xref: /freebsd/sys/netinet/tcp_subr.c (revision d5fc25e5d6c52b306312784663ccad85923a9c76)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #ifdef INET6
49 #include <sys/domain.h>
50 #endif
51 #include <sys/priv.h>
52 #include <sys/proc.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 #include <sys/random.h>
57 #include <sys/vimage.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/route.h>
62 #include <net/if.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #include <netinet6/nd6.h>
80 #endif
81 #include <netinet/ip_icmp.h>
82 #include <netinet/tcp.h>
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #include <netinet/tcp_offload.h>
89 #ifdef INET6
90 #include <netinet6/tcp6_var.h>
91 #endif
92 #include <netinet/tcpip.h>
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #include <netinet/vinet.h>
97 #include <netinet6/ip6protosw.h>
98 #include <netinet6/vinet6.h>
99 
100 #ifdef IPSEC
101 #include <netipsec/ipsec.h>
102 #include <netipsec/xform.h>
103 #ifdef INET6
104 #include <netipsec/ipsec6.h>
105 #endif
106 #include <netipsec/key.h>
107 #include <sys/syslog.h>
108 #endif /*IPSEC*/
109 
110 #include <machine/in_cksum.h>
111 #include <sys/md5.h>
112 
113 #include <security/mac/mac_framework.h>
114 
115 #ifdef VIMAGE_GLOBALS
116 int	tcp_mssdflt;
117 #ifdef INET6
118 int	tcp_v6mssdflt;
119 #endif
120 int	tcp_minmss;
121 int	tcp_do_rfc1323;
122 static int	icmp_may_rst;
123 static int	tcp_isn_reseed_interval;
124 static int	tcp_inflight_enable;
125 static int	tcp_inflight_rttthresh;
126 static int	tcp_inflight_min;
127 static int	tcp_inflight_max;
128 static int	tcp_inflight_stab;
129 #endif
130 
131 static int
132 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
133 {
134 	INIT_VNET_INET(curvnet);
135 	int error, new;
136 
137 	new = V_tcp_mssdflt;
138 	error = sysctl_handle_int(oidp, &new, 0, req);
139 	if (error == 0 && req->newptr) {
140 		if (new < TCP_MINMSS)
141 			error = EINVAL;
142 		else
143 			V_tcp_mssdflt = new;
144 	}
145 	return (error);
146 }
147 
148 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
149     CTLTYPE_INT|CTLFLAG_RW, tcp_mssdflt, 0,
150     &sysctl_net_inet_tcp_mss_check, "I",
151     "Default TCP Maximum Segment Size");
152 
153 #ifdef INET6
154 static int
155 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
156 {
157 	INIT_VNET_INET(curvnet);
158 	int error, new;
159 
160 	new = V_tcp_v6mssdflt;
161 	error = sysctl_handle_int(oidp, &new, 0, req);
162 	if (error == 0 && req->newptr) {
163 		if (new < TCP_MINMSS)
164 			error = EINVAL;
165 		else
166 			V_tcp_v6mssdflt = new;
167 	}
168 	return (error);
169 }
170 
171 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
172     CTLTYPE_INT|CTLFLAG_RW, tcp_v6mssdflt, 0,
173     &sysctl_net_inet_tcp_mss_v6_check, "I",
174    "Default TCP Maximum Segment Size for IPv6");
175 #endif
176 
177 /*
178  * Minimum MSS we accept and use. This prevents DoS attacks where
179  * we are forced to a ridiculous low MSS like 20 and send hundreds
180  * of packets instead of one. The effect scales with the available
181  * bandwidth and quickly saturates the CPU and network interface
182  * with packet generation and sending. Set to zero to disable MINMSS
183  * checking. This setting prevents us from sending too small packets.
184  */
185 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
186     CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
187 
188 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
189     CTLFLAG_RW, tcp_do_rfc1323, 0,
190     "Enable rfc1323 (high performance TCP) extensions");
191 
192 static int	tcp_log_debug = 0;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
194     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
195 
196 static int	tcp_tcbhashsize = 0;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
198     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
199 
200 static int	do_tcpdrain = 1;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
202     "Enable tcp_drain routine for extra help when low on mbufs");
203 
204 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
205     CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs");
206 
207 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
208     CTLFLAG_RW, icmp_may_rst, 0,
209     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
210 
211 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
212     CTLFLAG_RW, tcp_isn_reseed_interval, 0,
213     "Seconds between reseeding of ISN secret");
214 
215 /*
216  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
217  * 1024 exists only for debugging.  A good production default would be
218  * something like 6100.
219  */
220 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
221     "TCP inflight data limiting");
222 
223 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
224     CTLFLAG_RW, tcp_inflight_enable, 0,
225     "Enable automatic TCP inflight data limiting");
226 
227 static int	tcp_inflight_debug = 0;
228 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
229     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
230 
231 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, rttthresh,
232     CTLTYPE_INT|CTLFLAG_RW, tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks,
233     "I", "RTT threshold below which inflight will deactivate itself");
234 
235 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
236     CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
237 
238 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
239     CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
240 
241 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
242     CTLFLAG_RW, tcp_inflight_stab, 0,
243     "Inflight Algorithm Stabilization 20 = 2 packets");
244 
245 #ifdef VIMAGE_GLOBALS
246 uma_zone_t sack_hole_zone;
247 #endif
248 
249 static struct inpcb *tcp_notify(struct inpcb *, int);
250 static void	tcp_isn_tick(void *);
251 
252 /*
253  * Target size of TCP PCB hash tables. Must be a power of two.
254  *
255  * Note that this can be overridden by the kernel environment
256  * variable net.inet.tcp.tcbhashsize
257  */
258 #ifndef TCBHASHSIZE
259 #define TCBHASHSIZE	512
260 #endif
261 
262 /*
263  * XXX
264  * Callouts should be moved into struct tcp directly.  They are currently
265  * separate because the tcpcb structure is exported to userland for sysctl
266  * parsing purposes, which do not know about callouts.
267  */
268 struct tcpcb_mem {
269 	struct	tcpcb		tcb;
270 	struct	tcp_timer	tt;
271 };
272 
273 #ifdef VIMAGE_GLOBALS
274 static uma_zone_t tcpcb_zone;
275 #endif
276 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
277 struct callout isn_callout;
278 static struct mtx isn_mtx;
279 
280 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
281 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
282 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
283 
284 /*
285  * TCP initialization.
286  */
287 static void
288 tcp_zone_change(void *tag)
289 {
290 	INIT_VNET_INET(curvnet);
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 #ifdef VIMAGE
362 	V_tcbinfo.ipi_vnet = curvnet;
363 #endif
364 	V_tcbinfo.ipi_listhead = &V_tcb;
365 	hashsize = TCBHASHSIZE;
366 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
367 	if (!powerof2(hashsize)) {
368 		printf("WARNING: TCB hash size not a power of 2\n");
369 		hashsize = 512; /* safe default */
370 	}
371 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
372 	    &V_tcbinfo.ipi_hashmask);
373 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
374 	    &V_tcbinfo.ipi_porthashmask);
375 	V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
376 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
377 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
378 	/*
379 	 * These have to be type stable for the benefit of the timers.
380 	 */
381 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
382 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
383 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
384 	tcp_tw_init();
385 	syncache_init();
386 	tcp_hc_init();
387 	tcp_reass_init();
388 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
389 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
390 
391 	/* Skip initialization of globals for non-default instances. */
392 	if (!IS_DEFAULT_VNET(curvnet))
393 		return;
394 
395 	/* XXX virtualize those bellow? */
396 	tcp_delacktime = TCPTV_DELACK;
397 	tcp_keepinit = TCPTV_KEEP_INIT;
398 	tcp_keepidle = TCPTV_KEEP_IDLE;
399 	tcp_keepintvl = TCPTV_KEEPINTVL;
400 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
401 	tcp_msl = TCPTV_MSL;
402 	tcp_rexmit_min = TCPTV_MIN;
403 	if (tcp_rexmit_min < 1)
404 		tcp_rexmit_min = 1;
405 	tcp_rexmit_slop = TCPTV_CPU_VAR;
406 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
407 	tcp_tcbhashsize = hashsize;
408 
409 #ifdef INET6
410 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
411 #else /* INET6 */
412 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
413 #endif /* INET6 */
414 	if (max_protohdr < TCP_MINPROTOHDR)
415 		max_protohdr = TCP_MINPROTOHDR;
416 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
417 		panic("tcp_init");
418 #undef TCP_MINPROTOHDR
419 
420 	ISN_LOCK_INIT();
421 	callout_init(&isn_callout, CALLOUT_MPSAFE);
422 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
423 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
424 		SHUTDOWN_PRI_DEFAULT);
425 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
426 		EVENTHANDLER_PRI_ANY);
427 }
428 
429 void
430 tcp_fini(void *xtp)
431 {
432 
433 	callout_stop(&isn_callout);
434 }
435 
436 /*
437  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
438  * tcp_template used to store this data in mbufs, but we now recopy it out
439  * of the tcpcb each time to conserve mbufs.
