn_offset for each connection that is set up.
1408  * Second, the function tcp_isn_tick fires once per clock tick
1409  * and increments isn_offset as necessary so that sequence numbers
1410  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1411  * random positive increments serve only to ensure that the same
1412  * exact sequence number is never sent out twice (as could otherwise
1413  * happen when a port is recycled in less than the system tick
1414  * interval.)
1415  *
1416  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1417  * between seeding of isn_secret.  This is normally set to zero,
1418  * as reseeding should not be necessary.
1419  *
1420  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1421  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1422  * general, this means holding an exclusive (write) lock.
1423  */
1424 
1425 #define ISN_BYTES_PER_SECOND 1048576
1426 #define ISN_STATIC_INCREMENT 4096
1427 #define ISN_RANDOM_INCREMENT (4096 - 1)
1428 
1429 static u_char isn_secret[32];
1430 static int isn_last_reseed;
1431 static u_int32_t isn_offset, isn_offset_old;
1432 static MD5_CTX isn_ctx;
1433 
1434 tcp_seq
1435 tcp_new_isn(struct tcpcb *tp)
1436 {
1437 	u_int32_t md5_buffer[4];
1438 	tcp_seq new_isn;
1439 
1440 	INP_LOCK_ASSERT(tp->t_inpcb);
1441 
1442 	ISN_LOCK();
1443 	/* Seed if this is the first use, reseed if requested. */
1444 	if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1445 	     (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1446 		< (u_int)ticks))) {
1447 		read_random(&isn_secret, sizeof(isn_secret));
1448 		isn_last_reseed = ticks;
1449 	}
1450 
1451 	/* Compute the md5 hash and return the ISN. */
1452 	MD5Init(&isn_ctx);
1453 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1454 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1455 #ifdef INET6
1456 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1457 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1458 			  sizeof(struct in6_addr));
1459 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1460 			  sizeof(struct in6_addr));
1461 	} else
1462 #endif
1463 	{
1464 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1465 			  sizeof(struct in_addr));
1466 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1467 			  sizeof(struct in_addr));
1468 	}
1469 	MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1470 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1471 	new_isn = (tcp_seq) md5_buffer[0];
1472 	isn_offset += ISN_STATIC_INCREMENT +
1473 		(arc4random() & ISN_RANDOM_INCREMENT);
1474 	new_isn += isn_offset;
1475 	ISN_UNLOCK();
1476 	return (new_isn);
1477 }
1478 
1479 /*
1480  * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1481  * to keep time flowing at a relatively constant rate.  If the random
1482  * increments have already pushed us past the projected offset, do nothing.
1483  */
1484 static void
1485 tcp_isn_tick(void *xtp)
1486 {
1487 	u_int32_t projected_offset;
1488 
1489 	ISN_LOCK();
1490 	projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1491 
1492 	if (projected_offset > isn_offset)
1493 		isn_offset = projected_offset;
1494 
1495 	isn_offset_old = isn_offset;
1496 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1497 	ISN_UNLOCK();
1498 }
1499 
1500 /*
1501  * When a specific ICMP unreachable message is received and the
1502  * connection state is SYN-SENT, drop the connection.  This behavior
1503  * is controlled by the icmp_may_rst sysctl.
1504  */
1505 struct inpcb *
1506 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1507 {
1508 	struct tcpcb *tp;
1509 
1510 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
1511 	INP_LOCK_ASSERT(inp);
1512 
1513 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1514 	    (inp->inp_vflag & INP_DROPPED))
1515 		return (inp);
1516 
1517 	tp = intotcpcb(inp);
1518 	if (tp->t_state != TCPS_SYN_SENT)
1519 		return (inp);
1520 
1521 	tp = tcp_drop(tp, errno);
1522 	if (tp != NULL)
1523 		return (inp);
1524 	else
1525 		return (NULL);
1526 }
1527 
1528 /*
1529  * When `need fragmentation' ICMP is received, update our idea of the MSS
1530  * based on the new value in the route.  Also nudge TCP to send something,
1531  * since we know the packet we just sent was dropped.
