xref: /freebsd/sys/netinet/tcp_subr.c (revision 8957464be7644ab4da9559ab479c9a996fbcbd29)
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_mac.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.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 #include <sys/vimage.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/route.h>
63 #include <net/if.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 <netinet/vinet.h>
98 #include <netinet6/ip6protosw.h>
99 #include <netinet6/vinet6.h>
100 
101 #ifdef IPSEC
102 #include <netipsec/ipsec.h>
103 #include <netipsec/xform.h>
104 #ifdef INET6
105 #include <netipsec/ipsec6.h>
106 #endif
107 #include <netipsec/key.h>
108 #include <sys/syslog.h>
109 #endif /*IPSEC*/
110 
111 #include <machine/in_cksum.h>
112 #include <sys/md5.h>
113 
114 #include <security/mac/mac_framework.h>
115 
116 #ifdef VIMAGE_GLOBALS
117 int	tcp_mssdflt;
118 #ifdef INET6
119 int	tcp_v6mssdflt;
120 #endif
121 int	tcp_minmss;
122 int	tcp_do_rfc1323;
123 static int	icmp_may_rst;
124 static int	tcp_isn_reseed_interval;
125 static int	tcp_inflight_enable;
126 static int	tcp_inflight_rttthresh;
127 static int	tcp_inflight_min;
128 static int	tcp_inflight_max;
129 static int	tcp_inflight_stab;
130 #endif
131 
132 static int
133 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
134 {
135 	INIT_VNET_INET(curvnet);
136 	int error, new;
137 
138 	new = V_tcp_mssdflt;
139 	error = sysctl_handle_int(oidp, &new, 0, req);
140 	if (error == 0 && req->newptr) {
141 		if (new < TCP_MINMSS)
142 			error = EINVAL;
143 		else
144 			V_tcp_mssdflt = new;
145 	}
146 	return (error);
147 }
148 
149 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
150     CTLTYPE_INT|CTLFLAG_RW, tcp_mssdflt, 0,
151     &sysctl_net_inet_tcp_mss_check, "I",
152     "Default TCP Maximum Segment Size");
153 
154 #ifdef INET6
155 static int
156 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
157 {
158 	INIT_VNET_INET(curvnet);
159 	int error, new;
160 
161 	new = V_tcp_v6mssdflt;
162 	error = sysctl_handle_int(oidp, &new, 0, req);
163 	if (error == 0 && req->newptr) {
164 		if (new < TCP_MINMSS)
165 			error = EINVAL;
166 		else
167 			V_tcp_v6mssdflt = new;
168 	}
169 	return (error);
170 }
171 
172 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
173     CTLTYPE_INT|CTLFLAG_RW, tcp_v6mssdflt, 0,
174     &sysctl_net_inet_tcp_mss_v6_check, "I",
175    "Default TCP Maximum Segment Size for IPv6");
176 #endif
177 
178 /*
179  * Minimum MSS we accept and use. This prevents DoS attacks where
180  * we are forced to a ridiculous low MSS like 20 and send hundreds
181  * of packets instead of one. The effect scales with the available
182  * bandwidth and quickly saturates the CPU and network interface
183  * with packet generation and sending. Set to zero to disable MINMSS
184  * checking. This setting prevents us from sending too small packets.
185  */
186 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
187     CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
188 
189 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
190     CTLFLAG_RW, tcp_do_rfc1323, 0,
191     "Enable rfc1323 (high performance TCP) extensions");
192 
193 static int	tcp_log_debug = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
195     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
196 
197 static int	tcp_tcbhashsize = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
199     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
200 
201 static int	do_tcpdrain = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
203     "Enable tcp_drain routine for extra help when low on mbufs");
204 
205 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
206     CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs");
207 
208 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
209     CTLFLAG_RW, icmp_may_rst, 0,
210     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
211 
212 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
213     CTLFLAG_RW, tcp_isn_reseed_interval, 0,
214     "Seconds between reseeding of ISN secret");
215 
216 /*
217  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
218  * 1024 exists only for debugging.  A good production default would be
219  * something like 6100.
220  */
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222     "TCP inflight data limiting");
223 
224 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
225     CTLFLAG_RW, tcp_inflight_enable, 0,
226     "Enable automatic TCP inflight data limiting");
227 
228 static int	tcp_inflight_debug = 0;
229 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
230     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
231 
232 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, rttthresh,
233     CTLTYPE_INT|CTLFLAG_RW, tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks,
234     "I", "RTT threshold below which inflight will deactivate itself");
235 
236 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
237     CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
238 
239 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
240     CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
241 
242 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
243     CTLFLAG_RW, tcp_inflight_stab, 0,
244     "Inflight Algorithm Stabilization 20 = 2 packets");
245 
246 #ifdef VIMAGE_GLOBALS
247 uma_zone_t sack_hole_zone;
248 #endif
249 
250 static struct inpcb *tcp_notify(struct inpcb *, int);
251 static void	tcp_isn_tick(void *);
252 
253 /*
254  * Target size of TCP PCB hash tables. Must be a power of two.
255  *
256  * Note that this can be overridden by the kernel environment
257  * variable net.inet.tcp.tcbhashsize
258  */
259 #ifndef TCBHASHSIZE
260 #define TCBHASHSIZE	512
261 #endif
262 
263 /*
264  * XXX
265  * Callouts should be moved into struct tcp directly.  They are currently
266  * separate because the tcpcb structure is exported to userland for sysctl
267  * parsing purposes, which do not know about callouts.
268  */
269 struct tcpcb_mem {
270 	struct	tcpcb		tcb;
271 	struct	tcp_timer	tt;
272 };
273 
274 #ifdef VIMAGE_GLOBALS
275 static uma_zone_t tcpcb_zone;
276 #endif
277 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
278 struct callout isn_callout;
279 static struct mtx isn_mtx;
280 
281 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
282 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
283 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
284 
285 /*
286  * TCP initialization.
287  */
288 static void
289 tcp_zone_change(void *tag)
290 {
291 	INIT_VNET_INET(curvnet);
292 
293 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
294 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
295 	tcp_tw_zone_change();
296 }
297 
298 static int
299 tcp_inpcb_init(void *mem, int size, int flags)
300 {
301 	struct inpcb *inp = mem;
302 
303 	INP_LOCK_INIT(inp, "inp", "tcpinp");
304 	return (0);
305 }
306 
307 void
308 tcp_init(void)
309 {
310 	INIT_VNET_INET(curvnet);
311 	int hashsize;
312 
313 	V_blackhole = 0;
314 	V_tcp_delack_enabled = 1;
315 	V_drop_synfin = 0;
316 	V_tcp_do_rfc3042 = 1;
317 	V_tcp_do_rfc3390 = 1;
318 	V_tcp_do_ecn = 0;
319 	V_tcp_ecn_maxretries = 1;
320 	V_tcp_insecure_rst = 0;
321 	V_tcp_do_autorcvbuf = 1;
322 	V_tcp_autorcvbuf_inc = 16*1024;
323 	V_tcp_autorcvbuf_max = 256*1024;
324 	V_tcp_do_rfc3465 = 1;
325 	V_tcp_abc_l_var = 2;
326 
327 	V_tcp_mssdflt = TCP_MSS;
328 #ifdef INET6
329 	V_tcp_v6mssdflt = TCP6_MSS;
330 #endif
331 	V_tcp_minmss = TCP_MINMSS;
332 	V_tcp_do_rfc1323 = 1;
333 	V_icmp_may_rst = 1;
334 	V_tcp_isn_reseed_interval = 0;
335 	V_tcp_inflight_enable = 1;
336 	V_tcp_inflight_min = 6144;
337 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
338 	V_tcp_inflight_stab = 20;
339 
340 	V_path_mtu_discovery = 1;
341 	V_ss_fltsz = 1;
342 	V_ss_fltsz_local = 4;
343 	V_tcp_do_newreno = 1;
344 	V_tcp_do_tso = 1;
345 	V_tcp_do_autosndbuf = 1;
346 	V_tcp_autosndbuf_inc = 8*1024;
347 	V_tcp_autosndbuf_max = 256*1024;
348 
349 	V_nolocaltimewait = 0;
350 
351 	V_tcp_do_sack = 1;
352 	V_tcp_sack_maxholes = 128;
353 	V_tcp_sack_globalmaxholes = 65536;
354 	V_tcp_sack_globalholes = 0;
355 
356 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
357 
358 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
359 
360 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
361 	LIST_INIT(&V_tcb);
362 	V_tcbinfo.ipi_listhead = &V_tcb;
363 	hashsize = TCBHASHSIZE;
364 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
365 	if (!powerof2(hashsize)) {
366 		printf("WARNING: TCB hash size not a power of 2\n");
367 		hashsize = 512; /* safe default */
368 	}
369 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
370 	    &V_tcbinfo.ipi_hashmask);
371 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
372 	    &V_tcbinfo.ipi_porthashmask);
373 	V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
374 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
375 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
376 	/*
377 	 * These have to be type stable for the benefit of the timers.
