xref: /freebsd/sys/netinet/tcp_subr.c (revision 3823d5e198425b4f5e5a80267d195769d1063773)
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;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
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;
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 	if (tcp_soreceive_stream) {
402 #ifdef INET
403 		tcp_usrreqs.pru_soreceive = soreceive_stream;
404 #endif
405 #ifdef INET6
406 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
407 #endif /* INET6 */
408 	}
409 
410 #ifdef INET6
411 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
412 #else /* INET6 */
413 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
414 #endif /* INET6 */
415 	if (max_protohdr < TCP_MINPROTOHDR)
416 		max_protohdr = TCP_MINPROTOHDR;
417 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
418 		panic("tcp_init");
419 #undef TCP_MINPROTOHDR
420 
421 	ISN_LOCK_INIT();
422 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
423 		SHUTDOWN_PRI_DEFAULT);
424 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
425 		EVENTHANDLER_PRI_ANY);
426 }
427 
428 #ifdef VIMAGE
429 void
430 tcp_destroy(void)
431 {
432 	int error;
433 
434 	tcp_hc_destroy();
435 	syncache_destroy();
436 	tcp_tw_destroy();
437 	in_pcbinfo_destroy(&V_tcbinfo);
438 	uma_zdestroy(V_sack_hole_zone);
439 	uma_zdestroy(V_tcpcb_zone);
440 
441 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
442 	if (error != 0) {
443 		printf("%s: WARNING: unable to deregister helper hook "
444 		    "type=%d, id=%d: error %d returned\n", __func__,
445 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
446 	}
447 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
448 	if (error != 0) {
449 		printf("%s: WARNING: unable to deregister helper hook "
450 		    "type=%d, id=%d: error %d returned\n", __func__,
451 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
452 	}
453 }
454 #endif
455 
456 void
457 tcp_fini(void *xtp)
458 {
459 
460 }
461 
462 /*
463  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
464  * tcp_template used to store this data in mbufs, but we now recopy it out
465  * of the tcpcb each time to conserve mbufs.
466  */
467 void
468 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
469 {
470 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
471 
472 	INP_WLOCK_ASSERT(inp);
473 
474 #ifdef INET6
475 	if ((inp->inp_vflag & INP_IPV6) != 0) {
476 		struct ip6_hdr *ip6;
477 
478 		ip6 = (struct ip6_hdr *)ip_ptr;
479 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
480 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
481 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
482 			(IPV6_VERSION & IPV6_VERSION_MASK);
483 		ip6->ip6_nxt = IPPROTO_TCP;
484 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
485 		ip6->ip6_src = inp->in6p_laddr;
486 		ip6->ip6_dst = inp->in6p_faddr;
487 	}
488 #endif /* INET6 */
489 #if defined(INET6) && defined(INET)
490 	else
491 #endif
492 #ifdef INET
493 	{
494 		struct ip *ip;
495 
496 		ip = (struct ip *)ip_ptr;
497 		ip->ip_v = IPVERSION;
498 		ip->ip_hl = 5;
499 		ip->ip_tos = inp->inp_ip_tos;
500 		ip->ip_len = 0;
501 		ip->ip_id = 0;
502 		ip->ip_off = 0;
503 		ip->ip_ttl = inp->inp_ip_ttl;
504 		ip->ip_sum = 0;
505 		ip->ip_p = IPPROTO_TCP;
506 		ip->ip_src = inp->inp_laddr;
507 		ip->ip_dst = inp->inp_faddr;
508 	}
509 #endif /* INET */
510 	th->th_sport = inp->inp_lport;
511 	th->th_dport = inp->inp_fport;
512 	th->th_seq = 0;
513 	th->th_ack = 0;
514 	th->th_x2 = 0;
515 	th->th_off = 5;
516 	th->th_flags = 0;
517 	th->th_win = 0;
518 	th->th_urp = 0;
519 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
520 }
521 
522 /*
523  * Create template to be used to send tcp packets on a connection.
524  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
525  * use for this function is in keepalives, which use tcp_respond.
526  */
527 struct tcptemp *
528 tcpip_maketemplate(struct inpcb *inp)
529 {
530 	struct tcptemp *t;
531 
532 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
533 	if (t == NULL)
534 		return (NULL);
535 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
536 	return (t);
537 }
538 
539 /*
540  * Send a single message to the TCP at address specified by
541  * the given TCP/IP header.  If m == NULL, then we make a copy
542  * of the tcpiphdr at th and send directly to the addressed host.
543  * This is used to force keep alive messages out using the TCP
544  * template for a connection.  If flags are given then we send
545  * a message back to the TCP which originated the segment th,
546  * and discard the mbuf containing it and any other attached mbufs.
547  *
548  * In any case the ack and sequence number of the transmitted
549  * segment are as specified by the parameters.
550  *
551  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
552  */
553 void
554 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
555     tcp_seq ack, tcp_seq seq, int flags)
556 {
557 	int tlen;
558 	int win = 0;
559 	struct ip *ip;
560 	struct tcphdr *nth;
561 #ifdef INET6
562 	struct ip6_hdr *ip6;
563 	int isipv6;
564 #endif /* INET6 */
565 	int ipflags = 0;
566 	struct inpcb *inp;
567 
568 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
569 
570 #ifdef INET6
571 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
572 	ip6 = ipgen;
573 #endif /* INET6 */
574 	ip = ipgen;
575 
576 	if (tp != NULL) {
577 		inp = tp->t_inpcb;
578 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
579 		INP_WLOCK_ASSERT(inp);
580 	} else
581 		inp = NULL;
582 
583 	if (tp != NULL) {
584 		if (!(flags & TH_RST)) {
585 			win = sbspace(&inp->inp_socket->so_rcv);
586 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
587 				win = (long)TCP_MAXWIN << tp->rcv_scale;
588 		}
589 	}
590 	if (m == NULL) {
591 		m = m_gethdr(M_NOWAIT, MT_DATA);
592 		if (m == NULL)
593 			return;
594 		tlen = 0;
595 		m->m_data += max_linkhdr;
596 #ifdef INET6
597 		if (isipv6) {
598 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
599 			      sizeof(struct ip6_hdr));
600 			ip6 = mtod(m, struct ip6_hdr *);
601 			nth = (struct tcphdr *)(ip6 + 1);
602 		} else
603 #endif /* INET6 */
604 		{
605 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
606 			ip = mtod(m, struct ip *);
607 			nth = (struct tcphdr *)(ip + 1);
608 		}
609 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
610 		flags = TH_ACK;
611 	} else {
612 		/*
613 		 *  reuse the mbuf.
614 		 * XXX MRT We inherrit the FIB, which is lucky.
615 		 */
616 		m_freem(m->m_next);
617 		m->m_next = NULL;
618 		m->m_data = (caddr_t)ipgen;
619 		/* m_len is set later */
620 		tlen = 0;
621 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
622 #ifdef INET6
623 		if (isipv6) {
624 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
625 			nth = (struct tcphdr *)(ip6 + 1);
626 		} else
627 #endif /* INET6 */
628 		{
629 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
630 			nth = (struct tcphdr *)(ip + 1);
631 		}
632 		if (th != nth) {
633 			/*
634 			 * this is usually a case when an extension header
635 			 * exists between the IPv6 header and the
636 			 * TCP header.
637 			 */
638 			nth->th_sport = th->th_sport;
639 			nth->th_dport = th->th_dport;
640 		}
641 		xchg(nth->th_dport, nth->th_sport, uint16_t);
642 #undef xchg
643 	}
644 #ifdef INET6
645 	if (isipv6) {
646 		ip6->ip6_flow = 0;
647 		ip6->ip6_vfc = IPV6_VERSION;
648 		ip6->ip6_nxt = IPPROTO_TCP;
649 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
650 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
651 	}
652 #endif
653 #if defined(INET) && defined(INET6)
654 	else
655 #endif
656 #ifdef INET
657 	{
658 		tlen += sizeof (struct tcpiphdr);
659 		ip->ip_len = htons(tlen);
660 		ip->ip_ttl = V_ip_defttl;
661 		if (V_path_mtu_discovery)
662 			ip->ip_off |= htons(IP_DF);
663 	}
664 #endif
665 	m->m_len = tlen;
666 	m->m_pkthdr.len = tlen;
667 	m->m_pkthdr.rcvif = NULL;
668 #ifdef MAC
669 	if (inp != NULL) {
670 		/*
671 		 * Packet is associated with a socket, so allow the
672 		 * label of the response to reflect the socket label.
