xref: /freebsd/sys/netinet/tcp_subr.c (revision fba3cde907930eed2adb8a320524bc250338c729)
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_kdtrace.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
49 #include <sys/jail.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #ifdef INET6
53 #include <sys/domain.h>
54 #endif
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/sdt.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/protosw.h>
61 #include <sys/random.h>
62 
63 #include <vm/uma.h>
64 
65 #include <net/route.h>
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/vnet.h>
69 
70 #include <netinet/cc.h>
71 #include <netinet/in.h>
72 #include <netinet/in_kdtrace.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/in_systm.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip.h>
77 #include <netinet/ip_icmp.h>
78 #include <netinet/ip_var.h>
79 #ifdef INET6
80 #include <netinet/ip6.h>
81 #include <netinet6/in6_pcb.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #include <netinet6/nd6.h>
85 #endif
86 
87 #include <netinet/tcp_fsm.h>
88 #include <netinet/tcp_seq.h>
89 #include <netinet/tcp_timer.h>
90 #include <netinet/tcp_var.h>
91 #include <netinet/tcp_syncache.h>
92 #ifdef INET6
93 #include <netinet6/tcp6_var.h>
94 #endif
95 #include <netinet/tcpip.h>
96 #ifdef TCPDEBUG
97 #include <netinet/tcp_debug.h>
98 #endif
99 #ifdef INET6
100 #include <netinet6/ip6protosw.h>
101 #endif
102 #ifdef TCP_OFFLOAD
103 #include <netinet/tcp_offload.h>
104 #endif
105 
106 #ifdef IPSEC
107 #include <netipsec/ipsec.h>
108 #include <netipsec/xform.h>
109 #ifdef INET6
110 #include <netipsec/ipsec6.h>
111 #endif
112 #include <netipsec/key.h>
113 #include <sys/syslog.h>
114 #endif /*IPSEC*/
115 
116 #include <machine/in_cksum.h>
117 #include <sys/md5.h>
118 
119 #include <security/mac/mac_framework.h>
120 
121 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
122 #ifdef INET6
123 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
124 #endif
125 
126 static int
127 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
128 {
129 	int error, new;
130 
131 	new = V_tcp_mssdflt;
132 	error = sysctl_handle_int(oidp, &new, 0, req);
133 	if (error == 0 && req->newptr) {
134 		if (new < TCP_MINMSS)
135 			error = EINVAL;
136 		else
137 			V_tcp_mssdflt = new;
138 	}
139 	return (error);
140 }
141 
142 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
143     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
144     &sysctl_net_inet_tcp_mss_check, "I",
145     "Default TCP Maximum Segment Size");
146 
147 #ifdef INET6
148 static int
149 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
150 {
151 	int error, new;
152 
153 	new = V_tcp_v6mssdflt;
154 	error = sysctl_handle_int(oidp, &new, 0, req);
155 	if (error == 0 && req->newptr) {
156 		if (new < TCP_MINMSS)
157 			error = EINVAL;
158 		else
159 			V_tcp_v6mssdflt = new;
160 	}
161 	return (error);
162 }
163 
164 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
165     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
166     &sysctl_net_inet_tcp_mss_v6_check, "I",
167    "Default TCP Maximum Segment Size for IPv6");
168 #endif /* INET6 */
169 
170 /*
171  * Minimum MSS we accept and use. This prevents DoS attacks where
172  * we are forced to a ridiculous low MSS like 20 and send hundreds
173  * of packets instead of one. The effect scales with the available
174  * bandwidth and quickly saturates the CPU and network interface
175  * with packet generation and sending. Set to zero to disable MINMSS
176  * checking. This setting prevents us from sending too small packets.
177  */
178 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
179 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
180      &VNET_NAME(tcp_minmss), 0,
181     "Minimum TCP Maximum Segment Size");
182 
183 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
184 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
185     &VNET_NAME(tcp_do_rfc1323), 0,
186     "Enable rfc1323 (high performance TCP) extensions");
187 
188 static int	tcp_log_debug = 0;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
190     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
191 
192 static int	tcp_tcbhashsize = 0;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
194     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
195 
196 static int	do_tcpdrain = 1;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
198     "Enable tcp_drain routine for extra help when low on mbufs");
199 
200 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
201     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
202 
203 static VNET_DEFINE(int, icmp_may_rst) = 1;
204 #define	V_icmp_may_rst			VNET(icmp_may_rst)
205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
206     &VNET_NAME(icmp_may_rst), 0,
207     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
208 
209 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
210 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
211 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
212     &VNET_NAME(tcp_isn_reseed_interval), 0,
213     "Seconds between reseeding of ISN secret");
214 
215 static int	tcp_soreceive_stream = 0;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
217     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
218 
219 #ifdef TCP_SIGNATURE
220 static int	tcp_sig_checksigs = 1;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
222     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
223 #endif
224 
225 VNET_DEFINE(uma_zone_t, sack_hole_zone);
226 #define	V_sack_hole_zone		VNET(sack_hole_zone)
227 
228 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
229 
230 static struct inpcb *tcp_notify(struct inpcb *, int);
231 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
232 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
233 		    void *ip4hdr, const void *ip6hdr);
234 
235 /*
236  * Target size of TCP PCB hash tables. Must be a power of two.
237  *
238  * Note that this can be overridden by the kernel environment
239  * variable net.inet.tcp.tcbhashsize
240  */
241 #ifndef TCBHASHSIZE
242 #define TCBHASHSIZE	0
243 #endif
244 
245 /*
246  * XXX
247  * Callouts should be moved into struct tcp directly.  They are currently
248  * separate because the tcpcb structure is exported to userland for sysctl
249  * parsing purposes, which do not know about callouts.
250  */
251 struct tcpcb_mem {
252 	struct	tcpcb		tcb;
253 	struct	tcp_timer	tt;
254 	struct	cc_var		ccv;
255 	struct	osd		osd;
256 };
257 
258 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
259 #define	V_tcpcb_zone			VNET(tcpcb_zone)
260 
261 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
262 static struct mtx isn_mtx;
263 
264 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
265 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
266 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
267 
268 /*
269  * TCP initialization.
270  */
271 static void
272 tcp_zone_change(void *tag)
273 {
274 
275 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
276 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
277 	tcp_tw_zone_change();
278 }
279 
280 static int
281 tcp_inpcb_init(void *mem, int size, int flags)
282 {
283 	struct inpcb *inp = mem;
284 
285 	INP_LOCK_INIT(inp, "inp", "tcpinp");
286 	return (0);
287 }
288 
289 /*
290  * Take a value and get the next power of 2 that doesn't overflow.
291  * Used to size the tcp_inpcb hash buckets.
292  */
293 static int
294 maketcp_hashsize(int size)
295 {
296 	int hashsize;
297 
298 	/*
299 	 * auto tune.
300 	 * get the next power of 2 higher than maxsockets.
301 	 */
302 	hashsize = 1 << fls(size);
303 	/* catch overflow, and just go one power of 2 smaller */
304 	if (hashsize < size) {
305 		hashsize = 1 << (fls(size) - 1);
306 	}
307 	return (hashsize);
308 }
309 
310 void
311 tcp_init(void)
312 {
313 	const char *tcbhash_tuneable;
314 	int hashsize;
315 
316 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
317 
318 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
319 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
320 		printf("%s: WARNING: unable to register helper hook\n", __func__);
321 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
322 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 		printf("%s: WARNING: unable to register helper hook\n", __func__);
324 
325 	hashsize = TCBHASHSIZE;
326 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
327 	if (hashsize == 0) {
328 		/*
329 		 * Auto tune the hash size based on maxsockets.
330 		 * A perfect hash would have a 1:1 mapping
331 		 * (hashsize = maxsockets) however it's been
332 		 * suggested that O(2) average is better.
333 		 */
334 		hashsize = maketcp_hashsize(maxsockets / 4);
335 		/*
336 		 * Our historical default is 512,
337 		 * do not autotune lower than this.
338 		 */
339 		if (hashsize < 512)
340 			hashsize = 512;
341 		if (bootverbose)
342 			printf("%s: %s auto tuned to %d\n", __func__,
343 			    tcbhash_tuneable, hashsize);
344 	}
345 	/*
346 	 * We require a hashsize to be a power of two.
347 	 * Previously if it was not a power of two we would just reset it
348 	 * back to 512, which could be a nasty surprise if you did not notice
349 	 * the error message.
350 	 * Instead what we do is clip it to the closest power of two lower
351 	 * than the specified hash value.
352 	 */
353 	if (!powerof2(hashsize)) {
354 		int oldhashsize = hashsize;
355 
356 		hashsize = maketcp_hashsize(hashsize);
357 		/* prevent absurdly low value */
358 		if (hashsize < 16)
359 			hashsize = 16;
360 		printf("%s: WARNING: TCB hash size not a power of 2, "
361 		    "clipped from %d to %d.\n", __func__, oldhashsize,
362 		    hashsize);
363 	}
364 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
365 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
366 	    IPI_HASHFIELDS_4TUPLE);
367 
368 	/*
369 	 * These have to be type stable for the benefit of the timers.
