xref: /freebsd/sys/netinet/tcp_subr.c (revision 830940567b49bb0c08dfaed40418999e76616909)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/jail.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/route.h>
62 #include <net/if.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet6/ip6protosw.h>
98 
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
102 #ifdef INET6
103 #include <netipsec/ipsec6.h>
104 #endif
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
107 #endif /*IPSEC*/
108 
109 #include <machine/in_cksum.h>
110 #include <sys/md5.h>
111 
112 #include <security/mac/mac_framework.h>
113 
114 VNET_DEFINE(int, tcp_mssdflt);
115 #ifdef INET6
116 VNET_DEFINE(int, tcp_v6mssdflt);
117 #endif
118 VNET_DEFINE(int, tcp_minmss);
119 VNET_DEFINE(int, tcp_do_rfc1323);
120 
121 static VNET_DEFINE(int, icmp_may_rst);
122 static VNET_DEFINE(int, tcp_isn_reseed_interval);
123 static VNET_DEFINE(int, tcp_inflight_enable);
124 static VNET_DEFINE(int, tcp_inflight_rttthresh);
125 static VNET_DEFINE(int, tcp_inflight_min);
126 static VNET_DEFINE(int, tcp_inflight_max);
127 static VNET_DEFINE(int, tcp_inflight_stab);
128 
129 #define	V_icmp_may_rst			VNET(icmp_may_rst)
130 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
131 #define	V_tcp_inflight_enable		VNET(tcp_inflight_enable)
132 #define	V_tcp_inflight_rttthresh	VNET(tcp_inflight_rttthresh)
133 #define	V_tcp_inflight_min		VNET(tcp_inflight_min)
134 #define	V_tcp_inflight_max		VNET(tcp_inflight_max)
135 #define	V_tcp_inflight_stab		VNET(tcp_inflight_stab)
136 
137 static int
138 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
139 {
140 	int error, new;
141 
142 	new = V_tcp_mssdflt;
143 	error = sysctl_handle_int(oidp, &new, 0, req);
144 	if (error == 0 && req->newptr) {
145 		if (new < TCP_MINMSS)
146 			error = EINVAL;
147 		else
148 			V_tcp_mssdflt = new;
149 	}
150 	return (error);
151 }
152 
153 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
154     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
155     &sysctl_net_inet_tcp_mss_check, "I",
156     "Default TCP Maximum Segment Size");
157 
158 #ifdef INET6
159 static int
160 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
161 {
162 	int error, new;
163 
164 	new = V_tcp_v6mssdflt;
165 	error = sysctl_handle_int(oidp, &new, 0, req);
166 	if (error == 0 && req->newptr) {
167 		if (new < TCP_MINMSS)
168 			error = EINVAL;
169 		else
170 			V_tcp_v6mssdflt = new;
171 	}
172 	return (error);
173 }
174 
175 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
176     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
177     &sysctl_net_inet_tcp_mss_v6_check, "I",
178    "Default TCP Maximum Segment Size for IPv6");
179 #endif
180 
181 static int
182 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
183 {
184 
185 	VNET_SYSCTL_ARG(req, arg1);
186 	return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
187 }
188 
189 /*
190  * Minimum MSS we accept and use. This prevents DoS attacks where
191  * we are forced to a ridiculous low MSS like 20 and send hundreds
192  * of packets instead of one. The effect scales with the available
193  * bandwidth and quickly saturates the CPU and network interface
194  * with packet generation and sending. Set to zero to disable MINMSS
195  * checking. This setting prevents us from sending too small packets.
196  */
197 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
198      &VNET_NAME(tcp_minmss), 0,
199     "Minmum TCP Maximum Segment Size");
200 
201 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
202     &VNET_NAME(tcp_do_rfc1323), 0,
203     "Enable rfc1323 (high performance TCP) extensions");
204 
205 static int	tcp_log_debug = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
207     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
208 
209 static int	tcp_tcbhashsize = 0;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
211     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
212 
213 static int	do_tcpdrain = 1;
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
215     "Enable tcp_drain routine for extra help when low on mbufs");
216 
217 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
218     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
219 
220 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
221     &VNET_NAME(icmp_may_rst), 0,
222     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
223 
224 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
225     &VNET_NAME(tcp_isn_reseed_interval), 0,
226     "Seconds between reseeding of ISN secret");
227 
228 /*
229  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
230  * 1024 exists only for debugging.  A good production default would be
231  * something like 6100.
232  */
233 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
234     "TCP inflight data limiting");
235 
236 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
237     &VNET_NAME(tcp_inflight_enable), 0,
238     "Enable automatic TCP inflight data limiting");
239 
240 static int	tcp_inflight_debug = 0;
241 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
242     &tcp_inflight_debug, 0,
243     "Debug TCP inflight calculations");
244 
245 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
246     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
247     vnet_sysctl_msec_to_ticks, "I",
248     "RTT threshold below which inflight will deactivate itself");
249 
250 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
251     &VNET_NAME(tcp_inflight_min), 0,
252     "Lower-bound for TCP inflight window");
253 
254 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
255     &VNET_NAME(tcp_inflight_max), 0,
256     "Upper-bound for TCP inflight window");
257 
258 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
259     &VNET_NAME(tcp_inflight_stab), 0,
260     "Inflight Algorithm Stabilization 20 = 2 packets");
261 
262 VNET_DEFINE(uma_zone_t, sack_hole_zone);
263 #define	V_sack_hole_zone		VNET(sack_hole_zone)
264 
265 static struct inpcb *tcp_notify(struct inpcb *, int);
266 static void	tcp_isn_tick(void *);
267 
268 /*
269  * Target size of TCP PCB hash tables. Must be a power of two.
270  *
271  * Note that this can be overridden by the kernel environment
272  * variable net.inet.tcp.tcbhashsize
273  */
274 #ifndef TCBHASHSIZE
275 #define TCBHASHSIZE	512
276 #endif
277 
278 /*
279  * XXX
280  * Callouts should be moved into struct tcp directly.  They are currently
281  * separate because the tcpcb structure is exported to userland for sysctl
282  * parsing purposes, which do not know about callouts.
283  */
284 struct tcpcb_mem {
285 	struct	tcpcb		tcb;
286 	struct	tcp_timer	tt;
287 };
288 
289 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
290 #define	V_tcpcb_zone			VNET(tcpcb_zone)
291 
292 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
293 struct callout isn_callout;
294 static struct mtx isn_mtx;
295 
296 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
297 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
298 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
299 
300 /*
301  * TCP initialization.
302  */
303 static void
304 tcp_zone_change(void *tag)
305 {
306 
307 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
308 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
309 	tcp_tw_zone_change();
310 }
311 
312 static int
313 tcp_inpcb_init(void *mem, int size, int flags)
314 {
315 	struct inpcb *inp = mem;
316 
317 	INP_LOCK_INIT(inp, "inp", "tcpinp");
318 	return (0);
319 }
320 
321 void
322 tcp_init(void)
323 {
324 	int hashsize;
325 
326 	V_blackhole = 0;
327 	V_tcp_delack_enabled = 1;
328 	V_drop_synfin = 0;
329 	V_tcp_do_rfc3042 = 1;
330 	V_tcp_do_rfc3390 = 1;
331 	V_tcp_do_ecn = 0;
332 	V_tcp_ecn_maxretries = 1;
333 	V_tcp_insecure_rst = 0;
334 	V_tcp_do_autorcvbuf = 1;
335 	V_tcp_autorcvbuf_inc = 16*1024;
336 	V_tcp_autorcvbuf_max = 256*1024;
337 	V_tcp_do_rfc3465 = 1;
338 	V_tcp_abc_l_var = 2;
339 
340 	V_tcp_mssdflt = TCP_MSS;
341 #ifdef INET6
342 	V_tcp_v6mssdflt = TCP6_MSS;
343 #endif
344 	V_tcp_minmss = TCP_MINMSS;
345 	V_tcp_do_rfc1323 = 1;
346 	V_icmp_may_rst = 1;
347 	V_tcp_isn_reseed_interval = 0;
348 	V_tcp_inflight_enable = 1;
349 	V_tcp_inflight_min = 6144;
350 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
351 	V_tcp_inflight_stab = 20;
352 
353 	V_path_mtu_discovery = 1;
354 	V_ss_fltsz = 1;
355 	V_ss_fltsz_local = 4;
356 	V_tcp_do_newreno = 1;
357 	V_tcp_do_tso = 1;
358 	V_tcp_do_autosndbuf = 1;
359 	V_tcp_autosndbuf_inc = 8*1024;
360 	V_tcp_autosndbuf_max = 256*1024;
361 
362 	V_nolocaltimewait = 0;
363 
364 	V_tcp_do_sack = 1;
365 	V_tcp_sack_maxholes = 128;
366 	V_tcp_sack_globalmaxholes = 65536;
367 	V_tcp_sack_globalholes = 0;
368 
369 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
370 
371 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
372 
373 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
374 	LIST_INIT(&V_tcb);
375 #ifdef VIMAGE
376 	V_tcbinfo.ipi_vnet = curvnet;
377 #endif
378 	V_tcbinfo.ipi_listhead = &V_tcb;
379 	hashsize = TCBHASHSIZE;
380 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
381 	if (!powerof2(hashsize)) {
382 		printf("WARNING: TCB hash size not a power of 2\n");
383 		hashsize = 512; /* safe default */
384 	}
385 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
386 	    &V_tcbinfo.ipi_hashmask);
387 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
388 	    &V_tcbinfo.ipi_porthashmask);
389 	V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
390 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
391 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
392 	/*
393 	 * These have to be type stable for the benefit of the timers.
