xref: /freebsd/sys/netinet/tcp_subr.c (revision cc3f4b99653c34ae64f8a1fddea370abefef680e)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/hhook.h>
45 #include <sys/kernel.h>
46 #include <sys/khelp.h>
47 #include <sys/sysctl.h>
48 #include <sys/jail.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #ifdef INET6
52 #include <sys/domain.h>
53 #endif
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/sdt.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/random.h>
61 
62 #include <vm/uma.h>
63 
64 #include <net/route.h>
65 #include <net/if.h>
66 #include <net/if_var.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/cc.h>
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/in_var.h>
75 #include <netinet/ip.h>
76 #include <netinet/ip_icmp.h>
77 #include <netinet/ip_var.h>
78 #ifdef INET6
79 #include <netinet/ip6.h>
80 #include <netinet6/in6_pcb.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #include <netinet6/nd6.h>
84 #endif
85 
86 #include <netinet/tcp_fsm.h>
87 #include <netinet/tcp_seq.h>
88 #include <netinet/tcp_timer.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/tcp_syncache.h>
91 #ifdef INET6
92 #include <netinet6/tcp6_var.h>
93 #endif
94 #include <netinet/tcpip.h>
95 #ifdef TCPDEBUG
96 #include <netinet/tcp_debug.h>
97 #endif
98 #ifdef INET6
99 #include <netinet6/ip6protosw.h>
100 #endif
101 #ifdef TCP_OFFLOAD
102 #include <netinet/tcp_offload.h>
103 #endif
104 
105 #ifdef IPSEC
106 #include <netipsec/ipsec.h>
107 #include <netipsec/xform.h>
108 #ifdef INET6
109 #include <netipsec/ipsec6.h>
110 #endif
111 #include <netipsec/key.h>
112 #include <sys/syslog.h>
113 #endif /*IPSEC*/
114 
115 #include <machine/in_cksum.h>
116 #include <sys/md5.h>
117 
118 #include <security/mac/mac_framework.h>
119 
120 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
121 #ifdef INET6
122 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
123 #endif
124 
125 static int
126 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
127 {
128 	int error, new;
129 
130 	new = V_tcp_mssdflt;
131 	error = sysctl_handle_int(oidp, &new, 0, req);
132 	if (error == 0 && req->newptr) {
133 		if (new < TCP_MINMSS)
134 			error = EINVAL;
135 		else
136 			V_tcp_mssdflt = new;
137 	}
138 	return (error);
139 }
140 
141 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
142     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
143     &sysctl_net_inet_tcp_mss_check, "I",
144     "Default TCP Maximum Segment Size");
145 
146 #ifdef INET6
147 static int
148 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
149 {
150 	int error, new;
151 
152 	new = V_tcp_v6mssdflt;
153 	error = sysctl_handle_int(oidp, &new, 0, req);
154 	if (error == 0 && req->newptr) {
155 		if (new < TCP_MINMSS)
156 			error = EINVAL;
157 		else
158 			V_tcp_v6mssdflt = new;
159 	}
160 	return (error);
161 }
162 
163 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
164     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
165     &sysctl_net_inet_tcp_mss_v6_check, "I",
166    "Default TCP Maximum Segment Size for IPv6");
167 #endif /* INET6 */
168 
169 /*
170  * Minimum MSS we accept and use. This prevents DoS attacks where
171  * we are forced to a ridiculous low MSS like 20 and send hundreds
172  * of packets instead of one. The effect scales with the available
173  * bandwidth and quickly saturates the CPU and network interface
174  * with packet generation and sending. Set to zero to disable MINMSS
175  * checking. This setting prevents us from sending too small packets.
176  */
177 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
179      &VNET_NAME(tcp_minmss), 0,
180     "Minimum TCP Maximum Segment Size");
181 
182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
184     &VNET_NAME(tcp_do_rfc1323), 0,
185     "Enable rfc1323 (high performance TCP) extensions");
186 
187 static int	tcp_log_debug = 0;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
189     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
190 
191 static int	tcp_tcbhashsize = 0;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
193     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
194 
195 static int	do_tcpdrain = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
197     "Enable tcp_drain routine for extra help when low on mbufs");
198 
199 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
200     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
201 
202 static VNET_DEFINE(int, icmp_may_rst) = 1;
203 #define	V_icmp_may_rst			VNET(icmp_may_rst)
204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
205     &VNET_NAME(icmp_may_rst), 0,
206     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
207 
208 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
209 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
211     &VNET_NAME(tcp_isn_reseed_interval), 0,
212     "Seconds between reseeding of ISN secret");
213 
214 static int	tcp_soreceive_stream = 0;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
216     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
217 
218 #ifdef TCP_SIGNATURE
219 static int	tcp_sig_checksigs = 1;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
221     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
222 #endif
223 
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define	V_sack_hole_zone		VNET(sack_hole_zone)
226 
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
228 
229 static struct inpcb *tcp_notify(struct inpcb *, int);
230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
232 		    void *ip4hdr, const void *ip6hdr);
233 
234 /*
235  * Target size of TCP PCB hash tables. Must be a power of two.
236  *
237  * Note that this can be overridden by the kernel environment
238  * variable net.inet.tcp.tcbhashsize
239  */
240 #ifndef TCBHASHSIZE
241 #define TCBHASHSIZE	0
242 #endif
243 
244 /*
245  * XXX
246  * Callouts should be moved into struct tcp directly.  They are currently
247  * separate because the tcpcb structure is exported to userland for sysctl
248  * parsing purposes, which do not know about callouts.
249  */
250 struct tcpcb_mem {
251 	struct	tcpcb		tcb;
252 	struct	tcp_timer	tt;
253 	struct	cc_var		ccv;
254 	struct	osd		osd;
255 };
256 
257 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
258 #define	V_tcpcb_zone			VNET(tcpcb_zone)
259 
260 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
261 static struct mtx isn_mtx;
262 
263 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
264 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
265 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
266 
267 /*
268  * TCP initialization.
269  */
270 static void
271 tcp_zone_change(void *tag)
272 {
273 
274 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
275 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
276 	tcp_tw_zone_change();
277 }
278 
279 static int
280 tcp_inpcb_init(void *mem, int size, int flags)
281 {
282 	struct inpcb *inp = mem;
283 
284 	INP_LOCK_INIT(inp, "inp", "tcpinp");
285 	return (0);
286 }
287 
288 /*
289  * Take a value and get the next power of 2 that doesn't overflow.
290  * Used to size the tcp_inpcb hash buckets.
291  */
292 static int
293 maketcp_hashsize(int size)
294 {
295 	int hashsize;
296 
297 	/*
298 	 * auto tune.
299 	 * get the next power of 2 higher than maxsockets.
300 	 */
301 	hashsize = 1 << fls(size);
302 	/* catch overflow, and just go one power of 2 smaller */
303 	if (hashsize < size) {
304 		hashsize = 1 << (fls(size) - 1);
305 	}
306 	return (hashsize);
307 }
308 
309 void
310 tcp_init(void)
311 {
312 	const char *tcbhash_tuneable;
313 	int hashsize;
314 
315 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
316 
317 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
318 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 		printf("%s: WARNING: unable to register helper hook\n", __func__);
320 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
321 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
322 		printf("%s: WARNING: unable to register helper hook\n", __func__);
323 
324 	hashsize = TCBHASHSIZE;
325 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
326 	if (hashsize == 0) {
327 		/*
328 		 * Auto tune the hash size based on maxsockets.
329 		 * A perfect hash would have a 1:1 mapping
330 		 * (hashsize = maxsockets) however it's been
331 		 * suggested that O(2) average is better.
332 		 */
333 		hashsize = maketcp_hashsize(maxsockets / 4);
334 		/*
335 		 * Our historical default is 512,
336 		 * do not autotune lower than this.
337 		 */
338 		if (hashsize < 512)
339 			hashsize = 512;
340 		if (bootverbose)
341 			printf("%s: %s auto tuned to %d\n", __func__,
342 			    tcbhash_tuneable, hashsize);
343 	}
344 	/*
345 	 * We require a hashsize to be a power of two.
346 	 * Previously if it was not a power of two we would just reset it
347 	 * back to 512, which could be a nasty surprise if you did not notice
348 	 * the error message.
349 	 * Instead what we do is clip it to the closest power of two lower
350 	 * than the specified hash value.
351 	 */
352 	if (!powerof2(hashsize)) {
353 		int oldhashsize = hashsize;
354 
355 		hashsize = maketcp_hashsize(hashsize);
356 		/* prevent absurdly low value */
357 		if (hashsize < 16)
358 			hashsize = 16;
359 		printf("%s: WARNING: TCB hash size not a power of 2, "
360 		    "clipped from %d to %d.\n", __func__, oldhashsize,
361 		    hashsize);
362 	}
363 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
364 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
365 	    IPI_HASHFIELDS_4TUPLE);
366 
367 	/*
368 	 * These have to be type stable for the benefit of the timers.