370 	 */
371 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
372 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
373 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
374 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
375 
376 	tcp_tw_init();
377 	syncache_init();
378 	tcp_hc_init();
379 	tcp_reass_init();
380 
381 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
382 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
383 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384 
385 	/* Skip initialization of globals for non-default instances. */
386 	if (!IS_DEFAULT_VNET(curvnet))
387 		return;
388 
389 	/* XXX virtualize those bellow? */
390 	tcp_delacktime = TCPTV_DELACK;
391 	tcp_keepinit = TCPTV_KEEP_INIT;
392 	tcp_keepidle = TCPTV_KEEP_IDLE;
393 	tcp_keepintvl = TCPTV_KEEPINTVL;
394 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
395 	tcp_msl = TCPTV_MSL;
396 	tcp_rexmit_min = TCPTV_MIN;
397 	if (tcp_rexmit_min < 1)
398 		tcp_rexmit_min = 1;
399 	tcp_rexmit_slop = TCPTV_CPU_VAR;
400 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
401 	tcp_tcbhashsize = hashsize;
402 
403 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
404 	if (tcp_soreceive_stream) {
405 #ifdef INET
406 		tcp_usrreqs.pru_soreceive = soreceive_stream;
407 #endif
408 #ifdef INET6
409 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
410 #endif /* INET6 */
411 	}
412 
413 #ifdef INET6
414 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
415 #else /* INET6 */
416 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
417 #endif /* INET6 */
418 	if (max_protohdr < TCP_MINPROTOHDR)
419 		max_protohdr = TCP_MINPROTOHDR;
420 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
421 		panic("tcp_init");
422 #undef TCP_MINPROTOHDR
423 
424 	ISN_LOCK_INIT();
425 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
426 		SHUTDOWN_PRI_DEFAULT);
427 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
428 		EVENTHANDLER_PRI_ANY);
429 }
430 
431 #ifdef VIMAGE
432 void
433 tcp_destroy(void)
434 {
435 
436 	tcp_reass_destroy();
437 	tcp_hc_destroy();
438 	syncache_destroy();
439 	tcp_tw_destroy();
440 	in_pcbinfo_destroy(&V_tcbinfo);
441 	uma_zdestroy(V_sack_hole_zone);
442 	uma_zdestroy(V_tcpcb_zone);
443 }
444 #endif
445 
446 void
447 tcp_fini(void *xtp)
448 {
449 
450 }
451 
452 /*
453  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
454  * tcp_template used to store this data in mbufs, but we now recopy it out
455  * of the tcpcb each time to conserve mbufs.
456  */
457 void
458 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
459 {
460 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
461 
462 	INP_WLOCK_ASSERT(inp);
463 
464 #ifdef INET6
465 	if ((inp->inp_vflag & INP_IPV6) != 0) {
466 		struct ip6_hdr *ip6;
467 
468 		ip6 = (struct ip6_hdr *)ip_ptr;
469 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
470 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
471 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
472 			(IPV6_VERSION & IPV6_VERSION_MASK);
473 		ip6->ip6_nxt = IPPROTO_TCP;
474 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
475 		ip6->ip6_src = inp->in6p_laddr;
476 		ip6->ip6_dst = inp->in6p_faddr;
477 	}
478 #endif /* INET6 */
479 #if defined(INET6) && defined(INET)
480 	else
481 #endif
482 #ifdef INET
483 	{
484 		struct ip *ip;
485 
486 		ip = (struct ip *)ip_ptr;
487 		ip->ip_v = IPVERSION;
488 		ip->ip_hl = 5;
489 		ip->ip_tos = inp->inp_ip_tos;
490 		ip->ip_len = 0;
491 		ip->ip_id = 0;
492 		ip->ip_off = 0;
493 		ip->ip_ttl = inp->inp_ip_ttl;
494 		ip->ip_sum = 0;
495 		ip->ip_p = IPPROTO_TCP;
496 		ip->ip_src = inp->inp_laddr;
497 		ip->ip_dst = inp->inp_faddr;
498 	}
499 #endif /* INET */
500 	th->th_sport = inp->inp_lport;
501 	th->th_dport = inp->inp_fport;
502 	th->th_seq = 0;
503 	th->th_ack = 0;
504 	th->th_x2 = 0;
505 	th->th_off = 5;
506 	th->th_flags = 0;
507 	th->th_win = 0;
508 	th->th_urp = 0;
509 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
510 }
511 
512 /*
513  * Create template to be used to send tcp packets on a connection.
514  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
515  * use for this function is in keepalives, which use tcp_respond.
516  */
517 struct tcptemp *
518 tcpip_maketemplate(struct inpcb *inp)
519 {
520 	struct tcptemp *t;
521 
522 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
523 	if (t == NULL)
524 		return (NULL);
525 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
526 	return (t);
527 }
528 
529 /*
530  * Send a single message to the TCP at address specified by
531  * the given TCP/IP header.  If m == NULL, then we make a copy
532  * of the tcpiphdr at ti and send directly to the addressed host.
533  * This is used to force keep alive messages out using the TCP
534  * template for a connection.  If flags are given then we send
535  * a message back to the TCP which originated the * segment ti,
536  * and discard the mbuf containing it and any other attached mbufs.
537  *
538  * In any case the ack and sequence number of the transmitted
539  * segment are as specified by the parameters.
540  *
541  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
542  */
543 void
544 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
545     tcp_seq ack, tcp_seq seq, int flags)
546 {
547 	int tlen;
548 	int win = 0;
549 	struct ip *ip;
550 	struct tcphdr *nth;
551 #ifdef INET6
552 	struct ip6_hdr *ip6;
553 	int isipv6;
554 #endif /* INET6 */
555 	int ipflags = 0;
556 	struct inpcb *inp;
557 
558 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
559 
560 #ifdef INET6
561 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
562 	ip6 = ipgen;
563 #endif /* INET6 */
564 	ip = ipgen;
565 
566 	if (tp != NULL) {
567 		inp = tp->t_inpcb;
568 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
569 		INP_WLOCK_ASSERT(inp);
570 	} else
571 		inp = NULL;
572 
573 	if (tp != NULL) {
574 		if (!(flags & TH_RST)) {
575 			win = sbspace(&inp->inp_socket->so_rcv);
576 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
577 				win = (long)TCP_MAXWIN << tp->rcv_scale;
578 		}
579 	}
580 	if (m == NULL) {
581 		m = m_gethdr(M_NOWAIT, MT_DATA);
582 		if (m == NULL)
583 			return;
584 		tlen = 0;
585 		m->m_data += max_linkhdr;
586 #ifdef INET6
587 		if (isipv6) {
588 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
589 			      sizeof(struct ip6_hdr));
590 			ip6 = mtod(m, struct ip6_hdr *);
591 			nth = (struct tcphdr *)(ip6 + 1);
592 		} else
593 #endif /* INET6 */
594 		{
595 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
596 			ip = mtod(m, struct ip *);
597 			nth = (struct tcphdr *)(ip + 1);
598 		}
599 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
600 		flags = TH_ACK;
601 	} else {
602 		/*
603 		 *  reuse the mbuf.
604 		 * XXX MRT We inherrit the FIB, which is lucky.
605 		 */
606 		m_freem(m->m_next);
607 		m->m_next = NULL;
608 		m->m_data = (caddr_t)ipgen;
609 		/* m_len is set later */
610 		tlen = 0;
611 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
612 #ifdef INET6
613 		if (isipv6) {
614 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
615 			nth = (struct tcphdr *)(ip6 + 1);
616 		} else
617 #endif /* INET6 */
618 		{
619 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
620 			nth = (struct tcphdr *)(ip + 1);
621 		}
622 		if (th != nth) {
623 			/*
624 			 * this is usually a case when an extension header
625 			 * exists between the IPv6 header and the
626 			 * TCP header.
627 			 */
628 			nth->th_sport = th->th_sport;
629 			nth->th_dport = th->th_dport;
630 		}
631 		xchg(nth->th_dport, nth->th_sport, uint16_t);
632 #undef xchg
633 	}
634 #ifdef INET6
635 	if (isipv6) {
636 		ip6->ip6_flow = 0;
637 		ip6->ip6_vfc = IPV6_VERSION;
638 		ip6->ip6_nxt = IPPROTO_TCP;
639 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
640 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
641 	}
642 #endif
643 #if defined(INET) && defined(INET6)
644 	else
645 #endif
646 #ifdef INET
647 	{
648 		tlen += sizeof (struct tcpiphdr);
649 		ip->ip_len = htons(tlen);
650 		ip->ip_ttl = V_ip_defttl;
651 		if (V_path_mtu_discovery)
652 			ip->ip_off |= htons(IP_DF);
653 	}
654 #endif
655 	m->m_len = tlen;
656 	m->m_pkthdr.len = tlen;
657 	m->m_pkthdr.rcvif = NULL;
658 #ifdef MAC
659 	if (inp != NULL) {
660 		/*
661 		 * Packet is associated with a socket, so allow the
662 		 * label of the response to reflect the socket label.
663 		 */
664 		INP_WLOCK_ASSERT(inp);
665 		mac_inpcb_create_mbuf(inp, m);
666 	} else {
667 		/*
668 		 * Packet is not associated with a socket, so possibly
669 		 * update the label in place.
