xref: /freebsd/sys/netinet/tcp_subr.c (revision 9162f64b58d01ec01481d60b6cdc06ffd8e8c7fc)
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 uma_zone_t sack_hole_zone;
247 
248 static struct inpcb *tcp_notify(struct inpcb *, int);
249 static void	tcp_isn_tick(void *);
250 
251 /*
252  * Target size of TCP PCB hash tables. Must be a power of two.
253  *
254  * Note that this can be overridden by the kernel environment
255  * variable net.inet.tcp.tcbhashsize
256  */
257 #ifndef TCBHASHSIZE
258 #define TCBHASHSIZE	512
259 #endif
260 
261 /*
262  * XXX
263  * Callouts should be moved into struct tcp directly.  They are currently
264  * separate because the tcpcb structure is exported to userland for sysctl
265  * parsing purposes, which do not know about callouts.
266  */
267 struct tcpcb_mem {
268 	struct	tcpcb		tcb;
269 	struct	tcp_timer	tt;
270 };
271 
272 static uma_zone_t tcpcb_zone;
273 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
274 struct callout isn_callout;
275 static struct mtx isn_mtx;
276 
277 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
278 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
279 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
280 
281 /*
282  * TCP initialization.
283  */
284 static void
285 tcp_zone_change(void *tag)
286 {
287 
288 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
289 	uma_zone_set_max(tcpcb_zone, maxsockets);
290 	tcp_tw_zone_change();
291 }
292 
293 static int
294 tcp_inpcb_init(void *mem, int size, int flags)
295 {
296 	struct inpcb *inp = mem;
297 
298 	INP_LOCK_INIT(inp, "inp", "tcpinp");
299 	return (0);
300 }
301 
302 void
303 tcp_init(void)
304 {
305 	INIT_VNET_INET(curvnet);
306 	int hashsize;
307 
308 	V_blackhole = 0;
309 	V_tcp_delack_enabled = 1;
310 	V_drop_synfin = 0;
311 	V_tcp_do_rfc3042 = 1;
312 	V_tcp_do_rfc3390 = 1;
313 	V_tcp_do_ecn = 0;
314 	V_tcp_ecn_maxretries = 1;
315 	V_tcp_insecure_rst = 0;
316 	V_tcp_do_autorcvbuf = 1;
317 	V_tcp_autorcvbuf_inc = 16*1024;
318 	V_tcp_autorcvbuf_max = 256*1024;
319 
320 	V_tcp_mssdflt = TCP_MSS;
321 #ifdef INET6
322 	V_tcp_v6mssdflt = TCP6_MSS;
323 #endif
324 	V_tcp_minmss = TCP_MINMSS;
325 	V_tcp_do_rfc1323 = 1;
326 	V_icmp_may_rst = 1;
327 	V_tcp_isn_reseed_interval = 0;
328 	V_tcp_inflight_enable = 1;
329 	V_tcp_inflight_min = 6144;
330 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
331 	V_tcp_inflight_stab = 20;
332 
333 	V_path_mtu_discovery = 1;
334 	V_ss_fltsz = 1;
335 	V_ss_fltsz_local = 4;
336 	V_tcp_do_newreno = 1;
337 	V_tcp_do_tso = 1;
338 	V_tcp_do_autosndbuf = 1;
339 	V_tcp_autosndbuf_inc = 8*1024;
340 	V_tcp_autosndbuf_max = 256*1024;
341 
342 	V_nolocaltimewait = 0;
343 
344 	V_tcp_do_sack = 1;
345 	V_tcp_sack_maxholes = 128;
346 	V_tcp_sack_globalmaxholes = 65536;
347 	V_tcp_sack_globalholes = 0;
348 
349 	tcp_delacktime = TCPTV_DELACK;
350 	tcp_keepinit = TCPTV_KEEP_INIT;
351 	tcp_keepidle = TCPTV_KEEP_IDLE;
352 	tcp_keepintvl = TCPTV_KEEPINTVL;
353 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
354 	tcp_msl = TCPTV_MSL;
355 	tcp_rexmit_min = TCPTV_MIN;
356 	if (tcp_rexmit_min < 1)
357 		tcp_rexmit_min = 1;
358 	tcp_rexmit_slop = TCPTV_CPU_VAR;
359 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
360 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
361 
362 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
363 
364 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
365 	LIST_INIT(&V_tcb);
366 	V_tcbinfo.ipi_listhead = &V_tcb;
367 	hashsize = TCBHASHSIZE;
368 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
369 	if (!powerof2(hashsize)) {
370 		printf("WARNING: TCB hash size not a power of 2\n");
371 		hashsize = 512; /* safe default */
372 	}
373 	tcp_tcbhashsize = hashsize;
374 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
375 	    &V_tcbinfo.ipi_hashmask);
376 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
377 	    &V_tcbinfo.ipi_porthashmask);
378 	V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
379 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
380 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
381 #ifdef INET6
382 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
383 #else /* INET6 */
384 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
385 #endif /* INET6 */
386 	if (max_protohdr < TCP_MINPROTOHDR)
387 		max_protohdr = TCP_MINPROTOHDR;
388 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
389 		panic("tcp_init");
390 #undef TCP_MINPROTOHDR
391 	/*
392 	 * These have to be type stable for the benefit of the timers.
393 	 */
394 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
395 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
396 	uma_zone_set_max(tcpcb_zone, maxsockets);
397 	tcp_tw_init();
398 	syncache_init();
399 	tcp_hc_init();
400 	tcp_reass_init();
401 	ISN_LOCK_INIT();
402 	callout_init(&isn_callout, CALLOUT_MPSAFE);
403 	tcp_isn_tick(NULL);
404 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
405 		SHUTDOWN_PRI_DEFAULT);
406 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
407 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
408 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
409 		EVENTHANDLER_PRI_ANY);
410 }
411 
412 void
413 tcp_fini(void *xtp)
414 {
415 
416 	callout_stop(&isn_callout);
417 }
418 
419 /*
420  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
421  * tcp_template used to store this data in mbufs, but we now recopy it out
422  * of the tcpcb each time to conserve mbufs.
423  */
424 void
425 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
426 {
427 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
428 
429 	INP_WLOCK_ASSERT(inp);
430 
431 #ifdef INET6
432 	if ((inp->inp_vflag & INP_IPV6) != 0) {
433 		struct ip6_hdr *ip6;
434 
435 		ip6 = (struct ip6_hdr *)ip_ptr;
436 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
437 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
438 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
439 			(IPV6_VERSION & IPV6_VERSION_MASK);
440 		ip6->ip6_nxt = IPPROTO_TCP;
441 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
442 		ip6->ip6_src = inp->in6p_laddr;
443 		ip6->ip6_dst = inp->in6p_faddr;
444 	} else
445 #endif
446 	{
447 		struct ip *ip;
448 
449 		ip = (struct ip *)ip_ptr;
450 		ip->ip_v = IPVERSION;
451 		ip->ip_hl = 5;
452 		ip->ip_tos = inp->inp_ip_tos;
453 		ip->ip_len = 0;
454 		ip->ip_id = 0;
455 		ip->ip_off = 0;
456 		ip->ip_ttl = inp->inp_ip_ttl;
457 		ip->ip_sum = 0;
458 		ip->ip_p = IPPROTO_TCP;
459 		ip->ip_src = inp->inp_laddr;
460 		ip->ip_dst = inp->inp_faddr;
461 	}
462 	th->th_sport = inp->inp_lport;
463 	th->th_dport = inp->inp_fport;
464 	th->th_seq = 0;
465 	th->th_ack = 0;
466 	th->th_x2 = 0;
467 	th->th_off = 5;
468 	th->th_flags = 0;
469 	th->th_win = 0;
470 	th->th_urp = 0;
471 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
472 }
473 
474 /*
475  * Create template to be used to send tcp packets on a connection.
476  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
477  * use for this function is in keepalives, which use tcp_respond.
478  */
479 struct tcptemp *
480 tcpip_maketemplate(struct inpcb *inp)
481 {
482 	struct tcptemp *t;
483 
484 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
485 	if (t == NULL)
486 		return (NULL);
487 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
488 	return (t);
489 }
490 
491 /*
492  * Send a single message to the TCP at address specified by
493  * the given TCP/IP header.  If m == NULL, then we make a copy
494  * of the tcpiphdr at ti and send directly to the addressed host.
495  * This is used to force keep alive messages out using the TCP
496  * template for a connection.  If flags are given then we send
497  * a message back to the TCP which originated the * segment ti,
498  * and discard the mbuf containing it and any other attached mbufs.