673 		 */
674 		INP_WLOCK_ASSERT(inp);
675 		mac_inpcb_create_mbuf(inp, m);
676 	} else {
677 		/*
678 		 * Packet is not associated with a socket, so possibly
679 		 * update the label in place.
680 		 */
681 		mac_netinet_tcp_reply(m);
682 	}
683 #endif
684 	nth->th_seq = htonl(seq);
685 	nth->th_ack = htonl(ack);
686 	nth->th_x2 = 0;
687 	nth->th_off = sizeof (struct tcphdr) >> 2;
688 	nth->th_flags = flags;
689 	if (tp != NULL)
690 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
691 	else
692 		nth->th_win = htons((u_short)win);
693 	nth->th_urp = 0;
694 
695 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
696 #ifdef INET6
697 	if (isipv6) {
698 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
699 		nth->th_sum = in6_cksum_pseudo(ip6,
700 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
701 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
702 		    NULL, NULL);
703 	}
704 #endif /* INET6 */
705 #if defined(INET6) && defined(INET)
706 	else
707 #endif
708 #ifdef INET
709 	{
710 		m->m_pkthdr.csum_flags = CSUM_TCP;
711 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
712 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
713 	}
714 #endif /* INET */
715 #ifdef TCPDEBUG
716 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
717 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
718 #endif
719 	if (flags & TH_RST)
720 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
721 		    tp, nth);
722 
723 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
724 #ifdef INET6
725 	if (isipv6)
726 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
727 #endif /* INET6 */
728 #if defined(INET) && defined(INET6)
729 	else
730 #endif
731 #ifdef INET
732 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
733 #endif
734 }
735 
736 /*
737  * Create a new TCP control block, making an
738  * empty reassembly queue and hooking it to the argument
739  * protocol control block.  The `inp' parameter must have
740  * come from the zone allocator set up in tcp_init().
741  */
742 struct tcpcb *
743 tcp_newtcpcb(struct inpcb *inp)
744 {
745 	struct tcpcb_mem *tm;
746 	struct tcpcb *tp;
747 #ifdef INET6
748 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
749 #endif /* INET6 */
750 
751 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
752 	if (tm == NULL)
753 		return (NULL);
754 	tp = &tm->tcb;
755 
756 	/* Initialise cc_var struct for this tcpcb. */
757 	tp->ccv = &tm->ccv;
758 	tp->ccv->type = IPPROTO_TCP;
759 	tp->ccv->ccvc.tcp = tp;
760 
761 	/*
762 	 * Use the current system default CC algorithm.
763 	 */
764 	CC_LIST_RLOCK();
765 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
766 	CC_ALGO(tp) = CC_DEFAULT();
767 	CC_LIST_RUNLOCK();
768 
769 	if (CC_ALGO(tp)->cb_init != NULL)
770 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
771 			uma_zfree(V_tcpcb_zone, tm);
772 			return (NULL);
773 		}
774 
775 	tp->osd = &tm->osd;
776 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
777 		uma_zfree(V_tcpcb_zone, tm);
778 		return (NULL);
779 	}
780 
781 #ifdef VIMAGE
782 	tp->t_vnet = inp->inp_vnet;
783 #endif
784 	tp->t_timers = &tm->tt;
785 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
786 	tp->t_maxseg = tp->t_maxopd =
787 #ifdef INET6
788 		isipv6 ? V_tcp_v6mssdflt :
789 #endif /* INET6 */
790 		V_tcp_mssdflt;
791 
792 	/* Set up our timeouts. */
793 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
794 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
795 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
796 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
797 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
798 
799 	if (V_tcp_do_rfc1323)
800 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
801 	if (V_tcp_do_sack)
802 		tp->t_flags |= TF_SACK_PERMIT;
803 	TAILQ_INIT(&tp->snd_holes);
804 	tp->t_inpcb = inp;	/* XXX */
805 	/*
806 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
807 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
808 	 * reasonable initial retransmit time.
809 	 */
810 	tp->t_srtt = TCPTV_SRTTBASE;
811 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
812 	tp->t_rttmin = tcp_rexmit_min;
813 	tp->t_rxtcur = TCPTV_RTOBASE;
814 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
815 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
816 	tp->t_rcvtime = ticks;
817 	/*
818 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
819 	 * because the socket may be bound to an IPv6 wildcard address,
820 	 * which may match an IPv4-mapped IPv6 address.
821 	 */
822 	inp->inp_ip_ttl = V_ip_defttl;
823 	inp->inp_ppcb = tp;
824 	return (tp);		/* XXX */
825 }
826 
827 /*
828  * Switch the congestion control algorithm back to NewReno for any active
829  * control blocks using an algorithm which is about to go away.
830  * This ensures the CC framework can allow the unload to proceed without leaving
831  * any dangling pointers which would trigger a panic.
832  * Returning non-zero would inform the CC framework that something went wrong
833  * and it would be unsafe to allow the unload to proceed. However, there is no
834  * way for this to occur with this implementation so we always return zero.
835  */
836 int
837 tcp_ccalgounload(struct cc_algo *unload_algo)
838 {
839 	struct cc_algo *tmpalgo;
840 	struct inpcb *inp;
841 	struct tcpcb *tp;
842 	VNET_ITERATOR_DECL(vnet_iter);
843 
844 	/*
845 	 * Check all active control blocks across all network stacks and change
846 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
847 	 * requires cleanup code to be run, call it.
848 	 */
849 	VNET_LIST_RLOCK();
850 	VNET_FOREACH(vnet_iter) {
851 		CURVNET_SET(vnet_iter);
852 		INP_INFO_RLOCK(&V_tcbinfo);
853 		/*
854 		 * New connections already part way through being initialised
855 		 * with the CC algo we're removing will not race with this code
856 		 * because the INP_INFO_WLOCK is held during initialisation. We
857 		 * therefore don't enter the loop below until the connection
858 		 * list has stabilised.
859 		 */
860 		LIST_FOREACH(inp, &V_tcb, inp_list) {
861 			INP_WLOCK(inp);
862 			/* Important to skip tcptw structs. */
863 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
864 			    (tp = intotcpcb(inp)) != NULL) {
865 				/*
866 				 * By holding INP_WLOCK here, we are assured
867 				 * that the connection is not currently
868 				 * executing inside the CC module's functions
869 				 * i.e. it is safe to make the switch back to
870 				 * NewReno.
871 				 */
872 				if (CC_ALGO(tp) == unload_algo) {
873 					tmpalgo = CC_ALGO(tp);
874 					/* NewReno does not require any init. */
875 					CC_ALGO(tp) = &newreno_cc_algo;
876 					if (tmpalgo->cb_destroy != NULL)
877 						tmpalgo->cb_destroy(tp->ccv);
878 				}
879 			}
880 			INP_WUNLOCK(inp);
881 		}
882 		INP_INFO_RUNLOCK(&V_tcbinfo);
883 		CURVNET_RESTORE();
884 	}
885 	VNET_LIST_RUNLOCK();
886 
887 	return (0);
888 }
889 
890 /*
891  * Drop a TCP connection, reporting
892  * the specified error.  If connection is synchronized,
893  * then send a RST to peer.
894  */
895 struct tcpcb *
896 tcp_drop(struct tcpcb *tp, int errno)
897 {
898 	struct socket *so = tp->t_inpcb->inp_socket;
899 
900 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
901 	INP_WLOCK_ASSERT(tp->t_inpcb);
902 
903 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
904 		tcp_state_change(tp, TCPS_CLOSED);
905 		(void) tcp_output(tp);
906 		TCPSTAT_INC(tcps_drops);
907 	} else
908 		TCPSTAT_INC(tcps_conndrops);
909 	if (errno == ETIMEDOUT && tp->t_softerror)
910 		errno = tp->t_softerror;
911 	so->so_error = errno;
912 	return (tcp_close(tp));
913 }
914 
915 void
916 tcp_discardcb(struct tcpcb *tp)
917 {
918 	struct inpcb *inp = tp->t_inpcb;
919 	struct socket *so = inp->inp_socket;
920 #ifdef INET6
921 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
922 #endif /* INET6 */
923 
924 	INP_WLOCK_ASSERT(inp);
925 
926 	/*
927 	 * Make sure that all of our timers are stopped before we delete the
928 	 * PCB.
