xref: /freebsd/sys/netinet/tcp_subr.c (revision 0e1497aefd602cea581d2380d22e67dfdcac6b4e)
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/cc.h>
66 #include <netinet/in.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/ip.h>
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #endif
72 #include <netinet/in_pcb.h>
73 #ifdef INET6
74 #include <netinet6/in6_pcb.h>
75 #endif
76 #include <netinet/in_var.h>
77 #include <netinet/ip_var.h>
78 #ifdef INET6
79 #include <netinet6/ip6_var.h>
80 #include <netinet6/scope6_var.h>
81 #include <netinet6/nd6.h>
82 #endif
83 #include <netinet/ip_icmp.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 	struct cc_var		ccv;
242 };
243 
244 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
245 #define	V_tcpcb_zone			VNET(tcpcb_zone)
246 
247 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
248 struct callout isn_callout;
249 static struct mtx isn_mtx;
250 
251 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
252 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
253 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
254 
255 /*
256  * TCP initialization.
257  */
258 static void
259 tcp_zone_change(void *tag)
260 {
261 
262 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
263 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
264 	tcp_tw_zone_change();
265 }
266 
267 static int
268 tcp_inpcb_init(void *mem, int size, int flags)
269 {
270 	struct inpcb *inp = mem;
271 
272 	INP_LOCK_INIT(inp, "inp", "tcpinp");
273 	return (0);
274 }
275 
276 void
277 tcp_init(void)
278 {
279 	int hashsize;
280 
281 	hashsize = TCBHASHSIZE;
282 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
283 	if (!powerof2(hashsize)) {
284 		printf("WARNING: TCB hash size not a power of 2\n");
285 		hashsize = 512; /* safe default */
286 	}
287 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
288 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE);
289 
290 	/*
291 	 * These have to be type stable for the benefit of the timers.
292 	 */
293 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
294 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
296 
297 	tcp_tw_init();
298 	syncache_init();
299 	tcp_hc_init();
300 	tcp_reass_init();
301 
302 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
303 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
304 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 
306 	/* Skip initialization of globals for non-default instances. */
307 	if (!IS_DEFAULT_VNET(curvnet))
308 		return;
309 
310 	/* XXX virtualize those bellow? */
311 	tcp_delacktime = TCPTV_DELACK;
312 	tcp_keepinit = TCPTV_KEEP_INIT;
313 	tcp_keepidle = TCPTV_KEEP_IDLE;
314 	tcp_keepintvl = TCPTV_KEEPINTVL;
315 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
316 	tcp_msl = TCPTV_MSL;
317 	tcp_rexmit_min = TCPTV_MIN;
318 	if (tcp_rexmit_min < 1)
319 		tcp_rexmit_min = 1;
320 	tcp_rexmit_slop = TCPTV_CPU_VAR;
321 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
322 	tcp_tcbhashsize = hashsize;
323 
324 #ifdef TCP_SORECEIVE_STREAM
325 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
326 	if (tcp_soreceive_stream) {
327 		tcp_usrreqs.pru_soreceive = soreceive_stream;
328 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
329 	}
330 #endif
331 
332 #ifdef INET6
333 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
334 #else /* INET6 */
335 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
336 #endif /* INET6 */
337 	if (max_protohdr < TCP_MINPROTOHDR)
338 		max_protohdr = TCP_MINPROTOHDR;
339 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
340 		panic("tcp_init");
341 #undef TCP_MINPROTOHDR
342 
343 	ISN_LOCK_INIT();
344 	callout_init(&isn_callout, CALLOUT_MPSAFE);
345 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
346 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
347 		SHUTDOWN_PRI_DEFAULT);
348 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
349 		EVENTHANDLER_PRI_ANY);
350 }
351 
352 #ifdef VIMAGE
353 void
354 tcp_destroy(void)
355 {
356 
357 	tcp_reass_destroy();
358 	tcp_hc_destroy();
359 	syncache_destroy();
360 	tcp_tw_destroy();
361 	in_pcbinfo_destroy(&V_tcbinfo);
362 	uma_zdestroy(V_sack_hole_zone);
363 	uma_zdestroy(V_tcpcb_zone);
364 }
365 #endif
366 
367 void
368 tcp_fini(void *xtp)
369 {
370 
371 	callout_stop(&isn_callout);
372 }
373 
374 /*
375  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
376  * tcp_template used to store this data in mbufs, but we now recopy it out
377  * of the tcpcb each time to conserve mbufs.
378  */
379 void
380 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
381 {
382 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
383 
384 	INP_WLOCK_ASSERT(inp);
385 
386 #ifdef INET6
387 	if ((inp->inp_vflag & INP_IPV6) != 0) {
388 		struct ip6_hdr *ip6;
389 
390 		ip6 = (struct ip6_hdr *)ip_ptr;
391 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
392 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
393 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
394 			(IPV6_VERSION & IPV6_VERSION_MASK);
395 		ip6->ip6_nxt = IPPROTO_TCP;
396 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
397 		ip6->ip6_src = inp->in6p_laddr;
398 		ip6->ip6_dst = inp->in6p_faddr;
399 	} else
400 #endif
401 	{
402 		struct ip *ip;
403 
404 		ip = (struct ip *)ip_ptr;
405 		ip->ip_v = IPVERSION;
406 		ip->ip_hl = 5;
407 		ip->ip_tos = inp->inp_ip_tos;
408 		ip->ip_len = 0;
409 		ip->ip_id = 0;
410 		ip->ip_off = 0;
411 		ip->ip_ttl = inp->inp_ip_ttl;
412 		ip->ip_sum = 0;
413 		ip->ip_p = IPPROTO_TCP;
414 		ip->ip_src = inp->inp_laddr;
415 		ip->ip_dst = inp->inp_faddr;
416 	}
417 	th->th_sport = inp->inp_lport;
418 	th->th_dport = inp->inp_fport;
419 	th->th_seq = 0;
420 	th->th_ack = 0;
421 	th->th_x2 = 0;
422 	th->th_off = 5;
423 	th->th_flags = 0;
424 	th->th_win = 0;
425 	th->th_urp = 0;
426 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
427 }
428 
429 /*
430  * Create template to be used to send tcp packets on a connection.
431  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
432  * use for this function is in keepalives, which use tcp_respond.
433  */
434 struct tcptemp *
435 tcpip_maketemplate(struct inpcb *inp)
436 {
437 	struct tcptemp *t;
438 
439 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
440 	if (t == NULL)
441 		return (NULL);
442 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
443 	return (t);
444 }
445 
446 /*
447  * Send a single message to the TCP at address specified by
448  * the given TCP/IP header.  If m == NULL, then we make a copy
449  * of the tcpiphdr at ti and send directly to the addressed host.
450  * This is used to force keep alive messages out using the TCP
451  * template for a connection.  If flags are given then we send
452  * a message back to the TCP which originated the * segment ti,
453  * and discard the mbuf containing it and any other attached mbufs.
454  *
455  * In any case the ack and sequence number of the transmitted
456  * segment are as specified by the parameters.
457  *
458  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
459  */
460 void
461 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
462     tcp_seq ack, tcp_seq seq, int flags)
463 {
464 	int tlen;
465 	int win = 0;
466 	struct ip *ip;
467 	struct tcphdr *nth;
468 #ifdef INET6
469 	struct ip6_hdr *ip6;
470 	int isipv6;
471 #endif /* INET6 */
472 	int ipflags = 0;
473 	struct inpcb *inp;
474 
475 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
476 
477 #ifdef INET6
478 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
479 	ip6 = ipgen;
480 #endif /* INET6 */
481 	ip = ipgen;
482 
483 	if (tp != NULL) {
484 		inp = tp->t_inpcb;
485 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
486 		INP_WLOCK_ASSERT(inp);
487 	} else
488 		inp = NULL;
489 
490 	if (tp != NULL) {
491 		if (!(flags & TH_RST)) {
492 			win = sbspace(&inp->inp_socket->so_rcv);
493 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
494 				win = (long)TCP_MAXWIN << tp->rcv_scale;
495 		}
496 	}
497 	if (m == NULL) {
498 		m = m_gethdr(M_DONTWAIT, MT_DATA);
499 		if (m == NULL)
500 			return;
501 		tlen = 0;
502 		m->m_data += max_linkhdr;
503 #ifdef INET6
504 		if (isipv6) {
505 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
506 			      sizeof(struct ip6_hdr));
507 			ip6 = mtod(m, struct ip6_hdr *);
508 			nth = (struct tcphdr *)(ip6 + 1);
509 		} else
510 #endif /* INET6 */
511 	      {
512 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
513 		ip = mtod(m, struct ip *);
514 		nth = (struct tcphdr *)(ip + 1);
515 	      }
516 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
517 		flags = TH_ACK;
518 	} else {
519 		/*
520 		 *  reuse the mbuf.