369 	 */
370 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
371 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
372 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
373 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
374 
375 	tcp_tw_init();
376 	syncache_init();
377 	tcp_hc_init();
378 
379 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
380 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
381 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
382 
383 	/* Skip initialization of globals for non-default instances. */
384 	if (!IS_DEFAULT_VNET(curvnet))
385 		return;
386 
387 	/* XXX virtualize those bellow? */
388 	tcp_delacktime = TCPTV_DELACK;
389 	tcp_keepinit = TCPTV_KEEP_INIT;
390 	tcp_keepidle = TCPTV_KEEP_IDLE;
391 	tcp_keepintvl = TCPTV_KEEPINTVL;
392 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
393 	tcp_msl = TCPTV_MSL;
394 	tcp_rexmit_min = TCPTV_MIN;
395 	if (tcp_rexmit_min < 1)
396 		tcp_rexmit_min = 1;
397 	tcp_rexmit_slop = TCPTV_CPU_VAR;
398 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
399 	tcp_tcbhashsize = hashsize;
400 
401 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
402 	if (tcp_soreceive_stream) {
403 #ifdef INET
404 		tcp_usrreqs.pru_soreceive = soreceive_stream;
405 #endif
406 #ifdef INET6
407 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
408 #endif /* INET6 */
409 	}
410 
411 #ifdef INET6
412 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
413 #else /* INET6 */
414 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
415 #endif /* INET6 */
416 	if (max_protohdr < TCP_MINPROTOHDR)
417 		max_protohdr = TCP_MINPROTOHDR;
418 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
419 		panic("tcp_init");
420 #undef TCP_MINPROTOHDR
421 
422 	ISN_LOCK_INIT();
423 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
424 		SHUTDOWN_PRI_DEFAULT);
425 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
426 		EVENTHANDLER_PRI_ANY);
427 }
428 
429 #ifdef VIMAGE
430 void
431 tcp_destroy(void)
432 {
433 	int error;
434 
435 	tcp_hc_destroy();
436 	syncache_destroy();
437 	tcp_tw_destroy();
438 	in_pcbinfo_destroy(&V_tcbinfo);
439 	uma_zdestroy(V_sack_hole_zone);
440 	uma_zdestroy(V_tcpcb_zone);
441 
442 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
443 	if (error != 0) {
444 		printf("%s: WARNING: unable to deregister helper hook "
445 		    "type=%d, id=%d: error %d returned\n", __func__,
446 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
447 	}
448 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
449 	if (error != 0) {
450 		printf("%s: WARNING: unable to deregister helper hook "
451 		    "type=%d, id=%d: error %d returned\n", __func__,
452 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
453 	}
454 }
455 #endif
456 
457 void
458 tcp_fini(void *xtp)
459 {
460 
461 }
462 
463 /*
464  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
465  * tcp_template used to store this data in mbufs, but we now recopy it out
466  * of the tcpcb each time to conserve mbufs.
467  */
468 void
469 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
470 {
471 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
472 
473 	INP_WLOCK_ASSERT(inp);
474 
475 #ifdef INET6
476 	if ((inp->inp_vflag & INP_IPV6) != 0) {
477 		struct ip6_hdr *ip6;
478 
479 		ip6 = (struct ip6_hdr *)ip_ptr;
480 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
481 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
482 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
483 			(IPV6_VERSION & IPV6_VERSION_MASK);
484 		ip6->ip6_nxt = IPPROTO_TCP;
485 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
486 		ip6->ip6_src = inp->in6p_laddr;
487 		ip6->ip6_dst = inp->in6p_faddr;
488 	}
489 #endif /* INET6 */
490 #if defined(INET6) && defined(INET)
491 	else
492 #endif
493 #ifdef INET
494 	{
495 		struct ip *ip;
496 
497 		ip = (struct ip *)ip_ptr;
498 		ip->ip_v = IPVERSION;
499 		ip->ip_hl = 5;
500 		ip->ip_tos = inp->inp_ip_tos;
501 		ip->ip_len = 0;
502 		ip->ip_id = 0;
503 		ip->ip_off = 0;
504 		ip->ip_ttl = inp->inp_ip_ttl;
505 		ip->ip_sum = 0;
506 		ip->ip_p = IPPROTO_TCP;
507 		ip->ip_src = inp->inp_laddr;
508 		ip->ip_dst = inp->inp_faddr;
509 	}
510 #endif /* INET */
511 	th->th_sport = inp->inp_lport;
512 	th->th_dport = inp->inp_fport;
513 	th->th_seq = 0;
514 	th->th_ack = 0;
515 	th->th_x2 = 0;
516 	th->th_off = 5;
517 	th->th_flags = 0;
518 	th->th_win = 0;
519 	th->th_urp = 0;
520 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
521 }
522 
523 /*
524  * Create template to be used to send tcp packets on a connection.
525  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
526  * use for this function is in keepalives, which use tcp_respond.
527  */
528 struct tcptemp *
529 tcpip_maketemplate(struct inpcb *inp)
530 {
531 	struct tcptemp *t;
532 
533 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
534 	if (t == NULL)
535 		return (NULL);
536 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
537 	return (t);
538 }
539 
540 /*
541  * Send a single message to the TCP at address specified by
542  * the given TCP/IP header.  If m == NULL, then we make a copy
543  * of the tcpiphdr at ti and send directly to the addressed host.
544  * This is used to force keep alive messages out using the TCP
545  * template for a connection.  If flags are given then we send
546  * a message back to the TCP which originated the * segment ti,
547  * and discard the mbuf containing it and any other attached mbufs.
548  *
549  * In any case the ack and sequence number of the transmitted
550  * segment are as specified by the parameters.
551  *
552  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
553  */
554 void
555 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
556     tcp_seq ack, tcp_seq seq, int flags)
557 {
558 	int tlen;
559 	int win = 0;
560 	struct ip *ip;
561 	struct tcphdr *nth;
562 #ifdef INET6
563 	struct ip6_hdr *ip6;
564 	int isipv6;
565 #endif /* INET6 */
566 	int ipflags = 0;
567 	struct inpcb *inp;
568 
569 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
570 
571 #ifdef INET6
572 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
573 	ip6 = ipgen;
574 #endif /* INET6 */
575 	ip = ipgen;
576 
577 	if (tp != NULL) {
578 		inp = tp->t_inpcb;
579 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
580 		INP_WLOCK_ASSERT(inp);
581 	} else
582 		inp = NULL;
583 
584 	if (tp != NULL) {
585 		if (!(flags & TH_RST)) {
586 			win = sbspace(&inp->inp_socket->so_rcv);
587 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
588 				win = (long)TCP_MAXWIN << tp->rcv_scale;
589 		}
590 	}
591 	if (m == NULL) {
592 		m = m_gethdr(M_NOWAIT, MT_DATA);
593 		if (m == NULL)
594 			return;
595 		tlen = 0;
596 		m->m_data += max_linkhdr;
597 #ifdef INET6
598 		if (isipv6) {
599 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
600 			      sizeof(struct ip6_hdr));
601 			ip6 = mtod(m, struct ip6_hdr *);
602 			nth = (struct tcphdr *)(ip6 + 1);
603 		} else
604 #endif /* INET6 */
605 		{
606 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
607 			ip = mtod(m, struct ip *);
608 			nth = (struct tcphdr *)(ip + 1);
609 		}
610 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
611 		flags = TH_ACK;
612 	} else {
613 		/*
614 		 *  reuse the mbuf.
615 		 * XXX MRT We inherrit the FIB, which is lucky.
616 		 */
617 		m_freem(m->m_next);
618 		m->m_next = NULL;
619 		m->m_data = (caddr_t)ipgen;
620 		/* m_len is set later */
621 		tlen = 0;
622 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
623 #ifdef INET6
624 		if (isipv6) {
625 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
626 			nth = (struct tcphdr *)(ip6 + 1);
627 		} else
628 #endif /* INET6 */
629 		{
630 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
631 			nth = (struct tcphdr *)(ip + 1);
632 		}
633 		if (th != nth) {
634 			/*
635 			 * this is usually a case when an extension header
636 			 * exists between the IPv6 header and the
637 			 * TCP header.
638 			 */
639 			nth->th_sport = th->th_sport;
640 			nth->th_dport = th->th_dport;
641 		}
642 		xchg(nth->th_dport, nth->th_sport, uint16_t);
643 #undef xchg
644 	}
645 #ifdef INET6
646 	if (isipv6) {
647 		ip6->ip6_flow = 0;
648 		ip6->ip6_vfc = IPV6_VERSION;
649 		ip6->ip6_nxt = IPPROTO_TCP;
650 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
651 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
652 	}
653 #endif
654 #if defined(INET) && defined(INET6)
655 	else
656 #endif
657 #ifdef INET
658 	{
659 		tlen += sizeof (struct tcpiphdr);
660 		ip->ip_len = htons(tlen);
661 		ip->ip_ttl = V_ip_defttl;
662 		if (V_path_mtu_discovery)
663 			ip->ip_off |= htons(IP_DF);
664 	}
665 #endif
666 	m->m_len = tlen;
667 	m->m_pkthdr.len = tlen;
668 	m->m_pkthdr.rcvif = NULL;
669 #ifdef MAC
670 	if (inp != NULL) {
671 		/*
672 		 * Packet is associated with a socket, so allow the
673 		 * label of the response to reflect the socket label.
674 		 */
675 		INP_WLOCK_ASSERT(inp);
676 		mac_inpcb_create_mbuf(inp, m);
677 	} else {
678 		/*
679 		 * Packet is not associated with a socket, so possibly
680 		 * update the label in place.