378 	 */
379 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
380 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
381 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
382 	tcp_tw_init();
383 	syncache_init();
384 	tcp_hc_init();
385 	tcp_reass_init();
386 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
387 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
388 
389 	/* Skip initialization of globals for non-default instances. */
390 	if (!IS_DEFAULT_VNET(curvnet))
391 		return;
392 
393 	/* XXX virtualize those bellow? */
394 	tcp_delacktime = TCPTV_DELACK;
395 	tcp_keepinit = TCPTV_KEEP_INIT;
396 	tcp_keepidle = TCPTV_KEEP_IDLE;
397 	tcp_keepintvl = TCPTV_KEEPINTVL;
398 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
399 	tcp_msl = TCPTV_MSL;
400 	tcp_rexmit_min = TCPTV_MIN;
401 	if (tcp_rexmit_min < 1)
402 		tcp_rexmit_min = 1;
403 	tcp_rexmit_slop = TCPTV_CPU_VAR;
404 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
405 	tcp_tcbhashsize = hashsize;
406 
407 #ifdef INET6
408 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
409 #else /* INET6 */
410 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
411 #endif /* INET6 */
412 	if (max_protohdr < TCP_MINPROTOHDR)
413 		max_protohdr = TCP_MINPROTOHDR;
414 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
415 		panic("tcp_init");
416 #undef TCP_MINPROTOHDR
417 
418 	ISN_LOCK_INIT();
419 	callout_init(&isn_callout, CALLOUT_MPSAFE);
420 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
421 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
422 		SHUTDOWN_PRI_DEFAULT);
423 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
424 		EVENTHANDLER_PRI_ANY);
425 }
426 
427 void
428 tcp_fini(void *xtp)
429 {
430 
431 	callout_stop(&isn_callout);
432 }
433 
434 /*
435  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
436  * tcp_template used to store this data in mbufs, but we now recopy it out
437  * of the tcpcb each time to conserve mbufs.
438  */
439 void
440 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
441 {
442 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
443 
444 	INP_WLOCK_ASSERT(inp);
445 
446 #ifdef INET6
447 	if ((inp->inp_vflag & INP_IPV6) != 0) {
448 		struct ip6_hdr *ip6;
449 
450 		ip6 = (struct ip6_hdr *)ip_ptr;
451 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
452 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
453 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
454 			(IPV6_VERSION & IPV6_VERSION_MASK);
455 		ip6->ip6_nxt = IPPROTO_TCP;
456 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
457 		ip6->ip6_src = inp->in6p_laddr;
458 		ip6->ip6_dst = inp->in6p_faddr;
459 	} else
460 #endif
461 	{
462 		struct ip *ip;
463 
464 		ip = (struct ip *)ip_ptr;
465 		ip->ip_v = IPVERSION;
466 		ip->ip_hl = 5;
467 		ip->ip_tos = inp->inp_ip_tos;
468 		ip->ip_len = 0;
469 		ip->ip_id = 0;
470 		ip->ip_off = 0;
471 		ip->ip_ttl = inp->inp_ip_ttl;
472 		ip->ip_sum = 0;
473 		ip->ip_p = IPPROTO_TCP;
474 		ip->ip_src = inp->inp_laddr;
475 		ip->ip_dst = inp->inp_faddr;
476 	}
477 	th->th_sport = inp->inp_lport;
478 	th->th_dport = inp->inp_fport;
479 	th->th_seq = 0;
480 	th->th_ack = 0;
481 	th->th_x2 = 0;
482 	th->th_off = 5;
483 	th->th_flags = 0;
484 	th->th_win = 0;
485 	th->th_urp = 0;
486 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
487 }
488 
489 /*
490  * Create template to be used to send tcp packets on a connection.
491  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
492  * use for this function is in keepalives, which use tcp_respond.
493  */
494 struct tcptemp *
495 tcpip_maketemplate(struct inpcb *inp)
496 {
497 	struct tcptemp *t;
498 
499 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
500 	if (t == NULL)
501 		return (NULL);
502 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
503 	return (t);
504 }
505 
506 /*
507  * Send a single message to the TCP at address specified by
508  * the given TCP/IP header.  If m == NULL, then we make a copy
509  * of the tcpiphdr at ti and send directly to the addressed host.
510  * This is used to force keep alive messages out using the TCP
511  * template for a connection.  If flags are given then we send
512  * a message back to the TCP which originated the * segment ti,
513  * and discard the mbuf containing it and any other attached mbufs.
514  *
515  * In any case the ack and sequence number of the transmitted
516  * segment are as specified by the parameters.
517  *
518  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
519  */
520 void
521 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
522     tcp_seq ack, tcp_seq seq, int flags)
523 {
524 	INIT_VNET_INET(curvnet);
525 	int tlen;
526 	int win = 0;
527 	struct ip *ip;
528 	struct tcphdr *nth;
529 #ifdef INET6
530 	struct ip6_hdr *ip6;
531 	int isipv6;
532 #endif /* INET6 */
533 	int ipflags = 0;
534 	struct inpcb *inp;
535 
536 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
537 
538 #ifdef INET6
539 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
540 	ip6 = ipgen;
541 #endif /* INET6 */
542 	ip = ipgen;
543 
544 	if (tp != NULL) {
545 		inp = tp->t_inpcb;
546 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
547 		INP_WLOCK_ASSERT(inp);
548 	} else
549 		inp = NULL;
550 
551 	if (tp != NULL) {
552 		if (!(flags & TH_RST)) {
553 			win = sbspace(&inp->inp_socket->so_rcv);
554 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
555 				win = (long)TCP_MAXWIN << tp->rcv_scale;
556 		}
557 	}
558 	if (m == NULL) {
559 		m = m_gethdr(M_DONTWAIT, MT_DATA);
560 		if (m == NULL)
561 			return;
562 		tlen = 0;
563 		m->m_data += max_linkhdr;
564 #ifdef INET6
565 		if (isipv6) {
566 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
567 			      sizeof(struct ip6_hdr));
568 			ip6 = mtod(m, struct ip6_hdr *);
569 			nth = (struct tcphdr *)(ip6 + 1);
570 		} else
571 #endif /* INET6 */
572 	      {
573 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
574 		ip = mtod(m, struct ip *);
575 		nth = (struct tcphdr *)(ip + 1);
576 	      }
577 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
578 		flags = TH_ACK;
579 	} else {
580 		/*
581 		 *  reuse the mbuf.
582 		 * XXX MRT We inherrit the FIB, which is lucky.
583 		 */
584 		m_freem(m->m_next);
585 		m->m_next = NULL;
586 		m->m_data = (caddr_t)ipgen;
587 		/* m_len is set later */
588 		tlen = 0;
589 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
590 #ifdef INET6
591 		if (isipv6) {
592 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
593 			nth = (struct tcphdr *)(ip6 + 1);
594 		} else
595 #endif /* INET6 */
596 	      {
597 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
598 		nth = (struct tcphdr *)(ip + 1);
599 	      }
600 		if (th != nth) {
601 			/*
602 			 * this is usually a case when an extension header
603 			 * exists between the IPv6 header and the
604 			 * TCP header.
605 			 */
606 			nth->th_sport = th->th_sport;
607 			nth->th_dport = th->th_dport;
608 		}
609 		xchg(nth->th_dport, nth->th_sport, uint16_t);
610 #undef xchg
611 	}
612 #ifdef INET6
613 	if (isipv6) {
614 		ip6->ip6_flow = 0;
615 		ip6->ip6_vfc = IPV6_VERSION;
616 		ip6->ip6_nxt = IPPROTO_TCP;
617 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
618 						tlen));
619 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
620 	} else
621 #endif
622 	{
623 		tlen += sizeof (struct tcpiphdr);
624 		ip->ip_len = tlen;
625 		ip->ip_ttl = V_ip_defttl;
626 		if (V_path_mtu_discovery)
627 			ip->ip_off |= IP_DF;
628 	}
629 	m->m_len = tlen;
630 	m->m_pkthdr.len = tlen;
631 	m->m_pkthdr.rcvif = NULL;
632 #ifdef MAC
633 	if (inp != NULL) {
634 		/*
635 		 * Packet is associated with a socket, so allow the
636 		 * label of the response to reflect the socket label.
637 		 */
638 		INP_WLOCK_ASSERT(inp);
639 		mac_inpcb_create_mbuf(inp, m);
640 	} else {
641 		/*
642 		 * Packet is not associated with a socket, so possibly
643 		 * update the label in place.
644 		 */
645 		mac_netinet_tcp_reply(m);
646 	}
647 #endif
648 	nth->th_seq = htonl(seq);
649 	nth->th_ack = htonl(ack);
650 	nth->th_x2 = 0;
651 	nth->th_off = sizeof (struct tcphdr) >> 2;
652 	nth->th_flags = flags;
653 	if (tp != NULL)
654 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
655 	else
656 		nth->th_win = htons((u_short)win);
657 	nth->th_urp = 0;
658 #ifdef INET6
659 	if (isipv6) {
660 		nth->th_sum = 0;
661 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
662 					sizeof(struct ip6_hdr),
663 					tlen - sizeof(struct ip6_hdr));
664 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
665 		    NULL, NULL);
666 	} else
667 #endif /* INET6 */
668 	{
669 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
670 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
671 		m->m_pkthdr.csum_flags = CSUM_TCP;
672 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
673 	}
674 #ifdef TCPDEBUG
675 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
676 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
677 #endif
678 #ifdef INET6
679 	if (isipv6)
680 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
681 	else
682 #endif /* INET6 */
683 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
684 }
685 
686 /*
687  * Create a new TCP control block, making an
688  * empty reassembly queue and hooking it to the argument
689  * protocol control block.  The `inp' parameter must have
690  * come from the zone allocator set up in tcp_init().