394 	 */
395 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
396 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
397 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
398 	tcp_tw_init();
399 	syncache_init();
400 	tcp_hc_init();
401 	tcp_reass_init();
402 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
403 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
404 
405 	/* Skip initialization of globals for non-default instances. */
406 	if (!IS_DEFAULT_VNET(curvnet))
407 		return;
408 
409 	/* XXX virtualize those bellow? */
410 	tcp_delacktime = TCPTV_DELACK;
411 	tcp_keepinit = TCPTV_KEEP_INIT;
412 	tcp_keepidle = TCPTV_KEEP_IDLE;
413 	tcp_keepintvl = TCPTV_KEEPINTVL;
414 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
415 	tcp_msl = TCPTV_MSL;
416 	tcp_rexmit_min = TCPTV_MIN;
417 	if (tcp_rexmit_min < 1)
418 		tcp_rexmit_min = 1;
419 	tcp_rexmit_slop = TCPTV_CPU_VAR;
420 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
421 	tcp_tcbhashsize = hashsize;
422 
423 #ifdef INET6
424 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
425 #else /* INET6 */
426 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
427 #endif /* INET6 */
428 	if (max_protohdr < TCP_MINPROTOHDR)
429 		max_protohdr = TCP_MINPROTOHDR;
430 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
431 		panic("tcp_init");
432 #undef TCP_MINPROTOHDR
433 
434 	ISN_LOCK_INIT();
435 	callout_init(&isn_callout, CALLOUT_MPSAFE);
436 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
437 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
438 		SHUTDOWN_PRI_DEFAULT);
439 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
440 		EVENTHANDLER_PRI_ANY);
441 }
442 
443 #ifdef VIMAGE
444 void
445 tcp_destroy(void)
446 {
447 
448 	tcp_tw_destroy();
449 	tcp_hc_destroy();
450 	syncache_destroy();
451 
452 	/* XXX check that hashes are empty! */
453 	hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
454 	    V_tcbinfo.ipi_hashmask);
455 	hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
456 	    V_tcbinfo.ipi_porthashmask);
457 	INP_INFO_LOCK_DESTROY(&V_tcbinfo);
458 }
459 #endif
460 
461 void
462 tcp_fini(void *xtp)
463 {
464 
465 	callout_stop(&isn_callout);
466 }
467 
468 /*
469  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
470  * tcp_template used to store this data in mbufs, but we now recopy it out
471  * of the tcpcb each time to conserve mbufs.
472  */
473 void
474 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
475 {
476 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
477 
478 	INP_WLOCK_ASSERT(inp);
479 
480 #ifdef INET6
481 	if ((inp->inp_vflag & INP_IPV6) != 0) {
482 		struct ip6_hdr *ip6;
483 
484 		ip6 = (struct ip6_hdr *)ip_ptr;
485 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
486 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
487 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
488 			(IPV6_VERSION & IPV6_VERSION_MASK);
489 		ip6->ip6_nxt = IPPROTO_TCP;
490 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
491 		ip6->ip6_src = inp->in6p_laddr;
492 		ip6->ip6_dst = inp->in6p_faddr;
493 	} else
494 #endif
495 	{
496 		struct ip *ip;
497 
498 		ip = (struct ip *)ip_ptr;
499 		ip->ip_v = IPVERSION;
500 		ip->ip_hl = 5;
501 		ip->ip_tos = inp->inp_ip_tos;
502 		ip->ip_len = 0;
503 		ip->ip_id = 0;
504 		ip->ip_off = 0;
505 		ip->ip_ttl = inp->inp_ip_ttl;
506 		ip->ip_sum = 0;
507 		ip->ip_p = IPPROTO_TCP;
508 		ip->ip_src = inp->inp_laddr;
509 		ip->ip_dst = inp->inp_faddr;
510 	}
511 	th->th_sport = inp->inp_lport;
512 	th->th_dport = inp->inp_fport;
513 	th->th_seq = 0;
514 	th->th_ack = 0;
515 	th->th_x2 = 0;
516 	th->th_off = 5;
517 	th->th_flags = 0;
518 	th->th_win = 0;
519 	th->th_urp = 0;
520 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
521 }
522 
523 /*
524  * Create template to be used to send tcp packets on a connection.
525  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
526  * use for this function is in keepalives, which use tcp_respond.
527  */
528 struct tcptemp *
529 tcpip_maketemplate(struct inpcb *inp)
530 {
531 	struct tcptemp *t;
532 
533 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
534 	if (t == NULL)
535 		return (NULL);
536 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
537 	return (t);
538 }
539 
540 /*
541  * Send a single message to the TCP at address specified by
542  * the given TCP/IP header.  If m == NULL, then we make a copy
543  * of the tcpiphdr at ti and send directly to the addressed host.
544  * This is used to force keep alive messages out using the TCP
545  * template for a connection.  If flags are given then we send
546  * a message back to the TCP which originated the * segment ti,
547  * and discard the mbuf containing it and any other attached mbufs.
548  *
549  * In any case the ack and sequence number of the transmitted
550  * segment are as specified by the parameters.
551  *
552  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
553  */
554 void
555 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
556     tcp_seq ack, tcp_seq seq, int flags)
557 {
558 	int tlen;
559 	int win = 0;
560 	struct ip *ip;
561 	struct tcphdr *nth;
562 #ifdef INET6
563 	struct ip6_hdr *ip6;
564 	int isipv6;
565 #endif /* INET6 */
566 	int ipflags = 0;
567 	struct inpcb *inp;
568 
569 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
570 
571 #ifdef INET6
572 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
573 	ip6 = ipgen;
574 #endif /* INET6 */
575 	ip = ipgen;
576 
577 	if (tp != NULL) {
578 		inp = tp->t_inpcb;
579 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
580 		INP_WLOCK_ASSERT(inp);
581 	} else
582 		inp = NULL;
583 
584 	if (tp != NULL) {
585 		if (!(flags & TH_RST)) {
586 			win = sbspace(&inp->inp_socket->so_rcv);
587 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
588 				win = (long)TCP_MAXWIN << tp->rcv_scale;
589 		}
590 	}
591 	if (m == NULL) {
592 		m = m_gethdr(M_DONTWAIT, MT_DATA);
593 		if (m == NULL)
594 			return;
595 		tlen = 0;
596 		m->m_data += max_linkhdr;
597 #ifdef INET6
598 		if (isipv6) {
599 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
600 			      sizeof(struct ip6_hdr));
601 			ip6 = mtod(m, struct ip6_hdr *);
602 			nth = (struct tcphdr *)(ip6 + 1);
603 		} else
604 #endif /* INET6 */
605 	      {
606 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
607 		ip = mtod(m, struct ip *);
608 		nth = (struct tcphdr *)(ip + 1);
609 	      }
610 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
611 		flags = TH_ACK;
612 	} else {
613 		/*
614 		 *  reuse the mbuf.
615 		 * XXX MRT We inherrit the FIB, which is lucky.
616 		 */
617 		m_freem(m->m_next);
618 		m->m_next = NULL;
619 		m->m_data = (caddr_t)ipgen;
620 		/* m_len is set later */
621 		tlen = 0;
622 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
623 #ifdef INET6
624 		if (isipv6) {
625 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
626 			nth = (struct tcphdr *)(ip6 + 1);
627 		} else
628 #endif /* INET6 */
629 	      {
630 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
631 		nth = (struct tcphdr *)(ip + 1);
632 	      }
633 		if (th != nth) {
634 			/*
635 			 * this is usually a case when an extension header
636 			 * exists between the IPv6 header and the
637 			 * TCP header.
638 			 */
639 			nth->th_sport = th->th_sport;
640 			nth->th_dport = th->th_dport;
641 		}
642 		xchg(nth->th_dport, nth->th_sport, uint16_t);
643 #undef xchg
644 	}
645 #ifdef INET6
646 	if (isipv6) {
647 		ip6->ip6_flow = 0;
648 		ip6->ip6_vfc = IPV6_VERSION;
649 		ip6->ip6_nxt = IPPROTO_TCP;
650 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
651 						tlen));
652 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
653 	} else
654 #endif
655 	{
656 		tlen += sizeof (struct tcpiphdr);
657 		ip->ip_len = tlen;
658 		ip->ip_ttl = V_ip_defttl;
659 		if (V_path_mtu_discovery)
660 			ip->ip_off |= IP_DF;
661 	}
662 	m->m_len = tlen;
663 	m->m_pkthdr.len = tlen;
664 	m->m_pkthdr.rcvif = NULL;
665 #ifdef MAC
666 	if (inp != NULL) {
667 		/*
668 		 * Packet is associated with a socket, so allow the
669 		 * label of the response to reflect the socket label.