670 		 */
671 		mac_netinet_tcp_reply(m);
672 	}
673 #endif
674 	nth->th_seq = htonl(seq);
675 	nth->th_ack = htonl(ack);
676 	nth->th_x2 = 0;
677 	nth->th_off = sizeof (struct tcphdr) >> 2;
678 	nth->th_flags = flags;
679 	if (tp != NULL)
680 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
681 	else
682 		nth->th_win = htons((u_short)win);
683 	nth->th_urp = 0;
684 
685 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
686 #ifdef INET6
687 	if (isipv6) {
688 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
689 		nth->th_sum = in6_cksum_pseudo(ip6,
690 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
691 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
692 		    NULL, NULL);
693 	}
694 #endif /* INET6 */
695 #if defined(INET6) && defined(INET)
696 	else
697 #endif
698 #ifdef INET
699 	{
700 		m->m_pkthdr.csum_flags = CSUM_TCP;
701 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
702 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
703 	}
704 #endif /* INET */
705 #ifdef TCPDEBUG
706 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
707 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
708 #endif
709 	if (flags & TH_RST)
710 		TCP_PROBE5(accept_refused, NULL, NULL, m->m_data, tp, nth);
711 
712 	TCP_PROBE5(send, NULL, tp, m->m_data, tp, nth);
713 #ifdef INET6
714 	if (isipv6)
715 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
716 #endif /* INET6 */
717 #if defined(INET) && defined(INET6)
718 	else
719 #endif
720 #ifdef INET
721 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
722 #endif
723 }
724 
725 /*
726  * Create a new TCP control block, making an
727  * empty reassembly queue and hooking it to the argument
728  * protocol control block.  The `inp' parameter must have
729  * come from the zone allocator set up in tcp_init().
730  */
731 struct tcpcb *
732 tcp_newtcpcb(struct inpcb *inp)
733 {
734 	struct tcpcb_mem *tm;
735 	struct tcpcb *tp;
736 #ifdef INET6
737 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
738 #endif /* INET6 */
739 
740 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
741 	if (tm == NULL)
742 		return (NULL);
743 	tp = &tm->tcb;
744 
745 	/* Initialise cc_var struct for this tcpcb. */
746 	tp->ccv = &tm->ccv;
747 	tp->ccv->type = IPPROTO_TCP;
748 	tp->ccv->ccvc.tcp = tp;
749 
750 	/*
751 	 * Use the current system default CC algorithm.
752 	 */
753 	CC_LIST_RLOCK();
754 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
755 	CC_ALGO(tp) = CC_DEFAULT();
756 	CC_LIST_RUNLOCK();
757 
758 	if (CC_ALGO(tp)->cb_init != NULL)
759 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
760 			uma_zfree(V_tcpcb_zone, tm);
761 			return (NULL);
762 		}
763 
764 	tp->osd = &tm->osd;
765 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
766 		uma_zfree(V_tcpcb_zone, tm);
767 		return (NULL);
768 	}
769 
770 #ifdef VIMAGE
771 	tp->t_vnet = inp->inp_vnet;
772 #endif
773 	tp->t_timers = &tm->tt;
774 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
775 	tp->t_maxseg = tp->t_maxopd =
776 #ifdef INET6
777 		isipv6 ? V_tcp_v6mssdflt :
778 #endif /* INET6 */
779 		V_tcp_mssdflt;
780 
781 	/* Set up our timeouts. */
782 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
783 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
784 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
785 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
786 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
787 
788 	if (V_tcp_do_rfc1323)
789 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
790 	if (V_tcp_do_sack)
791 		tp->t_flags |= TF_SACK_PERMIT;
792 	TAILQ_INIT(&tp->snd_holes);
793 	tp->t_inpcb = inp;	/* XXX */
794 	/*
795 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
796 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
797 	 * reasonable initial retransmit time.
798 	 */
799 	tp->t_srtt = TCPTV_SRTTBASE;
800 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
801 	tp->t_rttmin = tcp_rexmit_min;
802 	tp->t_rxtcur = TCPTV_RTOBASE;
803 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
804 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
805 	tp->t_rcvtime = ticks;
806 	/*
807 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
808 	 * because the socket may be bound to an IPv6 wildcard address,
809 	 * which may match an IPv4-mapped IPv6 address.
810 	 */
811 	inp->inp_ip_ttl = V_ip_defttl;
812 	inp->inp_ppcb = tp;
813 	return (tp);		/* XXX */
814 }
815 
816 /*
817  * Switch the congestion control algorithm back to NewReno for any active
818  * control blocks using an algorithm which is about to go away.
819  * This ensures the CC framework can allow the unload to proceed without leaving
820  * any dangling pointers which would trigger a panic.
821  * Returning non-zero would inform the CC framework that something went wrong
822  * and it would be unsafe to allow the unload to proceed. However, there is no
823  * way for this to occur with this implementation so we always return zero.
824  */
825 int
826 tcp_ccalgounload(struct cc_algo *unload_algo)
827 {
828 	struct cc_algo *tmpalgo;
829 	struct inpcb *inp;
830 	struct tcpcb *tp;
831 	VNET_ITERATOR_DECL(vnet_iter);
832 
833 	/*
834 	 * Check all active control blocks across all network stacks and change
835 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
836 	 * requires cleanup code to be run, call it.
837 	 */
838 	VNET_LIST_RLOCK();
839 	VNET_FOREACH(vnet_iter) {
840 		CURVNET_SET(vnet_iter);
841 		INP_INFO_RLOCK(&V_tcbinfo);
842 		/*
843 		 * New connections already part way through being initialised
844 		 * with the CC algo we're removing will not race with this code
845 		 * because the INP_INFO_WLOCK is held during initialisation. We
846 		 * therefore don't enter the loop below until the connection
847 		 * list has stabilised.
848 		 */
849 		LIST_FOREACH(inp, &V_tcb, inp_list) {
850 			INP_WLOCK(inp);
851 			/* Important to skip tcptw structs. */
852 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
853 			    (tp = intotcpcb(inp)) != NULL) {
854 				/*
855 				 * By holding INP_WLOCK here, we are assured
856 				 * that the connection is not currently
857 				 * executing inside the CC module's functions
858 				 * i.e. it is safe to make the switch back to
859 				 * NewReno.
860 				 */
861 				if (CC_ALGO(tp) == unload_algo) {
862 					tmpalgo = CC_ALGO(tp);
863 					/* NewReno does not require any init. */
864 					CC_ALGO(tp) = &newreno_cc_algo;
865 					if (tmpalgo->cb_destroy != NULL)
866 						tmpalgo->cb_destroy(tp->ccv);
867 				}
868 			}
869 			INP_WUNLOCK(inp);
870 		}
871 		INP_INFO_RUNLOCK(&V_tcbinfo);
872 		CURVNET_RESTORE();
873 	}
874 	VNET_LIST_RUNLOCK();
875 
876 	return (0);
877 }
878 
879 /*
880  * Drop a TCP connection, reporting
881  * the specified error.  If connection is synchronized,
882  * then send a RST to peer.
883  */
884 struct tcpcb *
885 tcp_drop(struct tcpcb *tp, int errno)
886 {
887 	struct socket *so = tp->t_inpcb->inp_socket;
888 
889 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
890 	INP_WLOCK_ASSERT(tp->t_inpcb);
891 
892 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
893 		tcp_state_change(tp, TCPS_CLOSED);
894 		(void) tcp_output(tp);
895 		TCPSTAT_INC(tcps_drops);
896 	} else
897 		TCPSTAT_INC(tcps_conndrops);
898 	if (errno == ETIMEDOUT && tp->t_softerror)
899 		errno = tp->t_softerror;
900 	so->so_error = errno;
901 	return (tcp_close(tp));
902 }
903 
904 void
905 tcp_discardcb(struct tcpcb *tp)
906 {
907 	struct inpcb *inp = tp->t_inpcb;
908 	struct socket *so = inp->inp_socket;
909 #ifdef INET6
910 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
911 #endif /* INET6 */
912 
913 	INP_WLOCK_ASSERT(inp);
914 
915 	/*
916 	 * Make sure that all of our timers are stopped before we delete the
917 	 * PCB.
918 	 *
919 	 * XXXRW: Really, we would like to use callout_drain() here in order
920 	 * to avoid races experienced in tcp_timer.c where a timer is already
921 	 * executing at this point.  However, we can't, both because we're
922 	 * running in a context where we can't sleep, and also because we
923 	 * hold locks required by the timers.  What we instead need to do is
924 	 * test to see if callout_drain() is required, and if so, defer some
925 	 * portion of the remainder of tcp_discardcb() to an asynchronous
926 	 * context that can callout_drain() and then continue.  Some care
927 	 * will be required to ensure that no further processing takes place
928 	 * on the tcpcb, even though it hasn't been freed (a flag?).
929 	 */
930 	callout_stop(&tp->t_timers->tt_rexmt);
931 	callout_stop(&tp->t_timers->tt_persist);
932 	callout_stop(&tp->t_timers->tt_keep);
933 	callout_stop(&tp->t_timers->tt_2msl);
934 	callout_stop(&tp->t_timers->tt_delack);
935 
936 	/*
937 	 * If we got enough samples through the srtt filter,
938 	 * save the rtt and rttvar in the routing entry.
939 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
940 	 * 4 samples is enough for the srtt filter to converge
941 	 * to within enough % of the correct value; fewer samples
942 	 * and we could save a bogus rtt. The danger is not high
943 	 * as tcp quickly recovers from everything.