1532  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1533  */
1534 struct inpcb *
1535 tcp_mtudisc(struct inpcb *inp, int errno)
1536 {
1537 	struct tcpcb *tp;
1538 	struct socket *so = inp->inp_socket;
1539 	u_int maxmtu;
1540 	u_int romtu;
1541 	int mss;
1542 #ifdef INET6
1543 	int isipv6;
1544 #endif /* INET6 */
1545 
1546 	INP_LOCK_ASSERT(inp);
1547 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1548 	    (inp->inp_vflag & INP_DROPPED))
1549 		return (inp);
1550 
1551 	tp = intotcpcb(inp);
1552 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1553 
1554 #ifdef INET6
1555 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1556 #endif
1557 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1558 	romtu =
1559 #ifdef INET6
1560 	    isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
1561 #endif /* INET6 */
1562 	    tcp_maxmtu(&inp->inp_inc, NULL);
1563 	if (!maxmtu)
1564 		maxmtu = romtu;
1565 	else
1566 		maxmtu = min(maxmtu, romtu);
1567 	if (!maxmtu) {
1568 		tp->t_maxopd = tp->t_maxseg =
1569 #ifdef INET6
1570 			isipv6 ? tcp_v6mssdflt :
1571 #endif /* INET6 */
1572 			tcp_mssdflt;
1573 		return (inp);
1574 	}
1575 	mss = maxmtu -
1576 #ifdef INET6
1577 		(isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1578 #endif /* INET6 */
1579 		 sizeof(struct tcpiphdr)
1580 #ifdef INET6
1581 		 )
1582 #endif /* INET6 */
1583 		;
1584 
1585 	/*
1586 	 * XXX - The above conditional probably violates the TCP
1587 	 * spec.  The problem is that, since we don't know the
1588 	 * other end's MSS, we are supposed to use a conservative
1589 	 * default.  But, if we do that, then MTU discovery will
1590 	 * never actually take place, because the conservative
1591 	 * default is much less than the MTUs typically seen
1592 	 * on the Internet today.  For the moment, we'll sweep
1593 	 * this under the carpet.
1594 	 *
1595 	 * The conservative default might not actually be a problem
1596 	 * if the only case this occurs is when sending an initial
1597 	 * SYN with options and data to a host we've never talked
1598 	 * to before.  Then, they will reply with an MSS value which
1599 	 * will get recorded and the new parameters should get
1600 	 * recomputed.  For Further Study.
1601 	 */
1602 	if (tp->t_maxopd <= mss)
1603 		return (inp);
1604 	tp->t_maxopd = mss;
1605 
1606 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1607 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1608 		mss -= TCPOLEN_TSTAMP_APPA;
1609 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1610 	if (mss > MCLBYTES)
1611 		mss &= ~(MCLBYTES-1);
1612 #else
1613 	if (mss > MCLBYTES)
1614 		mss = mss / MCLBYTES * MCLBYTES;
1615 #endif
1616 	if (so->so_snd.sb_hiwat < mss)
1617 		mss = so->so_snd.sb_hiwat;
1618 
1619 	tp->t_maxseg = mss;
1620 
1621 	tcpstat.tcps_mturesent++;
1622 	tp->t_rtttime = 0;
1623 	tp->snd_nxt = tp->snd_una;
1624 	tcp_free_sackholes(tp);
1625 	tp->snd_recover = tp->snd_max;
1626 	if (tp->sack_enable)
1627 		EXIT_FASTRECOVERY(tp);
1628 	tcp_output(tp);
1629 	return (inp);
1630 }
1631 
1632 /*
1633  * Look-up the routing entry to the peer of this inpcb.  If no route
1634  * is found and it cannot be allocated, then return NULL.  This routine
1635  * is called by TCP routines that access the rmx structure and by tcp_mss
1636  * to get the interface MTU.