691  */
692 struct tcpcb *
693 tcp_newtcpcb(struct inpcb *inp)
694 {
695 	INIT_VNET_INET(inp->inp_vnet);
696 	struct tcpcb_mem *tm;
697 	struct tcpcb *tp;
698 #ifdef INET6
699 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
700 #endif /* INET6 */
701 
702 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
703 	if (tm == NULL)
704 		return (NULL);
705 	tp = &tm->tcb;
706 	tp->t_timers = &tm->tt;
707 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
708 	tp->t_maxseg = tp->t_maxopd =
709 #ifdef INET6
710 		isipv6 ? V_tcp_v6mssdflt :
711 #endif /* INET6 */
712 		V_tcp_mssdflt;
713 
714 	/* Set up our timeouts. */
715 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
716 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
717 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
718 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
719 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
720 
721 	if (V_tcp_do_rfc1323)
722 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
723 	if (V_tcp_do_sack)
724 		tp->t_flags |= TF_SACK_PERMIT;
725 	TAILQ_INIT(&tp->snd_holes);
726 	tp->t_inpcb = inp;	/* XXX */
727 	/*
728 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
729 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
730 	 * reasonable initial retransmit time.
731 	 */
732 	tp->t_srtt = TCPTV_SRTTBASE;
733 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
734 	tp->t_rttmin = tcp_rexmit_min;
735 	tp->t_rxtcur = TCPTV_RTOBASE;
736 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
737 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
738 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
739 	tp->t_rcvtime = ticks;
740 	tp->t_bw_rtttime = ticks;
741 	/*
742 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
743 	 * because the socket may be bound to an IPv6 wildcard address,
744 	 * which may match an IPv4-mapped IPv6 address.
745 	 */
746 	inp->inp_ip_ttl = V_ip_defttl;
747 	inp->inp_ppcb = tp;
748 	return (tp);		/* XXX */
749 }
750 
751 /*
752  * Drop a TCP connection, reporting
753  * the specified error.  If connection is synchronized,
754  * then send a RST to peer.
755  */
756 struct tcpcb *
757 tcp_drop(struct tcpcb *tp, int errno)
758 {
759 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
760 	struct socket *so = tp->t_inpcb->inp_socket;
761 
762 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
763 	INP_WLOCK_ASSERT(tp->t_inpcb);
764 
765 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
766 		tp->t_state = TCPS_CLOSED;
767 		(void) tcp_output_reset(tp);
768 		TCPSTAT_INC(tcps_drops);
769 	} else
770 		TCPSTAT_INC(tcps_conndrops);
771 	if (errno == ETIMEDOUT && tp->t_softerror)
772 		errno = tp->t_softerror;
773 	so->so_error = errno;
774 	return (tcp_close(tp));
775 }
776 
777 void
778 tcp_discardcb(struct tcpcb *tp)
779 {
780 	INIT_VNET_INET(tp->t_vnet);
781 	struct tseg_qent *q;
782 	struct inpcb *inp = tp->t_inpcb;
783 	struct socket *so = inp->inp_socket;
784 #ifdef INET6
785 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
786 #endif /* INET6 */
787 
788 	INP_WLOCK_ASSERT(inp);
789 
790 	/*
791 	 * Make sure that all of our timers are stopped before we
792 	 * delete the PCB.
793 	 */
794 	callout_stop(&tp->t_timers->tt_rexmt);
795 	callout_stop(&tp->t_timers->tt_persist);
796 	callout_stop(&tp->t_timers->tt_keep);
797 	callout_stop(&tp->t_timers->tt_2msl);
798 	callout_stop(&tp->t_timers->tt_delack);
799 
800 	/*
801 	 * If we got enough samples through the srtt filter,
802 	 * save the rtt and rttvar in the routing entry.
803 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
804 	 * 4 samples is enough for the srtt filter to converge
805 	 * to within enough % of the correct value; fewer samples
806 	 * and we could save a bogus rtt. The danger is not high
807 	 * as tcp quickly recovers from everything.
808 	 * XXX: Works very well but needs some more statistics!
809 	 */
810 	if (tp->t_rttupdated >= 4) {
811 		struct hc_metrics_lite metrics;
812 		u_long ssthresh;
813 
814 		bzero(&metrics, sizeof(metrics));
815 		/*
816 		 * Update the ssthresh always when the conditions below
817 		 * are satisfied. This gives us better new start value
818 		 * for the congestion avoidance for new connections.
819 		 * ssthresh is only set if packet loss occured on a session.
820 		 *
821 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
822 		 * being torn down.  Ideally this code would not use 'so'.
823 		 */
824 		ssthresh = tp->snd_ssthresh;
825 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
826 			/*
827 			 * convert the limit from user data bytes to
828 			 * packets then to packet data bytes.
829 			 */
830 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
831 			if (ssthresh < 2)
832 				ssthresh = 2;
833 			ssthresh *= (u_long)(tp->t_maxseg +
834 #ifdef INET6
835 				      (isipv6 ? sizeof (struct ip6_hdr) +
836 					       sizeof (struct tcphdr) :
837 #endif
838 				       sizeof (struct tcpiphdr)
839 #ifdef INET6
840 				       )
841 #endif
842 				      );
843 		} else
844 			ssthresh = 0;
845 		metrics.rmx_ssthresh = ssthresh;
846 
847 		metrics.rmx_rtt = tp->t_srtt;
848 		metrics.rmx_rttvar = tp->t_rttvar;
849 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
850 		metrics.rmx_bandwidth = tp->snd_bandwidth;
851 		metrics.rmx_cwnd = tp->snd_cwnd;
852 		metrics.rmx_sendpipe = 0;
853 		metrics.rmx_recvpipe = 0;
854 
855 		tcp_hc_update(&inp->inp_inc, &metrics);
856 	}
857 
858 	/* free the reassembly queue, if any */
859 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
860 		LIST_REMOVE(q, tqe_q);
861 		m_freem(q->tqe_m);
862 		uma_zfree(V_tcp_reass_zone, q);
863 		tp->t_segqlen--;
864 		V_tcp_reass_qsize--;
865 	}
866 	/* Disconnect offload device, if any. */
867 	tcp_offload_detach(tp);
868 
869 	tcp_free_sackholes(tp);
870 	inp->inp_ppcb = NULL;
871 	tp->t_inpcb = NULL;
872 	uma_zfree(V_tcpcb_zone, tp);
873 }
874 
875 /*
876  * Attempt to close a TCP control block, marking it as dropped, and freeing
877  * the socket if we hold the only reference.
878  */
879 struct tcpcb *
880 tcp_close(struct tcpcb *tp)
881 {
882 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
883 	struct inpcb *inp = tp->t_inpcb;
884 	struct socket *so;
885 
886 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
887 	INP_WLOCK_ASSERT(inp);
888 
889 	/* Notify any offload devices of listener close */
890 	if (tp->t_state == TCPS_LISTEN)
891 		tcp_offload_listen_close(tp);
892 	in_pcbdrop(inp);
893 	TCPSTAT_INC(tcps_closed);
894 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
895 	so = inp->inp_socket;
896 	soisdisconnected(so);
897 	if (inp->inp_flags & INP_SOCKREF) {
898 		KASSERT(so->so_state & SS_PROTOREF,
899 		    ("tcp_close: !SS_PROTOREF"));
900 		inp->inp_flags &= ~INP_SOCKREF;
901 		INP_WUNLOCK(inp);
902 		ACCEPT_LOCK();
903 		SOCK_LOCK(so);
904 		so->so_state &= ~SS_PROTOREF;
905 		sofree(so);
906 		return (NULL);
907 	}
908 	return (tp);
909 }
910 
911 void
912 tcp_drain(void)
913 {
914 	VNET_ITERATOR_DECL(vnet_iter);
915 
916 	if (!do_tcpdrain)
917 		return;
918 
919 	VNET_LIST_RLOCK();
920 	VNET_FOREACH(vnet_iter) {
921 		CURVNET_SET(vnet_iter);
922 		INIT_VNET_INET(vnet_iter);
923 		struct inpcb *inpb;
924 		struct tcpcb *tcpb;
925 		struct tseg_qent *te;
926 
927 	/*
928 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
929 	 * if there is one...
930 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
931 	 *      reassembly queue should be flushed, but in a situation
932 	 *	where we're really low on mbufs, this is potentially
933 	 *	usefull.
