xref: /freebsd/sys/netinet/tcp_subr.c (revision 37a60314613bc8baf1a46d9981e7c2f674dbe1cd)
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/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/protosw.h>
59 #include <sys/random.h>
60 
61 #include <vm/uma.h>
62 
63 #include <net/route.h>
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/cc.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet/ip6.h>
77 #include <netinet6/in6_pcb.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #ifdef INET6
89 #include <netinet6/tcp6_var.h>
90 #endif
91 #include <netinet/tcpip.h>
92 #ifdef TCPDEBUG
93 #include <netinet/tcp_debug.h>
94 #endif
95 #ifdef INET6
96 #include <netinet6/ip6protosw.h>
97 #endif
98 #ifdef TCP_OFFLOAD
99 #include <netinet/tcp_offload.h>
100 #endif
101 
102 #ifdef IPSEC
103 #include <netipsec/ipsec.h>
104 #include <netipsec/xform.h>
105 #ifdef INET6
106 #include <netipsec/ipsec6.h>
107 #endif
108 #include <netipsec/key.h>
109 #include <sys/syslog.h>
110 #endif /*IPSEC*/
111 
112 #include <machine/in_cksum.h>
113 #include <sys/md5.h>
114 
115 #include <security/mac/mac_framework.h>
116 
117 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
118 #ifdef INET6
119 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
120 #endif
121 
122 static int
123 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
124 {
125 	int error, new;
126 
127 	new = V_tcp_mssdflt;
128 	error = sysctl_handle_int(oidp, &new, 0, req);
129 	if (error == 0 && req->newptr) {
130 		if (new < TCP_MINMSS)
131 			error = EINVAL;
132 		else
133 			V_tcp_mssdflt = new;
134 	}
135 	return (error);
136 }
137 
138 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
139     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
140     &sysctl_net_inet_tcp_mss_check, "I",
141     "Default TCP Maximum Segment Size");
142 
143 #ifdef INET6
144 static int
145 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
146 {
147 	int error, new;
148 
149 	new = V_tcp_v6mssdflt;
150 	error = sysctl_handle_int(oidp, &new, 0, req);
151 	if (error == 0 && req->newptr) {
152 		if (new < TCP_MINMSS)
153 			error = EINVAL;
154 		else
155 			V_tcp_v6mssdflt = new;
156 	}
157 	return (error);
158 }
159 
160 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
161     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
162     &sysctl_net_inet_tcp_mss_v6_check, "I",
163    "Default TCP Maximum Segment Size for IPv6");
164 #endif /* INET6 */
165 
166 /*
167  * Minimum MSS we accept and use. This prevents DoS attacks where
168  * we are forced to a ridiculous low MSS like 20 and send hundreds
169  * of packets instead of one. The effect scales with the available
170  * bandwidth and quickly saturates the CPU and network interface
171  * with packet generation and sending. Set to zero to disable MINMSS
172  * checking. This setting prevents us from sending too small packets.
173  */
174 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
176      &VNET_NAME(tcp_minmss), 0,
177     "Minmum TCP Maximum Segment Size");
178 
179 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
180 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
181     &VNET_NAME(tcp_do_rfc1323), 0,
182     "Enable rfc1323 (high performance TCP) extensions");
183 
184 static int	tcp_log_debug = 0;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
186     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
187 
188 static int	tcp_tcbhashsize = 0;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
190     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
191 
192 static int	do_tcpdrain = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
194     "Enable tcp_drain routine for extra help when low on mbufs");
195 
196 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
197     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
198 
199 static VNET_DEFINE(int, icmp_may_rst) = 1;
200 #define	V_icmp_may_rst			VNET(icmp_may_rst)
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
202     &VNET_NAME(icmp_may_rst), 0,
203     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
204 
205 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
206 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
207 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
208     &VNET_NAME(tcp_isn_reseed_interval), 0,
209     "Seconds between reseeding of ISN secret");
210 
211 static int	tcp_soreceive_stream = 0;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
213     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
214 
215 #ifdef TCP_SIGNATURE
216 static int	tcp_sig_checksigs = 1;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
218     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
219 #endif
220 
221 VNET_DEFINE(uma_zone_t, sack_hole_zone);
222 #define	V_sack_hole_zone		VNET(sack_hole_zone)
223 
224 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
225 
226 static struct inpcb *tcp_notify(struct inpcb *, int);
227 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
228 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
229 		    void *ip4hdr, const void *ip6hdr);
230 
231 /*
232  * Target size of TCP PCB hash tables. Must be a power of two.
233  *
234  * Note that this can be overridden by the kernel environment
235  * variable net.inet.tcp.tcbhashsize
236  */
237 #ifndef TCBHASHSIZE
238 #define TCBHASHSIZE	0
239 #endif
240 
241 /*
242  * XXX
243  * Callouts should be moved into struct tcp directly.  They are currently
244  * separate because the tcpcb structure is exported to userland for sysctl
245  * parsing purposes, which do not know about callouts.
246  */
247 struct tcpcb_mem {
248 	struct	tcpcb		tcb;
249 	struct	tcp_timer	tt;
250 	struct	cc_var		ccv;
251 	struct	osd		osd;
252 };
253 
254 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
255 #define	V_tcpcb_zone			VNET(tcpcb_zone)
256 
257 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
258 static struct mtx isn_mtx;
259 
260 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
261 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
262 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
263 
264 /*
265  * TCP initialization.
266  */
267 static void
268 tcp_zone_change(void *tag)
269 {
270 
271 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
272 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
273 	tcp_tw_zone_change();
274 }
275 
276 static int
277 tcp_inpcb_init(void *mem, int size, int flags)
278 {
279 	struct inpcb *inp = mem;
280 
281 	INP_LOCK_INIT(inp, "inp", "tcpinp");
282 	return (0);
283 }
284 
285 /*
286  * Take a value and get the next power of 2 that doesn't overflow.
287  * Used to size the tcp_inpcb hash buckets.
288  */
289 static int
290 maketcp_hashsize(int size)
291 {
292 	int hashsize;
293 
294 	/*
295 	 * auto tune.
296 	 * get the next power of 2 higher than maxsockets.
297 	 */
298 	hashsize = 1 << fls(size);
299 	/* catch overflow, and just go one power of 2 smaller */
300 	if (hashsize < size) {
301 		hashsize = 1 << (fls(size) - 1);
302 	}
303 	return (hashsize);
304 }
305 
306 void
307 tcp_init(void)
308 {
309 	const char *tcbhash_tuneable;
310 	int hashsize;
311 
312 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
313 
314 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
315 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
316 		printf("%s: WARNING: unable to register helper hook\n", __func__);
317 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
318 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 		printf("%s: WARNING: unable to register helper hook\n", __func__);
320 
321 	hashsize = TCBHASHSIZE;
322 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
323 	if (hashsize == 0) {
324 		/*
325 		 * Auto tune the hash size based on maxsockets.
326 		 * A perfect hash would have a 1:1 mapping
327 		 * (hashsize = maxsockets) however it's been
328 		 * suggested that O(2) average is better.
329 		 */
330 		hashsize = maketcp_hashsize(maxsockets / 4);
331 		/*
332 		 * Our historical default is 512,
333 		 * do not autotune lower than this.
334 		 */
335 		if (hashsize < 512)
336 			hashsize = 512;
337 		if (bootverbose)
338 			printf("%s: %s auto tuned to %d\n", __func__,
339 			    tcbhash_tuneable, hashsize);
340 	}
341 	/*
342 	 * We require a hashsize to be a power of two.
343 	 * Previously if it was not a power of two we would just reset it
344 	 * back to 512, which could be a nasty surprise if you did not notice
345 	 * the error message.
346 	 * Instead what we do is clip it to the closest power of two lower
347 	 * than the specified hash value.
348 	 */
349 	if (!powerof2(hashsize)) {
350 		int oldhashsize = hashsize;
351 
352 		hashsize = maketcp_hashsize(hashsize);
353 		/* prevent absurdly low value */
354 		if (hashsize < 16)
355 			hashsize = 16;
356 		printf("%s: WARNING: TCB hash size not a power of 2, "
357 		    "clipped from %d to %d.\n", __func__, oldhashsize,
358 		    hashsize);
359 	}
360 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
361 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
362 	    IPI_HASHFIELDS_4TUPLE);
363 
364 	/*
365 	 * These have to be type stable for the benefit of the timers.