521 		 * XXX MRT We inherrit the FIB, which is lucky.
522 		 */
523 		m_freem(m->m_next);
524 		m->m_next = NULL;
525 		m->m_data = (caddr_t)ipgen;
526 		/* m_len is set later */
527 		tlen = 0;
528 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
529 #ifdef INET6
530 		if (isipv6) {
531 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
532 			nth = (struct tcphdr *)(ip6 + 1);
533 		} else
534 #endif /* INET6 */
535 	      {
536 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
537 		nth = (struct tcphdr *)(ip + 1);
538 	      }
539 		if (th != nth) {
540 			/*
541 			 * this is usually a case when an extension header
542 			 * exists between the IPv6 header and the
543 			 * TCP header.
544 			 */
545 			nth->th_sport = th->th_sport;
546 			nth->th_dport = th->th_dport;
547 		}
548 		xchg(nth->th_dport, nth->th_sport, uint16_t);
549 #undef xchg
550 	}
551 #ifdef INET6
552 	if (isipv6) {
553 		ip6->ip6_flow = 0;
554 		ip6->ip6_vfc = IPV6_VERSION;
555 		ip6->ip6_nxt = IPPROTO_TCP;
556 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
557 						tlen));
558 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
559 	} else
560 #endif
561 	{
562 		tlen += sizeof (struct tcpiphdr);
563 		ip->ip_len = tlen;
564 		ip->ip_ttl = V_ip_defttl;
565 		if (V_path_mtu_discovery)
566 			ip->ip_off |= IP_DF;
567 	}
568 	m->m_len = tlen;
569 	m->m_pkthdr.len = tlen;
570 	m->m_pkthdr.rcvif = NULL;
571 #ifdef MAC
572 	if (inp != NULL) {
573 		/*
574 		 * Packet is associated with a socket, so allow the
575 		 * label of the response to reflect the socket label.
576 		 */
577 		INP_WLOCK_ASSERT(inp);
578 		mac_inpcb_create_mbuf(inp, m);
579 	} else {
580 		/*
581 		 * Packet is not associated with a socket, so possibly
582 		 * update the label in place.
583 		 */
584 		mac_netinet_tcp_reply(m);
585 	}
586 #endif
587 	nth->th_seq = htonl(seq);
588 	nth->th_ack = htonl(ack);
589 	nth->th_x2 = 0;
590 	nth->th_off = sizeof (struct tcphdr) >> 2;
591 	nth->th_flags = flags;
592 	if (tp != NULL)
593 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
594 	else
595 		nth->th_win = htons((u_short)win);
596 	nth->th_urp = 0;
597 #ifdef INET6
598 	if (isipv6) {
599 		nth->th_sum = 0;
600 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
601 					sizeof(struct ip6_hdr),
602 					tlen - sizeof(struct ip6_hdr));
603 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
604 		    NULL, NULL);
605 	} else
606 #endif /* INET6 */
607 	{
608 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
609 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
610 		m->m_pkthdr.csum_flags = CSUM_TCP;
611 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
612 	}
613 #ifdef TCPDEBUG
614 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
615 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
616 #endif
617 #ifdef INET6
618 	if (isipv6)
619 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
620 	else
621 #endif /* INET6 */
622 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
623 }
624 
625 /*
626  * Create a new TCP control block, making an
627  * empty reassembly queue and hooking it to the argument
628  * protocol control block.  The `inp' parameter must have
629  * come from the zone allocator set up in tcp_init().
630  */
631 struct tcpcb *
632 tcp_newtcpcb(struct inpcb *inp)
633 {
634 	struct tcpcb_mem *tm;
635 	struct tcpcb *tp;
636 #ifdef INET6
637 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
638 #endif /* INET6 */
639 
640 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
641 	if (tm == NULL)
642 		return (NULL);
643 	tp = &tm->tcb;
644 
645 	/* Initialise cc_var struct for this tcpcb. */
646 	tp->ccv = &tm->ccv;
647 	tp->ccv->type = IPPROTO_TCP;
648 	tp->ccv->ccvc.tcp = tp;
649 
650 	/*
651 	 * Use the current system default CC algorithm.
652 	 */
653 	CC_LIST_RLOCK();
654 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
655 	CC_ALGO(tp) = CC_DEFAULT();
656 	CC_LIST_RUNLOCK();
657 
658 	if (CC_ALGO(tp)->cb_init != NULL)
659 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
660 			uma_zfree(V_tcpcb_zone, tm);
661 			return (NULL);
662 		}
663 
664 #ifdef VIMAGE
665 	tp->t_vnet = inp->inp_vnet;
666 #endif
667 	tp->t_timers = &tm->tt;
668 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
669 	tp->t_maxseg = tp->t_maxopd =
670 #ifdef INET6
671 		isipv6 ? V_tcp_v6mssdflt :
672 #endif /* INET6 */
673 		V_tcp_mssdflt;
674 
675 	/* Set up our timeouts. */
676 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
677 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
678 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
679 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
680 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
681 
682 	if (V_tcp_do_rfc1323)
683 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
684 	if (V_tcp_do_sack)
685 		tp->t_flags |= TF_SACK_PERMIT;
686 	TAILQ_INIT(&tp->snd_holes);
687 	tp->t_inpcb = inp;	/* XXX */
688 	/*
689 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
690 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
691 	 * reasonable initial retransmit time.
692 	 */
693 	tp->t_srtt = TCPTV_SRTTBASE;
694 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
695 	tp->t_rttmin = tcp_rexmit_min;
696 	tp->t_rxtcur = TCPTV_RTOBASE;
697 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
698 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
699 	tp->t_rcvtime = ticks;
700 	/*
701 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
702 	 * because the socket may be bound to an IPv6 wildcard address,
703 	 * which may match an IPv4-mapped IPv6 address.
704 	 */
705 	inp->inp_ip_ttl = V_ip_defttl;
706 	inp->inp_ppcb = tp;
707 	return (tp);		/* XXX */
708 }
709 
710 /*
711  * Switch the congestion control algorithm back to NewReno for any active
712  * control blocks using an algorithm which is about to go away.
713  * This ensures the CC framework can allow the unload to proceed without leaving
714  * any dangling pointers which would trigger a panic.
715  * Returning non-zero would inform the CC framework that something went wrong
716  * and it would be unsafe to allow the unload to proceed. However, there is no
717  * way for this to occur with this implementation so we always return zero.
718  */
719 int
720 tcp_ccalgounload(struct cc_algo *unload_algo)
721 {
722 	struct cc_algo *tmpalgo;
723 	struct inpcb *inp;
724 	struct tcpcb *tp;
725 	VNET_ITERATOR_DECL(vnet_iter);
726 
727 	/*
728 	 * Check all active control blocks across all network stacks and change
729 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
730 	 * requires cleanup code to be run, call it.
731 	 */
732 	VNET_LIST_RLOCK();
733 	VNET_FOREACH(vnet_iter) {
734 		CURVNET_SET(vnet_iter);
735 		INP_INFO_RLOCK(&V_tcbinfo);
736 		/*
737 		 * New connections already part way through being initialised
738 		 * with the CC algo we're removing will not race with this code
739 		 * because the INP_INFO_WLOCK is held during initialisation. We
740 		 * therefore don't enter the loop below until the connection
741 		 * list has stabilised.