944 	 * XXX: Works very well but needs some more statistics!
945 	 */
946 	if (tp->t_rttupdated >= 4) {
947 		struct hc_metrics_lite metrics;
948 		u_long ssthresh;
949 
950 		bzero(&metrics, sizeof(metrics));
951 		/*
952 		 * Update the ssthresh always when the conditions below
953 		 * are satisfied. This gives us better new start value
954 		 * for the congestion avoidance for new connections.
955 		 * ssthresh is only set if packet loss occured on a session.
956 		 *
957 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
958 		 * being torn down.  Ideally this code would not use 'so'.
959 		 */
960 		ssthresh = tp->snd_ssthresh;
961 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
962 			/*
963 			 * convert the limit from user data bytes to
964 			 * packets then to packet data bytes.
965 			 */
966 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
967 			if (ssthresh < 2)
968 				ssthresh = 2;
969 			ssthresh *= (u_long)(tp->t_maxseg +
970 #ifdef INET6
971 			    (isipv6 ? sizeof (struct ip6_hdr) +
972 				sizeof (struct tcphdr) :
973 #endif
974 				sizeof (struct tcpiphdr)
975 #ifdef INET6
976 			    )
977 #endif
978 			    );
979 		} else
980 			ssthresh = 0;
981 		metrics.rmx_ssthresh = ssthresh;
982 
983 		metrics.rmx_rtt = tp->t_srtt;
984 		metrics.rmx_rttvar = tp->t_rttvar;
985 		metrics.rmx_cwnd = tp->snd_cwnd;
986 		metrics.rmx_sendpipe = 0;
987 		metrics.rmx_recvpipe = 0;
988 
989 		tcp_hc_update(&inp->inp_inc, &metrics);
990 	}
991 
992 	/* free the reassembly queue, if any */
993 	tcp_reass_flush(tp);
994 
995 #ifdef TCP_OFFLOAD
996 	/* Disconnect offload device, if any. */
997 	if (tp->t_flags & TF_TOE)
998 		tcp_offload_detach(tp);
999 #endif
1000 
1001 	tcp_free_sackholes(tp);
1002 
1003 	/* Allow the CC algorithm to clean up after itself. */
1004 	if (CC_ALGO(tp)->cb_destroy != NULL)
1005 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1006 
1007 	khelp_destroy_osd(tp->osd);
1008 
1009 	CC_ALGO(tp) = NULL;
1010 	inp->inp_ppcb = NULL;
1011 	tp->t_inpcb = NULL;
1012 	uma_zfree(V_tcpcb_zone, tp);
1013 }
1014 
1015 /*
1016  * Attempt to close a TCP control block, marking it as dropped, and freeing
1017  * the socket if we hold the only reference.
1018  */
1019 struct tcpcb *
1020 tcp_close(struct tcpcb *tp)
1021 {
1022 	struct inpcb *inp = tp->t_inpcb;
1023 	struct socket *so;
1024 
1025 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1026 	INP_WLOCK_ASSERT(inp);
1027 
1028 #ifdef TCP_OFFLOAD
1029 	if (tp->t_state == TCPS_LISTEN)
1030 		tcp_offload_listen_stop(tp);
1031 #endif
1032 	in_pcbdrop(inp);
1033 	TCPSTAT_INC(tcps_closed);
1034 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1035 	so = inp->inp_socket;
1036 	soisdisconnected(so);
1037 	if (inp->inp_flags & INP_SOCKREF) {
1038 		KASSERT(so->so_state & SS_PROTOREF,
1039 		    ("tcp_close: !SS_PROTOREF"));
1040 		inp->inp_flags &= ~INP_SOCKREF;
1041 		INP_WUNLOCK(inp);
1042 		ACCEPT_LOCK();
1043 		SOCK_LOCK(so);
1044 		so->so_state &= ~SS_PROTOREF;
1045 		sofree(so);
1046 		return (NULL);
1047 	}
1048 	return (tp);
1049 }
1050 
1051 void
1052 tcp_drain(void)
1053 {
1054 	VNET_ITERATOR_DECL(vnet_iter);
1055 
1056 	if (!do_tcpdrain)
1057 		return;
1058 
1059 	VNET_LIST_RLOCK_NOSLEEP();
1060 	VNET_FOREACH(vnet_iter) {
1061 		CURVNET_SET(vnet_iter);
1062 		struct inpcb *inpb;
1063 		struct tcpcb *tcpb;
1064 
1065 	/*
1066 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1067 	 * if there is one...
1068 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1069 	 *      reassembly queue should be flushed, but in a situation
1070 	 *	where we're really low on mbufs, this is potentially
1071 	 *	useful.
1072 	 */
1073 		INP_INFO_RLOCK(&V_tcbinfo);
1074 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1075 			if (inpb->inp_flags & INP_TIMEWAIT)
1076 				continue;
1077 			INP_WLOCK(inpb);
1078 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1079 				tcp_reass_flush(tcpb);
1080 				tcp_clean_sackreport(tcpb);
1081 			}
1082 			INP_WUNLOCK(inpb);
1083 		}
1084 		INP_INFO_RUNLOCK(&V_tcbinfo);
1085 		CURVNET_RESTORE();
1086 	}
1087 	VNET_LIST_RUNLOCK_NOSLEEP();
1088 }
1089 
1090 /*
1091  * Notify a tcp user of an asynchronous error;
1092  * store error as soft error, but wake up user
1093  * (for now, won't do anything until can select for soft error).
1094  *
1095  * Do not wake up user since there currently is no mechanism for
1096  * reporting soft errors (yet - a kqueue filter may be added).
1097  */
1098 static struct inpcb *
1099 tcp_notify(struct inpcb *inp, int error)
1100 {
1101 	struct tcpcb *tp;
1102 
1103 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1104 	INP_WLOCK_ASSERT(inp);
1105 
1106 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1107 	    (inp->inp_flags & INP_DROPPED))
1108 		return (inp);
1109 
1110 	tp = intotcpcb(inp);
1111 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1112 
1113 	/*
1114 	 * Ignore some errors if we are hooked up.
1115 	 * If connection hasn't completed, has retransmitted several times,
1116 	 * and receives a second error, give up now.  This is better
1117 	 * than waiting a long time to establish a connection that
1118 	 * can never complete.
1119 	 */
1120 	if (tp->t_state == TCPS_ESTABLISHED &&
1121 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1122 	     error == EHOSTDOWN)) {
1123 		return (inp);
1124 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1125 	    tp->t_softerror) {
1126 		tp = tcp_drop(tp, error);
1127 		if (tp != NULL)
1128 			return (inp);
1129 		else
1130 			return (NULL);
1131 	} else {
1132 		tp->t_softerror = error;
1133 		return (inp);
1134 	}
1135 #if 0
1136 	wakeup( &so->so_timeo);
1137 	sorwakeup(so);
1138 	sowwakeup(so);
1139 #endif
1140 }
1141 
1142 static int
1143 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1144 {
1145 	int error, i, m, n, pcb_count;
1146 	struct inpcb *inp, **inp_list;
1147 	inp_gen_t gencnt;
1148 	struct xinpgen xig;
1149 
1150 	/*
1151 	 * The process of preparing the TCB list is too time-consuming and
1152 	 * resource-intensive to repeat twice on every request.
1153 	 */
1154 	if (req->oldptr == NULL) {
1155 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1156 		n += imax(n / 8, 10);
1157 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1158 		return (0);
1159 	}
1160 
1161 	if (req->newptr != NULL)
1162 		return (EPERM);
1163 
1164 	/*
1165 	 * OK, now we're committed to doing something.
1166 	 */
1167 	INP_INFO_RLOCK(&V_tcbinfo);
1168 	gencnt = V_tcbinfo.ipi_gencnt;
1169 	n = V_tcbinfo.ipi_count;
1170 	INP_INFO_RUNLOCK(&V_tcbinfo);
1171 
1172 	m = syncache_pcbcount();
1173 
1174 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1175 		+ (n + m) * sizeof(struct xtcpcb));
1176 	if (error != 0)
1177 		return (error);
1178 
1179 	xig.xig_len = sizeof xig;
1180 	xig.xig_count = n + m;
1181 	xig.xig_gen = gencnt;
1182 	xig.xig_sogen = so_gencnt;
1183 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1184 	if (error)
1185 		return (error);
1186 
1187 	error = syncache_pcblist(req, m, &pcb_count);
1188 	if (error)
1189 		return (error);
1190 
1191 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1192 	if (inp_list == NULL)
1193 		return (ENOMEM);
1194 
1195 	INP_INFO_RLOCK(&V_tcbinfo);
1196 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1197 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1198 		INP_WLOCK(inp);
1199 		if (inp->inp_gencnt <= gencnt) {
1200 			/*
1201 			 * XXX: This use of cr_cansee(), introduced with
1202 			 * TCP state changes, is not quite right, but for
1203 			 * now, better than nothing.