440  */
441 void
442 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
443 {
444 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
445 
446 	INP_WLOCK_ASSERT(inp);
447 
448 #ifdef INET6
449 	if ((inp->inp_vflag & INP_IPV6) != 0) {
450 		struct ip6_hdr *ip6;
451 
452 		ip6 = (struct ip6_hdr *)ip_ptr;
453 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
454 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
455 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
456 			(IPV6_VERSION & IPV6_VERSION_MASK);
457 		ip6->ip6_nxt = IPPROTO_TCP;
458 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
459 		ip6->ip6_src = inp->in6p_laddr;
460 		ip6->ip6_dst = inp->in6p_faddr;
461 	} else
462 #endif
463 	{
464 		struct ip *ip;
465 
466 		ip = (struct ip *)ip_ptr;
467 		ip->ip_v = IPVERSION;
468 		ip->ip_hl = 5;
469 		ip->ip_tos = inp->inp_ip_tos;
470 		ip->ip_len = 0;
471 		ip->ip_id = 0;
472 		ip->ip_off = 0;
473 		ip->ip_ttl = inp->inp_ip_ttl;
474 		ip->ip_sum = 0;
475 		ip->ip_p = IPPROTO_TCP;
476 		ip->ip_src = inp->inp_laddr;
477 		ip->ip_dst = inp->inp_faddr;
478 	}
479 	th->th_sport = inp->inp_lport;
480 	th->th_dport = inp->inp_fport;
481 	th->th_seq = 0;
482 	th->th_ack = 0;
483 	th->th_x2 = 0;
484 	th->th_off = 5;
485 	th->th_flags = 0;
486 	th->th_win = 0;
487 	th->th_urp = 0;
488 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
489 }
490 
491 /*
492  * Create template to be used to send tcp packets on a connection.
493  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
494  * use for this function is in keepalives, which use tcp_respond.
495  */
496 struct tcptemp *
497 tcpip_maketemplate(struct inpcb *inp)
498 {
499 	struct tcptemp *t;
500 
501 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
502 	if (t == NULL)
503 		return (NULL);
504 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
505 	return (t);
506 }
507 
508 /*
509  * Send a single message to the TCP at address specified by
510  * the given TCP/IP header.  If m == NULL, then we make a copy
511  * of the tcpiphdr at ti and send directly to the addressed host.
512  * This is used to force keep alive messages out using the TCP
513  * template for a connection.  If flags are given then we send
514  * a message back to the TCP which originated the * segment ti,
515  * and discard the mbuf containing it and any other attached mbufs.
516  *
517  * In any case the ack and sequence number of the transmitted
518  * segment are as specified by the parameters.
519  *
520  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
521  */
522 void
523 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
524     tcp_seq ack, tcp_seq seq, int flags)
525 {
526 	INIT_VNET_INET(curvnet);
527 	int tlen;
528 	int win = 0;
529 	struct ip *ip;
530 	struct tcphdr *nth;
531 #ifdef INET6
532 	struct ip6_hdr *ip6;
533 	int isipv6;
534 #endif /* INET6 */
535 	int ipflags = 0;
536 	struct inpcb *inp;
537 
538 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
539 
540 #ifdef INET6
541 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
542 	ip6 = ipgen;
543 #endif /* INET6 */
544 	ip = ipgen;
545 
546 	if (tp != NULL) {
547 		inp = tp->t_inpcb;
548 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
549 		INP_WLOCK_ASSERT(inp);
550 	} else
551 		inp = NULL;
552 
553 	if (tp != NULL) {
554 		if (!(flags & TH_RST)) {
555 			win = sbspace(&inp->inp_socket->so_rcv);
556 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
557 				win = (long)TCP_MAXWIN << tp->rcv_scale;
558 		}
559 	}
560 	if (m == NULL) {
561 		m = m_gethdr(M_DONTWAIT, MT_DATA);
562 		if (m == NULL)
563 			return;
564 		tlen = 0;
565 		m->m_data += max_linkhdr;
566 #ifdef INET6
567 		if (isipv6) {
568 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
569 			      sizeof(struct ip6_hdr));
570 			ip6 = mtod(m, struct ip6_hdr *);
571 			nth = (struct tcphdr *)(ip6 + 1);
572 		} else
573 #endif /* INET6 */
574 	      {
575 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
576 		ip = mtod(m, struct ip *);
577 		nth = (struct tcphdr *)(ip + 1);
578 	      }
579 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
580 		flags = TH_ACK;
581 	} else {
582 		/*
583 		 *  reuse the mbuf.
584 		 * XXX MRT We inherrit the FIB, which is lucky.
585 		 */
586 		m_freem(m->m_next);
587 		m->m_next = NULL;
588 		m->m_data = (caddr_t)ipgen;
589 		/* m_len is set later */
590 		tlen = 0;
591 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
592 #ifdef INET6
593 		if (isipv6) {
594 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
595 			nth = (struct tcphdr *)(ip6 + 1);
596 		} else
597 #endif /* INET6 */
598 	      {
599 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
600 		nth = (struct tcphdr *)(ip + 1);
601 	      }
602 		if (th != nth) {
603 			/*
604 			 * this is usually a case when an extension header
605 			 * exists between the IPv6 header and the
606 			 * TCP header.
607 			 */
608 			nth->th_sport = th->th_sport;
609 			nth->th_dport = th->th_dport;
610 		}
611 		xchg(nth->th_dport, nth->th_sport, uint16_t);
612 #undef xchg
613 	}
614 #ifdef INET6
615 	if (isipv6) {
616 		ip6->ip6_flow = 0;
617 		ip6->ip6_vfc = IPV6_VERSION;
618 		ip6->ip6_nxt = IPPROTO_TCP;
619 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
620 						tlen));
621 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
622 	} else
623 #endif
624 	{
625 		tlen += sizeof (struct tcpiphdr);
626 		ip->ip_len = tlen;
627 		ip->ip_ttl = V_ip_defttl;
628 		if (V_path_mtu_discovery)
629 			ip->ip_off |= IP_DF;
630 	}
631 	m->m_len = tlen;
632 	m->m_pkthdr.len = tlen;
633 	m->m_pkthdr.rcvif = NULL;
634 #ifdef MAC
635 	if (inp != NULL) {
636 		/*
637 		 * Packet is associated with a socket, so allow the
638 		 * label of the response to reflect the socket label.
639 		 */
640 		INP_WLOCK_ASSERT(inp);
641 		mac_inpcb_create_mbuf(inp, m);
642 	} else {
643 		/*
644 		 * Packet is not associated with a socket, so possibly
645 		 * update the label in place.
646 		 */
647 		mac_netinet_tcp_reply(m);
648 	}
649 #endif
650 	nth->th_seq = htonl(seq);
651 	nth->th_ack = htonl(ack);
652 	nth->th_x2 = 0;
653 	nth->th_off = sizeof (struct tcphdr) >> 2;
654 	nth->th_flags = flags;
655 	if (tp != NULL)
656 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
657 	else
658 		nth->th_win = htons((u_short)win);
659 	nth->th_urp = 0;
660 #ifdef INET6
661 	if (isipv6) {
662 		nth->th_sum = 0;
663 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
664 					sizeof(struct ip6_hdr),
665 					tlen - sizeof(struct ip6_hdr));
666 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
667 		    NULL, NULL);
668 	} else
669 #endif /* INET6 */
670 	{
671 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
672 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
673 		m->m_pkthdr.csum_flags = CSUM_TCP;
674 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
675 	}
676 #ifdef TCPDEBUG
677 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
678 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
679 #endif
680 #ifdef INET6
681 	if (isipv6)
682 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
683 	else
684 #endif /* INET6 */
685 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
686 }
687 
688 /*
689  * Create a new TCP control block, making an
690  * empty reassembly queue and hooking it to the argument
691  * protocol control block.  The `inp' parameter must have
692  * come from the zone allocator set up in tcp_init().
693  */
694 struct tcpcb *
695 tcp_newtcpcb(struct inpcb *inp)
696 {
697 	INIT_VNET_INET(inp->inp_vnet);
698 	struct tcpcb_mem *tm;
699 	struct tcpcb *tp;
700 #ifdef INET6
701 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
702 #endif /* INET6 */
703 
704 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
705 	if (tm == NULL)
706 		return (NULL);
707 	tp = &tm->tcb;
708 #ifdef VIMAGE
709 	tp->t_vnet = inp->inp_vnet;
710 #endif
711 	tp->t_timers = &tm->tt;
712 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
713 	tp->t_maxseg = tp->t_maxopd =
714 #ifdef INET6
715 		isipv6 ? V_tcp_v6mssdflt :
716 #endif /* INET6 */
717 		V_tcp_mssdflt;
718 
719 	/* Set up our timeouts. */
720 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
721 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
722 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
723 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
724 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
725 
726 	if (V_tcp_do_rfc1323)
727 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
728 	if (V_tcp_do_sack)
729 		tp->t_flags |= TF_SACK_PERMIT;
730 	TAILQ_INIT(&tp->snd_holes);
731 	tp->t_inpcb = inp;	/* XXX */
732 	/*
733 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
734 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
735 	 * reasonable initial retransmit time.