366 	 */
367 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
368 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
369 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
370 
371 	tcp_tw_init();
372 	syncache_init();
373 	tcp_hc_init();
374 	tcp_reass_init();
375 
376 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
377 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
378 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
379 
380 	/* Skip initialization of globals for non-default instances. */
381 	if (!IS_DEFAULT_VNET(curvnet))
382 		return;
383 
384 	/* XXX virtualize those bellow? */
385 	tcp_delacktime = TCPTV_DELACK;
386 	tcp_keepinit = TCPTV_KEEP_INIT;
387 	tcp_keepidle = TCPTV_KEEP_IDLE;
388 	tcp_keepintvl = TCPTV_KEEPINTVL;
389 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
390 	tcp_msl = TCPTV_MSL;
391 	tcp_rexmit_min = TCPTV_MIN;
392 	if (tcp_rexmit_min < 1)
393 		tcp_rexmit_min = 1;
394 	tcp_rexmit_slop = TCPTV_CPU_VAR;
395 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
396 	tcp_tcbhashsize = hashsize;
397 
398 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
399 	if (tcp_soreceive_stream) {
400 #ifdef INET
401 		tcp_usrreqs.pru_soreceive = soreceive_stream;
402 #endif
403 #ifdef INET6
404 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
405 #endif /* INET6 */
406 	}
407 
408 #ifdef INET6
409 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
410 #else /* INET6 */
411 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
412 #endif /* INET6 */
413 	if (max_protohdr < TCP_MINPROTOHDR)
414 		max_protohdr = TCP_MINPROTOHDR;
415 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
416 		panic("tcp_init");
417 #undef TCP_MINPROTOHDR
418 
419 	ISN_LOCK_INIT();
420 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
421 		SHUTDOWN_PRI_DEFAULT);
422 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
423 		EVENTHANDLER_PRI_ANY);
424 }
425 
426 #ifdef VIMAGE
427 void
428 tcp_destroy(void)
429 {
430 
431 	tcp_reass_destroy();
432 	tcp_hc_destroy();
433 	syncache_destroy();
434 	tcp_tw_destroy();
435 	in_pcbinfo_destroy(&V_tcbinfo);
436 	uma_zdestroy(V_sack_hole_zone);
437 	uma_zdestroy(V_tcpcb_zone);
438 }
439 #endif
440 
441 void
442 tcp_fini(void *xtp)
443 {
444 
445 }
446 
447 /*
448  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
449  * tcp_template used to store this data in mbufs, but we now recopy it out
450  * of the tcpcb each time to conserve mbufs.
451  */
452 void
453 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
454 {
455 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
456 
457 	INP_WLOCK_ASSERT(inp);
458 
459 #ifdef INET6
460 	if ((inp->inp_vflag & INP_IPV6) != 0) {
461 		struct ip6_hdr *ip6;
462 
463 		ip6 = (struct ip6_hdr *)ip_ptr;
464 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
465 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
466 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
467 			(IPV6_VERSION & IPV6_VERSION_MASK);
468 		ip6->ip6_nxt = IPPROTO_TCP;
469 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
470 		ip6->ip6_src = inp->in6p_laddr;
471 		ip6->ip6_dst = inp->in6p_faddr;
472 	}
473 #endif /* INET6 */
474 #if defined(INET6) && defined(INET)
475 	else
476 #endif
477 #ifdef INET
478 	{
479 		struct ip *ip;
480 
481 		ip = (struct ip *)ip_ptr;
482 		ip->ip_v = IPVERSION;
483 		ip->ip_hl = 5;
484 		ip->ip_tos = inp->inp_ip_tos;
485 		ip->ip_len = 0;
486 		ip->ip_id = 0;
487 		ip->ip_off = 0;
488 		ip->ip_ttl = inp->inp_ip_ttl;
489 		ip->ip_sum = 0;
490 		ip->ip_p = IPPROTO_TCP;
491 		ip->ip_src = inp->inp_laddr;
492 		ip->ip_dst = inp->inp_faddr;
493 	}
494 #endif /* INET */
495 	th->th_sport = inp->inp_lport;
496 	th->th_dport = inp->inp_fport;
497 	th->th_seq = 0;
498 	th->th_ack = 0;
499 	th->th_x2 = 0;
500 	th->th_off = 5;
501 	th->th_flags = 0;
502 	th->th_win = 0;
503 	th->th_urp = 0;
504 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
505 }
506 
507 /*
508  * Create template to be used to send tcp packets on a connection.
509  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
510  * use for this function is in keepalives, which use tcp_respond.
511  */
512 struct tcptemp *
513 tcpip_maketemplate(struct inpcb *inp)
514 {
515 	struct tcptemp *t;
516 
517 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
518 	if (t == NULL)
519 		return (NULL);
520 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
521 	return (t);
522 }
523 
524 /*
525  * Send a single message to the TCP at address specified by
526  * the given TCP/IP header.  If m == NULL, then we make a copy
527  * of the tcpiphdr at ti and send directly to the addressed host.
528  * This is used to force keep alive messages out using the TCP
529  * template for a connection.  If flags are given then we send
530  * a message back to the TCP which originated the * segment ti,
531  * and discard the mbuf containing it and any other attached mbufs.
532  *
533  * In any case the ack and sequence number of the transmitted
534  * segment are as specified by the parameters.
535  *
536  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
537  */
538 void
539 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
540     tcp_seq ack, tcp_seq seq, int flags)
541 {
542 	int tlen;
543 	int win = 0;
544 	struct ip *ip;
545 	struct tcphdr *nth;
546 #ifdef INET6
547 	struct ip6_hdr *ip6;
548 	int isipv6;
549 #endif /* INET6 */
550 	int ipflags = 0;
551 	struct inpcb *inp;
552 
553 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
554 
555 #ifdef INET6
556 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
557 	ip6 = ipgen;
558 #endif /* INET6 */
559 	ip = ipgen;
560 
561 	if (tp != NULL) {
562 		inp = tp->t_inpcb;
563 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
564 		INP_WLOCK_ASSERT(inp);
565 	} else
566 		inp = NULL;
567 
568 	if (tp != NULL) {
569 		if (!(flags & TH_RST)) {
570 			win = sbspace(&inp->inp_socket->so_rcv);
571 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
572 				win = (long)TCP_MAXWIN << tp->rcv_scale;
573 		}
574 	}
575 	if (m == NULL) {
576 		m = m_gethdr(M_NOWAIT, MT_DATA);
577 		if (m == NULL)
578 			return;
579 		tlen = 0;
580 		m->m_data += max_linkhdr;
581 #ifdef INET6
582 		if (isipv6) {
583 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
584 			      sizeof(struct ip6_hdr));
585 			ip6 = mtod(m, struct ip6_hdr *);
586 			nth = (struct tcphdr *)(ip6 + 1);
587 		} else
588 #endif /* INET6 */
589 		{
590 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
591 			ip = mtod(m, struct ip *);
592 			nth = (struct tcphdr *)(ip + 1);
593 		}
594 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
595 		flags = TH_ACK;
596 	} else {
597 		/*
598 		 *  reuse the mbuf.
599 		 * XXX MRT We inherrit the FIB, which is lucky.
600 		 */
601 		m_freem(m->m_next);
602 		m->m_next = NULL;
603 		m->m_data = (caddr_t)ipgen;
604 		/* m_len is set later */
605 		tlen = 0;
606 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
607 #ifdef INET6
608 		if (isipv6) {
609 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
610 			nth = (struct tcphdr *)(ip6 + 1);
611 		} else
612 #endif /* INET6 */
613 		{
614 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
615 			nth = (struct tcphdr *)(ip + 1);
616 		}
617 		if (th != nth) {
618 			/*
619 			 * this is usually a case when an extension header
620 			 * exists between the IPv6 header and the
621 			 * TCP header.
622 			 */
623 			nth->th_sport = th->th_sport;
624 			nth->th_dport = th->th_dport;
625 		}
626 		xchg(nth->th_dport, nth->th_sport, uint16_t);
627 #undef xchg
628 	}
629 #ifdef INET6
630 	if (isipv6) {
631 		ip6->ip6_flow = 0;
632 		ip6->ip6_vfc = IPV6_VERSION;
633 		ip6->ip6_nxt = IPPROTO_TCP;
634 		ip6->ip6_plen = 0;		/* Set in ip6_output(). */
635 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
636 	}
637 #endif
638 #if defined(INET) && defined(INET6)
639 	else
640 #endif
641 #ifdef INET
642 	{
643 		tlen += sizeof (struct tcpiphdr);
644 		ip->ip_len = htons(tlen);
645 		ip->ip_ttl = V_ip_defttl;
646 		if (V_path_mtu_discovery)
647 			ip->ip_off |= htons(IP_DF);
648 	}
649 #endif
650 	m->m_len = tlen;
651 	m->m_pkthdr.len = tlen;
652 	m->m_pkthdr.rcvif = NULL;
653 #ifdef MAC
654 	if (inp != NULL) {
655 		/*
656 		 * Packet is associated with a socket, so allow the
657 		 * label of the response to reflect the socket label.