929 	 *
930 	 * XXXRW: Really, we would like to use callout_drain() here in order
931 	 * to avoid races experienced in tcp_timer.c where a timer is already
932 	 * executing at this point.  However, we can't, both because we're
933 	 * running in a context where we can't sleep, and also because we
934 	 * hold locks required by the timers.  What we instead need to do is
935 	 * test to see if callout_drain() is required, and if so, defer some
936 	 * portion of the remainder of tcp_discardcb() to an asynchronous
937 	 * context that can callout_drain() and then continue.  Some care
938 	 * will be required to ensure that no further processing takes place
939 	 * on the tcpcb, even though it hasn't been freed (a flag?).
940 	 */
941 	callout_stop(&tp->t_timers->tt_rexmt);
942 	callout_stop(&tp->t_timers->tt_persist);
943 	callout_stop(&tp->t_timers->tt_keep);
944 	callout_stop(&tp->t_timers->tt_2msl);
945 	callout_stop(&tp->t_timers->tt_delack);
946 
947 	/*
948 	 * If we got enough samples through the srtt filter,
949 	 * save the rtt and rttvar in the routing entry.
950 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
951 	 * 4 samples is enough for the srtt filter to converge
952 	 * to within enough % of the correct value; fewer samples
953 	 * and we could save a bogus rtt. The danger is not high
954 	 * as tcp quickly recovers from everything.
955 	 * XXX: Works very well but needs some more statistics!
956 	 */
957 	if (tp->t_rttupdated >= 4) {
958 		struct hc_metrics_lite metrics;
959 		u_long ssthresh;
960 
961 		bzero(&metrics, sizeof(metrics));
962 		/*
963 		 * Update the ssthresh always when the conditions below
964 		 * are satisfied. This gives us better new start value
965 		 * for the congestion avoidance for new connections.
966 		 * ssthresh is only set if packet loss occured on a session.
967 		 *
968 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
969 		 * being torn down.  Ideally this code would not use 'so'.
970 		 */
971 		ssthresh = tp->snd_ssthresh;
972 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
973 			/*
974 			 * convert the limit from user data bytes to
975 			 * packets then to packet data bytes.
976 			 */
977 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
978 			if (ssthresh < 2)
979 				ssthresh = 2;
980 			ssthresh *= (u_long)(tp->t_maxseg +
981 #ifdef INET6
982 			    (isipv6 ? sizeof (struct ip6_hdr) +
983 				sizeof (struct tcphdr) :
984 #endif
985 				sizeof (struct tcpiphdr)
986 #ifdef INET6
987 			    )
988 #endif
989 			    );
990 		} else
991 			ssthresh = 0;
992 		metrics.rmx_ssthresh = ssthresh;
993 
994 		metrics.rmx_rtt = tp->t_srtt;
995 		metrics.rmx_rttvar = tp->t_rttvar;
996 		metrics.rmx_cwnd = tp->snd_cwnd;
997 		metrics.rmx_sendpipe = 0;
998 		metrics.rmx_recvpipe = 0;
999 
1000 		tcp_hc_update(&inp->inp_inc, &metrics);
1001 	}
1002 
1003 	/* free the reassembly queue, if any */
1004 	tcp_reass_flush(tp);
1005 
1006 #ifdef TCP_OFFLOAD
1007 	/* Disconnect offload device, if any. */
1008 	if (tp->t_flags & TF_TOE)
1009 		tcp_offload_detach(tp);
1010 #endif
1011 
1012 	tcp_free_sackholes(tp);
1013 
1014 	/* Allow the CC algorithm to clean up after itself. */
1015 	if (CC_ALGO(tp)->cb_destroy != NULL)
1016 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1017 
1018 	khelp_destroy_osd(tp->osd);
1019 
1020 	CC_ALGO(tp) = NULL;
1021 	inp->inp_ppcb = NULL;
1022 	tp->t_inpcb = NULL;
1023 	uma_zfree(V_tcpcb_zone, tp);
1024 }
1025 
1026 /*
1027  * Attempt to close a TCP control block, marking it as dropped, and freeing
1028  * the socket if we hold the only reference.
1029  */
1030 struct tcpcb *
1031 tcp_close(struct tcpcb *tp)
1032 {
1033 	struct inpcb *inp = tp->t_inpcb;
1034 	struct socket *so;
1035 
1036 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1037 	INP_WLOCK_ASSERT(inp);
1038 
1039 #ifdef TCP_OFFLOAD
1040 	if (tp->t_state == TCPS_LISTEN)
1041 		tcp_offload_listen_stop(tp);
1042 #endif
1043 	in_pcbdrop(inp);
1044 	TCPSTAT_INC(tcps_closed);
1045 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1046 	so = inp->inp_socket;
1047 	soisdisconnected(so);
1048 	if (inp->inp_flags & INP_SOCKREF) {
1049 		KASSERT(so->so_state & SS_PROTOREF,
1050 		    ("tcp_close: !SS_PROTOREF"));
1051 		inp->inp_flags &= ~INP_SOCKREF;
1052 		INP_WUNLOCK(inp);
1053 		ACCEPT_LOCK();
1054 		SOCK_LOCK(so);
1055 		so->so_state &= ~SS_PROTOREF;
1056 		sofree(so);
1057 		return (NULL);
1058 	}
1059 	return (tp);
1060 }
1061 
1062 void
1063 tcp_drain(void)
1064 {
1065 	VNET_ITERATOR_DECL(vnet_iter);
1066 
1067 	if (!do_tcpdrain)
1068 		return;
1069 
1070 	VNET_LIST_RLOCK_NOSLEEP();
1071 	VNET_FOREACH(vnet_iter) {
1072 		CURVNET_SET(vnet_iter);
1073 		struct inpcb *inpb;
1074 		struct tcpcb *tcpb;
1075 
1076 	/*
1077 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1078 	 * if there is one...
1079 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1080 	 *      reassembly queue should be flushed, but in a situation
1081 	 *	where we're really low on mbufs, this is potentially
1082 	 *	useful.
1083 	 */
1084 		INP_INFO_RLOCK(&V_tcbinfo);
1085 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1086 			if (inpb->inp_flags & INP_TIMEWAIT)
1087 				continue;
1088 			INP_WLOCK(inpb);
1089 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1090 				tcp_reass_flush(tcpb);
1091 				tcp_clean_sackreport(tcpb);
1092 			}
1093 			INP_WUNLOCK(inpb);
1094 		}
1095 		INP_INFO_RUNLOCK(&V_tcbinfo);
1096 		CURVNET_RESTORE();
1097 	}
1098 	VNET_LIST_RUNLOCK_NOSLEEP();
1099 }
1100 
1101 /*
1102  * Notify a tcp user of an asynchronous error;
1103  * store error as soft error, but wake up user
1104  * (for now, won't do anything until can select for soft error).
1105  *
1106  * Do not wake up user since there currently is no mechanism for
1107  * reporting soft errors (yet - a kqueue filter may be added).
1108  */
1109 static struct inpcb *
1110 tcp_notify(struct inpcb *inp, int error)
1111 {
1112 	struct tcpcb *tp;
1113 
1114 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1115 	INP_WLOCK_ASSERT(inp);
1116 
1117 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1118 	    (inp->inp_flags & INP_DROPPED))
1119 		return (inp);
1120 
1121 	tp = intotcpcb(inp);
1122 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1123 
1124 	/*
1125 	 * Ignore some errors if we are hooked up.
1126 	 * If connection hasn't completed, has retransmitted several times,
1127 	 * and receives a second error, give up now.  This is better
1128 	 * than waiting a long time to establish a connection that
1129 	 * can never complete.