736 	 */
737 	tp->t_srtt = TCPTV_SRTTBASE;
738 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
739 	tp->t_rttmin = tcp_rexmit_min;
740 	tp->t_rxtcur = TCPTV_RTOBASE;
741 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
742 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
743 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
744 	tp->t_rcvtime = ticks;
745 	tp->t_bw_rtttime = ticks;
746 	/*
747 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
748 	 * because the socket may be bound to an IPv6 wildcard address,
749 	 * which may match an IPv4-mapped IPv6 address.
750 	 */
751 	inp->inp_ip_ttl = V_ip_defttl;
752 	inp->inp_ppcb = tp;
753 	return (tp);		/* XXX */
754 }
755 
756 /*
757  * Drop a TCP connection, reporting
758  * the specified error.  If connection is synchronized,
759  * then send a RST to peer.
760  */
761 struct tcpcb *
762 tcp_drop(struct tcpcb *tp, int errno)
763 {
764 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
765 	struct socket *so = tp->t_inpcb->inp_socket;
766 
767 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
768 	INP_WLOCK_ASSERT(tp->t_inpcb);
769 
770 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
771 		tp->t_state = TCPS_CLOSED;
772 		(void) tcp_output_reset(tp);
773 		TCPSTAT_INC(tcps_drops);
774 	} else
775 		TCPSTAT_INC(tcps_conndrops);
776 	if (errno == ETIMEDOUT && tp->t_softerror)
777 		errno = tp->t_softerror;
778 	so->so_error = errno;
779 	return (tcp_close(tp));
780 }
781 
782 void
783 tcp_discardcb(struct tcpcb *tp)
784 {
785 	INIT_VNET_INET(tp->t_vnet);
786 	struct tseg_qent *q;
787 	struct inpcb *inp = tp->t_inpcb;
788 	struct socket *so = inp->inp_socket;
789 #ifdef INET6
790 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
791 #endif /* INET6 */
792 
793 	INP_WLOCK_ASSERT(inp);
794 
795 	/*
796 	 * Make sure that all of our timers are stopped before we
797 	 * delete the PCB.
798 	 */
799 	callout_stop(&tp->t_timers->tt_rexmt);
800 	callout_stop(&tp->t_timers->tt_persist);
801 	callout_stop(&tp->t_timers->tt_keep);
802 	callout_stop(&tp->t_timers->tt_2msl);
803 	callout_stop(&tp->t_timers->tt_delack);
804 
805 	/*
806 	 * If we got enough samples through the srtt filter,
807 	 * save the rtt and rttvar in the routing entry.
808 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
809 	 * 4 samples is enough for the srtt filter to converge
810 	 * to within enough % of the correct value; fewer samples
811 	 * and we could save a bogus rtt. The danger is not high
812 	 * as tcp quickly recovers from everything.
813 	 * XXX: Works very well but needs some more statistics!
814 	 */
815 	if (tp->t_rttupdated >= 4) {
816 		struct hc_metrics_lite metrics;
817 		u_long ssthresh;
818 
819 		bzero(&metrics, sizeof(metrics));
820 		/*
821 		 * Update the ssthresh always when the conditions below
822 		 * are satisfied. This gives us better new start value
823 		 * for the congestion avoidance for new connections.
824 		 * ssthresh is only set if packet loss occured on a session.
825 		 *
826 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
827 		 * being torn down.  Ideally this code would not use 'so'.
828 		 */
829 		ssthresh = tp->snd_ssthresh;
830 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
831 			/*
832 			 * convert the limit from user data bytes to
833 			 * packets then to packet data bytes.
834 			 */
835 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
836 			if (ssthresh < 2)
837 				ssthresh = 2;
838 			ssthresh *= (u_long)(tp->t_maxseg +
839 #ifdef INET6
840 				      (isipv6 ? sizeof (struct ip6_hdr) +
841 					       sizeof (struct tcphdr) :
842 #endif
843 				       sizeof (struct tcpiphdr)
844 #ifdef INET6
845 				       )
846 #endif
847 				      );
848 		} else
849 			ssthresh = 0;
850 		metrics.rmx_ssthresh = ssthresh;
851 
852 		metrics.rmx_rtt = tp->t_srtt;
853 		metrics.rmx_rttvar = tp->t_rttvar;
854 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
855 		metrics.rmx_bandwidth = tp->snd_bandwidth;
856 		metrics.rmx_cwnd = tp->snd_cwnd;
857 		metrics.rmx_sendpipe = 0;
858 		metrics.rmx_recvpipe = 0;
859 
860 		tcp_hc_update(&inp->inp_inc, &metrics);
861 	}
862 
863 	/* free the reassembly queue, if any */
864 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
865 		LIST_REMOVE(q, tqe_q);
866 		m_freem(q->tqe_m);
867 		uma_zfree(V_tcp_reass_zone, q);
868 		tp->t_segqlen--;
869 		V_tcp_reass_qsize--;
870 	}
871 	/* Disconnect offload device, if any. */
872 	tcp_offload_detach(tp);
873 
874 	tcp_free_sackholes(tp);
875 	inp->inp_ppcb = NULL;
876 	tp->t_inpcb = NULL;
877 	uma_zfree(V_tcpcb_zone, tp);
878 }
879 
880 /*
881  * Attempt to close a TCP control block, marking it as dropped, and freeing
882  * the socket if we hold the only reference.
883  */
884 struct tcpcb *
885 tcp_close(struct tcpcb *tp)
886 {
887 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
888 	struct inpcb *inp = tp->t_inpcb;
889 	struct socket *so;
890 
891 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
892 	INP_WLOCK_ASSERT(inp);
893 
894 	/* Notify any offload devices of listener close */
895 	if (tp->t_state == TCPS_LISTEN)
896 		tcp_offload_listen_close(tp);
897 	in_pcbdrop(inp);
898 	TCPSTAT_INC(tcps_closed);
899 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
900 	so = inp->inp_socket;
901 	soisdisconnected(so);
902 	if (inp->inp_flags & INP_SOCKREF) {
903 		KASSERT(so->so_state & SS_PROTOREF,
904 		    ("tcp_close: !SS_PROTOREF"));
905 		inp->inp_flags &= ~INP_SOCKREF;
906 		INP_WUNLOCK(inp);
907 		ACCEPT_LOCK();
908 		SOCK_LOCK(so);
909 		so->so_state &= ~SS_PROTOREF;
910 		sofree(so);
911 		return (NULL);
912 	}
913 	return (tp);
914 }
915 
916 void
917 tcp_drain(void)
918 {
919 	VNET_ITERATOR_DECL(vnet_iter);
920 
921 	if (!do_tcpdrain)
922 		return;
923 
924 	VNET_LIST_RLOCK();
925 	VNET_FOREACH(vnet_iter) {
926 		CURVNET_SET(vnet_iter);
927 		INIT_VNET_INET(vnet_iter);
928 		struct inpcb *inpb;
929 		struct tcpcb *tcpb;
930 		struct tseg_qent *te;
931 
932 	/*
933 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
934 	 * if there is one...
935 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
936 	 *      reassembly queue should be flushed, but in a situation
937 	 *	where we're really low on mbufs, this is potentially
938 	 *	usefull.
939 	 */
940 		INP_INFO_RLOCK(&V_tcbinfo);
941 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
942 			if (inpb->inp_flags & INP_TIMEWAIT)
943 				continue;
944 			INP_WLOCK(inpb);
945 			if ((tcpb = intotcpcb(inpb)) != NULL) {
946 				while ((te = LIST_FIRST(&tcpb->t_segq))
947 			            != NULL) {
948 					LIST_REMOVE(te, tqe_q);
949 					m_freem(te->tqe_m);
950 					uma_zfree(V_tcp_reass_zone, te);
951 					tcpb->t_segqlen--;
952 					V_tcp_reass_qsize--;
953 				}
954 				tcp_clean_sackreport(tcpb);
955 			}
956 			INP_WUNLOCK(inpb);
957 		}
958 		INP_INFO_RUNLOCK(&V_tcbinfo);
959 		CURVNET_RESTORE();
960 	}
961 	VNET_LIST_RUNLOCK();
962 }
963 
964 /*
965  * Notify a tcp user of an asynchronous error;
966  * store error as soft error, but wake up user
967  * (for now, won't do anything until can select for soft error).
968  *
969  * Do not wake up user since there currently is no mechanism for
970  * reporting soft errors (yet - a kqueue filter may be added).