658 		 */
659 		INP_WLOCK_ASSERT(inp);
660 		mac_inpcb_create_mbuf(inp, m);
661 	} else {
662 		/*
663 		 * Packet is not associated with a socket, so possibly
664 		 * update the label in place.
665 		 */
666 		mac_netinet_tcp_reply(m);
667 	}
668 #endif
669 	nth->th_seq = htonl(seq);
670 	nth->th_ack = htonl(ack);
671 	nth->th_x2 = 0;
672 	nth->th_off = sizeof (struct tcphdr) >> 2;
673 	nth->th_flags = flags;
674 	if (tp != NULL)
675 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
676 	else
677 		nth->th_win = htons((u_short)win);
678 	nth->th_urp = 0;
679 
680 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
681 #ifdef INET6
682 	if (isipv6) {
683 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
684 		nth->th_sum = in6_cksum_pseudo(ip6,
685 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
686 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
687 		    NULL, NULL);
688 	}
689 #endif /* INET6 */
690 #if defined(INET6) && defined(INET)
691 	else
692 #endif
693 #ifdef INET
694 	{
695 		m->m_pkthdr.csum_flags = CSUM_TCP;
696 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
697 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
698 	}
699 #endif /* INET */
700 #ifdef TCPDEBUG
701 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
702 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
703 #endif
704 #ifdef INET6
705 	if (isipv6)
706 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
707 #endif /* INET6 */
708 #if defined(INET) && defined(INET6)
709 	else
710 #endif
711 #ifdef INET
712 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
713 #endif
714 }
715 
716 /*
717  * Create a new TCP control block, making an
718  * empty reassembly queue and hooking it to the argument
719  * protocol control block.  The `inp' parameter must have
720  * come from the zone allocator set up in tcp_init().
721  */
722 struct tcpcb *
723 tcp_newtcpcb(struct inpcb *inp)
724 {
725 	struct tcpcb_mem *tm;
726 	struct tcpcb *tp;
727 #ifdef INET6
728 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
729 #endif /* INET6 */
730 
731 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
732 	if (tm == NULL)
733 		return (NULL);
734 	tp = &tm->tcb;
735 
736 	/* Initialise cc_var struct for this tcpcb. */
737 	tp->ccv = &tm->ccv;
738 	tp->ccv->type = IPPROTO_TCP;
739 	tp->ccv->ccvc.tcp = tp;
740 
741 	/*
742 	 * Use the current system default CC algorithm.
743 	 */
744 	CC_LIST_RLOCK();
745 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
746 	CC_ALGO(tp) = CC_DEFAULT();
747 	CC_LIST_RUNLOCK();
748 
749 	if (CC_ALGO(tp)->cb_init != NULL)
750 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
751 			uma_zfree(V_tcpcb_zone, tm);
752 			return (NULL);
753 		}
754 
755 	tp->osd = &tm->osd;
756 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
757 		uma_zfree(V_tcpcb_zone, tm);
758 		return (NULL);
759 	}
760 
761 #ifdef VIMAGE
762 	tp->t_vnet = inp->inp_vnet;
763 #endif
764 	tp->t_timers = &tm->tt;
765 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
766 	tp->t_maxseg = tp->t_maxopd =
767 #ifdef INET6
768 		isipv6 ? V_tcp_v6mssdflt :
769 #endif /* INET6 */
770 		V_tcp_mssdflt;
771 
772 	/* Set up our timeouts. */
773 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
774 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
775 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
776 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
777 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
778 
779 	if (V_tcp_do_rfc1323)
780 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
781 	if (V_tcp_do_sack)
782 		tp->t_flags |= TF_SACK_PERMIT;
783 	TAILQ_INIT(&tp->snd_holes);
784 	tp->t_inpcb = inp;	/* XXX */
785 	/*
786 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
787 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
788 	 * reasonable initial retransmit time.
789 	 */
790 	tp->t_srtt = TCPTV_SRTTBASE;
791 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
792 	tp->t_rttmin = tcp_rexmit_min;
793 	tp->t_rxtcur = TCPTV_RTOBASE;
794 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
795 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
796 	tp->t_rcvtime = ticks;
797 	/*
798 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
799 	 * because the socket may be bound to an IPv6 wildcard address,
800 	 * which may match an IPv4-mapped IPv6 address.
801 	 */
802 	inp->inp_ip_ttl = V_ip_defttl;
803 	inp->inp_ppcb = tp;
804 	return (tp);		/* XXX */
805 }
806 
807 /*
808  * Switch the congestion control algorithm back to NewReno for any active
809  * control blocks using an algorithm which is about to go away.
810  * This ensures the CC framework can allow the unload to proceed without leaving
811  * any dangling pointers which would trigger a panic.
812  * Returning non-zero would inform the CC framework that something went wrong
813  * and it would be unsafe to allow the unload to proceed. However, there is no
814  * way for this to occur with this implementation so we always return zero.
815  */
816 int
817 tcp_ccalgounload(struct cc_algo *unload_algo)
818 {
819 	struct cc_algo *tmpalgo;
820 	struct inpcb *inp;
821 	struct tcpcb *tp;
822 	VNET_ITERATOR_DECL(vnet_iter);
823 
824 	/*
825 	 * Check all active control blocks across all network stacks and change
826 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
827 	 * requires cleanup code to be run, call it.
828 	 */
829 	VNET_LIST_RLOCK();
830 	VNET_FOREACH(vnet_iter) {
831 		CURVNET_SET(vnet_iter);
832 		INP_INFO_RLOCK(&V_tcbinfo);
833 		/*
834 		 * New connections already part way through being initialised
835 		 * with the CC algo we're removing will not race with this code
836 		 * because the INP_INFO_WLOCK is held during initialisation. We
837 		 * therefore don't enter the loop below until the connection
838 		 * list has stabilised.
839 		 */
840 		LIST_FOREACH(inp, &V_tcb, inp_list) {
841 			INP_WLOCK(inp);
842 			/* Important to skip tcptw structs. */
843 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
844 			    (tp = intotcpcb(inp)) != NULL) {
845 				/*
846 				 * By holding INP_WLOCK here, we are assured
847 				 * that the connection is not currently
848 				 * executing inside the CC module's functions
849 				 * i.e. it is safe to make the switch back to
850 				 * NewReno.
851 				 */
852 				if (CC_ALGO(tp) == unload_algo) {
853 					tmpalgo = CC_ALGO(tp);
854 					/* NewReno does not require any init. */
855 					CC_ALGO(tp) = &newreno_cc_algo;
856 					if (tmpalgo->cb_destroy != NULL)
857 						tmpalgo->cb_destroy(tp->ccv);
858 				}
859 			}
860 			INP_WUNLOCK(inp);
861 		}
862 		INP_INFO_RUNLOCK(&V_tcbinfo);
863 		CURVNET_RESTORE();
864 	}
865 	VNET_LIST_RUNLOCK();
866 
867 	return (0);
868 }
869 
870 /*
871  * Drop a TCP connection, reporting
872  * the specified error.  If connection is synchronized,
873  * then send a RST to peer.
874  */
875 struct tcpcb *
876 tcp_drop(struct tcpcb *tp, int errno)
877 {
878 	struct socket *so = tp->t_inpcb->inp_socket;
879 
880 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
881 	INP_WLOCK_ASSERT(tp->t_inpcb);
882 
883 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
884 		tp->t_state = TCPS_CLOSED;
885 		(void) tcp_output(tp);
886 		TCPSTAT_INC(tcps_drops);
887 	} else
888 		TCPSTAT_INC(tcps_conndrops);
889 	if (errno == ETIMEDOUT && tp->t_softerror)
890 		errno = tp->t_softerror;
891 	so->so_error = errno;
892 	return (tcp_close(tp));
893 }
894 
895 void
896 tcp_discardcb(struct tcpcb *tp)
897 {
898 	struct inpcb *inp = tp->t_inpcb;
899 	struct socket *so = inp->inp_socket;
900 #ifdef INET6
901 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
902 #endif /* INET6 */
903 
904 	INP_WLOCK_ASSERT(inp);
905 
906 	/*
907 	 * Make sure that all of our timers are stopped before we delete the
908 	 * PCB.