499  *
500  * In any case the ack and sequence number of the transmitted
501  * segment are as specified by the parameters.
502  *
503  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
504  */
505 void
506 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
507     tcp_seq ack, tcp_seq seq, int flags)
508 {
509 	INIT_VNET_INET(curvnet);
510 	int tlen;
511 	int win = 0;
512 	struct ip *ip;
513 	struct tcphdr *nth;
514 #ifdef INET6
515 	struct ip6_hdr *ip6;
516 	int isipv6;
517 #endif /* INET6 */
518 	int ipflags = 0;
519 	struct inpcb *inp;
520 
521 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
522 
523 #ifdef INET6
524 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
525 	ip6 = ipgen;
526 #endif /* INET6 */
527 	ip = ipgen;
528 
529 	if (tp != NULL) {
530 		inp = tp->t_inpcb;
531 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
532 		INP_WLOCK_ASSERT(inp);
533 	} else
534 		inp = NULL;
535 
536 	if (tp != NULL) {
537 		if (!(flags & TH_RST)) {
538 			win = sbspace(&inp->inp_socket->so_rcv);
539 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
540 				win = (long)TCP_MAXWIN << tp->rcv_scale;
541 		}
542 	}
543 	if (m == NULL) {
544 		m = m_gethdr(M_DONTWAIT, MT_DATA);
545 		if (m == NULL)
546 			return;
547 		tlen = 0;
548 		m->m_data += max_linkhdr;
549 #ifdef INET6
550 		if (isipv6) {
551 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
552 			      sizeof(struct ip6_hdr));
553 			ip6 = mtod(m, struct ip6_hdr *);
554 			nth = (struct tcphdr *)(ip6 + 1);
555 		} else
556 #endif /* INET6 */
557 	      {
558 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
559 		ip = mtod(m, struct ip *);
560 		nth = (struct tcphdr *)(ip + 1);
561 	      }
562 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
563 		flags = TH_ACK;
564 	} else {
565 		/*
566 		 *  reuse the mbuf.
567 		 * XXX MRT We inherrit the FIB, which is lucky.
568 		 */
569 		m_freem(m->m_next);
570 		m->m_next = NULL;
571 		m->m_data = (caddr_t)ipgen;
572 		/* m_len is set later */
573 		tlen = 0;
574 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
575 #ifdef INET6
576 		if (isipv6) {
577 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
578 			nth = (struct tcphdr *)(ip6 + 1);
579 		} else
580 #endif /* INET6 */
581 	      {
582 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
583 		nth = (struct tcphdr *)(ip + 1);
584 	      }
585 		if (th != nth) {
586 			/*
587 			 * this is usually a case when an extension header
588 			 * exists between the IPv6 header and the
589 			 * TCP header.
590 			 */
591 			nth->th_sport = th->th_sport;
592 			nth->th_dport = th->th_dport;
593 		}
594 		xchg(nth->th_dport, nth->th_sport, n_short);
595 #undef xchg
596 	}
597 #ifdef INET6
598 	if (isipv6) {
599 		ip6->ip6_flow = 0;
600 		ip6->ip6_vfc = IPV6_VERSION;
601 		ip6->ip6_nxt = IPPROTO_TCP;
602 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
603 						tlen));
604 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
605 	} else
606 #endif
607 	{
608 		tlen += sizeof (struct tcpiphdr);
609 		ip->ip_len = tlen;
610 		ip->ip_ttl = V_ip_defttl;
611 		if (V_path_mtu_discovery)
612 			ip->ip_off |= IP_DF;
613 	}
614 	m->m_len = tlen;
615 	m->m_pkthdr.len = tlen;
616 	m->m_pkthdr.rcvif = NULL;
617 #ifdef MAC
618 	if (inp != NULL) {
619 		/*
620 		 * Packet is associated with a socket, so allow the
621 		 * label of the response to reflect the socket label.
622 		 */
623 		INP_WLOCK_ASSERT(inp);
624 		mac_inpcb_create_mbuf(inp, m);
625 	} else {
626 		/*
627 		 * Packet is not associated with a socket, so possibly
628 		 * update the label in place.
629 		 */
630 		mac_netinet_tcp_reply(m);
631 	}
632 #endif
633 	nth->th_seq = htonl(seq);
634 	nth->th_ack = htonl(ack);
635 	nth->th_x2 = 0;
636 	nth->th_off = sizeof (struct tcphdr) >> 2;
637 	nth->th_flags = flags;
638 	if (tp != NULL)
639 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
640 	else
641 		nth->th_win = htons((u_short)win);
642 	nth->th_urp = 0;
643 #ifdef INET6
644 	if (isipv6) {
645 		nth->th_sum = 0;
646 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
647 					sizeof(struct ip6_hdr),
648 					tlen - sizeof(struct ip6_hdr));
649 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
650 		    NULL, NULL);
651 	} else
652 #endif /* INET6 */
653 	{
654 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
655 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
656 		m->m_pkthdr.csum_flags = CSUM_TCP;
657 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
658 	}
659 #ifdef TCPDEBUG
660 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
661 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
662 #endif
663 #ifdef INET6
664 	if (isipv6)
665 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
666 	else
667 #endif /* INET6 */
668 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
669 }
670 
671 /*
672  * Create a new TCP control block, making an
673  * empty reassembly queue and hooking it to the argument
674  * protocol control block.  The `inp' parameter must have
675  * come from the zone allocator set up in tcp_init().
676  */
677 struct tcpcb *
678 tcp_newtcpcb(struct inpcb *inp)
679 {
680 	INIT_VNET_INET(inp->inp_vnet);
681 	struct tcpcb_mem *tm;
682 	struct tcpcb *tp;
683 #ifdef INET6
684 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
685 #endif /* INET6 */
686 
687 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
688 	if (tm == NULL)
689 		return (NULL);
690 	tp = &tm->tcb;
691 	tp->t_timers = &tm->tt;
692 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
693 	tp->t_maxseg = tp->t_maxopd =
694 #ifdef INET6
695 		isipv6 ? V_tcp_v6mssdflt :
696 #endif /* INET6 */
697 		V_tcp_mssdflt;
698 
699 	/* Set up our timeouts. */
700 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
701 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
702 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
703 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
704 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
705 
706 	if (V_tcp_do_rfc1323)
707 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
708 	if (V_tcp_do_sack)
709 		tp->t_flags |= TF_SACK_PERMIT;
710 	TAILQ_INIT(&tp->snd_holes);
711 	tp->t_inpcb = inp;	/* XXX */
712 	/*
713 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
714 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
715 	 * reasonable initial retransmit time.
716 	 */
717 	tp->t_srtt = TCPTV_SRTTBASE;
718 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
719 	tp->t_rttmin = tcp_rexmit_min;
720 	tp->t_rxtcur = TCPTV_RTOBASE;
721 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
722 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
723 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
724 	tp->t_rcvtime = ticks;
725 	tp->t_bw_rtttime = ticks;
726 	/*
727 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
728 	 * because the socket may be bound to an IPv6 wildcard address,
729 	 * which may match an IPv4-mapped IPv6 address.
730 	 */
731 	inp->inp_ip_ttl = V_ip_defttl;
732 	inp->inp_ppcb = tp;
733 	return (tp);		/* XXX */
734 }
735 
736 /*
737  * Drop a TCP connection, reporting
738  * the specified error.  If connection is synchronized,
739  * then send a RST to peer.
740  */
741 struct tcpcb *
742 tcp_drop(struct tcpcb *tp, int errno)
743 {
744 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
745 	struct socket *so = tp->t_inpcb->inp_socket;
746 
747 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
748 	INP_WLOCK_ASSERT(tp->t_inpcb);
749 
750 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
751 		tp->t_state = TCPS_CLOSED;
752 		(void) tcp_output_reset(tp);
753 		V_tcpstat.tcps_drops++;
754 	} else
755 		V_tcpstat.tcps_conndrops++;
756 	if (errno == ETIMEDOUT && tp->t_softerror)
757 		errno = tp->t_softerror;
758 	so->so_error = errno;
759 	return (tcp_close(tp));
760 }
761 
762 void
763 tcp_discardcb(struct tcpcb *tp)
764 {
765 	INIT_VNET_INET(tp->t_vnet);
766 	struct tseg_qent *q;
767 	struct inpcb *inp = tp->t_inpcb;
768 	struct socket *so = inp->inp_socket;
769 #ifdef INET6
770 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
771 #endif /* INET6 */
772 
773 	INP_WLOCK_ASSERT(inp);
774 
775 	/*
776 	 * Make sure that all of our timers are stopped before we
777 	 * delete the PCB.