971  */
972 static struct inpcb *
973 tcp_notify(struct inpcb *inp, int error)
974 {
975 	struct tcpcb *tp;
976 #ifdef INVARIANTS
977 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
978 #endif
979 
980 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
981 	INP_WLOCK_ASSERT(inp);
982 
983 	if ((inp->inp_flags & INP_TIMEWAIT) ||
984 	    (inp->inp_flags & INP_DROPPED))
985 		return (inp);
986 
987 	tp = intotcpcb(inp);
988 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
989 
990 	/*
991 	 * Ignore some errors if we are hooked up.
992 	 * If connection hasn't completed, has retransmitted several times,
993 	 * and receives a second error, give up now.  This is better
994 	 * than waiting a long time to establish a connection that
995 	 * can never complete.
996 	 */
997 	if (tp->t_state == TCPS_ESTABLISHED &&
998 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
999 	     error == EHOSTDOWN)) {
1000 		return (inp);
1001 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1002 	    tp->t_softerror) {
1003 		tp = tcp_drop(tp, error);
1004 		if (tp != NULL)
1005 			return (inp);
1006 		else
1007 			return (NULL);
1008 	} else {
1009 		tp->t_softerror = error;
1010 		return (inp);
1011 	}
1012 #if 0
1013 	wakeup( &so->so_timeo);
1014 	sorwakeup(so);
1015 	sowwakeup(so);
1016 #endif
1017 }
1018 
1019 static int
1020 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1021 {
1022 	INIT_VNET_INET(curvnet);
1023 	int error, i, m, n, pcb_count;
1024 	struct inpcb *inp, **inp_list;
1025 	inp_gen_t gencnt;
1026 	struct xinpgen xig;
1027 
1028 	/*
1029 	 * The process of preparing the TCB list is too time-consuming and
1030 	 * resource-intensive to repeat twice on every request.
1031 	 */
1032 	if (req->oldptr == NULL) {
1033 		m = syncache_pcbcount();
1034 		n = V_tcbinfo.ipi_count;
1035 		req->oldidx = 2 * (sizeof xig)
1036 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1037 		return (0);
1038 	}
1039 
1040 	if (req->newptr != NULL)
1041 		return (EPERM);
1042 
1043 	/*
1044 	 * OK, now we're committed to doing something.
1045 	 */
1046 	INP_INFO_RLOCK(&V_tcbinfo);
1047 	gencnt = V_tcbinfo.ipi_gencnt;
1048 	n = V_tcbinfo.ipi_count;
1049 	INP_INFO_RUNLOCK(&V_tcbinfo);
1050 
1051 	m = syncache_pcbcount();
1052 
1053 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1054 		+ (n + m) * sizeof(struct xtcpcb));
1055 	if (error != 0)
1056 		return (error);
1057 
1058 	xig.xig_len = sizeof xig;
1059 	xig.xig_count = n + m;
1060 	xig.xig_gen = gencnt;
1061 	xig.xig_sogen = so_gencnt;
1062 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1063 	if (error)
1064 		return (error);
1065 
1066 	error = syncache_pcblist(req, m, &pcb_count);
1067 	if (error)
1068 		return (error);
1069 
1070 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1071 	if (inp_list == NULL)
1072 		return (ENOMEM);
1073 
1074 	INP_INFO_RLOCK(&V_tcbinfo);
1075 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1076 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1077 		INP_RLOCK(inp);
1078 		if (inp->inp_gencnt <= gencnt) {
1079 			/*
1080 			 * XXX: This use of cr_cansee(), introduced with
1081 			 * TCP state changes, is not quite right, but for
1082 			 * now, better than nothing.
1083 			 */
1084 			if (inp->inp_flags & INP_TIMEWAIT) {
1085 				if (intotw(inp) != NULL)
1086 					error = cr_cansee(req->td->td_ucred,
1087 					    intotw(inp)->tw_cred);
1088 				else
1089 					error = EINVAL;	/* Skip this inp. */
1090 			} else
1091 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1092 			if (error == 0)
1093 				inp_list[i++] = inp;
1094 		}
1095 		INP_RUNLOCK(inp);
1096 	}
1097 	INP_INFO_RUNLOCK(&V_tcbinfo);
1098 	n = i;
1099 
1100 	error = 0;
1101 	for (i = 0; i < n; i++) {
1102 		inp = inp_list[i];
1103 		INP_RLOCK(inp);
1104 		if (inp->inp_gencnt <= gencnt) {
1105 			struct xtcpcb xt;
1106 			void *inp_ppcb;
1107 
1108 			bzero(&xt, sizeof(xt));
1109 			xt.xt_len = sizeof xt;
1110 			/* XXX should avoid extra copy */
1111 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1112 			inp_ppcb = inp->inp_ppcb;
1113 			if (inp_ppcb == NULL)
1114 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1115 			else if (inp->inp_flags & INP_TIMEWAIT) {
1116 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1117 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1118 			} else
1119 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1120 			if (inp->inp_socket != NULL)
1121 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1122 			else {
1123 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1124 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1125 			}
1126 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1127 			INP_RUNLOCK(inp);
1128 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1129 		} else
1130 			INP_RUNLOCK(inp);
1131 
1132 	}
1133 	if (!error) {
1134 		/*
1135 		 * Give the user an updated idea of our state.
1136 		 * If the generation differs from what we told
1137 		 * her before, she knows that something happened
1138 		 * while we were processing this request, and it
1139 		 * might be necessary to retry.
1140 		 */
1141 		INP_INFO_RLOCK(&V_tcbinfo);
1142 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1143 		xig.xig_sogen = so_gencnt;
1144 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1145 		INP_INFO_RUNLOCK(&V_tcbinfo);
1146 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1147 	}
1148 	free(inp_list, M_TEMP);
1149 	return (error);
1150 }
1151 
1152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1153     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1154 
1155 static int
1156 tcp_getcred(SYSCTL_HANDLER_ARGS)
1157 {
1158 	INIT_VNET_INET(curvnet);
1159 	struct xucred xuc;
1160 	struct sockaddr_in addrs[2];
1161 	struct inpcb *inp;
1162 	int error;
1163 
1164 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1165 	if (error)
1166 		return (error);
1167 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1168 	if (error)
1169 		return (error);
1170 	INP_INFO_RLOCK(&V_tcbinfo);
1171 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1172 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1173 	if (inp != NULL) {
1174 		INP_RLOCK(inp);
1175 		INP_INFO_RUNLOCK(&V_tcbinfo);
1176 		if (inp->inp_socket == NULL)
1177 			error = ENOENT;
1178 		if (error == 0)
1179 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1180 		if (error == 0)
1181 			cru2x(inp->inp_cred, &xuc);
1182 		INP_RUNLOCK(inp);
1183 	} else {
1184 		INP_INFO_RUNLOCK(&V_tcbinfo);
1185 		error = ENOENT;
1186 	}
1187 	if (error == 0)
1188 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1189 	return (error);
1190 }
1191 
1192 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1193     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1194     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1195 
1196 #ifdef INET6
1197 static int
1198 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1199 {
1200 	INIT_VNET_INET(curvnet);
1201 	INIT_VNET_INET6(curvnet);
1202 	struct xucred xuc;
1203 	struct sockaddr_in6 addrs[2];
1204 	struct inpcb *inp;
1205 	int error, mapped = 0;
1206 
1207 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1208 	if (error)
1209 		return (error);
1210 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1211 	if (error)
1212 		return (error);
1213 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1214 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1215 		return (error);
1216 	}
1217 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1218 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1219 			mapped = 1;
1220 		else
1221 			return (EINVAL);
1222 	}
1223 
1224 	INP_INFO_RLOCK(&V_tcbinfo);
1225 	if (mapped == 1)
1226 		inp = in_pcblookup_hash(&V_tcbinfo,
1227 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1228 			addrs[1].sin6_port,
1229 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1230 			addrs[0].sin6_port,
1231 			0, NULL);
1232 	else
1233 		inp = in6_pcblookup_hash(&V_tcbinfo,
1234 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1235 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1236 	if (inp != NULL) {
1237 		INP_RLOCK(inp);
1238 		INP_INFO_RUNLOCK(&V_tcbinfo);
1239 		if (inp->inp_socket == NULL)
1240 			error = ENOENT;
1241 		if (error == 0)
1242 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1243 		if (error == 0)
1244 			cru2x(inp->inp_cred, &xuc);
1245 		INP_RUNLOCK(inp);
1246 	} else {
1247 		INP_INFO_RUNLOCK(&V_tcbinfo);
1248 		error = ENOENT;
1249 	}
1250 	if (error == 0)
1251 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1252 	return (error);
1253 }
1254 
1255 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1256     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1257     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1258 #endif
1259 
1260 
1261 void
1262 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1263 {
1264 	INIT_VNET_INET(curvnet);
1265 	struct ip *ip = vip;
1266 	struct tcphdr *th;
1267 	struct in_addr faddr;
1268 	struct inpcb *inp;
1269 	struct tcpcb *tp;
1270 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1271 	struct icmp *icp;
1272 	struct in_conninfo inc;
1273 	tcp_seq icmp_tcp_seq;
1274 	int mtu;
1275 
1276 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1277 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1278 		return;
1279 
1280 	if (cmd == PRC_MSGSIZE)
1281 		notify = tcp_mtudisc;
1282 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1283 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1284 		notify = tcp_drop_syn_sent;
1285 	/*
1286 	 * Redirects don't need to be handled up here.
