xref: /freebsd/sys/netinet/tcp_subr.c (revision 195ebc7e9e4b129de810833791a19dfb4349d6a9)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 #include <sys/vimage.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/route.h>
63 #include <net/if.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet/vinet.h>
98 #include <netinet6/ip6protosw.h>
99 #include <netinet6/vinet6.h>
100 
101 #ifdef IPSEC
102 #include <netipsec/ipsec.h>
103 #include <netipsec/xform.h>
104 #ifdef INET6
105 #include <netipsec/ipsec6.h>
106 #endif
107 #include <netipsec/key.h>
108 #include <sys/syslog.h>
109 #endif /*IPSEC*/
110 
111 #include <machine/in_cksum.h>
112 #include <sys/md5.h>
113 
114 #include <security/mac/mac_framework.h>
115 
116 #ifdef VIMAGE_GLOBALS
117 int	tcp_mssdflt;
118 #ifdef INET6
119 int	tcp_v6mssdflt;
120 #endif
121 int	tcp_minmss;
122 int	tcp_do_rfc1323;
123 static int	icmp_may_rst;
124 static int	tcp_isn_reseed_interval;
125 static int	tcp_inflight_enable;
126 static int	tcp_inflight_rttthresh;
127 static int	tcp_inflight_min;
128 static int	tcp_inflight_max;
129 static int	tcp_inflight_stab;
130 #endif
131 
132 static int
133 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
134 {
135 	INIT_VNET_INET(curvnet);
136 	int error, new;
137 
138 	new = V_tcp_mssdflt;
139 	error = sysctl_handle_int(oidp, &new, 0, req);
140 	if (error == 0 && req->newptr) {
141 		if (new < TCP_MINMSS)
142 			error = EINVAL;
143 		else
144 			V_tcp_mssdflt = new;
145 	}
146 	return (error);
147 }
148 
149 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
150     CTLTYPE_INT|CTLFLAG_RW, tcp_mssdflt, 0,
151     &sysctl_net_inet_tcp_mss_check, "I",
152     "Default TCP Maximum Segment Size");
153 
154 #ifdef INET6
155 static int
156 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
157 {
158 	INIT_VNET_INET(curvnet);
159 	int error, new;
160 
161 	new = V_tcp_v6mssdflt;
162 	error = sysctl_handle_int(oidp, &new, 0, req);
163 	if (error == 0 && req->newptr) {
164 		if (new < TCP_MINMSS)
165 			error = EINVAL;
166 		else
167 			V_tcp_v6mssdflt = new;
168 	}
169 	return (error);
170 }
171 
172 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
173     CTLTYPE_INT|CTLFLAG_RW, tcp_v6mssdflt, 0,
174     &sysctl_net_inet_tcp_mss_v6_check, "I",
175    "Default TCP Maximum Segment Size for IPv6");
176 #endif
177 
178 /*
179  * Minimum MSS we accept and use. This prevents DoS attacks where
180  * we are forced to a ridiculous low MSS like 20 and send hundreds
181  * of packets instead of one. The effect scales with the available
182  * bandwidth and quickly saturates the CPU and network interface
183  * with packet generation and sending. Set to zero to disable MINMSS
184  * checking. This setting prevents us from sending too small packets.
185  */
186 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
187     CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
188 
189 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
190     CTLFLAG_RW, tcp_do_rfc1323, 0,
191     "Enable rfc1323 (high performance TCP) extensions");
192 
193 static int	tcp_log_debug = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
195     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
196 
197 static int	tcp_tcbhashsize = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
199     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
200 
201 static int	do_tcpdrain = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
203     "Enable tcp_drain routine for extra help when low on mbufs");
204 
205 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
206     CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs");
207 
208 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
209     CTLFLAG_RW, icmp_may_rst, 0,
210     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
211 
212 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
213     CTLFLAG_RW, tcp_isn_reseed_interval, 0,
214     "Seconds between reseeding of ISN secret");
215 
216 /*
217  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
218  * 1024 exists only for debugging.  A good production default would be
219  * something like 6100.
220  */
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222     "TCP inflight data limiting");
223 
224 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
225     CTLFLAG_RW, tcp_inflight_enable, 0,
226     "Enable automatic TCP inflight data limiting");
227 
228 static int	tcp_inflight_debug = 0;
229 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
230     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
231 
232 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, rttthresh,
233     CTLTYPE_INT|CTLFLAG_RW, tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks,
234     "I", "RTT threshold below which inflight will deactivate itself");
235 
236 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
237     CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
238 
239 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
240     CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
241 
242 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
243     CTLFLAG_RW, tcp_inflight_stab, 0,
244     "Inflight Algorithm Stabilization 20 = 2 packets");
245 
246 #ifdef VIMAGE_GLOBALS
247 uma_zone_t sack_hole_zone;
248 #endif
249 
250 static struct inpcb *tcp_notify(struct inpcb *, int);
251 static void	tcp_isn_tick(void *);
252 
253 /*
254  * Target size of TCP PCB hash tables. Must be a power of two.
255  *
256  * Note that this can be overridden by the kernel environment
257  * variable net.inet.tcp.tcbhashsize
258  */
259 #ifndef TCBHASHSIZE
260 #define TCBHASHSIZE	512
261 #endif
262 
263 /*
264  * XXX
265  * Callouts should be moved into struct tcp directly.  They are currently
266  * separate because the tcpcb structure is exported to userland for sysctl
267  * parsing purposes, which do not know about callouts.
268  */
269 struct tcpcb_mem {
270 	struct	tcpcb		tcb;
271 	struct	tcp_timer	tt;
272 };
273 
274 #ifdef VIMAGE_GLOBALS
275 static uma_zone_t tcpcb_zone;
276 #endif
277 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
278 struct callout isn_callout;
279 static struct mtx isn_mtx;
280 
281 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
282 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
283 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
284 
285 /*
286  * TCP initialization.
287  */
288 static void
289 tcp_zone_change(void *tag)
290 {
291 	INIT_VNET_INET(curvnet);
292 
293 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
294 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
295 	tcp_tw_zone_change();
296 }
297 
298 static int
299 tcp_inpcb_init(void *mem, int size, int flags)
300 {
301 	struct inpcb *inp = mem;
302 
303 	INP_LOCK_INIT(inp, "inp", "tcpinp");
304 	return (0);
305 }
306 
307 void
308 tcp_init(void)
309 {
310 	INIT_VNET_INET(curvnet);
311 	int hashsize;
312 
313 	V_blackhole = 0;
314 	V_tcp_delack_enabled = 1;
315 	V_drop_synfin = 0;
316 	V_tcp_do_rfc3042 = 1;
317 	V_tcp_do_rfc3390 = 1;
318 	V_tcp_do_ecn = 0;
319 	V_tcp_ecn_maxretries = 1;
320 	V_tcp_insecure_rst = 0;
321 	V_tcp_do_autorcvbuf = 1;
322 	V_tcp_autorcvbuf_inc = 16*1024;
323 	V_tcp_autorcvbuf_max = 256*1024;
324 	V_tcp_do_rfc3465 = 1;
325 	V_tcp_abc_l_var = 2;
326 
327 	V_tcp_mssdflt = TCP_MSS;
328 #ifdef INET6
329 	V_tcp_v6mssdflt = TCP6_MSS;
330 #endif
331 	V_tcp_minmss = TCP_MINMSS;
332 	V_tcp_do_rfc1323 = 1;
333 	V_icmp_may_rst = 1;
334 	V_tcp_isn_reseed_interval = 0;
335 	V_tcp_inflight_enable = 1;
336 	V_tcp_inflight_min = 6144;
337 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
338 	V_tcp_inflight_stab = 20;
339 
340 	V_path_mtu_discovery = 1;
341 	V_ss_fltsz = 1;
342 	V_ss_fltsz_local = 4;
343 	V_tcp_do_newreno = 1;
344 	V_tcp_do_tso = 1;
345 	V_tcp_do_autosndbuf = 1;
346 	V_tcp_autosndbuf_inc = 8*1024;
347 	V_tcp_autosndbuf_max = 256*1024;
348 
349 	V_nolocaltimewait = 0;
350 
351 	V_tcp_do_sack = 1;
352 	V_tcp_sack_maxholes = 128;
353 	V_tcp_sack_globalmaxholes = 65536;
354 	V_tcp_sack_globalholes = 0;
355 
356 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
357 
358 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
359 
360 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
361 	LIST_INIT(&V_tcb);
362 #ifdef VIMAGE
363 	V_tcbinfo.ipi_vnet = curvnet;
364 #endif
365 	V_tcbinfo.ipi_listhead = &V_tcb;
366 	hashsize = TCBHASHSIZE;
367 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
368 	if (!powerof2(hashsize)) {
369 		printf("WARNING: TCB hash size not a power of 2\n");
370 		hashsize = 512; /* safe default */
371 	}
372 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
373 	    &V_tcbinfo.ipi_hashmask);
374 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
375 	    &V_tcbinfo.ipi_porthashmask);
376 	V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
377 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
378 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
379 	/*
380 	 * These have to be type stable for the benefit of the timers.