934 	 */
935 		INP_INFO_RLOCK(&V_tcbinfo);
936 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
937 			if (inpb->inp_flags & INP_TIMEWAIT)
938 				continue;
939 			INP_WLOCK(inpb);
940 			if ((tcpb = intotcpcb(inpb)) != NULL) {
941 				while ((te = LIST_FIRST(&tcpb->t_segq))
942 			            != NULL) {
943 					LIST_REMOVE(te, tqe_q);
944 					m_freem(te->tqe_m);
945 					uma_zfree(V_tcp_reass_zone, te);
946 					tcpb->t_segqlen--;
947 					V_tcp_reass_qsize--;
948 				}
949 				tcp_clean_sackreport(tcpb);
950 			}
951 			INP_WUNLOCK(inpb);
952 		}
953 		INP_INFO_RUNLOCK(&V_tcbinfo);
954 		CURVNET_RESTORE();
955 	}
956 	VNET_LIST_RUNLOCK();
957 }
958 
959 /*
960  * Notify a tcp user of an asynchronous error;
961  * store error as soft error, but wake up user
962  * (for now, won't do anything until can select for soft error).
963  *
964  * Do not wake up user since there currently is no mechanism for
965  * reporting soft errors (yet - a kqueue filter may be added).
966  */
967 static struct inpcb *
968 tcp_notify(struct inpcb *inp, int error)
969 {
970 	struct tcpcb *tp;
971 #ifdef INVARIANTS
972 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
973 #endif
974 
975 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
976 	INP_WLOCK_ASSERT(inp);
977 
978 	if ((inp->inp_flags & INP_TIMEWAIT) ||
979 	    (inp->inp_flags & INP_DROPPED))
980 		return (inp);
981 
982 	tp = intotcpcb(inp);
983 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
984 
985 	/*
986 	 * Ignore some errors if we are hooked up.
987 	 * If connection hasn't completed, has retransmitted several times,
988 	 * and receives a second error, give up now.  This is better
989 	 * than waiting a long time to establish a connection that
990 	 * can never complete.
991 	 */
992 	if (tp->t_state == TCPS_ESTABLISHED &&
993 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
994 	     error == EHOSTDOWN)) {
995 		return (inp);
996 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
997 	    tp->t_softerror) {
998 		tp = tcp_drop(tp, error);
999 		if (tp != NULL)
1000 			return (inp);
1001 		else
1002 			return (NULL);
1003 	} else {
1004 		tp->t_softerror = error;
1005 		return (inp);
1006 	}
1007 #if 0
1008 	wakeup( &so->so_timeo);
1009 	sorwakeup(so);
1010 	sowwakeup(so);
1011 #endif
1012 }
1013 
1014 static int
1015 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1016 {
1017 	INIT_VNET_INET(curvnet);
1018 	int error, i, m, n, pcb_count;
1019 	struct inpcb *inp, **inp_list;
1020 	inp_gen_t gencnt;
1021 	struct xinpgen xig;
1022 
1023 	/*
1024 	 * The process of preparing the TCB list is too time-consuming and
1025 	 * resource-intensive to repeat twice on every request.
1026 	 */
1027 	if (req->oldptr == NULL) {
1028 		m = syncache_pcbcount();
1029 		n = V_tcbinfo.ipi_count;
1030 		req->oldidx = 2 * (sizeof xig)
1031 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1032 		return (0);
1033 	}
1034 
1035 	if (req->newptr != NULL)
1036 		return (EPERM);
1037 
1038 	/*
1039 	 * OK, now we're committed to doing something.
1040 	 */
1041 	INP_INFO_RLOCK(&V_tcbinfo);
1042 	gencnt = V_tcbinfo.ipi_gencnt;
1043 	n = V_tcbinfo.ipi_count;
1044 	INP_INFO_RUNLOCK(&V_tcbinfo);
1045 
1046 	m = syncache_pcbcount();
1047 
1048 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1049 		+ (n + m) * sizeof(struct xtcpcb));
1050 	if (error != 0)
1051 		return (error);
1052 
1053 	xig.xig_len = sizeof xig;
1054 	xig.xig_count = n + m;
1055 	xig.xig_gen = gencnt;
1056 	xig.xig_sogen = so_gencnt;
1057 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1058 	if (error)
1059 		return (error);
1060 
1061 	error = syncache_pcblist(req, m, &pcb_count);
1062 	if (error)
1063 		return (error);
1064 
1065 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1066 	if (inp_list == NULL)
1067 		return (ENOMEM);
1068 
1069 	INP_INFO_RLOCK(&V_tcbinfo);
1070 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1071 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1072 		INP_RLOCK(inp);
1073 		if (inp->inp_gencnt <= gencnt) {
1074 			/*
1075 			 * XXX: This use of cr_cansee(), introduced with
1076 			 * TCP state changes, is not quite right, but for
1077 			 * now, better than nothing.
1078 			 */
1079 			if (inp->inp_flags & INP_TIMEWAIT) {
1080 				if (intotw(inp) != NULL)
1081 					error = cr_cansee(req->td->td_ucred,
1082 					    intotw(inp)->tw_cred);
1083 				else
1084 					error = EINVAL;	/* Skip this inp. */
1085 			} else
1086 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1087 			if (error == 0)
1088 				inp_list[i++] = inp;
1089 		}
1090 		INP_RUNLOCK(inp);
1091 	}
1092 	INP_INFO_RUNLOCK(&V_tcbinfo);
1093 	n = i;
1094 
1095 	error = 0;
1096 	for (i = 0; i < n; i++) {
1097 		inp = inp_list[i];
1098 		INP_RLOCK(inp);
1099 		if (inp->inp_gencnt <= gencnt) {
1100 			struct xtcpcb xt;
1101 			void *inp_ppcb;
1102 
1103 			bzero(&xt, sizeof(xt));
1104 			xt.xt_len = sizeof xt;
1105 			/* XXX should avoid extra copy */
1106 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1107 			inp_ppcb = inp->inp_ppcb;
1108 			if (inp_ppcb == NULL)
1109 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1110 			else if (inp->inp_flags & INP_TIMEWAIT) {
1111 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1112 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1113 			} else
1114 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1115 			if (inp->inp_socket != NULL)
1116 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1117 			else {
1118 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1119 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1120 			}
1121 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1122 			INP_RUNLOCK(inp);
1123 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1124 		} else
1125 			INP_RUNLOCK(inp);
1126 
1127 	}
1128 	if (!error) {
1129 		/*
1130 		 * Give the user an updated idea of our state.
1131 		 * If the generation differs from what we told
1132 		 * her before, she knows that something happened
1133 		 * while we were processing this request, and it
1134 		 * might be necessary to retry.
1135 		 */
1136 		INP_INFO_RLOCK(&V_tcbinfo);
1137 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1138 		xig.xig_sogen = so_gencnt;
1139 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1140 		INP_INFO_RUNLOCK(&V_tcbinfo);
1141 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1142 	}
1143 	free(inp_list, M_TEMP);
1144 	return (error);
1145 }
1146 
1147 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1148     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1149 
1150 static int
1151 tcp_getcred(SYSCTL_HANDLER_ARGS)
1152 {
1153 	INIT_VNET_INET(curvnet);
1154 	struct xucred xuc;
1155 	struct sockaddr_in addrs[2];
1156 	struct inpcb *inp;
1157 	int error;
1158 
1159 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1160 	if (error)
1161 		return (error);
1162 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1163 	if (error)
1164 		return (error);
1165 	INP_INFO_RLOCK(&V_tcbinfo);
1166 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1167 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1168 	if (inp != NULL) {
1169 		INP_RLOCK(inp);
1170 		INP_INFO_RUNLOCK(&V_tcbinfo);
1171 		if (inp->inp_socket == NULL)
1172 			error = ENOENT;
1173 		if (error == 0)
1174 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1175 		if (error == 0)
1176 			cru2x(inp->inp_cred, &xuc);
1177 		INP_RUNLOCK(inp);
1178 	} else {
1179 		INP_INFO_RUNLOCK(&V_tcbinfo);
1180 		error = ENOENT;
1181 	}
1182 	if (error == 0)
1183 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1184 	return (error);
1185 }
1186 
1187 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1188     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1189     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1190 
1191 #ifdef INET6
1192 static int
1193 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1194 {
1195 	INIT_VNET_INET(curvnet);
1196 	INIT_VNET_INET6(curvnet);
1197 	struct xucred xuc;
1198 	struct sockaddr_in6 addrs[2];
1199 	struct inpcb *inp;
1200 	int error, mapped = 0;
1201 
1202 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1203 	if (error)
1204 		return (error);
1205 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1206 	if (error)
1207 		return (error);
1208 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1209 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1210 		return (error);
1211 	}
1212 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1213 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1214 			mapped = 1;
1215 		else
1216 			return (EINVAL);
1217 	}
1218 
1219 	INP_INFO_RLOCK(&V_tcbinfo);
1220 	if (mapped == 1)
1221 		inp = in_pcblookup_hash(&V_tcbinfo,
1222 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1223 			addrs[1].sin6_port,
1224 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1225 			addrs[0].sin6_port,
1226 			0, NULL);
1227 	else
1228 		inp = in6_pcblookup_hash(&V_tcbinfo,
1229 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1230 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1231 	if (inp != NULL) {
1232 		INP_RLOCK(inp);
1233 		INP_INFO_RUNLOCK(&V_tcbinfo);
1234 		if (inp->inp_socket == NULL)
1235 			error = ENOENT;
1236 		if (error == 0)
1237 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1238 		if (error == 0)
1239 			cru2x(inp->inp_cred, &xuc);
1240 		INP_RUNLOCK(inp);
1241 	} else {
1242 		INP_INFO_RUNLOCK(&V_tcbinfo);
1243 		error = ENOENT;
1244 	}
1245 	if (error == 0)
1246 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1247 	return (error);
1248 }
1249 
1250 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1251     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1252     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1253 #endif
1254 
1255 
1256 void
1257 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1258 {
1259 	INIT_VNET_INET(curvnet);
1260 	struct ip *ip = vip;
1261 	struct tcphdr *th;
1262 	struct in_addr faddr;
1263 	struct inpcb *inp;
1264 	struct tcpcb *tp;
1265 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1266 	struct icmp *icp;
1267 	struct in_conninfo inc;
1268 	tcp_seq icmp_tcp_seq;
1269 	int mtu;
1270 
1271 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1272 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1273 		return;
1274 
1275 	if (cmd == PRC_MSGSIZE)
1276 		notify = tcp_mtudisc;
1277 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1278 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1279 		notify = tcp_drop_syn_sent;
1280 	/*
1281 	 * Redirects don't need to be handled up here.
