xref: /freebsd/sys/netinet/tcp_subr.c (revision c0020399a650364d0134f79f3fa319f84064372d)
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 
292 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
293 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
294 	tcp_tw_zone_change();
295 }
296 
297 static int
298 tcp_inpcb_init(void *mem, int size, int flags)
299 {
300 	struct inpcb *inp = mem;
301 
302 	INP_LOCK_INIT(inp, "inp", "tcpinp");
303 	return (0);
304 }
305 
306 void
307 tcp_init(void)
308 {
309 	INIT_VNET_INET(curvnet);
310 	int hashsize;
311 
312 	V_blackhole = 0;
313 	V_tcp_delack_enabled = 1;
314 	V_drop_synfin = 0;
315 	V_tcp_do_rfc3042 = 1;
316 	V_tcp_do_rfc3390 = 1;
317 	V_tcp_do_ecn = 0;
318 	V_tcp_ecn_maxretries = 1;
319 	V_tcp_insecure_rst = 0;
320 	V_tcp_do_autorcvbuf = 1;
321 	V_tcp_autorcvbuf_inc = 16*1024;
322 	V_tcp_autorcvbuf_max = 256*1024;
323 	V_tcp_do_rfc3465 = 1;
324 	V_tcp_abc_l_var = 2;
325 
326 	V_tcp_mssdflt = TCP_MSS;
327 #ifdef INET6
328 	V_tcp_v6mssdflt = TCP6_MSS;
329 #endif
330 	V_tcp_minmss = TCP_MINMSS;
331 	V_tcp_do_rfc1323 = 1;
332 	V_icmp_may_rst = 1;
333 	V_tcp_isn_reseed_interval = 0;
334 	V_tcp_inflight_enable = 1;
335 	V_tcp_inflight_min = 6144;
336 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
337 	V_tcp_inflight_stab = 20;
338 
339 	V_path_mtu_discovery = 1;
340 	V_ss_fltsz = 1;
341 	V_ss_fltsz_local = 4;
342 	V_tcp_do_newreno = 1;
343 	V_tcp_do_tso = 1;
344 	V_tcp_do_autosndbuf = 1;
345 	V_tcp_autosndbuf_inc = 8*1024;
346 	V_tcp_autosndbuf_max = 256*1024;
347 
348 	V_nolocaltimewait = 0;
349 
350 	V_tcp_do_sack = 1;
351 	V_tcp_sack_maxholes = 128;
352 	V_tcp_sack_globalmaxholes = 65536;
353 	V_tcp_sack_globalholes = 0;
354 
355 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
356 
357 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
358 
359 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
360 	LIST_INIT(&V_tcb);
361 	V_tcbinfo.ipi_listhead = &V_tcb;
362 	hashsize = TCBHASHSIZE;
363 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
364 	if (!powerof2(hashsize)) {
365 		printf("WARNING: TCB hash size not a power of 2\n");
366 		hashsize = 512; /* safe default */
367 	}
368 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
369 	    &V_tcbinfo.ipi_hashmask);
370 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
371 	    &V_tcbinfo.ipi_porthashmask);
372 	V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
373 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
374 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
375 	/*
376 	 * These have to be type stable for the benefit of the timers.
377 	 */
378 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
379 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
380 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
381 	tcp_tw_init();
382 	syncache_init();
383 	tcp_hc_init();
384 	tcp_reass_init();
385 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
386 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
387 
388 	/* Skip initialization of globals for non-default instances. */
389 	if (!IS_DEFAULT_VNET(curvnet))
390 		return;
391 
392 	/* XXX virtualize those bellow? */
393 	tcp_delacktime = TCPTV_DELACK;
394 	tcp_keepinit = TCPTV_KEEP_INIT;
395 	tcp_keepidle = TCPTV_KEEP_IDLE;
396 	tcp_keepintvl = TCPTV_KEEPINTVL;
397 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
398 	tcp_msl = TCPTV_MSL;
399 	tcp_rexmit_min = TCPTV_MIN;
400 	if (tcp_rexmit_min < 1)
401 		tcp_rexmit_min = 1;
402 	tcp_rexmit_slop = TCPTV_CPU_VAR;
403 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
404 	tcp_tcbhashsize = hashsize;
405 
406 #ifdef INET6
407 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
408 #else /* INET6 */
409 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
410 #endif /* INET6 */
411 	if (max_protohdr < TCP_MINPROTOHDR)
412 		max_protohdr = TCP_MINPROTOHDR;
413 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
414 		panic("tcp_init");
415 #undef TCP_MINPROTOHDR
416 
417 	ISN_LOCK_INIT();
418 	callout_init(&isn_callout, CALLOUT_MPSAFE);
419 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
420 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
421 		SHUTDOWN_PRI_DEFAULT);
422 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
423 		EVENTHANDLER_PRI_ANY);
424 }
425 
426 void
427 tcp_fini(void *xtp)
428 {
429 
430 	callout_stop(&isn_callout);
431 }
432 
433 /*
434  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
435  * tcp_template used to store this data in mbufs, but we now recopy it out
436  * of the tcpcb each time to conserve mbufs.
437  */
438 void
439 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
440 {
441 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
442 
443 	INP_WLOCK_ASSERT(inp);
444 
445 #ifdef INET6
446 	if ((inp->inp_vflag & INP_IPV6) != 0) {
447 		struct ip6_hdr *ip6;
448 
449 		ip6 = (struct ip6_hdr *)ip_ptr;
450 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
451 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
452 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
453 			(IPV6_VERSION & IPV6_VERSION_MASK);
454 		ip6->ip6_nxt = IPPROTO_TCP;
455 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
456 		ip6->ip6_src = inp->in6p_laddr;
457 		ip6->ip6_dst = inp->in6p_faddr;
458 	} else
459 #endif
460 	{
461 		struct ip *ip;
462 
463 		ip = (struct ip *)ip_ptr;
464 		ip->ip_v = IPVERSION;
465 		ip->ip_hl = 5;
466 		ip->ip_tos = inp->inp_ip_tos;
467 		ip->ip_len = 0;
468 		ip->ip_id = 0;
469 		ip->ip_off = 0;
470 		ip->ip_ttl = inp->inp_ip_ttl;
471 		ip->ip_sum = 0;
472 		ip->ip_p = IPPROTO_TCP;
473 		ip->ip_src = inp->inp_laddr;
474 		ip->ip_dst = inp->inp_faddr;
475 	}
476 	th->th_sport = inp->inp_lport;
477 	th->th_dport = inp->inp_fport;
478 	th->th_seq = 0;
479 	th->th_ack = 0;
480 	th->th_x2 = 0;
481 	th->th_off = 5;
482 	th->th_flags = 0;
483 	th->th_win = 0;
484 	th->th_urp = 0;
485 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
486 }
487 
488 /*
489  * Create template to be used to send tcp packets on a connection.
490  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
491  * use for this function is in keepalives, which use tcp_respond.
492  */
493 struct tcptemp *
494 tcpip_maketemplate(struct inpcb *inp)
495 {
496 	struct tcptemp *t;
497 
498 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
499 	if (t == NULL)
500 		return (NULL);
501 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
502 	return (t);
503 }
504 
505 /*
506  * Send a single message to the TCP at address specified by
507  * the given TCP/IP header.  If m == NULL, then we make a copy
508  * of the tcpiphdr at ti and send directly to the addressed host.
509  * This is used to force keep alive messages out using the TCP
510  * template for a connection.  If flags are given then we send
511  * a message back to the TCP which originated the * segment ti,
512  * and discard the mbuf containing it and any other attached mbufs.
513  *
514  * In any case the ack and sequence number of the transmitted
515  * segment are as specified by the parameters.