381 	 */
382 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
383 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
385 	tcp_tw_init();
386 	syncache_init();
387 	tcp_hc_init();
388 	tcp_reass_init();
389 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
390 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
391 
392 	/* Skip initialization of globals for non-default instances. */
393 	if (!IS_DEFAULT_VNET(curvnet))
394 		return;
395 
396 	/* XXX virtualize those bellow? */
397 	tcp_delacktime = TCPTV_DELACK;
398 	tcp_keepinit = TCPTV_KEEP_INIT;
399 	tcp_keepidle = TCPTV_KEEP_IDLE;
400 	tcp_keepintvl = TCPTV_KEEPINTVL;
401 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
402 	tcp_msl = TCPTV_MSL;
403 	tcp_rexmit_min = TCPTV_MIN;
404 	if (tcp_rexmit_min < 1)
405 		tcp_rexmit_min = 1;
406 	tcp_rexmit_slop = TCPTV_CPU_VAR;
407 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
408 	tcp_tcbhashsize = hashsize;
409 
410 #ifdef INET6
411 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
412 #else /* INET6 */
413 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
414 #endif /* INET6 */
415 	if (max_protohdr < TCP_MINPROTOHDR)
416 		max_protohdr = TCP_MINPROTOHDR;
417 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
418 		panic("tcp_init");
419 #undef TCP_MINPROTOHDR
420 
421 	ISN_LOCK_INIT();
422 	callout_init(&isn_callout, CALLOUT_MPSAFE);
423 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
424 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
425 		SHUTDOWN_PRI_DEFAULT);
426 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
427 		EVENTHANDLER_PRI_ANY);
428 }
429 
430 void
431 tcp_fini(void *xtp)
432 {
433 
434 	callout_stop(&isn_callout);
435 }
436 
437 /*
438  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
439  * tcp_template used to store this data in mbufs, but we now recopy it out
440  * of the tcpcb each time to conserve mbufs.
441  */
442 void
443 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
444 {
445 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
446 
447 	INP_WLOCK_ASSERT(inp);
448 
449 #ifdef INET6
450 	if ((inp->inp_vflag & INP_IPV6) != 0) {
451 		struct ip6_hdr *ip6;
452 
453 		ip6 = (struct ip6_hdr *)ip_ptr;
454 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
455 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
456 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
457 			(IPV6_VERSION & IPV6_VERSION_MASK);
458 		ip6->ip6_nxt = IPPROTO_TCP;
459 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
460 		ip6->ip6_src = inp->in6p_laddr;
461 		ip6->ip6_dst = inp->in6p_faddr;
462 	} else
463 #endif
464 	{
465 		struct ip *ip;
466 
467 		ip = (struct ip *)ip_ptr;
468 		ip->ip_v = IPVERSION;
469 		ip->ip_hl = 5;
470 		ip->ip_tos = inp->inp_ip_tos;
471 		ip->ip_len = 0;
472 		ip->ip_id = 0;
473 		ip->ip_off = 0;
474 		ip->ip_ttl = inp->inp_ip_ttl;
475 		ip->ip_sum = 0;
476 		ip->ip_p = IPPROTO_TCP;
477 		ip->ip_src = inp->inp_laddr;
478 		ip->ip_dst = inp->inp_faddr;
479 	}
480 	th->th_sport = inp->inp_lport;
481 	th->th_dport = inp->inp_fport;
482 	th->th_seq = 0;
483 	th->th_ack = 0;
484 	th->th_x2 = 0;
485 	th->th_off = 5;
486 	th->th_flags = 0;
487 	th->th_win = 0;
488 	th->th_urp = 0;
489 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
490 }
491 
492 /*
493  * Create template to be used to send tcp packets on a connection.
494  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
495  * use for this function is in keepalives, which use tcp_respond.
496  */
497 struct tcptemp *
498 tcpip_maketemplate(struct inpcb *inp)
499 {
500 	struct tcptemp *t;
501 
502 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
503 	if (t == NULL)
504 		return (NULL);
505 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
506 	return (t);
507 }
508 
509 /*
510  * Send a single message to the TCP at address specified by
511  * the given TCP/IP header.  If m == NULL, then we make a copy
512  * of the tcpiphdr at ti and send directly to the addressed host.
513  * This is used to force keep alive messages out using the TCP
514  * template for a connection.  If flags are given then we send
515  * a message back to the TCP which originated the * segment ti,
516  * and discard the mbuf containing it and any other attached mbufs.
517  *
518  * In any case the ack and sequence number of the transmitted
519  * segment are as specified by the parameters.
520  *
521  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
522  */
523 void
524 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
525     tcp_seq ack, tcp_seq seq, int flags)
526 {
527 	INIT_VNET_INET(curvnet);
528 	int tlen;
529 	int win = 0;
530 	struct ip *ip;
531 	struct tcphdr *nth;
532 #ifdef INET6
533 	struct ip6_hdr *ip6;
534 	int isipv6;
535 #endif /* INET6 */
536 	int ipflags = 0;
537 	struct inpcb *inp;
538 
539 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
540 
541 #ifdef INET6
542 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
543 	ip6 = ipgen;
544 #endif /* INET6 */
545 	ip = ipgen;
546 
547 	if (tp != NULL) {
548 		inp = tp->t_inpcb;
549 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
550 		INP_WLOCK_ASSERT(inp);
551 	} else
552 		inp = NULL;
553 
554 	if (tp != NULL) {
555 		if (!(flags & TH_RST)) {
556 			win = sbspace(&inp->inp_socket->so_rcv);
557 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
558 				win = (long)TCP_MAXWIN << tp->rcv_scale;
559 		}
560 	}
561 	if (m == NULL) {
562 		m = m_gethdr(M_DONTWAIT, MT_DATA);
563 		if (m == NULL)
564 			return;
565 		tlen = 0;
566 		m->m_data += max_linkhdr;
567 #ifdef INET6
568 		if (isipv6) {
569 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
570 			      sizeof(struct ip6_hdr));
571 			ip6 = mtod(m, struct ip6_hdr *);
572 			nth = (struct tcphdr *)(ip6 + 1);
573 		} else
574 #endif /* INET6 */
575 	      {
576 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
577 		ip = mtod(m, struct ip *);
578 		nth = (struct tcphdr *)(ip + 1);
579 	      }
580 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
581 		flags = TH_ACK;
582 	} else {
583 		/*
584 		 *  reuse the mbuf.
585 		 * XXX MRT We inherrit the FIB, which is lucky.
586 		 */
587 		m_freem(m->m_next);
588 		m->m_next = NULL;
589 		m->m_data = (caddr_t)ipgen;
590 		/* m_len is set later */
591 		tlen = 0;
592 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
593 #ifdef INET6
594 		if (isipv6) {
595 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
596 			nth = (struct tcphdr *)(ip6 + 1);
597 		} else
598 #endif /* INET6 */
599 	      {
600 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
601 		nth = (struct tcphdr *)(ip + 1);
602 	      }
603 		if (th != nth) {
604 			/*
605 			 * this is usually a case when an extension header
606 			 * exists between the IPv6 header and the
607 			 * TCP header.
608 			 */
609 			nth->th_sport = th->th_sport;
610 			nth->th_dport = th->th_dport;
611 		}
612 		xchg(nth->th_dport, nth->th_sport, uint16_t);
613 #undef xchg
614 	}
615 #ifdef INET6
616 	if (isipv6) {
617 		ip6->ip6_flow = 0;
618 		ip6->ip6_vfc = IPV6_VERSION;
619 		ip6->ip6_nxt = IPPROTO_TCP;
620 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
621 						tlen));
622 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
623 	} else
624 #endif
625 	{
626 		tlen += sizeof (struct tcpiphdr);
627 		ip->ip_len = tlen;
628 		ip->ip_ttl = V_ip_defttl;
629 		if (V_path_mtu_discovery)
630 			ip->ip_off |= IP_DF;
631 	}
632 	m->m_len = tlen;
633 	m->m_pkthdr.len = tlen;
634 	m->m_pkthdr.rcvif = NULL;
635 #ifdef MAC
636 	if (inp != NULL) {
637 		/*
638 		 * Packet is associated with a socket, so allow the
639 		 * label of the response to reflect the socket label.
640 		 */
641 		INP_WLOCK_ASSERT(inp);
642 		mac_inpcb_create_mbuf(inp, m);
643 	} else {
644 		/*
645 		 * Packet is not associated with a socket, so possibly
646 		 * update the label in place.
647 		 */
648 		mac_netinet_tcp_reply(m);
649 	}
650 #endif
651 	nth->th_seq = htonl(seq);
652 	nth->th_ack = htonl(ack);
653 	nth->th_x2 = 0;
654 	nth->th_off = sizeof (struct tcphdr) >> 2;
655 	nth->th_flags = flags;
656 	if (tp != NULL)
657 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
658 	else
659 		nth->th_win = htons((u_short)win);
660 	nth->th_urp = 0;
661 #ifdef INET6
662 	if (isipv6) {
663 		nth->th_sum = 0;
664 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
665 					sizeof(struct ip6_hdr),
666 					tlen - sizeof(struct ip6_hdr));
667 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
668 		    NULL, NULL);
669 	} else
670 #endif /* INET6 */
671 	{
672 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
673 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
674 		m->m_pkthdr.csum_flags = CSUM_TCP;
675 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
676 	}
677 #ifdef TCPDEBUG
678 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
679 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
680 #endif
681 #ifdef INET6
682 	if (isipv6)
683 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
684 	else
685 #endif /* INET6 */
686 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
687 }
688 
689 /*
690  * Create a new TCP control block, making an
691  * empty reassembly queue and hooking it to the argument
692  * protocol control block.  The `inp' parameter must have
693  * come from the zone allocator set up in tcp_init().
