xref: /freebsd/sys/netinet/tcp_subr.c (revision 66e576525d35c68fcb86f142ebaa5a448555c0c7)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.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/jail.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/route.h>
62 #include <net/if.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet6/ip6protosw.h>
98 
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
102 #ifdef INET6
103 #include <netipsec/ipsec6.h>
104 #endif
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
107 #endif /*IPSEC*/
108 
109 #include <machine/in_cksum.h>
110 #include <sys/md5.h>
111 
112 #include <security/mac/mac_framework.h>
113 
114 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
115 #ifdef INET6
116 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
117 #endif
118 
119 static int
120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
121 {
122 	int error, new;
123 
124 	new = V_tcp_mssdflt;
125 	error = sysctl_handle_int(oidp, &new, 0, req);
126 	if (error == 0 && req->newptr) {
127 		if (new < TCP_MINMSS)
128 			error = EINVAL;
129 		else
130 			V_tcp_mssdflt = new;
131 	}
132 	return (error);
133 }
134 
135 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
136     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
137     &sysctl_net_inet_tcp_mss_check, "I",
138     "Default TCP Maximum Segment Size");
139 
140 #ifdef INET6
141 static int
142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
143 {
144 	int error, new;
145 
146 	new = V_tcp_v6mssdflt;
147 	error = sysctl_handle_int(oidp, &new, 0, req);
148 	if (error == 0 && req->newptr) {
149 		if (new < TCP_MINMSS)
150 			error = EINVAL;
151 		else
152 			V_tcp_v6mssdflt = new;
153 	}
154 	return (error);
155 }
156 
157 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
158     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
159     &sysctl_net_inet_tcp_mss_v6_check, "I",
160    "Default TCP Maximum Segment Size for IPv6");
161 #endif
162 
163 /*
164  * Minimum MSS we accept and use. This prevents DoS attacks where
165  * we are forced to a ridiculous low MSS like 20 and send hundreds
166  * of packets instead of one. The effect scales with the available
167  * bandwidth and quickly saturates the CPU and network interface
168  * with packet generation and sending. Set to zero to disable MINMSS
169  * checking. This setting prevents us from sending too small packets.
170  */
171 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
172 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
173      &VNET_NAME(tcp_minmss), 0,
174     "Minmum TCP Maximum Segment Size");
175 
176 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
177 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
178     &VNET_NAME(tcp_do_rfc1323), 0,
179     "Enable rfc1323 (high performance TCP) extensions");
180 
181 static int	tcp_log_debug = 0;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
183     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
184 
185 static int	tcp_tcbhashsize = 0;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
187     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
188 
189 static int	do_tcpdrain = 1;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
191     "Enable tcp_drain routine for extra help when low on mbufs");
192 
193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
194     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
195 
196 static VNET_DEFINE(int, icmp_may_rst) = 1;
197 #define	V_icmp_may_rst			VNET(icmp_may_rst)
198 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
199     &VNET_NAME(icmp_may_rst), 0,
200     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
201 
202 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
203 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
205     &VNET_NAME(tcp_isn_reseed_interval), 0,
206     "Seconds between reseeding of ISN secret");
207 
208 #ifdef TCP_SORECEIVE_STREAM
209 static int	tcp_soreceive_stream = 0;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
211     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
212 #endif
213 
214 VNET_DEFINE(uma_zone_t, sack_hole_zone);
215 #define	V_sack_hole_zone		VNET(sack_hole_zone)
216 
217 static struct inpcb *tcp_notify(struct inpcb *, int);
218 static void	tcp_isn_tick(void *);
219 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
220 		    void *ip4hdr, const void *ip6hdr);
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	tcp_timer	tt;
241 };
242 
243 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
244 #define	V_tcpcb_zone			VNET(tcpcb_zone)
245 
246 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
247 struct callout isn_callout;
248 static struct mtx isn_mtx;
249 
250 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
251 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
252 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
253 
254 /*
255  * TCP initialization.
256  */
257 static void
258 tcp_zone_change(void *tag)
259 {
260 
261 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
262 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
263 	tcp_tw_zone_change();
264 }
265 
266 static int
267 tcp_inpcb_init(void *mem, int size, int flags)
268 {
269 	struct inpcb *inp = mem;
270 
271 	INP_LOCK_INIT(inp, "inp", "tcpinp");
272 	return (0);
273 }
274 
275 void
276 tcp_init(void)
277 {
278 	int hashsize;
279 
280 	hashsize = TCBHASHSIZE;
281 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
282 	if (!powerof2(hashsize)) {
283 		printf("WARNING: TCB hash size not a power of 2\n");
284 		hashsize = 512; /* safe default */
285 	}
286 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
287 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE);
288 
289 	/*
290 	 * These have to be type stable for the benefit of the timers.
291 	 */
292 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
293 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
294 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
295 
296 	tcp_tw_init();
297 	syncache_init();
298 	tcp_hc_init();
299 	tcp_reass_init();
300 
301 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
302 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
303 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
304 
305 	/* Skip initialization of globals for non-default instances. */
306 	if (!IS_DEFAULT_VNET(curvnet))
307 		return;
308 
309 	/* XXX virtualize those bellow? */
310 	tcp_delacktime = TCPTV_DELACK;
311 	tcp_keepinit = TCPTV_KEEP_INIT;
312 	tcp_keepidle = TCPTV_KEEP_IDLE;
313 	tcp_keepintvl = TCPTV_KEEPINTVL;
314 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
315 	tcp_msl = TCPTV_MSL;
316 	tcp_rexmit_min = TCPTV_MIN;
317 	if (tcp_rexmit_min < 1)
318 		tcp_rexmit_min = 1;
319 	tcp_rexmit_slop = TCPTV_CPU_VAR;
320 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
321 	tcp_tcbhashsize = hashsize;
322 
323 #ifdef TCP_SORECEIVE_STREAM
324 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
325 	if (tcp_soreceive_stream) {
326 		tcp_usrreqs.pru_soreceive = soreceive_stream;
327 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
328 	}
329 #endif
330 
331 #ifdef INET6
332 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
333 #else /* INET6 */
334 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
335 #endif /* INET6 */
336 	if (max_protohdr < TCP_MINPROTOHDR)
337 		max_protohdr = TCP_MINPROTOHDR;
338 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
339 		panic("tcp_init");
340 #undef TCP_MINPROTOHDR
341 
342 	ISN_LOCK_INIT();
343 	callout_init(&isn_callout, CALLOUT_MPSAFE);
344 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
345 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
346 		SHUTDOWN_PRI_DEFAULT);
347 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
348 		EVENTHANDLER_PRI_ANY);
349 }
350 
351 #ifdef VIMAGE
352 void
353 tcp_destroy(void)
354 {
355 
356 	tcp_reass_destroy();
357 	tcp_hc_destroy();
358 	syncache_destroy();
359 	tcp_tw_destroy();
360 	in_pcbinfo_destroy(&V_tcbinfo);
361 	uma_zdestroy(V_sack_hole_zone);
362 	uma_zdestroy(V_tcpcb_zone);
363 }
364 #endif
365 
366 void
367 tcp_fini(void *xtp)
368 {
369 
370 	callout_stop(&isn_callout);
371 }
372 
373 /*
374  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
375  * tcp_template used to store this data in mbufs, but we now recopy it out
376  * of the tcpcb each time to conserve mbufs.