516  *
517  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
518  */
519 void
520 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
521     tcp_seq ack, tcp_seq seq, int flags)
522 {
523 	INIT_VNET_INET(curvnet);
524 	int tlen;
525 	int win = 0;
526 	struct ip *ip;
527 	struct tcphdr *nth;
528 #ifdef INET6
529 	struct ip6_hdr *ip6;
530 	int isipv6;
531 #endif /* INET6 */
532 	int ipflags = 0;
533 	struct inpcb *inp;
534 
535 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
536 
537 #ifdef INET6
538 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
539 	ip6 = ipgen;
540 #endif /* INET6 */
541 	ip = ipgen;
542 
543 	if (tp != NULL) {
544 		inp = tp->t_inpcb;
545 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
546 		INP_WLOCK_ASSERT(inp);
547 	} else
548 		inp = NULL;
549 
550 	if (tp != NULL) {
551 		if (!(flags & TH_RST)) {
552 			win = sbspace(&inp->inp_socket->so_rcv);
553 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
554 				win = (long)TCP_MAXWIN << tp->rcv_scale;
555 		}
556 	}
557 	if (m == NULL) {
558 		m = m_gethdr(M_DONTWAIT, MT_DATA);
559 		if (m == NULL)
560 			return;
561 		tlen = 0;
562 		m->m_data += max_linkhdr;
563 #ifdef INET6
564 		if (isipv6) {
565 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
566 			      sizeof(struct ip6_hdr));
567 			ip6 = mtod(m, struct ip6_hdr *);
568 			nth = (struct tcphdr *)(ip6 + 1);
569 		} else
570 #endif /* INET6 */
571 	      {
572 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
573 		ip = mtod(m, struct ip *);
574 		nth = (struct tcphdr *)(ip + 1);
575 	      }
576 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
577 		flags = TH_ACK;
578 	} else {
579 		/*
580 		 *  reuse the mbuf.
581 		 * XXX MRT We inherrit the FIB, which is lucky.
582 		 */
583 		m_freem(m->m_next);
584 		m->m_next = NULL;
585 		m->m_data = (caddr_t)ipgen;
586 		/* m_len is set later */
587 		tlen = 0;
588 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
589 #ifdef INET6
590 		if (isipv6) {
591 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
592 			nth = (struct tcphdr *)(ip6 + 1);
593 		} else
594 #endif /* INET6 */
595 	      {
596 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
597 		nth = (struct tcphdr *)(ip + 1);
598 	      }
599 		if (th != nth) {
600 			/*
601 			 * this is usually a case when an extension header
602 			 * exists between the IPv6 header and the
603 			 * TCP header.
604 			 */
605 			nth->th_sport = th->th_sport;
606 			nth->th_dport = th->th_dport;
607 		}
608 		xchg(nth->th_dport, nth->th_sport, uint16_t);
609 #undef xchg
610 	}
611 #ifdef INET6
612 	if (isipv6) {
613 		ip6->ip6_flow = 0;
614 		ip6->ip6_vfc = IPV6_VERSION;
615 		ip6->ip6_nxt = IPPROTO_TCP;
616 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
617 						tlen));
618 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
619 	} else
620 #endif
621 	{
622 		tlen += sizeof (struct tcpiphdr);
623 		ip->ip_len = tlen;
624 		ip->ip_ttl = V_ip_defttl;
625 		if (V_path_mtu_discovery)
626 			ip->ip_off |= IP_DF;
627 	}
628 	m->m_len = tlen;
629 	m->m_pkthdr.len = tlen;
630 	m->m_pkthdr.rcvif = NULL;
631 #ifdef MAC
632 	if (inp != NULL) {
633 		/*
634 		 * Packet is associated with a socket, so allow the
635 		 * label of the response to reflect the socket label.
636 		 */
637 		INP_WLOCK_ASSERT(inp);
638 		mac_inpcb_create_mbuf(inp, m);
639 	} else {
640 		/*
641 		 * Packet is not associated with a socket, so possibly
642 		 * update the label in place.
643 		 */
644 		mac_netinet_tcp_reply(m);
645 	}
646 #endif
647 	nth->th_seq = htonl(seq);
648 	nth->th_ack = htonl(ack);
649 	nth->th_x2 = 0;
650 	nth->th_off = sizeof (struct tcphdr) >> 2;
651 	nth->th_flags = flags;
652 	if (tp != NULL)
653 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
654 	else
655 		nth->th_win = htons((u_short)win);
656 	nth->th_urp = 0;
657 #ifdef INET6
658 	if (isipv6) {
659 		nth->th_sum = 0;
660 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
661 					sizeof(struct ip6_hdr),
662 					tlen - sizeof(struct ip6_hdr));
663 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
664 		    NULL, NULL);
665 	} else
666 #endif /* INET6 */
667 	{
668 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
669 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
670 		m->m_pkthdr.csum_flags = CSUM_TCP;
671 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
672 	}
673 #ifdef TCPDEBUG
674 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
675 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
676 #endif
677 #ifdef INET6
678 	if (isipv6)
679 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
680 	else
681 #endif /* INET6 */
682 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
683 }
684 
685 /*
686  * Create a new TCP control block, making an
687  * empty reassembly queue and hooking it to the argument
688  * protocol control block.  The `inp' parameter must have
689  * come from the zone allocator set up in tcp_init().
690  */
691 struct tcpcb *
692 tcp_newtcpcb(struct inpcb *inp)
693 {
694 	INIT_VNET_INET(inp->inp_vnet);
695 	struct tcpcb_mem *tm;
696 	struct tcpcb *tp;
697 #ifdef INET6
698 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
699 #endif /* INET6 */
700 
701 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
702 	if (tm == NULL)
703 		return (NULL);
704 	tp = &tm->tcb;
705 	tp->t_timers = &tm->tt;
706 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
707 	tp->t_maxseg = tp->t_maxopd =
708 #ifdef INET6
709 		isipv6 ? V_tcp_v6mssdflt :
710 #endif /* INET6 */
711 		V_tcp_mssdflt;
712 
713 	/* Set up our timeouts. */
714 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
715 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
716 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
717 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
718 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
719 
720 	if (V_tcp_do_rfc1323)
721 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
722 	if (V_tcp_do_sack)
723 		tp->t_flags |= TF_SACK_PERMIT;
724 	TAILQ_INIT(&tp->snd_holes);
725 	tp->t_inpcb = inp;	/* XXX */
726 	/*
727 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
728 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
729 	 * reasonable initial retransmit time.
730 	 */
731 	tp->t_srtt = TCPTV_SRTTBASE;
732 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
733 	tp->t_rttmin = tcp_rexmit_min;
734 	tp->t_rxtcur = TCPTV_RTOBASE;
735 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
736 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
737 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
738 	tp->t_rcvtime = ticks;
739 	tp->t_bw_rtttime = ticks;
740 	/*
741 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
742 	 * because the socket may be bound to an IPv6 wildcard address,
743 	 * which may match an IPv4-mapped IPv6 address.
744 	 */
745 	inp->inp_ip_ttl = V_ip_defttl;
746 	inp->inp_ppcb = tp;
747 	return (tp);		/* XXX */
748 }
749 
750 /*
751  * Drop a TCP connection, reporting
752  * the specified error.  If connection is synchronized,
753  * then send a RST to peer.
754  */
755 struct tcpcb *
756 tcp_drop(struct tcpcb *tp, int errno)
757 {
758 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
759 	struct socket *so = tp->t_inpcb->inp_socket;
760 
761 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
762 	INP_WLOCK_ASSERT(tp->t_inpcb);
763 
764 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
765 		tp->t_state = TCPS_CLOSED;
766 		(void) tcp_output_reset(tp);
767 		TCPSTAT_INC(tcps_drops);
768 	} else
769 		TCPSTAT_INC(tcps_conndrops);
770 	if (errno == ETIMEDOUT && tp->t_softerror)
771 		errno = tp->t_softerror;
772 	so->so_error = errno;
773 	return (tcp_close(tp));
774 }
775 
776 void
777 tcp_discardcb(struct tcpcb *tp)
778 {
779 	INIT_VNET_INET(tp->t_vnet);
780 	struct tseg_qent *q;
781 	struct inpcb *inp = tp->t_inpcb;
782 	struct socket *so = inp->inp_socket;
783 #ifdef INET6
784 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
785 #endif /* INET6 */
786 
787 	INP_WLOCK_ASSERT(inp);
788 
789 	/*
790 	 * Make sure that all of our timers are stopped before we
791 	 * delete the PCB.
792 	 */
793 	callout_stop(&tp->t_timers->tt_rexmt);
794 	callout_stop(&tp->t_timers->tt_persist);
795 	callout_stop(&tp->t_timers->tt_keep);
796 	callout_stop(&tp->t_timers->tt_2msl);
797 	callout_stop(&tp->t_timers->tt_delack);
798 
799 	/*
800 	 * If we got enough samples through the srtt filter,
801 	 * save the rtt and rttvar in the routing entry.
802 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
803 	 * 4 samples is enough for the srtt filter to converge
804 	 * to within enough % of the correct value; fewer samples
805 	 * and we could save a bogus rtt. The danger is not high
806 	 * as tcp quickly recovers from everything.