670 		 */
671 		INP_WLOCK_ASSERT(inp);
672 		mac_inpcb_create_mbuf(inp, m);
673 	} else {
674 		/*
675 		 * Packet is not associated with a socket, so possibly
676 		 * update the label in place.
677 		 */
678 		mac_netinet_tcp_reply(m);
679 	}
680 #endif
681 	nth->th_seq = htonl(seq);
682 	nth->th_ack = htonl(ack);
683 	nth->th_x2 = 0;
684 	nth->th_off = sizeof (struct tcphdr) >> 2;
685 	nth->th_flags = flags;
686 	if (tp != NULL)
687 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
688 	else
689 		nth->th_win = htons((u_short)win);
690 	nth->th_urp = 0;
691 #ifdef INET6
692 	if (isipv6) {
693 		nth->th_sum = 0;
694 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
695 					sizeof(struct ip6_hdr),
696 					tlen - sizeof(struct ip6_hdr));
697 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
698 		    NULL, NULL);
699 	} else
700 #endif /* INET6 */
701 	{
702 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
703 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
704 		m->m_pkthdr.csum_flags = CSUM_TCP;
705 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
706 	}
707 #ifdef TCPDEBUG
708 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
709 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
710 #endif
711 #ifdef INET6
712 	if (isipv6)
713 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
714 	else
715 #endif /* INET6 */
716 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
717 }
718 
719 /*
720  * Create a new TCP control block, making an
721  * empty reassembly queue and hooking it to the argument
722  * protocol control block.  The `inp' parameter must have
723  * come from the zone allocator set up in tcp_init().
724  */
725 struct tcpcb *
726 tcp_newtcpcb(struct inpcb *inp)
727 {
728 	struct tcpcb_mem *tm;
729 	struct tcpcb *tp;
730 #ifdef INET6
731 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
732 #endif /* INET6 */
733 
734 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
735 	if (tm == NULL)
736 		return (NULL);
737 	tp = &tm->tcb;
738 #ifdef VIMAGE
739 	tp->t_vnet = inp->inp_vnet;
740 #endif
741 	tp->t_timers = &tm->tt;
742 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
743 	tp->t_maxseg = tp->t_maxopd =
744 #ifdef INET6
745 		isipv6 ? V_tcp_v6mssdflt :
746 #endif /* INET6 */
747 		V_tcp_mssdflt;
748 
749 	/* Set up our timeouts. */
750 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
751 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
752 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
753 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
754 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
755 
756 	if (V_tcp_do_rfc1323)
757 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
758 	if (V_tcp_do_sack)
759 		tp->t_flags |= TF_SACK_PERMIT;
760 	TAILQ_INIT(&tp->snd_holes);
761 	tp->t_inpcb = inp;	/* XXX */
762 	/*
763 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
764 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
765 	 * reasonable initial retransmit time.
766 	 */
767 	tp->t_srtt = TCPTV_SRTTBASE;
768 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
769 	tp->t_rttmin = tcp_rexmit_min;
770 	tp->t_rxtcur = TCPTV_RTOBASE;
771 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
772 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
773 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
774 	tp->t_rcvtime = ticks;
775 	tp->t_bw_rtttime = ticks;
776 	/*
777 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
778 	 * because the socket may be bound to an IPv6 wildcard address,
779 	 * which may match an IPv4-mapped IPv6 address.
780 	 */
781 	inp->inp_ip_ttl = V_ip_defttl;
782 	inp->inp_ppcb = tp;
783 	return (tp);		/* XXX */
784 }
785 
786 /*
787  * Drop a TCP connection, reporting
788  * the specified error.  If connection is synchronized,
789  * then send a RST to peer.
790  */
791 struct tcpcb *
792 tcp_drop(struct tcpcb *tp, int errno)
793 {
794 	struct socket *so = tp->t_inpcb->inp_socket;
795 
796 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
797 	INP_WLOCK_ASSERT(tp->t_inpcb);
798 
799 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
800 		tp->t_state = TCPS_CLOSED;
801 		(void) tcp_output_reset(tp);
802 		TCPSTAT_INC(tcps_drops);
803 	} else
804 		TCPSTAT_INC(tcps_conndrops);
805 	if (errno == ETIMEDOUT && tp->t_softerror)
806 		errno = tp->t_softerror;
807 	so->so_error = errno;
808 	return (tcp_close(tp));
809 }
810 
811 void
812 tcp_discardcb(struct tcpcb *tp)
813 {
814 	struct tseg_qent *q;
815 	struct inpcb *inp = tp->t_inpcb;
816 	struct socket *so = inp->inp_socket;
817 #ifdef INET6
818 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
819 #endif /* INET6 */
820 
821 	INP_WLOCK_ASSERT(inp);
822 
823 	/*
824 	 * Make sure that all of our timers are stopped before we
825 	 * delete the PCB.
826 	 */
827 	callout_stop(&tp->t_timers->tt_rexmt);
828 	callout_stop(&tp->t_timers->tt_persist);
829 	callout_stop(&tp->t_timers->tt_keep);
830 	callout_stop(&tp->t_timers->tt_2msl);
831 	callout_stop(&tp->t_timers->tt_delack);
832 
833 	/*
834 	 * If we got enough samples through the srtt filter,
835 	 * save the rtt and rttvar in the routing entry.
836 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
837 	 * 4 samples is enough for the srtt filter to converge
838 	 * to within enough % of the correct value; fewer samples
839 	 * and we could save a bogus rtt. The danger is not high
840 	 * as tcp quickly recovers from everything.
841 	 * XXX: Works very well but needs some more statistics!
842 	 */
843 	if (tp->t_rttupdated >= 4) {
844 		struct hc_metrics_lite metrics;
845 		u_long ssthresh;
846 
847 		bzero(&metrics, sizeof(metrics));
848 		/*
849 		 * Update the ssthresh always when the conditions below
850 		 * are satisfied. This gives us better new start value
851 		 * for the congestion avoidance for new connections.
852 		 * ssthresh is only set if packet loss occured on a session.
853 		 *
854 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
855 		 * being torn down.  Ideally this code would not use 'so'.
856 		 */
857 		ssthresh = tp->snd_ssthresh;
858 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
859 			/*
860 			 * convert the limit from user data bytes to
861 			 * packets then to packet data bytes.
862 			 */
863 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
864 			if (ssthresh < 2)
865 				ssthresh = 2;
866 			ssthresh *= (u_long)(tp->t_maxseg +
867 #ifdef INET6
868 				      (isipv6 ? sizeof (struct ip6_hdr) +
869 					       sizeof (struct tcphdr) :
870 #endif
871 				       sizeof (struct tcpiphdr)
872 #ifdef INET6
873 				       )
874 #endif
875 				      );
876 		} else
877 			ssthresh = 0;
878 		metrics.rmx_ssthresh = ssthresh;
879 
880 		metrics.rmx_rtt = tp->t_srtt;
881 		metrics.rmx_rttvar = tp->t_rttvar;
882 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
883 		metrics.rmx_bandwidth = tp->snd_bandwidth;
884 		metrics.rmx_cwnd = tp->snd_cwnd;
885 		metrics.rmx_sendpipe = 0;
886 		metrics.rmx_recvpipe = 0;
887 
888 		tcp_hc_update(&inp->inp_inc, &metrics);
889 	}
890 
891 	/* free the reassembly queue, if any */
892 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
893 		LIST_REMOVE(q, tqe_q);
894 		m_freem(q->tqe_m);
895 		uma_zfree(V_tcp_reass_zone, q);
896 		tp->t_segqlen--;
897 		V_tcp_reass_qsize--;
898 	}
899 	/* Disconnect offload device, if any. */
900 	tcp_offload_detach(tp);
901 
902 	tcp_free_sackholes(tp);
903 	inp->inp_ppcb = NULL;
904 	tp->t_inpcb = NULL;
905 	uma_zfree(V_tcpcb_zone, tp);
906 }
907 
908 /*
909  * Attempt to close a TCP control block, marking it as dropped, and freeing
910  * the socket if we hold the only reference.
911  */
912 struct tcpcb *
913 tcp_close(struct tcpcb *tp)
914 {
915 	struct inpcb *inp = tp->t_inpcb;
916 	struct socket *so;
917 
918 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
919 	INP_WLOCK_ASSERT(inp);
920 
921 	/* Notify any offload devices of listener close */
922 	if (tp->t_state == TCPS_LISTEN)
923 		tcp_offload_listen_close(tp);
924 	in_pcbdrop(inp);
925 	TCPSTAT_INC(tcps_closed);
926 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
927 	so = inp->inp_socket;
928 	soisdisconnected(so);
929 	if (inp->inp_flags & INP_SOCKREF) {
930 		KASSERT(so->so_state & SS_PROTOREF,
931 		    ("tcp_close: !SS_PROTOREF"));
932 		inp->inp_flags &= ~INP_SOCKREF;
933 		INP_WUNLOCK(inp);
934 		ACCEPT_LOCK();
935 		SOCK_LOCK(so);
936 		so->so_state &= ~SS_PROTOREF;
937 		sofree(so);
938 		return (NULL);
939 	}
940 	return (tp);
941 }
942 
943 void
944 tcp_drain(void)
945 {
946 	VNET_ITERATOR_DECL(vnet_iter);
947 
948 	if (!do_tcpdrain)
949 		return;
950 
951 	VNET_LIST_RLOCK_NOSLEEP();
952 	VNET_FOREACH(vnet_iter) {
953 		CURVNET_SET(vnet_iter);
954 		struct inpcb *inpb;
955 		struct tcpcb *tcpb;
956 		struct tseg_qent *te;
957 
958 	/*
959 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
960 	 * if there is one...