742 		 */
743 		LIST_FOREACH(inp, &V_tcb, inp_list) {
744 			INP_WLOCK(inp);
745 			/* Important to skip tcptw structs. */
746 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
747 			    (tp = intotcpcb(inp)) != NULL) {
748 				/*
749 				 * By holding INP_WLOCK here, we are assured
750 				 * that the connection is not currently
751 				 * executing inside the CC module's functions
752 				 * i.e. it is safe to make the switch back to
753 				 * NewReno.
754 				 */
755 				if (CC_ALGO(tp) == unload_algo) {
756 					tmpalgo = CC_ALGO(tp);
757 					/* NewReno does not require any init. */
758 					CC_ALGO(tp) = &newreno_cc_algo;
759 					if (tmpalgo->cb_destroy != NULL)
760 						tmpalgo->cb_destroy(tp->ccv);
761 				}
762 			}
763 			INP_WUNLOCK(inp);
764 		}
765 		INP_INFO_RUNLOCK(&V_tcbinfo);
766 		CURVNET_RESTORE();
767 	}
768 	VNET_LIST_RUNLOCK();
769 
770 	return (0);
771 }
772 
773 /*
774  * Drop a TCP connection, reporting
775  * the specified error.  If connection is synchronized,
776  * then send a RST to peer.
777  */
778 struct tcpcb *
779 tcp_drop(struct tcpcb *tp, int errno)
780 {
781 	struct socket *so = tp->t_inpcb->inp_socket;
782 
783 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
784 	INP_WLOCK_ASSERT(tp->t_inpcb);
785 
786 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
787 		tp->t_state = TCPS_CLOSED;
788 		(void) tcp_output_reset(tp);
789 		TCPSTAT_INC(tcps_drops);
790 	} else
791 		TCPSTAT_INC(tcps_conndrops);
792 	if (errno == ETIMEDOUT && tp->t_softerror)
793 		errno = tp->t_softerror;
794 	so->so_error = errno;
795 	return (tcp_close(tp));
796 }
797 
798 void
799 tcp_discardcb(struct tcpcb *tp)
800 {
801 	struct inpcb *inp = tp->t_inpcb;
802 	struct socket *so = inp->inp_socket;
803 #ifdef INET6
804 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
805 #endif /* INET6 */
806 
807 	INP_WLOCK_ASSERT(inp);
808 
809 	/*
810 	 * Make sure that all of our timers are stopped before we delete the
811 	 * PCB.
812 	 *
813 	 * XXXRW: Really, we would like to use callout_drain() here in order
814 	 * to avoid races experienced in tcp_timer.c where a timer is already
815 	 * executing at this point.  However, we can't, both because we're
816 	 * running in a context where we can't sleep, and also because we
817 	 * hold locks required by the timers.  What we instead need to do is
818 	 * test to see if callout_drain() is required, and if so, defer some
819 	 * portion of the remainder of tcp_discardcb() to an asynchronous
820 	 * context that can callout_drain() and then continue.  Some care
821 	 * will be required to ensure that no further processing takes place
822 	 * on the tcpcb, even though it hasn't been freed (a flag?).
823 	 */
824 	callout_stop(&tp->t_timers->tt_rexmt);
825 	callout_stop(&tp->t_timers->tt_persist);
826 	callout_stop(&tp->t_timers->tt_keep);
827 	callout_stop(&tp->t_timers->tt_2msl);
828 	callout_stop(&tp->t_timers->tt_delack);
829 
830 	/*
831 	 * If we got enough samples through the srtt filter,
832 	 * save the rtt and rttvar in the routing entry.
833 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
834 	 * 4 samples is enough for the srtt filter to converge
835 	 * to within enough % of the correct value; fewer samples
836 	 * and we could save a bogus rtt. The danger is not high
837 	 * as tcp quickly recovers from everything.
838 	 * XXX: Works very well but needs some more statistics!
839 	 */
840 	if (tp->t_rttupdated >= 4) {
841 		struct hc_metrics_lite metrics;
842 		u_long ssthresh;
843 
844 		bzero(&metrics, sizeof(metrics));
845 		/*
846 		 * Update the ssthresh always when the conditions below
847 		 * are satisfied. This gives us better new start value
848 		 * for the congestion avoidance for new connections.
849 		 * ssthresh is only set if packet loss occured on a session.
850 		 *
851 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
852 		 * being torn down.  Ideally this code would not use 'so'.
853 		 */
854 		ssthresh = tp->snd_ssthresh;
855 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
856 			/*
857 			 * convert the limit from user data bytes to
858 			 * packets then to packet data bytes.
859 			 */
860 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
861 			if (ssthresh < 2)
862 				ssthresh = 2;
863 			ssthresh *= (u_long)(tp->t_maxseg +
864 #ifdef INET6
865 				      (isipv6 ? sizeof (struct ip6_hdr) +
866 					       sizeof (struct tcphdr) :
867 #endif
868 				       sizeof (struct tcpiphdr)
869 #ifdef INET6
870 				       )
871 #endif
872 				      );
873 		} else
874 			ssthresh = 0;
875 		metrics.rmx_ssthresh = ssthresh;
876 
877 		metrics.rmx_rtt = tp->t_srtt;
878 		metrics.rmx_rttvar = tp->t_rttvar;
879 		metrics.rmx_cwnd = tp->snd_cwnd;
880 		metrics.rmx_sendpipe = 0;
881 		metrics.rmx_recvpipe = 0;
882 
883 		tcp_hc_update(&inp->inp_inc, &metrics);
884 	}
885 
886 	/* free the reassembly queue, if any */
887 	tcp_reass_flush(tp);
888 	/* Disconnect offload device, if any. */
889 	tcp_offload_detach(tp);
890 
891 	tcp_free_sackholes(tp);
892 
893 	/* Allow the CC algorithm to clean up after itself. */
894 	if (CC_ALGO(tp)->cb_destroy != NULL)
895 		CC_ALGO(tp)->cb_destroy(tp->ccv);
896 
897 	CC_ALGO(tp) = NULL;
898 	inp->inp_ppcb = NULL;
899 	tp->t_inpcb = NULL;
900 	uma_zfree(V_tcpcb_zone, tp);
901 }
902 
903 /*
904  * Attempt to close a TCP control block, marking it as dropped, and freeing
905  * the socket if we hold the only reference.
906  */
907 struct tcpcb *
908 tcp_close(struct tcpcb *tp)
909 {
910 	struct inpcb *inp = tp->t_inpcb;
911 	struct socket *so;
912 
913 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
914 	INP_WLOCK_ASSERT(inp);
915 
916 	/* Notify any offload devices of listener close */
917 	if (tp->t_state == TCPS_LISTEN)
918 		tcp_offload_listen_close(tp);
919 	in_pcbdrop(inp);
920 	TCPSTAT_INC(tcps_closed);
921 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
922 	so = inp->inp_socket;
923 	soisdisconnected(so);
924 	if (inp->inp_flags & INP_SOCKREF) {
925 		KASSERT(so->so_state & SS_PROTOREF,
926 		    ("tcp_close: !SS_PROTOREF"));
927 		inp->inp_flags &= ~INP_SOCKREF;
928 		INP_WUNLOCK(inp);
929 		ACCEPT_LOCK();
930 		SOCK_LOCK(so);
931 		so->so_state &= ~SS_PROTOREF;
932 		sofree(so);
933 		return (NULL);
934 	}
935 	return (tp);
936 }
937 
938 void
939 tcp_drain(void)
940 {
941 	VNET_ITERATOR_DECL(vnet_iter);
942 
943 	if (!do_tcpdrain)
944 		return;
945 
946 	VNET_LIST_RLOCK_NOSLEEP();
947 	VNET_FOREACH(vnet_iter) {
948 		CURVNET_SET(vnet_iter);
949 		struct inpcb *inpb;
950 		struct tcpcb *tcpb;
951 
952 	/*
953 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
954 	 * if there is one...
955 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
956 	 *      reassembly queue should be flushed, but in a situation
957 	 *	where we're really low on mbufs, this is potentially
958 	 *	usefull.