807 	 * XXX: Works very well but needs some more statistics!
808 	 */
809 	if (tp->t_rttupdated >= 4) {
810 		struct hc_metrics_lite metrics;
811 		u_long ssthresh;
812 
813 		bzero(&metrics, sizeof(metrics));
814 		/*
815 		 * Update the ssthresh always when the conditions below
816 		 * are satisfied. This gives us better new start value
817 		 * for the congestion avoidance for new connections.
818 		 * ssthresh is only set if packet loss occured on a session.
819 		 *
820 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
821 		 * being torn down.  Ideally this code would not use 'so'.
822 		 */
823 		ssthresh = tp->snd_ssthresh;
824 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
825 			/*
826 			 * convert the limit from user data bytes to
827 			 * packets then to packet data bytes.
828 			 */
829 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
830 			if (ssthresh < 2)
831 				ssthresh = 2;
832 			ssthresh *= (u_long)(tp->t_maxseg +
833 #ifdef INET6
834 				      (isipv6 ? sizeof (struct ip6_hdr) +
835 					       sizeof (struct tcphdr) :
836 #endif
837 				       sizeof (struct tcpiphdr)
838 #ifdef INET6
839 				       )
840 #endif
841 				      );
842 		} else
843 			ssthresh = 0;
844 		metrics.rmx_ssthresh = ssthresh;
845 
846 		metrics.rmx_rtt = tp->t_srtt;
847 		metrics.rmx_rttvar = tp->t_rttvar;
848 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
849 		metrics.rmx_bandwidth = tp->snd_bandwidth;
850 		metrics.rmx_cwnd = tp->snd_cwnd;
851 		metrics.rmx_sendpipe = 0;
852 		metrics.rmx_recvpipe = 0;
853 
854 		tcp_hc_update(&inp->inp_inc, &metrics);
855 	}
856 
857 	/* free the reassembly queue, if any */
858 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
859 		LIST_REMOVE(q, tqe_q);
860 		m_freem(q->tqe_m);
861 		uma_zfree(V_tcp_reass_zone, q);
862 		tp->t_segqlen--;
863 		V_tcp_reass_qsize--;
864 	}
865 	/* Disconnect offload device, if any. */
866 	tcp_offload_detach(tp);
867 
868 	tcp_free_sackholes(tp);
869 	inp->inp_ppcb = NULL;
870 	tp->t_inpcb = NULL;
871 	uma_zfree(V_tcpcb_zone, tp);
872 }
873 
874 /*
875  * Attempt to close a TCP control block, marking it as dropped, and freeing
876  * the socket if we hold the only reference.
877  */
878 struct tcpcb *
879 tcp_close(struct tcpcb *tp)
880 {
881 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
882 	struct inpcb *inp = tp->t_inpcb;
883 	struct socket *so;
884 
885 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
886 	INP_WLOCK_ASSERT(inp);
887 
888 	/* Notify any offload devices of listener close */
889 	if (tp->t_state == TCPS_LISTEN)
890 		tcp_offload_listen_close(tp);
891 	in_pcbdrop(inp);
892 	TCPSTAT_INC(tcps_closed);
893 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
894 	so = inp->inp_socket;
895 	soisdisconnected(so);
896 	if (inp->inp_flags & INP_SOCKREF) {
897 		KASSERT(so->so_state & SS_PROTOREF,
898 		    ("tcp_close: !SS_PROTOREF"));
899 		inp->inp_flags &= ~INP_SOCKREF;
900 		INP_WUNLOCK(inp);
901 		ACCEPT_LOCK();
902 		SOCK_LOCK(so);
903 		so->so_state &= ~SS_PROTOREF;
904 		sofree(so);
905 		return (NULL);
906 	}
907 	return (tp);
908 }
909 
910 void
911 tcp_drain(void)
912 {
913 	VNET_ITERATOR_DECL(vnet_iter);
914 
915 	if (!do_tcpdrain)
916 		return;
917 
918 	VNET_LIST_RLOCK();
919 	VNET_FOREACH(vnet_iter) {
920 		CURVNET_SET(vnet_iter);
921 		INIT_VNET_INET(vnet_iter);
922 		struct inpcb *inpb;
923 		struct tcpcb *tcpb;
924 		struct tseg_qent *te;
925 
926 	/*
927 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
928 	 * if there is one...
929 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
930 	 *      reassembly queue should be flushed, but in a situation
931 	 *	where we're really low on mbufs, this is potentially
932 	 *	usefull.
933 	 */
934 		INP_INFO_RLOCK(&V_tcbinfo);
935 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
936 			if (inpb->inp_flags & INP_TIMEWAIT)
937 				continue;
938 			INP_WLOCK(inpb);
939 			if ((tcpb = intotcpcb(inpb)) != NULL) {
940 				while ((te = LIST_FIRST(&tcpb->t_segq))
941 			            != NULL) {
942 					LIST_REMOVE(te, tqe_q);
943 					m_freem(te->tqe_m);
944 					uma_zfree(V_tcp_reass_zone, te);
945 					tcpb->t_segqlen--;
946 					V_tcp_reass_qsize--;
947 				}
948 				tcp_clean_sackreport(tcpb);
949 			}
950 			INP_WUNLOCK(inpb);
951 		}
952 		INP_INFO_RUNLOCK(&V_tcbinfo);
953 		CURVNET_RESTORE();
954 	}
955 	VNET_LIST_RUNLOCK();
956 }
957 
958 /*
959  * Notify a tcp user of an asynchronous error;
960  * store error as soft error, but wake up user
961  * (for now, won't do anything until can select for soft error).
962  *
963  * Do not wake up user since there currently is no mechanism for
964  * reporting soft errors (yet - a kqueue filter may be added).
965  */
966 static struct inpcb *
967 tcp_notify(struct inpcb *inp, int error)
968 {
969 	struct tcpcb *tp;
970 #ifdef INVARIANTS
971 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
972 #endif
973 
974 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
975 	INP_WLOCK_ASSERT(inp);
976 
977 	if ((inp->inp_flags & INP_TIMEWAIT) ||
978 	    (inp->inp_flags & INP_DROPPED))
979 		return (inp);
980 
981 	tp = intotcpcb(inp);
982 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
983 
984 	/*
985 	 * Ignore some errors if we are hooked up.
986 	 * If connection hasn't completed, has retransmitted several times,
987 	 * and receives a second error, give up now.  This is better
988 	 * than waiting a long time to establish a connection that
989 	 * can never complete.
990 	 */
991 	if (tp->t_state == TCPS_ESTABLISHED &&
992 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
993 	     error == EHOSTDOWN)) {
994 		return (inp);
995 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
996 	    tp->t_softerror) {
997 		tp = tcp_drop(tp, error);
998 		if (tp != NULL)
999 			return (inp);
1000 		else
1001 			return (NULL);
1002 	} else {
1003 		tp->t_softerror = error;
1004 		return (inp);
1005 	}
1006 #if 0
1007 	wakeup( &so->so_timeo);
1008 	sorwakeup(so);
1009 	sowwakeup(so);
1010 #endif
1011 }
1012 
1013 static int
1014 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1015 {
1016 	INIT_VNET_INET(curvnet);
1017 	int error, i, m, n, pcb_count;
1018 	struct inpcb *inp, **inp_list;
1019 	inp_gen_t gencnt;
1020 	struct xinpgen xig;
1021 
1022 	/*
1023 	 * The process of preparing the TCB list is too time-consuming and
1024 	 * resource-intensive to repeat twice on every request.
1025 	 */
1026 	if (req->oldptr == NULL) {
1027 		m = syncache_pcbcount();
1028 		n = V_tcbinfo.ipi_count;
1029 		req->oldidx = 2 * (sizeof xig)
1030 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1031 		return (0);
1032 	}
1033 
1034 	if (req->newptr != NULL)
1035 		return (EPERM);
1036 
1037 	/*
1038 	 * OK, now we're committed to doing something.