1287 	 */
1288 	else if (PRC_IS_REDIRECT(cmd))
1289 		return;
1290 	/*
1291 	 * Source quench is depreciated.
1292 	 */
1293 	else if (cmd == PRC_QUENCH)
1294 		return;
1295 	/*
1296 	 * Hostdead is ugly because it goes linearly through all PCBs.
1297 	 * XXX: We never get this from ICMP, otherwise it makes an
1298 	 * excellent DoS attack on machines with many connections.
1299 	 */
1300 	else if (cmd == PRC_HOSTDEAD)
1301 		ip = NULL;
1302 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1303 		return;
1304 	if (ip != NULL) {
1305 		icp = (struct icmp *)((caddr_t)ip
1306 				      - offsetof(struct icmp, icmp_ip));
1307 		th = (struct tcphdr *)((caddr_t)ip
1308 				       + (ip->ip_hl << 2));
1309 		INP_INFO_WLOCK(&V_tcbinfo);
1310 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1311 		    ip->ip_src, th->th_sport, 0, NULL);
1312 		if (inp != NULL)  {
1313 			INP_WLOCK(inp);
1314 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1315 			    !(inp->inp_flags & INP_DROPPED) &&
1316 			    !(inp->inp_socket == NULL)) {
1317 				icmp_tcp_seq = htonl(th->th_seq);
1318 				tp = intotcpcb(inp);
1319 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1320 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1321 					if (cmd == PRC_MSGSIZE) {
1322 					    /*
1323 					     * MTU discovery:
1324 					     * If we got a needfrag set the MTU
1325 					     * in the route to the suggested new
1326 					     * value (if given) and then notify.
1327 					     */
1328 					    bzero(&inc, sizeof(inc));
1329 					    inc.inc_faddr = faddr;
1330 					    inc.inc_fibnum =
1331 						inp->inp_inc.inc_fibnum;
1332 
1333 					    mtu = ntohs(icp->icmp_nextmtu);
1334 					    /*
1335 					     * If no alternative MTU was
1336 					     * proposed, try the next smaller
1337 					     * one.  ip->ip_len has already
1338 					     * been swapped in icmp_input().
1339 					     */
1340 					    if (!mtu)
1341 						mtu = ip_next_mtu(ip->ip_len,
1342 						 1);
1343 					    if (mtu < max(296, V_tcp_minmss
1344 						 + sizeof(struct tcpiphdr)))
1345 						mtu = 0;
1346 					    if (!mtu)
1347 						mtu = V_tcp_mssdflt
1348 						 + sizeof(struct tcpiphdr);
1349 					    /*
1350 					     * Only cache the the MTU if it
1351 					     * is smaller than the interface
1352 					     * or route MTU.  tcp_mtudisc()
1353 					     * will do right thing by itself.
1354 					     */
1355 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1356 						tcp_hc_updatemtu(&inc, mtu);
1357 					}
1358 
1359 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1360 				}
1361 			}
1362 			if (inp != NULL)
1363 				INP_WUNLOCK(inp);
1364 		} else {
1365 			bzero(&inc, sizeof(inc));
1366 			inc.inc_fport = th->th_dport;
1367 			inc.inc_lport = th->th_sport;
1368 			inc.inc_faddr = faddr;
1369 			inc.inc_laddr = ip->ip_src;
1370 			syncache_unreach(&inc, th);
1371 		}
1372 		INP_INFO_WUNLOCK(&V_tcbinfo);
1373 	} else
1374 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1375 }
1376 
1377 #ifdef INET6
1378 void
1379 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1380 {
1381 	INIT_VNET_INET(curvnet);
1382 	struct tcphdr th;
1383 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1384 	struct ip6_hdr *ip6;
1385 	struct mbuf *m;
1386 	struct ip6ctlparam *ip6cp = NULL;
1387 	const struct sockaddr_in6 *sa6_src = NULL;
1388 	int off;
1389 	struct tcp_portonly {
1390 		u_int16_t th_sport;
1391 		u_int16_t th_dport;
1392 	} *thp;
1393 
1394 	if (sa->sa_family != AF_INET6 ||
1395 	    sa->sa_len != sizeof(struct sockaddr_in6))
1396 		return;
1397 
1398 	if (cmd == PRC_MSGSIZE)
1399 		notify = tcp_mtudisc;
1400 	else if (!PRC_IS_REDIRECT(cmd) &&
1401 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1402 		return;
1403 	/* Source quench is depreciated. */
1404 	else if (cmd == PRC_QUENCH)
1405 		return;
1406 
1407 	/* if the parameter is from icmp6, decode it. */
1408 	if (d != NULL) {
1409 		ip6cp = (struct ip6ctlparam *)d;
1410 		m = ip6cp->ip6c_m;
1411 		ip6 = ip6cp->ip6c_ip6;
1412 		off = ip6cp->ip6c_off;
1413 		sa6_src = ip6cp->ip6c_src;
1414 	} else {
1415 		m = NULL;
1416 		ip6 = NULL;
1417 		off = 0;	/* fool gcc */
1418 		sa6_src = &sa6_any;
1419 	}
1420 
1421 	if (ip6 != NULL) {
1422 		struct in_conninfo inc;
1423 		/*
1424 		 * XXX: We assume that when IPV6 is non NULL,
1425 		 * M and OFF are valid.
1426 		 */
1427 
1428 		/* check if we can safely examine src and dst ports */
1429 		if (m->m_pkthdr.len < off + sizeof(*thp))
1430 			return;
1431 
1432 		bzero(&th, sizeof(th));
1433 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1434 
1435 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1436 		    (struct sockaddr *)ip6cp->ip6c_src,
1437 		    th.th_sport, cmd, NULL, notify);
1438 
1439 		bzero(&inc, sizeof(inc));
1440 		inc.inc_fport = th.th_dport;
1441 		inc.inc_lport = th.th_sport;
1442 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1443 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1444 		inc.inc_flags |= INC_ISIPV6;
1445 		INP_INFO_WLOCK(&V_tcbinfo);
1446 		syncache_unreach(&inc, &th);
1447 		INP_INFO_WUNLOCK(&V_tcbinfo);
1448 	} else
1449 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1450 			      0, cmd, NULL, notify);
1451 }
1452 #endif /* INET6 */
1453 
1454 
1455 /*
1456  * Following is where TCP initial sequence number generation occurs.
1457  *
1458  * There are two places where we must use initial sequence numbers:
1459  * 1.  In SYN-ACK packets.
1460  * 2.  In SYN packets.
1461  *
1462  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1463  * tcp_syncache.c for details.
1464  *
1465  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1466  * depends on this property.  In addition, these ISNs should be
1467  * unguessable so as to prevent connection hijacking.  To satisfy
1468  * the requirements of this situation, the algorithm outlined in
1469  * RFC 1948 is used, with only small modifications.
1470  *
1471  * Implementation details:
1472  *
1473  * Time is based off the system timer, and is corrected so that it
1474  * increases by one megabyte per second.  This allows for proper
1475  * recycling on high speed LANs while still leaving over an hour
1476  * before rollover.
1477  *
1478  * As reading the *exact* system time is too expensive to be done
1479  * whenever setting up a TCP connection, we increment the time
1480  * offset in two ways.  First, a small random positive increment
1481  * is added to isn_offset for each connection that is set up.
1482  * Second, the function tcp_isn_tick fires once per clock tick
1483  * and increments isn_offset as necessary so that sequence numbers
1484  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1485  * random positive increments serve only to ensure that the same
1486  * exact sequence number is never sent out twice (as could otherwise
1487  * happen when a port is recycled in less than the system tick
1488  * interval.)
1489  *
1490  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1491  * between seeding of isn_secret.  This is normally set to zero,
1492  * as reseeding should not be necessary.
1493  *
1494  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1495  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1496  * general, this means holding an exclusive (write) lock.