377  */
378 void
379 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
380 {
381 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
382 
383 	INP_WLOCK_ASSERT(inp);
384 
385 #ifdef INET6
386 	if ((inp->inp_vflag & INP_IPV6) != 0) {
387 		struct ip6_hdr *ip6;
388 
389 		ip6 = (struct ip6_hdr *)ip_ptr;
390 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
391 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
392 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
393 			(IPV6_VERSION & IPV6_VERSION_MASK);
394 		ip6->ip6_nxt = IPPROTO_TCP;
395 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
396 		ip6->ip6_src = inp->in6p_laddr;
397 		ip6->ip6_dst = inp->in6p_faddr;
398 	} else
399 #endif
400 	{
401 		struct ip *ip;
402 
403 		ip = (struct ip *)ip_ptr;
404 		ip->ip_v = IPVERSION;
405 		ip->ip_hl = 5;
406 		ip->ip_tos = inp->inp_ip_tos;
407 		ip->ip_len = 0;
408 		ip->ip_id = 0;
409 		ip->ip_off = 0;
410 		ip->ip_ttl = inp->inp_ip_ttl;
411 		ip->ip_sum = 0;
412 		ip->ip_p = IPPROTO_TCP;
413 		ip->ip_src = inp->inp_laddr;
414 		ip->ip_dst = inp->inp_faddr;
415 	}
416 	th->th_sport = inp->inp_lport;
417 	th->th_dport = inp->inp_fport;
418 	th->th_seq = 0;
419 	th->th_ack = 0;
420 	th->th_x2 = 0;
421 	th->th_off = 5;
422 	th->th_flags = 0;
423 	th->th_win = 0;
424 	th->th_urp = 0;
425 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
426 }
427 
428 /*
429  * Create template to be used to send tcp packets on a connection.
430  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
431  * use for this function is in keepalives, which use tcp_respond.
432  */
433 struct tcptemp *
434 tcpip_maketemplate(struct inpcb *inp)
435 {
436 	struct tcptemp *t;
437 
438 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
439 	if (t == NULL)
440 		return (NULL);
441 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
442 	return (t);
443 }
444 
445 /*
446  * Send a single message to the TCP at address specified by
447  * the given TCP/IP header.  If m == NULL, then we make a copy
448  * of the tcpiphdr at ti and send directly to the addressed host.
449  * This is used to force keep alive messages out using the TCP
450  * template for a connection.  If flags are given then we send
451  * a message back to the TCP which originated the * segment ti,
452  * and discard the mbuf containing it and any other attached mbufs.
453  *
454  * In any case the ack and sequence number of the transmitted
455  * segment are as specified by the parameters.
456  *
457  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
458  */
459 void
460 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
461     tcp_seq ack, tcp_seq seq, int flags)
462 {
463 	int tlen;
464 	int win = 0;
465 	struct ip *ip;
466 	struct tcphdr *nth;
467 #ifdef INET6
468 	struct ip6_hdr *ip6;
469 	int isipv6;
470 #endif /* INET6 */
471 	int ipflags = 0;
472 	struct inpcb *inp;
473 
474 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
475 
476 #ifdef INET6
477 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
478 	ip6 = ipgen;
479 #endif /* INET6 */
480 	ip = ipgen;
481 
482 	if (tp != NULL) {
483 		inp = tp->t_inpcb;
484 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
485 		INP_WLOCK_ASSERT(inp);
486 	} else
487 		inp = NULL;
488 
489 	if (tp != NULL) {
490 		if (!(flags & TH_RST)) {
491 			win = sbspace(&inp->inp_socket->so_rcv);
492 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
493 				win = (long)TCP_MAXWIN << tp->rcv_scale;
494 		}
495 	}
496 	if (m == NULL) {
497 		m = m_gethdr(M_DONTWAIT, MT_DATA);
498 		if (m == NULL)
499 			return;
500 		tlen = 0;
501 		m->m_data += max_linkhdr;
502 #ifdef INET6
503 		if (isipv6) {
504 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
505 			      sizeof(struct ip6_hdr));
506 			ip6 = mtod(m, struct ip6_hdr *);
507 			nth = (struct tcphdr *)(ip6 + 1);
508 		} else
509 #endif /* INET6 */
510 	      {
511 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
512 		ip = mtod(m, struct ip *);
513 		nth = (struct tcphdr *)(ip + 1);
514 	      }
515 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
516 		flags = TH_ACK;
517 	} else {
518 		/*
519 		 *  reuse the mbuf.
520 		 * XXX MRT We inherrit the FIB, which is lucky.
521 		 */
522 		m_freem(m->m_next);
523 		m->m_next = NULL;
524 		m->m_data = (caddr_t)ipgen;
525 		/* m_len is set later */
526 		tlen = 0;
527 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
528 #ifdef INET6
529 		if (isipv6) {
530 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
531 			nth = (struct tcphdr *)(ip6 + 1);
532 		} else
533 #endif /* INET6 */
534 	      {
535 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
536 		nth = (struct tcphdr *)(ip + 1);
537 	      }
538 		if (th != nth) {
539 			/*
540 			 * this is usually a case when an extension header
541 			 * exists between the IPv6 header and the
542 			 * TCP header.
543 			 */
544 			nth->th_sport = th->th_sport;
545 			nth->th_dport = th->th_dport;
546 		}
547 		xchg(nth->th_dport, nth->th_sport, uint16_t);
548 #undef xchg
549 	}
550 #ifdef INET6
551 	if (isipv6) {
552 		ip6->ip6_flow = 0;
553 		ip6->ip6_vfc = IPV6_VERSION;
554 		ip6->ip6_nxt = IPPROTO_TCP;
555 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
556 						tlen));
557 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
558 	} else
559 #endif
560 	{
561 		tlen += sizeof (struct tcpiphdr);
562 		ip->ip_len = tlen;
563 		ip->ip_ttl = V_ip_defttl;
564 		if (V_path_mtu_discovery)
565 			ip->ip_off |= IP_DF;
566 	}
567 	m->m_len = tlen;
568 	m->m_pkthdr.len = tlen;
569 	m->m_pkthdr.rcvif = NULL;
570 #ifdef MAC
571 	if (inp != NULL) {
572 		/*
573 		 * Packet is associated with a socket, so allow the
574 		 * label of the response to reflect the socket label.
575 		 */
576 		INP_WLOCK_ASSERT(inp);
577 		mac_inpcb_create_mbuf(inp, m);
578 	} else {
579 		/*
580 		 * Packet is not associated with a socket, so possibly
581 		 * update the label in place.
582 		 */
583 		mac_netinet_tcp_reply(m);
584 	}
585 #endif
586 	nth->th_seq = htonl(seq);
587 	nth->th_ack = htonl(ack);
588 	nth->th_x2 = 0;
589 	nth->th_off = sizeof (struct tcphdr) >> 2;
590 	nth->th_flags = flags;
591 	if (tp != NULL)
592 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
593 	else
594 		nth->th_win = htons((u_short)win);
595 	nth->th_urp = 0;
596 #ifdef INET6
597 	if (isipv6) {
598 		nth->th_sum = 0;
599 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
600 					sizeof(struct ip6_hdr),
601 					tlen - sizeof(struct ip6_hdr));
602 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
603 		    NULL, NULL);
604 	} else
605 #endif /* INET6 */
606 	{
607 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
608 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
609 		m->m_pkthdr.csum_flags = CSUM_TCP;
610 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
611 	}
612 #ifdef TCPDEBUG
613 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
614 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
615 #endif
616 #ifdef INET6
617 	if (isipv6)
618 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
619 	else
620 #endif /* INET6 */
621 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
622 }
623 
624 /*
625  * Create a new TCP control block, making an
626  * empty reassembly queue and hooking it to the argument
627  * protocol control block.  The `inp' parameter must have
628  * come from the zone allocator set up in tcp_init().
629  */
630 struct tcpcb *
631 tcp_newtcpcb(struct inpcb *inp)
632 {
633 	struct tcpcb_mem *tm;
634 	struct tcpcb *tp;
635 #ifdef INET6
636 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
637 #endif /* INET6 */
638 
639 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
640 	if (tm == NULL)
641 		return (NULL);
642 	tp = &tm->tcb;
643 #ifdef VIMAGE
644 	tp->t_vnet = inp->inp_vnet;
645 #endif
646 	tp->t_timers = &tm->tt;
647 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
648 	tp->t_maxseg = tp->t_maxopd =
649 #ifdef INET6
650 		isipv6 ? V_tcp_v6mssdflt :
651 #endif /* INET6 */
652 		V_tcp_mssdflt;
653 
654 	/* Set up our timeouts. */
655 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
656 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
657 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
658 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
659 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
660 
661 	if (V_tcp_do_rfc1323)
662 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
663 	if (V_tcp_do_sack)
664 		tp->t_flags |= TF_SACK_PERMIT;
665 	TAILQ_INIT(&tp->snd_holes);
666 	tp->t_inpcb = inp;	/* XXX */
667 	/*
668 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
669 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
670 	 * reasonable initial retransmit time.