1039 	 */
1040 	INP_INFO_RLOCK(&V_tcbinfo);
1041 	gencnt = V_tcbinfo.ipi_gencnt;
1042 	n = V_tcbinfo.ipi_count;
1043 	INP_INFO_RUNLOCK(&V_tcbinfo);
1044 
1045 	m = syncache_pcbcount();
1046 
1047 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1048 		+ (n + m) * sizeof(struct xtcpcb));
1049 	if (error != 0)
1050 		return (error);
1051 
1052 	xig.xig_len = sizeof xig;
1053 	xig.xig_count = n + m;
1054 	xig.xig_gen = gencnt;
1055 	xig.xig_sogen = so_gencnt;
1056 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1057 	if (error)
1058 		return (error);
1059 
1060 	error = syncache_pcblist(req, m, &pcb_count);
1061 	if (error)
1062 		return (error);
1063 
1064 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1065 	if (inp_list == NULL)
1066 		return (ENOMEM);
1067 
1068 	INP_INFO_RLOCK(&V_tcbinfo);
1069 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1070 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1071 		INP_RLOCK(inp);
1072 		if (inp->inp_gencnt <= gencnt) {
1073 			/*
1074 			 * XXX: This use of cr_cansee(), introduced with
1075 			 * TCP state changes, is not quite right, but for
1076 			 * now, better than nothing.
1077 			 */
1078 			if (inp->inp_flags & INP_TIMEWAIT) {
1079 				if (intotw(inp) != NULL)
1080 					error = cr_cansee(req->td->td_ucred,
1081 					    intotw(inp)->tw_cred);
1082 				else
1083 					error = EINVAL;	/* Skip this inp. */
1084 			} else
1085 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1086 			if (error == 0)
1087 				inp_list[i++] = inp;
1088 		}
1089 		INP_RUNLOCK(inp);
1090 	}
1091 	INP_INFO_RUNLOCK(&V_tcbinfo);
1092 	n = i;
1093 
1094 	error = 0;
1095 	for (i = 0; i < n; i++) {
1096 		inp = inp_list[i];
1097 		INP_RLOCK(inp);
1098 		if (inp->inp_gencnt <= gencnt) {
1099 			struct xtcpcb xt;
1100 			void *inp_ppcb;
1101 
1102 			bzero(&xt, sizeof(xt));
1103 			xt.xt_len = sizeof xt;
1104 			/* XXX should avoid extra copy */
1105 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1106 			inp_ppcb = inp->inp_ppcb;
1107 			if (inp_ppcb == NULL)
1108 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1109 			else if (inp->inp_flags & INP_TIMEWAIT) {
1110 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1111 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1112 			} else
1113 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1114 			if (inp->inp_socket != NULL)
1115 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1116 			else {
1117 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1118 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1119 			}
1120 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1121 			INP_RUNLOCK(inp);
1122 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1123 		} else
1124 			INP_RUNLOCK(inp);
1125 
1126 	}
1127 	if (!error) {
1128 		/*
1129 		 * Give the user an updated idea of our state.
1130 		 * If the generation differs from what we told
1131 		 * her before, she knows that something happened
1132 		 * while we were processing this request, and it
1133 		 * might be necessary to retry.
1134 		 */
1135 		INP_INFO_RLOCK(&V_tcbinfo);
1136 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1137 		xig.xig_sogen = so_gencnt;
1138 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1139 		INP_INFO_RUNLOCK(&V_tcbinfo);
1140 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1141 	}
1142 	free(inp_list, M_TEMP);
1143 	return (error);
1144 }
1145 
1146 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1147     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1148 
1149 static int
1150 tcp_getcred(SYSCTL_HANDLER_ARGS)
1151 {
1152 	INIT_VNET_INET(curvnet);
1153 	struct xucred xuc;
1154 	struct sockaddr_in addrs[2];
1155 	struct inpcb *inp;
1156 	int error;
1157 
1158 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1159 	if (error)
1160 		return (error);
1161 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1162 	if (error)
1163 		return (error);
1164 	INP_INFO_RLOCK(&V_tcbinfo);
1165 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1166 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1167 	if (inp != NULL) {
1168 		INP_RLOCK(inp);
1169 		INP_INFO_RUNLOCK(&V_tcbinfo);
1170 		if (inp->inp_socket == NULL)
1171 			error = ENOENT;
1172 		if (error == 0)
1173 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1174 		if (error == 0)
1175 			cru2x(inp->inp_cred, &xuc);
1176 		INP_RUNLOCK(inp);
1177 	} else {
1178 		INP_INFO_RUNLOCK(&V_tcbinfo);
1179 		error = ENOENT;
1180 	}
1181 	if (error == 0)
1182 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1183 	return (error);
1184 }
1185 
1186 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1187     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1188     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1189 
1190 #ifdef INET6
1191 static int
1192 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1193 {
1194 	INIT_VNET_INET(curvnet);
1195 	INIT_VNET_INET6(curvnet);
1196 	struct xucred xuc;
1197 	struct sockaddr_in6 addrs[2];
1198 	struct inpcb *inp;
1199 	int error, mapped = 0;
1200 
1201 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1202 	if (error)
1203 		return (error);
1204 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1205 	if (error)
1206 		return (error);
1207 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1208 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1209 		return (error);
1210 	}
1211 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1212 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1213 			mapped = 1;
1214 		else
1215 			return (EINVAL);
1216 	}
1217 
1218 	INP_INFO_RLOCK(&V_tcbinfo);
1219 	if (mapped == 1)
1220 		inp = in_pcblookup_hash(&V_tcbinfo,
1221 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1222 			addrs[1].sin6_port,
1223 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1224 			addrs[0].sin6_port,
1225 			0, NULL);
1226 	else
1227 		inp = in6_pcblookup_hash(&V_tcbinfo,
1228 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1229 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1230 	if (inp != NULL) {
1231 		INP_RLOCK(inp);
1232 		INP_INFO_RUNLOCK(&V_tcbinfo);
1233 		if (inp->inp_socket == NULL)
1234 			error = ENOENT;
1235 		if (error == 0)
1236 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1237 		if (error == 0)
1238 			cru2x(inp->inp_cred, &xuc);
1239 		INP_RUNLOCK(inp);
1240 	} else {
1241 		INP_INFO_RUNLOCK(&V_tcbinfo);
1242 		error = ENOENT;
1243 	}
1244 	if (error == 0)
1245 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1246 	return (error);
1247 }
1248 
1249 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1250     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1251     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1252 #endif
1253 
1254 
1255 void
1256 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1257 {
1258 	INIT_VNET_INET(curvnet);
1259 	struct ip *ip = vip;
1260 	struct tcphdr *th;
1261 	struct in_addr faddr;
1262 	struct inpcb *inp;
1263 	struct tcpcb *tp;
1264 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1265 	struct icmp *icp;
1266 	struct in_conninfo inc;
1267 	tcp_seq icmp_tcp_seq;
1268 	int mtu;
1269 
1270 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1271 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1272 		return;
1273 
1274 	if (cmd == PRC_MSGSIZE)
1275 		notify = tcp_mtudisc;
1276 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1277 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1278 		notify = tcp_drop_syn_sent;
1279 	/*
1280 	 * Redirects don't need to be handled up here.
1281 	 */
1282 	else if (PRC_IS_REDIRECT(cmd))
1283 		return;
1284 	/*
1285 	 * Source quench is depreciated.
1286 	 */
1287 	else if (cmd == PRC_QUENCH)
1288 		return;
1289 	/*
1290 	 * Hostdead is ugly because it goes linearly through all PCBs.
1291 	 * XXX: We never get this from ICMP, otherwise it makes an
1292 	 * excellent DoS attack on machines with many connections.
1293 	 */
1294 	else if (cmd == PRC_HOSTDEAD)
1295 		ip = NULL;
1296 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1297 		return;
1298 	if (ip != NULL) {
1299 		icp = (struct icmp *)((caddr_t)ip
1300 				      - offsetof(struct icmp, icmp_ip));
1301 		th = (struct tcphdr *)((caddr_t)ip
1302 				       + (ip->ip_hl << 2));
1303 		INP_INFO_WLOCK(&V_tcbinfo);
1304 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1305 		    ip->ip_src, th->th_sport, 0, NULL);
1306 		if (inp != NULL)  {
1307 			INP_WLOCK(inp);
1308 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1309 			    !(inp->inp_flags & INP_DROPPED) &&
1310 			    !(inp->inp_socket == NULL)) {
1311 				icmp_tcp_seq = htonl(th->th_seq);
1312 				tp = intotcpcb(inp);
1313 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1314 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1315 					if (cmd == PRC_MSGSIZE) {
1316 					    /*
1317 					     * MTU discovery:
1318 					     * If we got a needfrag set the MTU
1319 					     * in the route to the suggested new
1320 					     * value (if given) and then notify.