1637  */
1638 u_long
1639 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1640 {
1641 	struct route sro;
1642 	struct sockaddr_in *dst;
1643 	struct ifnet *ifp;
1644 	u_long maxmtu = 0;
1645 
1646 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1647 
1648 	bzero(&sro, sizeof(sro));
1649 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1650 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1651 		dst->sin_family = AF_INET;
1652 		dst->sin_len = sizeof(*dst);
1653 		dst->sin_addr = inc->inc_faddr;
1654 		rtalloc_ign(&sro, RTF_CLONING);
1655 	}
1656 	if (sro.ro_rt != NULL) {
1657 		ifp = sro.ro_rt->rt_ifp;
1658 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1659 			maxmtu = ifp->if_mtu;
1660 		else
1661 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1662 
1663 		/* Report additional interface capabilities. */
1664 		if (flags != NULL) {
1665 			if (ifp->if_capenable & IFCAP_TSO4 &&
1666 			    ifp->if_hwassist & CSUM_TSO)
1667 				*flags |= CSUM_TSO;
1668 		}
1669 		RTFREE(sro.ro_rt);
1670 	}
1671 	return (maxmtu);
1672 }
1673 
1674 #ifdef INET6
1675 u_long
1676 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1677 {
1678 	struct route_in6 sro6;
1679 	struct ifnet *ifp;
1680 	u_long maxmtu = 0;
1681 
1682 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1683 
1684 	bzero(&sro6, sizeof(sro6));
1685 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1686 		sro6.ro_dst.sin6_family = AF_INET6;
1687 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1688 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1689 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1690 	}
1691 	if (sro6.ro_rt != NULL) {
1692 		ifp = sro6.ro_rt->rt_ifp;
1693 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1694 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1695 		else
1696 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1697 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1698 
1699 		/* Report additional interface capabilities. */
1700 		if (flags != NULL) {
1701 			if (ifp->if_capenable & IFCAP_TSO6 &&
1702 			    ifp->if_hwassist & CSUM_TSO)
1703 				*flags |= CSUM_TSO;
1704 		}
1705 		RTFREE(sro6.ro_rt);
1706 	}
1707 
1708 	return (maxmtu);
1709 }
1710 #endif /* INET6 */
1711 
1712 #ifdef IPSEC
1713 /* compute ESP/AH header size for TCP, including outer IP header. */
1714 size_t
1715 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1716 {
1717 	struct inpcb *inp;
1718 	struct mbuf *m;
1719 	size_t hdrsiz;
1720 	struct ip *ip;
1721 #ifdef INET6
1722 	struct ip6_hdr *ip6;
1723 #endif
1724 	struct tcphdr *th;
1725 
1726 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1727 		return (0);
1728 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1729 	if (!m)
1730 		return (0);
1731 
1732 #ifdef INET6
1733 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1734 		ip6 = mtod(m, struct ip6_hdr *);
1735 		th = (struct tcphdr *)(ip6 + 1);
1736 		m->m_pkthdr.len = m->m_len =
1737 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1738 		tcpip_fillheaders(inp, ip6, th);
1739 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1740 	} else
1741 #endif /* INET6 */
1742 	{
1743 		ip = mtod(m, struct ip *);
1744 		th = (struct tcphdr *)(ip + 1);
1745 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1746 		tcpip_fillheaders(inp, ip, th);
1747 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1748 	}
1749 
1750 	m_free(m);
1751 	return (hdrsiz);
1752 }
1753 #endif /*IPSEC*/
1754 
1755 /*
1756  * Move a TCP connection into TIME_WAIT state.
1757  *    tcbinfo is locked.
1758  *    inp is locked, and is unlocked before returning.
1759  */
1760 void
1761 tcp_twstart(struct tcpcb *tp)
1762 {
1763 	struct tcptw *tw;
1764 	struct inpcb *inp = tp->t_inpcb;
1765 	int acknow;
1766 	struct socket *so;
1767 
1768 	INP_INFO_WLOCK_ASSERT(&tcbinfo);	/* tcp_timer_2msl_reset(). */
1769 	INP_LOCK_ASSERT(inp);
1770 
1771 	if (nolocaltimewait && in_localip(inp->inp_faddr)) {
1772 		tp = tcp_close(tp);
1773 		if (tp != NULL)
1774 			INP_UNLOCK(inp);
1775 		return;
1776 	}
1777 
1778 	tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1779 	if (tw == NULL) {
1780 		tw = tcp_timer_2msl_tw(1);
1781 		if (tw == NULL) {
1782 			tp = tcp_close(tp);
1783 			if (tp != NULL)
1784 				INP_UNLOCK(inp);
1785 			return;
1786 		}
1787 	}
1788 	tw->tw_inpcb = inp;
1789 
1790 	/*
1791 	 * Recover last window size sent.
1792 	 */
1793 	tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1794 
1795 	/*
1796 	 * Set t_recent if timestamps are used on the connection.
1797 	 */
1798 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1799 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1800 		tw->t_recent = tp->ts_recent;
1801 	else
1802 		tw->t_recent = 0;
1803 
1804 	tw->snd_nxt = tp->snd_nxt;
1805 	tw->rcv_nxt = tp->rcv_nxt;
1806 	tw->iss     = tp->iss;
1807 	tw->irs     = tp->irs;
1808 	tw->t_starttime = tp->t_starttime;
1809 	tw->tw_time = 0;
1810 
1811 /* XXX
1812  * If this code will
1813  * be used for fin-wait-2 state also, then we may need
1814  * a ts_recent from the last segment.
1815  */
1816 	acknow = tp->t_flags & TF_ACKNOW;
1817 
1818 	/*
1819 	 * First, discard tcpcb state, which includes stopping its timers and
1820 	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
1821 	 * that work is now done in the caller.