961 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
962 	 *      reassembly queue should be flushed, but in a situation
963 	 *	where we're really low on mbufs, this is potentially
964 	 *	usefull.
965 	 */
966 		INP_INFO_RLOCK(&V_tcbinfo);
967 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
968 			if (inpb->inp_flags & INP_TIMEWAIT)
969 				continue;
970 			INP_WLOCK(inpb);
971 			if ((tcpb = intotcpcb(inpb)) != NULL) {
972 				while ((te = LIST_FIRST(&tcpb->t_segq))
973 			            != NULL) {
974 					LIST_REMOVE(te, tqe_q);
975 					m_freem(te->tqe_m);
976 					uma_zfree(V_tcp_reass_zone, te);
977 					tcpb->t_segqlen--;
978 					V_tcp_reass_qsize--;
979 				}
980 				tcp_clean_sackreport(tcpb);
981 			}
982 			INP_WUNLOCK(inpb);
983 		}
984 		INP_INFO_RUNLOCK(&V_tcbinfo);
985 		CURVNET_RESTORE();
986 	}
987 	VNET_LIST_RUNLOCK_NOSLEEP();
988 }
989 
990 /*
991  * Notify a tcp user of an asynchronous error;
992  * store error as soft error, but wake up user
993  * (for now, won't do anything until can select for soft error).
994  *
995  * Do not wake up user since there currently is no mechanism for
996  * reporting soft errors (yet - a kqueue filter may be added).
997  */
998 static struct inpcb *
999 tcp_notify(struct inpcb *inp, int error)
1000 {
1001 	struct tcpcb *tp;
1002 
1003 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1004 	INP_WLOCK_ASSERT(inp);
1005 
1006 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1007 	    (inp->inp_flags & INP_DROPPED))
1008 		return (inp);
1009 
1010 	tp = intotcpcb(inp);
1011 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1012 
1013 	/*
1014 	 * Ignore some errors if we are hooked up.
1015 	 * If connection hasn't completed, has retransmitted several times,
1016 	 * and receives a second error, give up now.  This is better
1017 	 * than waiting a long time to establish a connection that
1018 	 * can never complete.
1019 	 */
1020 	if (tp->t_state == TCPS_ESTABLISHED &&
1021 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1022 	     error == EHOSTDOWN)) {
1023 		return (inp);
1024 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1025 	    tp->t_softerror) {
1026 		tp = tcp_drop(tp, error);
1027 		if (tp != NULL)
1028 			return (inp);
1029 		else
1030 			return (NULL);
1031 	} else {
1032 		tp->t_softerror = error;
1033 		return (inp);
1034 	}
1035 #if 0
1036 	wakeup( &so->so_timeo);
1037 	sorwakeup(so);
1038 	sowwakeup(so);
1039 #endif
1040 }
1041 
1042 static int
1043 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1044 {
1045 	int error, i, m, n, pcb_count;
1046 	struct inpcb *inp, **inp_list;
1047 	inp_gen_t gencnt;
1048 	struct xinpgen xig;
1049 
1050 	/*
1051 	 * The process of preparing the TCB list is too time-consuming and
1052 	 * resource-intensive to repeat twice on every request.
1053 	 */
1054 	if (req->oldptr == NULL) {
1055 		m = syncache_pcbcount();
1056 		n = V_tcbinfo.ipi_count;
1057 		req->oldidx = 2 * (sizeof xig)
1058 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1059 		return (0);
1060 	}
1061 
1062 	if (req->newptr != NULL)
1063 		return (EPERM);
1064 
1065 	/*
1066 	 * OK, now we're committed to doing something.
1067 	 */
1068 	INP_INFO_RLOCK(&V_tcbinfo);
1069 	gencnt = V_tcbinfo.ipi_gencnt;
1070 	n = V_tcbinfo.ipi_count;
1071 	INP_INFO_RUNLOCK(&V_tcbinfo);
1072 
1073 	m = syncache_pcbcount();
1074 
1075 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1076 		+ (n + m) * sizeof(struct xtcpcb));
1077 	if (error != 0)
1078 		return (error);
1079 
1080 	xig.xig_len = sizeof xig;
1081 	xig.xig_count = n + m;
1082 	xig.xig_gen = gencnt;
1083 	xig.xig_sogen = so_gencnt;
1084 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1085 	if (error)
1086 		return (error);
1087 
1088 	error = syncache_pcblist(req, m, &pcb_count);
1089 	if (error)
1090 		return (error);
1091 
1092 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1093 	if (inp_list == NULL)
1094 		return (ENOMEM);
1095 
1096 	INP_INFO_RLOCK(&V_tcbinfo);
1097 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1098 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1099 		INP_RLOCK(inp);
1100 		if (inp->inp_gencnt <= gencnt) {
1101 			/*
1102 			 * XXX: This use of cr_cansee(), introduced with
1103 			 * TCP state changes, is not quite right, but for
1104 			 * now, better than nothing.
1105 			 */
1106 			if (inp->inp_flags & INP_TIMEWAIT) {
1107 				if (intotw(inp) != NULL)
1108 					error = cr_cansee(req->td->td_ucred,
1109 					    intotw(inp)->tw_cred);
1110 				else
1111 					error = EINVAL;	/* Skip this inp. */
1112 			} else
1113 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1114 			if (error == 0)
1115 				inp_list[i++] = inp;
1116 		}
1117 		INP_RUNLOCK(inp);
1118 	}
1119 	INP_INFO_RUNLOCK(&V_tcbinfo);
1120 	n = i;
1121 
1122 	error = 0;
1123 	for (i = 0; i < n; i++) {
1124 		inp = inp_list[i];
1125 		INP_RLOCK(inp);
1126 		if (inp->inp_gencnt <= gencnt) {
1127 			struct xtcpcb xt;
1128 			void *inp_ppcb;
1129 
1130 			bzero(&xt, sizeof(xt));
1131 			xt.xt_len = sizeof xt;
1132 			/* XXX should avoid extra copy */
1133 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1134 			inp_ppcb = inp->inp_ppcb;
1135 			if (inp_ppcb == NULL)
1136 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1137 			else if (inp->inp_flags & INP_TIMEWAIT) {
1138 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1139 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1140 			} else
1141 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1142 			if (inp->inp_socket != NULL)
1143 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1144 			else {
1145 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1146 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1147 			}
1148 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1149 			INP_RUNLOCK(inp);
1150 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1151 		} else
1152 			INP_RUNLOCK(inp);
1153 
1154 	}
1155 	if (!error) {
1156 		/*
1157 		 * Give the user an updated idea of our state.
1158 		 * If the generation differs from what we told
1159 		 * her before, she knows that something happened
1160 		 * while we were processing this request, and it
1161 		 * might be necessary to retry.
1162 		 */
1163 		INP_INFO_RLOCK(&V_tcbinfo);
1164 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1165 		xig.xig_sogen = so_gencnt;
1166 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1167 		INP_INFO_RUNLOCK(&V_tcbinfo);
1168 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1169 	}
1170 	free(inp_list, M_TEMP);
1171 	return (error);
1172 }
1173 
1174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1175     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1176 
1177 static int
1178 tcp_getcred(SYSCTL_HANDLER_ARGS)
1179 {
1180 	struct xucred xuc;
1181 	struct sockaddr_in addrs[2];
1182 	struct inpcb *inp;
1183 	int error;
1184 
1185 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1186 	if (error)
1187 		return (error);
1188 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1189 	if (error)
1190 		return (error);
1191 	INP_INFO_RLOCK(&V_tcbinfo);
1192 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1193 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1194 	if (inp != NULL) {
1195 		INP_RLOCK(inp);
1196 		INP_INFO_RUNLOCK(&V_tcbinfo);
1197 		if (inp->inp_socket == NULL)
1198 			error = ENOENT;
1199 		if (error == 0)
1200 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1201 		if (error == 0)
1202 			cru2x(inp->inp_cred, &xuc);
1203 		INP_RUNLOCK(inp);
1204 	} else {
1205 		INP_INFO_RUNLOCK(&V_tcbinfo);
1206 		error = ENOENT;
1207 	}
1208 	if (error == 0)
1209 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1210 	return (error);
1211 }
1212 
1213 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1214     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1215     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1216 
1217 #ifdef INET6
1218 static int
1219 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1220 {
1221 	struct xucred xuc;
1222 	struct sockaddr_in6 addrs[2];
1223 	struct inpcb *inp;
1224 	int error, mapped = 0;
1225 
1226 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1227 	if (error)
1228 		return (error);
1229 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1230 	if (error)
1231 		return (error);
1232 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1233 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1234 		return (error);
1235 	}
1236 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1237 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1238 			mapped = 1;
1239 		else
1240 			return (EINVAL);
1241 	}
1242 
1243 	INP_INFO_RLOCK(&V_tcbinfo);
1244 	if (mapped == 1)
1245 		inp = in_pcblookup_hash(&V_tcbinfo,
1246 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1247 			addrs[1].sin6_port,
1248 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1249 			addrs[0].sin6_port,
1250 			0, NULL);
1251 	else
1252 		inp = in6_pcblookup_hash(&V_tcbinfo,
1253 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1254 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1255 	if (inp != NULL) {
1256 		INP_RLOCK(inp);
1257 		INP_INFO_RUNLOCK(&V_tcbinfo);
1258 		if (inp->inp_socket == NULL)
1259 			error = ENOENT;
1260 		if (error == 0)
1261 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1262 		if (error == 0)
1263 			cru2x(inp->inp_cred, &xuc);
1264 		INP_RUNLOCK(inp);
1265 	} else {
1266 		INP_INFO_RUNLOCK(&V_tcbinfo);
1267 		error = ENOENT;
1268 	}
1269 	if (error == 0)
1270 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1271 	return (error);
1272 }
1273 
1274 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1275     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1276     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1277 #endif
1278 
1279 
1280 void
1281 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1282 {
1283 	struct ip *ip = vip;
1284 	struct tcphdr *th;
1285 	struct in_addr faddr;
1286 	struct inpcb *inp;
1287 	struct tcpcb *tp;
1288 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1289 	struct icmp *icp;
1290 	struct in_conninfo inc;
1291 	tcp_seq icmp_tcp_seq;
1292 	int mtu;
1293 
1294 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1295 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1296 		return;
1297 
1298 	if (cmd == PRC_MSGSIZE)
1299 		notify = tcp_mtudisc;
1300 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1301 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1302 		notify = tcp_drop_syn_sent;
1303 	/*
1304 	 * Redirects don't need to be handled up here.