778 	 */
779 	callout_stop(&tp->t_timers->tt_rexmt);
780 	callout_stop(&tp->t_timers->tt_persist);
781 	callout_stop(&tp->t_timers->tt_keep);
782 	callout_stop(&tp->t_timers->tt_2msl);
783 	callout_stop(&tp->t_timers->tt_delack);
784 
785 	/*
786 	 * If we got enough samples through the srtt filter,
787 	 * save the rtt and rttvar in the routing entry.
788 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
789 	 * 4 samples is enough for the srtt filter to converge
790 	 * to within enough % of the correct value; fewer samples
791 	 * and we could save a bogus rtt. The danger is not high
792 	 * as tcp quickly recovers from everything.
793 	 * XXX: Works very well but needs some more statistics!
794 	 */
795 	if (tp->t_rttupdated >= 4) {
796 		struct hc_metrics_lite metrics;
797 		u_long ssthresh;
798 
799 		bzero(&metrics, sizeof(metrics));
800 		/*
801 		 * Update the ssthresh always when the conditions below
802 		 * are satisfied. This gives us better new start value
803 		 * for the congestion avoidance for new connections.
804 		 * ssthresh is only set if packet loss occured on a session.
805 		 *
806 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
807 		 * being torn down.  Ideally this code would not use 'so'.
808 		 */
809 		ssthresh = tp->snd_ssthresh;
810 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
811 			/*
812 			 * convert the limit from user data bytes to
813 			 * packets then to packet data bytes.
814 			 */
815 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
816 			if (ssthresh < 2)
817 				ssthresh = 2;
818 			ssthresh *= (u_long)(tp->t_maxseg +
819 #ifdef INET6
820 				      (isipv6 ? sizeof (struct ip6_hdr) +
821 					       sizeof (struct tcphdr) :
822 #endif
823 				       sizeof (struct tcpiphdr)
824 #ifdef INET6
825 				       )
826 #endif
827 				      );
828 		} else
829 			ssthresh = 0;
830 		metrics.rmx_ssthresh = ssthresh;
831 
832 		metrics.rmx_rtt = tp->t_srtt;
833 		metrics.rmx_rttvar = tp->t_rttvar;
834 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
835 		metrics.rmx_bandwidth = tp->snd_bandwidth;
836 		metrics.rmx_cwnd = tp->snd_cwnd;
837 		metrics.rmx_sendpipe = 0;
838 		metrics.rmx_recvpipe = 0;
839 
840 		tcp_hc_update(&inp->inp_inc, &metrics);
841 	}
842 
843 	/* free the reassembly queue, if any */
844 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
845 		LIST_REMOVE(q, tqe_q);
846 		m_freem(q->tqe_m);
847 		uma_zfree(tcp_reass_zone, q);
848 		tp->t_segqlen--;
849 		V_tcp_reass_qsize--;
850 	}
851 	/* Disconnect offload device, if any. */
852 	tcp_offload_detach(tp);
853 
854 	tcp_free_sackholes(tp);
855 	inp->inp_ppcb = NULL;
856 	tp->t_inpcb = NULL;
857 	uma_zfree(tcpcb_zone, tp);
858 }
859 
860 /*
861  * Attempt to close a TCP control block, marking it as dropped, and freeing
862  * the socket if we hold the only reference.
863  */
864 struct tcpcb *
865 tcp_close(struct tcpcb *tp)
866 {
867 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
868 	struct inpcb *inp = tp->t_inpcb;
869 	struct socket *so;
870 
871 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
872 	INP_WLOCK_ASSERT(inp);
873 
874 	/* Notify any offload devices of listener close */
875 	if (tp->t_state == TCPS_LISTEN)
876 		tcp_offload_listen_close(tp);
877 	in_pcbdrop(inp);
878 	V_tcpstat.tcps_closed++;
879 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
880 	so = inp->inp_socket;
881 	soisdisconnected(so);
882 	if (inp->inp_vflag & INP_SOCKREF) {
883 		KASSERT(so->so_state & SS_PROTOREF,
884 		    ("tcp_close: !SS_PROTOREF"));
885 		inp->inp_vflag &= ~INP_SOCKREF;
886 		INP_WUNLOCK(inp);
887 		ACCEPT_LOCK();
888 		SOCK_LOCK(so);
889 		so->so_state &= ~SS_PROTOREF;
890 		sofree(so);
891 		return (NULL);
892 	}
893 	return (tp);
894 }
895 
896 void
897 tcp_drain(void)
898 {
899 	VNET_ITERATOR_DECL(vnet_iter);
900 
901 	if (!do_tcpdrain)
902 		return;
903 
904 	VNET_LIST_RLOCK();
905 	VNET_FOREACH(vnet_iter) {
906 		CURVNET_SET(vnet_iter);
907 		INIT_VNET_INET(vnet_iter);
908 		struct inpcb *inpb;
909 		struct tcpcb *tcpb;
910 		struct tseg_qent *te;
911 
912 	/*
913 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
914 	 * if there is one...
915 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
916 	 *      reassembly queue should be flushed, but in a situation
917 	 *	where we're really low on mbufs, this is potentially
918 	 *	usefull.
919 	 */
920 		INP_INFO_RLOCK(&V_tcbinfo);
921 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
922 			if (inpb->inp_vflag & INP_TIMEWAIT)
923 				continue;
924 			INP_WLOCK(inpb);
925 			if ((tcpb = intotcpcb(inpb)) != NULL) {
926 				while ((te = LIST_FIRST(&tcpb->t_segq))
927 			            != NULL) {
928 					LIST_REMOVE(te, tqe_q);
929 					m_freem(te->tqe_m);
930 					uma_zfree(tcp_reass_zone, te);
931 					tcpb->t_segqlen--;
932 					V_tcp_reass_qsize--;
933 				}
934 				tcp_clean_sackreport(tcpb);
935 			}
936 			INP_WUNLOCK(inpb);
937 		}
938 		INP_INFO_RUNLOCK(&V_tcbinfo);
939 		CURVNET_RESTORE();
940 	}
941 	VNET_LIST_RUNLOCK();
942 }
943 
944 /*
945  * Notify a tcp user of an asynchronous error;
946  * store error as soft error, but wake up user
947  * (for now, won't do anything until can select for soft error).
948  *
949  * Do not wake up user since there currently is no mechanism for
950  * reporting soft errors (yet - a kqueue filter may be added).
951  */
952 static struct inpcb *
953 tcp_notify(struct inpcb *inp, int error)
954 {
955 	struct tcpcb *tp;
956 #ifdef INVARIANTS
957 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
958 #endif
959 
960 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
961 	INP_WLOCK_ASSERT(inp);
962 
963 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
964 	    (inp->inp_vflag & INP_DROPPED))
965 		return (inp);
966 
967 	tp = intotcpcb(inp);
968 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
969 
970 	/*
971 	 * Ignore some errors if we are hooked up.
972 	 * If connection hasn't completed, has retransmitted several times,
973 	 * and receives a second error, give up now.  This is better
974 	 * than waiting a long time to establish a connection that
975 	 * can never complete.
976 	 */
977 	if (tp->t_state == TCPS_ESTABLISHED &&
978 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
979 	     error == EHOSTDOWN)) {
980 		return (inp);
981 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
982 	    tp->t_softerror) {
983 		tp = tcp_drop(tp, error);
984 		if (tp != NULL)
985 			return (inp);
986 		else
987 			return (NULL);
988 	} else {
989 		tp->t_softerror = error;
990 		return (inp);
991 	}
992 #if 0
993 	wakeup( &so->so_timeo);
994 	sorwakeup(so);
995 	sowwakeup(so);
996 #endif
997 }
998 
999 static int
1000 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1001 {
1002 	INIT_VNET_INET(curvnet);
1003 	int error, i, m, n, pcb_count;
1004 	struct inpcb *inp, **inp_list;
1005 	inp_gen_t gencnt;
1006 	struct xinpgen xig;
1007 
1008 	/*
1009 	 * The process of preparing the TCB list is too time-consuming and
1010 	 * resource-intensive to repeat twice on every request.
1011 	 */
1012 	if (req->oldptr == NULL) {
1013 		m = syncache_pcbcount();
1014 		n = V_tcbinfo.ipi_count;
1015 		req->oldidx = 2 * (sizeof xig)
1016 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1017 		return (0);
1018 	}
1019 
1020 	if (req->newptr != NULL)
1021 		return (EPERM);
1022 
1023 	/*
1024 	 * OK, now we're committed to doing something.