671 	 */
672 	tp->t_srtt = TCPTV_SRTTBASE;
673 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
674 	tp->t_rttmin = tcp_rexmit_min;
675 	tp->t_rxtcur = TCPTV_RTOBASE;
676 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
677 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
678 	tp->t_rcvtime = ticks;
679 	/*
680 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
681 	 * because the socket may be bound to an IPv6 wildcard address,
682 	 * which may match an IPv4-mapped IPv6 address.
683 	 */
684 	inp->inp_ip_ttl = V_ip_defttl;
685 	inp->inp_ppcb = tp;
686 	return (tp);		/* XXX */
687 }
688 
689 /*
690  * Drop a TCP connection, reporting
691  * the specified error.  If connection is synchronized,
692  * then send a RST to peer.
693  */
694 struct tcpcb *
695 tcp_drop(struct tcpcb *tp, int errno)
696 {
697 	struct socket *so = tp->t_inpcb->inp_socket;
698 
699 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
700 	INP_WLOCK_ASSERT(tp->t_inpcb);
701 
702 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
703 		tp->t_state = TCPS_CLOSED;
704 		(void) tcp_output_reset(tp);
705 		TCPSTAT_INC(tcps_drops);
706 	} else
707 		TCPSTAT_INC(tcps_conndrops);
708 	if (errno == ETIMEDOUT && tp->t_softerror)
709 		errno = tp->t_softerror;
710 	so->so_error = errno;
711 	return (tcp_close(tp));
712 }
713 
714 void
715 tcp_discardcb(struct tcpcb *tp)
716 {
717 	struct tseg_qent *q;
718 	struct inpcb *inp = tp->t_inpcb;
719 	struct socket *so = inp->inp_socket;
720 #ifdef INET6
721 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
722 #endif /* INET6 */
723 
724 	INP_WLOCK_ASSERT(inp);
725 
726 	/*
727 	 * Make sure that all of our timers are stopped before we delete the
728 	 * PCB.
729 	 *
730 	 * XXXRW: Really, we would like to use callout_drain() here in order
731 	 * to avoid races experienced in tcp_timer.c where a timer is already
732 	 * executing at this point.  However, we can't, both because we're
733 	 * running in a context where we can't sleep, and also because we
734 	 * hold locks required by the timers.  What we instead need to do is
735 	 * test to see if callout_drain() is required, and if so, defer some
736 	 * portion of the remainder of tcp_discardcb() to an asynchronous
737 	 * context that can callout_drain() and then continue.  Some care
738 	 * will be required to ensure that no further processing takes place
739 	 * on the tcpcb, even though it hasn't been freed (a flag?).
740 	 */
741 	callout_stop(&tp->t_timers->tt_rexmt);
742 	callout_stop(&tp->t_timers->tt_persist);
743 	callout_stop(&tp->t_timers->tt_keep);
744 	callout_stop(&tp->t_timers->tt_2msl);
745 	callout_stop(&tp->t_timers->tt_delack);
746 
747 	/*
748 	 * If we got enough samples through the srtt filter,
749 	 * save the rtt and rttvar in the routing entry.
750 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
751 	 * 4 samples is enough for the srtt filter to converge
752 	 * to within enough % of the correct value; fewer samples
753 	 * and we could save a bogus rtt. The danger is not high
754 	 * as tcp quickly recovers from everything.
755 	 * XXX: Works very well but needs some more statistics!
756 	 */
757 	if (tp->t_rttupdated >= 4) {
758 		struct hc_metrics_lite metrics;
759 		u_long ssthresh;
760 
761 		bzero(&metrics, sizeof(metrics));
762 		/*
763 		 * Update the ssthresh always when the conditions below
764 		 * are satisfied. This gives us better new start value
765 		 * for the congestion avoidance for new connections.
766 		 * ssthresh is only set if packet loss occured on a session.
767 		 *
768 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
769 		 * being torn down.  Ideally this code would not use 'so'.
770 		 */
771 		ssthresh = tp->snd_ssthresh;
772 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
773 			/*
774 			 * convert the limit from user data bytes to
775 			 * packets then to packet data bytes.
776 			 */
777 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
778 			if (ssthresh < 2)
779 				ssthresh = 2;
780 			ssthresh *= (u_long)(tp->t_maxseg +
781 #ifdef INET6
782 				      (isipv6 ? sizeof (struct ip6_hdr) +
783 					       sizeof (struct tcphdr) :
784 #endif
785 				       sizeof (struct tcpiphdr)
786 #ifdef INET6
787 				       )
788 #endif
789 				      );
790 		} else
791 			ssthresh = 0;
792 		metrics.rmx_ssthresh = ssthresh;
793 
794 		metrics.rmx_rtt = tp->t_srtt;
795 		metrics.rmx_rttvar = tp->t_rttvar;
796 		metrics.rmx_cwnd = tp->snd_cwnd;
797 		metrics.rmx_sendpipe = 0;
798 		metrics.rmx_recvpipe = 0;
799 
800 		tcp_hc_update(&inp->inp_inc, &metrics);
801 	}
802 
803 	/* free the reassembly queue, if any */
804 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
805 		LIST_REMOVE(q, tqe_q);
806 		m_freem(q->tqe_m);
807 		uma_zfree(V_tcp_reass_zone, q);
808 		tp->t_segqlen--;
809 		V_tcp_reass_qsize--;
810 	}
811 	/* Disconnect offload device, if any. */
812 	tcp_offload_detach(tp);
813 
814 	tcp_free_sackholes(tp);
815 	inp->inp_ppcb = NULL;
816 	tp->t_inpcb = NULL;
817 	uma_zfree(V_tcpcb_zone, tp);
818 }
819 
820 /*
821  * Attempt to close a TCP control block, marking it as dropped, and freeing
822  * the socket if we hold the only reference.
823  */
824 struct tcpcb *
825 tcp_close(struct tcpcb *tp)
826 {
827 	struct inpcb *inp = tp->t_inpcb;
828 	struct socket *so;
829 
830 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
831 	INP_WLOCK_ASSERT(inp);
832 
833 	/* Notify any offload devices of listener close */
834 	if (tp->t_state == TCPS_LISTEN)
835 		tcp_offload_listen_close(tp);
836 	in_pcbdrop(inp);
837 	TCPSTAT_INC(tcps_closed);
838 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
839 	so = inp->inp_socket;
840 	soisdisconnected(so);
841 	if (inp->inp_flags & INP_SOCKREF) {
842 		KASSERT(so->so_state & SS_PROTOREF,
843 		    ("tcp_close: !SS_PROTOREF"));
844 		inp->inp_flags &= ~INP_SOCKREF;
845 		INP_WUNLOCK(inp);
846 		ACCEPT_LOCK();
847 		SOCK_LOCK(so);
848 		so->so_state &= ~SS_PROTOREF;
849 		sofree(so);
850 		return (NULL);
851 	}
852 	return (tp);
853 }
854 
855 void
856 tcp_drain(void)
857 {
858 	VNET_ITERATOR_DECL(vnet_iter);
859 
860 	if (!do_tcpdrain)
861 		return;
862 
863 	VNET_LIST_RLOCK_NOSLEEP();
864 	VNET_FOREACH(vnet_iter) {
865 		CURVNET_SET(vnet_iter);
866 		struct inpcb *inpb;
867 		struct tcpcb *tcpb;
868 		struct tseg_qent *te;
869 
870 	/*
871 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
872 	 * if there is one...
873 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
874 	 *      reassembly queue should be flushed, but in a situation
875 	 *	where we're really low on mbufs, this is potentially
876 	 *	usefull.