694  */
695 struct tcpcb *
696 tcp_newtcpcb(struct inpcb *inp)
697 {
698 	INIT_VNET_INET(inp->inp_vnet);
699 	struct tcpcb_mem *tm;
700 	struct tcpcb *tp;
701 #ifdef INET6
702 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
703 #endif /* INET6 */
704 
705 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
706 	if (tm == NULL)
707 		return (NULL);
708 	tp = &tm->tcb;
709 #ifdef VIMAGE
710 	tp->t_vnet = inp->inp_vnet;
711 #endif
712 	tp->t_timers = &tm->tt;
713 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
714 	tp->t_maxseg = tp->t_maxopd =
715 #ifdef INET6
716 		isipv6 ? V_tcp_v6mssdflt :
717 #endif /* INET6 */
718 		V_tcp_mssdflt;
719 
720 	/* Set up our timeouts. */
721 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
722 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
723 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
724 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
725 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
726 
727 	if (V_tcp_do_rfc1323)
728 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
729 	if (V_tcp_do_sack)
730 		tp->t_flags |= TF_SACK_PERMIT;
731 	TAILQ_INIT(&tp->snd_holes);
732 	tp->t_inpcb = inp;	/* XXX */
733 	/*
734 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
735 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
736 	 * reasonable initial retransmit time.
737 	 */
738 	tp->t_srtt = TCPTV_SRTTBASE;
739 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
740 	tp->t_rttmin = tcp_rexmit_min;
741 	tp->t_rxtcur = TCPTV_RTOBASE;
742 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
743 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
744 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
745 	tp->t_rcvtime = ticks;
746 	tp->t_bw_rtttime = ticks;
747 	/*
748 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
749 	 * because the socket may be bound to an IPv6 wildcard address,
750 	 * which may match an IPv4-mapped IPv6 address.
751 	 */
752 	inp->inp_ip_ttl = V_ip_defttl;
753 	inp->inp_ppcb = tp;
754 	return (tp);		/* XXX */
755 }
756 
757 /*
758  * Drop a TCP connection, reporting
759  * the specified error.  If connection is synchronized,
760  * then send a RST to peer.
761  */
762 struct tcpcb *
763 tcp_drop(struct tcpcb *tp, int errno)
764 {
765 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
766 	struct socket *so = tp->t_inpcb->inp_socket;
767 
768 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
769 	INP_WLOCK_ASSERT(tp->t_inpcb);
770 
771 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
772 		tp->t_state = TCPS_CLOSED;
773 		(void) tcp_output_reset(tp);
774 		TCPSTAT_INC(tcps_drops);
775 	} else
776 		TCPSTAT_INC(tcps_conndrops);
777 	if (errno == ETIMEDOUT && tp->t_softerror)
778 		errno = tp->t_softerror;
779 	so->so_error = errno;
780 	return (tcp_close(tp));
781 }
782 
783 void
784 tcp_discardcb(struct tcpcb *tp)
785 {
786 	INIT_VNET_INET(tp->t_vnet);
787 	struct tseg_qent *q;
788 	struct inpcb *inp = tp->t_inpcb;
789 	struct socket *so = inp->inp_socket;
790 #ifdef INET6
791 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
792 #endif /* INET6 */
793 
794 	INP_WLOCK_ASSERT(inp);
795 
796 	/*
797 	 * Make sure that all of our timers are stopped before we
798 	 * delete the PCB.
799 	 */
800 	callout_stop(&tp->t_timers->tt_rexmt);
801 	callout_stop(&tp->t_timers->tt_persist);
802 	callout_stop(&tp->t_timers->tt_keep);
803 	callout_stop(&tp->t_timers->tt_2msl);
804 	callout_stop(&tp->t_timers->tt_delack);
805 
806 	/*
807 	 * If we got enough samples through the srtt filter,
808 	 * save the rtt and rttvar in the routing entry.
809 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
810 	 * 4 samples is enough for the srtt filter to converge
811 	 * to within enough % of the correct value; fewer samples
812 	 * and we could save a bogus rtt. The danger is not high
813 	 * as tcp quickly recovers from everything.
814 	 * XXX: Works very well but needs some more statistics!
815 	 */
816 	if (tp->t_rttupdated >= 4) {
817 		struct hc_metrics_lite metrics;
818 		u_long ssthresh;
819 
820 		bzero(&metrics, sizeof(metrics));
821 		/*
822 		 * Update the ssthresh always when the conditions below
823 		 * are satisfied. This gives us better new start value
824 		 * for the congestion avoidance for new connections.
825 		 * ssthresh is only set if packet loss occured on a session.
826 		 *
827 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
828 		 * being torn down.  Ideally this code would not use 'so'.
829 		 */
830 		ssthresh = tp->snd_ssthresh;
831 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
832 			/*
833 			 * convert the limit from user data bytes to
834 			 * packets then to packet data bytes.
835 			 */
836 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
837 			if (ssthresh < 2)
838 				ssthresh = 2;
839 			ssthresh *= (u_long)(tp->t_maxseg +
840 #ifdef INET6
841 				      (isipv6 ? sizeof (struct ip6_hdr) +
842 					       sizeof (struct tcphdr) :
843 #endif
844 				       sizeof (struct tcpiphdr)
845 #ifdef INET6
846 				       )
847 #endif
848 				      );
849 		} else
850 			ssthresh = 0;
851 		metrics.rmx_ssthresh = ssthresh;
852 
853 		metrics.rmx_rtt = tp->t_srtt;
854 		metrics.rmx_rttvar = tp->t_rttvar;
855 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
856 		metrics.rmx_bandwidth = tp->snd_bandwidth;
857 		metrics.rmx_cwnd = tp->snd_cwnd;
858 		metrics.rmx_sendpipe = 0;
859 		metrics.rmx_recvpipe = 0;
860 
861 		tcp_hc_update(&inp->inp_inc, &metrics);
862 	}
863 
864 	/* free the reassembly queue, if any */
865 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
866 		LIST_REMOVE(q, tqe_q);
867 		m_freem(q->tqe_m);
868 		uma_zfree(V_tcp_reass_zone, q);
869 		tp->t_segqlen--;
870 		V_tcp_reass_qsize--;
871 	}
872 	/* Disconnect offload device, if any. */
873 	tcp_offload_detach(tp);
874 
875 	tcp_free_sackholes(tp);
876 	inp->inp_ppcb = NULL;
877 	tp->t_inpcb = NULL;
878 	uma_zfree(V_tcpcb_zone, tp);
879 }
880 
881 /*
882  * Attempt to close a TCP control block, marking it as dropped, and freeing
883  * the socket if we hold the only reference.
884  */
885 struct tcpcb *
886 tcp_close(struct tcpcb *tp)
887 {
888 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
889 	struct inpcb *inp = tp->t_inpcb;
890 	struct socket *so;
891 
892 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
893 	INP_WLOCK_ASSERT(inp);
894 
895 	/* Notify any offload devices of listener close */
896 	if (tp->t_state == TCPS_LISTEN)
897 		tcp_offload_listen_close(tp);
898 	in_pcbdrop(inp);
899 	TCPSTAT_INC(tcps_closed);
900 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
901 	so = inp->inp_socket;
902 	soisdisconnected(so);
903 	if (inp->inp_flags & INP_SOCKREF) {
904 		KASSERT(so->so_state & SS_PROTOREF,
905 		    ("tcp_close: !SS_PROTOREF"));
906 		inp->inp_flags &= ~INP_SOCKREF;
907 		INP_WUNLOCK(inp);
908 		ACCEPT_LOCK();
909 		SOCK_LOCK(so);
910 		so->so_state &= ~SS_PROTOREF;
911 		sofree(so);
912 		return (NULL);
913 	}
914 	return (tp);
915 }
916 
917 void
918 tcp_drain(void)
919 {
920 	VNET_ITERATOR_DECL(vnet_iter);
921 
922 	if (!do_tcpdrain)
923 		return;
924 
925 	VNET_LIST_RLOCK();
926 	VNET_FOREACH(vnet_iter) {
927 		CURVNET_SET(vnet_iter);
928 		INIT_VNET_INET(vnet_iter);
929 		struct inpcb *inpb;
930 		struct tcpcb *tcpb;
931 		struct tseg_qent *te;
932 
933 	/*
934 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
935 	 * if there is one...
936 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
937 	 *      reassembly queue should be flushed, but in a situation
938 	 *	where we're really low on mbufs, this is potentially
939 	 *	usefull.