1282 	 */
1283 	else if (PRC_IS_REDIRECT(cmd))
1284 		return;
1285 	/*
1286 	 * Source quench is depreciated.
1287 	 */
1288 	else if (cmd == PRC_QUENCH)
1289 		return;
1290 	/*
1291 	 * Hostdead is ugly because it goes linearly through all PCBs.
1292 	 * XXX: We never get this from ICMP, otherwise it makes an
1293 	 * excellent DoS attack on machines with many connections.
1294 	 */
1295 	else if (cmd == PRC_HOSTDEAD)
1296 		ip = NULL;
1297 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1298 		return;
1299 	if (ip != NULL) {
1300 		icp = (struct icmp *)((caddr_t)ip
1301 				      - offsetof(struct icmp, icmp_ip));
1302 		th = (struct tcphdr *)((caddr_t)ip
1303 				       + (ip->ip_hl << 2));
1304 		INP_INFO_WLOCK(&V_tcbinfo);
1305 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1306 		    ip->ip_src, th->th_sport, 0, NULL);
1307 		if (inp != NULL)  {
1308 			INP_WLOCK(inp);
1309 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1310 			    !(inp->inp_flags & INP_DROPPED) &&
1311 			    !(inp->inp_socket == NULL)) {
1312 				icmp_tcp_seq = htonl(th->th_seq);
1313 				tp = intotcpcb(inp);
1314 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1315 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1316 					if (cmd == PRC_MSGSIZE) {
1317 					    /*
1318 					     * MTU discovery:
1319 					     * If we got a needfrag set the MTU
1320 					     * in the route to the suggested new
1321 					     * value (if given) and then notify.
1322 					     */
1323 					    bzero(&inc, sizeof(inc));
1324 					    inc.inc_faddr = faddr;
1325 					    inc.inc_fibnum =
1326 						inp->inp_inc.inc_fibnum;
1327 
1328 					    mtu = ntohs(icp->icmp_nextmtu);
1329 					    /*
1330 					     * If no alternative MTU was
1331 					     * proposed, try the next smaller
1332 					     * one.  ip->ip_len has already
1333 					     * been swapped in icmp_input().
1334 					     */
1335 					    if (!mtu)
1336 						mtu = ip_next_mtu(ip->ip_len,
1337 						 1);
1338 					    if (mtu < max(296, V_tcp_minmss
1339 						 + sizeof(struct tcpiphdr)))
1340 						mtu = 0;
1341 					    if (!mtu)
1342 						mtu = V_tcp_mssdflt
1343 						 + sizeof(struct tcpiphdr);
1344 					    /*
1345 					     * Only cache the the MTU if it
1346 					     * is smaller than the interface
1347 					     * or route MTU.  tcp_mtudisc()
1348 					     * will do right thing by itself.
1349 					     */
1350 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1351 						tcp_hc_updatemtu(&inc, mtu);
1352 					}
1353 
1354 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1355 				}
1356 			}
1357 			if (inp != NULL)
1358 				INP_WUNLOCK(inp);
1359 		} else {
1360 			bzero(&inc, sizeof(inc));
1361 			inc.inc_fport = th->th_dport;
1362 			inc.inc_lport = th->th_sport;
1363 			inc.inc_faddr = faddr;
1364 			inc.inc_laddr = ip->ip_src;
1365 			syncache_unreach(&inc, th);
1366 		}
1367 		INP_INFO_WUNLOCK(&V_tcbinfo);
1368 	} else
1369 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1370 }
1371 
1372 #ifdef INET6
1373 void
1374 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1375 {
1376 	INIT_VNET_INET(curvnet);
1377 	struct tcphdr th;
1378 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1379 	struct ip6_hdr *ip6;
1380 	struct mbuf *m;
1381 	struct ip6ctlparam *ip6cp = NULL;
1382 	const struct sockaddr_in6 *sa6_src = NULL;
1383 	int off;
1384 	struct tcp_portonly {
1385 		u_int16_t th_sport;
1386 		u_int16_t th_dport;
1387 	} *thp;
1388 
1389 	if (sa->sa_family != AF_INET6 ||
1390 	    sa->sa_len != sizeof(struct sockaddr_in6))
1391 		return;
1392 
1393 	if (cmd == PRC_MSGSIZE)
1394 		notify = tcp_mtudisc;
1395 	else if (!PRC_IS_REDIRECT(cmd) &&
1396 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1397 		return;
1398 	/* Source quench is depreciated. */
1399 	else if (cmd == PRC_QUENCH)
1400 		return;
1401 
1402 	/* if the parameter is from icmp6, decode it. */
1403 	if (d != NULL) {
1404 		ip6cp = (struct ip6ctlparam *)d;
1405 		m = ip6cp->ip6c_m;
1406 		ip6 = ip6cp->ip6c_ip6;
1407 		off = ip6cp->ip6c_off;
1408 		sa6_src = ip6cp->ip6c_src;
1409 	} else {
1410 		m = NULL;
1411 		ip6 = NULL;
1412 		off = 0;	/* fool gcc */
1413 		sa6_src = &sa6_any;
1414 	}
1415 
1416 	if (ip6 != NULL) {
1417 		struct in_conninfo inc;
1418 		/*
1419 		 * XXX: We assume that when IPV6 is non NULL,
1420 		 * M and OFF are valid.
1421 		 */
1422 
1423 		/* check if we can safely examine src and dst ports */
1424 		if (m->m_pkthdr.len < off + sizeof(*thp))
1425 			return;
1426 
1427 		bzero(&th, sizeof(th));
1428 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1429 
1430 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1431 		    (struct sockaddr *)ip6cp->ip6c_src,
1432 		    th.th_sport, cmd, NULL, notify);
1433 
1434 		bzero(&inc, sizeof(inc));
1435 		inc.inc_fport = th.th_dport;
1436 		inc.inc_lport = th.th_sport;
1437 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1438 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1439 		inc.inc_flags |= INC_ISIPV6;
1440 		INP_INFO_WLOCK(&V_tcbinfo);
1441 		syncache_unreach(&inc, &th);
1442 		INP_INFO_WUNLOCK(&V_tcbinfo);
1443 	} else
1444 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1445 			      0, cmd, NULL, notify);
1446 }
1447 #endif /* INET6 */
1448 
1449 
1450 /*
1451  * Following is where TCP initial sequence number generation occurs.
1452  *
1453  * There are two places where we must use initial sequence numbers:
1454  * 1.  In SYN-ACK packets.
1455  * 2.  In SYN packets.
1456  *
1457  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1458  * tcp_syncache.c for details.
1459  *
1460  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1461  * depends on this property.  In addition, these ISNs should be
1462  * unguessable so as to prevent connection hijacking.  To satisfy
1463  * the requirements of this situation, the algorithm outlined in
1464  * RFC 1948 is used, with only small modifications.
1465  *
1466  * Implementation details:
1467  *
1468  * Time is based off the system timer, and is corrected so that it
1469  * increases by one megabyte per second.  This allows for proper
1470  * recycling on high speed LANs while still leaving over an hour
1471  * before rollover.
1472  *
1473  * As reading the *exact* system time is too expensive to be done
1474  * whenever setting up a TCP connection, we increment the time
1475  * offset in two ways.  First, a small random positive increment
1476  * is added to isn_offset for each connection that is set up.
1477  * Second, the function tcp_isn_tick fires once per clock tick
1478  * and increments isn_offset as necessary so that sequence numbers
1479  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1480  * random positive increments serve only to ensure that the same
1481  * exact sequence number is never sent out twice (as could otherwise
1482  * happen when a port is recycled in less than the system tick
1483  * interval.)
1484  *
1485  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1486  * between seeding of isn_secret.  This is normally set to zero,
1487  * as reseeding should not be necessary.
1488  *
1489  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1490  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1491  * general, this means holding an exclusive (write) lock.