1497  */
1498 
1499 #define ISN_BYTES_PER_SECOND 1048576
1500 #define ISN_STATIC_INCREMENT 4096
1501 #define ISN_RANDOM_INCREMENT (4096 - 1)
1502 
1503 #ifdef VIMAGE_GLOBALS
1504 static u_char isn_secret[32];
1505 static int isn_last_reseed;
1506 static u_int32_t isn_offset, isn_offset_old;
1507 #endif
1508 
1509 tcp_seq
1510 tcp_new_isn(struct tcpcb *tp)
1511 {
1512 	INIT_VNET_INET(tp->t_vnet);
1513 	MD5_CTX isn_ctx;
1514 	u_int32_t md5_buffer[4];
1515 	tcp_seq new_isn;
1516 
1517 	INP_WLOCK_ASSERT(tp->t_inpcb);
1518 
1519 	ISN_LOCK();
1520 	/* Seed if this is the first use, reseed if requested. */
1521 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1522 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1523 		< (u_int)ticks))) {
1524 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1525 		V_isn_last_reseed = ticks;
1526 	}
1527 
1528 	/* Compute the md5 hash and return the ISN. */
1529 	MD5Init(&isn_ctx);
1530 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1531 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1532 #ifdef INET6
1533 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1534 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1535 			  sizeof(struct in6_addr));
1536 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1537 			  sizeof(struct in6_addr));
1538 	} else
1539 #endif
1540 	{
1541 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1542 			  sizeof(struct in_addr));
1543 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1544 			  sizeof(struct in_addr));
1545 	}
1546 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1547 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1548 	new_isn = (tcp_seq) md5_buffer[0];
1549 	V_isn_offset += ISN_STATIC_INCREMENT +
1550 		(arc4random() & ISN_RANDOM_INCREMENT);
1551 	new_isn += V_isn_offset;
1552 	ISN_UNLOCK();
1553 	return (new_isn);
1554 }
1555 
1556 /*
1557  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1558  * to keep time flowing at a relatively constant rate.  If the random
1559  * increments have already pushed us past the projected offset, do nothing.
1560  */
1561 static void
1562 tcp_isn_tick(void *xtp)
1563 {
1564 	VNET_ITERATOR_DECL(vnet_iter);
1565 	u_int32_t projected_offset;
1566 
1567 	VNET_LIST_RLOCK();
1568 	ISN_LOCK();
1569 	VNET_FOREACH(vnet_iter) {
1570 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1571 		INIT_VNET_INET(curvnet);
1572 		projected_offset =
1573 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1574 
1575 		if (SEQ_GT(projected_offset, V_isn_offset))
1576 			V_isn_offset = projected_offset;
1577 
1578 		V_isn_offset_old = V_isn_offset;
1579 		CURVNET_RESTORE();
1580 	}
1581 	ISN_UNLOCK();
1582 	VNET_LIST_RUNLOCK();
1583 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1584 }
1585 
1586 /*
1587  * When a specific ICMP unreachable message is received and the
1588  * connection state is SYN-SENT, drop the connection.  This behavior
1589  * is controlled by the icmp_may_rst sysctl.
1590  */
1591 struct inpcb *
1592 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1593 {
1594 #ifdef INVARIANTS
1595 	INIT_VNET_INET(inp->inp_vnet);
1596 #endif
1597 	struct tcpcb *tp;
1598 
1599 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1600 	INP_WLOCK_ASSERT(inp);
1601 
1602 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1603 	    (inp->inp_flags & INP_DROPPED))
1604 		return (inp);
1605 
1606 	tp = intotcpcb(inp);
1607 	if (tp->t_state != TCPS_SYN_SENT)
1608 		return (inp);
1609 
1610 	tp = tcp_drop(tp, errno);
1611 	if (tp != NULL)
1612 		return (inp);
1613 	else
1614 		return (NULL);
1615 }
1616 
1617 /*
1618  * When `need fragmentation' ICMP is received, update our idea of the MSS
1619  * based on the new value in the route.  Also nudge TCP to send something,
1620  * since we know the packet we just sent was dropped.
1621  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1622  */
1623 struct inpcb *
1624 tcp_mtudisc(struct inpcb *inp, int errno)
1625 {
1626 	INIT_VNET_INET(inp->inp_vnet);
1627 	struct tcpcb *tp;
1628 	struct socket *so;
1629 
1630 	INP_WLOCK_ASSERT(inp);
1631 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1632 	    (inp->inp_flags & INP_DROPPED))
1633 		return (inp);
1634 
1635 	tp = intotcpcb(inp);
1636 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1637 
1638 	tcp_mss_update(tp, -1, NULL, NULL);
1639 
1640 	so = inp->inp_socket;
1641 	SOCKBUF_LOCK(&so->so_snd);
1642 	/* If the mss is larger than the socket buffer, decrease the mss. */
1643 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1644 		tp->t_maxseg = so->so_snd.sb_hiwat;
1645 	SOCKBUF_UNLOCK(&so->so_snd);
1646 
1647 	TCPSTAT_INC(tcps_mturesent);
1648 	tp->t_rtttime = 0;
1649 	tp->snd_nxt = tp->snd_una;
1650 	tcp_free_sackholes(tp);
1651 	tp->snd_recover = tp->snd_max;
1652 	if (tp->t_flags & TF_SACK_PERMIT)
1653 		EXIT_FASTRECOVERY(tp);
1654 	tcp_output_send(tp);
1655 	return (inp);
1656 }
1657 
1658 /*
1659  * Look-up the routing entry to the peer of this inpcb.  If no route
1660  * is found and it cannot be allocated, then return 0.  This routine
1661  * is called by TCP routines that access the rmx structure and by
1662  * tcp_mss_update to get the peer/interface MTU.
1663  */
1664 u_long
1665 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1666 {
1667 	struct route sro;
1668 	struct sockaddr_in *dst;
1669 	struct ifnet *ifp;
1670 	u_long maxmtu = 0;
1671 
1672 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1673 
1674 	bzero(&sro, sizeof(sro));
1675 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1676 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1677 		dst->sin_family = AF_INET;
1678 		dst->sin_len = sizeof(*dst);
1679 		dst->sin_addr = inc->inc_faddr;
1680 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1681 	}
1682 	if (sro.ro_rt != NULL) {
1683 		ifp = sro.ro_rt->rt_ifp;
1684 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1685 			maxmtu = ifp->if_mtu;
1686 		else
1687 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1688 
1689 		/* Report additional interface capabilities. */
1690 		if (flags != NULL) {
1691 			if (ifp->if_capenable & IFCAP_TSO4 &&
1692 			    ifp->if_hwassist & CSUM_TSO)
1693 				*flags |= CSUM_TSO;
1694 		}
1695 		RTFREE(sro.ro_rt);
1696 	}
1697 	return (maxmtu);
1698 }
1699 
1700 #ifdef INET6
1701 u_long
1702 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1703 {
1704 	struct route_in6 sro6;
1705 	struct ifnet *ifp;
1706 	u_long maxmtu = 0;
1707 
1708 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1709 
1710 	bzero(&sro6, sizeof(sro6));
1711 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1712 		sro6.ro_dst.sin6_family = AF_INET6;
1713 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1714 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1715 		rtalloc_ign((struct route *)&sro6, 0);
1716 	}
1717 	if (sro6.ro_rt != NULL) {
1718 		ifp = sro6.ro_rt->rt_ifp;
1719 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1720 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1721 		else
1722 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1723 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1724 
1725 		/* Report additional interface capabilities. */
1726 		if (flags != NULL) {
1727 			if (ifp->if_capenable & IFCAP_TSO6 &&
1728 			    ifp->if_hwassist & CSUM_TSO)
1729 				*flags |= CSUM_TSO;
1730 		}
1731 		RTFREE(sro6.ro_rt);
1732 	}
1733 
1734 	return (maxmtu);
1735 }
1736 #endif /* INET6 */
1737 
1738 #ifdef IPSEC
1739 /* compute ESP/AH header size for TCP, including outer IP header. */
1740 size_t
1741 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1742 {
1743 	struct inpcb *inp;
1744 	struct mbuf *m;
1745 	size_t hdrsiz;
1746 	struct ip *ip;
1747 #ifdef INET6
1748 	struct ip6_hdr *ip6;
1749 #endif
1750 	struct tcphdr *th;
1751 
1752 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1753 		return (0);
1754 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1755 	if (!m)
1756 		return (0);
1757 
1758 #ifdef INET6
1759 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1760 		ip6 = mtod(m, struct ip6_hdr *);
1761 		th = (struct tcphdr *)(ip6 + 1);
1762 		m->m_pkthdr.len = m->m_len =
1763 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1764 		tcpip_fillheaders(inp, ip6, th);
1765 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1766 	} else
1767 #endif /* INET6 */
1768 	{
1769 		ip = mtod(m, struct ip *);
1770 		th = (struct tcphdr *)(ip + 1);
1771 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1772 		tcpip_fillheaders(inp, ip, th);
1773 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1774 	}
1775 
1776 	m_free(m);
1777 	return (hdrsiz);
1778 }
1779 #endif /* IPSEC */
1780 
1781 /*
1782  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1783  *
1784  * This code attempts to calculate the bandwidth-delay product as a
1785  * means of determining the optimal window size to maximize bandwidth,
1786  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1787  * routers.  This code also does a fairly good job keeping RTTs in check
1788  * across slow links like modems.  We implement an algorithm which is very
1789  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1790  * transmitter side of a TCP connection and so only effects the transmit
1791  * side of the connection.