940 	 */
941 		INP_INFO_RLOCK(&V_tcbinfo);
942 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
943 			if (inpb->inp_flags & INP_TIMEWAIT)
944 				continue;
945 			INP_WLOCK(inpb);
946 			if ((tcpb = intotcpcb(inpb)) != NULL) {
947 				while ((te = LIST_FIRST(&tcpb->t_segq))
948 			            != NULL) {
949 					LIST_REMOVE(te, tqe_q);
950 					m_freem(te->tqe_m);
951 					uma_zfree(V_tcp_reass_zone, te);
952 					tcpb->t_segqlen--;
953 					V_tcp_reass_qsize--;
954 				}
955 				tcp_clean_sackreport(tcpb);
956 			}
957 			INP_WUNLOCK(inpb);
958 		}
959 		INP_INFO_RUNLOCK(&V_tcbinfo);
960 		CURVNET_RESTORE();
961 	}
962 	VNET_LIST_RUNLOCK();
963 }
964 
965 /*
966  * Notify a tcp user of an asynchronous error;
967  * store error as soft error, but wake up user
968  * (for now, won't do anything until can select for soft error).
969  *
970  * Do not wake up user since there currently is no mechanism for
971  * reporting soft errors (yet - a kqueue filter may be added).
972  */
973 static struct inpcb *
974 tcp_notify(struct inpcb *inp, int error)
975 {
976 	struct tcpcb *tp;
977 #ifdef INVARIANTS
978 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
979 #endif
980 
981 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
982 	INP_WLOCK_ASSERT(inp);
983 
984 	if ((inp->inp_flags & INP_TIMEWAIT) ||
985 	    (inp->inp_flags & INP_DROPPED))
986 		return (inp);
987 
988 	tp = intotcpcb(inp);
989 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
990 
991 	/*
992 	 * Ignore some errors if we are hooked up.
993 	 * If connection hasn't completed, has retransmitted several times,
994 	 * and receives a second error, give up now.  This is better
995 	 * than waiting a long time to establish a connection that
996 	 * can never complete.
997 	 */
998 	if (tp->t_state == TCPS_ESTABLISHED &&
999 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1000 	     error == EHOSTDOWN)) {
1001 		return (inp);
1002 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1003 	    tp->t_softerror) {
1004 		tp = tcp_drop(tp, error);
1005 		if (tp != NULL)
1006 			return (inp);
1007 		else
1008 			return (NULL);
1009 	} else {
1010 		tp->t_softerror = error;
1011 		return (inp);
1012 	}
1013 #if 0
1014 	wakeup( &so->so_timeo);
1015 	sorwakeup(so);
1016 	sowwakeup(so);
1017 #endif
1018 }
1019 
1020 static int
1021 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1022 {
1023 	INIT_VNET_INET(curvnet);
1024 	int error, i, m, n, pcb_count;
1025 	struct inpcb *inp, **inp_list;
1026 	inp_gen_t gencnt;
1027 	struct xinpgen xig;
1028 
1029 	/*
1030 	 * The process of preparing the TCB list is too time-consuming and
1031 	 * resource-intensive to repeat twice on every request.
1032 	 */
1033 	if (req->oldptr == NULL) {
1034 		m = syncache_pcbcount();
1035 		n = V_tcbinfo.ipi_count;
1036 		req->oldidx = 2 * (sizeof xig)
1037 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1038 		return (0);
1039 	}
1040 
1041 	if (req->newptr != NULL)
1042 		return (EPERM);
1043 
1044 	/*
1045 	 * OK, now we're committed to doing something.
1046 	 */
1047 	INP_INFO_RLOCK(&V_tcbinfo);
1048 	gencnt = V_tcbinfo.ipi_gencnt;
1049 	n = V_tcbinfo.ipi_count;
1050 	INP_INFO_RUNLOCK(&V_tcbinfo);
1051 
1052 	m = syncache_pcbcount();
1053 
1054 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1055 		+ (n + m) * sizeof(struct xtcpcb));
1056 	if (error != 0)
1057 		return (error);
1058 
1059 	xig.xig_len = sizeof xig;
1060 	xig.xig_count = n + m;
1061 	xig.xig_gen = gencnt;
1062 	xig.xig_sogen = so_gencnt;
1063 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1064 	if (error)
1065 		return (error);
1066 
1067 	error = syncache_pcblist(req, m, &pcb_count);
1068 	if (error)
1069 		return (error);
1070 
1071 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1072 	if (inp_list == NULL)
1073 		return (ENOMEM);
1074 
1075 	INP_INFO_RLOCK(&V_tcbinfo);
1076 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1077 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1078 		INP_RLOCK(inp);
1079 		if (inp->inp_gencnt <= gencnt) {
1080 			/*
1081 			 * XXX: This use of cr_cansee(), introduced with
1082 			 * TCP state changes, is not quite right, but for
1083 			 * now, better than nothing.
1084 			 */
1085 			if (inp->inp_flags & INP_TIMEWAIT) {
1086 				if (intotw(inp) != NULL)
1087 					error = cr_cansee(req->td->td_ucred,
1088 					    intotw(inp)->tw_cred);
1089 				else
1090 					error = EINVAL;	/* Skip this inp. */
1091 			} else
1092 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1093 			if (error == 0)
1094 				inp_list[i++] = inp;
1095 		}
1096 		INP_RUNLOCK(inp);
1097 	}
1098 	INP_INFO_RUNLOCK(&V_tcbinfo);
1099 	n = i;
1100 
1101 	error = 0;
1102 	for (i = 0; i < n; i++) {
1103 		inp = inp_list[i];
1104 		INP_RLOCK(inp);
1105 		if (inp->inp_gencnt <= gencnt) {
1106 			struct xtcpcb xt;
1107 			void *inp_ppcb;
1108 
1109 			bzero(&xt, sizeof(xt));
1110 			xt.xt_len = sizeof xt;
1111 			/* XXX should avoid extra copy */
1112 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1113 			inp_ppcb = inp->inp_ppcb;
1114 			if (inp_ppcb == NULL)
1115 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1116 			else if (inp->inp_flags & INP_TIMEWAIT) {
1117 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1118 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1119 			} else
1120 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1121 			if (inp->inp_socket != NULL)
1122 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1123 			else {
1124 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1125 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1126 			}
1127 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1128 			INP_RUNLOCK(inp);
1129 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1130 		} else
1131 			INP_RUNLOCK(inp);
1132 
1133 	}
1134 	if (!error) {
1135 		/*
1136 		 * Give the user an updated idea of our state.
1137 		 * If the generation differs from what we told
1138 		 * her before, she knows that something happened
1139 		 * while we were processing this request, and it
1140 		 * might be necessary to retry.
1141 		 */
1142 		INP_INFO_RLOCK(&V_tcbinfo);
1143 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1144 		xig.xig_sogen = so_gencnt;
1145 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1146 		INP_INFO_RUNLOCK(&V_tcbinfo);
1147 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1148 	}
1149 	free(inp_list, M_TEMP);
1150 	return (error);
1151 }
1152 
1153 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1154     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1155 
1156 static int
1157 tcp_getcred(SYSCTL_HANDLER_ARGS)
1158 {
1159 	INIT_VNET_INET(curvnet);
1160 	struct xucred xuc;
1161 	struct sockaddr_in addrs[2];
1162 	struct inpcb *inp;
1163 	int error;
1164 
1165 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1166 	if (error)
1167 		return (error);
1168 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1169 	if (error)
1170 		return (error);
1171 	INP_INFO_RLOCK(&V_tcbinfo);
1172 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1173 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1174 	if (inp != NULL) {
1175 		INP_RLOCK(inp);
1176 		INP_INFO_RUNLOCK(&V_tcbinfo);
1177 		if (inp->inp_socket == NULL)
1178 			error = ENOENT;
1179 		if (error == 0)
1180 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1181 		if (error == 0)
1182 			cru2x(inp->inp_cred, &xuc);
1183 		INP_RUNLOCK(inp);
1184 	} else {
1185 		INP_INFO_RUNLOCK(&V_tcbinfo);
1186 		error = ENOENT;
1187 	}
1188 	if (error == 0)
1189 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1190 	return (error);
1191 }
1192 
1193 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1194     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1195     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1196 
1197 #ifdef INET6
1198 static int
1199 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1200 {
1201 	INIT_VNET_INET(curvnet);
1202 	INIT_VNET_INET6(curvnet);
1203 	struct xucred xuc;
1204 	struct sockaddr_in6 addrs[2];
1205 	struct inpcb *inp;
1206 	int error, mapped = 0;
1207 
1208 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1209 	if (error)
1210 		return (error);
1211 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1212 	if (error)
1213 		return (error);
1214 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1215 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1216 		return (error);
1217 	}
1218 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1219 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1220 			mapped = 1;
1221 		else
1222 			return (EINVAL);
1223 	}
1224 
1225 	INP_INFO_RLOCK(&V_tcbinfo);
1226 	if (mapped == 1)
1227 		inp = in_pcblookup_hash(&V_tcbinfo,
1228 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1229 			addrs[1].sin6_port,
1230 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1231 			addrs[0].sin6_port,
1232 			0, NULL);
1233 	else
1234 		inp = in6_pcblookup_hash(&V_tcbinfo,
1235 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1236 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1237 	if (inp != NULL) {
1238 		INP_RLOCK(inp);
1239 		INP_INFO_RUNLOCK(&V_tcbinfo);
1240 		if (inp->inp_socket == NULL)
1241 			error = ENOENT;
1242 		if (error == 0)
1243 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1244 		if (error == 0)
1245 			cru2x(inp->inp_cred, &xuc);
1246 		INP_RUNLOCK(inp);
1247 	} else {
1248 		INP_INFO_RUNLOCK(&V_tcbinfo);
1249 		error = ENOENT;
1250 	}
1251 	if (error == 0)
1252 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1253 	return (error);
1254 }
1255 
1256 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1257     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1258     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1259 #endif
1260 
1261 
1262 void
1263 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1264 {
1265 	INIT_VNET_INET(curvnet);
1266 	struct ip *ip = vip;
1267 	struct tcphdr *th;
1268 	struct in_addr faddr;
1269 	struct inpcb *inp;
1270 	struct tcpcb *tp;
1271 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1272 	struct icmp *icp;
1273 	struct in_conninfo inc;
1274 	tcp_seq icmp_tcp_seq;
1275 	int mtu;
1276 
1277 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1278 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1279 		return;
1280 
1281 	if (cmd == PRC_MSGSIZE)
1282 		notify = tcp_mtudisc;
1283 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1284 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1285 		notify = tcp_drop_syn_sent;
1286 	/*
1287 	 * Redirects don't need to be handled up here.