1130 	 */
1131 	if (tp->t_state == TCPS_ESTABLISHED &&
1132 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1133 	     error == EHOSTDOWN)) {
1134 		return (inp);
1135 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1136 	    tp->t_softerror) {
1137 		tp = tcp_drop(tp, error);
1138 		if (tp != NULL)
1139 			return (inp);
1140 		else
1141 			return (NULL);
1142 	} else {
1143 		tp->t_softerror = error;
1144 		return (inp);
1145 	}
1146 #if 0
1147 	wakeup( &so->so_timeo);
1148 	sorwakeup(so);
1149 	sowwakeup(so);
1150 #endif
1151 }
1152 
1153 static int
1154 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1155 {
1156 	int error, i, m, n, pcb_count;
1157 	struct inpcb *inp, **inp_list;
1158 	inp_gen_t gencnt;
1159 	struct xinpgen xig;
1160 
1161 	/*
1162 	 * The process of preparing the TCB list is too time-consuming and
1163 	 * resource-intensive to repeat twice on every request.
1164 	 */
1165 	if (req->oldptr == NULL) {
1166 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1167 		n += imax(n / 8, 10);
1168 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1169 		return (0);
1170 	}
1171 
1172 	if (req->newptr != NULL)
1173 		return (EPERM);
1174 
1175 	/*
1176 	 * OK, now we're committed to doing something.
1177 	 */
1178 	INP_INFO_RLOCK(&V_tcbinfo);
1179 	gencnt = V_tcbinfo.ipi_gencnt;
1180 	n = V_tcbinfo.ipi_count;
1181 	INP_INFO_RUNLOCK(&V_tcbinfo);
1182 
1183 	m = syncache_pcbcount();
1184 
1185 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1186 		+ (n + m) * sizeof(struct xtcpcb));
1187 	if (error != 0)
1188 		return (error);
1189 
1190 	xig.xig_len = sizeof xig;
1191 	xig.xig_count = n + m;
1192 	xig.xig_gen = gencnt;
1193 	xig.xig_sogen = so_gencnt;
1194 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1195 	if (error)
1196 		return (error);
1197 
1198 	error = syncache_pcblist(req, m, &pcb_count);
1199 	if (error)
1200 		return (error);
1201 
1202 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1203 	if (inp_list == NULL)
1204 		return (ENOMEM);
1205 
1206 	INP_INFO_RLOCK(&V_tcbinfo);
1207 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1208 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1209 		INP_WLOCK(inp);
1210 		if (inp->inp_gencnt <= gencnt) {
1211 			/*
1212 			 * XXX: This use of cr_cansee(), introduced with
1213 			 * TCP state changes, is not quite right, but for
1214 			 * now, better than nothing.
1215 			 */
1216 			if (inp->inp_flags & INP_TIMEWAIT) {
1217 				if (intotw(inp) != NULL)
1218 					error = cr_cansee(req->td->td_ucred,
1219 					    intotw(inp)->tw_cred);
1220 				else
1221 					error = EINVAL;	/* Skip this inp. */
1222 			} else
1223 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1224 			if (error == 0) {
1225 				in_pcbref(inp);
1226 				inp_list[i++] = inp;
1227 			}
1228 		}
1229 		INP_WUNLOCK(inp);
1230 	}
1231 	INP_INFO_RUNLOCK(&V_tcbinfo);
1232 	n = i;
1233 
1234 	error = 0;
1235 	for (i = 0; i < n; i++) {
1236 		inp = inp_list[i];
1237 		INP_RLOCK(inp);
1238 		if (inp->inp_gencnt <= gencnt) {
1239 			struct xtcpcb xt;
1240 			void *inp_ppcb;
1241 
1242 			bzero(&xt, sizeof(xt));
1243 			xt.xt_len = sizeof xt;
1244 			/* XXX should avoid extra copy */
1245 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1246 			inp_ppcb = inp->inp_ppcb;
1247 			if (inp_ppcb == NULL)
1248 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1249 			else if (inp->inp_flags & INP_TIMEWAIT) {
1250 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1251 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1252 			} else {
1253 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1254 				if (xt.xt_tp.t_timers)
1255 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1256 			}
1257 			if (inp->inp_socket != NULL)
1258 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1259 			else {
1260 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1261 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1262 			}
1263 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1264 			INP_RUNLOCK(inp);
1265 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1266 		} else
1267 			INP_RUNLOCK(inp);
1268 	}
1269 	INP_INFO_WLOCK(&V_tcbinfo);
1270 	for (i = 0; i < n; i++) {
1271 		inp = inp_list[i];
1272 		INP_RLOCK(inp);
1273 		if (!in_pcbrele_rlocked(inp))
1274 			INP_RUNLOCK(inp);
1275 	}
1276 	INP_INFO_WUNLOCK(&V_tcbinfo);
1277 
1278 	if (!error) {
1279 		/*
1280 		 * Give the user an updated idea of our state.
1281 		 * If the generation differs from what we told
1282 		 * her before, she knows that something happened
1283 		 * while we were processing this request, and it
1284 		 * might be necessary to retry.
1285 		 */
1286 		INP_INFO_RLOCK(&V_tcbinfo);
1287 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1288 		xig.xig_sogen = so_gencnt;
1289 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1290 		INP_INFO_RUNLOCK(&V_tcbinfo);
1291 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1292 	}
1293 	free(inp_list, M_TEMP);
1294 	return (error);
1295 }
1296 
1297 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1298     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1299     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1300 
1301 #ifdef INET
1302 static int
1303 tcp_getcred(SYSCTL_HANDLER_ARGS)
1304 {
1305 	struct xucred xuc;
1306 	struct sockaddr_in addrs[2];
1307 	struct inpcb *inp;
1308 	int error;
1309 
1310 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1311 	if (error)
1312 		return (error);
1313 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1314 	if (error)
1315 		return (error);
1316 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1317 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1318 	if (inp != NULL) {
1319 		if (inp->inp_socket == NULL)
1320 			error = ENOENT;
1321 		if (error == 0)
1322 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1323 		if (error == 0)
1324 			cru2x(inp->inp_cred, &xuc);
1325 		INP_RUNLOCK(inp);
1326 	} else
1327 		error = ENOENT;
1328 	if (error == 0)
1329 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1330 	return (error);
1331 }
1332 
1333 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1334     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1335     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1336 #endif /* INET */
1337 
1338 #ifdef INET6
1339 static int
1340 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1341 {
1342 	struct xucred xuc;
1343 	struct sockaddr_in6 addrs[2];
1344 	struct inpcb *inp;
1345 	int error;
1346 #ifdef INET
1347 	int mapped = 0;
1348 #endif
1349 
1350 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1351 	if (error)
1352 		return (error);
1353 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1354 	if (error)
1355 		return (error);
1356 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1357 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1358 		return (error);
1359 	}
1360 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1361 #ifdef INET
1362 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1363 			mapped = 1;
1364 		else
1365 #endif
1366 			return (EINVAL);
1367 	}
1368 
1369 #ifdef INET
1370 	if (mapped == 1)
1371 		inp = in_pcblookup(&V_tcbinfo,
1372 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1373 			addrs[1].sin6_port,
1374 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1375 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1376 	else
1377 #endif
1378 		inp = in6_pcblookup(&V_tcbinfo,
1379 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1380 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1381 			INPLOOKUP_RLOCKPCB, NULL);
1382 	if (inp != NULL) {
1383 		if (inp->inp_socket == NULL)
1384 			error = ENOENT;
1385 		if (error == 0)
1386 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1387 		if (error == 0)
1388 			cru2x(inp->inp_cred, &xuc);
1389 		INP_RUNLOCK(inp);
1390 	} else
1391 		error = ENOENT;
1392 	if (error == 0)
1393 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1394 	return (error);
1395 }
1396 
1397 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1398     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1399     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1400 #endif /* INET6 */
1401 
1402 
1403 #ifdef INET
1404 void
1405 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1406 {
1407 	struct ip *ip = vip;
1408 	struct tcphdr *th;
1409 	struct in_addr faddr;
1410 	struct inpcb *inp;
1411 	struct tcpcb *tp;
1412 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1413 	struct icmp *icp;
1414 	struct in_conninfo inc;
1415 	tcp_seq icmp_tcp_seq;
1416 	int mtu;
1417 
1418 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1419 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1420 		return;
1421 
1422 	if (cmd == PRC_MSGSIZE)
1423 		notify = tcp_mtudisc_notify;
1424 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1425 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1426 		notify = tcp_drop_syn_sent;
1427 	/*
1428 	 * Redirects don't need to be handled up here.