909 	 *
910 	 * XXXRW: Really, we would like to use callout_drain() here in order
911 	 * to avoid races experienced in tcp_timer.c where a timer is already
912 	 * executing at this point.  However, we can't, both because we're
913 	 * running in a context where we can't sleep, and also because we
914 	 * hold locks required by the timers.  What we instead need to do is
915 	 * test to see if callout_drain() is required, and if so, defer some
916 	 * portion of the remainder of tcp_discardcb() to an asynchronous
917 	 * context that can callout_drain() and then continue.  Some care
918 	 * will be required to ensure that no further processing takes place
919 	 * on the tcpcb, even though it hasn't been freed (a flag?).
920 	 */
921 	callout_stop(&tp->t_timers->tt_rexmt);
922 	callout_stop(&tp->t_timers->tt_persist);
923 	callout_stop(&tp->t_timers->tt_keep);
924 	callout_stop(&tp->t_timers->tt_2msl);
925 	callout_stop(&tp->t_timers->tt_delack);
926 
927 	/*
928 	 * If we got enough samples through the srtt filter,
929 	 * save the rtt and rttvar in the routing entry.
930 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
931 	 * 4 samples is enough for the srtt filter to converge
932 	 * to within enough % of the correct value; fewer samples
933 	 * and we could save a bogus rtt. The danger is not high
934 	 * as tcp quickly recovers from everything.
935 	 * XXX: Works very well but needs some more statistics!
936 	 */
937 	if (tp->t_rttupdated >= 4) {
938 		struct hc_metrics_lite metrics;
939 		u_long ssthresh;
940 
941 		bzero(&metrics, sizeof(metrics));
942 		/*
943 		 * Update the ssthresh always when the conditions below
944 		 * are satisfied. This gives us better new start value
945 		 * for the congestion avoidance for new connections.
946 		 * ssthresh is only set if packet loss occured on a session.
947 		 *
948 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
949 		 * being torn down.  Ideally this code would not use 'so'.
950 		 */
951 		ssthresh = tp->snd_ssthresh;
952 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
953 			/*
954 			 * convert the limit from user data bytes to
955 			 * packets then to packet data bytes.
956 			 */
957 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
958 			if (ssthresh < 2)
959 				ssthresh = 2;
960 			ssthresh *= (u_long)(tp->t_maxseg +
961 #ifdef INET6
962 			    (isipv6 ? sizeof (struct ip6_hdr) +
963 				sizeof (struct tcphdr) :
964 #endif
965 				sizeof (struct tcpiphdr)
966 #ifdef INET6
967 			    )
968 #endif
969 			    );
970 		} else
971 			ssthresh = 0;
972 		metrics.rmx_ssthresh = ssthresh;
973 
974 		metrics.rmx_rtt = tp->t_srtt;
975 		metrics.rmx_rttvar = tp->t_rttvar;
976 		metrics.rmx_cwnd = tp->snd_cwnd;
977 		metrics.rmx_sendpipe = 0;
978 		metrics.rmx_recvpipe = 0;
979 
980 		tcp_hc_update(&inp->inp_inc, &metrics);
981 	}
982 
983 	/* free the reassembly queue, if any */
984 	tcp_reass_flush(tp);
985 
986 #ifdef TCP_OFFLOAD
987 	/* Disconnect offload device, if any. */
988 	if (tp->t_flags & TF_TOE)
989 		tcp_offload_detach(tp);
990 #endif
991 
992 	tcp_free_sackholes(tp);
993 
994 	/* Allow the CC algorithm to clean up after itself. */
995 	if (CC_ALGO(tp)->cb_destroy != NULL)
996 		CC_ALGO(tp)->cb_destroy(tp->ccv);
997 
998 	khelp_destroy_osd(tp->osd);
999 
1000 	CC_ALGO(tp) = NULL;
1001 	inp->inp_ppcb = NULL;
1002 	tp->t_inpcb = NULL;
1003 	uma_zfree(V_tcpcb_zone, tp);
1004 }
1005 
1006 /*
1007  * Attempt to close a TCP control block, marking it as dropped, and freeing
1008  * the socket if we hold the only reference.
1009  */
1010 struct tcpcb *
1011 tcp_close(struct tcpcb *tp)
1012 {
1013 	struct inpcb *inp = tp->t_inpcb;
1014 	struct socket *so;
1015 
1016 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1017 	INP_WLOCK_ASSERT(inp);
1018 
1019 #ifdef TCP_OFFLOAD
1020 	if (tp->t_state == TCPS_LISTEN)
1021 		tcp_offload_listen_stop(tp);
1022 #endif
1023 	in_pcbdrop(inp);
1024 	TCPSTAT_INC(tcps_closed);
1025 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1026 	so = inp->inp_socket;
1027 	soisdisconnected(so);
1028 	if (inp->inp_flags & INP_SOCKREF) {
1029 		KASSERT(so->so_state & SS_PROTOREF,
1030 		    ("tcp_close: !SS_PROTOREF"));
1031 		inp->inp_flags &= ~INP_SOCKREF;
1032 		INP_WUNLOCK(inp);
1033 		ACCEPT_LOCK();
1034 		SOCK_LOCK(so);
1035 		so->so_state &= ~SS_PROTOREF;
1036 		sofree(so);
1037 		return (NULL);
1038 	}
1039 	return (tp);
1040 }
1041 
1042 void
1043 tcp_drain(void)
1044 {
1045 	VNET_ITERATOR_DECL(vnet_iter);
1046 
1047 	if (!do_tcpdrain)
1048 		return;
1049 
1050 	VNET_LIST_RLOCK_NOSLEEP();
1051 	VNET_FOREACH(vnet_iter) {
1052 		CURVNET_SET(vnet_iter);
1053 		struct inpcb *inpb;
1054 		struct tcpcb *tcpb;
1055 
1056 	/*
1057 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1058 	 * if there is one...
1059 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1060 	 *      reassembly queue should be flushed, but in a situation
1061 	 *	where we're really low on mbufs, this is potentially
1062 	 *	usefull.
1063 	 */
1064 		INP_INFO_RLOCK(&V_tcbinfo);
1065 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1066 			if (inpb->inp_flags & INP_TIMEWAIT)
1067 				continue;
1068 			INP_WLOCK(inpb);
1069 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1070 				tcp_reass_flush(tcpb);
1071 				tcp_clean_sackreport(tcpb);
1072 			}
1073 			INP_WUNLOCK(inpb);
1074 		}
1075 		INP_INFO_RUNLOCK(&V_tcbinfo);
1076 		CURVNET_RESTORE();
1077 	}
1078 	VNET_LIST_RUNLOCK_NOSLEEP();
1079 }
1080 
1081 /*
1082  * Notify a tcp user of an asynchronous error;
1083  * store error as soft error, but wake up user
1084  * (for now, won't do anything until can select for soft error).
1085  *
1086  * Do not wake up user since there currently is no mechanism for
1087  * reporting soft errors (yet - a kqueue filter may be added).
1088  */
1089 static struct inpcb *
1090 tcp_notify(struct inpcb *inp, int error)
1091 {
1092 	struct tcpcb *tp;
1093 
1094 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1095 	INP_WLOCK_ASSERT(inp);
1096 
1097 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1098 	    (inp->inp_flags & INP_DROPPED))
1099 		return (inp);
1100 
1101 	tp = intotcpcb(inp);
1102 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1103 
1104 	/*
1105 	 * Ignore some errors if we are hooked up.
1106 	 * If connection hasn't completed, has retransmitted several times,
1107 	 * and receives a second error, give up now.  This is better
1108 	 * than waiting a long time to establish a connection that
1109 	 * can never complete.
1110 	 */
1111 	if (tp->t_state == TCPS_ESTABLISHED &&
1112 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1113 	     error == EHOSTDOWN)) {
1114 		return (inp);
1115 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1116 	    tp->t_softerror) {
1117 		tp = tcp_drop(tp, error);
1118 		if (tp != NULL)
1119 			return (inp);
1120 		else
1121 			return (NULL);
1122 	} else {
1123 		tp->t_softerror = error;
1124 		return (inp);
1125 	}
1126 #if 0
1127 	wakeup( &so->so_timeo);
1128 	sorwakeup(so);
1129 	sowwakeup(so);
1130 #endif
1131 }
1132 
1133 static int
1134 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1135 {
1136 	int error, i, m, n, pcb_count;
1137 	struct inpcb *inp, **inp_list;
1138 	inp_gen_t gencnt;
1139 	struct xinpgen xig;
1140 
1141 	/*
1142 	 * The process of preparing the TCB list is too time-consuming and
1143 	 * resource-intensive to repeat twice on every request.