877 	 */
878 		INP_INFO_RLOCK(&V_tcbinfo);
879 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
880 			if (inpb->inp_flags & INP_TIMEWAIT)
881 				continue;
882 			INP_WLOCK(inpb);
883 			if ((tcpb = intotcpcb(inpb)) != NULL) {
884 				while ((te = LIST_FIRST(&tcpb->t_segq))
885 			            != NULL) {
886 					LIST_REMOVE(te, tqe_q);
887 					m_freem(te->tqe_m);
888 					uma_zfree(V_tcp_reass_zone, te);
889 					tcpb->t_segqlen--;
890 					V_tcp_reass_qsize--;
891 				}
892 				tcp_clean_sackreport(tcpb);
893 			}
894 			INP_WUNLOCK(inpb);
895 		}
896 		INP_INFO_RUNLOCK(&V_tcbinfo);
897 		CURVNET_RESTORE();
898 	}
899 	VNET_LIST_RUNLOCK_NOSLEEP();
900 }
901 
902 /*
903  * Notify a tcp user of an asynchronous error;
904  * store error as soft error, but wake up user
905  * (for now, won't do anything until can select for soft error).
906  *
907  * Do not wake up user since there currently is no mechanism for
908  * reporting soft errors (yet - a kqueue filter may be added).
909  */
910 static struct inpcb *
911 tcp_notify(struct inpcb *inp, int error)
912 {
913 	struct tcpcb *tp;
914 
915 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
916 	INP_WLOCK_ASSERT(inp);
917 
918 	if ((inp->inp_flags & INP_TIMEWAIT) ||
919 	    (inp->inp_flags & INP_DROPPED))
920 		return (inp);
921 
922 	tp = intotcpcb(inp);
923 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
924 
925 	/*
926 	 * Ignore some errors if we are hooked up.
927 	 * If connection hasn't completed, has retransmitted several times,
928 	 * and receives a second error, give up now.  This is better
929 	 * than waiting a long time to establish a connection that
930 	 * can never complete.
931 	 */
932 	if (tp->t_state == TCPS_ESTABLISHED &&
933 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
934 	     error == EHOSTDOWN)) {
935 		return (inp);
936 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
937 	    tp->t_softerror) {
938 		tp = tcp_drop(tp, error);
939 		if (tp != NULL)
940 			return (inp);
941 		else
942 			return (NULL);
943 	} else {
944 		tp->t_softerror = error;
945 		return (inp);
946 	}
947 #if 0
948 	wakeup( &so->so_timeo);
949 	sorwakeup(so);
950 	sowwakeup(so);
951 #endif
952 }
953 
954 static int
955 tcp_pcblist(SYSCTL_HANDLER_ARGS)
956 {
957 	int error, i, m, n, pcb_count;
958 	struct inpcb *inp, **inp_list;
959 	inp_gen_t gencnt;
960 	struct xinpgen xig;
961 
962 	/*
963 	 * The process of preparing the TCB list is too time-consuming and
964 	 * resource-intensive to repeat twice on every request.
965 	 */
966 	if (req->oldptr == NULL) {
967 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
968 		n += imax(n / 8, 10);
969 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
970 		return (0);
971 	}
972 
973 	if (req->newptr != NULL)
974 		return (EPERM);
975 
976 	/*
977 	 * OK, now we're committed to doing something.
978 	 */
979 	INP_INFO_RLOCK(&V_tcbinfo);
980 	gencnt = V_tcbinfo.ipi_gencnt;
981 	n = V_tcbinfo.ipi_count;
982 	INP_INFO_RUNLOCK(&V_tcbinfo);
983 
984 	m = syncache_pcbcount();
985 
986 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
987 		+ (n + m) * sizeof(struct xtcpcb));
988 	if (error != 0)
989 		return (error);
990 
991 	xig.xig_len = sizeof xig;
992 	xig.xig_count = n + m;
993 	xig.xig_gen = gencnt;
994 	xig.xig_sogen = so_gencnt;
995 	error = SYSCTL_OUT(req, &xig, sizeof xig);
996 	if (error)
997 		return (error);
998 
999 	error = syncache_pcblist(req, m, &pcb_count);
1000 	if (error)
1001 		return (error);
1002 
1003 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1004 	if (inp_list == NULL)
1005 		return (ENOMEM);
1006 
1007 	INP_INFO_RLOCK(&V_tcbinfo);
1008 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1009 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1010 		INP_WLOCK(inp);
1011 		if (inp->inp_gencnt <= gencnt) {
1012 			/*
1013 			 * XXX: This use of cr_cansee(), introduced with
1014 			 * TCP state changes, is not quite right, but for
1015 			 * now, better than nothing.
1016 			 */
1017 			if (inp->inp_flags & INP_TIMEWAIT) {
1018 				if (intotw(inp) != NULL)
1019 					error = cr_cansee(req->td->td_ucred,
1020 					    intotw(inp)->tw_cred);
1021 				else
1022 					error = EINVAL;	/* Skip this inp. */
1023 			} else
1024 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1025 			if (error == 0) {
1026 				in_pcbref(inp);
1027 				inp_list[i++] = inp;
1028 			}
1029 		}
1030 		INP_WUNLOCK(inp);
1031 	}
1032 	INP_INFO_RUNLOCK(&V_tcbinfo);
1033 	n = i;
1034 
1035 	error = 0;
1036 	for (i = 0; i < n; i++) {
1037 		inp = inp_list[i];
1038 		INP_RLOCK(inp);
1039 		if (inp->inp_gencnt <= gencnt) {
1040 			struct xtcpcb xt;
1041 			void *inp_ppcb;
1042 
1043 			bzero(&xt, sizeof(xt));
1044 			xt.xt_len = sizeof xt;
1045 			/* XXX should avoid extra copy */
1046 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1047 			inp_ppcb = inp->inp_ppcb;
1048 			if (inp_ppcb == NULL)
1049 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1050 			else if (inp->inp_flags & INP_TIMEWAIT) {
1051 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1052 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1053 			} else {
1054 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1055 				if (xt.xt_tp.t_timers)
1056 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1057 			}
1058 			if (inp->inp_socket != NULL)
1059 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1060 			else {
1061 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1062 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1063 			}
1064 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1065 			INP_RUNLOCK(inp);
1066 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1067 		} else
1068 			INP_RUNLOCK(inp);
1069 	}
1070 	INP_INFO_WLOCK(&V_tcbinfo);
1071 	for (i = 0; i < n; i++) {
1072 		inp = inp_list[i];
1073 		INP_WLOCK(inp);
1074 		if (!in_pcbrele(inp))
1075 			INP_WUNLOCK(inp);
1076 	}
1077 	INP_INFO_WUNLOCK(&V_tcbinfo);
1078 
1079 	if (!error) {
1080 		/*
1081 		 * Give the user an updated idea of our state.
1082 		 * If the generation differs from what we told
1083 		 * her before, she knows that something happened
1084 		 * while we were processing this request, and it
1085 		 * might be necessary to retry.