959 	 */
960 		INP_INFO_RLOCK(&V_tcbinfo);
961 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
962 			if (inpb->inp_flags & INP_TIMEWAIT)
963 				continue;
964 			INP_WLOCK(inpb);
965 			if ((tcpb = intotcpcb(inpb)) != NULL) {
966 				tcp_reass_flush(tcpb);
967 				tcp_clean_sackreport(tcpb);
968 			}
969 			INP_WUNLOCK(inpb);
970 		}
971 		INP_INFO_RUNLOCK(&V_tcbinfo);
972 		CURVNET_RESTORE();
973 	}
974 	VNET_LIST_RUNLOCK_NOSLEEP();
975 }
976 
977 /*
978  * Notify a tcp user of an asynchronous error;
979  * store error as soft error, but wake up user
980  * (for now, won't do anything until can select for soft error).
981  *
982  * Do not wake up user since there currently is no mechanism for
983  * reporting soft errors (yet - a kqueue filter may be added).
984  */
985 static struct inpcb *
986 tcp_notify(struct inpcb *inp, int error)
987 {
988 	struct tcpcb *tp;
989 
990 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
991 	INP_WLOCK_ASSERT(inp);
992 
993 	if ((inp->inp_flags & INP_TIMEWAIT) ||
994 	    (inp->inp_flags & INP_DROPPED))
995 		return (inp);
996 
997 	tp = intotcpcb(inp);
998 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
999 
1000 	/*
1001 	 * Ignore some errors if we are hooked up.
1002 	 * If connection hasn't completed, has retransmitted several times,
1003 	 * and receives a second error, give up now.  This is better
1004 	 * than waiting a long time to establish a connection that
1005 	 * can never complete.
1006 	 */
1007 	if (tp->t_state == TCPS_ESTABLISHED &&
1008 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1009 	     error == EHOSTDOWN)) {
1010 		return (inp);
1011 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1012 	    tp->t_softerror) {
1013 		tp = tcp_drop(tp, error);
1014 		if (tp != NULL)
1015 			return (inp);
1016 		else
1017 			return (NULL);
1018 	} else {
1019 		tp->t_softerror = error;
1020 		return (inp);
1021 	}
1022 #if 0
1023 	wakeup( &so->so_timeo);
1024 	sorwakeup(so);
1025 	sowwakeup(so);
1026 #endif
1027 }
1028 
1029 static int
1030 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1031 {
1032 	int error, i, m, n, pcb_count;
1033 	struct inpcb *inp, **inp_list;
1034 	inp_gen_t gencnt;
1035 	struct xinpgen xig;
1036 
1037 	/*
1038 	 * The process of preparing the TCB list is too time-consuming and
1039 	 * resource-intensive to repeat twice on every request.
1040 	 */
1041 	if (req->oldptr == NULL) {
1042 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1043 		n += imax(n / 8, 10);
1044 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1045 		return (0);
1046 	}
1047 
1048 	if (req->newptr != NULL)
1049 		return (EPERM);
1050 
1051 	/*
1052 	 * OK, now we're committed to doing something.
1053 	 */
1054 	INP_INFO_RLOCK(&V_tcbinfo);
1055 	gencnt = V_tcbinfo.ipi_gencnt;
1056 	n = V_tcbinfo.ipi_count;
1057 	INP_INFO_RUNLOCK(&V_tcbinfo);
1058 
1059 	m = syncache_pcbcount();
1060 
1061 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1062 		+ (n + m) * sizeof(struct xtcpcb));
1063 	if (error != 0)
1064 		return (error);
1065 
1066 	xig.xig_len = sizeof xig;
1067 	xig.xig_count = n + m;
1068 	xig.xig_gen = gencnt;
1069 	xig.xig_sogen = so_gencnt;
1070 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1071 	if (error)
1072 		return (error);
1073 
1074 	error = syncache_pcblist(req, m, &pcb_count);
1075 	if (error)
1076 		return (error);
1077 
1078 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1079 	if (inp_list == NULL)
1080 		return (ENOMEM);
1081 
1082 	INP_INFO_RLOCK(&V_tcbinfo);
1083 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1084 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1085 		INP_WLOCK(inp);
1086 		if (inp->inp_gencnt <= gencnt) {
1087 			/*
1088 			 * XXX: This use of cr_cansee(), introduced with
1089 			 * TCP state changes, is not quite right, but for
1090 			 * now, better than nothing.
1091 			 */
1092 			if (inp->inp_flags & INP_TIMEWAIT) {
1093 				if (intotw(inp) != NULL)
1094 					error = cr_cansee(req->td->td_ucred,
1095 					    intotw(inp)->tw_cred);
1096 				else
1097 					error = EINVAL;	/* Skip this inp. */
1098 			} else
1099 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1100 			if (error == 0) {
1101 				in_pcbref(inp);
1102 				inp_list[i++] = inp;
1103 			}
1104 		}
1105 		INP_WUNLOCK(inp);
1106 	}
1107 	INP_INFO_RUNLOCK(&V_tcbinfo);
1108 	n = i;
1109 
1110 	error = 0;
1111 	for (i = 0; i < n; i++) {
1112 		inp = inp_list[i];
1113 		INP_RLOCK(inp);
1114 		if (inp->inp_gencnt <= gencnt) {
1115 			struct xtcpcb xt;
1116 			void *inp_ppcb;
1117 
1118 			bzero(&xt, sizeof(xt));
1119 			xt.xt_len = sizeof xt;
1120 			/* XXX should avoid extra copy */
1121 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1122 			inp_ppcb = inp->inp_ppcb;
1123 			if (inp_ppcb == NULL)
1124 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1125 			else if (inp->inp_flags & INP_TIMEWAIT) {
1126 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1127 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1128 			} else {
1129 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1130 				if (xt.xt_tp.t_timers)
1131 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1132 			}
1133 			if (inp->inp_socket != NULL)
1134 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1135 			else {
1136 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1137 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1138 			}
1139 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1140 			INP_RUNLOCK(inp);
1141 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1142 		} else
1143 			INP_RUNLOCK(inp);
1144 	}
1145 	INP_INFO_WLOCK(&V_tcbinfo);
1146 	for (i = 0; i < n; i++) {
1147 		inp = inp_list[i];
1148 		INP_WLOCK(inp);
1149 		if (!in_pcbrele(inp))
1150 			INP_WUNLOCK(inp);
1151 	}
1152 	INP_INFO_WUNLOCK(&V_tcbinfo);
1153 
1154 	if (!error) {
1155 		/*
1156 		 * Give the user an updated idea of our state.
1157 		 * If the generation differs from what we told
1158 		 * her before, she knows that something happened
1159 		 * while we were processing this request, and it
1160 		 * might be necessary to retry.