1144 	 */
1145 	if (req->oldptr == NULL) {
1146 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1147 		n += imax(n / 8, 10);
1148 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1149 		return (0);
1150 	}
1151 
1152 	if (req->newptr != NULL)
1153 		return (EPERM);
1154 
1155 	/*
1156 	 * OK, now we're committed to doing something.
1157 	 */
1158 	INP_INFO_RLOCK(&V_tcbinfo);
1159 	gencnt = V_tcbinfo.ipi_gencnt;
1160 	n = V_tcbinfo.ipi_count;
1161 	INP_INFO_RUNLOCK(&V_tcbinfo);
1162 
1163 	m = syncache_pcbcount();
1164 
1165 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1166 		+ (n + m) * sizeof(struct xtcpcb));
1167 	if (error != 0)
1168 		return (error);
1169 
1170 	xig.xig_len = sizeof xig;
1171 	xig.xig_count = n + m;
1172 	xig.xig_gen = gencnt;
1173 	xig.xig_sogen = so_gencnt;
1174 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1175 	if (error)
1176 		return (error);
1177 
1178 	error = syncache_pcblist(req, m, &pcb_count);
1179 	if (error)
1180 		return (error);
1181 
1182 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1183 	if (inp_list == NULL)
1184 		return (ENOMEM);
1185 
1186 	INP_INFO_RLOCK(&V_tcbinfo);
1187 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1188 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1189 		INP_WLOCK(inp);
1190 		if (inp->inp_gencnt <= gencnt) {
1191 			/*
1192 			 * XXX: This use of cr_cansee(), introduced with
1193 			 * TCP state changes, is not quite right, but for
1194 			 * now, better than nothing.
1195 			 */
1196 			if (inp->inp_flags & INP_TIMEWAIT) {
1197 				if (intotw(inp) != NULL)
1198 					error = cr_cansee(req->td->td_ucred,
1199 					    intotw(inp)->tw_cred);
1200 				else
1201 					error = EINVAL;	/* Skip this inp. */
1202 			} else
1203 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1204 			if (error == 0) {
1205 				in_pcbref(inp);
1206 				inp_list[i++] = inp;
1207 			}
1208 		}
1209 		INP_WUNLOCK(inp);
1210 	}
1211 	INP_INFO_RUNLOCK(&V_tcbinfo);
1212 	n = i;
1213 
1214 	error = 0;
1215 	for (i = 0; i < n; i++) {
1216 		inp = inp_list[i];
1217 		INP_RLOCK(inp);
1218 		if (inp->inp_gencnt <= gencnt) {
1219 			struct xtcpcb xt;
1220 			void *inp_ppcb;
1221 
1222 			bzero(&xt, sizeof(xt));
1223 			xt.xt_len = sizeof xt;
1224 			/* XXX should avoid extra copy */
1225 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1226 			inp_ppcb = inp->inp_ppcb;
1227 			if (inp_ppcb == NULL)
1228 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1229 			else if (inp->inp_flags & INP_TIMEWAIT) {
1230 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1231 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1232 			} else {
1233 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1234 				if (xt.xt_tp.t_timers)
1235 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1236 			}
1237 			if (inp->inp_socket != NULL)
1238 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1239 			else {
1240 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1241 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1242 			}
1243 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1244 			INP_RUNLOCK(inp);
1245 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1246 		} else
1247 			INP_RUNLOCK(inp);
1248 	}
1249 	INP_INFO_WLOCK(&V_tcbinfo);
1250 	for (i = 0; i < n; i++) {
1251 		inp = inp_list[i];
1252 		INP_RLOCK(inp);
1253 		if (!in_pcbrele_rlocked(inp))
1254 			INP_RUNLOCK(inp);
1255 	}
1256 	INP_INFO_WUNLOCK(&V_tcbinfo);
1257 
1258 	if (!error) {
1259 		/*
1260 		 * Give the user an updated idea of our state.
1261 		 * If the generation differs from what we told
1262 		 * her before, she knows that something happened
1263 		 * while we were processing this request, and it
1264 		 * might be necessary to retry.
1265 		 */
1266 		INP_INFO_RLOCK(&V_tcbinfo);
1267 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1268 		xig.xig_sogen = so_gencnt;
1269 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1270 		INP_INFO_RUNLOCK(&V_tcbinfo);
1271 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1272 	}
1273 	free(inp_list, M_TEMP);
1274 	return (error);
1275 }
1276 
1277 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1278     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1279     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1280 
1281 #ifdef INET
1282 static int
1283 tcp_getcred(SYSCTL_HANDLER_ARGS)
1284 {
1285 	struct xucred xuc;
1286 	struct sockaddr_in addrs[2];
1287 	struct inpcb *inp;
1288 	int error;
1289 
1290 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1291 	if (error)
1292 		return (error);
1293 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1294 	if (error)
1295 		return (error);
1296 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1297 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1298 	if (inp != NULL) {
1299 		if (inp->inp_socket == NULL)
1300 			error = ENOENT;
1301 		if (error == 0)
1302 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1303 		if (error == 0)
1304 			cru2x(inp->inp_cred, &xuc);
1305 		INP_RUNLOCK(inp);
1306 	} else
1307 		error = ENOENT;
1308 	if (error == 0)
1309 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1310 	return (error);
1311 }
1312 
1313 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1314     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1315     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1316 #endif /* INET */
1317 
1318 #ifdef INET6
1319 static int
1320 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1321 {
1322 	struct xucred xuc;
1323 	struct sockaddr_in6 addrs[2];
1324 	struct inpcb *inp;
1325 	int error;
1326 #ifdef INET
1327 	int mapped = 0;
1328 #endif
1329 
1330 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1331 	if (error)
1332 		return (error);
1333 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1334 	if (error)
1335 		return (error);
1336 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1337 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1338 		return (error);
1339 	}
1340 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1341 #ifdef INET
1342 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1343 			mapped = 1;
1344 		else
1345 #endif
1346 			return (EINVAL);
1347 	}
1348 
1349 #ifdef INET
1350 	if (mapped == 1)
1351 		inp = in_pcblookup(&V_tcbinfo,
1352 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1353 			addrs[1].sin6_port,
1354 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1355 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1356 	else
1357 #endif
1358 		inp = in6_pcblookup(&V_tcbinfo,
1359 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1360 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1361 			INPLOOKUP_RLOCKPCB, NULL);
1362 	if (inp != NULL) {
1363 		if (inp->inp_socket == NULL)
1364 			error = ENOENT;
1365 		if (error == 0)
1366 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1367 		if (error == 0)
1368 			cru2x(inp->inp_cred, &xuc);
1369 		INP_RUNLOCK(inp);
1370 	} else
1371 		error = ENOENT;
1372 	if (error == 0)
1373 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1374 	return (error);
1375 }
1376 
1377 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1378     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1379     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1380 #endif /* INET6 */
1381 
1382 
1383 #ifdef INET
1384 void
1385 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1386 {
1387 	struct ip *ip = vip;
1388 	struct tcphdr *th;
1389 	struct in_addr faddr;
1390 	struct inpcb *inp;
1391 	struct tcpcb *tp;
1392 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1393 	struct icmp *icp;
1394 	struct in_conninfo inc;
1395 	tcp_seq icmp_tcp_seq;
1396 	int mtu;
1397 
1398 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1399 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1400 		return;
1401 
1402 	if (cmd == PRC_MSGSIZE)
1403 		notify = tcp_mtudisc_notify;
1404 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1405 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1406 		notify = tcp_drop_syn_sent;
1407 	/*
1408 	 * Redirects don't need to be handled up here.
1409 	 */
1410 	else if (PRC_IS_REDIRECT(cmd))
1411 		return;
1412 	/*
1413 	 * Source quench is depreciated.
1414 	 */
1415 	else if (cmd == PRC_QUENCH)
1416 		return;
1417 	/*
1418 	 * Hostdead is ugly because it goes linearly through all PCBs.
1419 	 * XXX: We never get this from ICMP, otherwise it makes an
1420 	 * excellent DoS attack on machines with many connections.
1421 	 */
1422 	else if (cmd == PRC_HOSTDEAD)
1423 		ip = NULL;
1424 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1425 		return;
1426 	if (ip != NULL) {
1427 		icp = (struct icmp *)((caddr_t)ip
1428 				      - offsetof(struct icmp, icmp_ip));
1429 		th = (struct tcphdr *)((caddr_t)ip
1430 				       + (ip->ip_hl << 2));
1431 		INP_INFO_WLOCK(&V_tcbinfo);
1432 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1433 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1434 		if (inp != NULL)  {
1435 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1436 			    !(inp->inp_flags & INP_DROPPED) &&
1437 			    !(inp->inp_socket == NULL)) {
1438 				icmp_tcp_seq = htonl(th->th_seq);
1439 				tp = intotcpcb(inp);
1440 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1441 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1442 					if (cmd == PRC_MSGSIZE) {
1443 					    /*
1444 					     * MTU discovery:
1445 					     * If we got a needfrag set the MTU
1446 					     * in the route to the suggested new
1447 					     * value (if given) and then notify.