1321 					     */
1322 					    bzero(&inc, sizeof(inc));
1323 					    inc.inc_faddr = faddr;
1324 					    inc.inc_fibnum =
1325 						inp->inp_inc.inc_fibnum;
1326 
1327 					    mtu = ntohs(icp->icmp_nextmtu);
1328 					    /*
1329 					     * If no alternative MTU was
1330 					     * proposed, try the next smaller
1331 					     * one.  ip->ip_len has already
1332 					     * been swapped in icmp_input().
1333 					     */
1334 					    if (!mtu)
1335 						mtu = ip_next_mtu(ip->ip_len,
1336 						 1);
1337 					    if (mtu < max(296, V_tcp_minmss
1338 						 + sizeof(struct tcpiphdr)))
1339 						mtu = 0;
1340 					    if (!mtu)
1341 						mtu = V_tcp_mssdflt
1342 						 + sizeof(struct tcpiphdr);
1343 					    /*
1344 					     * Only cache the the MTU if it
1345 					     * is smaller than the interface
1346 					     * or route MTU.  tcp_mtudisc()
1347 					     * will do right thing by itself.
1348 					     */
1349 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1350 						tcp_hc_updatemtu(&inc, mtu);
1351 					}
1352 
1353 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1354 				}
1355 			}
1356 			if (inp != NULL)
1357 				INP_WUNLOCK(inp);
1358 		} else {
1359 			bzero(&inc, sizeof(inc));
1360 			inc.inc_fport = th->th_dport;
1361 			inc.inc_lport = th->th_sport;
1362 			inc.inc_faddr = faddr;
1363 			inc.inc_laddr = ip->ip_src;
1364 			syncache_unreach(&inc, th);
1365 		}
1366 		INP_INFO_WUNLOCK(&V_tcbinfo);
1367 	} else
1368 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1369 }
1370 
1371 #ifdef INET6
1372 void
1373 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1374 {
1375 	INIT_VNET_INET(curvnet);
1376 	struct tcphdr th;
1377 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1378 	struct ip6_hdr *ip6;
1379 	struct mbuf *m;
1380 	struct ip6ctlparam *ip6cp = NULL;
1381 	const struct sockaddr_in6 *sa6_src = NULL;
1382 	int off;
1383 	struct tcp_portonly {
1384 		u_int16_t th_sport;
1385 		u_int16_t th_dport;
1386 	} *thp;
1387 
1388 	if (sa->sa_family != AF_INET6 ||
1389 	    sa->sa_len != sizeof(struct sockaddr_in6))
1390 		return;
1391 
1392 	if (cmd == PRC_MSGSIZE)
1393 		notify = tcp_mtudisc;
1394 	else if (!PRC_IS_REDIRECT(cmd) &&
1395 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1396 		return;
1397 	/* Source quench is depreciated. */
1398 	else if (cmd == PRC_QUENCH)
1399 		return;
1400 
1401 	/* if the parameter is from icmp6, decode it. */
1402 	if (d != NULL) {
1403 		ip6cp = (struct ip6ctlparam *)d;
1404 		m = ip6cp->ip6c_m;
1405 		ip6 = ip6cp->ip6c_ip6;
1406 		off = ip6cp->ip6c_off;
1407 		sa6_src = ip6cp->ip6c_src;
1408 	} else {
1409 		m = NULL;
1410 		ip6 = NULL;
1411 		off = 0;	/* fool gcc */
1412 		sa6_src = &sa6_any;
1413 	}
1414 
1415 	if (ip6 != NULL) {
1416 		struct in_conninfo inc;
1417 		/*
1418 		 * XXX: We assume that when IPV6 is non NULL,
1419 		 * M and OFF are valid.
1420 		 */
1421 
1422 		/* check if we can safely examine src and dst ports */
1423 		if (m->m_pkthdr.len < off + sizeof(*thp))
1424 			return;
1425 
1426 		bzero(&th, sizeof(th));
1427 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1428 
1429 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1430 		    (struct sockaddr *)ip6cp->ip6c_src,
1431 		    th.th_sport, cmd, NULL, notify);
1432 
1433 		bzero(&inc, sizeof(inc));
1434 		inc.inc_fport = th.th_dport;
1435 		inc.inc_lport = th.th_sport;
1436 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1437 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1438 		inc.inc_flags |= INC_ISIPV6;
1439 		INP_INFO_WLOCK(&V_tcbinfo);
1440 		syncache_unreach(&inc, &th);
1441 		INP_INFO_WUNLOCK(&V_tcbinfo);
1442 	} else
1443 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1444 			      0, cmd, NULL, notify);
1445 }
1446 #endif /* INET6 */
1447 
1448 
1449 /*
1450  * Following is where TCP initial sequence number generation occurs.
1451  *
1452  * There are two places where we must use initial sequence numbers:
1453  * 1.  In SYN-ACK packets.
1454  * 2.  In SYN packets.
1455  *
1456  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1457  * tcp_syncache.c for details.
1458  *
1459  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1460  * depends on this property.  In addition, these ISNs should be
1461  * unguessable so as to prevent connection hijacking.  To satisfy
1462  * the requirements of this situation, the algorithm outlined in
1463  * RFC 1948 is used, with only small modifications.
1464  *
1465  * Implementation details:
1466  *
1467  * Time is based off the system timer, and is corrected so that it
1468  * increases by one megabyte per second.  This allows for proper
1469  * recycling on high speed LANs while still leaving over an hour
1470  * before rollover.
1471  *
1472  * As reading the *exact* system time is too expensive to be done
1473  * whenever setting up a TCP connection, we increment the time
1474  * offset in two ways.  First, a small random positive increment
1475  * is added to isn_offset for each connection that is set up.
1476  * Second, the function tcp_isn_tick fires once per clock tick
1477  * and increments isn_offset as necessary so that sequence numbers
1478  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1479  * random positive increments serve only to ensure that the same
1480  * exact sequence number is never sent out twice (as could otherwise
1481  * happen when a port is recycled in less than the system tick
1482  * interval.)
1483  *
1484  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1485  * between seeding of isn_secret.  This is normally set to zero,
1486  * as reseeding should not be necessary.
1487  *
1488  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1489  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1490  * general, this means holding an exclusive (write) lock.
1491  */
1492 
1493 #define ISN_BYTES_PER_SECOND 1048576
1494 #define ISN_STATIC_INCREMENT 4096
1495 #define ISN_RANDOM_INCREMENT (4096 - 1)
1496 
1497 #ifdef VIMAGE_GLOBALS
1498 static u_char isn_secret[32];
1499 static int isn_last_reseed;
1500 static u_int32_t isn_offset, isn_offset_old;
1501 #endif
1502 
1503 tcp_seq
1504 tcp_new_isn(struct tcpcb *tp)
1505 {
1506 	INIT_VNET_INET(tp->t_vnet);
1507 	MD5_CTX isn_ctx;
1508 	u_int32_t md5_buffer[4];
1509 	tcp_seq new_isn;
1510 
1511 	INP_WLOCK_ASSERT(tp->t_inpcb);
1512 
1513 	ISN_LOCK();
1514 	/* Seed if this is the first use, reseed if requested. */
1515 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1516 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1517 		< (u_int)ticks))) {
1518 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1519 		V_isn_last_reseed = ticks;
1520 	}
1521 
1522 	/* Compute the md5 hash and return the ISN. */
1523 	MD5Init(&isn_ctx);
1524 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1525 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1526 #ifdef INET6
1527 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1528 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1529 			  sizeof(struct in6_addr));
1530 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1531 			  sizeof(struct in6_addr));
1532 	} else
1533 #endif
1534 	{
1535 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1536 			  sizeof(struct in_addr));
1537 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1538 			  sizeof(struct in_addr));
1539 	}
1540 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1541 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1542 	new_isn = (tcp_seq) md5_buffer[0];
1543 	V_isn_offset += ISN_STATIC_INCREMENT +
1544 		(arc4random() & ISN_RANDOM_INCREMENT);
1545 	new_isn += V_isn_offset;
1546 	ISN_UNLOCK();
1547 	return (new_isn);
1548 }
1549 
1550 /*
1551  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1552  * to keep time flowing at a relatively constant rate.  If the random
1553  * increments have already pushed us past the projected offset, do nothing.