681 		 */
682 		mac_netinet_tcp_reply(m);
683 	}
684 #endif
685 	nth->th_seq = htonl(seq);
686 	nth->th_ack = htonl(ack);
687 	nth->th_x2 = 0;
688 	nth->th_off = sizeof (struct tcphdr) >> 2;
689 	nth->th_flags = flags;
690 	if (tp != NULL)
691 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
692 	else
693 		nth->th_win = htons((u_short)win);
694 	nth->th_urp = 0;
695 
696 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
697 #ifdef INET6
698 	if (isipv6) {
699 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
700 		nth->th_sum = in6_cksum_pseudo(ip6,
701 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
702 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
703 		    NULL, NULL);
704 	}
705 #endif /* INET6 */
706 #if defined(INET6) && defined(INET)
707 	else
708 #endif
709 #ifdef INET
710 	{
711 		m->m_pkthdr.csum_flags = CSUM_TCP;
712 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
713 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
714 	}
715 #endif /* INET */
716 #ifdef TCPDEBUG
717 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
718 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
719 #endif
720 	if (flags & TH_RST)
721 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
722 		    tp, nth);
723 
724 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
725 #ifdef INET6
726 	if (isipv6)
727 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
728 #endif /* INET6 */
729 #if defined(INET) && defined(INET6)
730 	else
731 #endif
732 #ifdef INET
733 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
734 #endif
735 }
736 
737 /*
738  * Create a new TCP control block, making an
739  * empty reassembly queue and hooking it to the argument
740  * protocol control block.  The `inp' parameter must have
741  * come from the zone allocator set up in tcp_init().
742  */
743 struct tcpcb *
744 tcp_newtcpcb(struct inpcb *inp)
745 {
746 	struct tcpcb_mem *tm;
747 	struct tcpcb *tp;
748 #ifdef INET6
749 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
750 #endif /* INET6 */
751 
752 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
753 	if (tm == NULL)
754 		return (NULL);
755 	tp = &tm->tcb;
756 
757 	/* Initialise cc_var struct for this tcpcb. */
758 	tp->ccv = &tm->ccv;
759 	tp->ccv->type = IPPROTO_TCP;
760 	tp->ccv->ccvc.tcp = tp;
761 
762 	/*
763 	 * Use the current system default CC algorithm.
764 	 */
765 	CC_LIST_RLOCK();
766 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
767 	CC_ALGO(tp) = CC_DEFAULT();
768 	CC_LIST_RUNLOCK();
769 
770 	if (CC_ALGO(tp)->cb_init != NULL)
771 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
772 			uma_zfree(V_tcpcb_zone, tm);
773 			return (NULL);
774 		}
775 
776 	tp->osd = &tm->osd;
777 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
778 		uma_zfree(V_tcpcb_zone, tm);
779 		return (NULL);
780 	}
781 
782 #ifdef VIMAGE
783 	tp->t_vnet = inp->inp_vnet;
784 #endif
785 	tp->t_timers = &tm->tt;
786 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
787 	tp->t_maxseg = tp->t_maxopd =
788 #ifdef INET6
789 		isipv6 ? V_tcp_v6mssdflt :
790 #endif /* INET6 */
791 		V_tcp_mssdflt;
792 
793 	/* Set up our timeouts. */
794 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
795 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
796 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
797 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
798 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
799 
800 	if (V_tcp_do_rfc1323)
801 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
802 	if (V_tcp_do_sack)
803 		tp->t_flags |= TF_SACK_PERMIT;
804 	TAILQ_INIT(&tp->snd_holes);
805 	tp->t_inpcb = inp;	/* XXX */
806 	/*
807 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
808 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
809 	 * reasonable initial retransmit time.
810 	 */
811 	tp->t_srtt = TCPTV_SRTTBASE;
812 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
813 	tp->t_rttmin = tcp_rexmit_min;
814 	tp->t_rxtcur = TCPTV_RTOBASE;
815 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
816 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
817 	tp->t_rcvtime = ticks;
818 	/*
819 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
820 	 * because the socket may be bound to an IPv6 wildcard address,
821 	 * which may match an IPv4-mapped IPv6 address.
822 	 */
823 	inp->inp_ip_ttl = V_ip_defttl;
824 	inp->inp_ppcb = tp;
825 	return (tp);		/* XXX */
826 }
827 
828 /*
829  * Switch the congestion control algorithm back to NewReno for any active
830  * control blocks using an algorithm which is about to go away.
831  * This ensures the CC framework can allow the unload to proceed without leaving
832  * any dangling pointers which would trigger a panic.
833  * Returning non-zero would inform the CC framework that something went wrong
834  * and it would be unsafe to allow the unload to proceed. However, there is no
835  * way for this to occur with this implementation so we always return zero.
836  */
837 int
838 tcp_ccalgounload(struct cc_algo *unload_algo)
839 {
840 	struct cc_algo *tmpalgo;
841 	struct inpcb *inp;
842 	struct tcpcb *tp;
843 	VNET_ITERATOR_DECL(vnet_iter);
844 
845 	/*
846 	 * Check all active control blocks across all network stacks and change
847 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
848 	 * requires cleanup code to be run, call it.
849 	 */
850 	VNET_LIST_RLOCK();
851 	VNET_FOREACH(vnet_iter) {
852 		CURVNET_SET(vnet_iter);
853 		INP_INFO_RLOCK(&V_tcbinfo);
854 		/*
855 		 * New connections already part way through being initialised
856 		 * with the CC algo we're removing will not race with this code
857 		 * because the INP_INFO_WLOCK is held during initialisation. We
858 		 * therefore don't enter the loop below until the connection
859 		 * list has stabilised.
860 		 */
861 		LIST_FOREACH(inp, &V_tcb, inp_list) {
862 			INP_WLOCK(inp);
863 			/* Important to skip tcptw structs. */
864 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
865 			    (tp = intotcpcb(inp)) != NULL) {
866 				/*
867 				 * By holding INP_WLOCK here, we are assured
868 				 * that the connection is not currently
869 				 * executing inside the CC module's functions
870 				 * i.e. it is safe to make the switch back to
871 				 * NewReno.
872 				 */
873 				if (CC_ALGO(tp) == unload_algo) {
874 					tmpalgo = CC_ALGO(tp);
875 					/* NewReno does not require any init. */
876 					CC_ALGO(tp) = &newreno_cc_algo;
877 					if (tmpalgo->cb_destroy != NULL)
878 						tmpalgo->cb_destroy(tp->ccv);
879 				}
880 			}
881 			INP_WUNLOCK(inp);
882 		}
883 		INP_INFO_RUNLOCK(&V_tcbinfo);
884 		CURVNET_RESTORE();
885 	}
886 	VNET_LIST_RUNLOCK();
887 
888 	return (0);
889 }
890 
891 /*
892  * Drop a TCP connection, reporting
893  * the specified error.  If connection is synchronized,
894  * then send a RST to peer.
895  */
896 struct tcpcb *
897 tcp_drop(struct tcpcb *tp, int errno)
898 {
899 	struct socket *so = tp->t_inpcb->inp_socket;
900 
901 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
902 	INP_WLOCK_ASSERT(tp->t_inpcb);
903 
904 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
905 		tcp_state_change(tp, TCPS_CLOSED);
906 		(void) tcp_output(tp);
907 		TCPSTAT_INC(tcps_drops);
908 	} else
909 		TCPSTAT_INC(tcps_conndrops);
910 	if (errno == ETIMEDOUT && tp->t_softerror)
911 		errno = tp->t_softerror;
912 	so->so_error = errno;
913 	return (tcp_close(tp));
914 }
915 
916 void
917 tcp_discardcb(struct tcpcb *tp)
918 {
919 	struct inpcb *inp = tp->t_inpcb;
920 	struct socket *so = inp->inp_socket;
921 #ifdef INET6
922 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
923 #endif /* INET6 */
924 
925 	INP_WLOCK_ASSERT(inp);
926 
927 	/*
928 	 * Make sure that all of our timers are stopped before we delete the
929 	 * PCB.
930 	 *
931 	 * XXXRW: Really, we would like to use callout_drain() here in order
932 	 * to avoid races experienced in tcp_timer.c where a timer is already
933 	 * executing at this point.  However, we can't, both because we're
934 	 * running in a context where we can't sleep, and also because we
935 	 * hold locks required by the timers.  What we instead need to do is
936 	 * test to see if callout_drain() is required, and if so, defer some
937 	 * portion of the remainder of tcp_discardcb() to an asynchronous
938 	 * context that can callout_drain() and then continue.  Some care
939 	 * will be required to ensure that no further processing takes place
940 	 * on the tcpcb, even though it hasn't been freed (a flag?).
941 	 */
942 	callout_stop(&tp->t_timers->tt_rexmt);
943 	callout_stop(&tp->t_timers->tt_persist);
944 	callout_stop(&tp->t_timers->tt_keep);
945 	callout_stop(&tp->t_timers->tt_2msl);
946 	callout_stop(&tp->t_timers->tt_delack);
947 
948 	/*
949 	 * If we got enough samples through the srtt filter,
950 	 * save the rtt and rttvar in the routing entry.
951 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
952 	 * 4 samples is enough for the srtt filter to converge
953 	 * to within enough % of the correct value; fewer samples
954 	 * and we could save a bogus rtt. The danger is not high
955 	 * as tcp quickly recovers from everything.