1204 			 */
1205 			if (inp->inp_flags & INP_TIMEWAIT) {
1206 				if (intotw(inp) != NULL)
1207 					error = cr_cansee(req->td->td_ucred,
1208 					    intotw(inp)->tw_cred);
1209 				else
1210 					error = EINVAL;	/* Skip this inp. */
1211 			} else
1212 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1213 			if (error == 0) {
1214 				in_pcbref(inp);
1215 				inp_list[i++] = inp;
1216 			}
1217 		}
1218 		INP_WUNLOCK(inp);
1219 	}
1220 	INP_INFO_RUNLOCK(&V_tcbinfo);
1221 	n = i;
1222 
1223 	error = 0;
1224 	for (i = 0; i < n; i++) {
1225 		inp = inp_list[i];
1226 		INP_RLOCK(inp);
1227 		if (inp->inp_gencnt <= gencnt) {
1228 			struct xtcpcb xt;
1229 			void *inp_ppcb;
1230 
1231 			bzero(&xt, sizeof(xt));
1232 			xt.xt_len = sizeof xt;
1233 			/* XXX should avoid extra copy */
1234 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1235 			inp_ppcb = inp->inp_ppcb;
1236 			if (inp_ppcb == NULL)
1237 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1238 			else if (inp->inp_flags & INP_TIMEWAIT) {
1239 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1240 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1241 			} else {
1242 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1243 				if (xt.xt_tp.t_timers)
1244 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1245 			}
1246 			if (inp->inp_socket != NULL)
1247 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1248 			else {
1249 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1250 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1251 			}
1252 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1253 			INP_RUNLOCK(inp);
1254 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1255 		} else
1256 			INP_RUNLOCK(inp);
1257 	}
1258 	INP_INFO_WLOCK(&V_tcbinfo);
1259 	for (i = 0; i < n; i++) {
1260 		inp = inp_list[i];
1261 		INP_RLOCK(inp);
1262 		if (!in_pcbrele_rlocked(inp))
1263 			INP_RUNLOCK(inp);
1264 	}
1265 	INP_INFO_WUNLOCK(&V_tcbinfo);
1266 
1267 	if (!error) {
1268 		/*
1269 		 * Give the user an updated idea of our state.
1270 		 * If the generation differs from what we told
1271 		 * her before, she knows that something happened
1272 		 * while we were processing this request, and it
1273 		 * might be necessary to retry.
1274 		 */
1275 		INP_INFO_RLOCK(&V_tcbinfo);
1276 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1277 		xig.xig_sogen = so_gencnt;
1278 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1279 		INP_INFO_RUNLOCK(&V_tcbinfo);
1280 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1281 	}
1282 	free(inp_list, M_TEMP);
1283 	return (error);
1284 }
1285 
1286 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1287     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1288     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1289 
1290 #ifdef INET
1291 static int
1292 tcp_getcred(SYSCTL_HANDLER_ARGS)
1293 {
1294 	struct xucred xuc;
1295 	struct sockaddr_in addrs[2];
1296 	struct inpcb *inp;
1297 	int error;
1298 
1299 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1300 	if (error)
1301 		return (error);
1302 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1303 	if (error)
1304 		return (error);
1305 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1306 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1307 	if (inp != NULL) {
1308 		if (inp->inp_socket == NULL)
1309 			error = ENOENT;
1310 		if (error == 0)
1311 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1312 		if (error == 0)
1313 			cru2x(inp->inp_cred, &xuc);
1314 		INP_RUNLOCK(inp);
1315 	} else
1316 		error = ENOENT;
1317 	if (error == 0)
1318 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1319 	return (error);
1320 }
1321 
1322 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1323     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1324     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1325 #endif /* INET */
1326 
1327 #ifdef INET6
1328 static int
1329 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1330 {
1331 	struct xucred xuc;
1332 	struct sockaddr_in6 addrs[2];
1333 	struct inpcb *inp;
1334 	int error;
1335 #ifdef INET
1336 	int mapped = 0;
1337 #endif
1338 
1339 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1340 	if (error)
1341 		return (error);
1342 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1343 	if (error)
1344 		return (error);
1345 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1346 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1347 		return (error);
1348 	}
1349 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1350 #ifdef INET
1351 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1352 			mapped = 1;
1353 		else
1354 #endif
1355 			return (EINVAL);
1356 	}
1357 
1358 #ifdef INET
1359 	if (mapped == 1)
1360 		inp = in_pcblookup(&V_tcbinfo,
1361 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1362 			addrs[1].sin6_port,
1363 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1364 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1365 	else
1366 #endif
1367 		inp = in6_pcblookup(&V_tcbinfo,
1368 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1369 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1370 			INPLOOKUP_RLOCKPCB, NULL);
1371 	if (inp != NULL) {
1372 		if (inp->inp_socket == NULL)
1373 			error = ENOENT;
1374 		if (error == 0)
1375 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1376 		if (error == 0)
1377 			cru2x(inp->inp_cred, &xuc);
1378 		INP_RUNLOCK(inp);
1379 	} else
1380 		error = ENOENT;
1381 	if (error == 0)
1382 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1383 	return (error);
1384 }
1385 
1386 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1387     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1388     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1389 #endif /* INET6 */
1390 
1391 
1392 #ifdef INET
1393 void
1394 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1395 {
1396 	struct ip *ip = vip;
1397 	struct tcphdr *th;
1398 	struct in_addr faddr;
1399 	struct inpcb *inp;
1400 	struct tcpcb *tp;
1401 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1402 	struct icmp *icp;
1403 	struct in_conninfo inc;
1404 	tcp_seq icmp_tcp_seq;
1405 	int mtu;
1406 
1407 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1408 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1409 		return;
1410 
1411 	if (cmd == PRC_MSGSIZE)
1412 		notify = tcp_mtudisc_notify;
1413 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1414 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1415 		notify = tcp_drop_syn_sent;
1416 	/*
1417 	 * Redirects don't need to be handled up here.
1418 	 */
1419 	else if (PRC_IS_REDIRECT(cmd))
1420 		return;
1421 	/*
1422 	 * Source quench is depreciated.
1423 	 */
1424 	else if (cmd == PRC_QUENCH)
1425 		return;
1426 	/*
1427 	 * Hostdead is ugly because it goes linearly through all PCBs.
1428 	 * XXX: We never get this from ICMP, otherwise it makes an
1429 	 * excellent DoS attack on machines with many connections.
1430 	 */
1431 	else if (cmd == PRC_HOSTDEAD)
1432 		ip = NULL;
1433 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1434 		return;
1435 	if (ip != NULL) {
1436 		icp = (struct icmp *)((caddr_t)ip
1437 				      - offsetof(struct icmp, icmp_ip));
1438 		th = (struct tcphdr *)((caddr_t)ip
1439 				       + (ip->ip_hl << 2));
1440 		INP_INFO_WLOCK(&V_tcbinfo);
1441 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1442 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1443 		if (inp != NULL)  {
1444 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1445 			    !(inp->inp_flags & INP_DROPPED) &&
1446 			    !(inp->inp_socket == NULL)) {
1447 				icmp_tcp_seq = htonl(th->th_seq);
1448 				tp = intotcpcb(inp);
1449 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1450 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1451 					if (cmd == PRC_MSGSIZE) {
1452 					    /*
1453 					     * MTU discovery:
1454 					     * If we got a needfrag set the MTU
1455 					     * in the route to the suggested new
1456 					     * value (if given) and then notify.
1457 					     */
1458 					    bzero(&inc, sizeof(inc));
1459 					    inc.inc_faddr = faddr;
1460 					    inc.inc_fibnum =
1461 						inp->inp_inc.inc_fibnum;
1462 
1463 					    mtu = ntohs(icp->icmp_nextmtu);
1464 					    /*
1465 					     * If no alternative MTU was
1466 					     * proposed, try the next smaller
1467 					     * one.
1468 					     */
1469 					    if (!mtu)
1470 						mtu = ip_next_mtu(
1471 						 ntohs(ip->ip_len), 1);
1472 					    if (mtu < V_tcp_minmss
1473 						 + sizeof(struct tcpiphdr))
1474 						mtu = V_tcp_minmss
1475 						 + sizeof(struct tcpiphdr);
1476 					    /*
1477 					     * Only cache the MTU if it
1478 					     * is smaller than the interface
1479 					     * or route MTU.  tcp_mtudisc()
1480 					     * will do right thing by itself.