1554  */
1555 static void
1556 tcp_isn_tick(void *xtp)
1557 {
1558 	VNET_ITERATOR_DECL(vnet_iter);
1559 	u_int32_t projected_offset;
1560 
1561 	VNET_LIST_RLOCK();
1562 	ISN_LOCK();
1563 	VNET_FOREACH(vnet_iter) {
1564 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1565 		INIT_VNET_INET(curvnet);
1566 		projected_offset =
1567 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1568 
1569 		if (SEQ_GT(projected_offset, V_isn_offset))
1570 			V_isn_offset = projected_offset;
1571 
1572 		V_isn_offset_old = V_isn_offset;
1573 		CURVNET_RESTORE();
1574 	}
1575 	ISN_UNLOCK();
1576 	VNET_LIST_RUNLOCK();
1577 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1578 }
1579 
1580 /*
1581  * When a specific ICMP unreachable message is received and the
1582  * connection state is SYN-SENT, drop the connection.  This behavior
1583  * is controlled by the icmp_may_rst sysctl.
1584  */
1585 struct inpcb *
1586 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1587 {
1588 #ifdef INVARIANTS
1589 	INIT_VNET_INET(inp->inp_vnet);
1590 #endif
1591 	struct tcpcb *tp;
1592 
1593 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1594 	INP_WLOCK_ASSERT(inp);
1595 
1596 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1597 	    (inp->inp_flags & INP_DROPPED))
1598 		return (inp);
1599 
1600 	tp = intotcpcb(inp);
1601 	if (tp->t_state != TCPS_SYN_SENT)
1602 		return (inp);
1603 
1604 	tp = tcp_drop(tp, errno);
1605 	if (tp != NULL)
1606 		return (inp);
1607 	else
1608 		return (NULL);
1609 }
1610 
1611 /*
1612  * When `need fragmentation' ICMP is received, update our idea of the MSS
1613  * based on the new value in the route.  Also nudge TCP to send something,
1614  * since we know the packet we just sent was dropped.
1615  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1616  */
1617 struct inpcb *
1618 tcp_mtudisc(struct inpcb *inp, int errno)
1619 {
1620 	INIT_VNET_INET(inp->inp_vnet);
1621 	struct tcpcb *tp;
1622 	struct socket *so;
1623 
1624 	INP_WLOCK_ASSERT(inp);
1625 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1626 	    (inp->inp_flags & INP_DROPPED))
1627 		return (inp);
1628 
1629 	tp = intotcpcb(inp);
1630 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1631 
1632 	tcp_mss_update(tp, -1, NULL, NULL);
1633 
1634 	so = inp->inp_socket;
1635 	SOCKBUF_LOCK(&so->so_snd);
1636 	/* If the mss is larger than the socket buffer, decrease the mss. */
1637 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1638 		tp->t_maxseg = so->so_snd.sb_hiwat;
1639 	SOCKBUF_UNLOCK(&so->so_snd);
1640 
1641 	TCPSTAT_INC(tcps_mturesent);
1642 	tp->t_rtttime = 0;
1643 	tp->snd_nxt = tp->snd_una;
1644 	tcp_free_sackholes(tp);
1645 	tp->snd_recover = tp->snd_max;
1646 	if (tp->t_flags & TF_SACK_PERMIT)
1647 		EXIT_FASTRECOVERY(tp);
1648 	tcp_output_send(tp);
1649 	return (inp);
1650 }
1651 
1652 /*
1653  * Look-up the routing entry to the peer of this inpcb.  If no route
1654  * is found and it cannot be allocated, then return 0.  This routine
1655  * is called by TCP routines that access the rmx structure and by
1656  * tcp_mss_update to get the peer/interface MTU.
1657  */
1658 u_long
1659 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1660 {
1661 	struct route sro;
1662 	struct sockaddr_in *dst;
1663 	struct ifnet *ifp;
1664 	u_long maxmtu = 0;
1665 
1666 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1667 
1668 	bzero(&sro, sizeof(sro));
1669 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1670 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1671 		dst->sin_family = AF_INET;
1672 		dst->sin_len = sizeof(*dst);
1673 		dst->sin_addr = inc->inc_faddr;
1674 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1675 	}
1676 	if (sro.ro_rt != NULL) {
1677 		ifp = sro.ro_rt->rt_ifp;
1678 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1679 			maxmtu = ifp->if_mtu;
1680 		else
1681 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1682 
1683 		/* Report additional interface capabilities. */
1684 		if (flags != NULL) {
1685 			if (ifp->if_capenable & IFCAP_TSO4 &&
1686 			    ifp->if_hwassist & CSUM_TSO)
1687 				*flags |= CSUM_TSO;
1688 		}
1689 		RTFREE(sro.ro_rt);
1690 	}
1691 	return (maxmtu);
1692 }
1693 
1694 #ifdef INET6
1695 u_long
1696 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1697 {
1698 	struct route_in6 sro6;
1699 	struct ifnet *ifp;
1700 	u_long maxmtu = 0;
1701 
1702 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1703 
1704 	bzero(&sro6, sizeof(sro6));
1705 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1706 		sro6.ro_dst.sin6_family = AF_INET6;
1707 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1708 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1709 		rtalloc_ign((struct route *)&sro6, 0);
1710 	}
1711 	if (sro6.ro_rt != NULL) {
1712 		ifp = sro6.ro_rt->rt_ifp;
1713 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1714 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1715 		else
1716 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1717 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1718 
1719 		/* Report additional interface capabilities. */
1720 		if (flags != NULL) {
1721 			if (ifp->if_capenable & IFCAP_TSO6 &&
1722 			    ifp->if_hwassist & CSUM_TSO)
1723 				*flags |= CSUM_TSO;
1724 		}
1725 		RTFREE(sro6.ro_rt);
1726 	}
1727 
1728 	return (maxmtu);
1729 }
1730 #endif /* INET6 */
1731 
1732 #ifdef IPSEC
1733 /* compute ESP/AH header size for TCP, including outer IP header. */
1734 size_t
1735 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1736 {
1737 	struct inpcb *inp;
1738 	struct mbuf *m;
1739 	size_t hdrsiz;
1740 	struct ip *ip;
1741 #ifdef INET6
1742 	struct ip6_hdr *ip6;
1743 #endif
1744 	struct tcphdr *th;
1745 
1746 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1747 		return (0);
1748 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1749 	if (!m)
1750 		return (0);
1751 
1752 #ifdef INET6
1753 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1754 		ip6 = mtod(m, struct ip6_hdr *);
1755 		th = (struct tcphdr *)(ip6 + 1);
1756 		m->m_pkthdr.len = m->m_len =
1757 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1758 		tcpip_fillheaders(inp, ip6, th);
1759 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1760 	} else
1761 #endif /* INET6 */
1762 	{
1763 		ip = mtod(m, struct ip *);
1764 		th = (struct tcphdr *)(ip + 1);
1765 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1766 		tcpip_fillheaders(inp, ip, th);
1767 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1768 	}
1769 
1770 	m_free(m);
1771 	return (hdrsiz);
1772 }
1773 #endif /* IPSEC */
1774 
1775 /*
1776  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1777  *
1778  * This code attempts to calculate the bandwidth-delay product as a
1779  * means of determining the optimal window size to maximize bandwidth,
1780  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1781  * routers.  This code also does a fairly good job keeping RTTs in check
1782  * across slow links like modems.  We implement an algorithm which is very
1783  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1784  * transmitter side of a TCP connection and so only effects the transmit
1785  * side of the connection.
1786  *
1787  * BACKGROUND:  TCP makes no provision for the management of buffer space
1788  * at the end points or at the intermediate routers and switches.  A TCP
1789  * stream, whether using NewReno or not, will eventually buffer as
1790  * many packets as it is able and the only reason this typically works is
1791  * due to the fairly small default buffers made available for a connection
1792  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1793  * scaling it is now fairly easy for even a single TCP connection to blow-out
1794  * all available buffer space not only on the local interface, but on
1795  * intermediate routers and switches as well.  NewReno makes a misguided
1796  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1797  * then backing off, then steadily increasing the window again until another
1798  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1799  * is only made worse as network loads increase and the idea of intentionally
1800  * blowing out network buffers is, frankly, a terrible way to manage network
1801  * resources.