1086 		 */
1087 		INP_INFO_RLOCK(&V_tcbinfo);
1088 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1089 		xig.xig_sogen = so_gencnt;
1090 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1091 		INP_INFO_RUNLOCK(&V_tcbinfo);
1092 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1093 	}
1094 	free(inp_list, M_TEMP);
1095 	return (error);
1096 }
1097 
1098 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1099     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1100 
1101 static int
1102 tcp_getcred(SYSCTL_HANDLER_ARGS)
1103 {
1104 	struct xucred xuc;
1105 	struct sockaddr_in addrs[2];
1106 	struct inpcb *inp;
1107 	int error;
1108 
1109 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1110 	if (error)
1111 		return (error);
1112 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1113 	if (error)
1114 		return (error);
1115 	INP_INFO_RLOCK(&V_tcbinfo);
1116 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1117 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1118 	if (inp != NULL) {
1119 		INP_RLOCK(inp);
1120 		INP_INFO_RUNLOCK(&V_tcbinfo);
1121 		if (inp->inp_socket == NULL)
1122 			error = ENOENT;
1123 		if (error == 0)
1124 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1125 		if (error == 0)
1126 			cru2x(inp->inp_cred, &xuc);
1127 		INP_RUNLOCK(inp);
1128 	} else {
1129 		INP_INFO_RUNLOCK(&V_tcbinfo);
1130 		error = ENOENT;
1131 	}
1132 	if (error == 0)
1133 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1134 	return (error);
1135 }
1136 
1137 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1138     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1139     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1140 
1141 #ifdef INET6
1142 static int
1143 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1144 {
1145 	struct xucred xuc;
1146 	struct sockaddr_in6 addrs[2];
1147 	struct inpcb *inp;
1148 	int error, mapped = 0;
1149 
1150 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1151 	if (error)
1152 		return (error);
1153 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1154 	if (error)
1155 		return (error);
1156 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1157 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1158 		return (error);
1159 	}
1160 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1161 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1162 			mapped = 1;
1163 		else
1164 			return (EINVAL);
1165 	}
1166 
1167 	INP_INFO_RLOCK(&V_tcbinfo);
1168 	if (mapped == 1)
1169 		inp = in_pcblookup_hash(&V_tcbinfo,
1170 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1171 			addrs[1].sin6_port,
1172 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1173 			addrs[0].sin6_port,
1174 			0, NULL);
1175 	else
1176 		inp = in6_pcblookup_hash(&V_tcbinfo,
1177 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1178 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1179 	if (inp != NULL) {
1180 		INP_RLOCK(inp);
1181 		INP_INFO_RUNLOCK(&V_tcbinfo);
1182 		if (inp->inp_socket == NULL)
1183 			error = ENOENT;
1184 		if (error == 0)
1185 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1186 		if (error == 0)
1187 			cru2x(inp->inp_cred, &xuc);
1188 		INP_RUNLOCK(inp);
1189 	} else {
1190 		INP_INFO_RUNLOCK(&V_tcbinfo);
1191 		error = ENOENT;
1192 	}
1193 	if (error == 0)
1194 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1195 	return (error);
1196 }
1197 
1198 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1199     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1200     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1201 #endif
1202 
1203 
1204 void
1205 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1206 {
1207 	struct ip *ip = vip;
1208 	struct tcphdr *th;
1209 	struct in_addr faddr;
1210 	struct inpcb *inp;
1211 	struct tcpcb *tp;
1212 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1213 	struct icmp *icp;
1214 	struct in_conninfo inc;
1215 	tcp_seq icmp_tcp_seq;
1216 	int mtu;
1217 
1218 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1219 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1220 		return;
1221 
1222 	if (cmd == PRC_MSGSIZE)
1223 		notify = tcp_mtudisc;
1224 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1225 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1226 		notify = tcp_drop_syn_sent;
1227 	/*
1228 	 * Redirects don't need to be handled up here.
1229 	 */
1230 	else if (PRC_IS_REDIRECT(cmd))
1231 		return;
1232 	/*
1233 	 * Source quench is depreciated.
1234 	 */
1235 	else if (cmd == PRC_QUENCH)
1236 		return;
1237 	/*
1238 	 * Hostdead is ugly because it goes linearly through all PCBs.
1239 	 * XXX: We never get this from ICMP, otherwise it makes an
1240 	 * excellent DoS attack on machines with many connections.
1241 	 */
1242 	else if (cmd == PRC_HOSTDEAD)
1243 		ip = NULL;
1244 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1245 		return;
1246 	if (ip != NULL) {
1247 		icp = (struct icmp *)((caddr_t)ip
1248 				      - offsetof(struct icmp, icmp_ip));
1249 		th = (struct tcphdr *)((caddr_t)ip
1250 				       + (ip->ip_hl << 2));
1251 		INP_INFO_WLOCK(&V_tcbinfo);
1252 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1253 		    ip->ip_src, th->th_sport, 0, NULL);
1254 		if (inp != NULL)  {
1255 			INP_WLOCK(inp);
1256 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1257 			    !(inp->inp_flags & INP_DROPPED) &&
1258 			    !(inp->inp_socket == NULL)) {
1259 				icmp_tcp_seq = htonl(th->th_seq);
1260 				tp = intotcpcb(inp);
1261 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1262 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1263 					if (cmd == PRC_MSGSIZE) {
1264 					    /*
1265 					     * MTU discovery:
1266 					     * If we got a needfrag set the MTU
1267 					     * in the route to the suggested new
1268 					     * value (if given) and then notify.
1269 					     */
1270 					    bzero(&inc, sizeof(inc));
1271 					    inc.inc_faddr = faddr;
1272 					    inc.inc_fibnum =
1273 						inp->inp_inc.inc_fibnum;
1274 
1275 					    mtu = ntohs(icp->icmp_nextmtu);
1276 					    /*
1277 					     * If no alternative MTU was
1278 					     * proposed, try the next smaller
1279 					     * one.  ip->ip_len has already
1280 					     * been swapped in icmp_input().
1281 					     */
1282 					    if (!mtu)
1283 						mtu = ip_next_mtu(ip->ip_len,
1284 						 1);
1285 					    if (mtu < V_tcp_minmss
1286 						 + sizeof(struct tcpiphdr))
1287 						mtu = V_tcp_minmss
1288 						 + sizeof(struct tcpiphdr);
1289 					    /*
1290 					     * Only cache the the MTU if it
1291 					     * is smaller than the interface
1292 					     * or route MTU.  tcp_mtudisc()
1293 					     * will do right thing by itself.
1294 					     */
1295 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1296 						tcp_hc_updatemtu(&inc, mtu);
1297 					}
1298 
1299 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1300 				}
1301 			}
1302 			if (inp != NULL)
1303 				INP_WUNLOCK(inp);
1304 		} else {
1305 			bzero(&inc, sizeof(inc));
1306 			inc.inc_fport = th->th_dport;
1307 			inc.inc_lport = th->th_sport;
1308 			inc.inc_faddr = faddr;
1309 			inc.inc_laddr = ip->ip_src;
1310 			syncache_unreach(&inc, th);
1311 		}
1312 		INP_INFO_WUNLOCK(&V_tcbinfo);
1313 	} else
1314 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1315 }
1316 
1317 #ifdef INET6
1318 void
1319 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1320 {
1321 	struct tcphdr th;
1322 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1323 	struct ip6_hdr *ip6;
1324 	struct mbuf *m;
1325 	struct ip6ctlparam *ip6cp = NULL;
1326 	const struct sockaddr_in6 *sa6_src = NULL;
1327 	int off;
1328 	struct tcp_portonly {
1329 		u_int16_t th_sport;
1330 		u_int16_t th_dport;
1331 	} *thp;
1332 
1333 	if (sa->sa_family != AF_INET6 ||
1334 	    sa->sa_len != sizeof(struct sockaddr_in6))
1335 		return;
1336 
1337 	if (cmd == PRC_MSGSIZE)
1338 		notify = tcp_mtudisc;
1339 	else if (!PRC_IS_REDIRECT(cmd) &&
1340 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1341 		return;
1342 	/* Source quench is depreciated. */
1343 	else if (cmd == PRC_QUENCH)
1344 		return;
1345 
1346 	/* if the parameter is from icmp6, decode it. */
1347 	if (d != NULL) {
1348 		ip6cp = (struct ip6ctlparam *)d;
1349 		m = ip6cp->ip6c_m;
1350 		ip6 = ip6cp->ip6c_ip6;
1351 		off = ip6cp->ip6c_off;
1352 		sa6_src = ip6cp->ip6c_src;
1353 	} else {
1354 		m = NULL;
1355 		ip6 = NULL;
1356 		off = 0;	/* fool gcc */
1357 		sa6_src = &sa6_any;
1358 	}
1359 
1360 	if (ip6 != NULL) {
1361 		struct in_conninfo inc;
1362 		/*
1363 		 * XXX: We assume that when IPV6 is non NULL,
1364 		 * M and OFF are valid.