1481 					     */
1482 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1483 						tcp_hc_updatemtu(&inc, mtu);
1484 					    tcp_mtudisc(inp, mtu);
1485 					} else
1486 						inp = (*notify)(inp,
1487 						    inetctlerrmap[cmd]);
1488 				}
1489 			}
1490 			if (inp != NULL)
1491 				INP_WUNLOCK(inp);
1492 		} else {
1493 			bzero(&inc, sizeof(inc));
1494 			inc.inc_fport = th->th_dport;
1495 			inc.inc_lport = th->th_sport;
1496 			inc.inc_faddr = faddr;
1497 			inc.inc_laddr = ip->ip_src;
1498 			syncache_unreach(&inc, th);
1499 		}
1500 		INP_INFO_WUNLOCK(&V_tcbinfo);
1501 	} else
1502 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1503 }
1504 #endif /* INET */
1505 
1506 #ifdef INET6
1507 void
1508 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1509 {
1510 	struct tcphdr th;
1511 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1512 	struct ip6_hdr *ip6;
1513 	struct mbuf *m;
1514 	struct ip6ctlparam *ip6cp = NULL;
1515 	const struct sockaddr_in6 *sa6_src = NULL;
1516 	int off;
1517 	struct tcp_portonly {
1518 		u_int16_t th_sport;
1519 		u_int16_t th_dport;
1520 	} *thp;
1521 
1522 	if (sa->sa_family != AF_INET6 ||
1523 	    sa->sa_len != sizeof(struct sockaddr_in6))
1524 		return;
1525 
1526 	if (cmd == PRC_MSGSIZE)
1527 		notify = tcp_mtudisc_notify;
1528 	else if (!PRC_IS_REDIRECT(cmd) &&
1529 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1530 		return;
1531 	/* Source quench is depreciated. */
1532 	else if (cmd == PRC_QUENCH)
1533 		return;
1534 
1535 	/* if the parameter is from icmp6, decode it. */
1536 	if (d != NULL) {
1537 		ip6cp = (struct ip6ctlparam *)d;
1538 		m = ip6cp->ip6c_m;
1539 		ip6 = ip6cp->ip6c_ip6;
1540 		off = ip6cp->ip6c_off;
1541 		sa6_src = ip6cp->ip6c_src;
1542 	} else {
1543 		m = NULL;
1544 		ip6 = NULL;
1545 		off = 0;	/* fool gcc */
1546 		sa6_src = &sa6_any;
1547 	}
1548 
1549 	if (ip6 != NULL) {
1550 		struct in_conninfo inc;
1551 		/*
1552 		 * XXX: We assume that when IPV6 is non NULL,
1553 		 * M and OFF are valid.
1554 		 */
1555 
1556 		/* check if we can safely examine src and dst ports */
1557 		if (m->m_pkthdr.len < off + sizeof(*thp))
1558 			return;
1559 
1560 		bzero(&th, sizeof(th));
1561 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1562 
1563 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1564 		    (struct sockaddr *)ip6cp->ip6c_src,
1565 		    th.th_sport, cmd, NULL, notify);
1566 
1567 		bzero(&inc, sizeof(inc));
1568 		inc.inc_fport = th.th_dport;
1569 		inc.inc_lport = th.th_sport;
1570 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1571 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1572 		inc.inc_flags |= INC_ISIPV6;
1573 		INP_INFO_WLOCK(&V_tcbinfo);
1574 		syncache_unreach(&inc, &th);
1575 		INP_INFO_WUNLOCK(&V_tcbinfo);
1576 	} else
1577 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1578 			      0, cmd, NULL, notify);
1579 }
1580 #endif /* INET6 */
1581 
1582 
1583 /*
1584  * Following is where TCP initial sequence number generation occurs.
1585  *
1586  * There are two places where we must use initial sequence numbers:
1587  * 1.  In SYN-ACK packets.
1588  * 2.  In SYN packets.
1589  *
1590  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1591  * tcp_syncache.c for details.
1592  *
1593  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1594  * depends on this property.  In addition, these ISNs should be
1595  * unguessable so as to prevent connection hijacking.  To satisfy
1596  * the requirements of this situation, the algorithm outlined in
1597  * RFC 1948 is used, with only small modifications.
1598  *
1599  * Implementation details:
1600  *
1601  * Time is based off the system timer, and is corrected so that it
1602  * increases by one megabyte per second.  This allows for proper
1603  * recycling on high speed LANs while still leaving over an hour
1604  * before rollover.
1605  *
1606  * As reading the *exact* system time is too expensive to be done
1607  * whenever setting up a TCP connection, we increment the time
1608  * offset in two ways.  First, a small random positive increment
1609  * is added to isn_offset for each connection that is set up.
1610  * Second, the function tcp_isn_tick fires once per clock tick
1611  * and increments isn_offset as necessary so that sequence numbers
1612  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1613  * random positive increments serve only to ensure that the same
1614  * exact sequence number is never sent out twice (as could otherwise
1615  * happen when a port is recycled in less than the system tick
1616  * interval.)
1617  *
1618  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1619  * between seeding of isn_secret.  This is normally set to zero,
1620  * as reseeding should not be necessary.
1621  *
1622  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1623  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1624  * general, this means holding an exclusive (write) lock.
1625  */
1626 
1627 #define ISN_BYTES_PER_SECOND 1048576
1628 #define ISN_STATIC_INCREMENT 4096
1629 #define ISN_RANDOM_INCREMENT (4096 - 1)
1630 
1631 static VNET_DEFINE(u_char, isn_secret[32]);
1632 static VNET_DEFINE(int, isn_last);
1633 static VNET_DEFINE(int, isn_last_reseed);
1634 static VNET_DEFINE(u_int32_t, isn_offset);
1635 static VNET_DEFINE(u_int32_t, isn_offset_old);
1636 
1637 #define	V_isn_secret			VNET(isn_secret)
1638 #define	V_isn_last			VNET(isn_last)
1639 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1640 #define	V_isn_offset			VNET(isn_offset)
1641 #define	V_isn_offset_old		VNET(isn_offset_old)
1642 
1643 tcp_seq
1644 tcp_new_isn(struct tcpcb *tp)
1645 {
1646 	MD5_CTX isn_ctx;
1647 	u_int32_t md5_buffer[4];
1648 	tcp_seq new_isn;
1649 	u_int32_t projected_offset;
1650 
1651 	INP_WLOCK_ASSERT(tp->t_inpcb);
1652 
1653 	ISN_LOCK();
1654 	/* Seed if this is the first use, reseed if requested. */
1655 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1656 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1657 		< (u_int)ticks))) {
1658 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1659 		V_isn_last_reseed = ticks;
1660 	}
1661 
1662 	/* Compute the md5 hash and return the ISN. */
1663 	MD5Init(&isn_ctx);
1664 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1665 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1666 #ifdef INET6
1667 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1668 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1669 			  sizeof(struct in6_addr));
1670 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1671 			  sizeof(struct in6_addr));
1672 	} else
1673 #endif
1674 	{
1675 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1676 			  sizeof(struct in_addr));
1677 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1678 			  sizeof(struct in_addr));
1679 	}
1680 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1681 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1682 	new_isn = (tcp_seq) md5_buffer[0];
1683 	V_isn_offset += ISN_STATIC_INCREMENT +
1684 		(arc4random() & ISN_RANDOM_INCREMENT);
1685 	if (ticks != V_isn_last) {
1686 		projected_offset = V_isn_offset_old +
1687 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1688 		if (SEQ_GT(projected_offset, V_isn_offset))
1689 			V_isn_offset = projected_offset;
1690 		V_isn_offset_old = V_isn_offset;
1691 		V_isn_last = ticks;
1692 	}
1693 	new_isn += V_isn_offset;
1694 	ISN_UNLOCK();
1695 	return (new_isn);
1696 }
1697 
1698 /*
1699  * When a specific ICMP unreachable message is received and the
1700  * connection state is SYN-SENT, drop the connection.  This behavior
1701  * is controlled by the icmp_may_rst sysctl.
1702  */
1703 struct inpcb *
1704 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1705 {
1706 	struct tcpcb *tp;
1707 
1708 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1709 	INP_WLOCK_ASSERT(inp);
1710 
1711 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1712 	    (inp->inp_flags & INP_DROPPED))
1713 		return (inp);
1714 
1715 	tp = intotcpcb(inp);
1716 	if (tp->t_state != TCPS_SYN_SENT)
1717 		return (inp);
1718 
1719 	tp = tcp_drop(tp, errno);
1720 	if (tp != NULL)
1721 		return (inp);
1722 	else
1723 		return (NULL);
1724 }
1725 
1726 /*
1727  * When `need fragmentation' ICMP is received, update our idea of the MSS
1728  * based on the new value. Also nudge TCP to send something, since we
1729  * know the packet we just sent was dropped.
1730  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1731  */
1732 static struct inpcb *
1733 tcp_mtudisc_notify(struct inpcb *inp, int error)
1734 {
1735 
1736 	return (tcp_mtudisc(inp, -1));
1737 }
1738 
1739 struct inpcb *
1740 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1741 {
1742 	struct tcpcb *tp;
1743 	struct socket *so;
1744 
1745 	INP_WLOCK_ASSERT(inp);
1746 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1747 	    (inp->inp_flags & INP_DROPPED))
1748 		return (inp);
1749 
1750 	tp = intotcpcb(inp);
1751 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1752 
1753 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1754 
1755 	so = inp->inp_socket;
1756 	SOCKBUF_LOCK(&so->so_snd);
1757 	/* If the mss is larger than the socket buffer, decrease the mss. */
1758 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1759 		tp->t_maxseg = so->so_snd.sb_hiwat;
1760 	SOCKBUF_UNLOCK(&so->so_snd);
1761 
1762 	TCPSTAT_INC(tcps_mturesent);
1763 	tp->t_rtttime = 0;
1764 	tp->snd_nxt = tp->snd_una;
1765 	tcp_free_sackholes(tp);
1766 	tp->snd_recover = tp->snd_max;
1767 	if (tp->t_flags & TF_SACK_PERMIT)
1768 		EXIT_FASTRECOVERY(tp->t_flags);
1769 	tcp_output(tp);
1770 	return (inp);
1771 }
1772 
1773 #ifdef INET
1774 /*
1775  * Look-up the routing entry to the peer of this inpcb.  If no route
1776  * is found and it cannot be allocated, then return 0.  This routine
1777  * is called by TCP routines that access the rmx structure and by
1778  * tcp_mss_update to get the peer/interface MTU.