1492  */
1493 
1494 #define ISN_BYTES_PER_SECOND 1048576
1495 #define ISN_STATIC_INCREMENT 4096
1496 #define ISN_RANDOM_INCREMENT (4096 - 1)
1497 
1498 #ifdef VIMAGE_GLOBALS
1499 static u_char isn_secret[32];
1500 static int isn_last_reseed;
1501 static u_int32_t isn_offset, isn_offset_old;
1502 #endif
1503 
1504 tcp_seq
1505 tcp_new_isn(struct tcpcb *tp)
1506 {
1507 	INIT_VNET_INET(tp->t_vnet);
1508 	MD5_CTX isn_ctx;
1509 	u_int32_t md5_buffer[4];
1510 	tcp_seq new_isn;
1511 
1512 	INP_WLOCK_ASSERT(tp->t_inpcb);
1513 
1514 	ISN_LOCK();
1515 	/* Seed if this is the first use, reseed if requested. */
1516 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1517 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1518 		< (u_int)ticks))) {
1519 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1520 		V_isn_last_reseed = ticks;
1521 	}
1522 
1523 	/* Compute the md5 hash and return the ISN. */
1524 	MD5Init(&isn_ctx);
1525 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1526 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1527 #ifdef INET6
1528 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1529 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1530 			  sizeof(struct in6_addr));
1531 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1532 			  sizeof(struct in6_addr));
1533 	} else
1534 #endif
1535 	{
1536 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1537 			  sizeof(struct in_addr));
1538 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1539 			  sizeof(struct in_addr));
1540 	}
1541 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1542 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1543 	new_isn = (tcp_seq) md5_buffer[0];
1544 	V_isn_offset += ISN_STATIC_INCREMENT +
1545 		(arc4random() & ISN_RANDOM_INCREMENT);
1546 	new_isn += V_isn_offset;
1547 	ISN_UNLOCK();
1548 	return (new_isn);
1549 }
1550 
1551 /*
1552  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1553  * to keep time flowing at a relatively constant rate.  If the random
1554  * increments have already pushed us past the projected offset, do nothing.
1555  */
1556 static void
1557 tcp_isn_tick(void *xtp)
1558 {
1559 	VNET_ITERATOR_DECL(vnet_iter);
1560 	u_int32_t projected_offset;
1561 
1562 	VNET_LIST_RLOCK();
1563 	ISN_LOCK();
1564 	VNET_FOREACH(vnet_iter) {
1565 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1566 		INIT_VNET_INET(curvnet);
1567 		projected_offset =
1568 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1569 
1570 		if (SEQ_GT(projected_offset, V_isn_offset))
1571 			V_isn_offset = projected_offset;
1572 
1573 		V_isn_offset_old = V_isn_offset;
1574 		CURVNET_RESTORE();
1575 	}
1576 	ISN_UNLOCK();
1577 	VNET_LIST_RUNLOCK();
1578 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1579 }
1580 
1581 /*
1582  * When a specific ICMP unreachable message is received and the
1583  * connection state is SYN-SENT, drop the connection.  This behavior
1584  * is controlled by the icmp_may_rst sysctl.
1585  */
1586 struct inpcb *
1587 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1588 {
1589 #ifdef INVARIANTS
1590 	INIT_VNET_INET(inp->inp_vnet);
1591 #endif
1592 	struct tcpcb *tp;
1593 
1594 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1595 	INP_WLOCK_ASSERT(inp);
1596 
1597 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1598 	    (inp->inp_flags & INP_DROPPED))
1599 		return (inp);
1600 
1601 	tp = intotcpcb(inp);
1602 	if (tp->t_state != TCPS_SYN_SENT)
1603 		return (inp);
1604 
1605 	tp = tcp_drop(tp, errno);
1606 	if (tp != NULL)
1607 		return (inp);
1608 	else
1609 		return (NULL);
1610 }
1611 
1612 /*
1613  * When `need fragmentation' ICMP is received, update our idea of the MSS
1614  * based on the new value in the route.  Also nudge TCP to send something,
1615  * since we know the packet we just sent was dropped.
1616  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1617  */
1618 struct inpcb *
1619 tcp_mtudisc(struct inpcb *inp, int errno)
1620 {
1621 	INIT_VNET_INET(inp->inp_vnet);
1622 	struct tcpcb *tp;
1623 	struct socket *so;
1624 
1625 	INP_WLOCK_ASSERT(inp);
1626 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1627 	    (inp->inp_flags & INP_DROPPED))
1628 		return (inp);
1629 
1630 	tp = intotcpcb(inp);
1631 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1632 
1633 	tcp_mss_update(tp, -1, NULL, NULL);
1634 
1635 	so = inp->inp_socket;
1636 	SOCKBUF_LOCK(&so->so_snd);
1637 	/* If the mss is larger than the socket buffer, decrease the mss. */
1638 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1639 		tp->t_maxseg = so->so_snd.sb_hiwat;
1640 	SOCKBUF_UNLOCK(&so->so_snd);
1641 
1642 	TCPSTAT_INC(tcps_mturesent);
1643 	tp->t_rtttime = 0;
1644 	tp->snd_nxt = tp->snd_una;
1645 	tcp_free_sackholes(tp);
1646 	tp->snd_recover = tp->snd_max;
1647 	if (tp->t_flags & TF_SACK_PERMIT)
1648 		EXIT_FASTRECOVERY(tp);
1649 	tcp_output_send(tp);
1650 	return (inp);
1651 }
1652 
1653 /*
1654  * Look-up the routing entry to the peer of this inpcb.  If no route
1655  * is found and it cannot be allocated, then return 0.  This routine
1656  * is called by TCP routines that access the rmx structure and by
1657  * tcp_mss_update to get the peer/interface MTU.
1658  */
1659 u_long
1660 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1661 {
1662 	struct route sro;
1663 	struct sockaddr_in *dst;
1664 	struct ifnet *ifp;
1665 	u_long maxmtu = 0;
1666 
1667 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1668 
1669 	bzero(&sro, sizeof(sro));
1670 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1671 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1672 		dst->sin_family = AF_INET;
1673 		dst->sin_len = sizeof(*dst);
1674 		dst->sin_addr = inc->inc_faddr;
1675 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1676 	}
1677 	if (sro.ro_rt != NULL) {
1678 		ifp = sro.ro_rt->rt_ifp;
1679 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1680 			maxmtu = ifp->if_mtu;
1681 		else
1682 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1683 
1684 		/* Report additional interface capabilities. */
1685 		if (flags != NULL) {
1686 			if (ifp->if_capenable & IFCAP_TSO4 &&
1687 			    ifp->if_hwassist & CSUM_TSO)
1688 				*flags |= CSUM_TSO;
1689 		}
1690 		RTFREE(sro.ro_rt);
1691 	}
1692 	return (maxmtu);
1693 }
1694 
1695 #ifdef INET6
1696 u_long
1697 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1698 {
1699 	struct route_in6 sro6;
1700 	struct ifnet *ifp;
1701 	u_long maxmtu = 0;
1702 
1703 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1704 
1705 	bzero(&sro6, sizeof(sro6));
1706 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1707 		sro6.ro_dst.sin6_family = AF_INET6;
1708 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1709 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1710 		rtalloc_ign((struct route *)&sro6, 0);
1711 	}
1712 	if (sro6.ro_rt != NULL) {
1713 		ifp = sro6.ro_rt->rt_ifp;
1714 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1715 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1716 		else
1717 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1718 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1719 
1720 		/* Report additional interface capabilities. */
1721 		if (flags != NULL) {
1722 			if (ifp->if_capenable & IFCAP_TSO6 &&
1723 			    ifp->if_hwassist & CSUM_TSO)
1724 				*flags |= CSUM_TSO;
1725 		}
1726 		RTFREE(sro6.ro_rt);
1727 	}
1728 
1729 	return (maxmtu);
1730 }
1731 #endif /* INET6 */
1732 
1733 #ifdef IPSEC
1734 /* compute ESP/AH header size for TCP, including outer IP header. */
1735 size_t
1736 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1737 {
1738 	struct inpcb *inp;
1739 	struct mbuf *m;
1740 	size_t hdrsiz;
1741 	struct ip *ip;
1742 #ifdef INET6
1743 	struct ip6_hdr *ip6;
1744 #endif
1745 	struct tcphdr *th;
1746 
1747 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1748 		return (0);
1749 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1750 	if (!m)
1751 		return (0);
1752 
1753 #ifdef INET6
1754 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1755 		ip6 = mtod(m, struct ip6_hdr *);
1756 		th = (struct tcphdr *)(ip6 + 1);
1757 		m->m_pkthdr.len = m->m_len =
1758 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1759 		tcpip_fillheaders(inp, ip6, th);
1760 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1761 	} else
1762 #endif /* INET6 */
1763 	{
1764 		ip = mtod(m, struct ip *);
1765 		th = (struct tcphdr *)(ip + 1);
1766 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1767 		tcpip_fillheaders(inp, ip, th);
1768 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1769 	}
1770 
1771 	m_free(m);
1772 	return (hdrsiz);
1773 }
1774 #endif /* IPSEC */
1775 
1776 /*
1777  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1778  *
1779  * This code attempts to calculate the bandwidth-delay product as a
1780  * means of determining the optimal window size to maximize bandwidth,
1781  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1782  * routers.  This code also does a fairly good job keeping RTTs in check
1783  * across slow links like modems.  We implement an algorithm which is very
1784  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1785  * transmitter side of a TCP connection and so only effects the transmit
1786  * side of the connection.