1305 	 */
1306 	else if (PRC_IS_REDIRECT(cmd))
1307 		return;
1308 	/*
1309 	 * Source quench is depreciated.
1310 	 */
1311 	else if (cmd == PRC_QUENCH)
1312 		return;
1313 	/*
1314 	 * Hostdead is ugly because it goes linearly through all PCBs.
1315 	 * XXX: We never get this from ICMP, otherwise it makes an
1316 	 * excellent DoS attack on machines with many connections.
1317 	 */
1318 	else if (cmd == PRC_HOSTDEAD)
1319 		ip = NULL;
1320 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1321 		return;
1322 	if (ip != NULL) {
1323 		icp = (struct icmp *)((caddr_t)ip
1324 				      - offsetof(struct icmp, icmp_ip));
1325 		th = (struct tcphdr *)((caddr_t)ip
1326 				       + (ip->ip_hl << 2));
1327 		INP_INFO_WLOCK(&V_tcbinfo);
1328 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1329 		    ip->ip_src, th->th_sport, 0, NULL);
1330 		if (inp != NULL)  {
1331 			INP_WLOCK(inp);
1332 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1333 			    !(inp->inp_flags & INP_DROPPED) &&
1334 			    !(inp->inp_socket == NULL)) {
1335 				icmp_tcp_seq = htonl(th->th_seq);
1336 				tp = intotcpcb(inp);
1337 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1338 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1339 					if (cmd == PRC_MSGSIZE) {
1340 					    /*
1341 					     * MTU discovery:
1342 					     * If we got a needfrag set the MTU
1343 					     * in the route to the suggested new
1344 					     * value (if given) and then notify.
1345 					     */
1346 					    bzero(&inc, sizeof(inc));
1347 					    inc.inc_faddr = faddr;
1348 					    inc.inc_fibnum =
1349 						inp->inp_inc.inc_fibnum;
1350 
1351 					    mtu = ntohs(icp->icmp_nextmtu);
1352 					    /*
1353 					     * If no alternative MTU was
1354 					     * proposed, try the next smaller
1355 					     * one.  ip->ip_len has already
1356 					     * been swapped in icmp_input().
1357 					     */
1358 					    if (!mtu)
1359 						mtu = ip_next_mtu(ip->ip_len,
1360 						 1);
1361 					    if (mtu < max(296, V_tcp_minmss
1362 						 + sizeof(struct tcpiphdr)))
1363 						mtu = 0;
1364 					    if (!mtu)
1365 						mtu = V_tcp_mssdflt
1366 						 + sizeof(struct tcpiphdr);
1367 					    /*
1368 					     * Only cache the the MTU if it
1369 					     * is smaller than the interface
1370 					     * or route MTU.  tcp_mtudisc()
1371 					     * will do right thing by itself.
1372 					     */
1373 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1374 						tcp_hc_updatemtu(&inc, mtu);
1375 					}
1376 
1377 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1378 				}
1379 			}
1380 			if (inp != NULL)
1381 				INP_WUNLOCK(inp);
1382 		} else {
1383 			bzero(&inc, sizeof(inc));
1384 			inc.inc_fport = th->th_dport;
1385 			inc.inc_lport = th->th_sport;
1386 			inc.inc_faddr = faddr;
1387 			inc.inc_laddr = ip->ip_src;
1388 			syncache_unreach(&inc, th);
1389 		}
1390 		INP_INFO_WUNLOCK(&V_tcbinfo);
1391 	} else
1392 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1393 }
1394 
1395 #ifdef INET6
1396 void
1397 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1398 {
1399 	struct tcphdr th;
1400 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1401 	struct ip6_hdr *ip6;
1402 	struct mbuf *m;
1403 	struct ip6ctlparam *ip6cp = NULL;
1404 	const struct sockaddr_in6 *sa6_src = NULL;
1405 	int off;
1406 	struct tcp_portonly {
1407 		u_int16_t th_sport;
1408 		u_int16_t th_dport;
1409 	} *thp;
1410 
1411 	if (sa->sa_family != AF_INET6 ||
1412 	    sa->sa_len != sizeof(struct sockaddr_in6))
1413 		return;
1414 
1415 	if (cmd == PRC_MSGSIZE)
1416 		notify = tcp_mtudisc;
1417 	else if (!PRC_IS_REDIRECT(cmd) &&
1418 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1419 		return;
1420 	/* Source quench is depreciated. */
1421 	else if (cmd == PRC_QUENCH)
1422 		return;
1423 
1424 	/* if the parameter is from icmp6, decode it. */
1425 	if (d != NULL) {
1426 		ip6cp = (struct ip6ctlparam *)d;
1427 		m = ip6cp->ip6c_m;
1428 		ip6 = ip6cp->ip6c_ip6;
1429 		off = ip6cp->ip6c_off;
1430 		sa6_src = ip6cp->ip6c_src;
1431 	} else {
1432 		m = NULL;
1433 		ip6 = NULL;
1434 		off = 0;	/* fool gcc */
1435 		sa6_src = &sa6_any;
1436 	}
1437 
1438 	if (ip6 != NULL) {
1439 		struct in_conninfo inc;
1440 		/*
1441 		 * XXX: We assume that when IPV6 is non NULL,
1442 		 * M and OFF are valid.
1443 		 */
1444 
1445 		/* check if we can safely examine src and dst ports */
1446 		if (m->m_pkthdr.len < off + sizeof(*thp))
1447 			return;
1448 
1449 		bzero(&th, sizeof(th));
1450 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1451 
1452 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1453 		    (struct sockaddr *)ip6cp->ip6c_src,
1454 		    th.th_sport, cmd, NULL, notify);
1455 
1456 		bzero(&inc, sizeof(inc));
1457 		inc.inc_fport = th.th_dport;
1458 		inc.inc_lport = th.th_sport;
1459 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1460 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1461 		inc.inc_flags |= INC_ISIPV6;
1462 		INP_INFO_WLOCK(&V_tcbinfo);
1463 		syncache_unreach(&inc, &th);
1464 		INP_INFO_WUNLOCK(&V_tcbinfo);
1465 	} else
1466 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1467 			      0, cmd, NULL, notify);
1468 }
1469 #endif /* INET6 */
1470 
1471 
1472 /*
1473  * Following is where TCP initial sequence number generation occurs.
1474  *
1475  * There are two places where we must use initial sequence numbers:
1476  * 1.  In SYN-ACK packets.
1477  * 2.  In SYN packets.
1478  *
1479  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1480  * tcp_syncache.c for details.
1481  *
1482  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1483  * depends on this property.  In addition, these ISNs should be
1484  * unguessable so as to prevent connection hijacking.  To satisfy
1485  * the requirements of this situation, the algorithm outlined in
1486  * RFC 1948 is used, with only small modifications.
1487  *
1488  * Implementation details:
1489  *
1490  * Time is based off the system timer, and is corrected so that it
1491  * increases by one megabyte per second.  This allows for proper
1492  * recycling on high speed LANs while still leaving over an hour
1493  * before rollover.
1494  *
1495  * As reading the *exact* system time is too expensive to be done
1496  * whenever setting up a TCP connection, we increment the time
1497  * offset in two ways.  First, a small random positive increment
1498  * is added to isn_offset for each connection that is set up.