1025 	 */
1026 	INP_INFO_RLOCK(&V_tcbinfo);
1027 	gencnt = V_tcbinfo.ipi_gencnt;
1028 	n = V_tcbinfo.ipi_count;
1029 	INP_INFO_RUNLOCK(&V_tcbinfo);
1030 
1031 	m = syncache_pcbcount();
1032 
1033 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1034 		+ (n + m) * sizeof(struct xtcpcb));
1035 	if (error != 0)
1036 		return (error);
1037 
1038 	xig.xig_len = sizeof xig;
1039 	xig.xig_count = n + m;
1040 	xig.xig_gen = gencnt;
1041 	xig.xig_sogen = so_gencnt;
1042 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1043 	if (error)
1044 		return (error);
1045 
1046 	error = syncache_pcblist(req, m, &pcb_count);
1047 	if (error)
1048 		return (error);
1049 
1050 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1051 	if (inp_list == NULL)
1052 		return (ENOMEM);
1053 
1054 	INP_INFO_RLOCK(&V_tcbinfo);
1055 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1056 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1057 		INP_RLOCK(inp);
1058 		if (inp->inp_gencnt <= gencnt) {
1059 			/*
1060 			 * XXX: This use of cr_cansee(), introduced with
1061 			 * TCP state changes, is not quite right, but for
1062 			 * now, better than nothing.
1063 			 */
1064 			if (inp->inp_vflag & INP_TIMEWAIT) {
1065 				if (intotw(inp) != NULL)
1066 					error = cr_cansee(req->td->td_ucred,
1067 					    intotw(inp)->tw_cred);
1068 				else
1069 					error = EINVAL;	/* Skip this inp. */
1070 			} else
1071 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1072 			if (error == 0)
1073 				inp_list[i++] = inp;
1074 		}
1075 		INP_RUNLOCK(inp);
1076 	}
1077 	INP_INFO_RUNLOCK(&V_tcbinfo);
1078 	n = i;
1079 
1080 	error = 0;
1081 	for (i = 0; i < n; i++) {
1082 		inp = inp_list[i];
1083 		INP_RLOCK(inp);
1084 		if (inp->inp_gencnt <= gencnt) {
1085 			struct xtcpcb xt;
1086 			void *inp_ppcb;
1087 
1088 			bzero(&xt, sizeof(xt));
1089 			xt.xt_len = sizeof xt;
1090 			/* XXX should avoid extra copy */
1091 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1092 			inp_ppcb = inp->inp_ppcb;
1093 			if (inp_ppcb == NULL)
1094 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1095 			else if (inp->inp_vflag & INP_TIMEWAIT) {
1096 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1097 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1098 			} else
1099 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1100 			if (inp->inp_socket != NULL)
1101 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1102 			else {
1103 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1104 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1105 			}
1106 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1107 			INP_RUNLOCK(inp);
1108 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1109 		} else
1110 			INP_RUNLOCK(inp);
1111 
1112 	}
1113 	if (!error) {
1114 		/*
1115 		 * Give the user an updated idea of our state.
1116 		 * If the generation differs from what we told
1117 		 * her before, she knows that something happened
1118 		 * while we were processing this request, and it
1119 		 * might be necessary to retry.
1120 		 */
1121 		INP_INFO_RLOCK(&V_tcbinfo);
1122 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1123 		xig.xig_sogen = so_gencnt;
1124 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1125 		INP_INFO_RUNLOCK(&V_tcbinfo);
1126 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1127 	}
1128 	free(inp_list, M_TEMP);
1129 	return (error);
1130 }
1131 
1132 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1133     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1134 
1135 static int
1136 tcp_getcred(SYSCTL_HANDLER_ARGS)
1137 {
1138 	INIT_VNET_INET(curvnet);
1139 	struct xucred xuc;
1140 	struct sockaddr_in addrs[2];
1141 	struct inpcb *inp;
1142 	int error;
1143 
1144 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1145 	if (error)
1146 		return (error);
1147 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1148 	if (error)
1149 		return (error);
1150 	INP_INFO_RLOCK(&V_tcbinfo);
1151 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1152 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1153 	if (inp != NULL) {
1154 		INP_RLOCK(inp);
1155 		INP_INFO_RUNLOCK(&V_tcbinfo);
1156 		if (inp->inp_socket == NULL)
1157 			error = ENOENT;
1158 		if (error == 0)
1159 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1160 		if (error == 0)
1161 			cru2x(inp->inp_cred, &xuc);
1162 		INP_RUNLOCK(inp);
1163 	} else {
1164 		INP_INFO_RUNLOCK(&V_tcbinfo);
1165 		error = ENOENT;
1166 	}
1167 	if (error == 0)
1168 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1169 	return (error);
1170 }
1171 
1172 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1173     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1174     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1175 
1176 #ifdef INET6
1177 static int
1178 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1179 {
1180 	INIT_VNET_INET(curvnet);
1181 	INIT_VNET_INET6(curvnet);
1182 	struct xucred xuc;
1183 	struct sockaddr_in6 addrs[2];
1184 	struct inpcb *inp;
1185 	int error, mapped = 0;
1186 
1187 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1188 	if (error)
1189 		return (error);
1190 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1191 	if (error)
1192 		return (error);
1193 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1194 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1195 		return (error);
1196 	}
1197 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1198 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1199 			mapped = 1;
1200 		else
1201 			return (EINVAL);
1202 	}
1203 
1204 	INP_INFO_RLOCK(&V_tcbinfo);
1205 	if (mapped == 1)
1206 		inp = in_pcblookup_hash(&V_tcbinfo,
1207 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1208 			addrs[1].sin6_port,
1209 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1210 			addrs[0].sin6_port,
1211 			0, NULL);
1212 	else
1213 		inp = in6_pcblookup_hash(&V_tcbinfo,
1214 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1215 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1216 	if (inp != NULL) {
1217 		INP_RLOCK(inp);
1218 		INP_INFO_RUNLOCK(&V_tcbinfo);
1219 		if (inp->inp_socket == NULL)
1220 			error = ENOENT;
1221 		if (error == 0)
1222 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1223 		if (error == 0)
1224 			cru2x(inp->inp_cred, &xuc);
1225 		INP_RUNLOCK(inp);
1226 	} else {
1227 		INP_INFO_RUNLOCK(&V_tcbinfo);
1228 		error = ENOENT;
1229 	}
1230 	if (error == 0)
1231 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1232 	return (error);
1233 }
1234 
1235 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1236     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1237     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1238 #endif
1239 
1240 
1241 void
1242 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1243 {
1244 	INIT_VNET_INET(curvnet);
1245 	struct ip *ip = vip;
1246 	struct tcphdr *th;
1247 	struct in_addr faddr;
1248 	struct inpcb *inp;
1249 	struct tcpcb *tp;
1250 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1251 	struct icmp *icp;
1252 	struct in_conninfo inc;
1253 	tcp_seq icmp_tcp_seq;
1254 	int mtu;
1255 
1256 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1257 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1258 		return;
1259 
1260 	if (cmd == PRC_MSGSIZE)
1261 		notify = tcp_mtudisc;
1262 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1263 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1264 		notify = tcp_drop_syn_sent;
1265 	/*
1266 	 * Redirects don't need to be handled up here.
1267 	 */
1268 	else if (PRC_IS_REDIRECT(cmd))
1269 		return;
1270 	/*
1271 	 * Source quench is depreciated.
1272 	 */
1273 	else if (cmd == PRC_QUENCH)
1274 		return;
1275 	/*
1276 	 * Hostdead is ugly because it goes linearly through all PCBs.
1277 	 * XXX: We never get this from ICMP, otherwise it makes an
1278 	 * excellent DoS attack on machines with many connections.
1279 	 */
1280 	else if (cmd == PRC_HOSTDEAD)
1281 		ip = NULL;
1282 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1283 		return;
1284 	if (ip != NULL) {
1285 		icp = (struct icmp *)((caddr_t)ip
1286 				      - offsetof(struct icmp, icmp_ip));
1287 		th = (struct tcphdr *)((caddr_t)ip
1288 				       + (ip->ip_hl << 2));
1289 		INP_INFO_WLOCK(&V_tcbinfo);
1290 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1291 		    ip->ip_src, th->th_sport, 0, NULL);
1292 		if (inp != NULL)  {
1293 			INP_WLOCK(inp);
1294 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1295 			    !(inp->inp_vflag & INP_DROPPED) &&
1296 			    !(inp->inp_socket == NULL)) {
1297 				icmp_tcp_seq = htonl(th->th_seq);
1298 				tp = intotcpcb(inp);
1299 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1300 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1301 					if (cmd == PRC_MSGSIZE) {
1302 					    /*
1303 					     * MTU discovery:
1304 					     * If we got a needfrag set the MTU
1305 					     * in the route to the suggested new
1306 					     * value (if given) and then notify.