1802  *
1803  * It is far better to limit the transmit window prior to the failure
1804  * condition being achieved.  There are two general ways to do this:  First
1805  * you can 'scan' through different transmit window sizes and locate the
1806  * point where the RTT stops increasing, indicating that you have filled the
1807  * pipe, then scan backwards until you note that RTT stops decreasing, then
1808  * repeat ad-infinitum.  This method works in principle but has severe
1809  * implementation issues due to RTT variances, timer granularity, and
1810  * instability in the algorithm which can lead to many false positives and
1811  * create oscillations as well as interact badly with other TCP streams
1812  * implementing the same algorithm.
1813  *
1814  * The second method is to limit the window to the bandwidth delay product
1815  * of the link.  This is the method we implement.  RTT variances and our
1816  * own manipulation of the congestion window, bwnd, can potentially
1817  * destabilize the algorithm.  For this reason we have to stabilize the
1818  * elements used to calculate the window.  We do this by using the minimum
1819  * observed RTT, the long term average of the observed bandwidth, and
1820  * by adding two segments worth of slop.  It isn't perfect but it is able
1821  * to react to changing conditions and gives us a very stable basis on
1822  * which to extend the algorithm.
1823  */
1824 void
1825 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1826 {
1827 	INIT_VNET_INET(tp->t_vnet);
1828 	u_long bw;
1829 	u_long bwnd;
1830 	int save_ticks;
1831 
1832 	INP_WLOCK_ASSERT(tp->t_inpcb);
1833 
1834 	/*
1835 	 * If inflight_enable is disabled in the middle of a tcp connection,
1836 	 * make sure snd_bwnd is effectively disabled.
1837 	 */
1838 	if (V_tcp_inflight_enable == 0 ||
1839 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1840 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1841 		tp->snd_bandwidth = 0;
1842 		return;
1843 	}
1844 
1845 	/*
1846 	 * Figure out the bandwidth.  Due to the tick granularity this
1847 	 * is a very rough number and it MUST be averaged over a fairly
1848 	 * long period of time.  XXX we need to take into account a link
1849 	 * that is not using all available bandwidth, but for now our
1850 	 * slop will ramp us up if this case occurs and the bandwidth later
1851 	 * increases.
1852 	 *
1853 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1854 	 * effectively reset t_bw_rtttime for this case.
1855 	 */
1856 	save_ticks = ticks;
1857 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1858 		return;
1859 
1860 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1861 	    (save_ticks - tp->t_bw_rtttime);
1862 	tp->t_bw_rtttime = save_ticks;
1863 	tp->t_bw_rtseq = ack_seq;
1864 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1865 		return;
1866 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1867 
1868 	tp->snd_bandwidth = bw;
1869 
1870 	/*
1871 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1872 	 * segments.  The additional slop puts us squarely in the sweet
1873 	 * spot and also handles the bandwidth run-up case and stabilization.
1874 	 * Without the slop we could be locking ourselves into a lower
1875 	 * bandwidth.
1876 	 *
1877 	 * Situations Handled:
1878 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1879 	 *	    high speed LANs, allowing larger TCP buffers to be
1880 	 *	    specified, and also does a good job preventing
1881 	 *	    over-queueing of packets over choke points like modems
1882 	 *	    (at least for the transmit side).
1883 	 *
1884 	 *	(2) Is able to handle changing network loads (bandwidth
1885 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1886 	 *	    increases).
1887 	 *
1888 	 *	(3) Theoretically should stabilize in the face of multiple
1889 	 *	    connections implementing the same algorithm (this may need
1890 	 *	    a little work).
1891 	 *
1892 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1893 	 *	    be adjusted with a sysctl but typically only needs to be
1894 	 *	    on very slow connections.  A value no smaller then 5
1895 	 *	    should be used, but only reduce this default if you have
1896 	 *	    no other choice.
1897 	 */
1898 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1899 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1900 #undef USERTT
1901 
1902 	if (tcp_inflight_debug > 0) {
1903 		static int ltime;
1904 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1905 			ltime = ticks;
1906 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1907 			    tp,
1908 			    bw,
1909 			    tp->t_rttbest,
1910 			    tp->t_srtt,
1911 			    bwnd
1912 			);
1913 		}
1914 	}
1915 	if ((long)bwnd < V_tcp_inflight_min)
1916 		bwnd = V_tcp_inflight_min;
1917 	if (bwnd > V_tcp_inflight_max)
1918 		bwnd = V_tcp_inflight_max;
1919 	if ((long)bwnd < tp->t_maxseg * 2)
1920 		bwnd = tp->t_maxseg * 2;
1921 	tp->snd_bwnd = bwnd;
1922 }
1923 
1924 #ifdef TCP_SIGNATURE
1925 /*
1926  * Callback function invoked by m_apply() to digest TCP segment data
1927  * contained within an mbuf chain.
1928  */
1929 static int
1930 tcp_signature_apply(void *fstate, void *data, u_int len)
1931 {
1932 
1933 	MD5Update(fstate, (u_char *)data, len);
1934 	return (0);
1935 }
1936 
1937 /*
1938  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1939  *
1940  * Parameters:
1941  * m		pointer to head of mbuf chain
1942  * _unused
1943  * len		length of TCP segment data, excluding options
1944  * optlen	length of TCP segment options
1945  * buf		pointer to storage for computed MD5 digest
1946  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1947  *
1948  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1949  * When called from tcp_input(), we can be sure that th_sum has been
1950  * zeroed out and verified already.
1951  *
1952  * Return 0 if successful, otherwise return -1.
1953  *
1954  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1955  * search with the destination IP address, and a 'magic SPI' to be
1956  * determined by the application. This is hardcoded elsewhere to 1179
1957  * right now. Another branch of this code exists which uses the SPD to
1958  * specify per-application flows but it is unstable.
1959  */
1960 int
1961 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1962     u_char *buf, u_int direction)
1963 {
1964 	INIT_VNET_IPSEC(curvnet);
1965 	union sockaddr_union dst;
1966 	struct ippseudo ippseudo;
1967 	MD5_CTX ctx;
1968 	int doff;
1969 	struct ip *ip;
1970 	struct ipovly *ipovly;
1971 	struct secasvar *sav;
1972 	struct tcphdr *th;
1973 #ifdef INET6
1974 	struct ip6_hdr *ip6;
1975 	struct in6_addr in6;
1976 	char ip6buf[INET6_ADDRSTRLEN];
1977 	uint32_t plen;
1978 	uint16_t nhdr;
1979 #endif
1980 	u_short savecsum;
1981 
1982 	KASSERT(m != NULL, ("NULL mbuf chain"));
1983 	KASSERT(buf != NULL, ("NULL signature pointer"));
1984 
1985 	/* Extract the destination from the IP header in the mbuf. */
1986 	bzero(&dst, sizeof(union sockaddr_union));
1987 	ip = mtod(m, struct ip *);
1988 #ifdef INET6
1989 	ip6 = NULL;	/* Make the compiler happy. */
1990 #endif
1991 	switch (ip->ip_v) {
1992 	case IPVERSION:
1993 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1994 		dst.sa.sa_family = AF_INET;
1995 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1996 		    ip->ip_src : ip->ip_dst;
1997 		break;
1998 #ifdef INET6
1999 	case (IPV6_VERSION >> 4):
2000 		ip6 = mtod(m, struct ip6_hdr *);
2001 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2002 		dst.sa.sa_family = AF_INET6;
2003 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2004 		    ip6->ip6_src : ip6->ip6_dst;
2005 		break;
2006 #endif
2007 	default:
2008 		return (EINVAL);
2009 		/* NOTREACHED */
2010 		break;
2011 	}
2012 
2013 	/* Look up an SADB entry which matches the address of the peer. */
2014 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2015 	if (sav == NULL) {
2016 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2017 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2018 #ifdef INET6
2019 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2020 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2021 #endif
2022 			"(unsupported)"));
2023 		return (EINVAL);
2024 	}
2025 
2026 	MD5Init(&ctx);
2027 	/*
2028 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2029 	 *
2030 	 * XXX The ippseudo header MUST be digested in network byte order,
2031 	 * or else we'll fail the regression test. Assume all fields we've
2032 	 * been doing arithmetic on have been in host byte order.
2033 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2034 	 * tcp_output(), the underlying ip_len member has not yet been set.
2035 	 */
2036 	switch (ip->ip_v) {
2037 	case IPVERSION:
2038 		ipovly = (struct ipovly *)ip;
2039 		ippseudo.ippseudo_src = ipovly->ih_src;
2040 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2041 		ippseudo.ippseudo_pad = 0;
2042 		ippseudo.ippseudo_p = IPPROTO_TCP;
2043 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2044 		    optlen);
2045 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2046 
2047 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2048 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2049 		break;
2050 #ifdef INET6
2051 	/*
2052 	 * RFC 2385, 2.0  Proposal
2053 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2054 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2055 	 * extended next header value (to form 32 bits), and 32-bit segment
2056 	 * length.