1822 	 *
1823 	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
1824 	 * and might not be needed here any longer.
1825 	 */
1826 	tcp_discardcb(tp);
1827 	so = inp->inp_socket;
1828 	soisdisconnected(so);
1829 	SOCK_LOCK(so);
1830 	tw->tw_cred = crhold(so->so_cred);
1831 	tw->tw_so_options = so->so_options;
1832 	SOCK_UNLOCK(so);
1833 	if (acknow)
1834 		tcp_twrespond(tw, TH_ACK);
1835 	inp->inp_ppcb = tw;
1836 	inp->inp_vflag |= INP_TIMEWAIT;
1837 	tcp_timer_2msl_reset(tw, 0);
1838 
1839 	/*
1840 	 * If the inpcb owns the sole reference to the socket, then we can
1841 	 * detach and free the socket as it is not needed in time wait.
1842 	 */
1843 	if (inp->inp_vflag & INP_SOCKREF) {
1844 		KASSERT(so->so_state & SS_PROTOREF,
1845 		    ("tcp_twstart: !SS_PROTOREF"));
1846 		inp->inp_vflag &= ~INP_SOCKREF;
1847 		INP_UNLOCK(inp);
1848 		ACCEPT_LOCK();
1849 		SOCK_LOCK(so);
1850 		so->so_state &= ~SS_PROTOREF;
1851 		sofree(so);
1852 	} else
1853 		INP_UNLOCK(inp);
1854 }
1855 
1856 #if 0
1857 /*
1858  * The appromixate rate of ISN increase of Microsoft TCP stacks;
1859  * the actual rate is slightly higher due to the addition of
1860  * random positive increments.
1861  *
1862  * Most other new OSes use semi-randomized ISN values, so we
1863  * do not need to worry about them.
1864  */
1865 #define MS_ISN_BYTES_PER_SECOND		250000
1866 
1867 /*
1868  * Determine if the ISN we will generate has advanced beyond the last
1869  * sequence number used by the previous connection.  If so, indicate
1870  * that it is safe to recycle this tw socket by returning 1.
1871  */
1872 int
1873 tcp_twrecycleable(struct tcptw *tw)
1874 {
1875 	tcp_seq new_iss = tw->iss;
1876 	tcp_seq new_irs = tw->irs;
1877 
1878 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
1879 	new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1880 	new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1881 
1882 	if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1883 		return (1);
1884 	else
1885 		return (0);
1886 }
1887 #endif
1888 
1889 void
1890 tcp_twclose(struct tcptw *tw, int reuse)
1891 {
1892 	struct socket *so;
1893 	struct inpcb *inp;
1894 
1895 	/*
1896 	 * At this point, we are in one of two situations:
1897 	 *
1898 	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
1899 	 *     all state.
1900 	 *
1901 	 * (2) We have a socket -- if we own a reference, release it and
1902 	 *     notify the socket layer.
1903 	 */
1904 	inp = tw->tw_inpcb;
1905 	KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
1906 	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
1907 	INP_INFO_WLOCK_ASSERT(&tcbinfo);	/* tcp_timer_2msl_stop(). */
1908 	INP_LOCK_ASSERT(inp);
1909 
1910 	tw->tw_inpcb = NULL;
1911 	tcp_timer_2msl_stop(tw);
1912 	inp->inp_ppcb = NULL;
1913 	in_pcbdrop(inp);
1914 
1915 	so = inp->inp_socket;
1916 	if (so != NULL) {
1917 		/*
1918 		 * If there's a socket, handle two cases: first, we own a
1919 		 * strong reference, which we will now release, or we don't
1920 		 * in which case another reference exists (XXXRW: think
1921 		 * about this more), and we don't need to take action.
1922 		 */
1923 		if (inp->inp_vflag & INP_SOCKREF) {
1924 			inp->inp_vflag &= ~INP_SOCKREF;
1925 			INP_UNLOCK(inp);
1926 			ACCEPT_LOCK();
1927 			SOCK_LOCK(so);
1928 			KASSERT(so->so_state & SS_PROTOREF,
1929 			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
1930 			so->so_state &= ~SS_PROTOREF;
1931 			sofree(so);
1932 		} else {
1933 			/*
1934 			 * If we don't own the only reference, the socket and
1935 			 * inpcb need to be left around to be handled by
1936 			 * tcp_usr_detach() later.