1429 	 */
1430 	else if (PRC_IS_REDIRECT(cmd))
1431 		return;
1432 	/*
1433 	 * Source quench is depreciated.
1434 	 */
1435 	else if (cmd == PRC_QUENCH)
1436 		return;
1437 	/*
1438 	 * Hostdead is ugly because it goes linearly through all PCBs.
1439 	 * XXX: We never get this from ICMP, otherwise it makes an
1440 	 * excellent DoS attack on machines with many connections.
1441 	 */
1442 	else if (cmd == PRC_HOSTDEAD)
1443 		ip = NULL;
1444 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1445 		return;
1446 	if (ip != NULL) {
1447 		icp = (struct icmp *)((caddr_t)ip
1448 				      - offsetof(struct icmp, icmp_ip));
1449 		th = (struct tcphdr *)((caddr_t)ip
1450 				       + (ip->ip_hl << 2));
1451 		INP_INFO_WLOCK(&V_tcbinfo);
1452 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1453 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1454 		if (inp != NULL)  {
1455 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1456 			    !(inp->inp_flags & INP_DROPPED) &&
1457 			    !(inp->inp_socket == NULL)) {
1458 				icmp_tcp_seq = htonl(th->th_seq);
1459 				tp = intotcpcb(inp);
1460 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1461 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1462 					if (cmd == PRC_MSGSIZE) {
1463 					    /*
1464 					     * MTU discovery:
1465 					     * If we got a needfrag set the MTU
1466 					     * in the route to the suggested new
1467 					     * value (if given) and then notify.
1468 					     */
1469 					    bzero(&inc, sizeof(inc));
1470 					    inc.inc_faddr = faddr;
1471 					    inc.inc_fibnum =
1472 						inp->inp_inc.inc_fibnum;
1473 
1474 					    mtu = ntohs(icp->icmp_nextmtu);
1475 					    /*
1476 					     * If no alternative MTU was
1477 					     * proposed, try the next smaller
1478 					     * one.
1479 					     */
1480 					    if (!mtu)
1481 						mtu = ip_next_mtu(
1482 						 ntohs(ip->ip_len), 1);
1483 					    if (mtu < V_tcp_minmss
1484 						 + sizeof(struct tcpiphdr))
1485 						mtu = V_tcp_minmss
1486 						 + sizeof(struct tcpiphdr);
1487 					    /*
1488 					     * Only cache the MTU if it
1489 					     * is smaller than the interface
1490 					     * or route MTU.  tcp_mtudisc()
1491 					     * will do right thing by itself.
1492 					     */
1493 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1494 						tcp_hc_updatemtu(&inc, mtu);
1495 					    tcp_mtudisc(inp, mtu);
1496 					} else
1497 						inp = (*notify)(inp,
1498 						    inetctlerrmap[cmd]);
1499 				}
1500 			}
1501 			if (inp != NULL)
1502 				INP_WUNLOCK(inp);
1503 		} else {
1504 			bzero(&inc, sizeof(inc));
1505 			inc.inc_fport = th->th_dport;
1506 			inc.inc_lport = th->th_sport;
1507 			inc.inc_faddr = faddr;
1508 			inc.inc_laddr = ip->ip_src;
1509 			syncache_unreach(&inc, th);
1510 		}
1511 		INP_INFO_WUNLOCK(&V_tcbinfo);
1512 	} else
1513 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1514 }
1515 #endif /* INET */
1516 
1517 #ifdef INET6
1518 void
1519 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1520 {
1521 	struct tcphdr th;
1522 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1523 	struct ip6_hdr *ip6;
1524 	struct mbuf *m;
1525 	struct ip6ctlparam *ip6cp = NULL;
1526 	const struct sockaddr_in6 *sa6_src = NULL;
1527 	int off;
1528 	struct tcp_portonly {
1529 		u_int16_t th_sport;
1530 		u_int16_t th_dport;
1531 	} *thp;
1532 
1533 	if (sa->sa_family != AF_INET6 ||
1534 	    sa->sa_len != sizeof(struct sockaddr_in6))
1535 		return;
1536 
1537 	if (cmd == PRC_MSGSIZE)
1538 		notify = tcp_mtudisc_notify;
1539 	else if (!PRC_IS_REDIRECT(cmd) &&
1540 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1541 		return;
1542 	/* Source quench is depreciated. */
1543 	else if (cmd == PRC_QUENCH)
1544 		return;
1545 
1546 	/* if the parameter is from icmp6, decode it. */
1547 	if (d != NULL) {
1548 		ip6cp = (struct ip6ctlparam *)d;
1549 		m = ip6cp->ip6c_m;
1550 		ip6 = ip6cp->ip6c_ip6;
1551 		off = ip6cp->ip6c_off;
1552 		sa6_src = ip6cp->ip6c_src;
1553 	} else {
1554 		m = NULL;
1555 		ip6 = NULL;
1556 		off = 0;	/* fool gcc */
1557 		sa6_src = &sa6_any;
1558 	}
1559 
1560 	if (ip6 != NULL) {
1561 		struct in_conninfo inc;
1562 		/*
1563 		 * XXX: We assume that when IPV6 is non NULL,
1564 		 * M and OFF are valid.
1565 		 */
1566 
1567 		/* check if we can safely examine src and dst ports */
1568 		if (m->m_pkthdr.len < off + sizeof(*thp))
1569 			return;
1570 
1571 		bzero(&th, sizeof(th));
1572 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1573 
1574 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1575 		    (struct sockaddr *)ip6cp->ip6c_src,
1576 		    th.th_sport, cmd, NULL, notify);
1577 
1578 		bzero(&inc, sizeof(inc));
1579 		inc.inc_fport = th.th_dport;
1580 		inc.inc_lport = th.th_sport;
1581 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1582 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1583 		inc.inc_flags |= INC_ISIPV6;
1584 		INP_INFO_WLOCK(&V_tcbinfo);
1585 		syncache_unreach(&inc, &th);
1586 		INP_INFO_WUNLOCK(&V_tcbinfo);
1587 	} else
1588 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1589 			      0, cmd, NULL, notify);
1590 }
1591 #endif /* INET6 */
1592 
1593 
1594 /*
1595  * Following is where TCP initial sequence number generation occurs.
1596  *
1597  * There are two places where we must use initial sequence numbers:
1598  * 1.  In SYN-ACK packets.
1599  * 2.  In SYN packets.
1600  *
1601  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1602  * tcp_syncache.c for details.
1603  *
1604  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1605  * depends on this property.  In addition, these ISNs should be
1606  * unguessable so as to prevent connection hijacking.  To satisfy
1607  * the requirements of this situation, the algorithm outlined in
1608  * RFC 1948 is used, with only small modifications.
1609  *
1610  * Implementation details:
1611  *
1612  * Time is based off the system timer, and is corrected so that it
1613  * increases by one megabyte per second.  This allows for proper
1614  * recycling on high speed LANs while still leaving over an hour
1615  * before rollover.
1616  *
1617  * As reading the *exact* system time is too expensive to be done
1618  * whenever setting up a TCP connection, we increment the time
1619  * offset in two ways.  First, a small random positive increment
1620  * is added to isn_offset for each connection that is set up.
1621  * Second, the function tcp_isn_tick fires once per clock tick
1622  * and increments isn_offset as necessary so that sequence numbers
1623  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1624  * random positive increments serve only to ensure that the same
1625  * exact sequence number is never sent out twice (as could otherwise
1626  * happen when a port is recycled in less than the system tick
1627  * interval.)
1628  *
1629  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1630  * between seeding of isn_secret.  This is normally set to zero,
1631  * as reseeding should not be necessary.
1632  *
1633  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1634  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1635  * general, this means holding an exclusive (write) lock.