1288 	 */
1289 	else if (PRC_IS_REDIRECT(cmd))
1290 		return;
1291 	/*
1292 	 * Source quench is depreciated.
1293 	 */
1294 	else if (cmd == PRC_QUENCH)
1295 		return;
1296 	/*
1297 	 * Hostdead is ugly because it goes linearly through all PCBs.
1298 	 * XXX: We never get this from ICMP, otherwise it makes an
1299 	 * excellent DoS attack on machines with many connections.
1300 	 */
1301 	else if (cmd == PRC_HOSTDEAD)
1302 		ip = NULL;
1303 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1304 		return;
1305 	if (ip != NULL) {
1306 		icp = (struct icmp *)((caddr_t)ip
1307 				      - offsetof(struct icmp, icmp_ip));
1308 		th = (struct tcphdr *)((caddr_t)ip
1309 				       + (ip->ip_hl << 2));
1310 		INP_INFO_WLOCK(&V_tcbinfo);
1311 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1312 		    ip->ip_src, th->th_sport, 0, NULL);
1313 		if (inp != NULL)  {
1314 			INP_WLOCK(inp);
1315 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1316 			    !(inp->inp_flags & INP_DROPPED) &&
1317 			    !(inp->inp_socket == NULL)) {
1318 				icmp_tcp_seq = htonl(th->th_seq);
1319 				tp = intotcpcb(inp);
1320 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1321 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1322 					if (cmd == PRC_MSGSIZE) {
1323 					    /*
1324 					     * MTU discovery:
1325 					     * If we got a needfrag set the MTU
1326 					     * in the route to the suggested new
1327 					     * value (if given) and then notify.
1328 					     */
1329 					    bzero(&inc, sizeof(inc));
1330 					    inc.inc_faddr = faddr;
1331 					    inc.inc_fibnum =
1332 						inp->inp_inc.inc_fibnum;
1333 
1334 					    mtu = ntohs(icp->icmp_nextmtu);
1335 					    /*
1336 					     * If no alternative MTU was
1337 					     * proposed, try the next smaller
1338 					     * one.  ip->ip_len has already
1339 					     * been swapped in icmp_input().
1340 					     */
1341 					    if (!mtu)
1342 						mtu = ip_next_mtu(ip->ip_len,
1343 						 1);
1344 					    if (mtu < max(296, V_tcp_minmss
1345 						 + sizeof(struct tcpiphdr)))
1346 						mtu = 0;
1347 					    if (!mtu)
1348 						mtu = V_tcp_mssdflt
1349 						 + sizeof(struct tcpiphdr);
1350 					    /*
1351 					     * Only cache the the MTU if it
1352 					     * is smaller than the interface
1353 					     * or route MTU.  tcp_mtudisc()
1354 					     * will do right thing by itself.
1355 					     */
1356 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1357 						tcp_hc_updatemtu(&inc, mtu);
1358 					}
1359 
1360 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1361 				}
1362 			}
1363 			if (inp != NULL)
1364 				INP_WUNLOCK(inp);
1365 		} else {
1366 			bzero(&inc, sizeof(inc));
1367 			inc.inc_fport = th->th_dport;
1368 			inc.inc_lport = th->th_sport;
1369 			inc.inc_faddr = faddr;
1370 			inc.inc_laddr = ip->ip_src;
1371 			syncache_unreach(&inc, th);
1372 		}
1373 		INP_INFO_WUNLOCK(&V_tcbinfo);
1374 	} else
1375 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1376 }
1377 
1378 #ifdef INET6
1379 void
1380 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1381 {
1382 	INIT_VNET_INET(curvnet);
1383 	struct tcphdr th;
1384 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1385 	struct ip6_hdr *ip6;
1386 	struct mbuf *m;
1387 	struct ip6ctlparam *ip6cp = NULL;
1388 	const struct sockaddr_in6 *sa6_src = NULL;
1389 	int off;
1390 	struct tcp_portonly {
1391 		u_int16_t th_sport;
1392 		u_int16_t th_dport;
1393 	} *thp;
1394 
1395 	if (sa->sa_family != AF_INET6 ||
1396 	    sa->sa_len != sizeof(struct sockaddr_in6))
1397 		return;
1398 
1399 	if (cmd == PRC_MSGSIZE)
1400 		notify = tcp_mtudisc;
1401 	else if (!PRC_IS_REDIRECT(cmd) &&
1402 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1403 		return;
1404 	/* Source quench is depreciated. */
1405 	else if (cmd == PRC_QUENCH)
1406 		return;
1407 
1408 	/* if the parameter is from icmp6, decode it. */
1409 	if (d != NULL) {
1410 		ip6cp = (struct ip6ctlparam *)d;
1411 		m = ip6cp->ip6c_m;
1412 		ip6 = ip6cp->ip6c_ip6;
1413 		off = ip6cp->ip6c_off;
1414 		sa6_src = ip6cp->ip6c_src;
1415 	} else {
1416 		m = NULL;
1417 		ip6 = NULL;
1418 		off = 0;	/* fool gcc */
1419 		sa6_src = &sa6_any;
1420 	}
1421 
1422 	if (ip6 != NULL) {
1423 		struct in_conninfo inc;
1424 		/*
1425 		 * XXX: We assume that when IPV6 is non NULL,
1426 		 * M and OFF are valid.
1427 		 */
1428 
1429 		/* check if we can safely examine src and dst ports */
1430 		if (m->m_pkthdr.len < off + sizeof(*thp))
1431 			return;
1432 
1433 		bzero(&th, sizeof(th));
1434 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1435 
1436 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1437 		    (struct sockaddr *)ip6cp->ip6c_src,
1438 		    th.th_sport, cmd, NULL, notify);
1439 
1440 		bzero(&inc, sizeof(inc));
1441 		inc.inc_fport = th.th_dport;
1442 		inc.inc_lport = th.th_sport;
1443 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1444 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1445 		inc.inc_flags |= INC_ISIPV6;
1446 		INP_INFO_WLOCK(&V_tcbinfo);
1447 		syncache_unreach(&inc, &th);
1448 		INP_INFO_WUNLOCK(&V_tcbinfo);
1449 	} else
1450 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1451 			      0, cmd, NULL, notify);
1452 }
1453 #endif /* INET6 */
1454 
1455 
1456 /*
1457  * Following is where TCP initial sequence number generation occurs.
1458  *
1459  * There are two places where we must use initial sequence numbers:
1460  * 1.  In SYN-ACK packets.
1461  * 2.  In SYN packets.
1462  *
1463  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1464  * tcp_syncache.c for details.
1465  *
1466  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1467  * depends on this property.  In addition, these ISNs should be
1468  * unguessable so as to prevent connection hijacking.  To satisfy
1469  * the requirements of this situation, the algorithm outlined in
1470  * RFC 1948 is used, with only small modifications.
1471  *
1472  * Implementation details:
1473  *
1474  * Time is based off the system timer, and is corrected so that it
1475  * increases by one megabyte per second.  This allows for proper
1476  * recycling on high speed LANs while still leaving over an hour
1477  * before rollover.
1478  *
1479  * As reading the *exact* system time is too expensive to be done
1480  * whenever setting up a TCP connection, we increment the time
1481  * offset in two ways.  First, a small random positive increment
1482  * is added to isn_offset for each connection that is set up.
1483  * Second, the function tcp_isn_tick fires once per clock tick
1484  * and increments isn_offset as necessary so that sequence numbers
1485  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1486  * random positive increments serve only to ensure that the same
1487  * exact sequence number is never sent out twice (as could otherwise
1488  * happen when a port is recycled in less than the system tick
1489  * interval.)
1490  *
1491  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1492  * between seeding of isn_secret.  This is normally set to zero,
1493  * as reseeding should not be necessary.