1307 					     */
1308 					    bzero(&inc, sizeof(inc));
1309 					    inc.inc_faddr = faddr;
1310 					    inc.inc_fibnum =
1311 						inp->inp_inc.inc_fibnum;
1312 
1313 					    mtu = ntohs(icp->icmp_nextmtu);
1314 					    /*
1315 					     * If no alternative MTU was
1316 					     * proposed, try the next smaller
1317 					     * one.  ip->ip_len has already
1318 					     * been swapped in icmp_input().
1319 					     */
1320 					    if (!mtu)
1321 						mtu = ip_next_mtu(ip->ip_len,
1322 						 1);
1323 					    if (mtu < max(296, V_tcp_minmss
1324 						 + sizeof(struct tcpiphdr)))
1325 						mtu = 0;
1326 					    if (!mtu)
1327 						mtu = V_tcp_mssdflt
1328 						 + sizeof(struct tcpiphdr);
1329 					    /*
1330 					     * Only cache the the MTU if it
1331 					     * is smaller than the interface
1332 					     * or route MTU.  tcp_mtudisc()
1333 					     * will do right thing by itself.
1334 					     */
1335 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1336 						tcp_hc_updatemtu(&inc, mtu);
1337 					}
1338 
1339 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1340 				}
1341 			}
1342 			if (inp != NULL)
1343 				INP_WUNLOCK(inp);
1344 		} else {
1345 			bzero(&inc, sizeof(inc));
1346 			inc.inc_fport = th->th_dport;
1347 			inc.inc_lport = th->th_sport;
1348 			inc.inc_faddr = faddr;
1349 			inc.inc_laddr = ip->ip_src;
1350 			syncache_unreach(&inc, th);
1351 		}
1352 		INP_INFO_WUNLOCK(&V_tcbinfo);
1353 	} else
1354 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1355 }
1356 
1357 #ifdef INET6
1358 void
1359 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1360 {
1361 	INIT_VNET_INET(curvnet);
1362 	struct tcphdr th;
1363 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1364 	struct ip6_hdr *ip6;
1365 	struct mbuf *m;
1366 	struct ip6ctlparam *ip6cp = NULL;
1367 	const struct sockaddr_in6 *sa6_src = NULL;
1368 	int off;
1369 	struct tcp_portonly {
1370 		u_int16_t th_sport;
1371 		u_int16_t th_dport;
1372 	} *thp;
1373 
1374 	if (sa->sa_family != AF_INET6 ||
1375 	    sa->sa_len != sizeof(struct sockaddr_in6))
1376 		return;
1377 
1378 	if (cmd == PRC_MSGSIZE)
1379 		notify = tcp_mtudisc;
1380 	else if (!PRC_IS_REDIRECT(cmd) &&
1381 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1382 		return;
1383 	/* Source quench is depreciated. */
1384 	else if (cmd == PRC_QUENCH)
1385 		return;
1386 
1387 	/* if the parameter is from icmp6, decode it. */
1388 	if (d != NULL) {
1389 		ip6cp = (struct ip6ctlparam *)d;
1390 		m = ip6cp->ip6c_m;
1391 		ip6 = ip6cp->ip6c_ip6;
1392 		off = ip6cp->ip6c_off;
1393 		sa6_src = ip6cp->ip6c_src;
1394 	} else {
1395 		m = NULL;
1396 		ip6 = NULL;
1397 		off = 0;	/* fool gcc */
1398 		sa6_src = &sa6_any;
1399 	}
1400 
1401 	if (ip6 != NULL) {
1402 		struct in_conninfo inc;
1403 		/*
1404 		 * XXX: We assume that when IPV6 is non NULL,
1405 		 * M and OFF are valid.
1406 		 */
1407 
1408 		/* check if we can safely examine src and dst ports */
1409 		if (m->m_pkthdr.len < off + sizeof(*thp))
1410 			return;
1411 
1412 		bzero(&th, sizeof(th));
1413 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1414 
1415 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1416 		    (struct sockaddr *)ip6cp->ip6c_src,
1417 		    th.th_sport, cmd, NULL, notify);
1418 
1419 		bzero(&inc, sizeof(inc));
1420 		inc.inc_fport = th.th_dport;
1421 		inc.inc_lport = th.th_sport;
1422 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1423 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1424 		inc.inc_flags |= INC_ISIPV6;
1425 		INP_INFO_WLOCK(&V_tcbinfo);
1426 		syncache_unreach(&inc, &th);
1427 		INP_INFO_WUNLOCK(&V_tcbinfo);
1428 	} else
1429 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1430 			      0, cmd, NULL, notify);
1431 }
1432 #endif /* INET6 */
1433 
1434 
1435 /*
1436  * Following is where TCP initial sequence number generation occurs.
1437  *
1438  * There are two places where we must use initial sequence numbers:
1439  * 1.  In SYN-ACK packets.
1440  * 2.  In SYN packets.
1441  *
1442  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1443  * tcp_syncache.c for details.
1444  *
1445  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1446  * depends on this property.  In addition, these ISNs should be
1447  * unguessable so as to prevent connection hijacking.  To satisfy
1448  * the requirements of this situation, the algorithm outlined in
1449  * RFC 1948 is used, with only small modifications.
1450  *
1451  * Implementation details:
1452  *
1453  * Time is based off the system timer, and is corrected so that it
1454  * increases by one megabyte per second.  This allows for proper
1455  * recycling on high speed LANs while still leaving over an hour
1456  * before rollover.
1457  *
1458  * As reading the *exact* system time is too expensive to be done
1459  * whenever setting up a TCP connection, we increment the time
1460  * offset in two ways.  First, a small random positive increment
1461  * is added to isn_offset for each connection that is set up.
1462  * Second, the function tcp_isn_tick fires once per clock tick
1463  * and increments isn_offset as necessary so that sequence numbers
1464  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1465  * random positive increments serve only to ensure that the same
1466  * exact sequence number is never sent out twice (as could otherwise
1467  * happen when a port is recycled in less than the system tick
1468  * interval.)
1469  *
1470  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1471  * between seeding of isn_secret.  This is normally set to zero,
1472  * as reseeding should not be necessary.
1473  *
1474  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1475  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1476  * general, this means holding an exclusive (write) lock.
1477  */
1478 
1479 #define ISN_BYTES_PER_SECOND 1048576
1480 #define ISN_STATIC_INCREMENT 4096
1481 #define ISN_RANDOM_INCREMENT (4096 - 1)
1482 
1483 #ifdef VIMAGE_GLOBALS
1484 static u_char isn_secret[32];
1485 static int isn_last_reseed;
1486 static u_int32_t isn_offset, isn_offset_old;
1487 #endif
1488 
1489 tcp_seq
1490 tcp_new_isn(struct tcpcb *tp)
1491 {
1492 	INIT_VNET_INET(tp->t_vnet);
1493 	MD5_CTX isn_ctx;
1494 	u_int32_t md5_buffer[4];
1495 	tcp_seq new_isn;
1496 
1497 	INP_WLOCK_ASSERT(tp->t_inpcb);
1498 
1499 	ISN_LOCK();
1500 	/* Seed if this is the first use, reseed if requested. */
1501 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1502 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1503 		< (u_int)ticks))) {
1504 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1505 		V_isn_last_reseed = ticks;
1506 	}
1507 
1508 	/* Compute the md5 hash and return the ISN. */
1509 	MD5Init(&isn_ctx);
1510 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1511 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1512 #ifdef INET6
1513 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1514 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1515 			  sizeof(struct in6_addr));
1516 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1517 			  sizeof(struct in6_addr));
1518 	} else
1519 #endif
1520 	{
1521 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1522 			  sizeof(struct in_addr));
1523 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1524 			  sizeof(struct in_addr));
1525 	}
1526 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1527 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1528 	new_isn = (tcp_seq) md5_buffer[0];
1529 	V_isn_offset += ISN_STATIC_INCREMENT +
1530 		(arc4random() & ISN_RANDOM_INCREMENT);
1531 	new_isn += V_isn_offset;
1532 	ISN_UNLOCK();
1533 	return (new_isn);
1534 }
1535 
1536 /*
1537  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1538  * to keep time flowing at a relatively constant rate.  If the random
1539  * increments have already pushed us past the projected offset, do nothing.