1161 		 */
1162 		INP_INFO_RLOCK(&V_tcbinfo);
1163 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1164 		xig.xig_sogen = so_gencnt;
1165 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1166 		INP_INFO_RUNLOCK(&V_tcbinfo);
1167 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1168 	}
1169 	free(inp_list, M_TEMP);
1170 	return (error);
1171 }
1172 
1173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1174     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1175 
1176 static int
1177 tcp_getcred(SYSCTL_HANDLER_ARGS)
1178 {
1179 	struct xucred xuc;
1180 	struct sockaddr_in addrs[2];
1181 	struct inpcb *inp;
1182 	int error;
1183 
1184 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1185 	if (error)
1186 		return (error);
1187 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1188 	if (error)
1189 		return (error);
1190 	INP_INFO_RLOCK(&V_tcbinfo);
1191 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1192 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1193 	if (inp != NULL) {
1194 		INP_RLOCK(inp);
1195 		INP_INFO_RUNLOCK(&V_tcbinfo);
1196 		if (inp->inp_socket == NULL)
1197 			error = ENOENT;
1198 		if (error == 0)
1199 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1200 		if (error == 0)
1201 			cru2x(inp->inp_cred, &xuc);
1202 		INP_RUNLOCK(inp);
1203 	} else {
1204 		INP_INFO_RUNLOCK(&V_tcbinfo);
1205 		error = ENOENT;
1206 	}
1207 	if (error == 0)
1208 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1209 	return (error);
1210 }
1211 
1212 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1213     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1214     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1215 
1216 #ifdef INET6
1217 static int
1218 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1219 {
1220 	struct xucred xuc;
1221 	struct sockaddr_in6 addrs[2];
1222 	struct inpcb *inp;
1223 	int error, mapped = 0;
1224 
1225 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1226 	if (error)
1227 		return (error);
1228 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1229 	if (error)
1230 		return (error);
1231 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1232 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1233 		return (error);
1234 	}
1235 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1236 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1237 			mapped = 1;
1238 		else
1239 			return (EINVAL);
1240 	}
1241 
1242 	INP_INFO_RLOCK(&V_tcbinfo);
1243 	if (mapped == 1)
1244 		inp = in_pcblookup_hash(&V_tcbinfo,
1245 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1246 			addrs[1].sin6_port,
1247 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1248 			addrs[0].sin6_port,
1249 			0, NULL);
1250 	else
1251 		inp = in6_pcblookup_hash(&V_tcbinfo,
1252 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1253 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1254 	if (inp != NULL) {
1255 		INP_RLOCK(inp);
1256 		INP_INFO_RUNLOCK(&V_tcbinfo);
1257 		if (inp->inp_socket == NULL)
1258 			error = ENOENT;
1259 		if (error == 0)
1260 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1261 		if (error == 0)
1262 			cru2x(inp->inp_cred, &xuc);
1263 		INP_RUNLOCK(inp);
1264 	} else {
1265 		INP_INFO_RUNLOCK(&V_tcbinfo);
1266 		error = ENOENT;
1267 	}
1268 	if (error == 0)
1269 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1270 	return (error);
1271 }
1272 
1273 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1274     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1275     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1276 #endif
1277 
1278 
1279 void
1280 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1281 {
1282 	struct ip *ip = vip;
1283 	struct tcphdr *th;
1284 	struct in_addr faddr;
1285 	struct inpcb *inp;
1286 	struct tcpcb *tp;
1287 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1288 	struct icmp *icp;
1289 	struct in_conninfo inc;
1290 	tcp_seq icmp_tcp_seq;
1291 	int mtu;
1292 
1293 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1294 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1295 		return;
1296 
1297 	if (cmd == PRC_MSGSIZE)
1298 		notify = tcp_mtudisc;
1299 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1300 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1301 		notify = tcp_drop_syn_sent;
1302 	/*
1303 	 * Redirects don't need to be handled up here.
1304 	 */
1305 	else if (PRC_IS_REDIRECT(cmd))
1306 		return;
1307 	/*
1308 	 * Source quench is depreciated.
1309 	 */
1310 	else if (cmd == PRC_QUENCH)
1311 		return;
1312 	/*
1313 	 * Hostdead is ugly because it goes linearly through all PCBs.
1314 	 * XXX: We never get this from ICMP, otherwise it makes an
1315 	 * excellent DoS attack on machines with many connections.
1316 	 */
1317 	else if (cmd == PRC_HOSTDEAD)
1318 		ip = NULL;
1319 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1320 		return;
1321 	if (ip != NULL) {
1322 		icp = (struct icmp *)((caddr_t)ip
1323 				      - offsetof(struct icmp, icmp_ip));
1324 		th = (struct tcphdr *)((caddr_t)ip
1325 				       + (ip->ip_hl << 2));
1326 		INP_INFO_WLOCK(&V_tcbinfo);
1327 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1328 		    ip->ip_src, th->th_sport, 0, NULL);
1329 		if (inp != NULL)  {
1330 			INP_WLOCK(inp);
1331 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1332 			    !(inp->inp_flags & INP_DROPPED) &&
1333 			    !(inp->inp_socket == NULL)) {
1334 				icmp_tcp_seq = htonl(th->th_seq);
1335 				tp = intotcpcb(inp);
1336 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1337 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1338 					if (cmd == PRC_MSGSIZE) {
1339 					    /*
1340 					     * MTU discovery:
1341 					     * If we got a needfrag set the MTU
1342 					     * in the route to the suggested new
1343 					     * value (if given) and then notify.
1344 					     */
1345 					    bzero(&inc, sizeof(inc));
1346 					    inc.inc_faddr = faddr;
1347 					    inc.inc_fibnum =
1348 						inp->inp_inc.inc_fibnum;
1349 
1350 					    mtu = ntohs(icp->icmp_nextmtu);
1351 					    /*
1352 					     * If no alternative MTU was
1353 					     * proposed, try the next smaller
1354 					     * one.  ip->ip_len has already
1355 					     * been swapped in icmp_input().
1356 					     */
1357 					    if (!mtu)
1358 						mtu = ip_next_mtu(ip->ip_len,
1359 						 1);
1360 					    if (mtu < V_tcp_minmss
1361 						 + sizeof(struct tcpiphdr))
1362 						mtu = V_tcp_minmss
1363 						 + sizeof(struct tcpiphdr);
1364 					    /*
1365 					     * Only cache the the MTU if it
1366 					     * is smaller than the interface
1367 					     * or route MTU.  tcp_mtudisc()
1368 					     * will do right thing by itself.
1369 					     */
1370 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1371 						tcp_hc_updatemtu(&inc, mtu);
1372 					}
1373 
1374 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1375 				}
1376 			}
1377 			if (inp != NULL)
1378 				INP_WUNLOCK(inp);
1379 		} else {
1380 			bzero(&inc, sizeof(inc));
1381 			inc.inc_fport = th->th_dport;
1382 			inc.inc_lport = th->th_sport;
1383 			inc.inc_faddr = faddr;
1384 			inc.inc_laddr = ip->ip_src;
1385 			syncache_unreach(&inc, th);
1386 		}
1387 		INP_INFO_WUNLOCK(&V_tcbinfo);
1388 	} else
1389 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1390 }
1391 
1392 #ifdef INET6
1393 void
1394 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1395 {
1396 	struct tcphdr th;
1397 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1398 	struct ip6_hdr *ip6;
1399 	struct mbuf *m;
1400 	struct ip6ctlparam *ip6cp = NULL;
1401 	const struct sockaddr_in6 *sa6_src = NULL;
1402 	int off;
1403 	struct tcp_portonly {
1404 		u_int16_t th_sport;
1405 		u_int16_t th_dport;
1406 	} *thp;
1407 
1408 	if (sa->sa_family != AF_INET6 ||
1409 	    sa->sa_len != sizeof(struct sockaddr_in6))
1410 		return;
1411 
1412 	if (cmd == PRC_MSGSIZE)
1413 		notify = tcp_mtudisc;
1414 	else if (!PRC_IS_REDIRECT(cmd) &&
1415 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1416 		return;
1417 	/* Source quench is depreciated. */
1418 	else if (cmd == PRC_QUENCH)
1419 		return;
1420 
1421 	/* if the parameter is from icmp6, decode it. */
1422 	if (d != NULL) {
1423 		ip6cp = (struct ip6ctlparam *)d;
1424 		m = ip6cp->ip6c_m;
1425 		ip6 = ip6cp->ip6c_ip6;
1426 		off = ip6cp->ip6c_off;
1427 		sa6_src = ip6cp->ip6c_src;
1428 	} else {
1429 		m = NULL;
1430 		ip6 = NULL;
1431 		off = 0;	/* fool gcc */
1432 		sa6_src = &sa6_any;
1433 	}
1434 
1435 	if (ip6 != NULL) {
1436 		struct in_conninfo inc;
1437 		/*
1438 		 * XXX: We assume that when IPV6 is non NULL,
1439 		 * M and OFF are valid.