1494  *
1495  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1496  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1497  * general, this means holding an exclusive (write) lock.
1498  */
1499 
1500 #define ISN_BYTES_PER_SECOND 1048576
1501 #define ISN_STATIC_INCREMENT 4096
1502 #define ISN_RANDOM_INCREMENT (4096 - 1)
1503 
1504 #ifdef VIMAGE_GLOBALS
1505 static u_char isn_secret[32];
1506 static int isn_last_reseed;
1507 static u_int32_t isn_offset, isn_offset_old;
1508 #endif
1509 
1510 tcp_seq
1511 tcp_new_isn(struct tcpcb *tp)
1512 {
1513 	INIT_VNET_INET(tp->t_vnet);
1514 	MD5_CTX isn_ctx;
1515 	u_int32_t md5_buffer[4];
1516 	tcp_seq new_isn;
1517 
1518 	INP_WLOCK_ASSERT(tp->t_inpcb);
1519 
1520 	ISN_LOCK();
1521 	/* Seed if this is the first use, reseed if requested. */
1522 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1523 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1524 		< (u_int)ticks))) {
1525 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1526 		V_isn_last_reseed = ticks;
1527 	}
1528 
1529 	/* Compute the md5 hash and return the ISN. */
1530 	MD5Init(&isn_ctx);
1531 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1532 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1533 #ifdef INET6
1534 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1535 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1536 			  sizeof(struct in6_addr));
1537 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1538 			  sizeof(struct in6_addr));
1539 	} else
1540 #endif
1541 	{
1542 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1543 			  sizeof(struct in_addr));
1544 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1545 			  sizeof(struct in_addr));
1546 	}
1547 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1548 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1549 	new_isn = (tcp_seq) md5_buffer[0];
1550 	V_isn_offset += ISN_STATIC_INCREMENT +
1551 		(arc4random() & ISN_RANDOM_INCREMENT);
1552 	new_isn += V_isn_offset;
1553 	ISN_UNLOCK();
1554 	return (new_isn);
1555 }
1556 
1557 /*
1558  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1559  * to keep time flowing at a relatively constant rate.  If the random
1560  * increments have already pushed us past the projected offset, do nothing.
1561  */
1562 static void
1563 tcp_isn_tick(void *xtp)
1564 {
1565 	VNET_ITERATOR_DECL(vnet_iter);
1566 	u_int32_t projected_offset;
1567 
1568 	VNET_LIST_RLOCK();
1569 	ISN_LOCK();
1570 	VNET_FOREACH(vnet_iter) {
1571 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1572 		INIT_VNET_INET(curvnet);
1573 		projected_offset =
1574 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1575 
1576 		if (SEQ_GT(projected_offset, V_isn_offset))
1577 			V_isn_offset = projected_offset;
1578 
1579 		V_isn_offset_old = V_isn_offset;
1580 		CURVNET_RESTORE();
1581 	}
1582 	ISN_UNLOCK();
1583 	VNET_LIST_RUNLOCK();
1584 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1585 }
1586 
1587 /*
1588  * When a specific ICMP unreachable message is received and the
1589  * connection state is SYN-SENT, drop the connection.  This behavior
1590  * is controlled by the icmp_may_rst sysctl.
1591  */
1592 struct inpcb *
1593 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1594 {
1595 #ifdef INVARIANTS
1596 	INIT_VNET_INET(inp->inp_vnet);
1597 #endif
1598 	struct tcpcb *tp;
1599 
1600 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1601 	INP_WLOCK_ASSERT(inp);
1602 
1603 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1604 	    (inp->inp_flags & INP_DROPPED))
1605 		return (inp);
1606 
1607 	tp = intotcpcb(inp);
1608 	if (tp->t_state != TCPS_SYN_SENT)
1609 		return (inp);
1610 
1611 	tp = tcp_drop(tp, errno);
1612 	if (tp != NULL)
1613 		return (inp);
1614 	else
1615 		return (NULL);
1616 }
1617 
1618 /*
1619  * When `need fragmentation' ICMP is received, update our idea of the MSS
1620  * based on the new value in the route.  Also nudge TCP to send something,
1621  * since we know the packet we just sent was dropped.
1622  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1623  */
1624 struct inpcb *
1625 tcp_mtudisc(struct inpcb *inp, int errno)
1626 {
1627 	INIT_VNET_INET(inp->inp_vnet);
1628 	struct tcpcb *tp;
1629 	struct socket *so;
1630 
1631 	INP_WLOCK_ASSERT(inp);
1632 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1633 	    (inp->inp_flags & INP_DROPPED))
1634 		return (inp);
1635 
1636 	tp = intotcpcb(inp);
1637 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1638 
1639 	tcp_mss_update(tp, -1, NULL, NULL);
1640 
1641 	so = inp->inp_socket;
1642 	SOCKBUF_LOCK(&so->so_snd);
1643 	/* If the mss is larger than the socket buffer, decrease the mss. */
1644 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1645 		tp->t_maxseg = so->so_snd.sb_hiwat;
1646 	SOCKBUF_UNLOCK(&so->so_snd);
1647 
1648 	TCPSTAT_INC(tcps_mturesent);
1649 	tp->t_rtttime = 0;
1650 	tp->snd_nxt = tp->snd_una;
1651 	tcp_free_sackholes(tp);
1652 	tp->snd_recover = tp->snd_max;
1653 	if (tp->t_flags & TF_SACK_PERMIT)
1654 		EXIT_FASTRECOVERY(tp);
1655 	tcp_output_send(tp);
1656 	return (inp);
1657 }
1658 
1659 /*
1660  * Look-up the routing entry to the peer of this inpcb.  If no route
1661  * is found and it cannot be allocated, then return 0.  This routine
1662  * is called by TCP routines that access the rmx structure and by
1663  * tcp_mss_update to get the peer/interface MTU.
1664  */
1665 u_long
1666 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1667 {
1668 	struct route sro;
1669 	struct sockaddr_in *dst;
1670 	struct ifnet *ifp;
1671 	u_long maxmtu = 0;
1672 
1673 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1674 
1675 	bzero(&sro, sizeof(sro));
1676 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1677 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1678 		dst->sin_family = AF_INET;
1679 		dst->sin_len = sizeof(*dst);
1680 		dst->sin_addr = inc->inc_faddr;
1681 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1682 	}
1683 	if (sro.ro_rt != NULL) {
1684 		ifp = sro.ro_rt->rt_ifp;
1685 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1686 			maxmtu = ifp->if_mtu;
1687 		else
1688 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1689 
1690 		/* Report additional interface capabilities. */
1691 		if (flags != NULL) {
1692 			if (ifp->if_capenable & IFCAP_TSO4 &&
1693 			    ifp->if_hwassist & CSUM_TSO)
1694 				*flags |= CSUM_TSO;
1695 		}
1696 		RTFREE(sro.ro_rt);
1697 	}
1698 	return (maxmtu);
1699 }
1700 
1701 #ifdef INET6
1702 u_long
1703 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1704 {
1705 	struct route_in6 sro6;
1706 	struct ifnet *ifp;
1707 	u_long maxmtu = 0;
1708 
1709 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1710 
1711 	bzero(&sro6, sizeof(sro6));
1712 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1713 		sro6.ro_dst.sin6_family = AF_INET6;
1714 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1715 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1716 		rtalloc_ign((struct route *)&sro6, 0);
1717 	}
1718 	if (sro6.ro_rt != NULL) {
1719 		ifp = sro6.ro_rt->rt_ifp;
1720 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1721 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1722 		else
1723 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1724 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1725 
1726 		/* Report additional interface capabilities. */
1727 		if (flags != NULL) {
1728 			if (ifp->if_capenable & IFCAP_TSO6 &&
1729 			    ifp->if_hwassist & CSUM_TSO)
1730 				*flags |= CSUM_TSO;
1731 		}
1732 		RTFREE(sro6.ro_rt);
1733 	}
1734 
1735 	return (maxmtu);
1736 }
1737 #endif /* INET6 */
1738 
1739 #ifdef IPSEC
1740 /* compute ESP/AH header size for TCP, including outer IP header. */
1741 size_t
1742 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1743 {
1744 	struct inpcb *inp;
1745 	struct mbuf *m;
1746 	size_t hdrsiz;
1747 	struct ip *ip;
1748 #ifdef INET6
1749 	struct ip6_hdr *ip6;
1750 #endif
1751 	struct tcphdr *th;
1752 
1753 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1754 		return (0);
1755 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1756 	if (!m)
1757 		return (0);
1758 
1759 #ifdef INET6
1760 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1761 		ip6 = mtod(m, struct ip6_hdr *);
1762 		th = (struct tcphdr *)(ip6 + 1);
1763 		m->m_pkthdr.len = m->m_len =
1764 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1765 		tcpip_fillheaders(inp, ip6, th);
1766 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1767 	} else
1768 #endif /* INET6 */
1769 	{
1770 		ip = mtod(m, struct ip *);
1771 		th = (struct tcphdr *)(ip + 1);
1772 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1773 		tcpip_fillheaders(inp, ip, th);
1774 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1775 	}
1776 
1777 	m_free(m);
1778 	return (hdrsiz);
1779 }
1780 #endif /* IPSEC */
1781 
1782 /*
1783  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1784  *
1785  * This code attempts to calculate the bandwidth-delay product as a
1786  * means of determining the optimal window size to maximize bandwidth,
1787  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1788  * routers.  This code also does a fairly good job keeping RTTs in check
1789  * across slow links like modems.  We implement an algorithm which is very
1790  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1791  * transmitter side of a TCP connection and so only effects the transmit
1792  * side of the connection.