1937 			 */
1938 			INP_UNLOCK(inp);
1939 		}
1940 	} else {
1941 #ifdef INET6
1942 		if (inp->inp_vflag & INP_IPV6PROTO)
1943 			in6_pcbfree(inp);
1944 		else
1945 #endif
1946 			in_pcbfree(inp);
1947 	}
1948 	tcpstat.tcps_closed++;
1949 	crfree(tw->tw_cred);
1950 	tw->tw_cred = NULL;
1951 	if (reuse)
1952 		return;
1953 	uma_zfree(tcptw_zone, tw);
1954 }
1955 
1956 int
1957 tcp_twrespond(struct tcptw *tw, int flags)
1958 {
1959 	struct inpcb *inp = tw->tw_inpcb;
1960 	struct tcphdr *th;
1961 	struct mbuf *m;
1962 	struct ip *ip = NULL;
1963 	u_int8_t *optp;
1964 	u_int hdrlen, optlen;
1965 	int error;
1966 #ifdef INET6
1967 	struct ip6_hdr *ip6 = NULL;
1968 	int isipv6 = inp->inp_inc.inc_isipv6;
1969 #endif
1970 
1971 	INP_LOCK_ASSERT(inp);
1972 
1973 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1974 	if (m == NULL)
1975 		return (ENOBUFS);
1976 	m->m_data += max_linkhdr;
1977 
1978 #ifdef MAC
1979 	mac_create_mbuf_from_inpcb(inp, m);
1980 #endif
1981 
1982 #ifdef INET6
1983 	if (isipv6) {
1984 		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1985 		ip6 = mtod(m, struct ip6_hdr *);
1986 		th = (struct tcphdr *)(ip6 + 1);
1987 		tcpip_fillheaders(inp, ip6, th);
1988 	} else
1989 #endif
1990 	{
1991 		hdrlen = sizeof(struct tcpiphdr);
1992 		ip = mtod(m, struct ip *);
1993 		th = (struct tcphdr *)(ip + 1);
1994 		tcpip_fillheaders(inp, ip, th);
1995 	}
1996 	optp = (u_int8_t *)(th + 1);
1997 
1998 	/*
1999 	 * Send a timestamp and echo-reply if both our side and our peer
2000 	 * have sent timestamps in our SYN's and this is not a RST.
2001 	 */
2002 	if (tw->t_recent && flags == TH_ACK) {
2003 		u_int32_t *lp = (u_int32_t *)optp;
2004 
2005 		/* Form timestamp option as shown in appendix A of RFC 1323. */
2006 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
2007 		*lp++ = htonl(ticks);
2008 		*lp   = htonl(tw->t_recent);
2009 		optp += TCPOLEN_TSTAMP_APPA;
2010 	}
2011 
2012 	optlen = optp - (u_int8_t *)(th + 1);
2013 
2014 	m->m_len = hdrlen + optlen;
2015 	m->m_pkthdr.len = m->m_len;
2016 
2017 	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
2018 
2019 	th->th_seq = htonl(tw->snd_nxt);
2020 	th->th_ack = htonl(tw->rcv_nxt);
2021 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
2022 	th->th_flags = flags;
2023 	th->th_win = htons(tw->last_win);
2024 
2025 #ifdef INET6
2026 	if (isipv6) {
2027 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
2028 		    sizeof(struct tcphdr) + optlen);
2029 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2030 		error = ip6_output(m, inp->in6p_outputopts, NULL,
2031 		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
2032 	} else
2033 #endif
2034 	{
2035 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2036 		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
2037 		m->m_pkthdr.csum_flags = CSUM_TCP;
2038 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2039 		ip->ip_len = m->m_pkthdr.len;
2040 		if (path_mtu_discovery)
2041 			ip->ip_off |= IP_DF;
2042 		error = ip_output(m, inp->inp_options, NULL,
2043 		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
2044 		    NULL, inp);
2045 	}
2046 	if (flags & TH_ACK)
2047 		tcpstat.tcps_sndacks++;
2048 	else
2049 		tcpstat.tcps_sndctrl++;
2050 	tcpstat.tcps_sndtotal++;
2051 	return (error);
2052 }
2053 
2054 /*
2055  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
2056  *
2057  * This code attempts to calculate the bandwidth-delay product as a
2058  * means of determining the optimal window size to maximize bandwidth,
2059  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
2060  * routers.  This code also does a fairly good job keeping RTTs in check
2061  * across slow links like modems.  We implement an algorithm which is very
2062  * similar (but not meant to be) TCP/Vegas.  The code operates on the
2063  * transmitter side of a TCP connection and so only effects the transmit
2064  * side of the connection.