1440 		 */
1441 
1442 		/* check if we can safely examine src and dst ports */
1443 		if (m->m_pkthdr.len < off + sizeof(*thp))
1444 			return;
1445 
1446 		bzero(&th, sizeof(th));
1447 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1448 
1449 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1450 		    (struct sockaddr *)ip6cp->ip6c_src,
1451 		    th.th_sport, cmd, NULL, notify);
1452 
1453 		bzero(&inc, sizeof(inc));
1454 		inc.inc_fport = th.th_dport;
1455 		inc.inc_lport = th.th_sport;
1456 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1457 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1458 		inc.inc_flags |= INC_ISIPV6;
1459 		INP_INFO_WLOCK(&V_tcbinfo);
1460 		syncache_unreach(&inc, &th);
1461 		INP_INFO_WUNLOCK(&V_tcbinfo);
1462 	} else
1463 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1464 			      0, cmd, NULL, notify);
1465 }
1466 #endif /* INET6 */
1467 
1468 
1469 /*
1470  * Following is where TCP initial sequence number generation occurs.
1471  *
1472  * There are two places where we must use initial sequence numbers:
1473  * 1.  In SYN-ACK packets.
1474  * 2.  In SYN packets.
1475  *
1476  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1477  * tcp_syncache.c for details.
1478  *
1479  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1480  * depends on this property.  In addition, these ISNs should be
1481  * unguessable so as to prevent connection hijacking.  To satisfy
1482  * the requirements of this situation, the algorithm outlined in
1483  * RFC 1948 is used, with only small modifications.
1484  *
1485  * Implementation details:
1486  *
1487  * Time is based off the system timer, and is corrected so that it
1488  * increases by one megabyte per second.  This allows for proper
1489  * recycling on high speed LANs while still leaving over an hour
1490  * before rollover.
1491  *
1492  * As reading the *exact* system time is too expensive to be done
1493  * whenever setting up a TCP connection, we increment the time
1494  * offset in two ways.  First, a small random positive increment
1495  * is added to isn_offset for each connection that is set up.
1496  * Second, the function tcp_isn_tick fires once per clock tick
1497  * and increments isn_offset as necessary so that sequence numbers
1498  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1499  * random positive increments serve only to ensure that the same
1500  * exact sequence number is never sent out twice (as could otherwise
1501  * happen when a port is recycled in less than the system tick
1502  * interval.)
1503  *
1504  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1505  * between seeding of isn_secret.  This is normally set to zero,
1506  * as reseeding should not be necessary.
1507  *
1508  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1509  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1510  * general, this means holding an exclusive (write) lock.
1511  */
1512 
1513 #define ISN_BYTES_PER_SECOND 1048576
1514 #define ISN_STATIC_INCREMENT 4096
1515 #define ISN_RANDOM_INCREMENT (4096 - 1)
1516 
1517 static VNET_DEFINE(u_char, isn_secret[32]);
1518 static VNET_DEFINE(int, isn_last_reseed);
1519 static VNET_DEFINE(u_int32_t, isn_offset);
1520 static VNET_DEFINE(u_int32_t, isn_offset_old);
1521 
1522 #define	V_isn_secret			VNET(isn_secret)
1523 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1524 #define	V_isn_offset			VNET(isn_offset)
1525 #define	V_isn_offset_old		VNET(isn_offset_old)
1526 
1527 tcp_seq
1528 tcp_new_isn(struct tcpcb *tp)
1529 {
1530 	MD5_CTX isn_ctx;
1531 	u_int32_t md5_buffer[4];
1532 	tcp_seq new_isn;
1533 
1534 	INP_WLOCK_ASSERT(tp->t_inpcb);
1535 
1536 	ISN_LOCK();
1537 	/* Seed if this is the first use, reseed if requested. */
1538 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1539 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1540 		< (u_int)ticks))) {
1541 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1542 		V_isn_last_reseed = ticks;
1543 	}
1544 
1545 	/* Compute the md5 hash and return the ISN. */
1546 	MD5Init(&isn_ctx);
1547 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1548 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1549 #ifdef INET6
1550 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1551 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1552 			  sizeof(struct in6_addr));
1553 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1554 			  sizeof(struct in6_addr));
1555 	} else
1556 #endif
1557 	{
1558 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1559 			  sizeof(struct in_addr));
1560 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1561 			  sizeof(struct in_addr));
1562 	}
1563 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1564 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1565 	new_isn = (tcp_seq) md5_buffer[0];
1566 	V_isn_offset += ISN_STATIC_INCREMENT +
1567 		(arc4random() & ISN_RANDOM_INCREMENT);
1568 	new_isn += V_isn_offset;
1569 	ISN_UNLOCK();
1570 	return (new_isn);
1571 }
1572 
1573 /*
1574  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1575  * to keep time flowing at a relatively constant rate.  If the random
1576  * increments have already pushed us past the projected offset, do nothing.
1577  */
1578 static void
1579 tcp_isn_tick(void *xtp)
1580 {
1581 	VNET_ITERATOR_DECL(vnet_iter);
1582 	u_int32_t projected_offset;
1583 
1584 	VNET_LIST_RLOCK_NOSLEEP();
1585 	ISN_LOCK();
1586 	VNET_FOREACH(vnet_iter) {
1587 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1588 		projected_offset =
1589 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1590 
1591 		if (SEQ_GT(projected_offset, V_isn_offset))
1592 			V_isn_offset = projected_offset;
1593 
1594 		V_isn_offset_old = V_isn_offset;
1595 		CURVNET_RESTORE();
1596 	}
1597 	ISN_UNLOCK();
1598 	VNET_LIST_RUNLOCK_NOSLEEP();
1599 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1600 }
1601 
1602 /*
1603  * When a specific ICMP unreachable message is received and the
1604  * connection state is SYN-SENT, drop the connection.  This behavior
1605  * is controlled by the icmp_may_rst sysctl.
1606  */
1607 struct inpcb *
1608 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1609 {
1610 	struct tcpcb *tp;
1611 
1612 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1613 	INP_WLOCK_ASSERT(inp);
1614 
1615 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1616 	    (inp->inp_flags & INP_DROPPED))
1617 		return (inp);
1618 
1619 	tp = intotcpcb(inp);
1620 	if (tp->t_state != TCPS_SYN_SENT)
1621 		return (inp);
1622 
1623 	tp = tcp_drop(tp, errno);
1624 	if (tp != NULL)
1625 		return (inp);
1626 	else
1627 		return (NULL);
1628 }
1629 
1630 /*
1631  * When `need fragmentation' ICMP is received, update our idea of the MSS
1632  * based on the new value in the route.  Also nudge TCP to send something,
1633  * since we know the packet we just sent was dropped.
1634  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1635  */
1636 struct inpcb *
1637 tcp_mtudisc(struct inpcb *inp, int errno)
1638 {
1639 	struct tcpcb *tp;
1640 	struct socket *so;
1641 
1642 	INP_WLOCK_ASSERT(inp);
1643 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1644 	    (inp->inp_flags & INP_DROPPED))
1645 		return (inp);
1646 
1647 	tp = intotcpcb(inp);
1648 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1649 
1650 	tcp_mss_update(tp, -1, NULL, NULL);
1651 
1652 	so = inp->inp_socket;
1653 	SOCKBUF_LOCK(&so->so_snd);
1654 	/* If the mss is larger than the socket buffer, decrease the mss. */
1655 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1656 		tp->t_maxseg = so->so_snd.sb_hiwat;
1657 	SOCKBUF_UNLOCK(&so->so_snd);
1658 
1659 	TCPSTAT_INC(tcps_mturesent);
1660 	tp->t_rtttime = 0;
1661 	tp->snd_nxt = tp->snd_una;
1662 	tcp_free_sackholes(tp);
1663 	tp->snd_recover = tp->snd_max;
1664 	if (tp->t_flags & TF_SACK_PERMIT)
1665 		EXIT_FASTRECOVERY(tp->t_flags);
1666 	tcp_output_send(tp);
1667 	return (inp);
1668 }
1669 
1670 /*
1671  * Look-up the routing entry to the peer of this inpcb.  If no route
1672  * is found and it cannot be allocated, then return 0.  This routine
1673  * is called by TCP routines that access the rmx structure and by
1674  * tcp_mss_update to get the peer/interface MTU.