956 	 * XXX: Works very well but needs some more statistics!
957 	 */
958 	if (tp->t_rttupdated >= 4) {
959 		struct hc_metrics_lite metrics;
960 		u_long ssthresh;
961 
962 		bzero(&metrics, sizeof(metrics));
963 		/*
964 		 * Update the ssthresh always when the conditions below
965 		 * are satisfied. This gives us better new start value
966 		 * for the congestion avoidance for new connections.
967 		 * ssthresh is only set if packet loss occured on a session.
968 		 *
969 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
970 		 * being torn down.  Ideally this code would not use 'so'.
971 		 */
972 		ssthresh = tp->snd_ssthresh;
973 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
974 			/*
975 			 * convert the limit from user data bytes to
976 			 * packets then to packet data bytes.
977 			 */
978 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
979 			if (ssthresh < 2)
980 				ssthresh = 2;
981 			ssthresh *= (u_long)(tp->t_maxseg +
982 #ifdef INET6
983 			    (isipv6 ? sizeof (struct ip6_hdr) +
984 				sizeof (struct tcphdr) :
985 #endif
986 				sizeof (struct tcpiphdr)
987 #ifdef INET6
988 			    )
989 #endif
990 			    );
991 		} else
992 			ssthresh = 0;
993 		metrics.rmx_ssthresh = ssthresh;
994 
995 		metrics.rmx_rtt = tp->t_srtt;
996 		metrics.rmx_rttvar = tp->t_rttvar;
997 		metrics.rmx_cwnd = tp->snd_cwnd;
998 		metrics.rmx_sendpipe = 0;
999 		metrics.rmx_recvpipe = 0;
1000 
1001 		tcp_hc_update(&inp->inp_inc, &metrics);
1002 	}
1003 
1004 	/* free the reassembly queue, if any */
1005 	tcp_reass_flush(tp);
1006 
1007 #ifdef TCP_OFFLOAD
1008 	/* Disconnect offload device, if any. */
1009 	if (tp->t_flags & TF_TOE)
1010 		tcp_offload_detach(tp);
1011 #endif
1012 
1013 	tcp_free_sackholes(tp);
1014 
1015 	/* Allow the CC algorithm to clean up after itself. */
1016 	if (CC_ALGO(tp)->cb_destroy != NULL)
1017 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1018 
1019 	khelp_destroy_osd(tp->osd);
1020 
1021 	CC_ALGO(tp) = NULL;
1022 	inp->inp_ppcb = NULL;
1023 	tp->t_inpcb = NULL;
1024 	uma_zfree(V_tcpcb_zone, tp);
1025 }
1026 
1027 /*
1028  * Attempt to close a TCP control block, marking it as dropped, and freeing
1029  * the socket if we hold the only reference.
1030  */
1031 struct tcpcb *
1032 tcp_close(struct tcpcb *tp)
1033 {
1034 	struct inpcb *inp = tp->t_inpcb;
1035 	struct socket *so;
1036 
1037 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1038 	INP_WLOCK_ASSERT(inp);
1039 
1040 #ifdef TCP_OFFLOAD
1041 	if (tp->t_state == TCPS_LISTEN)
1042 		tcp_offload_listen_stop(tp);
1043 #endif
1044 	in_pcbdrop(inp);
1045 	TCPSTAT_INC(tcps_closed);
1046 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1047 	so = inp->inp_socket;
1048 	soisdisconnected(so);
1049 	if (inp->inp_flags & INP_SOCKREF) {
1050 		KASSERT(so->so_state & SS_PROTOREF,
1051 		    ("tcp_close: !SS_PROTOREF"));
1052 		inp->inp_flags &= ~INP_SOCKREF;
1053 		INP_WUNLOCK(inp);
1054 		ACCEPT_LOCK();
1055 		SOCK_LOCK(so);
1056 		so->so_state &= ~SS_PROTOREF;
1057 		sofree(so);
1058 		return (NULL);
1059 	}
1060 	return (tp);
1061 }
1062 
1063 void
1064 tcp_drain(void)
1065 {
1066 	VNET_ITERATOR_DECL(vnet_iter);
1067 
1068 	if (!do_tcpdrain)
1069 		return;
1070 
1071 	VNET_LIST_RLOCK_NOSLEEP();
1072 	VNET_FOREACH(vnet_iter) {
1073 		CURVNET_SET(vnet_iter);
1074 		struct inpcb *inpb;
1075 		struct tcpcb *tcpb;
1076 
1077 	/*
1078 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1079 	 * if there is one...
1080 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1081 	 *      reassembly queue should be flushed, but in a situation
1082 	 *	where we're really low on mbufs, this is potentially
1083 	 *	useful.
1084 	 */
1085 		INP_INFO_RLOCK(&V_tcbinfo);
1086 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1087 			if (inpb->inp_flags & INP_TIMEWAIT)
1088 				continue;
1089 			INP_WLOCK(inpb);
1090 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1091 				tcp_reass_flush(tcpb);
1092 				tcp_clean_sackreport(tcpb);
1093 			}
1094 			INP_WUNLOCK(inpb);
1095 		}
1096 		INP_INFO_RUNLOCK(&V_tcbinfo);
1097 		CURVNET_RESTORE();
1098 	}
1099 	VNET_LIST_RUNLOCK_NOSLEEP();
1100 }
1101 
1102 /*
1103  * Notify a tcp user of an asynchronous error;
1104  * store error as soft error, but wake up user
1105  * (for now, won't do anything until can select for soft error).
1106  *
1107  * Do not wake up user since there currently is no mechanism for
1108  * reporting soft errors (yet - a kqueue filter may be added).
1109  */
1110 static struct inpcb *
1111 tcp_notify(struct inpcb *inp, int error)
1112 {
1113 	struct tcpcb *tp;
1114 
1115 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1116 	INP_WLOCK_ASSERT(inp);
1117 
1118 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1119 	    (inp->inp_flags & INP_DROPPED))
1120 		return (inp);
1121 
1122 	tp = intotcpcb(inp);
1123 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1124 
1125 	/*
1126 	 * Ignore some errors if we are hooked up.
1127 	 * If connection hasn't completed, has retransmitted several times,
1128 	 * and receives a second error, give up now.  This is better
1129 	 * than waiting a long time to establish a connection that
1130 	 * can never complete.
1131 	 */
1132 	if (tp->t_state == TCPS_ESTABLISHED &&
1133 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1134 	     error == EHOSTDOWN)) {
1135 		return (inp);
1136 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1137 	    tp->t_softerror) {
1138 		tp = tcp_drop(tp, error);
1139 		if (tp != NULL)
1140 			return (inp);
1141 		else
1142 			return (NULL);
1143 	} else {
1144 		tp->t_softerror = error;
1145 		return (inp);
1146 	}
1147 #if 0
1148 	wakeup( &so->so_timeo);
1149 	sorwakeup(so);
1150 	sowwakeup(so);
1151 #endif
1152 }
1153 
1154 static int
1155 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1156 {
1157 	int error, i, m, n, pcb_count;
1158 	struct inpcb *inp, **inp_list;
1159 	inp_gen_t gencnt;
1160 	struct xinpgen xig;
1161 
1162 	/*
1163 	 * The process of preparing the TCB list is too time-consuming and
1164 	 * resource-intensive to repeat twice on every request.
1165 	 */
1166 	if (req->oldptr == NULL) {
1167 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1168 		n += imax(n / 8, 10);
1169 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1170 		return (0);
1171 	}
1172 
1173 	if (req->newptr != NULL)
1174 		return (EPERM);
1175 
1176 	/*
1177 	 * OK, now we're committed to doing something.
1178 	 */
1179 	INP_INFO_RLOCK(&V_tcbinfo);
1180 	gencnt = V_tcbinfo.ipi_gencnt;
1181 	n = V_tcbinfo.ipi_count;
1182 	INP_INFO_RUNLOCK(&V_tcbinfo);
1183 
1184 	m = syncache_pcbcount();
1185 
1186 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1187 		+ (n + m) * sizeof(struct xtcpcb));
1188 	if (error != 0)
1189 		return (error);
1190 
1191 	xig.xig_len = sizeof xig;
1192 	xig.xig_count = n + m;
1193 	xig.xig_gen = gencnt;
1194 	xig.xig_sogen = so_gencnt;
1195 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1196 	if (error)
1197 		return (error);
1198 
1199 	error = syncache_pcblist(req, m, &pcb_count);
1200 	if (error)
1201 		return (error);
1202 
1203 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1204 	if (inp_list == NULL)
1205 		return (ENOMEM);
1206 
1207 	INP_INFO_RLOCK(&V_tcbinfo);
1208 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1209 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1210 		INP_WLOCK(inp);
1211 		if (inp->inp_gencnt <= gencnt) {
1212 			/*
1213 			 * XXX: This use of cr_cansee(), introduced with
1214 			 * TCP state changes, is not quite right, but for
1215 			 * now, better than nothing.