2065  *
2066  * BACKGROUND:  TCP makes no provision for the management of buffer space
2067  * at the end points or at the intermediate routers and switches.  A TCP
2068  * stream, whether using NewReno or not, will eventually buffer as
2069  * many packets as it is able and the only reason this typically works is
2070  * due to the fairly small default buffers made available for a connection
2071  * (typicaly 16K or 32K).  As machines use larger windows and/or window
2072  * scaling it is now fairly easy for even a single TCP connection to blow-out
2073  * all available buffer space not only on the local interface, but on
2074  * intermediate routers and switches as well.  NewReno makes a misguided
2075  * attempt to 'solve' this problem by waiting for an actual failure to occur,
2076  * then backing off, then steadily increasing the window again until another
2077  * failure occurs, ad-infinitum.  This results in terrible oscillation that
2078  * is only made worse as network loads increase and the idea of intentionally
2079  * blowing out network buffers is, frankly, a terrible way to manage network
2080  * resources.
2081  *
2082  * It is far better to limit the transmit window prior to the failure
2083  * condition being achieved.  There are two general ways to do this:  First
2084  * you can 'scan' through different transmit window sizes and locate the
2085  * point where the RTT stops increasing, indicating that you have filled the
2086  * pipe, then scan backwards until you note that RTT stops decreasing, then
2087  * repeat ad-infinitum.  This method works in principle but has severe
2088  * implementation issues due to RTT variances, timer granularity, and
2089  * instability in the algorithm which can lead to many false positives and
2090  * create oscillations as well as interact badly with other TCP streams
2091  * implementing the same algorithm.
2092  *
2093  * The second method is to limit the window to the bandwidth delay product
2094  * of the link.  This is the method we implement.  RTT variances and our
2095  * own manipulation of the congestion window, bwnd, can potentially
2096  * destabilize the algorithm.  For this reason we have to stabilize the
2097  * elements used to calculate the window.  We do this by using the minimum
2098  * observed RTT, the long term average of the observed bandwidth, and
2099  * by adding two segments worth of slop.  It isn't perfect but it is able
2100  * to react to changing conditions and gives us a very stable basis on
2101  * which to extend the algorithm.
2102  */
2103 void
2104 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
2105 {
2106 	u_long bw;
2107 	u_long bwnd;
2108 	int save_ticks;
2109 
2110 	INP_LOCK_ASSERT(tp->t_inpcb);
2111 
2112 	/*
2113 	 * If inflight_enable is disabled in the middle of a tcp connection,
2114 	 * make sure snd_bwnd is effectively disabled.
2115 	 */
2116 	if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
2117 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2118 		tp->snd_bandwidth = 0;
2119 		return;
2120 	}
2121 
2122 	/*
2123 	 * Figure out the bandwidth.  Due to the tick granularity this
2124 	 * is a very rough number and it MUST be averaged over a fairly
2125 	 * long period of time.  XXX we need to take into account a link
2126 	 * that is not using all available bandwidth, but for now our
2127 	 * slop will ramp us up if this case occurs and the bandwidth later
2128 	 * increases.
2129 	 *
2130 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
2131 	 * effectively reset t_bw_rtttime for this case.
2132 	 */
2133 	save_ticks = ticks;
2134 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2135 		return;
2136 
2137 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2138 	    (save_ticks - tp->t_bw_rtttime);
2139 	tp->t_bw_rtttime = save_ticks;
2140 	tp->t_bw_rtseq = ack_seq;
2141 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2142 		return;
2143 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2144 
2145 	tp->snd_bandwidth = bw;
2146 
2147 	/*
2148 	 * Calculate the semi-static bandwidth delay product, plus two maximal
2149 	 * segments.  The additional slop puts us squarely in the sweet
2150 	 * spot and also handles the bandwidth run-up case and stabilization.
2151 	 * Without the slop we could be locking ourselves into a lower
2152 	 * bandwidth.
2153 	 *
2154 	 * Situations Handled:
2155 	 *	(1) Prevents over-queueing of packets on LANs, especially on
2156 	 *	    high speed LANs, allowing larger TCP buffers to be
2157 	 *	    specified, and also does a good job preventing
2158 	 *	    over-queueing of packets over choke points like modems
2159 	 *	    (at least for the transmit side).
2160 	 *
2161 	 *	(2) Is able to handle changing network loads (bandwidth
2162 	 *	    drops so bwnd drops, bandwidth increases so bwnd
2163 	 *	    increases).
2164 	 *
2165 	 *	(3) Theoretically should stabilize in the face of multiple
2166 	 *	    connections implementing the same algorithm (this may need
2167 	 *	    a little work).