1675  */
1676 u_long
1677 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1678 {
1679 	struct route sro;
1680 	struct sockaddr_in *dst;
1681 	struct ifnet *ifp;
1682 	u_long maxmtu = 0;
1683 
1684 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1685 
1686 	bzero(&sro, sizeof(sro));
1687 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1688 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1689 		dst->sin_family = AF_INET;
1690 		dst->sin_len = sizeof(*dst);
1691 		dst->sin_addr = inc->inc_faddr;
1692 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1693 	}
1694 	if (sro.ro_rt != NULL) {
1695 		ifp = sro.ro_rt->rt_ifp;
1696 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1697 			maxmtu = ifp->if_mtu;
1698 		else
1699 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1700 
1701 		/* Report additional interface capabilities. */
1702 		if (flags != NULL) {
1703 			if (ifp->if_capenable & IFCAP_TSO4 &&
1704 			    ifp->if_hwassist & CSUM_TSO)
1705 				*flags |= CSUM_TSO;
1706 		}
1707 		RTFREE(sro.ro_rt);
1708 	}
1709 	return (maxmtu);
1710 }
1711 
1712 #ifdef INET6
1713 u_long
1714 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1715 {
1716 	struct route_in6 sro6;
1717 	struct ifnet *ifp;
1718 	u_long maxmtu = 0;
1719 
1720 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1721 
1722 	bzero(&sro6, sizeof(sro6));
1723 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1724 		sro6.ro_dst.sin6_family = AF_INET6;
1725 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1726 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1727 		rtalloc_ign((struct route *)&sro6, 0);
1728 	}
1729 	if (sro6.ro_rt != NULL) {
1730 		ifp = sro6.ro_rt->rt_ifp;
1731 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1732 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1733 		else
1734 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1735 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1736 
1737 		/* Report additional interface capabilities. */
1738 		if (flags != NULL) {
1739 			if (ifp->if_capenable & IFCAP_TSO6 &&
1740 			    ifp->if_hwassist & CSUM_TSO)
1741 				*flags |= CSUM_TSO;
1742 		}
1743 		RTFREE(sro6.ro_rt);
1744 	}
1745 
1746 	return (maxmtu);
1747 }
1748 #endif /* INET6 */
1749 
1750 #ifdef IPSEC
1751 /* compute ESP/AH header size for TCP, including outer IP header. */
1752 size_t
1753 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1754 {
1755 	struct inpcb *inp;
1756 	struct mbuf *m;
1757 	size_t hdrsiz;
1758 	struct ip *ip;
1759 #ifdef INET6
1760 	struct ip6_hdr *ip6;
1761 #endif
1762 	struct tcphdr *th;
1763 
1764 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1765 		return (0);
1766 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1767 	if (!m)
1768 		return (0);
1769 
1770 #ifdef INET6
1771 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1772 		ip6 = mtod(m, struct ip6_hdr *);
1773 		th = (struct tcphdr *)(ip6 + 1);
1774 		m->m_pkthdr.len = m->m_len =
1775 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1776 		tcpip_fillheaders(inp, ip6, th);
1777 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1778 	} else
1779 #endif /* INET6 */
1780 	{
1781 		ip = mtod(m, struct ip *);
1782 		th = (struct tcphdr *)(ip + 1);
1783 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1784 		tcpip_fillheaders(inp, ip, th);
1785 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1786 	}
1787 
1788 	m_free(m);
1789 	return (hdrsiz);
1790 }
1791 #endif /* IPSEC */
1792 
1793 #ifdef TCP_SIGNATURE
1794 /*
1795  * Callback function invoked by m_apply() to digest TCP segment data
1796  * contained within an mbuf chain.
1797  */
1798 static int
1799 tcp_signature_apply(void *fstate, void *data, u_int len)
1800 {
1801 
1802 	MD5Update(fstate, (u_char *)data, len);
1803 	return (0);
1804 }
1805 
1806 /*
1807  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1808  *
1809  * Parameters:
1810  * m		pointer to head of mbuf chain
1811  * _unused
1812  * len		length of TCP segment data, excluding options
1813  * optlen	length of TCP segment options
1814  * buf		pointer to storage for computed MD5 digest
1815  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1816  *
1817  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1818  * When called from tcp_input(), we can be sure that th_sum has been
1819  * zeroed out and verified already.
1820  *
1821  * Return 0 if successful, otherwise return -1.
1822  *
1823  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1824  * search with the destination IP address, and a 'magic SPI' to be
1825  * determined by the application. This is hardcoded elsewhere to 1179
1826  * right now. Another branch of this code exists which uses the SPD to
1827  * specify per-application flows but it is unstable.
1828  */
1829 int
1830 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1831     u_char *buf, u_int direction)
1832 {
1833 	union sockaddr_union dst;
1834 	struct ippseudo ippseudo;
1835 	MD5_CTX ctx;
1836 	int doff;
1837 	struct ip *ip;
1838 	struct ipovly *ipovly;
1839 	struct secasvar *sav;
1840 	struct tcphdr *th;
1841 #ifdef INET6
1842 	struct ip6_hdr *ip6;
1843 	struct in6_addr in6;
1844 	char ip6buf[INET6_ADDRSTRLEN];
1845 	uint32_t plen;
1846 	uint16_t nhdr;
1847 #endif
1848 	u_short savecsum;
1849 
1850 	KASSERT(m != NULL, ("NULL mbuf chain"));
1851 	KASSERT(buf != NULL, ("NULL signature pointer"));
1852 
1853 	/* Extract the destination from the IP header in the mbuf. */
1854 	bzero(&dst, sizeof(union sockaddr_union));
1855 	ip = mtod(m, struct ip *);
1856 #ifdef INET6
1857 	ip6 = NULL;	/* Make the compiler happy. */
1858 #endif
1859 	switch (ip->ip_v) {
1860 	case IPVERSION:
1861 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1862 		dst.sa.sa_family = AF_INET;
1863 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1864 		    ip->ip_src : ip->ip_dst;
1865 		break;
1866 #ifdef INET6
1867 	case (IPV6_VERSION >> 4):
1868 		ip6 = mtod(m, struct ip6_hdr *);
1869 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1870 		dst.sa.sa_family = AF_INET6;
1871 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1872 		    ip6->ip6_src : ip6->ip6_dst;
1873 		break;
1874 #endif
1875 	default:
1876 		return (EINVAL);
1877 		/* NOTREACHED */
1878 		break;
1879 	}
1880 
1881 	/* Look up an SADB entry which matches the address of the peer. */
1882 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1883 	if (sav == NULL) {
1884 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1885 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1886 #ifdef INET6
1887 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
1888 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1889 #endif
1890 			"(unsupported)"));
1891 		return (EINVAL);
1892 	}
1893 
1894 	MD5Init(&ctx);
1895 	/*
1896 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1897 	 *
1898 	 * XXX The ippseudo header MUST be digested in network byte order,
1899 	 * or else we'll fail the regression test. Assume all fields we've
1900 	 * been doing arithmetic on have been in host byte order.
1901 	 * XXX One cannot depend on ipovly->ih_len here. When called from
1902 	 * tcp_output(), the underlying ip_len member has not yet been set.
1903 	 */
1904 	switch (ip->ip_v) {
1905 	case IPVERSION:
1906 		ipovly = (struct ipovly *)ip;
1907 		ippseudo.ippseudo_src = ipovly->ih_src;
1908 		ippseudo.ippseudo_dst = ipovly->ih_dst;
1909 		ippseudo.ippseudo_pad = 0;
1910 		ippseudo.ippseudo_p = IPPROTO_TCP;
1911 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1912 		    optlen);
1913 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1914 
1915 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1916 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1917 		break;
1918 #ifdef INET6
1919 	/*
1920 	 * RFC 2385, 2.0  Proposal
1921 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1922 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1923 	 * extended next header value (to form 32 bits), and 32-bit segment
1924 	 * length.
1925 	 * Note: Upper-Layer Packet Length comes before Next Header.