1793  *
1794  * BACKGROUND:  TCP makes no provision for the management of buffer space
1795  * at the end points or at the intermediate routers and switches.  A TCP
1796  * stream, whether using NewReno or not, will eventually buffer as
1797  * many packets as it is able and the only reason this typically works is
1798  * due to the fairly small default buffers made available for a connection
1799  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1800  * scaling it is now fairly easy for even a single TCP connection to blow-out
1801  * all available buffer space not only on the local interface, but on
1802  * intermediate routers and switches as well.  NewReno makes a misguided
1803  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1804  * then backing off, then steadily increasing the window again until another
1805  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1806  * is only made worse as network loads increase and the idea of intentionally
1807  * blowing out network buffers is, frankly, a terrible way to manage network
1808  * resources.
1809  *
1810  * It is far better to limit the transmit window prior to the failure
1811  * condition being achieved.  There are two general ways to do this:  First
1812  * you can 'scan' through different transmit window sizes and locate the
1813  * point where the RTT stops increasing, indicating that you have filled the
1814  * pipe, then scan backwards until you note that RTT stops decreasing, then
1815  * repeat ad-infinitum.  This method works in principle but has severe
1816  * implementation issues due to RTT variances, timer granularity, and
1817  * instability in the algorithm which can lead to many false positives and
1818  * create oscillations as well as interact badly with other TCP streams
1819  * implementing the same algorithm.
1820  *
1821  * The second method is to limit the window to the bandwidth delay product
1822  * of the link.  This is the method we implement.  RTT variances and our
1823  * own manipulation of the congestion window, bwnd, can potentially
1824  * destabilize the algorithm.  For this reason we have to stabilize the
1825  * elements used to calculate the window.  We do this by using the minimum
1826  * observed RTT, the long term average of the observed bandwidth, and
1827  * by adding two segments worth of slop.  It isn't perfect but it is able
1828  * to react to changing conditions and gives us a very stable basis on
1829  * which to extend the algorithm.
1830  */
1831 void
1832 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1833 {
1834 	INIT_VNET_INET(tp->t_vnet);
1835 	u_long bw;
1836 	u_long bwnd;
1837 	int save_ticks;
1838 
1839 	INP_WLOCK_ASSERT(tp->t_inpcb);
1840 
1841 	/*
1842 	 * If inflight_enable is disabled in the middle of a tcp connection,
1843 	 * make sure snd_bwnd is effectively disabled.
1844 	 */
1845 	if (V_tcp_inflight_enable == 0 ||
1846 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1847 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1848 		tp->snd_bandwidth = 0;
1849 		return;
1850 	}
1851 
1852 	/*
1853 	 * Figure out the bandwidth.  Due to the tick granularity this
1854 	 * is a very rough number and it MUST be averaged over a fairly
1855 	 * long period of time.  XXX we need to take into account a link
1856 	 * that is not using all available bandwidth, but for now our
1857 	 * slop will ramp us up if this case occurs and the bandwidth later
1858 	 * increases.
1859 	 *
1860 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1861 	 * effectively reset t_bw_rtttime for this case.
1862 	 */
1863 	save_ticks = ticks;
1864 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1865 		return;
1866 
1867 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1868 	    (save_ticks - tp->t_bw_rtttime);
1869 	tp->t_bw_rtttime = save_ticks;
1870 	tp->t_bw_rtseq = ack_seq;
1871 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1872 		return;
1873 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1874 
1875 	tp->snd_bandwidth = bw;
1876 
1877 	/*
1878 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1879 	 * segments.  The additional slop puts us squarely in the sweet
1880 	 * spot and also handles the bandwidth run-up case and stabilization.
1881 	 * Without the slop we could be locking ourselves into a lower
1882 	 * bandwidth.
1883 	 *
1884 	 * Situations Handled:
1885 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1886 	 *	    high speed LANs, allowing larger TCP buffers to be
1887 	 *	    specified, and also does a good job preventing
1888 	 *	    over-queueing of packets over choke points like modems
1889 	 *	    (at least for the transmit side).
1890 	 *
1891 	 *	(2) Is able to handle changing network loads (bandwidth
1892 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1893 	 *	    increases).
1894 	 *
1895 	 *	(3) Theoretically should stabilize in the face of multiple
1896 	 *	    connections implementing the same algorithm (this may need
1897 	 *	    a little work).
1898 	 *
1899 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1900 	 *	    be adjusted with a sysctl but typically only needs to be
1901 	 *	    on very slow connections.  A value no smaller then 5
1902 	 *	    should be used, but only reduce this default if you have
1903 	 *	    no other choice.
1904 	 */
1905 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1906 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1907 #undef USERTT
1908 
1909 	if (tcp_inflight_debug > 0) {
1910 		static int ltime;
1911 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1912 			ltime = ticks;
1913 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1914 			    tp,
1915 			    bw,
1916 			    tp->t_rttbest,
1917 			    tp->t_srtt,
1918 			    bwnd
1919 			);
1920 		}
1921 	}
1922 	if ((long)bwnd < V_tcp_inflight_min)
1923 		bwnd = V_tcp_inflight_min;
1924 	if (bwnd > V_tcp_inflight_max)
1925 		bwnd = V_tcp_inflight_max;
1926 	if ((long)bwnd < tp->t_maxseg * 2)
1927 		bwnd = tp->t_maxseg * 2;
1928 	tp->snd_bwnd = bwnd;
1929 }
1930 
1931 #ifdef TCP_SIGNATURE
1932 /*
1933  * Callback function invoked by m_apply() to digest TCP segment data
1934  * contained within an mbuf chain.
1935  */
1936 static int
1937 tcp_signature_apply(void *fstate, void *data, u_int len)
1938 {
1939 
1940 	MD5Update(fstate, (u_char *)data, len);
1941 	return (0);
1942 }
1943 
1944 /*
1945  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1946  *
1947  * Parameters:
1948  * m		pointer to head of mbuf chain
1949  * _unused
1950  * len		length of TCP segment data, excluding options
1951  * optlen	length of TCP segment options
1952  * buf		pointer to storage for computed MD5 digest
1953  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1954  *
1955  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1956  * When called from tcp_input(), we can be sure that th_sum has been
1957  * zeroed out and verified already.
1958  *
1959  * Return 0 if successful, otherwise return -1.
1960  *
1961  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1962  * search with the destination IP address, and a 'magic SPI' to be
1963  * determined by the application. This is hardcoded elsewhere to 1179
1964  * right now. Another branch of this code exists which uses the SPD to
1965  * specify per-application flows but it is unstable.
1966  */
1967 int
1968 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1969     u_char *buf, u_int direction)
1970 {
1971 	INIT_VNET_IPSEC(curvnet);
1972 	union sockaddr_union dst;
1973 	struct ippseudo ippseudo;
1974 	MD5_CTX ctx;
1975 	int doff;
1976 	struct ip *ip;
1977 	struct ipovly *ipovly;
1978 	struct secasvar *sav;
1979 	struct tcphdr *th;
1980 #ifdef INET6
1981 	struct ip6_hdr *ip6;
1982 	struct in6_addr in6;
1983 	char ip6buf[INET6_ADDRSTRLEN];
1984 	uint32_t plen;
1985 	uint16_t nhdr;
1986 #endif
1987 	u_short savecsum;
1988 
1989 	KASSERT(m != NULL, ("NULL mbuf chain"));
1990 	KASSERT(buf != NULL, ("NULL signature pointer"));
1991 
1992 	/* Extract the destination from the IP header in the mbuf. */
1993 	bzero(&dst, sizeof(union sockaddr_union));
1994 	ip = mtod(m, struct ip *);
1995 #ifdef INET6
1996 	ip6 = NULL;	/* Make the compiler happy. */
1997 #endif
1998 	switch (ip->ip_v) {
1999 	case IPVERSION:
2000 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2001 		dst.sa.sa_family = AF_INET;
2002 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2003 		    ip->ip_src : ip->ip_dst;
2004 		break;
2005 #ifdef INET6
2006 	case (IPV6_VERSION >> 4):
2007 		ip6 = mtod(m, struct ip6_hdr *);
2008 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2009 		dst.sa.sa_family = AF_INET6;
2010 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2011 		    ip6->ip6_src : ip6->ip6_dst;
2012 		break;
2013 #endif
2014 	default:
2015 		return (EINVAL);
2016 		/* NOTREACHED */
2017 		break;
2018 	}
2019 
2020 	/* Look up an SADB entry which matches the address of the peer. */
2021 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2022 	if (sav == NULL) {
2023 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2024 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2025 #ifdef INET6
2026 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2027 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2028 #endif
2029 			"(unsupported)"));
2030 		return (EINVAL);
2031 	}
2032 
2033 	MD5Init(&ctx);
2034 	/*
2035 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2036 	 *
2037 	 * XXX The ippseudo header MUST be digested in network byte order,
2038 	 * or else we'll fail the regression test. Assume all fields we've
2039 	 * been doing arithmetic on have been in host byte order.