2168 	 *
2169 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
2170 	 *	    be adjusted with a sysctl but typically only needs to be
2171 	 *	    on very slow connections.  A value no smaller then 5
2172 	 *	    should be used, but only reduce this default if you have
2173 	 *	    no other choice.
2174 	 */
2175 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
2176 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2177 #undef USERTT
2178 
2179 	if (tcp_inflight_debug > 0) {
2180 		static int ltime;
2181 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2182 			ltime = ticks;
2183 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2184 			    tp,
2185 			    bw,
2186 			    tp->t_rttbest,
2187 			    tp->t_srtt,
2188 			    bwnd
2189 			);
2190 		}
2191 	}
2192 	if ((long)bwnd < tcp_inflight_min)
2193 		bwnd = tcp_inflight_min;
2194 	if (bwnd > tcp_inflight_max)
2195 		bwnd = tcp_inflight_max;
2196 	if ((long)bwnd < tp->t_maxseg * 2)
2197 		bwnd = tp->t_maxseg * 2;
2198 	tp->snd_bwnd = bwnd;
2199 }
2200 
2201 #ifdef TCP_SIGNATURE
2202 /*
2203  * Callback function invoked by m_apply() to digest TCP segment data
2204  * contained within an mbuf chain.
2205  */
2206 static int
2207 tcp_signature_apply(void *fstate, void *data, u_int len)
2208 {
2209 
2210 	MD5Update(fstate, (u_char *)data, len);
2211 	return (0);
2212 }
2213 
2214 /*
2215  * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2216  *
2217  * Parameters:
2218  * m		pointer to head of mbuf chain
2219  * off0		offset to TCP header within the mbuf chain
2220  * len		length of TCP segment data, excluding options
2221  * optlen	length of TCP segment options
2222  * buf		pointer to storage for computed MD5 digest
2223  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2224  *
2225  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2226  * When called from tcp_input(), we can be sure that th_sum has been
2227  * zeroed out and verified already.
2228  *
2229  * This function is for IPv4 use only. Calling this function with an
2230  * IPv6 packet in the mbuf chain will yield undefined results.
2231  *
2232  * Return 0 if successful, otherwise return -1.
2233  *
2234  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2235  * search with the destination IP address, and a 'magic SPI' to be
2236  * determined by the application. This is hardcoded elsewhere to 1179
2237  * right now. Another branch of this code exists which uses the SPD to
2238  * specify per-application flows but it is unstable.
2239  */
2240 int
2241 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2242     u_char *buf, u_int direction)
2243 {
2244 	union sockaddr_union dst;
2245 	struct ippseudo ippseudo;
2246 	MD5_CTX ctx;
2247 	int doff;
2248 	struct ip *ip;
2249 	struct ipovly *ipovly;
2250 	struct secasvar *sav;
2251 	struct tcphdr *th;
2252 	u_short savecsum;
2253 
2254 	KASSERT(m != NULL, ("NULL mbuf chain"));
2255 	KASSERT(buf != NULL, ("NULL signature pointer"));
2256 
2257 	/* Extract the destination from the IP header in the mbuf. */
2258 	ip = mtod(m, struct ip *);
2259 	bzero(&dst, sizeof(union sockaddr_union));
2260 	dst.sa.sa_len = sizeof(struct sockaddr_in);
2261 	dst.sa.sa_family = AF_INET;
2262 	dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2263 	    ip->ip_src : ip->ip_dst;
2264 
2265 	/* Look up an SADB entry which matches the address of the peer. */
2266 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2267 	if (sav == NULL) {
2268 		printf("%s: SADB lookup failed for %s\n", __func__,
2269 		    inet_ntoa(dst.sin.sin_addr));
2270 		return (EINVAL);
2271 	}
2272 
2273 	MD5Init(&ctx);
2274 	ipovly = (struct ipovly *)ip;
2275 	th = (struct tcphdr *)((u_char *)ip + off0);
2276 	doff = off0 + sizeof(struct tcphdr) + optlen;
2277 
2278 	/*
2279 	 * Step 1: Update MD5 hash with IP pseudo-header.
2280 	 *
2281 	 * XXX The ippseudo header MUST be digested in network byte order,
2282 	 * or else we'll fail the regression test. Assume all fields we've
2283 	 * been doing arithmetic on have been in host byte order.
2284 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2285 	 * tcp_output(), the underlying ip_len member has not yet been set.