1792  *
1793  * BACKGROUND:  TCP makes no provision for the management of buffer space
1794  * at the end points or at the intermediate routers and switches.  A TCP
1795  * stream, whether using NewReno or not, will eventually buffer as
1796  * many packets as it is able and the only reason this typically works is
1797  * due to the fairly small default buffers made available for a connection
1798  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1799  * scaling it is now fairly easy for even a single TCP connection to blow-out
1800  * all available buffer space not only on the local interface, but on
1801  * intermediate routers and switches as well.  NewReno makes a misguided
1802  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1803  * then backing off, then steadily increasing the window again until another
1804  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1805  * is only made worse as network loads increase and the idea of intentionally
1806  * blowing out network buffers is, frankly, a terrible way to manage network
1807  * resources.
1808  *
1809  * It is far better to limit the transmit window prior to the failure
1810  * condition being achieved.  There are two general ways to do this:  First
1811  * you can 'scan' through different transmit window sizes and locate the
1812  * point where the RTT stops increasing, indicating that you have filled the
1813  * pipe, then scan backwards until you note that RTT stops decreasing, then
1814  * repeat ad-infinitum.  This method works in principle but has severe
1815  * implementation issues due to RTT variances, timer granularity, and
1816  * instability in the algorithm which can lead to many false positives and
1817  * create oscillations as well as interact badly with other TCP streams
1818  * implementing the same algorithm.
1819  *
1820  * The second method is to limit the window to the bandwidth delay product
1821  * of the link.  This is the method we implement.  RTT variances and our
1822  * own manipulation of the congestion window, bwnd, can potentially
1823  * destabilize the algorithm.  For this reason we have to stabilize the
1824  * elements used to calculate the window.  We do this by using the minimum
1825  * observed RTT, the long term average of the observed bandwidth, and
1826  * by adding two segments worth of slop.  It isn't perfect but it is able
1827  * to react to changing conditions and gives us a very stable basis on
1828  * which to extend the algorithm.
1829  */
1830 void
1831 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1832 {
1833 	INIT_VNET_INET(tp->t_vnet);
1834 	u_long bw;
1835 	u_long bwnd;
1836 	int save_ticks;
1837 
1838 	INP_WLOCK_ASSERT(tp->t_inpcb);
1839 
1840 	/*
1841 	 * If inflight_enable is disabled in the middle of a tcp connection,
1842 	 * make sure snd_bwnd is effectively disabled.
1843 	 */
1844 	if (V_tcp_inflight_enable == 0 ||
1845 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1846 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1847 		tp->snd_bandwidth = 0;
1848 		return;
1849 	}
1850 
1851 	/*
1852 	 * Figure out the bandwidth.  Due to the tick granularity this
1853 	 * is a very rough number and it MUST be averaged over a fairly
1854 	 * long period of time.  XXX we need to take into account a link
1855 	 * that is not using all available bandwidth, but for now our
1856 	 * slop will ramp us up if this case occurs and the bandwidth later
1857 	 * increases.
1858 	 *
1859 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1860 	 * effectively reset t_bw_rtttime for this case.
1861 	 */
1862 	save_ticks = ticks;
1863 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1864 		return;
1865 
1866 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1867 	    (save_ticks - tp->t_bw_rtttime);
1868 	tp->t_bw_rtttime = save_ticks;
1869 	tp->t_bw_rtseq = ack_seq;
1870 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1871 		return;
1872 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1873 
1874 	tp->snd_bandwidth = bw;
1875 
1876 	/*
1877 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1878 	 * segments.  The additional slop puts us squarely in the sweet
1879 	 * spot and also handles the bandwidth run-up case and stabilization.
1880 	 * Without the slop we could be locking ourselves into a lower
1881 	 * bandwidth.
1882 	 *
1883 	 * Situations Handled:
1884 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1885 	 *	    high speed LANs, allowing larger TCP buffers to be
1886 	 *	    specified, and also does a good job preventing
1887 	 *	    over-queueing of packets over choke points like modems
1888 	 *	    (at least for the transmit side).
1889 	 *
1890 	 *	(2) Is able to handle changing network loads (bandwidth
1891 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1892 	 *	    increases).
1893 	 *
1894 	 *	(3) Theoretically should stabilize in the face of multiple
1895 	 *	    connections implementing the same algorithm (this may need
1896 	 *	    a little work).
1897 	 *
1898 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1899 	 *	    be adjusted with a sysctl but typically only needs to be
1900 	 *	    on very slow connections.  A value no smaller then 5
1901 	 *	    should be used, but only reduce this default if you have
1902 	 *	    no other choice.
1903 	 */
1904 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1905 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1906 #undef USERTT
1907 
1908 	if (tcp_inflight_debug > 0) {
1909 		static int ltime;
1910 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1911 			ltime = ticks;
1912 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1913 			    tp,
1914 			    bw,
1915 			    tp->t_rttbest,
1916 			    tp->t_srtt,
1917 			    bwnd
1918 			);
1919 		}
1920 	}
1921 	if ((long)bwnd < V_tcp_inflight_min)
1922 		bwnd = V_tcp_inflight_min;
1923 	if (bwnd > V_tcp_inflight_max)
1924 		bwnd = V_tcp_inflight_max;
1925 	if ((long)bwnd < tp->t_maxseg * 2)
1926 		bwnd = tp->t_maxseg * 2;
1927 	tp->snd_bwnd = bwnd;
1928 }
1929 
1930 #ifdef TCP_SIGNATURE
1931 /*
1932  * Callback function invoked by m_apply() to digest TCP segment data
1933  * contained within an mbuf chain.
1934  */
1935 static int
1936 tcp_signature_apply(void *fstate, void *data, u_int len)
1937 {
1938 
1939 	MD5Update(fstate, (u_char *)data, len);
1940 	return (0);
1941 }
1942 
1943 /*
1944  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1945  *
1946  * Parameters:
1947  * m		pointer to head of mbuf chain
1948  * _unused
1949  * len		length of TCP segment data, excluding options
1950  * optlen	length of TCP segment options
1951  * buf		pointer to storage for computed MD5 digest
1952  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1953  *
1954  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1955  * When called from tcp_input(), we can be sure that th_sum has been
1956  * zeroed out and verified already.
1957  *
1958  * Return 0 if successful, otherwise return -1.
1959  *
1960  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1961  * search with the destination IP address, and a 'magic SPI' to be
1962  * determined by the application. This is hardcoded elsewhere to 1179
1963  * right now. Another branch of this code exists which uses the SPD to
1964  * specify per-application flows but it is unstable.
1965  */
1966 int
1967 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1968     u_char *buf, u_int direction)
1969 {
1970 	INIT_VNET_IPSEC(curvnet);
1971 	union sockaddr_union dst;
1972 	struct ippseudo ippseudo;
1973 	MD5_CTX ctx;
1974 	int doff;
1975 	struct ip *ip;
1976 	struct ipovly *ipovly;
1977 	struct secasvar *sav;
1978 	struct tcphdr *th;
1979 #ifdef INET6
1980 	struct ip6_hdr *ip6;
1981 	struct in6_addr in6;
1982 	char ip6buf[INET6_ADDRSTRLEN];
1983 	uint32_t plen;
1984 	uint16_t nhdr;
1985 #endif
1986 	u_short savecsum;
1987 
1988 	KASSERT(m != NULL, ("NULL mbuf chain"));
1989 	KASSERT(buf != NULL, ("NULL signature pointer"));
1990 
1991 	/* Extract the destination from the IP header in the mbuf. */
1992 	bzero(&dst, sizeof(union sockaddr_union));
1993 	ip = mtod(m, struct ip *);
1994 #ifdef INET6
1995 	ip6 = NULL;	/* Make the compiler happy. */
1996 #endif
1997 	switch (ip->ip_v) {
1998 	case IPVERSION:
1999 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2000 		dst.sa.sa_family = AF_INET;
2001 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2002 		    ip->ip_src : ip->ip_dst;
2003 		break;
2004 #ifdef INET6
2005 	case (IPV6_VERSION >> 4):
2006 		ip6 = mtod(m, struct ip6_hdr *);
2007 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2008 		dst.sa.sa_family = AF_INET6;
2009 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2010 		    ip6->ip6_src : ip6->ip6_dst;
2011 		break;
2012 #endif
2013 	default:
2014 		return (EINVAL);
2015 		/* NOTREACHED */
2016 		break;
2017 	}
2018 
2019 	/* Look up an SADB entry which matches the address of the peer. */
2020 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2021 	if (sav == NULL) {
2022 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2023 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2024 #ifdef INET6
2025 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2026 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2027 #endif
2028 			"(unsupported)"));
2029 		return (EINVAL);
2030 	}
2031 
2032 	MD5Init(&ctx);
2033 	/*
2034 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2035 	 *
2036 	 * XXX The ippseudo header MUST be digested in network byte order,
2037 	 * or else we'll fail the regression test. Assume all fields we've
2038 	 * been doing arithmetic on have been in host byte order.