1448 					     */
1449 					    bzero(&inc, sizeof(inc));
1450 					    inc.inc_faddr = faddr;
1451 					    inc.inc_fibnum =
1452 						inp->inp_inc.inc_fibnum;
1453 
1454 					    mtu = ntohs(icp->icmp_nextmtu);
1455 					    /*
1456 					     * If no alternative MTU was
1457 					     * proposed, try the next smaller
1458 					     * one.
1459 					     */
1460 					    if (!mtu)
1461 						mtu = ip_next_mtu(
1462 						 ntohs(ip->ip_len), 1);
1463 					    if (mtu < V_tcp_minmss
1464 						 + sizeof(struct tcpiphdr))
1465 						mtu = V_tcp_minmss
1466 						 + sizeof(struct tcpiphdr);
1467 					    /*
1468 					     * Only cache the MTU if it
1469 					     * is smaller than the interface
1470 					     * or route MTU.  tcp_mtudisc()
1471 					     * will do right thing by itself.
1472 					     */
1473 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1474 						tcp_hc_updatemtu(&inc, mtu);
1475 					    tcp_mtudisc(inp, mtu);
1476 					} else
1477 						inp = (*notify)(inp,
1478 						    inetctlerrmap[cmd]);
1479 				}
1480 			}
1481 			if (inp != NULL)
1482 				INP_WUNLOCK(inp);
1483 		} else {
1484 			bzero(&inc, sizeof(inc));
1485 			inc.inc_fport = th->th_dport;
1486 			inc.inc_lport = th->th_sport;
1487 			inc.inc_faddr = faddr;
1488 			inc.inc_laddr = ip->ip_src;
1489 			syncache_unreach(&inc, th);
1490 		}
1491 		INP_INFO_WUNLOCK(&V_tcbinfo);
1492 	} else
1493 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1494 }
1495 #endif /* INET */
1496 
1497 #ifdef INET6
1498 void
1499 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1500 {
1501 	struct tcphdr th;
1502 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1503 	struct ip6_hdr *ip6;
1504 	struct mbuf *m;
1505 	struct ip6ctlparam *ip6cp = NULL;
1506 	const struct sockaddr_in6 *sa6_src = NULL;
1507 	int off;
1508 	struct tcp_portonly {
1509 		u_int16_t th_sport;
1510 		u_int16_t th_dport;
1511 	} *thp;
1512 
1513 	if (sa->sa_family != AF_INET6 ||
1514 	    sa->sa_len != sizeof(struct sockaddr_in6))
1515 		return;
1516 
1517 	if (cmd == PRC_MSGSIZE)
1518 		notify = tcp_mtudisc_notify;
1519 	else if (!PRC_IS_REDIRECT(cmd) &&
1520 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1521 		return;
1522 	/* Source quench is depreciated. */
1523 	else if (cmd == PRC_QUENCH)
1524 		return;
1525 
1526 	/* if the parameter is from icmp6, decode it. */
1527 	if (d != NULL) {
1528 		ip6cp = (struct ip6ctlparam *)d;
1529 		m = ip6cp->ip6c_m;
1530 		ip6 = ip6cp->ip6c_ip6;
1531 		off = ip6cp->ip6c_off;
1532 		sa6_src = ip6cp->ip6c_src;
1533 	} else {
1534 		m = NULL;
1535 		ip6 = NULL;
1536 		off = 0;	/* fool gcc */
1537 		sa6_src = &sa6_any;
1538 	}
1539 
1540 	if (ip6 != NULL) {
1541 		struct in_conninfo inc;
1542 		/*
1543 		 * XXX: We assume that when IPV6 is non NULL,
1544 		 * M and OFF are valid.
1545 		 */
1546 
1547 		/* check if we can safely examine src and dst ports */
1548 		if (m->m_pkthdr.len < off + sizeof(*thp))
1549 			return;
1550 
1551 		bzero(&th, sizeof(th));
1552 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1553 
1554 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1555 		    (struct sockaddr *)ip6cp->ip6c_src,
1556 		    th.th_sport, cmd, NULL, notify);
1557 
1558 		bzero(&inc, sizeof(inc));
1559 		inc.inc_fport = th.th_dport;
1560 		inc.inc_lport = th.th_sport;
1561 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1562 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1563 		inc.inc_flags |= INC_ISIPV6;
1564 		INP_INFO_WLOCK(&V_tcbinfo);
1565 		syncache_unreach(&inc, &th);
1566 		INP_INFO_WUNLOCK(&V_tcbinfo);
1567 	} else
1568 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1569 			      0, cmd, NULL, notify);
1570 }
1571 #endif /* INET6 */
1572 
1573 
1574 /*
1575  * Following is where TCP initial sequence number generation occurs.
1576  *
1577  * There are two places where we must use initial sequence numbers:
1578  * 1.  In SYN-ACK packets.
1579  * 2.  In SYN packets.
1580  *
1581  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1582  * tcp_syncache.c for details.
1583  *
1584  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1585  * depends on this property.  In addition, these ISNs should be
1586  * unguessable so as to prevent connection hijacking.  To satisfy
1587  * the requirements of this situation, the algorithm outlined in
1588  * RFC 1948 is used, with only small modifications.
1589  *
1590  * Implementation details:
1591  *
1592  * Time is based off the system timer, and is corrected so that it
1593  * increases by one megabyte per second.  This allows for proper
1594  * recycling on high speed LANs while still leaving over an hour
1595  * before rollover.
1596  *
1597  * As reading the *exact* system time is too expensive to be done
1598  * whenever setting up a TCP connection, we increment the time
1599  * offset in two ways.  First, a small random positive increment
1600  * is added to isn_offset for each connection that is set up.
1601  * Second, the function tcp_isn_tick fires once per clock tick
1602  * and increments isn_offset as necessary so that sequence numbers
1603  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1604  * random positive increments serve only to ensure that the same
1605  * exact sequence number is never sent out twice (as could otherwise
1606  * happen when a port is recycled in less than the system tick
1607  * interval.)
1608  *
1609  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1610  * between seeding of isn_secret.  This is normally set to zero,
1611  * as reseeding should not be necessary.
1612  *
1613  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1614  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1615  * general, this means holding an exclusive (write) lock.
1616  */
1617 
1618 #define ISN_BYTES_PER_SECOND 1048576
1619 #define ISN_STATIC_INCREMENT 4096
1620 #define ISN_RANDOM_INCREMENT (4096 - 1)
1621 
1622 static VNET_DEFINE(u_char, isn_secret[32]);
1623 static VNET_DEFINE(int, isn_last);
1624 static VNET_DEFINE(int, isn_last_reseed);
1625 static VNET_DEFINE(u_int32_t, isn_offset);
1626 static VNET_DEFINE(u_int32_t, isn_offset_old);
1627 
1628 #define	V_isn_secret			VNET(isn_secret)
1629 #define	V_isn_last			VNET(isn_last)
1630 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1631 #define	V_isn_offset			VNET(isn_offset)
1632 #define	V_isn_offset_old		VNET(isn_offset_old)
1633 
1634 tcp_seq
1635 tcp_new_isn(struct tcpcb *tp)
1636 {
1637 	MD5_CTX isn_ctx;
1638 	u_int32_t md5_buffer[4];
1639 	tcp_seq new_isn;
1640 	u_int32_t projected_offset;
1641 
1642 	INP_WLOCK_ASSERT(tp->t_inpcb);
1643 
1644 	ISN_LOCK();
1645 	/* Seed if this is the first use, reseed if requested. */
1646 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1647 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1648 		< (u_int)ticks))) {
1649 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1650 		V_isn_last_reseed = ticks;
1651 	}
1652 
1653 	/* Compute the md5 hash and return the ISN. */
1654 	MD5Init(&isn_ctx);
1655 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1656 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1657 #ifdef INET6
1658 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1659 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1660 			  sizeof(struct in6_addr));
1661 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1662 			  sizeof(struct in6_addr));
1663 	} else
1664 #endif
1665 	{
1666 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1667 			  sizeof(struct in_addr));
1668 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1669 			  sizeof(struct in_addr));
1670 	}
1671 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1672 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1673 	new_isn = (tcp_seq) md5_buffer[0];
1674 	V_isn_offset += ISN_STATIC_INCREMENT +
1675 		(arc4random() & ISN_RANDOM_INCREMENT);
1676 	if (ticks != V_isn_last) {
1677 		projected_offset = V_isn_offset_old +
1678 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1679 		if (SEQ_GT(projected_offset, V_isn_offset))
1680 			V_isn_offset = projected_offset;
1681 		V_isn_offset_old = V_isn_offset;
1682 		V_isn_last = ticks;
1683 	}
1684 	new_isn += V_isn_offset;
1685 	ISN_UNLOCK();
1686 	return (new_isn);
1687 }
1688 
1689 /*
1690  * When a specific ICMP unreachable message is received and the
1691  * connection state is SYN-SENT, drop the connection.  This behavior
1692  * is controlled by the icmp_may_rst sysctl.