1540  */
1541 static void
1542 tcp_isn_tick(void *xtp)
1543 {
1544 	VNET_ITERATOR_DECL(vnet_iter);
1545 	u_int32_t projected_offset;
1546 
1547 	ISN_LOCK();
1548 	VNET_LIST_RLOCK();
1549 	VNET_FOREACH(vnet_iter) {
1550 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1551 		INIT_VNET_INET(curvnet);
1552 		projected_offset =
1553 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1554 
1555 		if (SEQ_GT(projected_offset, V_isn_offset))
1556 			V_isn_offset = projected_offset;
1557 
1558 		V_isn_offset_old = V_isn_offset;
1559 		CURVNET_RESTORE();
1560 	}
1561 	VNET_LIST_RUNLOCK();
1562 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1563 	ISN_UNLOCK();
1564 }
1565 
1566 /*
1567  * When a specific ICMP unreachable message is received and the
1568  * connection state is SYN-SENT, drop the connection.  This behavior
1569  * is controlled by the icmp_may_rst sysctl.
1570  */
1571 struct inpcb *
1572 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1573 {
1574 #ifdef INVARIANTS
1575 	INIT_VNET_INET(inp->inp_vnet);
1576 #endif
1577 	struct tcpcb *tp;
1578 
1579 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1580 	INP_WLOCK_ASSERT(inp);
1581 
1582 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1583 	    (inp->inp_vflag & INP_DROPPED))
1584 		return (inp);
1585 
1586 	tp = intotcpcb(inp);
1587 	if (tp->t_state != TCPS_SYN_SENT)
1588 		return (inp);
1589 
1590 	tp = tcp_drop(tp, errno);
1591 	if (tp != NULL)
1592 		return (inp);
1593 	else
1594 		return (NULL);
1595 }
1596 
1597 /*
1598  * When `need fragmentation' ICMP is received, update our idea of the MSS
1599  * based on the new value in the route.  Also nudge TCP to send something,
1600  * since we know the packet we just sent was dropped.
1601  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1602  */
1603 struct inpcb *
1604 tcp_mtudisc(struct inpcb *inp, int errno)
1605 {
1606 	INIT_VNET_INET(inp->inp_vnet);
1607 	struct tcpcb *tp;
1608 	struct socket *so;
1609 
1610 	INP_WLOCK_ASSERT(inp);
1611 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1612 	    (inp->inp_vflag & INP_DROPPED))
1613 		return (inp);
1614 
1615 	tp = intotcpcb(inp);
1616 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1617 
1618 	tcp_mss_update(tp, -1, NULL, NULL);
1619 
1620 	so = inp->inp_socket;
1621 	SOCKBUF_LOCK(&so->so_snd);
1622 	/* If the mss is larger than the socket buffer, decrease the mss. */
1623 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1624 		tp->t_maxseg = so->so_snd.sb_hiwat;
1625 	SOCKBUF_UNLOCK(&so->so_snd);
1626 
1627 	V_tcpstat.tcps_mturesent++;
1628 	tp->t_rtttime = 0;
1629 	tp->snd_nxt = tp->snd_una;
1630 	tcp_free_sackholes(tp);
1631 	tp->snd_recover = tp->snd_max;
1632 	if (tp->t_flags & TF_SACK_PERMIT)
1633 		EXIT_FASTRECOVERY(tp);
1634 	tcp_output_send(tp);
1635 	return (inp);
1636 }
1637 
1638 /*
1639  * Look-up the routing entry to the peer of this inpcb.  If no route
1640  * is found and it cannot be allocated, then return 0.  This routine
1641  * is called by TCP routines that access the rmx structure and by
1642  * tcp_mss_update to get the peer/interface MTU.
1643  */
1644 u_long
1645 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1646 {
1647 	struct route sro;
1648 	struct sockaddr_in *dst;
1649 	struct ifnet *ifp;
1650 	u_long maxmtu = 0;
1651 
1652 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1653 
1654 	bzero(&sro, sizeof(sro));
1655 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1656 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1657 		dst->sin_family = AF_INET;
1658 		dst->sin_len = sizeof(*dst);
1659 		dst->sin_addr = inc->inc_faddr;
1660 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1661 	}
1662 	if (sro.ro_rt != NULL) {
1663 		ifp = sro.ro_rt->rt_ifp;
1664 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1665 			maxmtu = ifp->if_mtu;
1666 		else
1667 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1668 
1669 		/* Report additional interface capabilities. */
1670 		if (flags != NULL) {
1671 			if (ifp->if_capenable & IFCAP_TSO4 &&
1672 			    ifp->if_hwassist & CSUM_TSO)
1673 				*flags |= CSUM_TSO;
1674 		}
1675 		RTFREE(sro.ro_rt);
1676 	}
1677 	return (maxmtu);
1678 }
1679 
1680 #ifdef INET6
1681 u_long
1682 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1683 {
1684 	struct route_in6 sro6;
1685 	struct ifnet *ifp;
1686 	u_long maxmtu = 0;
1687 
1688 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1689 
1690 	bzero(&sro6, sizeof(sro6));
1691 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1692 		sro6.ro_dst.sin6_family = AF_INET6;
1693 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1694 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1695 		rtalloc_ign((struct route *)&sro6, 0);
1696 	}
1697 	if (sro6.ro_rt != NULL) {
1698 		ifp = sro6.ro_rt->rt_ifp;
1699 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1700 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1701 		else
1702 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1703 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1704 
1705 		/* Report additional interface capabilities. */
1706 		if (flags != NULL) {
1707 			if (ifp->if_capenable & IFCAP_TSO6 &&
1708 			    ifp->if_hwassist & CSUM_TSO)
1709 				*flags |= CSUM_TSO;
1710 		}
1711 		RTFREE(sro6.ro_rt);
1712 	}
1713 
1714 	return (maxmtu);
1715 }
1716 #endif /* INET6 */
1717 
1718 #ifdef IPSEC
1719 /* compute ESP/AH header size for TCP, including outer IP header. */
1720 size_t
1721 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1722 {
1723 	struct inpcb *inp;
1724 	struct mbuf *m;
1725 	size_t hdrsiz;
1726 	struct ip *ip;
1727 #ifdef INET6
1728 	struct ip6_hdr *ip6;
1729 #endif
1730 	struct tcphdr *th;
1731 
1732 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1733 		return (0);
1734 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1735 	if (!m)
1736 		return (0);
1737 
1738 #ifdef INET6
1739 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1740 		ip6 = mtod(m, struct ip6_hdr *);
1741 		th = (struct tcphdr *)(ip6 + 1);
1742 		m->m_pkthdr.len = m->m_len =
1743 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1744 		tcpip_fillheaders(inp, ip6, th);
1745 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1746 	} else
1747 #endif /* INET6 */
1748 	{
1749 		ip = mtod(m, struct ip *);
1750 		th = (struct tcphdr *)(ip + 1);
1751 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1752 		tcpip_fillheaders(inp, ip, th);
1753 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1754 	}
1755 
1756 	m_free(m);
1757 	return (hdrsiz);
1758 }
1759 #endif /* IPSEC */
1760 
1761 /*
1762  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1763  *
1764  * This code attempts to calculate the bandwidth-delay product as a
1765  * means of determining the optimal window size to maximize bandwidth,
1766  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1767  * routers.  This code also does a fairly good job keeping RTTs in check
1768  * across slow links like modems.  We implement an algorithm which is very
1769  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1770  * transmitter side of a TCP connection and so only effects the transmit
1771  * side of the connection.
1772  *
1773  * BACKGROUND:  TCP makes no provision for the management of buffer space
1774  * at the end points or at the intermediate routers and switches.  A TCP
1775  * stream, whether using NewReno or not, will eventually buffer as
1776  * many packets as it is able and the only reason this typically works is
1777  * due to the fairly small default buffers made available for a connection
1778  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1779  * scaling it is now fairly easy for even a single TCP connection to blow-out
1780  * all available buffer space not only on the local interface, but on
1781  * intermediate routers and switches as well.  NewReno makes a misguided
1782  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1783  * then backing off, then steadily increasing the window again until another
1784  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1785  * is only made worse as network loads increase and the idea of intentionally
1786  * blowing out network buffers is, frankly, a terrible way to manage network
1787  * resources.