1365 		 */
1366 
1367 		/* check if we can safely examine src and dst ports */
1368 		if (m->m_pkthdr.len < off + sizeof(*thp))
1369 			return;
1370 
1371 		bzero(&th, sizeof(th));
1372 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1373 
1374 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1375 		    (struct sockaddr *)ip6cp->ip6c_src,
1376 		    th.th_sport, cmd, NULL, notify);
1377 
1378 		bzero(&inc, sizeof(inc));
1379 		inc.inc_fport = th.th_dport;
1380 		inc.inc_lport = th.th_sport;
1381 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1382 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1383 		inc.inc_flags |= INC_ISIPV6;
1384 		INP_INFO_WLOCK(&V_tcbinfo);
1385 		syncache_unreach(&inc, &th);
1386 		INP_INFO_WUNLOCK(&V_tcbinfo);
1387 	} else
1388 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1389 			      0, cmd, NULL, notify);
1390 }
1391 #endif /* INET6 */
1392 
1393 
1394 /*
1395  * Following is where TCP initial sequence number generation occurs.
1396  *
1397  * There are two places where we must use initial sequence numbers:
1398  * 1.  In SYN-ACK packets.
1399  * 2.  In SYN packets.
1400  *
1401  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1402  * tcp_syncache.c for details.
1403  *
1404  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1405  * depends on this property.  In addition, these ISNs should be
1406  * unguessable so as to prevent connection hijacking.  To satisfy
1407  * the requirements of this situation, the algorithm outlined in
1408  * RFC 1948 is used, with only small modifications.
1409  *
1410  * Implementation details:
1411  *
1412  * Time is based off the system timer, and is corrected so that it
1413  * increases by one megabyte per second.  This allows for proper
1414  * recycling on high speed LANs while still leaving over an hour
1415  * before rollover.
1416  *
1417  * As reading the *exact* system time is too expensive to be done
1418  * whenever setting up a TCP connection, we increment the time
1419  * offset in two ways.  First, a small random positive increment
1420  * is added to isn_offset for each connection that is set up.
1421  * Second, the function tcp_isn_tick fires once per clock tick
1422  * and increments isn_offset as necessary so that sequence numbers
1423  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1424  * random positive increments serve only to ensure that the same
1425  * exact sequence number is never sent out twice (as could otherwise
1426  * happen when a port is recycled in less than the system tick
1427  * interval.)
1428  *
1429  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1430  * between seeding of isn_secret.  This is normally set to zero,
1431  * as reseeding should not be necessary.
1432  *
1433  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1434  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1435  * general, this means holding an exclusive (write) lock.
1436  */
1437 
1438 #define ISN_BYTES_PER_SECOND 1048576
1439 #define ISN_STATIC_INCREMENT 4096
1440 #define ISN_RANDOM_INCREMENT (4096 - 1)
1441 
1442 static VNET_DEFINE(u_char, isn_secret[32]);
1443 static VNET_DEFINE(int, isn_last_reseed);
1444 static VNET_DEFINE(u_int32_t, isn_offset);
1445 static VNET_DEFINE(u_int32_t, isn_offset_old);
1446 
1447 #define	V_isn_secret			VNET(isn_secret)
1448 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1449 #define	V_isn_offset			VNET(isn_offset)
1450 #define	V_isn_offset_old		VNET(isn_offset_old)
1451 
1452 tcp_seq
1453 tcp_new_isn(struct tcpcb *tp)
1454 {
1455 	MD5_CTX isn_ctx;
1456 	u_int32_t md5_buffer[4];
1457 	tcp_seq new_isn;
1458 
1459 	INP_WLOCK_ASSERT(tp->t_inpcb);
1460 
1461 	ISN_LOCK();
1462 	/* Seed if this is the first use, reseed if requested. */
1463 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1464 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1465 		< (u_int)ticks))) {
1466 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1467 		V_isn_last_reseed = ticks;
1468 	}
1469 
1470 	/* Compute the md5 hash and return the ISN. */
1471 	MD5Init(&isn_ctx);
1472 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1473 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1474 #ifdef INET6
1475 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1476 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1477 			  sizeof(struct in6_addr));
1478 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1479 			  sizeof(struct in6_addr));
1480 	} else
1481 #endif
1482 	{
1483 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1484 			  sizeof(struct in_addr));
1485 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1486 			  sizeof(struct in_addr));
1487 	}
1488 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1489 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1490 	new_isn = (tcp_seq) md5_buffer[0];
1491 	V_isn_offset += ISN_STATIC_INCREMENT +
1492 		(arc4random() & ISN_RANDOM_INCREMENT);
1493 	new_isn += V_isn_offset;
1494 	ISN_UNLOCK();
1495 	return (new_isn);
1496 }
1497 
1498 /*
1499  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1500  * to keep time flowing at a relatively constant rate.  If the random
1501  * increments have already pushed us past the projected offset, do nothing.
1502  */
1503 static void
1504 tcp_isn_tick(void *xtp)
1505 {
1506 	VNET_ITERATOR_DECL(vnet_iter);
1507 	u_int32_t projected_offset;
1508 
1509 	VNET_LIST_RLOCK_NOSLEEP();
1510 	ISN_LOCK();
1511 	VNET_FOREACH(vnet_iter) {
1512 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1513 		projected_offset =
1514 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1515 
1516 		if (SEQ_GT(projected_offset, V_isn_offset))
1517 			V_isn_offset = projected_offset;
1518 
1519 		V_isn_offset_old = V_isn_offset;
1520 		CURVNET_RESTORE();
1521 	}
1522 	ISN_UNLOCK();
1523 	VNET_LIST_RUNLOCK_NOSLEEP();
1524 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1525 }
1526 
1527 /*
1528  * When a specific ICMP unreachable message is received and the
1529  * connection state is SYN-SENT, drop the connection.  This behavior
1530  * is controlled by the icmp_may_rst sysctl.
1531  */
1532 struct inpcb *
1533 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1534 {
1535 	struct tcpcb *tp;
1536 
1537 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1538 	INP_WLOCK_ASSERT(inp);
1539 
1540 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1541 	    (inp->inp_flags & INP_DROPPED))
1542 		return (inp);
1543 
1544 	tp = intotcpcb(inp);
1545 	if (tp->t_state != TCPS_SYN_SENT)
1546 		return (inp);
1547 
1548 	tp = tcp_drop(tp, errno);
1549 	if (tp != NULL)
1550 		return (inp);
1551 	else
1552 		return (NULL);
1553 }
1554 
1555 /*
1556  * When `need fragmentation' ICMP is received, update our idea of the MSS
1557  * based on the new value in the route.  Also nudge TCP to send something,
1558  * since we know the packet we just sent was dropped.
1559  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1560  */
1561 struct inpcb *
1562 tcp_mtudisc(struct inpcb *inp, int errno)
1563 {
1564 	struct tcpcb *tp;
1565 	struct socket *so;
1566 
1567 	INP_WLOCK_ASSERT(inp);
1568 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1569 	    (inp->inp_flags & INP_DROPPED))
1570 		return (inp);
1571 
1572 	tp = intotcpcb(inp);
1573 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1574 
1575 	tcp_mss_update(tp, -1, NULL, NULL);
1576 
1577 	so = inp->inp_socket;
1578 	SOCKBUF_LOCK(&so->so_snd);
1579 	/* If the mss is larger than the socket buffer, decrease the mss. */
1580 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1581 		tp->t_maxseg = so->so_snd.sb_hiwat;
1582 	SOCKBUF_UNLOCK(&so->so_snd);
1583 
1584 	TCPSTAT_INC(tcps_mturesent);
1585 	tp->t_rtttime = 0;
1586 	tp->snd_nxt = tp->snd_una;
1587 	tcp_free_sackholes(tp);
1588 	tp->snd_recover = tp->snd_max;
1589 	if (tp->t_flags & TF_SACK_PERMIT)
1590 		EXIT_FASTRECOVERY(tp);
1591 	tcp_output_send(tp);
1592 	return (inp);
1593 }
1594 
1595 /*
1596  * Look-up the routing entry to the peer of this inpcb.  If no route
1597  * is found and it cannot be allocated, then return 0.  This routine
1598  * is called by TCP routines that access the rmx structure and by
1599  * tcp_mss_update to get the peer/interface MTU.