2057 	 * Note: Upper-Layer Packet Length comes before Next Header.
2058 	 */
2059 	case (IPV6_VERSION >> 4):
2060 		in6 = ip6->ip6_src;
2061 		in6_clearscope(&in6);
2062 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2063 		in6 = ip6->ip6_dst;
2064 		in6_clearscope(&in6);
2065 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2066 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2067 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2068 		nhdr = 0;
2069 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2070 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2072 		nhdr = IPPROTO_TCP;
2073 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2074 
2075 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2076 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2077 		break;
2078 #endif
2079 	default:
2080 		return (EINVAL);
2081 		/* NOTREACHED */
2082 		break;
2083 	}
2084 
2085 
2086 	/*
2087 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2088 	 * The TCP checksum must be set to zero.
2089 	 */
2090 	savecsum = th->th_sum;
2091 	th->th_sum = 0;
2092 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2093 	th->th_sum = savecsum;
2094 
2095 	/*
2096 	 * Step 3: Update MD5 hash with TCP segment data.
2097 	 *         Use m_apply() to avoid an early m_pullup().
2098 	 */
2099 	if (len > 0)
2100 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2101 
2102 	/*
2103 	 * Step 4: Update MD5 hash with shared secret.
2104 	 */
2105 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2106 	MD5Final(buf, &ctx);
2107 
2108 	key_sa_recordxfer(sav, m);
2109 	KEY_FREESAV(&sav);
2110 	return (0);
2111 }
2112 #endif /* TCP_SIGNATURE */
2113 
2114 static int
2115 sysctl_drop(SYSCTL_HANDLER_ARGS)
2116 {
2117 	INIT_VNET_INET(curvnet);
2118 #ifdef INET6
2119 	INIT_VNET_INET6(curvnet);
2120 #endif
2121 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2122 	struct sockaddr_storage addrs[2];
2123 	struct inpcb *inp;
2124 	struct tcpcb *tp;
2125 	struct tcptw *tw;
2126 	struct sockaddr_in *fin, *lin;
2127 #ifdef INET6
2128 	struct sockaddr_in6 *fin6, *lin6;
2129 #endif
2130 	int error;
2131 
2132 	inp = NULL;
2133 	fin = lin = NULL;
2134 #ifdef INET6
2135 	fin6 = lin6 = NULL;
2136 #endif
2137 	error = 0;
2138 
2139 	if (req->oldptr != NULL || req->oldlen != 0)
2140 		return (EINVAL);
2141 	if (req->newptr == NULL)
2142 		return (EPERM);
2143 	if (req->newlen < sizeof(addrs))
2144 		return (ENOMEM);
2145 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2146 	if (error)
2147 		return (error);
2148 
2149 	switch (addrs[0].ss_family) {
2150 #ifdef INET6
2151 	case AF_INET6:
2152 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2153 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2154 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2155 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2156 			return (EINVAL);
2157 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2158 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2159 				return (EINVAL);
2160 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2161 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2162 			fin = (struct sockaddr_in *)&addrs[0];
2163 			lin = (struct sockaddr_in *)&addrs[1];
2164 			break;
2165 		}
2166 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2167 		if (error)
2168 			return (error);
2169 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2170 		if (error)
2171 			return (error);
2172 		break;
2173 #endif
2174 	case AF_INET:
2175 		fin = (struct sockaddr_in *)&addrs[0];
2176 		lin = (struct sockaddr_in *)&addrs[1];
2177 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2178 		    lin->sin_len != sizeof(struct sockaddr_in))
2179 			return (EINVAL);
2180 		break;
2181 	default:
2182 		return (EINVAL);
2183 	}
2184 	INP_INFO_WLOCK(&V_tcbinfo);
2185 	switch (addrs[0].ss_family) {
2186 #ifdef INET6
2187 	case AF_INET6:
2188 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2189 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2190 		    NULL);
2191 		break;
2192 #endif
2193 	case AF_INET:
2194 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2195 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2196 		break;
2197 	}
2198 	if (inp != NULL) {
2199 		INP_WLOCK(inp);
2200 		if (inp->inp_flags & INP_TIMEWAIT) {
2201 			/*
2202 			 * XXXRW: There currently exists a state where an
2203 			 * inpcb is present, but its timewait state has been
2204 			 * discarded.  For now, don't allow dropping of this
2205 			 * type of inpcb.
2206 			 */
2207 			tw = intotw(inp);
2208 			if (tw != NULL)
2209 				tcp_twclose(tw, 0);
2210 			else
2211 				INP_WUNLOCK(inp);
2212 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2213 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2214 			tp = intotcpcb(inp);
2215 			tp = tcp_drop(tp, ECONNABORTED);
2216 			if (tp != NULL)
2217 				INP_WUNLOCK(inp);
2218 		} else
2219 			INP_WUNLOCK(inp);
2220 	} else
2221 		error = ESRCH;
2222 	INP_INFO_WUNLOCK(&V_tcbinfo);
2223 	return (error);
2224 }
2225 
2226 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2227     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2228     0, sysctl_drop, "", "Drop TCP connection");
2229 
2230 /*
2231  * Generate a standardized TCP log line for use throughout the
2232  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2233  * allow use in the interrupt context.
2234  *
2235  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2236  * NB: The function may return NULL if memory allocation failed.
2237  *
2238  * Due to header inclusion and ordering limitations the struct ip
2239  * and ip6_hdr pointers have to be passed as void pointers.
2240  */
2241 char *
2242 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2243     const void *ip6hdr)
2244 {
2245 	char *s, *sp;
2246 	size_t size;
2247 	struct ip *ip;
2248 #ifdef INET6
2249 	const struct ip6_hdr *ip6;
2250 
2251 	ip6 = (const struct ip6_hdr *)ip6hdr;
2252 #endif /* INET6 */
2253 	ip = (struct ip *)ip4hdr;
2254 
2255 	/*
2256 	 * The log line looks like this:
2257 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2258 	 */
2259 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2260 	    sizeof(PRINT_TH_FLAGS) + 1 +
2261 #ifdef INET6
2262 	    2 * INET6_ADDRSTRLEN;
2263 #else
2264 	    2 * INET_ADDRSTRLEN;
2265 #endif /* INET6 */
2266 
2267 	/* Is logging enabled? */
2268 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2269 		return (NULL);
2270 
2271 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2272 	if (s == NULL)
2273 		return (NULL);
2274 
2275 	strcat(s, "TCP: [");
2276 	sp = s + strlen(s);
2277 
2278 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2279 		inet_ntoa_r(inc->inc_faddr, sp);
2280 		sp = s + strlen(s);
2281 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2282 		sp = s + strlen(s);
2283 		inet_ntoa_r(inc->inc_laddr, sp);
2284 		sp = s + strlen(s);
2285 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2286 #ifdef INET6
2287 	} else if (inc) {
2288 		ip6_sprintf(sp, &inc->inc6_faddr);
2289 		sp = s + strlen(s);
2290 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2291 		sp = s + strlen(s);
2292 		ip6_sprintf(sp, &inc->inc6_laddr);
2293 		sp = s + strlen(s);
2294 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2295 	} else if (ip6 && th) {
2296 		ip6_sprintf(sp, &ip6->ip6_src);
2297 		sp = s + strlen(s);
2298 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2299 		sp = s + strlen(s);
2300 		ip6_sprintf(sp, &ip6->ip6_dst);
2301 		sp = s + strlen(s);
2302 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2303 #endif /* INET6 */
2304 	} else if (ip && th) {
2305 		inet_ntoa_r(ip->ip_src, sp);
2306 		sp = s + strlen(s);
2307 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2308 		sp = s + strlen(s);
2309 		inet_ntoa_r(ip->ip_dst, sp);
2310 		sp = s + strlen(s);
2311 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2312 	} else {
2313 		free(s, M_TCPLOG);
2314 		return (NULL);
2315 	}
2316 	sp = s + strlen(s);
2317 	if (th)
2318 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2319 	if (*(s + size - 1) != '\0')
2320 		panic("%s: string too long", __func__);
2321 	return (s);
2322 }
2323