1499  * Second, the function tcp_isn_tick fires once per clock tick
1500  * and increments isn_offset as necessary so that sequence numbers
1501  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1502  * random positive increments serve only to ensure that the same
1503  * exact sequence number is never sent out twice (as could otherwise
1504  * happen when a port is recycled in less than the system tick
1505  * interval.)
1506  *
1507  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1508  * between seeding of isn_secret.  This is normally set to zero,
1509  * as reseeding should not be necessary.
1510  *
1511  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1512  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1513  * general, this means holding an exclusive (write) lock.
1514  */
1515 
1516 #define ISN_BYTES_PER_SECOND 1048576
1517 #define ISN_STATIC_INCREMENT 4096
1518 #define ISN_RANDOM_INCREMENT (4096 - 1)
1519 
1520 static VNET_DEFINE(u_char, isn_secret[32]);
1521 static VNET_DEFINE(int, isn_last_reseed);
1522 static VNET_DEFINE(u_int32_t, isn_offset);
1523 static VNET_DEFINE(u_int32_t, isn_offset_old);
1524 
1525 #define	V_isn_secret			VNET(isn_secret)
1526 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1527 #define	V_isn_offset			VNET(isn_offset)
1528 #define	V_isn_offset_old		VNET(isn_offset_old)
1529 
1530 tcp_seq
1531 tcp_new_isn(struct tcpcb *tp)
1532 {
1533 	MD5_CTX isn_ctx;
1534 	u_int32_t md5_buffer[4];
1535 	tcp_seq new_isn;
1536 
1537 	INP_WLOCK_ASSERT(tp->t_inpcb);
1538 
1539 	ISN_LOCK();
1540 	/* Seed if this is the first use, reseed if requested. */
1541 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1542 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1543 		< (u_int)ticks))) {
1544 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1545 		V_isn_last_reseed = ticks;
1546 	}
1547 
1548 	/* Compute the md5 hash and return the ISN. */
1549 	MD5Init(&isn_ctx);
1550 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1551 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1552 #ifdef INET6
1553 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1554 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1555 			  sizeof(struct in6_addr));
1556 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1557 			  sizeof(struct in6_addr));
1558 	} else
1559 #endif
1560 	{
1561 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1562 			  sizeof(struct in_addr));
1563 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1564 			  sizeof(struct in_addr));
1565 	}
1566 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1567 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1568 	new_isn = (tcp_seq) md5_buffer[0];
1569 	V_isn_offset += ISN_STATIC_INCREMENT +
1570 		(arc4random() & ISN_RANDOM_INCREMENT);
1571 	new_isn += V_isn_offset;
1572 	ISN_UNLOCK();
1573 	return (new_isn);
1574 }
1575 
1576 /*
1577  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1578  * to keep time flowing at a relatively constant rate.  If the random
1579  * increments have already pushed us past the projected offset, do nothing.
1580  */
1581 static void
1582 tcp_isn_tick(void *xtp)
1583 {
1584 	VNET_ITERATOR_DECL(vnet_iter);
1585 	u_int32_t projected_offset;
1586 
1587 	VNET_LIST_RLOCK_NOSLEEP();
1588 	ISN_LOCK();
1589 	VNET_FOREACH(vnet_iter) {
1590 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1591 		projected_offset =
1592 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1593 
1594 		if (SEQ_GT(projected_offset, V_isn_offset))
1595 			V_isn_offset = projected_offset;
1596 
1597 		V_isn_offset_old = V_isn_offset;
1598 		CURVNET_RESTORE();
1599 	}
1600 	ISN_UNLOCK();
1601 	VNET_LIST_RUNLOCK_NOSLEEP();
1602 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1603 }
1604 
1605 /*
1606  * When a specific ICMP unreachable message is received and the
1607  * connection state is SYN-SENT, drop the connection.  This behavior
1608  * is controlled by the icmp_may_rst sysctl.
1609  */
1610 struct inpcb *
1611 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1612 {
1613 	struct tcpcb *tp;
1614 
1615 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1616 	INP_WLOCK_ASSERT(inp);
1617 
1618 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1619 	    (inp->inp_flags & INP_DROPPED))
1620 		return (inp);
1621 
1622 	tp = intotcpcb(inp);
1623 	if (tp->t_state != TCPS_SYN_SENT)
1624 		return (inp);
1625 
1626 	tp = tcp_drop(tp, errno);
1627 	if (tp != NULL)
1628 		return (inp);
1629 	else
1630 		return (NULL);
1631 }
1632 
1633 /*
1634  * When `need fragmentation' ICMP is received, update our idea of the MSS
1635  * based on the new value in the route.  Also nudge TCP to send something,
1636  * since we know the packet we just sent was dropped.
1637  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1638  */
1639 struct inpcb *
1640 tcp_mtudisc(struct inpcb *inp, int errno)
1641 {
1642 	struct tcpcb *tp;
1643 	struct socket *so;
1644 
1645 	INP_WLOCK_ASSERT(inp);
1646 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1647 	    (inp->inp_flags & INP_DROPPED))
1648 		return (inp);
1649 
1650 	tp = intotcpcb(inp);
1651 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1652 
1653 	tcp_mss_update(tp, -1, NULL, NULL);
1654 
1655 	so = inp->inp_socket;
1656 	SOCKBUF_LOCK(&so->so_snd);
1657 	/* If the mss is larger than the socket buffer, decrease the mss. */
1658 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1659 		tp->t_maxseg = so->so_snd.sb_hiwat;
1660 	SOCKBUF_UNLOCK(&so->so_snd);
1661 
1662 	TCPSTAT_INC(tcps_mturesent);
1663 	tp->t_rtttime = 0;
1664 	tp->snd_nxt = tp->snd_una;
1665 	tcp_free_sackholes(tp);
1666 	tp->snd_recover = tp->snd_max;
1667 	if (tp->t_flags & TF_SACK_PERMIT)
1668 		EXIT_FASTRECOVERY(tp);
1669 	tcp_output_send(tp);
1670 	return (inp);
1671 }
1672 
1673 /*
1674  * Look-up the routing entry to the peer of this inpcb.  If no route
1675  * is found and it cannot be allocated, then return 0.  This routine
1676  * is called by TCP routines that access the rmx structure and by
1677  * tcp_mss_update to get the peer/interface MTU.
1678  */
1679 u_long
1680 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1681 {
1682 	struct route sro;
1683 	struct sockaddr_in *dst;
1684 	struct ifnet *ifp;
1685 	u_long maxmtu = 0;
1686 
1687 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1688 
1689 	bzero(&sro, sizeof(sro));
1690 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1691 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1692 		dst->sin_family = AF_INET;
1693 		dst->sin_len = sizeof(*dst);
1694 		dst->sin_addr = inc->inc_faddr;
1695 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1696 	}
1697 	if (sro.ro_rt != NULL) {
1698 		ifp = sro.ro_rt->rt_ifp;
1699 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1700 			maxmtu = ifp->if_mtu;
1701 		else
1702 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1703 
1704 		/* Report additional interface capabilities. */
1705 		if (flags != NULL) {
1706 			if (ifp->if_capenable & IFCAP_TSO4 &&
1707 			    ifp->if_hwassist & CSUM_TSO)
1708 				*flags |= CSUM_TSO;
1709 		}
1710 		RTFREE(sro.ro_rt);
1711 	}
1712 	return (maxmtu);
1713 }
1714 
1715 #ifdef INET6
1716 u_long
1717 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1718 {
1719 	struct route_in6 sro6;
1720 	struct ifnet *ifp;
1721 	u_long maxmtu = 0;
1722 
1723 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1724 
1725 	bzero(&sro6, sizeof(sro6));
1726 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1727 		sro6.ro_dst.sin6_family = AF_INET6;
1728 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1729 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1730 		rtalloc_ign((struct route *)&sro6, 0);
1731 	}
1732 	if (sro6.ro_rt != NULL) {
1733 		ifp = sro6.ro_rt->rt_ifp;
1734 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1735 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1736 		else
1737 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1738 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1739 
1740 		/* Report additional interface capabilities. */
1741 		if (flags != NULL) {
1742 			if (ifp->if_capenable & IFCAP_TSO6 &&
1743 			    ifp->if_hwassist & CSUM_TSO)
1744 				*flags |= CSUM_TSO;
1745 		}
1746 		RTFREE(sro6.ro_rt);
1747 	}
1748 
1749 	return (maxmtu);
1750 }
1751 #endif /* INET6 */
1752 
1753 #ifdef IPSEC
1754 /* compute ESP/AH header size for TCP, including outer IP header. */
1755 size_t
1756 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1757 {
1758 	struct inpcb *inp;
1759 	struct mbuf *m;
1760 	size_t hdrsiz;
1761 	struct ip *ip;
1762 #ifdef INET6
1763 	struct ip6_hdr *ip6;
1764 #endif
1765 	struct tcphdr *th;
1766 
1767 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1768 		return (0);
1769 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1770 	if (!m)
1771 		return (0);
1772 
1773 #ifdef INET6
1774 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1775 		ip6 = mtod(m, struct ip6_hdr *);
1776 		th = (struct tcphdr *)(ip6 + 1);
1777 		m->m_pkthdr.len = m->m_len =
1778 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1779 		tcpip_fillheaders(inp, ip6, th);
1780 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1781 	} else
1782 #endif /* INET6 */
1783 	{
1784 		ip = mtod(m, struct ip *);
1785 		th = (struct tcphdr *)(ip + 1);
1786 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1787 		tcpip_fillheaders(inp, ip, th);
1788 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1789 	}
1790 
1791 	m_free(m);
1792 	return (hdrsiz);
1793 }
1794 #endif /* IPSEC */
1795 
1796 /*
1797  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1798  *
1799  * This code attempts to calculate the bandwidth-delay product as a
1800  * means of determining the optimal window size to maximize bandwidth,
1801  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1802  * routers.  This code also does a fairly good job keeping RTTs in check
1803  * across slow links like modems.  We implement an algorithm which is very
1804  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1805  * transmitter side of a TCP connection and so only effects the transmit
1806  * side of the connection.