1787  *
1788  * BACKGROUND:  TCP makes no provision for the management of buffer space
1789  * at the end points or at the intermediate routers and switches.  A TCP
1790  * stream, whether using NewReno or not, will eventually buffer as
1791  * many packets as it is able and the only reason this typically works is
1792  * due to the fairly small default buffers made available for a connection
1793  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1794  * scaling it is now fairly easy for even a single TCP connection to blow-out
1795  * all available buffer space not only on the local interface, but on
1796  * intermediate routers and switches as well.  NewReno makes a misguided
1797  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1798  * then backing off, then steadily increasing the window again until another
1799  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1800  * is only made worse as network loads increase and the idea of intentionally
1801  * blowing out network buffers is, frankly, a terrible way to manage network
1802  * resources.
1803  *
1804  * It is far better to limit the transmit window prior to the failure
1805  * condition being achieved.  There are two general ways to do this:  First
1806  * you can 'scan' through different transmit window sizes and locate the
1807  * point where the RTT stops increasing, indicating that you have filled the
1808  * pipe, then scan backwards until you note that RTT stops decreasing, then
1809  * repeat ad-infinitum.  This method works in principle but has severe
1810  * implementation issues due to RTT variances, timer granularity, and
1811  * instability in the algorithm which can lead to many false positives and
1812  * create oscillations as well as interact badly with other TCP streams
1813  * implementing the same algorithm.
1814  *
1815  * The second method is to limit the window to the bandwidth delay product
1816  * of the link.  This is the method we implement.  RTT variances and our
1817  * own manipulation of the congestion window, bwnd, can potentially
1818  * destabilize the algorithm.  For this reason we have to stabilize the
1819  * elements used to calculate the window.  We do this by using the minimum
1820  * observed RTT, the long term average of the observed bandwidth, and
1821  * by adding two segments worth of slop.  It isn't perfect but it is able
1822  * to react to changing conditions and gives us a very stable basis on
1823  * which to extend the algorithm.
1824  */
1825 void
1826 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1827 {
1828 	INIT_VNET_INET(tp->t_vnet);
1829 	u_long bw;
1830 	u_long bwnd;
1831 	int save_ticks;
1832 
1833 	INP_WLOCK_ASSERT(tp->t_inpcb);
1834 
1835 	/*
1836 	 * If inflight_enable is disabled in the middle of a tcp connection,
1837 	 * make sure snd_bwnd is effectively disabled.
1838 	 */
1839 	if (V_tcp_inflight_enable == 0 ||
1840 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1841 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1842 		tp->snd_bandwidth = 0;
1843 		return;
1844 	}
1845 
1846 	/*
1847 	 * Figure out the bandwidth.  Due to the tick granularity this
1848 	 * is a very rough number and it MUST be averaged over a fairly
1849 	 * long period of time.  XXX we need to take into account a link
1850 	 * that is not using all available bandwidth, but for now our
1851 	 * slop will ramp us up if this case occurs and the bandwidth later
1852 	 * increases.
1853 	 *
1854 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1855 	 * effectively reset t_bw_rtttime for this case.
1856 	 */
1857 	save_ticks = ticks;
1858 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1859 		return;
1860 
1861 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1862 	    (save_ticks - tp->t_bw_rtttime);
1863 	tp->t_bw_rtttime = save_ticks;
1864 	tp->t_bw_rtseq = ack_seq;
1865 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1866 		return;
1867 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1868 
1869 	tp->snd_bandwidth = bw;
1870 
1871 	/*
1872 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1873 	 * segments.  The additional slop puts us squarely in the sweet
1874 	 * spot and also handles the bandwidth run-up case and stabilization.
1875 	 * Without the slop we could be locking ourselves into a lower
1876 	 * bandwidth.
1877 	 *
1878 	 * Situations Handled:
1879 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1880 	 *	    high speed LANs, allowing larger TCP buffers to be
1881 	 *	    specified, and also does a good job preventing
1882 	 *	    over-queueing of packets over choke points like modems
1883 	 *	    (at least for the transmit side).
1884 	 *
1885 	 *	(2) Is able to handle changing network loads (bandwidth
1886 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1887 	 *	    increases).
1888 	 *
1889 	 *	(3) Theoretically should stabilize in the face of multiple
1890 	 *	    connections implementing the same algorithm (this may need
1891 	 *	    a little work).
1892 	 *
1893 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1894 	 *	    be adjusted with a sysctl but typically only needs to be
1895 	 *	    on very slow connections.  A value no smaller then 5
1896 	 *	    should be used, but only reduce this default if you have
1897 	 *	    no other choice.
1898 	 */
1899 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1900 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1901 #undef USERTT
1902 
1903 	if (tcp_inflight_debug > 0) {
1904 		static int ltime;
1905 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1906 			ltime = ticks;
1907 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1908 			    tp,
1909 			    bw,
1910 			    tp->t_rttbest,
1911 			    tp->t_srtt,
1912 			    bwnd
1913 			);
1914 		}
1915 	}
1916 	if ((long)bwnd < V_tcp_inflight_min)
1917 		bwnd = V_tcp_inflight_min;
1918 	if (bwnd > V_tcp_inflight_max)
1919 		bwnd = V_tcp_inflight_max;
1920 	if ((long)bwnd < tp->t_maxseg * 2)
1921 		bwnd = tp->t_maxseg * 2;
1922 	tp->snd_bwnd = bwnd;
1923 }
1924 
1925 #ifdef TCP_SIGNATURE
1926 /*
1927  * Callback function invoked by m_apply() to digest TCP segment data
1928  * contained within an mbuf chain.
1929  */
1930 static int
1931 tcp_signature_apply(void *fstate, void *data, u_int len)
1932 {
1933 
1934 	MD5Update(fstate, (u_char *)data, len);
1935 	return (0);
1936 }
1937 
1938 /*
1939  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1940  *
1941  * Parameters:
1942  * m		pointer to head of mbuf chain
1943  * _unused
1944  * len		length of TCP segment data, excluding options
1945  * optlen	length of TCP segment options
1946  * buf		pointer to storage for computed MD5 digest
1947  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1948  *
1949  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1950  * When called from tcp_input(), we can be sure that th_sum has been
1951  * zeroed out and verified already.
1952  *
1953  * Return 0 if successful, otherwise return -1.
1954  *
1955  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1956  * search with the destination IP address, and a 'magic SPI' to be
1957  * determined by the application. This is hardcoded elsewhere to 1179
1958  * right now. Another branch of this code exists which uses the SPD to
1959  * specify per-application flows but it is unstable.
1960  */
1961 int
1962 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1963     u_char *buf, u_int direction)
1964 {
1965 	INIT_VNET_IPSEC(curvnet);
1966 	union sockaddr_union dst;
1967 	struct ippseudo ippseudo;
1968 	MD5_CTX ctx;
1969 	int doff;
1970 	struct ip *ip;
1971 	struct ipovly *ipovly;
1972 	struct secasvar *sav;
1973 	struct tcphdr *th;
1974 #ifdef INET6
1975 	struct ip6_hdr *ip6;
1976 	struct in6_addr in6;
1977 	char ip6buf[INET6_ADDRSTRLEN];
1978 	uint32_t plen;
1979 	uint16_t nhdr;
1980 #endif
1981 	u_short savecsum;
1982 
1983 	KASSERT(m != NULL, ("NULL mbuf chain"));
1984 	KASSERT(buf != NULL, ("NULL signature pointer"));
1985 
1986 	/* Extract the destination from the IP header in the mbuf. */
1987 	bzero(&dst, sizeof(union sockaddr_union));
1988 	ip = mtod(m, struct ip *);
1989 #ifdef INET6
1990 	ip6 = NULL;	/* Make the compiler happy. */
1991 #endif
1992 	switch (ip->ip_v) {
1993 	case IPVERSION:
1994 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1995 		dst.sa.sa_family = AF_INET;
1996 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1997 		    ip->ip_src : ip->ip_dst;
1998 		break;
1999 #ifdef INET6
2000 	case (IPV6_VERSION >> 4):
2001 		ip6 = mtod(m, struct ip6_hdr *);
2002 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2003 		dst.sa.sa_family = AF_INET6;
2004 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2005 		    ip6->ip6_src : ip6->ip6_dst;
2006 		break;
2007 #endif
2008 	default:
2009 		return (EINVAL);
2010 		/* NOTREACHED */
2011 		break;
2012 	}
2013 
2014 	/* Look up an SADB entry which matches the address of the peer. */
2015 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2016 	if (sav == NULL) {
2017 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2018 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2019 #ifdef INET6
2020 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2021 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2022 #endif
2023 			"(unsupported)"));
2024 		return (EINVAL);
2025 	}
2026 
2027 	MD5Init(&ctx);
2028 	/*
2029 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2030 	 *
2031 	 * XXX The ippseudo header MUST be digested in network byte order,
2032 	 * or else we'll fail the regression test. Assume all fields we've
2033 	 * been doing arithmetic on have been in host byte order.