1216 			 */
1217 			if (inp->inp_flags & INP_TIMEWAIT) {
1218 				if (intotw(inp) != NULL)
1219 					error = cr_cansee(req->td->td_ucred,
1220 					    intotw(inp)->tw_cred);
1221 				else
1222 					error = EINVAL;	/* Skip this inp. */
1223 			} else
1224 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1225 			if (error == 0) {
1226 				in_pcbref(inp);
1227 				inp_list[i++] = inp;
1228 			}
1229 		}
1230 		INP_WUNLOCK(inp);
1231 	}
1232 	INP_INFO_RUNLOCK(&V_tcbinfo);
1233 	n = i;
1234 
1235 	error = 0;
1236 	for (i = 0; i < n; i++) {
1237 		inp = inp_list[i];
1238 		INP_RLOCK(inp);
1239 		if (inp->inp_gencnt <= gencnt) {
1240 			struct xtcpcb xt;
1241 			void *inp_ppcb;
1242 
1243 			bzero(&xt, sizeof(xt));
1244 			xt.xt_len = sizeof xt;
1245 			/* XXX should avoid extra copy */
1246 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1247 			inp_ppcb = inp->inp_ppcb;
1248 			if (inp_ppcb == NULL)
1249 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1250 			else if (inp->inp_flags & INP_TIMEWAIT) {
1251 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1252 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1253 			} else {
1254 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1255 				if (xt.xt_tp.t_timers)
1256 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1257 			}
1258 			if (inp->inp_socket != NULL)
1259 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1260 			else {
1261 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1262 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1263 			}
1264 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1265 			INP_RUNLOCK(inp);
1266 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1267 		} else
1268 			INP_RUNLOCK(inp);
1269 	}
1270 	INP_INFO_WLOCK(&V_tcbinfo);
1271 	for (i = 0; i < n; i++) {
1272 		inp = inp_list[i];
1273 		INP_RLOCK(inp);
1274 		if (!in_pcbrele_rlocked(inp))
1275 			INP_RUNLOCK(inp);
1276 	}
1277 	INP_INFO_WUNLOCK(&V_tcbinfo);
1278 
1279 	if (!error) {
1280 		/*
1281 		 * Give the user an updated idea of our state.
1282 		 * If the generation differs from what we told
1283 		 * her before, she knows that something happened
1284 		 * while we were processing this request, and it
1285 		 * might be necessary to retry.
1286 		 */
1287 		INP_INFO_RLOCK(&V_tcbinfo);
1288 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1289 		xig.xig_sogen = so_gencnt;
1290 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1291 		INP_INFO_RUNLOCK(&V_tcbinfo);
1292 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1293 	}
1294 	free(inp_list, M_TEMP);
1295 	return (error);
1296 }
1297 
1298 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1299     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1300     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1301 
1302 #ifdef INET
1303 static int
1304 tcp_getcred(SYSCTL_HANDLER_ARGS)
1305 {
1306 	struct xucred xuc;
1307 	struct sockaddr_in addrs[2];
1308 	struct inpcb *inp;
1309 	int error;
1310 
1311 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1312 	if (error)
1313 		return (error);
1314 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1315 	if (error)
1316 		return (error);
1317 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1318 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1319 	if (inp != NULL) {
1320 		if (inp->inp_socket == NULL)
1321 			error = ENOENT;
1322 		if (error == 0)
1323 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1324 		if (error == 0)
1325 			cru2x(inp->inp_cred, &xuc);
1326 		INP_RUNLOCK(inp);
1327 	} else
1328 		error = ENOENT;
1329 	if (error == 0)
1330 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1331 	return (error);
1332 }
1333 
1334 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1335     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1336     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1337 #endif /* INET */
1338 
1339 #ifdef INET6
1340 static int
1341 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1342 {
1343 	struct xucred xuc;
1344 	struct sockaddr_in6 addrs[2];
1345 	struct inpcb *inp;
1346 	int error;
1347 #ifdef INET
1348 	int mapped = 0;
1349 #endif
1350 
1351 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1352 	if (error)
1353 		return (error);
1354 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1355 	if (error)
1356 		return (error);
1357 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1358 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1359 		return (error);
1360 	}
1361 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1362 #ifdef INET
1363 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1364 			mapped = 1;
1365 		else
1366 #endif
1367 			return (EINVAL);
1368 	}
1369 
1370 #ifdef INET
1371 	if (mapped == 1)
1372 		inp = in_pcblookup(&V_tcbinfo,
1373 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1374 			addrs[1].sin6_port,
1375 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1376 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1377 	else
1378 #endif
1379 		inp = in6_pcblookup(&V_tcbinfo,
1380 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1381 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1382 			INPLOOKUP_RLOCKPCB, NULL);
1383 	if (inp != NULL) {
1384 		if (inp->inp_socket == NULL)
1385 			error = ENOENT;
1386 		if (error == 0)
1387 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1388 		if (error == 0)
1389 			cru2x(inp->inp_cred, &xuc);
1390 		INP_RUNLOCK(inp);
1391 	} else
1392 		error = ENOENT;
1393 	if (error == 0)
1394 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1395 	return (error);
1396 }
1397 
1398 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1399     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1400     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1401 #endif /* INET6 */
1402 
1403 
1404 #ifdef INET
1405 void
1406 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1407 {
1408 	struct ip *ip = vip;
1409 	struct tcphdr *th;
1410 	struct in_addr faddr;
1411 	struct inpcb *inp;
1412 	struct tcpcb *tp;
1413 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1414 	struct icmp *icp;
1415 	struct in_conninfo inc;
1416 	tcp_seq icmp_tcp_seq;
1417 	int mtu;
1418 
1419 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1420 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1421 		return;
1422 
1423 	if (cmd == PRC_MSGSIZE)
1424 		notify = tcp_mtudisc_notify;
1425 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1426 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1427 		notify = tcp_drop_syn_sent;
1428 	/*
1429 	 * Redirects don't need to be handled up here.
1430 	 */
1431 	else if (PRC_IS_REDIRECT(cmd))
1432 		return;
1433 	/*
1434 	 * Source quench is depreciated.
1435 	 */
1436 	else if (cmd == PRC_QUENCH)
1437 		return;
1438 	/*
1439 	 * Hostdead is ugly because it goes linearly through all PCBs.
1440 	 * XXX: We never get this from ICMP, otherwise it makes an
1441 	 * excellent DoS attack on machines with many connections.
1442 	 */
1443 	else if (cmd == PRC_HOSTDEAD)
1444 		ip = NULL;
1445 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1446 		return;
1447 	if (ip != NULL) {
1448 		icp = (struct icmp *)((caddr_t)ip
1449 				      - offsetof(struct icmp, icmp_ip));
1450 		th = (struct tcphdr *)((caddr_t)ip
1451 				       + (ip->ip_hl << 2));
1452 		INP_INFO_WLOCK(&V_tcbinfo);
1453 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1454 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1455 		if (inp != NULL)  {
1456 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1457 			    !(inp->inp_flags & INP_DROPPED) &&
1458 			    !(inp->inp_socket == NULL)) {
1459 				icmp_tcp_seq = htonl(th->th_seq);
1460 				tp = intotcpcb(inp);
1461 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1462 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1463 					if (cmd == PRC_MSGSIZE) {
1464 					    /*
1465 					     * MTU discovery:
1466 					     * If we got a needfrag set the MTU
1467 					     * in the route to the suggested new
1468 					     * value (if given) and then notify.
1469 					     */
1470 					    bzero(&inc, sizeof(inc));
1471 					    inc.inc_faddr = faddr;
1472 					    inc.inc_fibnum =
1473 						inp->inp_inc.inc_fibnum;
1474 
1475 					    mtu = ntohs(icp->icmp_nextmtu);
1476 					    /*
1477 					     * If no alternative MTU was
1478 					     * proposed, try the next smaller
1479 					     * one.
1480 					     */
1481 					    if (!mtu)
1482 						mtu = ip_next_mtu(
1483 						 ntohs(ip->ip_len), 1);
1484 					    if (mtu < V_tcp_minmss
1485 						 + sizeof(struct tcpiphdr))
1486 						mtu = V_tcp_minmss
1487 						 + sizeof(struct tcpiphdr);
1488 					    /*
1489 					     * Only cache the MTU if it
1490 					     * is smaller than the interface
1491 					     * or route MTU.  tcp_mtudisc()
1492 					     * will do right thing by itself.
1493 					     */
1494 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1495 						tcp_hc_updatemtu(&inc, mtu);
1496 					    tcp_mtudisc(inp, mtu);
1497 					} else
1498 						inp = (*notify)(inp,
1499 						    inetctlerrmap[cmd]);
1500 				}
1501 			}
1502 			if (inp != NULL)
1503 				INP_WUNLOCK(inp);
1504 		} else {
1505 			bzero(&inc, sizeof(inc));
1506 			inc.inc_fport = th->th_dport;
1507 			inc.inc_lport = th->th_sport;
1508 			inc.inc_faddr = faddr;
1509 			inc.inc_laddr = ip->ip_src;
1510 			syncache_unreach(&inc, th);
1511 		}
1512 		INP_INFO_WUNLOCK(&V_tcbinfo);
1513 	} else
1514 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1515 }
1516 #endif /* INET */
1517 
1518 #ifdef INET6
1519 void
1520 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1521 {
1522 	struct tcphdr th;
1523 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1524 	struct ip6_hdr *ip6;
1525 	struct mbuf *m;
1526 	struct ip6ctlparam *ip6cp = NULL;
1527 	const struct sockaddr_in6 *sa6_src = NULL;
1528 	int off;
1529 	struct tcp_portonly {
1530 		u_int16_t th_sport;
1531 		u_int16_t th_dport;
1532 	} *thp;
1533 
1534 	if (sa->sa_family != AF_INET6 ||
1535 	    sa->sa_len != sizeof(struct sockaddr_in6))
1536 		return;
1537 
1538 	if (cmd == PRC_MSGSIZE)
1539 		notify = tcp_mtudisc_notify;
1540 	else if (!PRC_IS_REDIRECT(cmd) &&
1541 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1542 		return;
1543 	/* Source quench is depreciated. */
1544 	else if (cmd == PRC_QUENCH)
1545 		return;
1546 
1547 	/* if the parameter is from icmp6, decode it. */
1548 	if (d != NULL) {
1549 		ip6cp = (struct ip6ctlparam *)d;
1550 		m = ip6cp->ip6c_m;
1551 		ip6 = ip6cp->ip6c_ip6;
1552 		off = ip6cp->ip6c_off;
1553 		sa6_src = ip6cp->ip6c_src;
1554 	} else {
1555 		m = NULL;
1556 		ip6 = NULL;
1557 		off = 0;	/* fool gcc */
1558 		sa6_src = &sa6_any;
1559 	}
1560 
1561 	if (ip6 != NULL) {
1562 		struct in_conninfo inc;
1563 		/*
1564 		 * XXX: We assume that when IPV6 is non NULL,
1565 		 * M and OFF are valid.