1788  *
1789  * It is far better to limit the transmit window prior to the failure
1790  * condition being achieved.  There are two general ways to do this:  First
1791  * you can 'scan' through different transmit window sizes and locate the
1792  * point where the RTT stops increasing, indicating that you have filled the
1793  * pipe, then scan backwards until you note that RTT stops decreasing, then
1794  * repeat ad-infinitum.  This method works in principle but has severe
1795  * implementation issues due to RTT variances, timer granularity, and
1796  * instability in the algorithm which can lead to many false positives and
1797  * create oscillations as well as interact badly with other TCP streams
1798  * implementing the same algorithm.
1799  *
1800  * The second method is to limit the window to the bandwidth delay product
1801  * of the link.  This is the method we implement.  RTT variances and our
1802  * own manipulation of the congestion window, bwnd, can potentially
1803  * destabilize the algorithm.  For this reason we have to stabilize the
1804  * elements used to calculate the window.  We do this by using the minimum
1805  * observed RTT, the long term average of the observed bandwidth, and
1806  * by adding two segments worth of slop.  It isn't perfect but it is able
1807  * to react to changing conditions and gives us a very stable basis on
1808  * which to extend the algorithm.
1809  */
1810 void
1811 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1812 {
1813 	INIT_VNET_INET(tp->t_vnet);
1814 	u_long bw;
1815 	u_long bwnd;
1816 	int save_ticks;
1817 
1818 	INP_WLOCK_ASSERT(tp->t_inpcb);
1819 
1820 	/*
1821 	 * If inflight_enable is disabled in the middle of a tcp connection,
1822 	 * make sure snd_bwnd is effectively disabled.
1823 	 */
1824 	if (V_tcp_inflight_enable == 0 ||
1825 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1826 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1827 		tp->snd_bandwidth = 0;
1828 		return;
1829 	}
1830 
1831 	/*
1832 	 * Figure out the bandwidth.  Due to the tick granularity this
1833 	 * is a very rough number and it MUST be averaged over a fairly
1834 	 * long period of time.  XXX we need to take into account a link
1835 	 * that is not using all available bandwidth, but for now our
1836 	 * slop will ramp us up if this case occurs and the bandwidth later
1837 	 * increases.
1838 	 *
1839 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1840 	 * effectively reset t_bw_rtttime for this case.
1841 	 */
1842 	save_ticks = ticks;
1843 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1844 		return;
1845 
1846 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1847 	    (save_ticks - tp->t_bw_rtttime);
1848 	tp->t_bw_rtttime = save_ticks;
1849 	tp->t_bw_rtseq = ack_seq;
1850 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1851 		return;
1852 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1853 
1854 	tp->snd_bandwidth = bw;
1855 
1856 	/*
1857 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1858 	 * segments.  The additional slop puts us squarely in the sweet
1859 	 * spot and also handles the bandwidth run-up case and stabilization.
1860 	 * Without the slop we could be locking ourselves into a lower
1861 	 * bandwidth.
1862 	 *
1863 	 * Situations Handled:
1864 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1865 	 *	    high speed LANs, allowing larger TCP buffers to be
1866 	 *	    specified, and also does a good job preventing
1867 	 *	    over-queueing of packets over choke points like modems
1868 	 *	    (at least for the transmit side).
1869 	 *
1870 	 *	(2) Is able to handle changing network loads (bandwidth
1871 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1872 	 *	    increases).
1873 	 *
1874 	 *	(3) Theoretically should stabilize in the face of multiple
1875 	 *	    connections implementing the same algorithm (this may need
1876 	 *	    a little work).
1877 	 *
1878 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1879 	 *	    be adjusted with a sysctl but typically only needs to be
1880 	 *	    on very slow connections.  A value no smaller then 5
1881 	 *	    should be used, but only reduce this default if you have
1882 	 *	    no other choice.
1883 	 */
1884 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1885 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1886 #undef USERTT
1887 
1888 	if (tcp_inflight_debug > 0) {
1889 		static int ltime;
1890 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1891 			ltime = ticks;
1892 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1893 			    tp,
1894 			    bw,
1895 			    tp->t_rttbest,
1896 			    tp->t_srtt,
1897 			    bwnd
1898 			);
1899 		}
1900 	}
1901 	if ((long)bwnd < V_tcp_inflight_min)
1902 		bwnd = V_tcp_inflight_min;
1903 	if (bwnd > V_tcp_inflight_max)
1904 		bwnd = V_tcp_inflight_max;
1905 	if ((long)bwnd < tp->t_maxseg * 2)
1906 		bwnd = tp->t_maxseg * 2;
1907 	tp->snd_bwnd = bwnd;
1908 }
1909 
1910 #ifdef TCP_SIGNATURE
1911 /*
1912  * Callback function invoked by m_apply() to digest TCP segment data
1913  * contained within an mbuf chain.
1914  */
1915 static int
1916 tcp_signature_apply(void *fstate, void *data, u_int len)
1917 {
1918 
1919 	MD5Update(fstate, (u_char *)data, len);
1920 	return (0);
1921 }
1922 
1923 /*
1924  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1925  *
1926  * Parameters:
1927  * m		pointer to head of mbuf chain
1928  * _unused
1929  * len		length of TCP segment data, excluding options
1930  * optlen	length of TCP segment options
1931  * buf		pointer to storage for computed MD5 digest
1932  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1933  *
1934  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1935  * When called from tcp_input(), we can be sure that th_sum has been
1936  * zeroed out and verified already.
1937  *
1938  * Return 0 if successful, otherwise return -1.
1939  *
1940  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1941  * search with the destination IP address, and a 'magic SPI' to be
1942  * determined by the application. This is hardcoded elsewhere to 1179
1943  * right now. Another branch of this code exists which uses the SPD to
1944  * specify per-application flows but it is unstable.
1945  */
1946 int
1947 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1948     u_char *buf, u_int direction)
1949 {
1950 	INIT_VNET_IPSEC(curvnet);
1951 	union sockaddr_union dst;
1952 	struct ippseudo ippseudo;
1953 	MD5_CTX ctx;
1954 	int doff;
1955 	struct ip *ip;
1956 	struct ipovly *ipovly;
1957 	struct secasvar *sav;
1958 	struct tcphdr *th;
1959 #ifdef INET6
1960 	struct ip6_hdr *ip6;
1961 	struct in6_addr in6;
1962 	char ip6buf[INET6_ADDRSTRLEN];
1963 	uint32_t plen;
1964 	uint16_t nhdr;
1965 #endif
1966 	u_short savecsum;
1967 
1968 	KASSERT(m != NULL, ("NULL mbuf chain"));
1969 	KASSERT(buf != NULL, ("NULL signature pointer"));
1970 
1971 	/* Extract the destination from the IP header in the mbuf. */
1972 	bzero(&dst, sizeof(union sockaddr_union));
1973 	ip = mtod(m, struct ip *);
1974 #ifdef INET6
1975 	ip6 = NULL;	/* Make the compiler happy. */
1976 #endif
1977 	switch (ip->ip_v) {
1978 	case IPVERSION:
1979 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1980 		dst.sa.sa_family = AF_INET;
1981 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1982 		    ip->ip_src : ip->ip_dst;
1983 		break;
1984 #ifdef INET6
1985 	case (IPV6_VERSION >> 4):
1986 		ip6 = mtod(m, struct ip6_hdr *);
1987 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1988 		dst.sa.sa_family = AF_INET6;
1989 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1990 		    ip6->ip6_src : ip6->ip6_dst;
1991 		break;
1992 #endif
1993 	default:
1994 		return (EINVAL);
1995 		/* NOTREACHED */
1996 		break;
1997 	}
1998 
1999 	/* Look up an SADB entry which matches the address of the peer. */
2000 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2001 	if (sav == NULL) {
2002 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2003 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2004 #ifdef INET6
2005 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2006 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2007 #endif
2008 			"(unsupported)"));
2009 		return (EINVAL);
2010 	}
2011 
2012 	MD5Init(&ctx);
2013 	/*
2014 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2015 	 *
2016 	 * XXX The ippseudo header MUST be digested in network byte order,
2017 	 * or else we'll fail the regression test. Assume all fields we've
2018 	 * been doing arithmetic on have been in host byte order.
2019 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2020 	 * tcp_output(), the underlying ip_len member has not yet been set.
2021 	 */
2022 	switch (ip->ip_v) {
2023 	case IPVERSION:
2024 		ipovly = (struct ipovly *)ip;
2025 		ippseudo.ippseudo_src = ipovly->ih_src;
2026 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2027 		ippseudo.ippseudo_pad = 0;
2028 		ippseudo.ippseudo_p = IPPROTO_TCP;
2029 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2030 		    optlen);
2031 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2032 
2033 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2034 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2035 		break;
2036 #ifdef INET6
2037 	/*
2038 	 * RFC 2385, 2.0  Proposal
2039 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2040 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2041 	 * extended next header value (to form 32 bits), and 32-bit segment
2042 	 * length.