2040 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2041 	 * tcp_output(), the underlying ip_len member has not yet been set.
2042 	 */
2043 	switch (ip->ip_v) {
2044 	case IPVERSION:
2045 		ipovly = (struct ipovly *)ip;
2046 		ippseudo.ippseudo_src = ipovly->ih_src;
2047 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2048 		ippseudo.ippseudo_pad = 0;
2049 		ippseudo.ippseudo_p = IPPROTO_TCP;
2050 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2051 		    optlen);
2052 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2053 
2054 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2055 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2056 		break;
2057 #ifdef INET6
2058 	/*
2059 	 * RFC 2385, 2.0  Proposal
2060 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2061 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2062 	 * extended next header value (to form 32 bits), and 32-bit segment
2063 	 * length.
2064 	 * Note: Upper-Layer Packet Length comes before Next Header.
2065 	 */
2066 	case (IPV6_VERSION >> 4):
2067 		in6 = ip6->ip6_src;
2068 		in6_clearscope(&in6);
2069 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2070 		in6 = ip6->ip6_dst;
2071 		in6_clearscope(&in6);
2072 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2073 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2074 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2075 		nhdr = 0;
2076 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2077 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2078 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2079 		nhdr = IPPROTO_TCP;
2080 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2081 
2082 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2083 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2084 		break;
2085 #endif
2086 	default:
2087 		return (EINVAL);
2088 		/* NOTREACHED */
2089 		break;
2090 	}
2091 
2092 
2093 	/*
2094 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2095 	 * The TCP checksum must be set to zero.
2096 	 */
2097 	savecsum = th->th_sum;
2098 	th->th_sum = 0;
2099 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2100 	th->th_sum = savecsum;
2101 
2102 	/*
2103 	 * Step 3: Update MD5 hash with TCP segment data.
2104 	 *         Use m_apply() to avoid an early m_pullup().
2105 	 */
2106 	if (len > 0)
2107 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2108 
2109 	/*
2110 	 * Step 4: Update MD5 hash with shared secret.
2111 	 */
2112 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2113 	MD5Final(buf, &ctx);
2114 
2115 	key_sa_recordxfer(sav, m);
2116 	KEY_FREESAV(&sav);
2117 	return (0);
2118 }
2119 #endif /* TCP_SIGNATURE */
2120 
2121 static int
2122 sysctl_drop(SYSCTL_HANDLER_ARGS)
2123 {
2124 	INIT_VNET_INET(curvnet);
2125 #ifdef INET6
2126 	INIT_VNET_INET6(curvnet);
2127 #endif
2128 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2129 	struct sockaddr_storage addrs[2];
2130 	struct inpcb *inp;
2131 	struct tcpcb *tp;
2132 	struct tcptw *tw;
2133 	struct sockaddr_in *fin, *lin;
2134 #ifdef INET6
2135 	struct sockaddr_in6 *fin6, *lin6;
2136 #endif
2137 	int error;
2138 
2139 	inp = NULL;
2140 	fin = lin = NULL;
2141 #ifdef INET6
2142 	fin6 = lin6 = NULL;
2143 #endif
2144 	error = 0;
2145 
2146 	if (req->oldptr != NULL || req->oldlen != 0)
2147 		return (EINVAL);
2148 	if (req->newptr == NULL)
2149 		return (EPERM);
2150 	if (req->newlen < sizeof(addrs))
2151 		return (ENOMEM);
2152 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2153 	if (error)
2154 		return (error);
2155 
2156 	switch (addrs[0].ss_family) {
2157 #ifdef INET6
2158 	case AF_INET6:
2159 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2160 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2161 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2162 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2163 			return (EINVAL);
2164 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2165 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2166 				return (EINVAL);
2167 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2168 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2169 			fin = (struct sockaddr_in *)&addrs[0];
2170 			lin = (struct sockaddr_in *)&addrs[1];
2171 			break;
2172 		}
2173 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2174 		if (error)
2175 			return (error);
2176 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2177 		if (error)
2178 			return (error);
2179 		break;
2180 #endif
2181 	case AF_INET:
2182 		fin = (struct sockaddr_in *)&addrs[0];
2183 		lin = (struct sockaddr_in *)&addrs[1];
2184 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2185 		    lin->sin_len != sizeof(struct sockaddr_in))
2186 			return (EINVAL);
2187 		break;
2188 	default:
2189 		return (EINVAL);
2190 	}
2191 	INP_INFO_WLOCK(&V_tcbinfo);
2192 	switch (addrs[0].ss_family) {
2193 #ifdef INET6
2194 	case AF_INET6:
2195 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2196 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2197 		    NULL);
2198 		break;
2199 #endif
2200 	case AF_INET:
2201 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2202 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2203 		break;
2204 	}
2205 	if (inp != NULL) {
2206 		INP_WLOCK(inp);
2207 		if (inp->inp_flags & INP_TIMEWAIT) {
2208 			/*
2209 			 * XXXRW: There currently exists a state where an
2210 			 * inpcb is present, but its timewait state has been
2211 			 * discarded.  For now, don't allow dropping of this
2212 			 * type of inpcb.
2213 			 */
2214 			tw = intotw(inp);
2215 			if (tw != NULL)
2216 				tcp_twclose(tw, 0);
2217 			else
2218 				INP_WUNLOCK(inp);
2219 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2220 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2221 			tp = intotcpcb(inp);
2222 			tp = tcp_drop(tp, ECONNABORTED);
2223 			if (tp != NULL)
2224 				INP_WUNLOCK(inp);
2225 		} else
2226 			INP_WUNLOCK(inp);
2227 	} else
2228 		error = ESRCH;
2229 	INP_INFO_WUNLOCK(&V_tcbinfo);
2230 	return (error);
2231 }
2232 
2233 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2234     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2235     0, sysctl_drop, "", "Drop TCP connection");
2236 
2237 /*
2238  * Generate a standardized TCP log line for use throughout the
2239  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2240  * allow use in the interrupt context.
2241  *
2242  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2243  * NB: The function may return NULL if memory allocation failed.
2244  *
2245  * Due to header inclusion and ordering limitations the struct ip
2246  * and ip6_hdr pointers have to be passed as void pointers.
2247  */
2248 char *
2249 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2250     const void *ip6hdr)
2251 {
2252 	char *s, *sp;
2253 	size_t size;
2254 	struct ip *ip;
2255 #ifdef INET6
2256 	const struct ip6_hdr *ip6;
2257 
2258 	ip6 = (const struct ip6_hdr *)ip6hdr;
2259 #endif /* INET6 */
2260 	ip = (struct ip *)ip4hdr;
2261 
2262 	/*
2263 	 * The log line looks like this:
2264 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2265 	 */
2266 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2267 	    sizeof(PRINT_TH_FLAGS) + 1 +
2268 #ifdef INET6
2269 	    2 * INET6_ADDRSTRLEN;
2270 #else
2271 	    2 * INET_ADDRSTRLEN;
2272 #endif /* INET6 */
2273 
2274 	/* Is logging enabled? */
2275 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2276 		return (NULL);
2277 
2278 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2279 	if (s == NULL)
2280 		return (NULL);
2281 
2282 	strcat(s, "TCP: [");
2283 	sp = s + strlen(s);
2284 
2285 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2286 		inet_ntoa_r(inc->inc_faddr, sp);
2287 		sp = s + strlen(s);
2288 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2289 		sp = s + strlen(s);
2290 		inet_ntoa_r(inc->inc_laddr, sp);
2291 		sp = s + strlen(s);
2292 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2293 #ifdef INET6
2294 	} else if (inc) {
2295 		ip6_sprintf(sp, &inc->inc6_faddr);
2296 		sp = s + strlen(s);
2297 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2298 		sp = s + strlen(s);
2299 		ip6_sprintf(sp, &inc->inc6_laddr);
2300 		sp = s + strlen(s);
2301 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2302 	} else if (ip6 && th) {
2303 		ip6_sprintf(sp, &ip6->ip6_src);
2304 		sp = s + strlen(s);
2305 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2306 		sp = s + strlen(s);
2307 		ip6_sprintf(sp, &ip6->ip6_dst);
2308 		sp = s + strlen(s);
2309 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2310 #endif /* INET6 */
2311 	} else if (ip && th) {
2312 		inet_ntoa_r(ip->ip_src, sp);
2313 		sp = s + strlen(s);
2314 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2315 		sp = s + strlen(s);
2316 		inet_ntoa_r(ip->ip_dst, sp);
2317 		sp = s + strlen(s);
2318 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2319 	} else {
2320 		free(s, M_TCPLOG);
2321 		return (NULL);
2322 	}
2323 	sp = s + strlen(s);
2324 	if (th)
2325 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2326 	if (*(s + size - 1) != '\0')
2327 		panic("%s: string too long", __func__);
2328 	return (s);
2329 }
2330