1566 		 */
1567 
1568 		/* check if we can safely examine src and dst ports */
1569 		if (m->m_pkthdr.len < off + sizeof(*thp))
1570 			return;
1571 
1572 		bzero(&th, sizeof(th));
1573 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1574 
1575 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1576 		    (struct sockaddr *)ip6cp->ip6c_src,
1577 		    th.th_sport, cmd, NULL, notify);
1578 
1579 		bzero(&inc, sizeof(inc));
1580 		inc.inc_fport = th.th_dport;
1581 		inc.inc_lport = th.th_sport;
1582 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1583 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1584 		inc.inc_flags |= INC_ISIPV6;
1585 		INP_INFO_WLOCK(&V_tcbinfo);
1586 		syncache_unreach(&inc, &th);
1587 		INP_INFO_WUNLOCK(&V_tcbinfo);
1588 	} else
1589 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1590 			      0, cmd, NULL, notify);
1591 }
1592 #endif /* INET6 */
1593 
1594 
1595 /*
1596  * Following is where TCP initial sequence number generation occurs.
1597  *
1598  * There are two places where we must use initial sequence numbers:
1599  * 1.  In SYN-ACK packets.
1600  * 2.  In SYN packets.
1601  *
1602  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1603  * tcp_syncache.c for details.
1604  *
1605  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1606  * depends on this property.  In addition, these ISNs should be
1607  * unguessable so as to prevent connection hijacking.  To satisfy
1608  * the requirements of this situation, the algorithm outlined in
1609  * RFC 1948 is used, with only small modifications.
1610  *
1611  * Implementation details:
1612  *
1613  * Time is based off the system timer, and is corrected so that it
1614  * increases by one megabyte per second.  This allows for proper
1615  * recycling on high speed LANs while still leaving over an hour
1616  * before rollover.
1617  *
1618  * As reading the *exact* system time is too expensive to be done
1619  * whenever setting up a TCP connection, we increment the time
1620  * offset in two ways.  First, a small random positive increment
1621  * is added to isn_offset for each connection that is set up.
1622  * Second, the function tcp_isn_tick fires once per clock tick
1623  * and increments isn_offset as necessary so that sequence numbers
1624  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1625  * random positive increments serve only to ensure that the same
1626  * exact sequence number is never sent out twice (as could otherwise
1627  * happen when a port is recycled in less than the system tick
1628  * interval.)
1629  *
1630  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1631  * between seeding of isn_secret.  This is normally set to zero,
1632  * as reseeding should not be necessary.
1633  *
1634  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1635  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1636  * general, this means holding an exclusive (write) lock.
1637  */
1638 
1639 #define ISN_BYTES_PER_SECOND 1048576
1640 #define ISN_STATIC_INCREMENT 4096
1641 #define ISN_RANDOM_INCREMENT (4096 - 1)
1642 
1643 static VNET_DEFINE(u_char, isn_secret[32]);
1644 static VNET_DEFINE(int, isn_last);
1645 static VNET_DEFINE(int, isn_last_reseed);
1646 static VNET_DEFINE(u_int32_t, isn_offset);
1647 static VNET_DEFINE(u_int32_t, isn_offset_old);
1648 
1649 #define	V_isn_secret			VNET(isn_secret)
1650 #define	V_isn_last			VNET(isn_last)
1651 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1652 #define	V_isn_offset			VNET(isn_offset)
1653 #define	V_isn_offset_old		VNET(isn_offset_old)
1654 
1655 tcp_seq
1656 tcp_new_isn(struct tcpcb *tp)
1657 {
1658 	MD5_CTX isn_ctx;
1659 	u_int32_t md5_buffer[4];
1660 	tcp_seq new_isn;
1661 	u_int32_t projected_offset;
1662 
1663 	INP_WLOCK_ASSERT(tp->t_inpcb);
1664 
1665 	ISN_LOCK();
1666 	/* Seed if this is the first use, reseed if requested. */
1667 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1668 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1669 		< (u_int)ticks))) {
1670 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1671 		V_isn_last_reseed = ticks;
1672 	}
1673 
1674 	/* Compute the md5 hash and return the ISN. */
1675 	MD5Init(&isn_ctx);
1676 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1677 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1678 #ifdef INET6
1679 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1680 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1681 			  sizeof(struct in6_addr));
1682 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1683 			  sizeof(struct in6_addr));
1684 	} else
1685 #endif
1686 	{
1687 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1688 			  sizeof(struct in_addr));
1689 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1690 			  sizeof(struct in_addr));
1691 	}
1692 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1693 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1694 	new_isn = (tcp_seq) md5_buffer[0];
1695 	V_isn_offset += ISN_STATIC_INCREMENT +
1696 		(arc4random() & ISN_RANDOM_INCREMENT);
1697 	if (ticks != V_isn_last) {
1698 		projected_offset = V_isn_offset_old +
1699 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1700 		if (SEQ_GT(projected_offset, V_isn_offset))
1701 			V_isn_offset = projected_offset;
1702 		V_isn_offset_old = V_isn_offset;
1703 		V_isn_last = ticks;
1704 	}
1705 	new_isn += V_isn_offset;
1706 	ISN_UNLOCK();
1707 	return (new_isn);
1708 }
1709 
1710 /*
1711  * When a specific ICMP unreachable message is received and the
1712  * connection state is SYN-SENT, drop the connection.  This behavior
1713  * is controlled by the icmp_may_rst sysctl.
1714  */
1715 struct inpcb *
1716 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1717 {
1718 	struct tcpcb *tp;
1719 
1720 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1721 	INP_WLOCK_ASSERT(inp);
1722 
1723 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1724 	    (inp->inp_flags & INP_DROPPED))
1725 		return (inp);
1726 
1727 	tp = intotcpcb(inp);
1728 	if (tp->t_state != TCPS_SYN_SENT)
1729 		return (inp);
1730 
1731 	tp = tcp_drop(tp, errno);
1732 	if (tp != NULL)
1733 		return (inp);
1734 	else
1735 		return (NULL);
1736 }
1737 
1738 /*
1739  * When `need fragmentation' ICMP is received, update our idea of the MSS
1740  * based on the new value. Also nudge TCP to send something, since we
1741  * know the packet we just sent was dropped.
1742  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1743  */
1744 static struct inpcb *
1745 tcp_mtudisc_notify(struct inpcb *inp, int error)
1746 {
1747 
1748 	return (tcp_mtudisc(inp, -1));
1749 }
1750 
1751 struct inpcb *
1752 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1753 {
1754 	struct tcpcb *tp;
1755 	struct socket *so;
1756 
1757 	INP_WLOCK_ASSERT(inp);
1758 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1759 	    (inp->inp_flags & INP_DROPPED))
1760 		return (inp);
1761 
1762 	tp = intotcpcb(inp);
1763 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1764 
1765 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1766 
1767 	so = inp->inp_socket;
1768 	SOCKBUF_LOCK(&so->so_snd);
1769 	/* If the mss is larger than the socket buffer, decrease the mss. */
1770 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1771 		tp->t_maxseg = so->so_snd.sb_hiwat;
1772 	SOCKBUF_UNLOCK(&so->so_snd);
1773 
1774 	TCPSTAT_INC(tcps_mturesent);
1775 	tp->t_rtttime = 0;
1776 	tp->snd_nxt = tp->snd_una;
1777 	tcp_free_sackholes(tp);
1778 	tp->snd_recover = tp->snd_max;
1779 	if (tp->t_flags & TF_SACK_PERMIT)
1780 		EXIT_FASTRECOVERY(tp->t_flags);
1781 	tcp_output(tp);
1782 	return (inp);
1783 }
1784 
1785 #ifdef INET
1786 /*
1787  * Look-up the routing entry to the peer of this inpcb.  If no route
1788  * is found and it cannot be allocated, then return 0.  This routine
1789  * is called by TCP routines that access the rmx structure and by
1790  * tcp_mss_update to get the peer/interface MTU.