1636  */
1637 
1638 #define ISN_BYTES_PER_SECOND 1048576
1639 #define ISN_STATIC_INCREMENT 4096
1640 #define ISN_RANDOM_INCREMENT (4096 - 1)
1641 
1642 static VNET_DEFINE(u_char, isn_secret[32]);
1643 static VNET_DEFINE(int, isn_last);
1644 static VNET_DEFINE(int, isn_last_reseed);
1645 static VNET_DEFINE(u_int32_t, isn_offset);
1646 static VNET_DEFINE(u_int32_t, isn_offset_old);
1647 
1648 #define	V_isn_secret			VNET(isn_secret)
1649 #define	V_isn_last			VNET(isn_last)
1650 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1651 #define	V_isn_offset			VNET(isn_offset)
1652 #define	V_isn_offset_old		VNET(isn_offset_old)
1653 
1654 tcp_seq
1655 tcp_new_isn(struct tcpcb *tp)
1656 {
1657 	MD5_CTX isn_ctx;
1658 	u_int32_t md5_buffer[4];
1659 	tcp_seq new_isn;
1660 	u_int32_t projected_offset;
1661 
1662 	INP_WLOCK_ASSERT(tp->t_inpcb);
1663 
1664 	ISN_LOCK();
1665 	/* Seed if this is the first use, reseed if requested. */
1666 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1667 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1668 		< (u_int)ticks))) {
1669 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1670 		V_isn_last_reseed = ticks;
1671 	}
1672 
1673 	/* Compute the md5 hash and return the ISN. */
1674 	MD5Init(&isn_ctx);
1675 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1676 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1677 #ifdef INET6
1678 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1679 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1680 			  sizeof(struct in6_addr));
1681 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1682 			  sizeof(struct in6_addr));
1683 	} else
1684 #endif
1685 	{
1686 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1687 			  sizeof(struct in_addr));
1688 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1689 			  sizeof(struct in_addr));
1690 	}
1691 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1692 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1693 	new_isn = (tcp_seq) md5_buffer[0];
1694 	V_isn_offset += ISN_STATIC_INCREMENT +
1695 		(arc4random() & ISN_RANDOM_INCREMENT);
1696 	if (ticks != V_isn_last) {
1697 		projected_offset = V_isn_offset_old +
1698 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1699 		if (SEQ_GT(projected_offset, V_isn_offset))
1700 			V_isn_offset = projected_offset;
1701 		V_isn_offset_old = V_isn_offset;
1702 		V_isn_last = ticks;
1703 	}
1704 	new_isn += V_isn_offset;
1705 	ISN_UNLOCK();
1706 	return (new_isn);
1707 }
1708 
1709 /*
1710  * When a specific ICMP unreachable message is received and the
1711  * connection state is SYN-SENT, drop the connection.  This behavior
1712  * is controlled by the icmp_may_rst sysctl.
1713  */
1714 struct inpcb *
1715 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1716 {
1717 	struct tcpcb *tp;
1718 
1719 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1720 	INP_WLOCK_ASSERT(inp);
1721 
1722 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1723 	    (inp->inp_flags & INP_DROPPED))
1724 		return (inp);
1725 
1726 	tp = intotcpcb(inp);
1727 	if (tp->t_state != TCPS_SYN_SENT)
1728 		return (inp);
1729 
1730 	tp = tcp_drop(tp, errno);
1731 	if (tp != NULL)
1732 		return (inp);
1733 	else
1734 		return (NULL);
1735 }
1736 
1737 /*
1738  * When `need fragmentation' ICMP is received, update our idea of the MSS
1739  * based on the new value. Also nudge TCP to send something, since we
1740  * know the packet we just sent was dropped.
1741  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1742  */
1743 static struct inpcb *
1744 tcp_mtudisc_notify(struct inpcb *inp, int error)
1745 {
1746 
1747 	return (tcp_mtudisc(inp, -1));
1748 }
1749 
1750 struct inpcb *
1751 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1752 {
1753 	struct tcpcb *tp;
1754 	struct socket *so;
1755 
1756 	INP_WLOCK_ASSERT(inp);
1757 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1758 	    (inp->inp_flags & INP_DROPPED))
1759 		return (inp);
1760 
1761 	tp = intotcpcb(inp);
1762 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1763 
1764 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1765 
1766 	so = inp->inp_socket;
1767 	SOCKBUF_LOCK(&so->so_snd);
1768 	/* If the mss is larger than the socket buffer, decrease the mss. */
1769 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1770 		tp->t_maxseg = so->so_snd.sb_hiwat;
1771 	SOCKBUF_UNLOCK(&so->so_snd);
1772 
1773 	TCPSTAT_INC(tcps_mturesent);
1774 	tp->t_rtttime = 0;
1775 	tp->snd_nxt = tp->snd_una;
1776 	tcp_free_sackholes(tp);
1777 	tp->snd_recover = tp->snd_max;
1778 	if (tp->t_flags & TF_SACK_PERMIT)
1779 		EXIT_FASTRECOVERY(tp->t_flags);
1780 	tcp_output(tp);
1781 	return (inp);
1782 }
1783 
1784 #ifdef INET
1785 /*
1786  * Look-up the routing entry to the peer of this inpcb.  If no route
1787  * is found and it cannot be allocated, then return 0.  This routine
1788  * is called by TCP routines that access the rmx structure and by
1789  * tcp_mss_update to get the peer/interface MTU.
1790  */
1791 u_long
1792 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1793 {
1794 	struct route sro;
1795 	struct sockaddr_in *dst;
1796 	struct ifnet *ifp;
1797 	u_long maxmtu = 0;
1798 
1799 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1800 
1801 	bzero(&sro, sizeof(sro));
1802 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1803 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1804 		dst->sin_family = AF_INET;
1805 		dst->sin_len = sizeof(*dst);
1806 		dst->sin_addr = inc->inc_faddr;
1807 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1808 	}
1809 	if (sro.ro_rt != NULL) {
1810 		ifp = sro.ro_rt->rt_ifp;
1811 		if (sro.ro_rt->rt_mtu == 0)
1812 			maxmtu = ifp->if_mtu;
1813 		else
1814 			maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1815 
1816 		/* Report additional interface capabilities. */
1817 		if (cap != NULL) {
1818 			if (ifp->if_capenable & IFCAP_TSO4 &&
1819 			    ifp->if_hwassist & CSUM_TSO) {
1820 				cap->ifcap |= CSUM_TSO;
1821 				cap->tsomax = ifp->if_hw_tsomax;
1822 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1823 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1824 			}
1825 		}
1826 		RTFREE(sro.ro_rt);
1827 	}
1828 	return (maxmtu);
1829 }
1830 #endif /* INET */
1831 
1832 #ifdef INET6
1833 u_long
1834 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1835 {
1836 	struct route_in6 sro6;
1837 	struct ifnet *ifp;
1838 	u_long maxmtu = 0;
1839 
1840 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1841 
1842 	bzero(&sro6, sizeof(sro6));
1843 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1844 		sro6.ro_dst.sin6_family = AF_INET6;
1845 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1846 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1847 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1848 	}
1849 	if (sro6.ro_rt != NULL) {
1850 		ifp = sro6.ro_rt->rt_ifp;
1851 		if (sro6.ro_rt->rt_mtu == 0)
1852 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1853 		else
1854 			maxmtu = min(sro6.ro_rt->rt_mtu,
1855 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1856 
1857 		/* Report additional interface capabilities. */
1858 		if (cap != NULL) {
1859 			if (ifp->if_capenable & IFCAP_TSO6 &&
1860 			    ifp->if_hwassist & CSUM_TSO) {
1861 				cap->ifcap |= CSUM_TSO;
1862 				cap->tsomax = ifp->if_hw_tsomax;
1863 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1864 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1865 			}
1866 		}
1867 		RTFREE(sro6.ro_rt);
1868 	}
1869 
1870 	return (maxmtu);
1871 }
1872 #endif /* INET6 */
1873 
1874 #ifdef IPSEC
1875 /* compute ESP/AH header size for TCP, including outer IP header. */
1876 size_t
1877 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1878 {
1879 	struct inpcb *inp;
1880 	struct mbuf *m;
1881 	size_t hdrsiz;
1882 	struct ip *ip;
1883 #ifdef INET6
1884 	struct ip6_hdr *ip6;
1885 #endif
1886 	struct tcphdr *th;
1887 
1888 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1889 		return (0);
1890 	m = m_gethdr(M_NOWAIT, MT_DATA);
1891 	if (!m)
1892 		return (0);
1893 
1894 #ifdef INET6
1895 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1896 		ip6 = mtod(m, struct ip6_hdr *);
1897 		th = (struct tcphdr *)(ip6 + 1);
1898 		m->m_pkthdr.len = m->m_len =
1899 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1900 		tcpip_fillheaders(inp, ip6, th);
1901 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1902 	} else
1903 #endif /* INET6 */
1904 	{
1905 		ip = mtod(m, struct ip *);
1906 		th = (struct tcphdr *)(ip + 1);
1907 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1908 		tcpip_fillheaders(inp, ip, th);
1909 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1910 	}
1911 
1912 	m_free(m);
1913 	return (hdrsiz);
1914 }
1915 #endif /* IPSEC */
1916 
1917 #ifdef TCP_SIGNATURE
1918 /*
1919  * Callback function invoked by m_apply() to digest TCP segment data
1920  * contained within an mbuf chain.