2034 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2035 	 * tcp_output(), the underlying ip_len member has not yet been set.
2036 	 */
2037 	switch (ip->ip_v) {
2038 	case IPVERSION:
2039 		ipovly = (struct ipovly *)ip;
2040 		ippseudo.ippseudo_src = ipovly->ih_src;
2041 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2042 		ippseudo.ippseudo_pad = 0;
2043 		ippseudo.ippseudo_p = IPPROTO_TCP;
2044 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2045 		    optlen);
2046 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2047 
2048 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2049 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2050 		break;
2051 #ifdef INET6
2052 	/*
2053 	 * RFC 2385, 2.0  Proposal
2054 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2055 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2056 	 * extended next header value (to form 32 bits), and 32-bit segment
2057 	 * length.
2058 	 * Note: Upper-Layer Packet Length comes before Next Header.
2059 	 */
2060 	case (IPV6_VERSION >> 4):
2061 		in6 = ip6->ip6_src;
2062 		in6_clearscope(&in6);
2063 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2064 		in6 = ip6->ip6_dst;
2065 		in6_clearscope(&in6);
2066 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2067 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2068 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2069 		nhdr = 0;
2070 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2072 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2073 		nhdr = IPPROTO_TCP;
2074 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2075 
2076 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2077 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2078 		break;
2079 #endif
2080 	default:
2081 		return (EINVAL);
2082 		/* NOTREACHED */
2083 		break;
2084 	}
2085 
2086 
2087 	/*
2088 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2089 	 * The TCP checksum must be set to zero.
2090 	 */
2091 	savecsum = th->th_sum;
2092 	th->th_sum = 0;
2093 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2094 	th->th_sum = savecsum;
2095 
2096 	/*
2097 	 * Step 3: Update MD5 hash with TCP segment data.
2098 	 *         Use m_apply() to avoid an early m_pullup().
2099 	 */
2100 	if (len > 0)
2101 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2102 
2103 	/*
2104 	 * Step 4: Update MD5 hash with shared secret.
2105 	 */
2106 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2107 	MD5Final(buf, &ctx);
2108 
2109 	key_sa_recordxfer(sav, m);
2110 	KEY_FREESAV(&sav);
2111 	return (0);
2112 }
2113 #endif /* TCP_SIGNATURE */
2114 
2115 static int
2116 sysctl_drop(SYSCTL_HANDLER_ARGS)
2117 {
2118 	INIT_VNET_INET(curvnet);
2119 #ifdef INET6
2120 	INIT_VNET_INET6(curvnet);
2121 #endif
2122 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2123 	struct sockaddr_storage addrs[2];
2124 	struct inpcb *inp;
2125 	struct tcpcb *tp;
2126 	struct tcptw *tw;
2127 	struct sockaddr_in *fin, *lin;
2128 #ifdef INET6
2129 	struct sockaddr_in6 *fin6, *lin6;
2130 #endif
2131 	int error;
2132 
2133 	inp = NULL;
2134 	fin = lin = NULL;
2135 #ifdef INET6
2136 	fin6 = lin6 = NULL;
2137 #endif
2138 	error = 0;
2139 
2140 	if (req->oldptr != NULL || req->oldlen != 0)
2141 		return (EINVAL);
2142 	if (req->newptr == NULL)
2143 		return (EPERM);
2144 	if (req->newlen < sizeof(addrs))
2145 		return (ENOMEM);
2146 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2147 	if (error)
2148 		return (error);
2149 
2150 	switch (addrs[0].ss_family) {
2151 #ifdef INET6
2152 	case AF_INET6:
2153 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2154 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2155 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2156 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2157 			return (EINVAL);
2158 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2159 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2160 				return (EINVAL);
2161 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2162 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2163 			fin = (struct sockaddr_in *)&addrs[0];
2164 			lin = (struct sockaddr_in *)&addrs[1];
2165 			break;
2166 		}
2167 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2168 		if (error)
2169 			return (error);
2170 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2171 		if (error)
2172 			return (error);
2173 		break;
2174 #endif
2175 	case AF_INET:
2176 		fin = (struct sockaddr_in *)&addrs[0];
2177 		lin = (struct sockaddr_in *)&addrs[1];
2178 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2179 		    lin->sin_len != sizeof(struct sockaddr_in))
2180 			return (EINVAL);
2181 		break;
2182 	default:
2183 		return (EINVAL);
2184 	}
2185 	INP_INFO_WLOCK(&V_tcbinfo);
2186 	switch (addrs[0].ss_family) {
2187 #ifdef INET6
2188 	case AF_INET6:
2189 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2190 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2191 		    NULL);
2192 		break;
2193 #endif
2194 	case AF_INET:
2195 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2196 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2197 		break;
2198 	}
2199 	if (inp != NULL) {
2200 		INP_WLOCK(inp);
2201 		if (inp->inp_flags & INP_TIMEWAIT) {
2202 			/*
2203 			 * XXXRW: There currently exists a state where an
2204 			 * inpcb is present, but its timewait state has been
2205 			 * discarded.  For now, don't allow dropping of this
2206 			 * type of inpcb.
2207 			 */
2208 			tw = intotw(inp);
2209 			if (tw != NULL)
2210 				tcp_twclose(tw, 0);
2211 			else
2212 				INP_WUNLOCK(inp);
2213 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2214 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2215 			tp = intotcpcb(inp);
2216 			tp = tcp_drop(tp, ECONNABORTED);
2217 			if (tp != NULL)
2218 				INP_WUNLOCK(inp);
2219 		} else
2220 			INP_WUNLOCK(inp);
2221 	} else
2222 		error = ESRCH;
2223 	INP_INFO_WUNLOCK(&V_tcbinfo);
2224 	return (error);
2225 }
2226 
2227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2228     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2229     0, sysctl_drop, "", "Drop TCP connection");
2230 
2231 /*
2232  * Generate a standardized TCP log line for use throughout the
2233  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2234  * allow use in the interrupt context.
2235  *
2236  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2237  * NB: The function may return NULL if memory allocation failed.
2238  *
2239  * Due to header inclusion and ordering limitations the struct ip
2240  * and ip6_hdr pointers have to be passed as void pointers.
2241  */
2242 char *
2243 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2244     const void *ip6hdr)
2245 {
2246 	char *s, *sp;
2247 	size_t size;
2248 	struct ip *ip;
2249 #ifdef INET6
2250 	const struct ip6_hdr *ip6;
2251 
2252 	ip6 = (const struct ip6_hdr *)ip6hdr;
2253 #endif /* INET6 */
2254 	ip = (struct ip *)ip4hdr;
2255 
2256 	/*
2257 	 * The log line looks like this:
2258 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2259 	 */
2260 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2261 	    sizeof(PRINT_TH_FLAGS) + 1 +
2262 #ifdef INET6
2263 	    2 * INET6_ADDRSTRLEN;
2264 #else
2265 	    2 * INET_ADDRSTRLEN;
2266 #endif /* INET6 */
2267 
2268 	/* Is logging enabled? */
2269 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2270 		return (NULL);
2271 
2272 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2273 	if (s == NULL)
2274 		return (NULL);
2275 
2276 	strcat(s, "TCP: [");
2277 	sp = s + strlen(s);
2278 
2279 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2280 		inet_ntoa_r(inc->inc_faddr, sp);
2281 		sp = s + strlen(s);
2282 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2283 		sp = s + strlen(s);
2284 		inet_ntoa_r(inc->inc_laddr, sp);
2285 		sp = s + strlen(s);
2286 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2287 #ifdef INET6
2288 	} else if (inc) {
2289 		ip6_sprintf(sp, &inc->inc6_faddr);
2290 		sp = s + strlen(s);
2291 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2292 		sp = s + strlen(s);
2293 		ip6_sprintf(sp, &inc->inc6_laddr);
2294 		sp = s + strlen(s);
2295 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2296 	} else if (ip6 && th) {
2297 		ip6_sprintf(sp, &ip6->ip6_src);
2298 		sp = s + strlen(s);
2299 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2300 		sp = s + strlen(s);
2301 		ip6_sprintf(sp, &ip6->ip6_dst);
2302 		sp = s + strlen(s);
2303 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2304 #endif /* INET6 */
2305 	} else if (ip && th) {
2306 		inet_ntoa_r(ip->ip_src, sp);
2307 		sp = s + strlen(s);
2308 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2309 		sp = s + strlen(s);
2310 		inet_ntoa_r(ip->ip_dst, sp);
2311 		sp = s + strlen(s);
2312 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2313 	} else {
2314 		free(s, M_TCPLOG);
2315 		return (NULL);
2316 	}
2317 	sp = s + strlen(s);
2318 	if (th)
2319 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2320 	if (*(s + size - 1) != '\0')
2321 		panic("%s: string too long", __func__);
2322 	return (s);
2323 }
2324