1693  */
1694 struct inpcb *
1695 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1696 {
1697 	struct tcpcb *tp;
1698 
1699 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1700 	INP_WLOCK_ASSERT(inp);
1701 
1702 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1703 	    (inp->inp_flags & INP_DROPPED))
1704 		return (inp);
1705 
1706 	tp = intotcpcb(inp);
1707 	if (tp->t_state != TCPS_SYN_SENT)
1708 		return (inp);
1709 
1710 	tp = tcp_drop(tp, errno);
1711 	if (tp != NULL)
1712 		return (inp);
1713 	else
1714 		return (NULL);
1715 }
1716 
1717 /*
1718  * When `need fragmentation' ICMP is received, update our idea of the MSS
1719  * based on the new value. Also nudge TCP to send something, since we
1720  * know the packet we just sent was dropped.
1721  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1722  */
1723 static struct inpcb *
1724 tcp_mtudisc_notify(struct inpcb *inp, int error)
1725 {
1726 
1727 	return (tcp_mtudisc(inp, -1));
1728 }
1729 
1730 struct inpcb *
1731 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1732 {
1733 	struct tcpcb *tp;
1734 	struct socket *so;
1735 
1736 	INP_WLOCK_ASSERT(inp);
1737 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1738 	    (inp->inp_flags & INP_DROPPED))
1739 		return (inp);
1740 
1741 	tp = intotcpcb(inp);
1742 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1743 
1744 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1745 
1746 	so = inp->inp_socket;
1747 	SOCKBUF_LOCK(&so->so_snd);
1748 	/* If the mss is larger than the socket buffer, decrease the mss. */
1749 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1750 		tp->t_maxseg = so->so_snd.sb_hiwat;
1751 	SOCKBUF_UNLOCK(&so->so_snd);
1752 
1753 	TCPSTAT_INC(tcps_mturesent);
1754 	tp->t_rtttime = 0;
1755 	tp->snd_nxt = tp->snd_una;
1756 	tcp_free_sackholes(tp);
1757 	tp->snd_recover = tp->snd_max;
1758 	if (tp->t_flags & TF_SACK_PERMIT)
1759 		EXIT_FASTRECOVERY(tp->t_flags);
1760 	tcp_output(tp);
1761 	return (inp);
1762 }
1763 
1764 #ifdef INET
1765 /*
1766  * Look-up the routing entry to the peer of this inpcb.  If no route
1767  * is found and it cannot be allocated, then return 0.  This routine
1768  * is called by TCP routines that access the rmx structure and by
1769  * tcp_mss_update to get the peer/interface MTU.
1770  */
1771 u_long
1772 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1773 {
1774 	struct route sro;
1775 	struct sockaddr_in *dst;
1776 	struct ifnet *ifp;
1777 	u_long maxmtu = 0;
1778 
1779 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1780 
1781 	bzero(&sro, sizeof(sro));
1782 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1783 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1784 		dst->sin_family = AF_INET;
1785 		dst->sin_len = sizeof(*dst);
1786 		dst->sin_addr = inc->inc_faddr;
1787 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1788 	}
1789 	if (sro.ro_rt != NULL) {
1790 		ifp = sro.ro_rt->rt_ifp;
1791 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1792 			maxmtu = ifp->if_mtu;
1793 		else
1794 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1795 
1796 		/* Report additional interface capabilities. */
1797 		if (flags != NULL) {
1798 			if (ifp->if_capenable & IFCAP_TSO4 &&
1799 			    ifp->if_hwassist & CSUM_TSO)
1800 				*flags |= CSUM_TSO;
1801 		}
1802 		RTFREE(sro.ro_rt);
1803 	}
1804 	return (maxmtu);
1805 }
1806 #endif /* INET */
1807 
1808 #ifdef INET6
1809 u_long
1810 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1811 {
1812 	struct route_in6 sro6;
1813 	struct ifnet *ifp;
1814 	u_long maxmtu = 0;
1815 
1816 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1817 
1818 	bzero(&sro6, sizeof(sro6));
1819 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1820 		sro6.ro_dst.sin6_family = AF_INET6;
1821 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1822 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1823 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1824 	}
1825 	if (sro6.ro_rt != NULL) {
1826 		ifp = sro6.ro_rt->rt_ifp;
1827 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1828 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1829 		else
1830 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1831 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1832 
1833 		/* Report additional interface capabilities. */
1834 		if (flags != NULL) {
1835 			if (ifp->if_capenable & IFCAP_TSO6 &&
1836 			    ifp->if_hwassist & CSUM_TSO)
1837 				*flags |= CSUM_TSO;
1838 		}
1839 		RTFREE(sro6.ro_rt);
1840 	}
1841 
1842 	return (maxmtu);
1843 }
1844 #endif /* INET6 */
1845 
1846 #ifdef IPSEC
1847 /* compute ESP/AH header size for TCP, including outer IP header. */
1848 size_t
1849 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1850 {
1851 	struct inpcb *inp;
1852 	struct mbuf *m;
1853 	size_t hdrsiz;
1854 	struct ip *ip;
1855 #ifdef INET6
1856 	struct ip6_hdr *ip6;
1857 #endif
1858 	struct tcphdr *th;
1859 
1860 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1861 		return (0);
1862 	MGETHDR(m, M_NOWAIT, MT_DATA);
1863 	if (!m)
1864 		return (0);
1865 
1866 #ifdef INET6
1867 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1868 		ip6 = mtod(m, struct ip6_hdr *);
1869 		th = (struct tcphdr *)(ip6 + 1);
1870 		m->m_pkthdr.len = m->m_len =
1871 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1872 		tcpip_fillheaders(inp, ip6, th);
1873 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1874 	} else
1875 #endif /* INET6 */
1876 	{
1877 		ip = mtod(m, struct ip *);
1878 		th = (struct tcphdr *)(ip + 1);
1879 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1880 		tcpip_fillheaders(inp, ip, th);
1881 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1882 	}
1883 
1884 	m_free(m);
1885 	return (hdrsiz);
1886 }
1887 #endif /* IPSEC */
1888 
1889 #ifdef TCP_SIGNATURE
1890 /*
1891  * Callback function invoked by m_apply() to digest TCP segment data
1892  * contained within an mbuf chain.
1893  */
1894 static int
1895 tcp_signature_apply(void *fstate, void *data, u_int len)
1896 {
1897 
1898 	MD5Update(fstate, (u_char *)data, len);
1899 	return (0);
1900 }
1901 
1902 /*
1903  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1904  *
1905  * Parameters:
1906  * m		pointer to head of mbuf chain
1907  * _unused
1908  * len		length of TCP segment data, excluding options
1909  * optlen	length of TCP segment options
1910  * buf		pointer to storage for computed MD5 digest
1911  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1912  *
1913  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1914  * When called from tcp_input(), we can be sure that th_sum has been
1915  * zeroed out and verified already.
1916  *
1917  * Return 0 if successful, otherwise return -1.
1918  *
1919  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1920  * search with the destination IP address, and a 'magic SPI' to be
1921  * determined by the application. This is hardcoded elsewhere to 1179
1922  * right now. Another branch of this code exists which uses the SPD to
1923  * specify per-application flows but it is unstable.