1779  */
1780 u_long
1781 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1782 {
1783 	struct route sro;
1784 	struct sockaddr_in *dst;
1785 	struct ifnet *ifp;
1786 	u_long maxmtu = 0;
1787 
1788 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1789 
1790 	bzero(&sro, sizeof(sro));
1791 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1792 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1793 		dst->sin_family = AF_INET;
1794 		dst->sin_len = sizeof(*dst);
1795 		dst->sin_addr = inc->inc_faddr;
1796 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1797 	}
1798 	if (sro.ro_rt != NULL) {
1799 		ifp = sro.ro_rt->rt_ifp;
1800 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1801 			maxmtu = ifp->if_mtu;
1802 		else
1803 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1804 
1805 		/* Report additional interface capabilities. */
1806 		if (cap != NULL) {
1807 			if (ifp->if_capenable & IFCAP_TSO4 &&
1808 			    ifp->if_hwassist & CSUM_TSO)
1809 				cap->ifcap |= CSUM_TSO;
1810 				cap->tsomax = ifp->if_hw_tsomax;
1811 		}
1812 		RTFREE(sro.ro_rt);
1813 	}
1814 	return (maxmtu);
1815 }
1816 #endif /* INET */
1817 
1818 #ifdef INET6
1819 u_long
1820 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1821 {
1822 	struct route_in6 sro6;
1823 	struct ifnet *ifp;
1824 	u_long maxmtu = 0;
1825 
1826 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1827 
1828 	bzero(&sro6, sizeof(sro6));
1829 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1830 		sro6.ro_dst.sin6_family = AF_INET6;
1831 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1832 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1833 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1834 	}
1835 	if (sro6.ro_rt != NULL) {
1836 		ifp = sro6.ro_rt->rt_ifp;
1837 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1838 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1839 		else
1840 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1841 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1842 
1843 		/* Report additional interface capabilities. */
1844 		if (cap != NULL) {
1845 			if (ifp->if_capenable & IFCAP_TSO6 &&
1846 			    ifp->if_hwassist & CSUM_TSO)
1847 				cap->ifcap |= CSUM_TSO;
1848 				cap->tsomax = ifp->if_hw_tsomax;
1849 		}
1850 		RTFREE(sro6.ro_rt);
1851 	}
1852 
1853 	return (maxmtu);
1854 }
1855 #endif /* INET6 */
1856 
1857 #ifdef IPSEC
1858 /* compute ESP/AH header size for TCP, including outer IP header. */
1859 size_t
1860 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1861 {
1862 	struct inpcb *inp;
1863 	struct mbuf *m;
1864 	size_t hdrsiz;
1865 	struct ip *ip;
1866 #ifdef INET6
1867 	struct ip6_hdr *ip6;
1868 #endif
1869 	struct tcphdr *th;
1870 
1871 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1872 		return (0);
1873 	m = m_gethdr(M_NOWAIT, MT_DATA);
1874 	if (!m)
1875 		return (0);
1876 
1877 #ifdef INET6
1878 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1879 		ip6 = mtod(m, struct ip6_hdr *);
1880 		th = (struct tcphdr *)(ip6 + 1);
1881 		m->m_pkthdr.len = m->m_len =
1882 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1883 		tcpip_fillheaders(inp, ip6, th);
1884 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1885 	} else
1886 #endif /* INET6 */
1887 	{
1888 		ip = mtod(m, struct ip *);
1889 		th = (struct tcphdr *)(ip + 1);
1890 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1891 		tcpip_fillheaders(inp, ip, th);
1892 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1893 	}
1894 
1895 	m_free(m);
1896 	return (hdrsiz);
1897 }
1898 #endif /* IPSEC */
1899 
1900 #ifdef TCP_SIGNATURE
1901 /*
1902  * Callback function invoked by m_apply() to digest TCP segment data
1903  * contained within an mbuf chain.
1904  */
1905 static int
1906 tcp_signature_apply(void *fstate, void *data, u_int len)
1907 {
1908 
1909 	MD5Update(fstate, (u_char *)data, len);
1910 	return (0);
1911 }
1912 
1913 /*
1914  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1915  *
1916  * Parameters:
1917  * m		pointer to head of mbuf chain
1918  * _unused
1919  * len		length of TCP segment data, excluding options
1920  * optlen	length of TCP segment options
1921  * buf		pointer to storage for computed MD5 digest
1922  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1923  *
1924  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1925  * When called from tcp_input(), we can be sure that th_sum has been
1926  * zeroed out and verified already.
1927  *
1928  * Return 0 if successful, otherwise return -1.
1929  *
1930  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1931  * search with the destination IP address, and a 'magic SPI' to be
1932  * determined by the application. This is hardcoded elsewhere to 1179
1933  * right now. Another branch of this code exists which uses the SPD to
1934  * specify per-application flows but it is unstable.
1935  */
1936 int
1937 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1938     u_char *buf, u_int direction)
1939 {
1940 	union sockaddr_union dst;
1941 #ifdef INET
1942 	struct ippseudo ippseudo;
1943 #endif
1944 	MD5_CTX ctx;
1945 	int doff;
1946 	struct ip *ip;
1947 #ifdef INET
1948 	struct ipovly *ipovly;
1949 #endif
1950 	struct secasvar *sav;
1951 	struct tcphdr *th;
1952 #ifdef INET6
1953 	struct ip6_hdr *ip6;
1954 	struct in6_addr in6;
1955 	char ip6buf[INET6_ADDRSTRLEN];
1956 	uint32_t plen;
1957 	uint16_t nhdr;
1958 #endif
1959 	u_short savecsum;
1960 
1961 	KASSERT(m != NULL, ("NULL mbuf chain"));
1962 	KASSERT(buf != NULL, ("NULL signature pointer"));
1963 
1964 	/* Extract the destination from the IP header in the mbuf. */
1965 	bzero(&dst, sizeof(union sockaddr_union));
1966 	ip = mtod(m, struct ip *);
1967 #ifdef INET6
1968 	ip6 = NULL;	/* Make the compiler happy. */
1969 #endif
1970 	switch (ip->ip_v) {
1971 #ifdef INET
1972 	case IPVERSION:
1973 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1974 		dst.sa.sa_family = AF_INET;
1975 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1976 		    ip->ip_src : ip->ip_dst;
1977 		break;
1978 #endif
1979 #ifdef INET6
1980 	case (IPV6_VERSION >> 4):
1981 		ip6 = mtod(m, struct ip6_hdr *);
1982 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1983 		dst.sa.sa_family = AF_INET6;
1984 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1985 		    ip6->ip6_src : ip6->ip6_dst;
1986 		break;
1987 #endif
1988 	default:
1989 		return (EINVAL);
1990 		/* NOTREACHED */
1991 		break;
1992 	}
1993 
1994 	/* Look up an SADB entry which matches the address of the peer. */
1995 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1996 	if (sav == NULL) {
1997 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1998 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1999 #ifdef INET6
2000 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2001 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2002 #endif
2003 			"(unsupported)"));
2004 		return (EINVAL);
2005 	}
2006 
2007 	MD5Init(&ctx);
2008 	/*
2009 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2010 	 *
2011 	 * XXX The ippseudo header MUST be digested in network byte order,
2012 	 * or else we'll fail the regression test. Assume all fields we've
2013 	 * been doing arithmetic on have been in host byte order.
2014 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2015 	 * tcp_output(), the underlying ip_len member has not yet been set.
2016 	 */
2017 	switch (ip->ip_v) {
2018 #ifdef INET
2019 	case IPVERSION:
2020 		ipovly = (struct ipovly *)ip;
2021 		ippseudo.ippseudo_src = ipovly->ih_src;
2022 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2023 		ippseudo.ippseudo_pad = 0;
2024 		ippseudo.ippseudo_p = IPPROTO_TCP;
2025 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2026 		    optlen);
2027 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2028 
2029 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2030 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2031 		break;
2032 #endif
2033 #ifdef INET6
2034 	/*
2035 	 * RFC 2385, 2.0  Proposal
2036 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2037 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2038 	 * extended next header value (to form 32 bits), and 32-bit segment
2039 	 * length.
2040 	 * Note: Upper-Layer Packet Length comes before Next Header.
2041 	 */
2042 	case (IPV6_VERSION >> 4):
2043 		in6 = ip6->ip6_src;
2044 		in6_clearscope(&in6);
2045 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2046 		in6 = ip6->ip6_dst;
2047 		in6_clearscope(&in6);
2048 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2049 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2050 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2051 		nhdr = 0;
2052 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2053 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2054 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2055 		nhdr = IPPROTO_TCP;
2056 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2057 
2058 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2059 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2060 		break;
2061 #endif
2062 	default:
2063 		return (EINVAL);
2064 		/* NOTREACHED */
2065 		break;
2066 	}
2067 
2068 
2069 	/*
2070 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2071 	 * The TCP checksum must be set to zero.
2072 	 */
2073 	savecsum = th->th_sum;
2074 	th->th_sum = 0;
2075 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2076 	th->th_sum = savecsum;
2077 
2078 	/*
2079 	 * Step 3: Update MD5 hash with TCP segment data.
2080 	 *         Use m_apply() to avoid an early m_pullup().
2081 	 */
2082 	if (len > 0)
2083 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2084 
2085 	/*
2086 	 * Step 4: Update MD5 hash with shared secret.