1791  */
1792 u_long
1793 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1794 {
1795 	struct route sro;
1796 	struct sockaddr_in *dst;
1797 	struct ifnet *ifp;
1798 	u_long maxmtu = 0;
1799 
1800 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1801 
1802 	bzero(&sro, sizeof(sro));
1803 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1804 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1805 		dst->sin_family = AF_INET;
1806 		dst->sin_len = sizeof(*dst);
1807 		dst->sin_addr = inc->inc_faddr;
1808 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1809 	}
1810 	if (sro.ro_rt != NULL) {
1811 		ifp = sro.ro_rt->rt_ifp;
1812 		if (sro.ro_rt->rt_mtu == 0)
1813 			maxmtu = ifp->if_mtu;
1814 		else
1815 			maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1816 
1817 		/* Report additional interface capabilities. */
1818 		if (cap != NULL) {
1819 			if (ifp->if_capenable & IFCAP_TSO4 &&
1820 			    ifp->if_hwassist & CSUM_TSO) {
1821 				cap->ifcap |= CSUM_TSO;
1822 				cap->tsomax = ifp->if_hw_tsomax;
1823 			}
1824 		}
1825 		RTFREE(sro.ro_rt);
1826 	}
1827 	return (maxmtu);
1828 }
1829 #endif /* INET */
1830 
1831 #ifdef INET6
1832 u_long
1833 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1834 {
1835 	struct route_in6 sro6;
1836 	struct ifnet *ifp;
1837 	u_long maxmtu = 0;
1838 
1839 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1840 
1841 	bzero(&sro6, sizeof(sro6));
1842 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1843 		sro6.ro_dst.sin6_family = AF_INET6;
1844 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1845 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1846 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1847 	}
1848 	if (sro6.ro_rt != NULL) {
1849 		ifp = sro6.ro_rt->rt_ifp;
1850 		if (sro6.ro_rt->rt_mtu == 0)
1851 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1852 		else
1853 			maxmtu = min(sro6.ro_rt->rt_mtu,
1854 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1855 
1856 		/* Report additional interface capabilities. */
1857 		if (cap != NULL) {
1858 			if (ifp->if_capenable & IFCAP_TSO6 &&
1859 			    ifp->if_hwassist & CSUM_TSO) {
1860 				cap->ifcap |= CSUM_TSO;
1861 				cap->tsomax = ifp->if_hw_tsomax;
1862 			}
1863 		}
1864 		RTFREE(sro6.ro_rt);
1865 	}
1866 
1867 	return (maxmtu);
1868 }
1869 #endif /* INET6 */
1870 
1871 #ifdef IPSEC
1872 /* compute ESP/AH header size for TCP, including outer IP header. */
1873 size_t
1874 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1875 {
1876 	struct inpcb *inp;
1877 	struct mbuf *m;
1878 	size_t hdrsiz;
1879 	struct ip *ip;
1880 #ifdef INET6
1881 	struct ip6_hdr *ip6;
1882 #endif
1883 	struct tcphdr *th;
1884 
1885 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1886 		return (0);
1887 	m = m_gethdr(M_NOWAIT, MT_DATA);
1888 	if (!m)
1889 		return (0);
1890 
1891 #ifdef INET6
1892 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1893 		ip6 = mtod(m, struct ip6_hdr *);
1894 		th = (struct tcphdr *)(ip6 + 1);
1895 		m->m_pkthdr.len = m->m_len =
1896 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1897 		tcpip_fillheaders(inp, ip6, th);
1898 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1899 	} else
1900 #endif /* INET6 */
1901 	{
1902 		ip = mtod(m, struct ip *);
1903 		th = (struct tcphdr *)(ip + 1);
1904 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1905 		tcpip_fillheaders(inp, ip, th);
1906 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1907 	}
1908 
1909 	m_free(m);
1910 	return (hdrsiz);
1911 }
1912 #endif /* IPSEC */
1913 
1914 #ifdef TCP_SIGNATURE
1915 /*
1916  * Callback function invoked by m_apply() to digest TCP segment data
1917  * contained within an mbuf chain.
1918  */
1919 static int
1920 tcp_signature_apply(void *fstate, void *data, u_int len)
1921 {
1922 
1923 	MD5Update(fstate, (u_char *)data, len);
1924 	return (0);
1925 }
1926 
1927 /*
1928  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1929  *
1930  * Parameters:
1931  * m		pointer to head of mbuf chain
1932  * _unused
1933  * len		length of TCP segment data, excluding options
1934  * optlen	length of TCP segment options
1935  * buf		pointer to storage for computed MD5 digest
1936  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1937  *
1938  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1939  * When called from tcp_input(), we can be sure that th_sum has been
1940  * zeroed out and verified already.
1941  *
1942  * Return 0 if successful, otherwise return -1.
1943  *
1944  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1945  * search with the destination IP address, and a 'magic SPI' to be
1946  * determined by the application. This is hardcoded elsewhere to 1179
1947  * right now. Another branch of this code exists which uses the SPD to
1948  * specify per-application flows but it is unstable.
1949  */
1950 int
1951 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1952     u_char *buf, u_int direction)
1953 {
1954 	union sockaddr_union dst;
1955 #ifdef INET
1956 	struct ippseudo ippseudo;
1957 #endif
1958 	MD5_CTX ctx;
1959 	int doff;
1960 	struct ip *ip;
1961 #ifdef INET
1962 	struct ipovly *ipovly;
1963 #endif
1964 	struct secasvar *sav;
1965 	struct tcphdr *th;
1966 #ifdef INET6
1967 	struct ip6_hdr *ip6;
1968 	struct in6_addr in6;
1969 	char ip6buf[INET6_ADDRSTRLEN];
1970 	uint32_t plen;
1971 	uint16_t nhdr;
1972 #endif
1973 	u_short savecsum;
1974 
1975 	KASSERT(m != NULL, ("NULL mbuf chain"));
1976 	KASSERT(buf != NULL, ("NULL signature pointer"));
1977 
1978 	/* Extract the destination from the IP header in the mbuf. */
1979 	bzero(&dst, sizeof(union sockaddr_union));
1980 	ip = mtod(m, struct ip *);
1981 #ifdef INET6
1982 	ip6 = NULL;	/* Make the compiler happy. */
1983 #endif
1984 	switch (ip->ip_v) {
1985 #ifdef INET
1986 	case IPVERSION:
1987 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1988 		dst.sa.sa_family = AF_INET;
1989 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1990 		    ip->ip_src : ip->ip_dst;
1991 		break;
1992 #endif
1993 #ifdef INET6
1994 	case (IPV6_VERSION >> 4):
1995 		ip6 = mtod(m, struct ip6_hdr *);
1996 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1997 		dst.sa.sa_family = AF_INET6;
1998 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1999 		    ip6->ip6_src : ip6->ip6_dst;
2000 		break;
2001 #endif
2002 	default:
2003 		return (EINVAL);
2004 		/* NOTREACHED */
2005 		break;
2006 	}
2007 
2008 	/* Look up an SADB entry which matches the address of the peer. */
2009 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2010 	if (sav == NULL) {
2011 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2012 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2013 #ifdef INET6
2014 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2015 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2016 #endif
2017 			"(unsupported)"));
2018 		return (EINVAL);
2019 	}
2020 
2021 	MD5Init(&ctx);
2022 	/*
2023 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2024 	 *
2025 	 * XXX The ippseudo header MUST be digested in network byte order,
2026 	 * or else we'll fail the regression test. Assume all fields we've
2027 	 * been doing arithmetic on have been in host byte order.
2028 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2029 	 * tcp_output(), the underlying ip_len member has not yet been set.
2030 	 */
2031 	switch (ip->ip_v) {
2032 #ifdef INET
2033 	case IPVERSION:
2034 		ipovly = (struct ipovly *)ip;
2035 		ippseudo.ippseudo_src = ipovly->ih_src;
2036 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2037 		ippseudo.ippseudo_pad = 0;
2038 		ippseudo.ippseudo_p = IPPROTO_TCP;
2039 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2040 		    optlen);
2041 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2042 
2043 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2044 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2045 		break;
2046 #endif
2047 #ifdef INET6
2048 	/*
2049 	 * RFC 2385, 2.0  Proposal
2050 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2051 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2052 	 * extended next header value (to form 32 bits), and 32-bit segment
2053 	 * length.
2054 	 * Note: Upper-Layer Packet Length comes before Next Header.
2055 	 */
2056 	case (IPV6_VERSION >> 4):
2057 		in6 = ip6->ip6_src;
2058 		in6_clearscope(&in6);
2059 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2060 		in6 = ip6->ip6_dst;
2061 		in6_clearscope(&in6);
2062 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2063 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2064 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2065 		nhdr = 0;
2066 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2067 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2068 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2069 		nhdr = IPPROTO_TCP;
2070 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071 
2072 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2073 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2074 		break;
2075 #endif
2076 	default:
2077 		return (EINVAL);
2078 		/* NOTREACHED */
2079 		break;
2080 	}
2081 
2082 
2083 	/*
2084 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2085 	 * The TCP checksum must be set to zero.
2086 	 */
2087 	savecsum = th->th_sum;
2088 	th->th_sum = 0;
2089 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2090 	th->th_sum = savecsum;
2091 
2092 	/*
2093 	 * Step 3: Update MD5 hash with TCP segment data.
2094 	 *         Use m_apply() to avoid an early m_pullup().