2286 	 */
2287 	ippseudo.ippseudo_src = ipovly->ih_src;
2288 	ippseudo.ippseudo_dst = ipovly->ih_dst;
2289 	ippseudo.ippseudo_pad = 0;
2290 	ippseudo.ippseudo_p = IPPROTO_TCP;
2291 	ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2292 	MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2293 
2294 	/*
2295 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2296 	 * The TCP checksum must be set to zero.
2297 	 */
2298 	savecsum = th->th_sum;
2299 	th->th_sum = 0;
2300 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2301 	th->th_sum = savecsum;
2302 
2303 	/*
2304 	 * Step 3: Update MD5 hash with TCP segment data.
2305 	 *         Use m_apply() to avoid an early m_pullup().
2306 	 */
2307 	if (len > 0)
2308 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2309 
2310 	/*
2311 	 * Step 4: Update MD5 hash with shared secret.
2312 	 */
2313 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2314 	MD5Final(buf, &ctx);
2315 
2316 	key_sa_recordxfer(sav, m);
2317 	KEY_FREESAV(&sav);
2318 	return (0);
2319 }
2320 #endif /* TCP_SIGNATURE */
2321 
2322 static int
2323 sysctl_drop(SYSCTL_HANDLER_ARGS)
2324 {
2325 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2326 	struct sockaddr_storage addrs[2];
2327 	struct inpcb *inp;
2328 	struct tcpcb *tp;
2329 	struct tcptw *tw;
2330 	struct sockaddr_in *fin, *lin;
2331 #ifdef INET6
2332 	struct sockaddr_in6 *fin6, *lin6;
2333 	struct in6_addr f6, l6;
2334 #endif
2335 	int error;
2336 
2337 	inp = NULL;
2338 	fin = lin = NULL;
2339 #ifdef INET6
2340 	fin6 = lin6 = NULL;
2341 #endif
2342 	error = 0;
2343 
2344 	if (req->oldptr != NULL || req->oldlen != 0)
2345 		return (EINVAL);
2346 	if (req->newptr == NULL)
2347 		return (EPERM);
2348 	if (req->newlen < sizeof(addrs))
2349 		return (ENOMEM);
2350 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2351 	if (error)
2352 		return (error);
2353 
2354 	switch (addrs[0].ss_family) {
2355 #ifdef INET6
2356 	case AF_INET6:
2357 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2358 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2359 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2360 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2361 			return (EINVAL);
2362 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2363 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2364 				return (EINVAL);
2365 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2366 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2367 			fin = (struct sockaddr_in *)&addrs[0];
2368 			lin = (struct sockaddr_in *)&addrs[1];
2369 			break;
2370 		}
2371 		error = sa6_embedscope(fin6, ip6_use_defzone);
2372 		if (error)
2373 			return (error);
2374 		error = sa6_embedscope(lin6, ip6_use_defzone);
2375 		if (error)
2376 			return (error);
2377 		break;
2378 #endif
2379 	case AF_INET:
2380 		fin = (struct sockaddr_in *)&addrs[0];
2381 		lin = (struct sockaddr_in *)&addrs[1];
2382 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2383 		    lin->sin_len != sizeof(struct sockaddr_in))
2384 			return (EINVAL);
2385 		break;
2386 	default:
2387 		return (EINVAL);
2388 	}
2389 	INP_INFO_WLOCK(&tcbinfo);
2390 	switch (addrs[0].ss_family) {
2391 #ifdef INET6
2392 	case AF_INET6:
2393 		inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2394 		    &l6, lin6->sin6_port, 0, NULL);
2395 		break;
2396 #endif
2397 	case AF_INET:
2398 		inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2399 		    lin->sin_addr, lin->sin_port, 0, NULL);
2400 		break;
2401 	}
2402 	if (inp != NULL) {
2403 		INP_LOCK(inp);
2404 		if (inp->inp_vflag & INP_TIMEWAIT) {
2405 			/*
2406 			 * XXXRW: There currently exists a state where an
2407 			 * inpcb is present, but its timewait state has been
2408 			 * discarded.  For now, don't allow dropping of this
2409 			 * type of inpcb.
2410 			 */
2411 			tw = intotw(inp);
2412 			if (tw != NULL)
2413 				tcp_twclose(tw, 0);
2414 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2415 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2416 			tp = intotcpcb(inp);
2417 			tcp_drop(tp, ECONNABORTED);
2418 		}
2419 		INP_UNLOCK(inp);
2420 	} else
2421 		error = ESRCH;
2422 	INP_INFO_WUNLOCK(&tcbinfo);
2423 	return (error);
2424 }
2425 
2426 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2427     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2428     0, sysctl_drop, "", "Drop TCP connection");
2429