1600  */
1601 u_long
1602 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1603 {
1604 	struct route sro;
1605 	struct sockaddr_in *dst;
1606 	struct ifnet *ifp;
1607 	u_long maxmtu = 0;
1608 
1609 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1610 
1611 	bzero(&sro, sizeof(sro));
1612 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1613 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1614 		dst->sin_family = AF_INET;
1615 		dst->sin_len = sizeof(*dst);
1616 		dst->sin_addr = inc->inc_faddr;
1617 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1618 	}
1619 	if (sro.ro_rt != NULL) {
1620 		ifp = sro.ro_rt->rt_ifp;
1621 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1622 			maxmtu = ifp->if_mtu;
1623 		else
1624 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1625 
1626 		/* Report additional interface capabilities. */
1627 		if (flags != NULL) {
1628 			if (ifp->if_capenable & IFCAP_TSO4 &&
1629 			    ifp->if_hwassist & CSUM_TSO)
1630 				*flags |= CSUM_TSO;
1631 		}
1632 		RTFREE(sro.ro_rt);
1633 	}
1634 	return (maxmtu);
1635 }
1636 
1637 #ifdef INET6
1638 u_long
1639 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1640 {
1641 	struct route_in6 sro6;
1642 	struct ifnet *ifp;
1643 	u_long maxmtu = 0;
1644 
1645 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1646 
1647 	bzero(&sro6, sizeof(sro6));
1648 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1649 		sro6.ro_dst.sin6_family = AF_INET6;
1650 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1651 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1652 		rtalloc_ign((struct route *)&sro6, 0);
1653 	}
1654 	if (sro6.ro_rt != NULL) {
1655 		ifp = sro6.ro_rt->rt_ifp;
1656 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1657 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1658 		else
1659 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1660 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1661 
1662 		/* Report additional interface capabilities. */
1663 		if (flags != NULL) {
1664 			if (ifp->if_capenable & IFCAP_TSO6 &&
1665 			    ifp->if_hwassist & CSUM_TSO)
1666 				*flags |= CSUM_TSO;
1667 		}
1668 		RTFREE(sro6.ro_rt);
1669 	}
1670 
1671 	return (maxmtu);
1672 }
1673 #endif /* INET6 */
1674 
1675 #ifdef IPSEC
1676 /* compute ESP/AH header size for TCP, including outer IP header. */
1677 size_t
1678 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1679 {
1680 	struct inpcb *inp;
1681 	struct mbuf *m;
1682 	size_t hdrsiz;
1683 	struct ip *ip;
1684 #ifdef INET6
1685 	struct ip6_hdr *ip6;
1686 #endif
1687 	struct tcphdr *th;
1688 
1689 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1690 		return (0);
1691 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1692 	if (!m)
1693 		return (0);
1694 
1695 #ifdef INET6
1696 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1697 		ip6 = mtod(m, struct ip6_hdr *);
1698 		th = (struct tcphdr *)(ip6 + 1);
1699 		m->m_pkthdr.len = m->m_len =
1700 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1701 		tcpip_fillheaders(inp, ip6, th);
1702 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1703 	} else
1704 #endif /* INET6 */
1705 	{
1706 		ip = mtod(m, struct ip *);
1707 		th = (struct tcphdr *)(ip + 1);
1708 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1709 		tcpip_fillheaders(inp, ip, th);
1710 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1711 	}
1712 
1713 	m_free(m);
1714 	return (hdrsiz);
1715 }
1716 #endif /* IPSEC */
1717 
1718 #ifdef TCP_SIGNATURE
1719 /*
1720  * Callback function invoked by m_apply() to digest TCP segment data
1721  * contained within an mbuf chain.
1722  */
1723 static int
1724 tcp_signature_apply(void *fstate, void *data, u_int len)
1725 {
1726 
1727 	MD5Update(fstate, (u_char *)data, len);
1728 	return (0);
1729 }
1730 
1731 /*
1732  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1733  *
1734  * Parameters:
1735  * m		pointer to head of mbuf chain
1736  * _unused
1737  * len		length of TCP segment data, excluding options
1738  * optlen	length of TCP segment options
1739  * buf		pointer to storage for computed MD5 digest
1740  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1741  *
1742  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1743  * When called from tcp_input(), we can be sure that th_sum has been
1744  * zeroed out and verified already.
1745  *
1746  * Return 0 if successful, otherwise return -1.
1747  *
1748  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1749  * search with the destination IP address, and a 'magic SPI' to be
1750  * determined by the application. This is hardcoded elsewhere to 1179
1751  * right now. Another branch of this code exists which uses the SPD to
1752  * specify per-application flows but it is unstable.
1753  */
1754 int
1755 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1756     u_char *buf, u_int direction)
1757 {
1758 	union sockaddr_union dst;
1759 	struct ippseudo ippseudo;
1760 	MD5_CTX ctx;
1761 	int doff;
1762 	struct ip *ip;
1763 	struct ipovly *ipovly;
1764 	struct secasvar *sav;
1765 	struct tcphdr *th;
1766 #ifdef INET6
1767 	struct ip6_hdr *ip6;
1768 	struct in6_addr in6;
1769 	char ip6buf[INET6_ADDRSTRLEN];
1770 	uint32_t plen;
1771 	uint16_t nhdr;
1772 #endif
1773 	u_short savecsum;
1774 
1775 	KASSERT(m != NULL, ("NULL mbuf chain"));
1776 	KASSERT(buf != NULL, ("NULL signature pointer"));
1777 
1778 	/* Extract the destination from the IP header in the mbuf. */
1779 	bzero(&dst, sizeof(union sockaddr_union));
1780 	ip = mtod(m, struct ip *);
1781 #ifdef INET6
1782 	ip6 = NULL;	/* Make the compiler happy. */
1783 #endif
1784 	switch (ip->ip_v) {
1785 	case IPVERSION:
1786 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1787 		dst.sa.sa_family = AF_INET;
1788 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1789 		    ip->ip_src : ip->ip_dst;
1790 		break;
1791 #ifdef INET6
1792 	case (IPV6_VERSION >> 4):
1793 		ip6 = mtod(m, struct ip6_hdr *);
1794 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1795 		dst.sa.sa_family = AF_INET6;
1796 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1797 		    ip6->ip6_src : ip6->ip6_dst;
1798 		break;
1799 #endif
1800 	default:
1801 		return (EINVAL);
1802 		/* NOTREACHED */
1803 		break;
1804 	}
1805 
1806 	/* Look up an SADB entry which matches the address of the peer. */
1807 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1808 	if (sav == NULL) {
1809 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1810 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1811 #ifdef INET6
1812 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
1813 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1814 #endif
1815 			"(unsupported)"));
1816 		return (EINVAL);
1817 	}
1818 
1819 	MD5Init(&ctx);
1820 	/*
1821 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1822 	 *
1823 	 * XXX The ippseudo header MUST be digested in network byte order,
1824 	 * or else we'll fail the regression test. Assume all fields we've
1825 	 * been doing arithmetic on have been in host byte order.
1826 	 * XXX One cannot depend on ipovly->ih_len here. When called from
1827 	 * tcp_output(), the underlying ip_len member has not yet been set.
1828 	 */
1829 	switch (ip->ip_v) {
1830 	case IPVERSION:
1831 		ipovly = (struct ipovly *)ip;
1832 		ippseudo.ippseudo_src = ipovly->ih_src;
1833 		ippseudo.ippseudo_dst = ipovly->ih_dst;
1834 		ippseudo.ippseudo_pad = 0;
1835 		ippseudo.ippseudo_p = IPPROTO_TCP;
1836 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1837 		    optlen);
1838 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1839 
1840 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1841 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1842 		break;
1843 #ifdef INET6
1844 	/*
1845 	 * RFC 2385, 2.0  Proposal
1846 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1847 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1848 	 * extended next header value (to form 32 bits), and 32-bit segment
1849 	 * length.
1850 	 * Note: Upper-Layer Packet Length comes before Next Header.
1851 	 */
1852 	case (IPV6_VERSION >> 4):
1853 		in6 = ip6->ip6_src;
1854 		in6_clearscope(&in6);
1855 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1856 		in6 = ip6->ip6_dst;
1857 		in6_clearscope(&in6);
1858 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1859 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
1860 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1861 		nhdr = 0;
1862 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1863 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1864 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1865 		nhdr = IPPROTO_TCP;
1866 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1867 
1868 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1869 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1870 		break;
1871 #endif
1872 	default:
1873 		return (EINVAL);
1874 		/* NOTREACHED */
1875 		break;
1876 	}
1877 
1878 
1879 	/*
1880 	 * Step 2: Update MD5 hash with TCP header, excluding options.