2095 	 */
2096 	if (len > 0)
2097 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2098 
2099 	/*
2100 	 * Step 4: Update MD5 hash with shared secret.
2101 	 */
2102 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2103 	MD5Final(buf, &ctx);
2104 
2105 	key_sa_recordxfer(sav, m);
2106 	KEY_FREESAV(&sav);
2107 	return (0);
2108 }
2109 
2110 /*
2111  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2112  *
2113  * Parameters:
2114  * m		pointer to head of mbuf chain
2115  * len		length of TCP segment data, excluding options
2116  * optlen	length of TCP segment options
2117  * buf		pointer to storage for computed MD5 digest
2118  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2119  *
2120  * Return 1 if successful, otherwise return 0.
2121  */
2122 int
2123 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2124     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2125 {
2126 	char tmpdigest[TCP_SIGLEN];
2127 
2128 	if (tcp_sig_checksigs == 0)
2129 		return (1);
2130 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2131 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2132 
2133 			/*
2134 			 * If this socket is not expecting signature but
2135 			 * the segment contains signature just fail.
2136 			 */
2137 			TCPSTAT_INC(tcps_sig_err_sigopt);
2138 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2139 			return (0);
2140 		}
2141 
2142 		/* Signature is not expected, and not present in segment. */
2143 		return (1);
2144 	}
2145 
2146 	/*
2147 	 * If this socket is expecting signature but the segment does not
2148 	 * contain any just fail.
2149 	 */
2150 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2151 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2152 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2153 		return (0);
2154 	}
2155 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2156 	    IPSEC_DIR_INBOUND) == -1) {
2157 		TCPSTAT_INC(tcps_sig_err_buildsig);
2158 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2159 		return (0);
2160 	}
2161 
2162 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2163 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2164 		return (0);
2165 	}
2166 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2167 	return (1);
2168 }
2169 #endif /* TCP_SIGNATURE */
2170 
2171 static int
2172 sysctl_drop(SYSCTL_HANDLER_ARGS)
2173 {
2174 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2175 	struct sockaddr_storage addrs[2];
2176 	struct inpcb *inp;
2177 	struct tcpcb *tp;
2178 	struct tcptw *tw;
2179 	struct sockaddr_in *fin, *lin;
2180 #ifdef INET6
2181 	struct sockaddr_in6 *fin6, *lin6;
2182 #endif
2183 	int error;
2184 
2185 	inp = NULL;
2186 	fin = lin = NULL;
2187 #ifdef INET6
2188 	fin6 = lin6 = NULL;
2189 #endif
2190 	error = 0;
2191 
2192 	if (req->oldptr != NULL || req->oldlen != 0)
2193 		return (EINVAL);
2194 	if (req->newptr == NULL)
2195 		return (EPERM);
2196 	if (req->newlen < sizeof(addrs))
2197 		return (ENOMEM);
2198 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2199 	if (error)
2200 		return (error);
2201 
2202 	switch (addrs[0].ss_family) {
2203 #ifdef INET6
2204 	case AF_INET6:
2205 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2206 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2207 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2208 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2209 			return (EINVAL);
2210 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2211 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2212 				return (EINVAL);
2213 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2214 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2215 			fin = (struct sockaddr_in *)&addrs[0];
2216 			lin = (struct sockaddr_in *)&addrs[1];
2217 			break;
2218 		}
2219 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2220 		if (error)
2221 			return (error);
2222 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2223 		if (error)
2224 			return (error);
2225 		break;
2226 #endif
2227 #ifdef INET
2228 	case AF_INET:
2229 		fin = (struct sockaddr_in *)&addrs[0];
2230 		lin = (struct sockaddr_in *)&addrs[1];
2231 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2232 		    lin->sin_len != sizeof(struct sockaddr_in))
2233 			return (EINVAL);
2234 		break;
2235 #endif
2236 	default:
2237 		return (EINVAL);
2238 	}
2239 	INP_INFO_WLOCK(&V_tcbinfo);
2240 	switch (addrs[0].ss_family) {
2241 #ifdef INET6
2242 	case AF_INET6:
2243 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2244 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2245 		    INPLOOKUP_WLOCKPCB, NULL);
2246 		break;
2247 #endif
2248 #ifdef INET
2249 	case AF_INET:
2250 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2251 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2252 		break;
2253 #endif
2254 	}
2255 	if (inp != NULL) {
2256 		if (inp->inp_flags & INP_TIMEWAIT) {
2257 			/*
2258 			 * XXXRW: There currently exists a state where an
2259 			 * inpcb is present, but its timewait state has been
2260 			 * discarded.  For now, don't allow dropping of this
2261 			 * type of inpcb.
2262 			 */
2263 			tw = intotw(inp);
2264 			if (tw != NULL)
2265 				tcp_twclose(tw, 0);
2266 			else
2267 				INP_WUNLOCK(inp);
2268 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2269 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2270 			tp = intotcpcb(inp);
2271 			tp = tcp_drop(tp, ECONNABORTED);
2272 			if (tp != NULL)
2273 				INP_WUNLOCK(inp);
2274 		} else
2275 			INP_WUNLOCK(inp);
2276 	} else
2277 		error = ESRCH;
2278 	INP_INFO_WUNLOCK(&V_tcbinfo);
2279 	return (error);
2280 }
2281 
2282 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2283     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2284     0, sysctl_drop, "", "Drop TCP connection");
2285 
2286 /*
2287  * Generate a standardized TCP log line for use throughout the
2288  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2289  * allow use in the interrupt context.
2290  *
2291  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2292  * NB: The function may return NULL if memory allocation failed.
2293  *
2294  * Due to header inclusion and ordering limitations the struct ip
2295  * and ip6_hdr pointers have to be passed as void pointers.
2296  */
2297 char *
2298 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2299     const void *ip6hdr)
2300 {
2301 
2302 	/* Is logging enabled? */
2303 	if (tcp_log_in_vain == 0)
2304 		return (NULL);
2305 
2306 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2307 }
2308 
2309 char *
2310 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2311     const void *ip6hdr)
2312 {
2313 
2314 	/* Is logging enabled? */
2315 	if (tcp_log_debug == 0)
2316 		return (NULL);
2317 
2318 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2319 }
2320 
2321 static char *
2322 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2323     const void *ip6hdr)
2324 {
2325 	char *s, *sp;
2326 	size_t size;
2327 	struct ip *ip;
2328 #ifdef INET6
2329 	const struct ip6_hdr *ip6;
2330 
2331 	ip6 = (const struct ip6_hdr *)ip6hdr;
2332 #endif /* INET6 */
2333 	ip = (struct ip *)ip4hdr;
2334 
2335 	/*
2336 	 * The log line looks like this:
2337 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2338 	 */
2339 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2340 	    sizeof(PRINT_TH_FLAGS) + 1 +
2341 #ifdef INET6
2342 	    2 * INET6_ADDRSTRLEN;
2343 #else
2344 	    2 * INET_ADDRSTRLEN;
2345 #endif /* INET6 */
2346 
2347 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2348 	if (s == NULL)
2349 		return (NULL);
2350 
2351 	strcat(s, "TCP: [");
2352 	sp = s + strlen(s);
2353 
2354 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2355 		inet_ntoa_r(inc->inc_faddr, sp);
2356 		sp = s + strlen(s);
2357 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2358 		sp = s + strlen(s);
2359 		inet_ntoa_r(inc->inc_laddr, sp);
2360 		sp = s + strlen(s);
2361 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2362 #ifdef INET6
2363 	} else if (inc) {
2364 		ip6_sprintf(sp, &inc->inc6_faddr);
2365 		sp = s + strlen(s);
2366 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2367 		sp = s + strlen(s);
2368 		ip6_sprintf(sp, &inc->inc6_laddr);
2369 		sp = s + strlen(s);
2370 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2371 	} else if (ip6 && th) {
2372 		ip6_sprintf(sp, &ip6->ip6_src);
2373 		sp = s + strlen(s);
2374 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2375 		sp = s + strlen(s);
2376 		ip6_sprintf(sp, &ip6->ip6_dst);
2377 		sp = s + strlen(s);
2378 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2379 #endif /* INET6 */
2380 #ifdef INET
2381 	} else if (ip && th) {
2382 		inet_ntoa_r(ip->ip_src, sp);
2383 		sp = s + strlen(s);
2384 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2385 		sp = s + strlen(s);
2386 		inet_ntoa_r(ip->ip_dst, sp);
2387 		sp = s + strlen(s);
2388 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2389 #endif /* INET */
2390 	} else {
2391 		free(s, M_TCPLOG);
2392 		return (NULL);
2393 	}
2394 	sp = s + strlen(s);
2395 	if (th)
2396 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2397 	if (*(s + size - 1) != '\0')
2398 		panic("%s: string too long", __func__);
2399 	return (s);
2400 }
2401 
2402 /*
2403  * A subroutine which makes it easy to track TCP state changes with DTrace.
2404  * This function shouldn't be called for t_state initializations that don't
2405  * correspond to actual TCP state transitions.
2406  */
2407 void
2408 tcp_state_change(struct tcpcb *tp, int newstate)
2409 {
2410 #if defined(KDTRACE_HOOKS)
2411 	int pstate = tp->t_state;
2412 #endif
2413 
2414 	tp->t_state = newstate;
2415 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2416 }
2417