1807  *
1808  * BACKGROUND:  TCP makes no provision for the management of buffer space
1809  * at the end points or at the intermediate routers and switches.  A TCP
1810  * stream, whether using NewReno or not, will eventually buffer as
1811  * many packets as it is able and the only reason this typically works is
1812  * due to the fairly small default buffers made available for a connection
1813  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1814  * scaling it is now fairly easy for even a single TCP connection to blow-out
1815  * all available buffer space not only on the local interface, but on
1816  * intermediate routers and switches as well.  NewReno makes a misguided
1817  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1818  * then backing off, then steadily increasing the window again until another
1819  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1820  * is only made worse as network loads increase and the idea of intentionally
1821  * blowing out network buffers is, frankly, a terrible way to manage network
1822  * resources.
1823  *
1824  * It is far better to limit the transmit window prior to the failure
1825  * condition being achieved.  There are two general ways to do this:  First
1826  * you can 'scan' through different transmit window sizes and locate the
1827  * point where the RTT stops increasing, indicating that you have filled the
1828  * pipe, then scan backwards until you note that RTT stops decreasing, then
1829  * repeat ad-infinitum.  This method works in principle but has severe
1830  * implementation issues due to RTT variances, timer granularity, and
1831  * instability in the algorithm which can lead to many false positives and
1832  * create oscillations as well as interact badly with other TCP streams
1833  * implementing the same algorithm.
1834  *
1835  * The second method is to limit the window to the bandwidth delay product
1836  * of the link.  This is the method we implement.  RTT variances and our
1837  * own manipulation of the congestion window, bwnd, can potentially
1838  * destabilize the algorithm.  For this reason we have to stabilize the
1839  * elements used to calculate the window.  We do this by using the minimum
1840  * observed RTT, the long term average of the observed bandwidth, and
1841  * by adding two segments worth of slop.  It isn't perfect but it is able
1842  * to react to changing conditions and gives us a very stable basis on
1843  * which to extend the algorithm.
1844  */
1845 void
1846 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1847 {
1848 	u_long bw;
1849 	u_long bwnd;
1850 	int save_ticks;
1851 
1852 	INP_WLOCK_ASSERT(tp->t_inpcb);
1853 
1854 	/*
1855 	 * If inflight_enable is disabled in the middle of a tcp connection,
1856 	 * make sure snd_bwnd is effectively disabled.
1857 	 */
1858 	if (V_tcp_inflight_enable == 0 ||
1859 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1860 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1861 		tp->snd_bandwidth = 0;
1862 		return;
1863 	}
1864 
1865 	/*
1866 	 * Figure out the bandwidth.  Due to the tick granularity this
1867 	 * is a very rough number and it MUST be averaged over a fairly
1868 	 * long period of time.  XXX we need to take into account a link
1869 	 * that is not using all available bandwidth, but for now our
1870 	 * slop will ramp us up if this case occurs and the bandwidth later
1871 	 * increases.
1872 	 *
1873 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1874 	 * effectively reset t_bw_rtttime for this case.
1875 	 */
1876 	save_ticks = ticks;
1877 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1878 		return;
1879 
1880 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1881 	    (save_ticks - tp->t_bw_rtttime);
1882 	tp->t_bw_rtttime = save_ticks;
1883 	tp->t_bw_rtseq = ack_seq;
1884 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1885 		return;
1886 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1887 
1888 	tp->snd_bandwidth = bw;
1889 
1890 	/*
1891 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1892 	 * segments.  The additional slop puts us squarely in the sweet
1893 	 * spot and also handles the bandwidth run-up case and stabilization.
1894 	 * Without the slop we could be locking ourselves into a lower
1895 	 * bandwidth.
1896 	 *
1897 	 * Situations Handled:
1898 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1899 	 *	    high speed LANs, allowing larger TCP buffers to be
1900 	 *	    specified, and also does a good job preventing
1901 	 *	    over-queueing of packets over choke points like modems
1902 	 *	    (at least for the transmit side).
1903 	 *
1904 	 *	(2) Is able to handle changing network loads (bandwidth
1905 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1906 	 *	    increases).
1907 	 *
1908 	 *	(3) Theoretically should stabilize in the face of multiple
1909 	 *	    connections implementing the same algorithm (this may need
1910 	 *	    a little work).
1911 	 *
1912 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1913 	 *	    be adjusted with a sysctl but typically only needs to be
1914 	 *	    on very slow connections.  A value no smaller then 5
1915 	 *	    should be used, but only reduce this default if you have
1916 	 *	    no other choice.
1917 	 */
1918 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1919 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1920 #undef USERTT
1921 
1922 	if (tcp_inflight_debug > 0) {
1923 		static int ltime;
1924 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1925 			ltime = ticks;
1926 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1927 			    tp,
1928 			    bw,
1929 			    tp->t_rttbest,
1930 			    tp->t_srtt,
1931 			    bwnd
1932 			);
1933 		}
1934 	}
1935 	if ((long)bwnd < V_tcp_inflight_min)
1936 		bwnd = V_tcp_inflight_min;
1937 	if (bwnd > V_tcp_inflight_max)
1938 		bwnd = V_tcp_inflight_max;
1939 	if ((long)bwnd < tp->t_maxseg * 2)
1940 		bwnd = tp->t_maxseg * 2;
1941 	tp->snd_bwnd = bwnd;
1942 }
1943 
1944 #ifdef TCP_SIGNATURE
1945 /*
1946  * Callback function invoked by m_apply() to digest TCP segment data
1947  * contained within an mbuf chain.
1948  */
1949 static int
1950 tcp_signature_apply(void *fstate, void *data, u_int len)
1951 {
1952 
1953 	MD5Update(fstate, (u_char *)data, len);
1954 	return (0);
1955 }
1956 
1957 /*
1958  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1959  *
1960  * Parameters:
1961  * m		pointer to head of mbuf chain
1962  * _unused
1963  * len		length of TCP segment data, excluding options
1964  * optlen	length of TCP segment options
1965  * buf		pointer to storage for computed MD5 digest
1966  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1967  *
1968  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1969  * When called from tcp_input(), we can be sure that th_sum has been
1970  * zeroed out and verified already.
1971  *
1972  * Return 0 if successful, otherwise return -1.
1973  *
1974  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1975  * search with the destination IP address, and a 'magic SPI' to be
1976  * determined by the application. This is hardcoded elsewhere to 1179
1977  * right now. Another branch of this code exists which uses the SPD to
1978  * specify per-application flows but it is unstable.
1979  */
1980 int
1981 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1982     u_char *buf, u_int direction)
1983 {
1984 	union sockaddr_union dst;
1985 	struct ippseudo ippseudo;
1986 	MD5_CTX ctx;
1987 	int doff;
1988 	struct ip *ip;
1989 	struct ipovly *ipovly;
1990 	struct secasvar *sav;
1991 	struct tcphdr *th;
1992 #ifdef INET6
1993 	struct ip6_hdr *ip6;
1994 	struct in6_addr in6;
1995 	char ip6buf[INET6_ADDRSTRLEN];
1996 	uint32_t plen;
1997 	uint16_t nhdr;
1998 #endif
1999 	u_short savecsum;
2000 
2001 	KASSERT(m != NULL, ("NULL mbuf chain"));
2002 	KASSERT(buf != NULL, ("NULL signature pointer"));
2003 
2004 	/* Extract the destination from the IP header in the mbuf. */
2005 	bzero(&dst, sizeof(union sockaddr_union));
2006 	ip = mtod(m, struct ip *);
2007 #ifdef INET6
2008 	ip6 = NULL;	/* Make the compiler happy. */
2009 #endif
2010 	switch (ip->ip_v) {
2011 	case IPVERSION:
2012 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2013 		dst.sa.sa_family = AF_INET;
2014 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2015 		    ip->ip_src : ip->ip_dst;
2016 		break;
2017 #ifdef INET6
2018 	case (IPV6_VERSION >> 4):
2019 		ip6 = mtod(m, struct ip6_hdr *);
2020 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2021 		dst.sa.sa_family = AF_INET6;
2022 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2023 		    ip6->ip6_src : ip6->ip6_dst;
2024 		break;
2025 #endif
2026 	default:
2027 		return (EINVAL);
2028 		/* NOTREACHED */
2029 		break;
2030 	}
2031 
2032 	/* Look up an SADB entry which matches the address of the peer. */
2033 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2034 	if (sav == NULL) {
2035 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2036 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2037 #ifdef INET6
2038 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2039 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2040 #endif
2041 			"(unsupported)"));
2042 		return (EINVAL);
2043 	}
2044 
2045 	MD5Init(&ctx);
2046 	/*
2047 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2048 	 *
2049 	 * XXX The ippseudo header MUST be digested in network byte order,
2050 	 * or else we'll fail the regression test. Assume all fields we've
2051 	 * been doing arithmetic on have been in host byte order.