1881 	 * The TCP checksum must be set to zero.
1882 	 */
1883 	savecsum = th->th_sum;
1884 	th->th_sum = 0;
1885 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1886 	th->th_sum = savecsum;
1887 
1888 	/*
1889 	 * Step 3: Update MD5 hash with TCP segment data.
1890 	 *         Use m_apply() to avoid an early m_pullup().
1891 	 */
1892 	if (len > 0)
1893 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
1894 
1895 	/*
1896 	 * Step 4: Update MD5 hash with shared secret.
1897 	 */
1898 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1899 	MD5Final(buf, &ctx);
1900 
1901 	key_sa_recordxfer(sav, m);
1902 	KEY_FREESAV(&sav);
1903 	return (0);
1904 }
1905 #endif /* TCP_SIGNATURE */
1906 
1907 static int
1908 sysctl_drop(SYSCTL_HANDLER_ARGS)
1909 {
1910 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
1911 	struct sockaddr_storage addrs[2];
1912 	struct inpcb *inp;
1913 	struct tcpcb *tp;
1914 	struct tcptw *tw;
1915 	struct sockaddr_in *fin, *lin;
1916 #ifdef INET6
1917 	struct sockaddr_in6 *fin6, *lin6;
1918 #endif
1919 	int error;
1920 
1921 	inp = NULL;
1922 	fin = lin = NULL;
1923 #ifdef INET6
1924 	fin6 = lin6 = NULL;
1925 #endif
1926 	error = 0;
1927 
1928 	if (req->oldptr != NULL || req->oldlen != 0)
1929 		return (EINVAL);
1930 	if (req->newptr == NULL)
1931 		return (EPERM);
1932 	if (req->newlen < sizeof(addrs))
1933 		return (ENOMEM);
1934 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
1935 	if (error)
1936 		return (error);
1937 
1938 	switch (addrs[0].ss_family) {
1939 #ifdef INET6
1940 	case AF_INET6:
1941 		fin6 = (struct sockaddr_in6 *)&addrs[0];
1942 		lin6 = (struct sockaddr_in6 *)&addrs[1];
1943 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
1944 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
1945 			return (EINVAL);
1946 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
1947 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
1948 				return (EINVAL);
1949 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
1950 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
1951 			fin = (struct sockaddr_in *)&addrs[0];
1952 			lin = (struct sockaddr_in *)&addrs[1];
1953 			break;
1954 		}
1955 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
1956 		if (error)
1957 			return (error);
1958 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
1959 		if (error)
1960 			return (error);
1961 		break;
1962 #endif
1963 	case AF_INET:
1964 		fin = (struct sockaddr_in *)&addrs[0];
1965 		lin = (struct sockaddr_in *)&addrs[1];
1966 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
1967 		    lin->sin_len != sizeof(struct sockaddr_in))
1968 			return (EINVAL);
1969 		break;
1970 	default:
1971 		return (EINVAL);
1972 	}
1973 	INP_INFO_WLOCK(&V_tcbinfo);
1974 	switch (addrs[0].ss_family) {
1975 #ifdef INET6
1976 	case AF_INET6:
1977 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
1978 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
1979 		    NULL);
1980 		break;
1981 #endif
1982 	case AF_INET:
1983 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
1984 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
1985 		break;
1986 	}
1987 	if (inp != NULL) {
1988 		INP_WLOCK(inp);
1989 		if (inp->inp_flags & INP_TIMEWAIT) {
1990 			/*
1991 			 * XXXRW: There currently exists a state where an
1992 			 * inpcb is present, but its timewait state has been
1993 			 * discarded.  For now, don't allow dropping of this
1994 			 * type of inpcb.
1995 			 */
1996 			tw = intotw(inp);
1997 			if (tw != NULL)
1998 				tcp_twclose(tw, 0);
1999 			else
2000 				INP_WUNLOCK(inp);
2001 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2002 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2003 			tp = intotcpcb(inp);
2004 			tp = tcp_drop(tp, ECONNABORTED);
2005 			if (tp != NULL)
2006 				INP_WUNLOCK(inp);
2007 		} else
2008 			INP_WUNLOCK(inp);
2009 	} else
2010 		error = ESRCH;
2011 	INP_INFO_WUNLOCK(&V_tcbinfo);
2012 	return (error);
2013 }
2014 
2015 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2016     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2017     0, sysctl_drop, "", "Drop TCP connection");
2018 
2019 /*
2020  * Generate a standardized TCP log line for use throughout the
2021  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2022  * allow use in the interrupt context.
2023  *
2024  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2025  * NB: The function may return NULL if memory allocation failed.
2026  *
2027  * Due to header inclusion and ordering limitations the struct ip
2028  * and ip6_hdr pointers have to be passed as void pointers.
2029  */
2030 char *
2031 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2032     const void *ip6hdr)
2033 {
2034 
2035 	/* Is logging enabled? */
2036 	if (tcp_log_in_vain == 0)
2037 		return (NULL);
2038 
2039 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2040 }
2041 
2042 char *
2043 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2044     const void *ip6hdr)
2045 {
2046 
2047 	/* Is logging enabled? */
2048 	if (tcp_log_debug == 0)
2049 		return (NULL);
2050 
2051 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2052 }
2053 
2054 static char *
2055 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2056     const void *ip6hdr)
2057 {
2058 	char *s, *sp;
2059 	size_t size;
2060 	struct ip *ip;
2061 #ifdef INET6
2062 	const struct ip6_hdr *ip6;
2063 
2064 	ip6 = (const struct ip6_hdr *)ip6hdr;
2065 #endif /* INET6 */
2066 	ip = (struct ip *)ip4hdr;
2067 
2068 	/*
2069 	 * The log line looks like this:
2070 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2071 	 */
2072 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2073 	    sizeof(PRINT_TH_FLAGS) + 1 +
2074 #ifdef INET6
2075 	    2 * INET6_ADDRSTRLEN;
2076 #else
2077 	    2 * INET_ADDRSTRLEN;
2078 #endif /* INET6 */
2079 
2080 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2081 	if (s == NULL)
2082 		return (NULL);
2083 
2084 	strcat(s, "TCP: [");
2085 	sp = s + strlen(s);
2086 
2087 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2088 		inet_ntoa_r(inc->inc_faddr, sp);
2089 		sp = s + strlen(s);
2090 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2091 		sp = s + strlen(s);
2092 		inet_ntoa_r(inc->inc_laddr, sp);
2093 		sp = s + strlen(s);
2094 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2095 #ifdef INET6
2096 	} else if (inc) {
2097 		ip6_sprintf(sp, &inc->inc6_faddr);
2098 		sp = s + strlen(s);
2099 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2100 		sp = s + strlen(s);
2101 		ip6_sprintf(sp, &inc->inc6_laddr);
2102 		sp = s + strlen(s);
2103 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2104 	} else if (ip6 && th) {
2105 		ip6_sprintf(sp, &ip6->ip6_src);
2106 		sp = s + strlen(s);
2107 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2108 		sp = s + strlen(s);
2109 		ip6_sprintf(sp, &ip6->ip6_dst);
2110 		sp = s + strlen(s);
2111 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2112 #endif /* INET6 */
2113 	} else if (ip && th) {
2114 		inet_ntoa_r(ip->ip_src, sp);
2115 		sp = s + strlen(s);
2116 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2117 		sp = s + strlen(s);
2118 		inet_ntoa_r(ip->ip_dst, sp);
2119 		sp = s + strlen(s);
2120 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2121 	} else {
2122 		free(s, M_TCPLOG);
2123 		return (NULL);
2124 	}
2125 	sp = s + strlen(s);
2126 	if (th)
2127 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2128 	if (*(s + size - 1) != '\0')
2129 		panic("%s: string too long", __func__);
2130 	return (s);
2131 }
2132