2043 	 * Note: Upper-Layer Packet Length comes before Next Header.
2044 	 */
2045 	case (IPV6_VERSION >> 4):
2046 		in6 = ip6->ip6_src;
2047 		in6_clearscope(&in6);
2048 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2049 		in6 = ip6->ip6_dst;
2050 		in6_clearscope(&in6);
2051 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2052 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2053 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2054 		nhdr = 0;
2055 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2056 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2057 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2058 		nhdr = IPPROTO_TCP;
2059 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2060 
2061 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2062 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2063 		break;
2064 #endif
2065 	default:
2066 		return (EINVAL);
2067 		/* NOTREACHED */
2068 		break;
2069 	}
2070 
2071 
2072 	/*
2073 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2074 	 * The TCP checksum must be set to zero.
2075 	 */
2076 	savecsum = th->th_sum;
2077 	th->th_sum = 0;
2078 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2079 	th->th_sum = savecsum;
2080 
2081 	/*
2082 	 * Step 3: Update MD5 hash with TCP segment data.
2083 	 *         Use m_apply() to avoid an early m_pullup().
2084 	 */
2085 	if (len > 0)
2086 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2087 
2088 	/*
2089 	 * Step 4: Update MD5 hash with shared secret.
2090 	 */
2091 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2092 	MD5Final(buf, &ctx);
2093 
2094 	key_sa_recordxfer(sav, m);
2095 	KEY_FREESAV(&sav);
2096 	return (0);
2097 }
2098 #endif /* TCP_SIGNATURE */
2099 
2100 static int
2101 sysctl_drop(SYSCTL_HANDLER_ARGS)
2102 {
2103 	INIT_VNET_INET(curvnet);
2104 #ifdef INET6
2105 	INIT_VNET_INET6(curvnet);
2106 #endif
2107 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2108 	struct sockaddr_storage addrs[2];
2109 	struct inpcb *inp;
2110 	struct tcpcb *tp;
2111 	struct tcptw *tw;
2112 	struct sockaddr_in *fin, *lin;
2113 #ifdef INET6
2114 	struct sockaddr_in6 *fin6, *lin6;
2115 	struct in6_addr f6, l6;
2116 #endif
2117 	int error;
2118 
2119 	inp = NULL;
2120 	fin = lin = NULL;
2121 #ifdef INET6
2122 	fin6 = lin6 = NULL;
2123 #endif
2124 	error = 0;
2125 
2126 	if (req->oldptr != NULL || req->oldlen != 0)
2127 		return (EINVAL);
2128 	if (req->newptr == NULL)
2129 		return (EPERM);
2130 	if (req->newlen < sizeof(addrs))
2131 		return (ENOMEM);
2132 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2133 	if (error)
2134 		return (error);
2135 
2136 	switch (addrs[0].ss_family) {
2137 #ifdef INET6
2138 	case AF_INET6:
2139 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2140 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2141 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2142 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2143 			return (EINVAL);
2144 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2145 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2146 				return (EINVAL);
2147 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2148 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2149 			fin = (struct sockaddr_in *)&addrs[0];
2150 			lin = (struct sockaddr_in *)&addrs[1];
2151 			break;
2152 		}
2153 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2154 		if (error)
2155 			return (error);
2156 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2157 		if (error)
2158 			return (error);
2159 		break;
2160 #endif
2161 	case AF_INET:
2162 		fin = (struct sockaddr_in *)&addrs[0];
2163 		lin = (struct sockaddr_in *)&addrs[1];
2164 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2165 		    lin->sin_len != sizeof(struct sockaddr_in))
2166 			return (EINVAL);
2167 		break;
2168 	default:
2169 		return (EINVAL);
2170 	}
2171 	INP_INFO_WLOCK(&V_tcbinfo);
2172 	switch (addrs[0].ss_family) {
2173 #ifdef INET6
2174 	case AF_INET6:
2175 		inp = in6_pcblookup_hash(&V_tcbinfo, &f6, fin6->sin6_port,
2176 		    &l6, lin6->sin6_port, 0, NULL);
2177 		break;
2178 #endif
2179 	case AF_INET:
2180 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2181 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2182 		break;
2183 	}
2184 	if (inp != NULL) {
2185 		INP_WLOCK(inp);
2186 		if (inp->inp_vflag & INP_TIMEWAIT) {
2187 			/*
2188 			 * XXXRW: There currently exists a state where an
2189 			 * inpcb is present, but its timewait state has been
2190 			 * discarded.  For now, don't allow dropping of this
2191 			 * type of inpcb.
2192 			 */
2193 			tw = intotw(inp);
2194 			if (tw != NULL)
2195 				tcp_twclose(tw, 0);
2196 			else
2197 				INP_WUNLOCK(inp);
2198 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2199 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2200 			tp = intotcpcb(inp);
2201 			tp = tcp_drop(tp, ECONNABORTED);
2202 			if (tp != NULL)
2203 				INP_WUNLOCK(inp);
2204 		} else
2205 			INP_WUNLOCK(inp);
2206 	} else
2207 		error = ESRCH;
2208 	INP_INFO_WUNLOCK(&V_tcbinfo);
2209 	return (error);
2210 }
2211 
2212 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2213     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2214     0, sysctl_drop, "", "Drop TCP connection");
2215 
2216 /*
2217  * Generate a standardized TCP log line for use throughout the
2218  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2219  * allow use in the interrupt context.
2220  *
2221  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2222  * NB: The function may return NULL if memory allocation failed.
2223  *
2224  * Due to header inclusion and ordering limitations the struct ip
2225  * and ip6_hdr pointers have to be passed as void pointers.
2226  */
2227 char *
2228 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2229     const void *ip6hdr)
2230 {
2231 	char *s, *sp;
2232 	size_t size;
2233 	struct ip *ip;
2234 #ifdef INET6
2235 	const struct ip6_hdr *ip6;
2236 
2237 	ip6 = (const struct ip6_hdr *)ip6hdr;
2238 #endif /* INET6 */
2239 	ip = (struct ip *)ip4hdr;
2240 
2241 	/*
2242 	 * The log line looks like this:
2243 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2244 	 */
2245 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2246 	    sizeof(PRINT_TH_FLAGS) + 1 +
2247 #ifdef INET6
2248 	    2 * INET6_ADDRSTRLEN;
2249 #else
2250 	    2 * INET_ADDRSTRLEN;
2251 #endif /* INET6 */
2252 
2253 	/* Is logging enabled? */
2254 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2255 		return (NULL);
2256 
2257 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2258 	if (s == NULL)
2259 		return (NULL);
2260 
2261 	strcat(s, "TCP: [");
2262 	sp = s + strlen(s);
2263 
2264 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2265 		inet_ntoa_r(inc->inc_faddr, sp);
2266 		sp = s + strlen(s);
2267 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2268 		sp = s + strlen(s);
2269 		inet_ntoa_r(inc->inc_laddr, sp);
2270 		sp = s + strlen(s);
2271 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2272 #ifdef INET6
2273 	} else if (inc) {
2274 		ip6_sprintf(sp, &inc->inc6_faddr);
2275 		sp = s + strlen(s);
2276 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2277 		sp = s + strlen(s);
2278 		ip6_sprintf(sp, &inc->inc6_laddr);
2279 		sp = s + strlen(s);
2280 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2281 	} else if (ip6 && th) {
2282 		ip6_sprintf(sp, &ip6->ip6_src);
2283 		sp = s + strlen(s);
2284 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2285 		sp = s + strlen(s);
2286 		ip6_sprintf(sp, &ip6->ip6_dst);
2287 		sp = s + strlen(s);
2288 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2289 #endif /* INET6 */
2290 	} else if (ip && th) {
2291 		inet_ntoa_r(ip->ip_src, sp);
2292 		sp = s + strlen(s);
2293 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2294 		sp = s + strlen(s);
2295 		inet_ntoa_r(ip->ip_dst, sp);
2296 		sp = s + strlen(s);
2297 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2298 	} else {
2299 		free(s, M_TCPLOG);
2300 		return (NULL);
2301 	}
2302 	sp = s + strlen(s);
2303 	if (th)
2304 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2305 	if (*(s + size - 1) != '\0')
2306 		panic("%s: string too long", __func__);
2307 	return (s);
2308 }
2309