1926 	 */
1927 	case (IPV6_VERSION >> 4):
1928 		in6 = ip6->ip6_src;
1929 		in6_clearscope(&in6);
1930 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1931 		in6 = ip6->ip6_dst;
1932 		in6_clearscope(&in6);
1933 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1934 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
1935 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1936 		nhdr = 0;
1937 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1938 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1939 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1940 		nhdr = IPPROTO_TCP;
1941 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1942 
1943 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1944 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1945 		break;
1946 #endif
1947 	default:
1948 		return (EINVAL);
1949 		/* NOTREACHED */
1950 		break;
1951 	}
1952 
1953 
1954 	/*
1955 	 * Step 2: Update MD5 hash with TCP header, excluding options.
1956 	 * The TCP checksum must be set to zero.
1957 	 */
1958 	savecsum = th->th_sum;
1959 	th->th_sum = 0;
1960 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1961 	th->th_sum = savecsum;
1962 
1963 	/*
1964 	 * Step 3: Update MD5 hash with TCP segment data.
1965 	 *         Use m_apply() to avoid an early m_pullup().
1966 	 */
1967 	if (len > 0)
1968 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
1969 
1970 	/*
1971 	 * Step 4: Update MD5 hash with shared secret.
1972 	 */
1973 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1974 	MD5Final(buf, &ctx);
1975 
1976 	key_sa_recordxfer(sav, m);
1977 	KEY_FREESAV(&sav);
1978 	return (0);
1979 }
1980 #endif /* TCP_SIGNATURE */
1981 
1982 static int
1983 sysctl_drop(SYSCTL_HANDLER_ARGS)
1984 {
1985 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
1986 	struct sockaddr_storage addrs[2];
1987 	struct inpcb *inp;
1988 	struct tcpcb *tp;
1989 	struct tcptw *tw;
1990 	struct sockaddr_in *fin, *lin;
1991 #ifdef INET6
1992 	struct sockaddr_in6 *fin6, *lin6;
1993 #endif
1994 	int error;
1995 
1996 	inp = NULL;
1997 	fin = lin = NULL;
1998 #ifdef INET6
1999 	fin6 = lin6 = NULL;
2000 #endif
2001 	error = 0;
2002 
2003 	if (req->oldptr != NULL || req->oldlen != 0)
2004 		return (EINVAL);
2005 	if (req->newptr == NULL)
2006 		return (EPERM);
2007 	if (req->newlen < sizeof(addrs))
2008 		return (ENOMEM);
2009 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2010 	if (error)
2011 		return (error);
2012 
2013 	switch (addrs[0].ss_family) {
2014 #ifdef INET6
2015 	case AF_INET6:
2016 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2017 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2018 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2019 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2020 			return (EINVAL);
2021 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2022 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2023 				return (EINVAL);
2024 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2025 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2026 			fin = (struct sockaddr_in *)&addrs[0];
2027 			lin = (struct sockaddr_in *)&addrs[1];
2028 			break;
2029 		}
2030 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2031 		if (error)
2032 			return (error);
2033 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2034 		if (error)
2035 			return (error);
2036 		break;
2037 #endif
2038 	case AF_INET:
2039 		fin = (struct sockaddr_in *)&addrs[0];
2040 		lin = (struct sockaddr_in *)&addrs[1];
2041 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2042 		    lin->sin_len != sizeof(struct sockaddr_in))
2043 			return (EINVAL);
2044 		break;
2045 	default:
2046 		return (EINVAL);
2047 	}
2048 	INP_INFO_WLOCK(&V_tcbinfo);
2049 	switch (addrs[0].ss_family) {
2050 #ifdef INET6
2051 	case AF_INET6:
2052 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2053 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2054 		    NULL);
2055 		break;
2056 #endif
2057 	case AF_INET:
2058 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2059 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2060 		break;
2061 	}
2062 	if (inp != NULL) {
2063 		INP_WLOCK(inp);
2064 		if (inp->inp_flags & INP_TIMEWAIT) {
2065 			/*
2066 			 * XXXRW: There currently exists a state where an
2067 			 * inpcb is present, but its timewait state has been
2068 			 * discarded.  For now, don't allow dropping of this
2069 			 * type of inpcb.
2070 			 */
2071 			tw = intotw(inp);
2072 			if (tw != NULL)
2073 				tcp_twclose(tw, 0);
2074 			else
2075 				INP_WUNLOCK(inp);
2076 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2077 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2078 			tp = intotcpcb(inp);
2079 			tp = tcp_drop(tp, ECONNABORTED);
2080 			if (tp != NULL)
2081 				INP_WUNLOCK(inp);
2082 		} else
2083 			INP_WUNLOCK(inp);
2084 	} else
2085 		error = ESRCH;
2086 	INP_INFO_WUNLOCK(&V_tcbinfo);
2087 	return (error);
2088 }
2089 
2090 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2091     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2092     0, sysctl_drop, "", "Drop TCP connection");
2093 
2094 /*
2095  * Generate a standardized TCP log line for use throughout the
2096  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2097  * allow use in the interrupt context.
2098  *
2099  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2100  * NB: The function may return NULL if memory allocation failed.
2101  *
2102  * Due to header inclusion and ordering limitations the struct ip
2103  * and ip6_hdr pointers have to be passed as void pointers.
2104  */
2105 char *
2106 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2107     const void *ip6hdr)
2108 {
2109 
2110 	/* Is logging enabled? */
2111 	if (tcp_log_in_vain == 0)
2112 		return (NULL);
2113 
2114 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2115 }
2116 
2117 char *
2118 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2119     const void *ip6hdr)
2120 {
2121 
2122 	/* Is logging enabled? */
2123 	if (tcp_log_debug == 0)
2124 		return (NULL);
2125 
2126 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2127 }
2128 
2129 static char *
2130 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2131     const void *ip6hdr)
2132 {
2133 	char *s, *sp;
2134 	size_t size;
2135 	struct ip *ip;
2136 #ifdef INET6
2137 	const struct ip6_hdr *ip6;
2138 
2139 	ip6 = (const struct ip6_hdr *)ip6hdr;
2140 #endif /* INET6 */
2141 	ip = (struct ip *)ip4hdr;
2142 
2143 	/*
2144 	 * The log line looks like this:
2145 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2146 	 */
2147 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2148 	    sizeof(PRINT_TH_FLAGS) + 1 +
2149 #ifdef INET6
2150 	    2 * INET6_ADDRSTRLEN;
2151 #else
2152 	    2 * INET_ADDRSTRLEN;
2153 #endif /* INET6 */
2154 
2155 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2156 	if (s == NULL)
2157 		return (NULL);
2158 
2159 	strcat(s, "TCP: [");
2160 	sp = s + strlen(s);
2161 
2162 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2163 		inet_ntoa_r(inc->inc_faddr, sp);
2164 		sp = s + strlen(s);
2165 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2166 		sp = s + strlen(s);
2167 		inet_ntoa_r(inc->inc_laddr, sp);
2168 		sp = s + strlen(s);
2169 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2170 #ifdef INET6
2171 	} else if (inc) {
2172 		ip6_sprintf(sp, &inc->inc6_faddr);
2173 		sp = s + strlen(s);
2174 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2175 		sp = s + strlen(s);
2176 		ip6_sprintf(sp, &inc->inc6_laddr);
2177 		sp = s + strlen(s);
2178 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2179 	} else if (ip6 && th) {
2180 		ip6_sprintf(sp, &ip6->ip6_src);
2181 		sp = s + strlen(s);
2182 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2183 		sp = s + strlen(s);
2184 		ip6_sprintf(sp, &ip6->ip6_dst);
2185 		sp = s + strlen(s);
2186 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2187 #endif /* INET6 */
2188 	} else if (ip && th) {
2189 		inet_ntoa_r(ip->ip_src, sp);
2190 		sp = s + strlen(s);
2191 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2192 		sp = s + strlen(s);
2193 		inet_ntoa_r(ip->ip_dst, sp);
2194 		sp = s + strlen(s);
2195 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2196 	} else {
2197 		free(s, M_TCPLOG);
2198 		return (NULL);
2199 	}
2200 	sp = s + strlen(s);
2201 	if (th)
2202 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2203 	if (*(s + size - 1) != '\0')
2204 		panic("%s: string too long", __func__);
2205 	return (s);
2206 }
2207