1921  */
1922 static int
1923 tcp_signature_apply(void *fstate, void *data, u_int len)
1924 {
1925 
1926 	MD5Update(fstate, (u_char *)data, len);
1927 	return (0);
1928 }
1929 
1930 /*
1931  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1932  * search with the destination IP address, and a 'magic SPI' to be
1933  * determined by the application. This is hardcoded elsewhere to 1179
1934 */
1935 struct secasvar *
1936 tcp_get_sav(struct mbuf *m, u_int direction)
1937 {
1938 	union sockaddr_union dst;
1939 	struct secasvar *sav;
1940 	struct ip *ip;
1941 #ifdef INET6
1942 	struct ip6_hdr *ip6;
1943 	char ip6buf[INET6_ADDRSTRLEN];
1944 #endif
1945 
1946 	/* Extract the destination from the IP header in the mbuf. */
1947 	bzero(&dst, sizeof(union sockaddr_union));
1948 	ip = mtod(m, struct ip *);
1949 #ifdef INET6
1950 	ip6 = NULL;	/* Make the compiler happy. */
1951 #endif
1952 	switch (ip->ip_v) {
1953 #ifdef INET
1954 	case IPVERSION:
1955 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1956 		dst.sa.sa_family = AF_INET;
1957 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1958 		    ip->ip_src : ip->ip_dst;
1959 		break;
1960 #endif
1961 #ifdef INET6
1962 	case (IPV6_VERSION >> 4):
1963 		ip6 = mtod(m, struct ip6_hdr *);
1964 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1965 		dst.sa.sa_family = AF_INET6;
1966 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1967 		    ip6->ip6_src : ip6->ip6_dst;
1968 		break;
1969 #endif
1970 	default:
1971 		return (NULL);
1972 		/* NOTREACHED */
1973 		break;
1974 	}
1975 
1976 	/* Look up an SADB entry which matches the address of the peer. */
1977 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1978 	if (sav == NULL) {
1979 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1980 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1981 #ifdef INET6
1982 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
1983 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1984 #endif
1985 			"(unsupported)"));
1986 	}
1987 
1988 	return (sav);
1989 }
1990 
1991 /*
1992  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1993  *
1994  * Parameters:
1995  * m		pointer to head of mbuf chain
1996  * len		length of TCP segment data, excluding options
1997  * optlen	length of TCP segment options
1998  * buf		pointer to storage for computed MD5 digest
1999  * sav		pointer to security assosiation
2000  *
2001  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2002  * When called from tcp_input(), we can be sure that th_sum has been
2003  * zeroed out and verified already.
2004  *
2005  * Releases reference to SADB key before return.
2006  *
2007  * Return 0 if successful, otherwise return -1.
2008  *
2009  */
2010 int
2011 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2012     u_char *buf, struct secasvar *sav)
2013 {
2014 #ifdef INET
2015 	struct ippseudo ippseudo;
2016 #endif
2017 	MD5_CTX ctx;
2018 	int doff;
2019 	struct ip *ip;
2020 #ifdef INET
2021 	struct ipovly *ipovly;
2022 #endif
2023 	struct tcphdr *th;
2024 #ifdef INET6
2025 	struct ip6_hdr *ip6;
2026 	struct in6_addr in6;
2027 	uint32_t plen;
2028 	uint16_t nhdr;
2029 #endif
2030 	u_short savecsum;
2031 
2032 	KASSERT(m != NULL, ("NULL mbuf chain"));
2033 	KASSERT(buf != NULL, ("NULL signature pointer"));
2034 
2035 	/* Extract the destination from the IP header in the mbuf. */
2036 	ip = mtod(m, struct ip *);
2037 #ifdef INET6
2038 	ip6 = NULL;	/* Make the compiler happy. */
2039 #endif
2040 
2041 	MD5Init(&ctx);
2042 	/*
2043 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2044 	 *
2045 	 * XXX The ippseudo header MUST be digested in network byte order,
2046 	 * or else we'll fail the regression test. Assume all fields we've
2047 	 * been doing arithmetic on have been in host byte order.
2048 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2049 	 * tcp_output(), the underlying ip_len member has not yet been set.
2050 	 */
2051 	switch (ip->ip_v) {
2052 #ifdef INET
2053 	case IPVERSION:
2054 		ipovly = (struct ipovly *)ip;
2055 		ippseudo.ippseudo_src = ipovly->ih_src;
2056 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2057 		ippseudo.ippseudo_pad = 0;
2058 		ippseudo.ippseudo_p = IPPROTO_TCP;
2059 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2060 		    optlen);
2061 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2062 
2063 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2064 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2065 		break;
2066 #endif
2067 #ifdef INET6
2068 	/*
2069 	 * RFC 2385, 2.0  Proposal
2070 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2071 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2072 	 * extended next header value (to form 32 bits), and 32-bit segment
2073 	 * length.
2074 	 * Note: Upper-Layer Packet Length comes before Next Header.
2075 	 */
2076 	case (IPV6_VERSION >> 4):
2077 		in6 = ip6->ip6_src;
2078 		in6_clearscope(&in6);
2079 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2080 		in6 = ip6->ip6_dst;
2081 		in6_clearscope(&in6);
2082 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2083 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2084 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2085 		nhdr = 0;
2086 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2087 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2088 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2089 		nhdr = IPPROTO_TCP;
2090 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2091 
2092 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2093 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2094 		break;
2095 #endif
2096 	default:
2097 		return (-1);
2098 		/* NOTREACHED */
2099 		break;
2100 	}
2101 
2102 
2103 	/*
2104 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2105 	 * The TCP checksum must be set to zero.
2106 	 */
2107 	savecsum = th->th_sum;
2108 	th->th_sum = 0;
2109 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2110 	th->th_sum = savecsum;
2111 
2112 	/*
2113 	 * Step 3: Update MD5 hash with TCP segment data.
2114 	 *         Use m_apply() to avoid an early m_pullup().
2115 	 */
2116 	if (len > 0)
2117 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2118 
2119 	/*
2120 	 * Step 4: Update MD5 hash with shared secret.
2121 	 */
2122 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2123 	MD5Final(buf, &ctx);
2124 
2125 	key_sa_recordxfer(sav, m);
2126 	KEY_FREESAV(&sav);
2127 	return (0);
2128 }
2129 
2130 /*
2131  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2132  *
2133  * Return 0 if successful, otherwise return -1.
2134  */
2135 int
2136 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2137     u_char *buf, u_int direction)
2138 {
2139 	struct secasvar *sav;
2140 
2141 	if ((sav = tcp_get_sav(m, direction)) == NULL)
2142 		return (-1);
2143 
2144 	return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2145 }
2146 
2147 /*
2148  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2149  *
2150  * Parameters:
2151  * m		pointer to head of mbuf chain
2152  * len		length of TCP segment data, excluding options
2153  * optlen	length of TCP segment options
2154  * buf		pointer to storage for computed MD5 digest
2155  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2156  *
2157  * Return 1 if successful, otherwise return 0.
2158  */
2159 int
2160 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2161     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2162 {
2163 	char tmpdigest[TCP_SIGLEN];
2164 
2165 	if (tcp_sig_checksigs == 0)
2166 		return (1);
2167 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2168 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2169 
2170 			/*
2171 			 * If this socket is not expecting signature but
2172 			 * the segment contains signature just fail.