2052 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2053 	 * tcp_output(), the underlying ip_len member has not yet been set.
2054 	 */
2055 	switch (ip->ip_v) {
2056 	case IPVERSION:
2057 		ipovly = (struct ipovly *)ip;
2058 		ippseudo.ippseudo_src = ipovly->ih_src;
2059 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2060 		ippseudo.ippseudo_pad = 0;
2061 		ippseudo.ippseudo_p = IPPROTO_TCP;
2062 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2063 		    optlen);
2064 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2065 
2066 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2067 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2068 		break;
2069 #ifdef INET6
2070 	/*
2071 	 * RFC 2385, 2.0  Proposal
2072 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2073 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2074 	 * extended next header value (to form 32 bits), and 32-bit segment
2075 	 * length.
2076 	 * Note: Upper-Layer Packet Length comes before Next Header.
2077 	 */
2078 	case (IPV6_VERSION >> 4):
2079 		in6 = ip6->ip6_src;
2080 		in6_clearscope(&in6);
2081 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2082 		in6 = ip6->ip6_dst;
2083 		in6_clearscope(&in6);
2084 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2085 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2086 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2087 		nhdr = 0;
2088 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2089 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2090 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2091 		nhdr = IPPROTO_TCP;
2092 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2093 
2094 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2095 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2096 		break;
2097 #endif
2098 	default:
2099 		return (EINVAL);
2100 		/* NOTREACHED */
2101 		break;
2102 	}
2103 
2104 
2105 	/*
2106 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2107 	 * The TCP checksum must be set to zero.
2108 	 */
2109 	savecsum = th->th_sum;
2110 	th->th_sum = 0;
2111 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2112 	th->th_sum = savecsum;
2113 
2114 	/*
2115 	 * Step 3: Update MD5 hash with TCP segment data.
2116 	 *         Use m_apply() to avoid an early m_pullup().
2117 	 */
2118 	if (len > 0)
2119 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2120 
2121 	/*
2122 	 * Step 4: Update MD5 hash with shared secret.
2123 	 */
2124 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2125 	MD5Final(buf, &ctx);
2126 
2127 	key_sa_recordxfer(sav, m);
2128 	KEY_FREESAV(&sav);
2129 	return (0);
2130 }
2131 #endif /* TCP_SIGNATURE */
2132 
2133 static int
2134 sysctl_drop(SYSCTL_HANDLER_ARGS)
2135 {
2136 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2137 	struct sockaddr_storage addrs[2];
2138 	struct inpcb *inp;
2139 	struct tcpcb *tp;
2140 	struct tcptw *tw;
2141 	struct sockaddr_in *fin, *lin;
2142 #ifdef INET6
2143 	struct sockaddr_in6 *fin6, *lin6;
2144 #endif
2145 	int error;
2146 
2147 	inp = NULL;
2148 	fin = lin = NULL;
2149 #ifdef INET6
2150 	fin6 = lin6 = NULL;
2151 #endif
2152 	error = 0;
2153 
2154 	if (req->oldptr != NULL || req->oldlen != 0)
2155 		return (EINVAL);
2156 	if (req->newptr == NULL)
2157 		return (EPERM);
2158 	if (req->newlen < sizeof(addrs))
2159 		return (ENOMEM);
2160 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2161 	if (error)
2162 		return (error);
2163 
2164 	switch (addrs[0].ss_family) {
2165 #ifdef INET6
2166 	case AF_INET6:
2167 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2168 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2169 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2170 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2171 			return (EINVAL);
2172 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2173 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2174 				return (EINVAL);
2175 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2176 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2177 			fin = (struct sockaddr_in *)&addrs[0];
2178 			lin = (struct sockaddr_in *)&addrs[1];
2179 			break;
2180 		}
2181 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2182 		if (error)
2183 			return (error);
2184 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2185 		if (error)
2186 			return (error);
2187 		break;
2188 #endif
2189 	case AF_INET:
2190 		fin = (struct sockaddr_in *)&addrs[0];
2191 		lin = (struct sockaddr_in *)&addrs[1];
2192 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2193 		    lin->sin_len != sizeof(struct sockaddr_in))
2194 			return (EINVAL);
2195 		break;
2196 	default:
2197 		return (EINVAL);
2198 	}
2199 	INP_INFO_WLOCK(&V_tcbinfo);
2200 	switch (addrs[0].ss_family) {
2201 #ifdef INET6
2202 	case AF_INET6:
2203 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2204 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2205 		    NULL);
2206 		break;
2207 #endif
2208 	case AF_INET:
2209 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2210 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2211 		break;
2212 	}
2213 	if (inp != NULL) {
2214 		INP_WLOCK(inp);
2215 		if (inp->inp_flags & INP_TIMEWAIT) {
2216 			/*
2217 			 * XXXRW: There currently exists a state where an
2218 			 * inpcb is present, but its timewait state has been
2219 			 * discarded.  For now, don't allow dropping of this
2220 			 * type of inpcb.
2221 			 */
2222 			tw = intotw(inp);
2223 			if (tw != NULL)
2224 				tcp_twclose(tw, 0);
2225 			else
2226 				INP_WUNLOCK(inp);
2227 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2228 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2229 			tp = intotcpcb(inp);
2230 			tp = tcp_drop(tp, ECONNABORTED);
2231 			if (tp != NULL)
2232 				INP_WUNLOCK(inp);
2233 		} else
2234 			INP_WUNLOCK(inp);
2235 	} else
2236 		error = ESRCH;
2237 	INP_INFO_WUNLOCK(&V_tcbinfo);
2238 	return (error);
2239 }
2240 
2241 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2242     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2243     0, sysctl_drop, "", "Drop TCP connection");
2244 
2245 /*
2246  * Generate a standardized TCP log line for use throughout the
2247  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2248  * allow use in the interrupt context.
2249  *
2250  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2251  * NB: The function may return NULL if memory allocation failed.
2252  *
2253  * Due to header inclusion and ordering limitations the struct ip
2254  * and ip6_hdr pointers have to be passed as void pointers.
2255  */
2256 char *
2257 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2258     const void *ip6hdr)
2259 {
2260 	char *s, *sp;
2261 	size_t size;
2262 	struct ip *ip;
2263 #ifdef INET6
2264 	const struct ip6_hdr *ip6;
2265 
2266 	ip6 = (const struct ip6_hdr *)ip6hdr;
2267 #endif /* INET6 */
2268 	ip = (struct ip *)ip4hdr;
2269 
2270 	/*
2271 	 * The log line looks like this:
2272 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2273 	 */
2274 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2275 	    sizeof(PRINT_TH_FLAGS) + 1 +
2276 #ifdef INET6
2277 	    2 * INET6_ADDRSTRLEN;
2278 #else
2279 	    2 * INET_ADDRSTRLEN;
2280 #endif /* INET6 */
2281 
2282 	/* Is logging enabled? */
2283 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2284 		return (NULL);
2285 
2286 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2287 	if (s == NULL)
2288 		return (NULL);
2289 
2290 	strcat(s, "TCP: [");
2291 	sp = s + strlen(s);
2292 
2293 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2294 		inet_ntoa_r(inc->inc_faddr, sp);
2295 		sp = s + strlen(s);
2296 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2297 		sp = s + strlen(s);
2298 		inet_ntoa_r(inc->inc_laddr, sp);
2299 		sp = s + strlen(s);
2300 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2301 #ifdef INET6
2302 	} else if (inc) {
2303 		ip6_sprintf(sp, &inc->inc6_faddr);
2304 		sp = s + strlen(s);
2305 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2306 		sp = s + strlen(s);
2307 		ip6_sprintf(sp, &inc->inc6_laddr);
2308 		sp = s + strlen(s);
2309 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2310 	} else if (ip6 && th) {
2311 		ip6_sprintf(sp, &ip6->ip6_src);
2312 		sp = s + strlen(s);
2313 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2314 		sp = s + strlen(s);
2315 		ip6_sprintf(sp, &ip6->ip6_dst);
2316 		sp = s + strlen(s);
2317 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2318 #endif /* INET6 */
2319 	} else if (ip && th) {
2320 		inet_ntoa_r(ip->ip_src, sp);
2321 		sp = s + strlen(s);
2322 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2323 		sp = s + strlen(s);
2324 		inet_ntoa_r(ip->ip_dst, sp);
2325 		sp = s + strlen(s);
2326 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2327 	} else {
2328 		free(s, M_TCPLOG);
2329 		return (NULL);
2330 	}
2331 	sp = s + strlen(s);
2332 	if (th)
2333 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2334 	if (*(s + size - 1) != '\0')
2335 		panic("%s: string too long", __func__);
2336 	return (s);
2337 }
2338