2087 	 */
2088 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2089 	MD5Final(buf, &ctx);
2090 
2091 	key_sa_recordxfer(sav, m);
2092 	KEY_FREESAV(&sav);
2093 	return (0);
2094 }
2095 
2096 /*
2097  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2098  *
2099  * Parameters:
2100  * m		pointer to head of mbuf chain
2101  * len		length of TCP segment data, excluding options
2102  * optlen	length of TCP segment options
2103  * buf		pointer to storage for computed MD5 digest
2104  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2105  *
2106  * Return 1 if successful, otherwise return 0.
2107  */
2108 int
2109 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2110     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2111 {
2112 	char tmpdigest[TCP_SIGLEN];
2113 
2114 	if (tcp_sig_checksigs == 0)
2115 		return (1);
2116 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2117 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2118 
2119 			/*
2120 			 * If this socket is not expecting signature but
2121 			 * the segment contains signature just fail.
2122 			 */
2123 			TCPSTAT_INC(tcps_sig_err_sigopt);
2124 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2125 			return (0);
2126 		}
2127 
2128 		/* Signature is not expected, and not present in segment. */
2129 		return (1);
2130 	}
2131 
2132 	/*
2133 	 * If this socket is expecting signature but the segment does not
2134 	 * contain any just fail.
2135 	 */
2136 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2137 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2138 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2139 		return (0);
2140 	}
2141 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2142 	    IPSEC_DIR_INBOUND) == -1) {
2143 		TCPSTAT_INC(tcps_sig_err_buildsig);
2144 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2145 		return (0);
2146 	}
2147 
2148 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2149 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2150 		return (0);
2151 	}
2152 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2153 	return (1);
2154 }
2155 #endif /* TCP_SIGNATURE */
2156 
2157 static int
2158 sysctl_drop(SYSCTL_HANDLER_ARGS)
2159 {
2160 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2161 	struct sockaddr_storage addrs[2];
2162 	struct inpcb *inp;
2163 	struct tcpcb *tp;
2164 	struct tcptw *tw;
2165 	struct sockaddr_in *fin, *lin;
2166 #ifdef INET6
2167 	struct sockaddr_in6 *fin6, *lin6;
2168 #endif
2169 	int error;
2170 
2171 	inp = NULL;
2172 	fin = lin = NULL;
2173 #ifdef INET6
2174 	fin6 = lin6 = NULL;
2175 #endif
2176 	error = 0;
2177 
2178 	if (req->oldptr != NULL || req->oldlen != 0)
2179 		return (EINVAL);
2180 	if (req->newptr == NULL)
2181 		return (EPERM);
2182 	if (req->newlen < sizeof(addrs))
2183 		return (ENOMEM);
2184 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2185 	if (error)
2186 		return (error);
2187 
2188 	switch (addrs[0].ss_family) {
2189 #ifdef INET6
2190 	case AF_INET6:
2191 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2192 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2193 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2194 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2195 			return (EINVAL);
2196 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2197 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2198 				return (EINVAL);
2199 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2200 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2201 			fin = (struct sockaddr_in *)&addrs[0];
2202 			lin = (struct sockaddr_in *)&addrs[1];
2203 			break;
2204 		}
2205 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2206 		if (error)
2207 			return (error);
2208 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2209 		if (error)
2210 			return (error);
2211 		break;
2212 #endif
2213 #ifdef INET
2214 	case AF_INET:
2215 		fin = (struct sockaddr_in *)&addrs[0];
2216 		lin = (struct sockaddr_in *)&addrs[1];
2217 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2218 		    lin->sin_len != sizeof(struct sockaddr_in))
2219 			return (EINVAL);
2220 		break;
2221 #endif
2222 	default:
2223 		return (EINVAL);
2224 	}
2225 	INP_INFO_WLOCK(&V_tcbinfo);
2226 	switch (addrs[0].ss_family) {
2227 #ifdef INET6
2228 	case AF_INET6:
2229 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2230 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2231 		    INPLOOKUP_WLOCKPCB, NULL);
2232 		break;
2233 #endif
2234 #ifdef INET
2235 	case AF_INET:
2236 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2237 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2238 		break;
2239 #endif
2240 	}
2241 	if (inp != NULL) {
2242 		if (inp->inp_flags & INP_TIMEWAIT) {
2243 			/*
2244 			 * XXXRW: There currently exists a state where an
2245 			 * inpcb is present, but its timewait state has been
2246 			 * discarded.  For now, don't allow dropping of this
2247 			 * type of inpcb.
2248 			 */
2249 			tw = intotw(inp);
2250 			if (tw != NULL)
2251 				tcp_twclose(tw, 0);
2252 			else
2253 				INP_WUNLOCK(inp);
2254 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2255 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2256 			tp = intotcpcb(inp);
2257 			tp = tcp_drop(tp, ECONNABORTED);
2258 			if (tp != NULL)
2259 				INP_WUNLOCK(inp);
2260 		} else
2261 			INP_WUNLOCK(inp);
2262 	} else
2263 		error = ESRCH;
2264 	INP_INFO_WUNLOCK(&V_tcbinfo);
2265 	return (error);
2266 }
2267 
2268 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2269     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2270     0, sysctl_drop, "", "Drop TCP connection");
2271 
2272 /*
2273  * Generate a standardized TCP log line for use throughout the
2274  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2275  * allow use in the interrupt context.
2276  *
2277  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2278  * NB: The function may return NULL if memory allocation failed.
2279  *
2280  * Due to header inclusion and ordering limitations the struct ip
2281  * and ip6_hdr pointers have to be passed as void pointers.
2282  */
2283 char *
2284 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2285     const void *ip6hdr)
2286 {
2287 
2288 	/* Is logging enabled? */
2289 	if (tcp_log_in_vain == 0)
2290 		return (NULL);
2291 
2292 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2293 }
2294 
2295 char *
2296 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2297     const void *ip6hdr)
2298 {
2299 
2300 	/* Is logging enabled? */
2301 	if (tcp_log_debug == 0)
2302 		return (NULL);
2303 
2304 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2305 }
2306 
2307 static char *
2308 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2309     const void *ip6hdr)
2310 {
2311 	char *s, *sp;
2312 	size_t size;
2313 	struct ip *ip;
2314 #ifdef INET6
2315 	const struct ip6_hdr *ip6;
2316 
2317 	ip6 = (const struct ip6_hdr *)ip6hdr;
2318 #endif /* INET6 */
2319 	ip = (struct ip *)ip4hdr;
2320 
2321 	/*
2322 	 * The log line looks like this:
2323 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2324 	 */
2325 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2326 	    sizeof(PRINT_TH_FLAGS) + 1 +
2327 #ifdef INET6
2328 	    2 * INET6_ADDRSTRLEN;
2329 #else
2330 	    2 * INET_ADDRSTRLEN;
2331 #endif /* INET6 */
2332 
2333 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2334 	if (s == NULL)
2335 		return (NULL);
2336 
2337 	strcat(s, "TCP: [");
2338 	sp = s + strlen(s);
2339 
2340 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2341 		inet_ntoa_r(inc->inc_faddr, sp);
2342 		sp = s + strlen(s);
2343 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2344 		sp = s + strlen(s);
2345 		inet_ntoa_r(inc->inc_laddr, sp);
2346 		sp = s + strlen(s);
2347 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2348 #ifdef INET6
2349 	} else if (inc) {
2350 		ip6_sprintf(sp, &inc->inc6_faddr);
2351 		sp = s + strlen(s);
2352 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2353 		sp = s + strlen(s);
2354 		ip6_sprintf(sp, &inc->inc6_laddr);
2355 		sp = s + strlen(s);
2356 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2357 	} else if (ip6 && th) {
2358 		ip6_sprintf(sp, &ip6->ip6_src);
2359 		sp = s + strlen(s);
2360 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2361 		sp = s + strlen(s);
2362 		ip6_sprintf(sp, &ip6->ip6_dst);
2363 		sp = s + strlen(s);
2364 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2365 #endif /* INET6 */
2366 #ifdef INET
2367 	} else if (ip && th) {
2368 		inet_ntoa_r(ip->ip_src, sp);
2369 		sp = s + strlen(s);
2370 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2371 		sp = s + strlen(s);
2372 		inet_ntoa_r(ip->ip_dst, sp);
2373 		sp = s + strlen(s);
2374 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2375 #endif /* INET */
2376 	} else {
2377 		free(s, M_TCPLOG);
2378 		return (NULL);
2379 	}
2380 	sp = s + strlen(s);
2381 	if (th)
2382 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2383 	if (*(s + size - 1) != '\0')
2384 		panic("%s: string too long", __func__);
2385 	return (s);
2386 }
2387 
2388 /*
2389  * A subroutine which makes it easy to track TCP state changes with DTrace.
2390  * This function shouldn't be called for t_state initializations that don't
2391  * correspond to actual TCP state transitions.
2392  */
2393 void
2394 tcp_state_change(struct tcpcb *tp, int newstate)
2395 {
2396 #if defined(KDTRACE_HOOKS)
2397 	int pstate = tp->t_state;
2398 #endif
2399 
2400 	tp->t_state = newstate;
2401 	TCP_PROBE6(state_change, NULL, tp, NULL, tp, NULL, pstate);
2402 }
2403