2173 			 */
2174 			TCPSTAT_INC(tcps_sig_err_sigopt);
2175 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2176 			return (0);
2177 		}
2178 
2179 		/* Signature is not expected, and not present in segment. */
2180 		return (1);
2181 	}
2182 
2183 	/*
2184 	 * If this socket is expecting signature but the segment does not
2185 	 * contain any just fail.
2186 	 */
2187 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2188 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2189 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2190 		return (0);
2191 	}
2192 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2193 	    IPSEC_DIR_INBOUND) == -1) {
2194 		TCPSTAT_INC(tcps_sig_err_buildsig);
2195 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2196 		return (0);
2197 	}
2198 
2199 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2200 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2201 		return (0);
2202 	}
2203 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2204 	return (1);
2205 }
2206 #endif /* TCP_SIGNATURE */
2207 
2208 static int
2209 sysctl_drop(SYSCTL_HANDLER_ARGS)
2210 {
2211 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2212 	struct sockaddr_storage addrs[2];
2213 	struct inpcb *inp;
2214 	struct tcpcb *tp;
2215 	struct tcptw *tw;
2216 	struct sockaddr_in *fin, *lin;
2217 #ifdef INET6
2218 	struct sockaddr_in6 *fin6, *lin6;
2219 #endif
2220 	int error;
2221 
2222 	inp = NULL;
2223 	fin = lin = NULL;
2224 #ifdef INET6
2225 	fin6 = lin6 = NULL;
2226 #endif
2227 	error = 0;
2228 
2229 	if (req->oldptr != NULL || req->oldlen != 0)
2230 		return (EINVAL);
2231 	if (req->newptr == NULL)
2232 		return (EPERM);
2233 	if (req->newlen < sizeof(addrs))
2234 		return (ENOMEM);
2235 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2236 	if (error)
2237 		return (error);
2238 
2239 	switch (addrs[0].ss_family) {
2240 #ifdef INET6
2241 	case AF_INET6:
2242 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2243 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2244 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2245 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2246 			return (EINVAL);
2247 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2248 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2249 				return (EINVAL);
2250 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2251 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2252 			fin = (struct sockaddr_in *)&addrs[0];
2253 			lin = (struct sockaddr_in *)&addrs[1];
2254 			break;
2255 		}
2256 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2257 		if (error)
2258 			return (error);
2259 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2260 		if (error)
2261 			return (error);
2262 		break;
2263 #endif
2264 #ifdef INET
2265 	case AF_INET:
2266 		fin = (struct sockaddr_in *)&addrs[0];
2267 		lin = (struct sockaddr_in *)&addrs[1];
2268 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2269 		    lin->sin_len != sizeof(struct sockaddr_in))
2270 			return (EINVAL);
2271 		break;
2272 #endif
2273 	default:
2274 		return (EINVAL);
2275 	}
2276 	INP_INFO_WLOCK(&V_tcbinfo);
2277 	switch (addrs[0].ss_family) {
2278 #ifdef INET6
2279 	case AF_INET6:
2280 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2281 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2282 		    INPLOOKUP_WLOCKPCB, NULL);
2283 		break;
2284 #endif
2285 #ifdef INET
2286 	case AF_INET:
2287 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2288 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2289 		break;
2290 #endif
2291 	}
2292 	if (inp != NULL) {
2293 		if (inp->inp_flags & INP_TIMEWAIT) {
2294 			/*
2295 			 * XXXRW: There currently exists a state where an
2296 			 * inpcb is present, but its timewait state has been
2297 			 * discarded.  For now, don't allow dropping of this
2298 			 * type of inpcb.
2299 			 */
2300 			tw = intotw(inp);
2301 			if (tw != NULL)
2302 				tcp_twclose(tw, 0);
2303 			else
2304 				INP_WUNLOCK(inp);
2305 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2306 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2307 			tp = intotcpcb(inp);
2308 			tp = tcp_drop(tp, ECONNABORTED);
2309 			if (tp != NULL)
2310 				INP_WUNLOCK(inp);
2311 		} else
2312 			INP_WUNLOCK(inp);
2313 	} else
2314 		error = ESRCH;
2315 	INP_INFO_WUNLOCK(&V_tcbinfo);
2316 	return (error);
2317 }
2318 
2319 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2320     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2321     0, sysctl_drop, "", "Drop TCP connection");
2322 
2323 /*
2324  * Generate a standardized TCP log line for use throughout the
2325  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2326  * allow use in the interrupt context.
2327  *
2328  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2329  * NB: The function may return NULL if memory allocation failed.
2330  *
2331  * Due to header inclusion and ordering limitations the struct ip
2332  * and ip6_hdr pointers have to be passed as void pointers.
2333  */
2334 char *
2335 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2336     const void *ip6hdr)
2337 {
2338 
2339 	/* Is logging enabled? */
2340 	if (tcp_log_in_vain == 0)
2341 		return (NULL);
2342 
2343 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2344 }
2345 
2346 char *
2347 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2348     const void *ip6hdr)
2349 {
2350 
2351 	/* Is logging enabled? */
2352 	if (tcp_log_debug == 0)
2353 		return (NULL);
2354 
2355 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2356 }
2357 
2358 static char *
2359 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2360     const void *ip6hdr)
2361 {
2362 	char *s, *sp;
2363 	size_t size;
2364 	struct ip *ip;
2365 #ifdef INET6
2366 	const struct ip6_hdr *ip6;
2367 
2368 	ip6 = (const struct ip6_hdr *)ip6hdr;
2369 #endif /* INET6 */
2370 	ip = (struct ip *)ip4hdr;
2371 
2372 	/*
2373 	 * The log line looks like this:
2374 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2375 	 */
2376 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2377 	    sizeof(PRINT_TH_FLAGS) + 1 +
2378 #ifdef INET6
2379 	    2 * INET6_ADDRSTRLEN;
2380 #else
2381 	    2 * INET_ADDRSTRLEN;
2382 #endif /* INET6 */
2383 
2384 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2385 	if (s == NULL)
2386 		return (NULL);
2387 
2388 	strcat(s, "TCP: [");
2389 	sp = s + strlen(s);
2390 
2391 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2392 		inet_ntoa_r(inc->inc_faddr, sp);
2393 		sp = s + strlen(s);
2394 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2395 		sp = s + strlen(s);
2396 		inet_ntoa_r(inc->inc_laddr, sp);
2397 		sp = s + strlen(s);
2398 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2399 #ifdef INET6
2400 	} else if (inc) {
2401 		ip6_sprintf(sp, &inc->inc6_faddr);
2402 		sp = s + strlen(s);
2403 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2404 		sp = s + strlen(s);
2405 		ip6_sprintf(sp, &inc->inc6_laddr);
2406 		sp = s + strlen(s);
2407 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2408 	} else if (ip6 && th) {
2409 		ip6_sprintf(sp, &ip6->ip6_src);
2410 		sp = s + strlen(s);
2411 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2412 		sp = s + strlen(s);
2413 		ip6_sprintf(sp, &ip6->ip6_dst);
2414 		sp = s + strlen(s);
2415 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2416 #endif /* INET6 */
2417 #ifdef INET
2418 	} else if (ip && th) {
2419 		inet_ntoa_r(ip->ip_src, sp);
2420 		sp = s + strlen(s);
2421 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2422 		sp = s + strlen(s);
2423 		inet_ntoa_r(ip->ip_dst, sp);
2424 		sp = s + strlen(s);
2425 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2426 #endif /* INET */
2427 	} else {
2428 		free(s, M_TCPLOG);
2429 		return (NULL);
2430 	}
2431 	sp = s + strlen(s);
2432 	if (th)
2433 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2434 	if (*(s + size - 1) != '\0')
2435 		panic("%s: string too long", __func__);
2436 	return (s);
2437 }
2438 
2439 /*
2440  * A subroutine which makes it easy to track TCP state changes with DTrace.
2441  * This function shouldn't be called for t_state initializations that don't
2442  * correspond to actual TCP state transitions.
2443  */
2444 void
2445 tcp_state_change(struct tcpcb *tp, int newstate)
2446 {
2447 #if defined(KDTRACE_HOOKS)
2448 	int pstate = tp->t_state;
2449 #endif
2450 
2451 	tp->t_state = newstate;
2452 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2453 }
2454