2039 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2040 	 * tcp_output(), the underlying ip_len member has not yet been set.
2041 	 */
2042 	switch (ip->ip_v) {
2043 	case IPVERSION:
2044 		ipovly = (struct ipovly *)ip;
2045 		ippseudo.ippseudo_src = ipovly->ih_src;
2046 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2047 		ippseudo.ippseudo_pad = 0;
2048 		ippseudo.ippseudo_p = IPPROTO_TCP;
2049 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2050 		    optlen);
2051 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2052 
2053 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2054 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2055 		break;
2056 #ifdef INET6
2057 	/*
2058 	 * RFC 2385, 2.0  Proposal
2059 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2060 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2061 	 * extended next header value (to form 32 bits), and 32-bit segment
2062 	 * length.
2063 	 * Note: Upper-Layer Packet Length comes before Next Header.
2064 	 */
2065 	case (IPV6_VERSION >> 4):
2066 		in6 = ip6->ip6_src;
2067 		in6_clearscope(&in6);
2068 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2069 		in6 = ip6->ip6_dst;
2070 		in6_clearscope(&in6);
2071 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2072 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2073 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2074 		nhdr = 0;
2075 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2076 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2077 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2078 		nhdr = IPPROTO_TCP;
2079 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2080 
2081 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2082 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2083 		break;
2084 #endif
2085 	default:
2086 		return (EINVAL);
2087 		/* NOTREACHED */
2088 		break;
2089 	}
2090 
2091 
2092 	/*
2093 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2094 	 * The TCP checksum must be set to zero.
2095 	 */
2096 	savecsum = th->th_sum;
2097 	th->th_sum = 0;
2098 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2099 	th->th_sum = savecsum;
2100 
2101 	/*
2102 	 * Step 3: Update MD5 hash with TCP segment data.
2103 	 *         Use m_apply() to avoid an early m_pullup().
2104 	 */
2105 	if (len > 0)
2106 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2107 
2108 	/*
2109 	 * Step 4: Update MD5 hash with shared secret.
2110 	 */
2111 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2112 	MD5Final(buf, &ctx);
2113 
2114 	key_sa_recordxfer(sav, m);
2115 	KEY_FREESAV(&sav);
2116 	return (0);
2117 }
2118 #endif /* TCP_SIGNATURE */
2119 
2120 static int
2121 sysctl_drop(SYSCTL_HANDLER_ARGS)
2122 {
2123 	INIT_VNET_INET(curvnet);
2124 #ifdef INET6
2125 	INIT_VNET_INET6(curvnet);
2126 #endif
2127 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2128 	struct sockaddr_storage addrs[2];
2129 	struct inpcb *inp;
2130 	struct tcpcb *tp;
2131 	struct tcptw *tw;
2132 	struct sockaddr_in *fin, *lin;
2133 #ifdef INET6
2134 	struct sockaddr_in6 *fin6, *lin6;
2135 #endif
2136 	int error;
2137 
2138 	inp = NULL;
2139 	fin = lin = NULL;
2140 #ifdef INET6
2141 	fin6 = lin6 = NULL;
2142 #endif
2143 	error = 0;
2144 
2145 	if (req->oldptr != NULL || req->oldlen != 0)
2146 		return (EINVAL);
2147 	if (req->newptr == NULL)
2148 		return (EPERM);
2149 	if (req->newlen < sizeof(addrs))
2150 		return (ENOMEM);
2151 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2152 	if (error)
2153 		return (error);
2154 
2155 	switch (addrs[0].ss_family) {
2156 #ifdef INET6
2157 	case AF_INET6:
2158 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2159 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2160 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2161 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2162 			return (EINVAL);
2163 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2164 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2165 				return (EINVAL);
2166 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2167 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2168 			fin = (struct sockaddr_in *)&addrs[0];
2169 			lin = (struct sockaddr_in *)&addrs[1];
2170 			break;
2171 		}
2172 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2173 		if (error)
2174 			return (error);
2175 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2176 		if (error)
2177 			return (error);
2178 		break;
2179 #endif
2180 	case AF_INET:
2181 		fin = (struct sockaddr_in *)&addrs[0];
2182 		lin = (struct sockaddr_in *)&addrs[1];
2183 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2184 		    lin->sin_len != sizeof(struct sockaddr_in))
2185 			return (EINVAL);
2186 		break;
2187 	default:
2188 		return (EINVAL);
2189 	}
2190 	INP_INFO_WLOCK(&V_tcbinfo);
2191 	switch (addrs[0].ss_family) {
2192 #ifdef INET6
2193 	case AF_INET6:
2194 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2195 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2196 		    NULL);
2197 		break;
2198 #endif
2199 	case AF_INET:
2200 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2201 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2202 		break;
2203 	}
2204 	if (inp != NULL) {
2205 		INP_WLOCK(inp);
2206 		if (inp->inp_flags & INP_TIMEWAIT) {
2207 			/*
2208 			 * XXXRW: There currently exists a state where an
2209 			 * inpcb is present, but its timewait state has been
2210 			 * discarded.  For now, don't allow dropping of this
2211 			 * type of inpcb.
2212 			 */
2213 			tw = intotw(inp);
2214 			if (tw != NULL)
2215 				tcp_twclose(tw, 0);
2216 			else
2217 				INP_WUNLOCK(inp);
2218 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2219 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2220 			tp = intotcpcb(inp);
2221 			tp = tcp_drop(tp, ECONNABORTED);
2222 			if (tp != NULL)
2223 				INP_WUNLOCK(inp);
2224 		} else
2225 			INP_WUNLOCK(inp);
2226 	} else
2227 		error = ESRCH;
2228 	INP_INFO_WUNLOCK(&V_tcbinfo);
2229 	return (error);
2230 }
2231 
2232 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2233     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2234     0, sysctl_drop, "", "Drop TCP connection");
2235 
2236 /*
2237  * Generate a standardized TCP log line for use throughout the
2238  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2239  * allow use in the interrupt context.
2240  *
2241  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2242  * NB: The function may return NULL if memory allocation failed.
2243  *
2244  * Due to header inclusion and ordering limitations the struct ip
2245  * and ip6_hdr pointers have to be passed as void pointers.
2246  */
2247 char *
2248 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2249     const void *ip6hdr)
2250 {
2251 	char *s, *sp;
2252 	size_t size;
2253 	struct ip *ip;
2254 #ifdef INET6
2255 	const struct ip6_hdr *ip6;
2256 
2257 	ip6 = (const struct ip6_hdr *)ip6hdr;
2258 #endif /* INET6 */
2259 	ip = (struct ip *)ip4hdr;
2260 
2261 	/*
2262 	 * The log line looks like this:
2263 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2264 	 */
2265 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2266 	    sizeof(PRINT_TH_FLAGS) + 1 +
2267 #ifdef INET6
2268 	    2 * INET6_ADDRSTRLEN;
2269 #else
2270 	    2 * INET_ADDRSTRLEN;
2271 #endif /* INET6 */
2272 
2273 	/* Is logging enabled? */
2274 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2275 		return (NULL);
2276 
2277 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2278 	if (s == NULL)
2279 		return (NULL);
2280 
2281 	strcat(s, "TCP: [");
2282 	sp = s + strlen(s);
2283 
2284 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2285 		inet_ntoa_r(inc->inc_faddr, sp);
2286 		sp = s + strlen(s);
2287 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2288 		sp = s + strlen(s);
2289 		inet_ntoa_r(inc->inc_laddr, sp);
2290 		sp = s + strlen(s);
2291 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2292 #ifdef INET6
2293 	} else if (inc) {
2294 		ip6_sprintf(sp, &inc->inc6_faddr);
2295 		sp = s + strlen(s);
2296 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2297 		sp = s + strlen(s);
2298 		ip6_sprintf(sp, &inc->inc6_laddr);
2299 		sp = s + strlen(s);
2300 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2301 	} else if (ip6 && th) {
2302 		ip6_sprintf(sp, &ip6->ip6_src);
2303 		sp = s + strlen(s);
2304 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2305 		sp = s + strlen(s);
2306 		ip6_sprintf(sp, &ip6->ip6_dst);
2307 		sp = s + strlen(s);
2308 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2309 #endif /* INET6 */
2310 	} else if (ip && th) {
2311 		inet_ntoa_r(ip->ip_src, sp);
2312 		sp = s + strlen(s);
2313 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2314 		sp = s + strlen(s);
2315 		inet_ntoa_r(ip->ip_dst, sp);
2316 		sp = s + strlen(s);
2317 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2318 	} else {
2319 		free(s, M_TCPLOG);
2320 		return (NULL);
2321 	}
2322 	sp = s + strlen(s);
2323 	if (th)
2324 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2325 	if (*(s + size - 1) != '\0')
2326 		panic("%s: string too long", __func__);
2327 	return (s);
2328 }
2329