1924  */
1925 int
1926 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1927     u_char *buf, u_int direction)
1928 {
1929 	union sockaddr_union dst;
1930 #ifdef INET
1931 	struct ippseudo ippseudo;
1932 #endif
1933 	MD5_CTX ctx;
1934 	int doff;
1935 	struct ip *ip;
1936 #ifdef INET
1937 	struct ipovly *ipovly;
1938 #endif
1939 	struct secasvar *sav;
1940 	struct tcphdr *th;
1941 #ifdef INET6
1942 	struct ip6_hdr *ip6;
1943 	struct in6_addr in6;
1944 	char ip6buf[INET6_ADDRSTRLEN];
1945 	uint32_t plen;
1946 	uint16_t nhdr;
1947 #endif
1948 	u_short savecsum;
1949 
1950 	KASSERT(m != NULL, ("NULL mbuf chain"));
1951 	KASSERT(buf != NULL, ("NULL signature pointer"));
1952 
1953 	/* Extract the destination from the IP header in the mbuf. */
1954 	bzero(&dst, sizeof(union sockaddr_union));
1955 	ip = mtod(m, struct ip *);
1956 #ifdef INET6
1957 	ip6 = NULL;	/* Make the compiler happy. */
1958 #endif
1959 	switch (ip->ip_v) {
1960 #ifdef INET
1961 	case IPVERSION:
1962 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1963 		dst.sa.sa_family = AF_INET;
1964 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1965 		    ip->ip_src : ip->ip_dst;
1966 		break;
1967 #endif
1968 #ifdef INET6
1969 	case (IPV6_VERSION >> 4):
1970 		ip6 = mtod(m, struct ip6_hdr *);
1971 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1972 		dst.sa.sa_family = AF_INET6;
1973 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1974 		    ip6->ip6_src : ip6->ip6_dst;
1975 		break;
1976 #endif
1977 	default:
1978 		return (EINVAL);
1979 		/* NOTREACHED */
1980 		break;
1981 	}
1982 
1983 	/* Look up an SADB entry which matches the address of the peer. */
1984 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1985 	if (sav == NULL) {
1986 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1987 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1988 #ifdef INET6
1989 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
1990 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1991 #endif
1992 			"(unsupported)"));
1993 		return (EINVAL);
1994 	}
1995 
1996 	MD5Init(&ctx);
1997 	/*
1998 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1999 	 *
2000 	 * XXX The ippseudo header MUST be digested in network byte order,
2001 	 * or else we'll fail the regression test. Assume all fields we've
2002 	 * been doing arithmetic on have been in host byte order.
2003 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2004 	 * tcp_output(), the underlying ip_len member has not yet been set.
2005 	 */
2006 	switch (ip->ip_v) {
2007 #ifdef INET
2008 	case IPVERSION:
2009 		ipovly = (struct ipovly *)ip;
2010 		ippseudo.ippseudo_src = ipovly->ih_src;
2011 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2012 		ippseudo.ippseudo_pad = 0;
2013 		ippseudo.ippseudo_p = IPPROTO_TCP;
2014 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2015 		    optlen);
2016 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2017 
2018 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2019 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2020 		break;
2021 #endif
2022 #ifdef INET6
2023 	/*
2024 	 * RFC 2385, 2.0  Proposal
2025 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2026 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2027 	 * extended next header value (to form 32 bits), and 32-bit segment
2028 	 * length.
2029 	 * Note: Upper-Layer Packet Length comes before Next Header.
2030 	 */
2031 	case (IPV6_VERSION >> 4):
2032 		in6 = ip6->ip6_src;
2033 		in6_clearscope(&in6);
2034 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2035 		in6 = ip6->ip6_dst;
2036 		in6_clearscope(&in6);
2037 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2038 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2039 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2040 		nhdr = 0;
2041 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2042 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2043 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2044 		nhdr = IPPROTO_TCP;
2045 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2046 
2047 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2048 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2049 		break;
2050 #endif
2051 	default:
2052 		return (EINVAL);
2053 		/* NOTREACHED */
2054 		break;
2055 	}
2056 
2057 
2058 	/*
2059 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2060 	 * The TCP checksum must be set to zero.
2061 	 */
2062 	savecsum = th->th_sum;
2063 	th->th_sum = 0;
2064 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2065 	th->th_sum = savecsum;
2066 
2067 	/*
2068 	 * Step 3: Update MD5 hash with TCP segment data.
2069 	 *         Use m_apply() to avoid an early m_pullup().
2070 	 */
2071 	if (len > 0)
2072 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2073 
2074 	/*
2075 	 * Step 4: Update MD5 hash with shared secret.
2076 	 */
2077 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2078 	MD5Final(buf, &ctx);
2079 
2080 	key_sa_recordxfer(sav, m);
2081 	KEY_FREESAV(&sav);
2082 	return (0);
2083 }
2084 
2085 /*
2086  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2087  *
2088  * Parameters:
2089  * m		pointer to head of mbuf chain
2090  * len		length of TCP segment data, excluding options
2091  * optlen	length of TCP segment options
2092  * buf		pointer to storage for computed MD5 digest
2093  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2094  *
2095  * Return 1 if successful, otherwise return 0.
2096  */
2097 int
2098 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2099     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2100 {
2101 	char tmpdigest[TCP_SIGLEN];
2102 
2103 	if (tcp_sig_checksigs == 0)
2104 		return (1);
2105 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2106 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2107 
2108 			/*
2109 			 * If this socket is not expecting signature but
2110 			 * the segment contains signature just fail.
2111 			 */
2112 			TCPSTAT_INC(tcps_sig_err_sigopt);
2113 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2114 			return (0);
2115 		}
2116 
2117 		/* Signature is not expected, and not present in segment. */
2118 		return (1);
2119 	}
2120 
2121 	/*
2122 	 * If this socket is expecting signature but the segment does not
2123 	 * contain any just fail.
2124 	 */
2125 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2126 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2127 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2128 		return (0);
2129 	}
2130 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2131 	    IPSEC_DIR_INBOUND) == -1) {
2132 		TCPSTAT_INC(tcps_sig_err_buildsig);
2133 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2134 		return (0);
2135 	}
2136 
2137 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2138 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2139 		return (0);
2140 	}
2141 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2142 	return (1);
2143 }
2144 #endif /* TCP_SIGNATURE */
2145 
2146 static int
2147 sysctl_drop(SYSCTL_HANDLER_ARGS)
2148 {
2149 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2150 	struct sockaddr_storage addrs[2];
2151 	struct inpcb *inp;
2152 	struct tcpcb *tp;
2153 	struct tcptw *tw;
2154 	struct sockaddr_in *fin, *lin;
2155 #ifdef INET6
2156 	struct sockaddr_in6 *fin6, *lin6;
2157 #endif
2158 	int error;
2159 
2160 	inp = NULL;
2161 	fin = lin = NULL;
2162 #ifdef INET6
2163 	fin6 = lin6 = NULL;
2164 #endif
2165 	error = 0;
2166 
2167 	if (req->oldptr != NULL || req->oldlen != 0)
2168 		return (EINVAL);
2169 	if (req->newptr == NULL)
2170 		return (EPERM);
2171 	if (req->newlen < sizeof(addrs))
2172 		return (ENOMEM);
2173 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2174 	if (error)
2175 		return (error);
2176 
2177 	switch (addrs[0].ss_family) {
2178 #ifdef INET6
2179 	case AF_INET6:
2180 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2181 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2182 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2183 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2184 			return (EINVAL);
2185 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2186 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2187 				return (EINVAL);
2188 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2189 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2190 			fin = (struct sockaddr_in *)&addrs[0];
2191 			lin = (struct sockaddr_in *)&addrs[1];
2192 			break;
2193 		}
2194 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2195 		if (error)
2196 			return (error);
2197 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2198 		if (error)
2199 			return (error);
2200 		break;
2201 #endif
2202 #ifdef INET
2203 	case AF_INET:
2204 		fin = (struct sockaddr_in *)&addrs[0];
2205 		lin = (struct sockaddr_in *)&addrs[1];
2206 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2207 		    lin->sin_len != sizeof(struct sockaddr_in))
2208 			return (EINVAL);
2209 		break;
2210 #endif
2211 	default:
2212 		return (EINVAL);
2213 	}
2214 	INP_INFO_WLOCK(&V_tcbinfo);
2215 	switch (addrs[0].ss_family) {
2216 #ifdef INET6
2217 	case AF_INET6:
2218 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2219 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2220 		    INPLOOKUP_WLOCKPCB, NULL);
2221 		break;
2222 #endif
2223 #ifdef INET
2224 	case AF_INET:
2225 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2226 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2227 		break;
2228 #endif
2229 	}
2230 	if (inp != NULL) {
2231 		if (inp->inp_flags & INP_TIMEWAIT) {
2232 			/*
2233 			 * XXXRW: There currently exists a state where an
2234 			 * inpcb is present, but its timewait state has been
2235 			 * discarded.  For now, don't allow dropping of this
2236 			 * type of inpcb.
2237 			 */
2238 			tw = intotw(inp);
2239 			if (tw != NULL)
2240 				tcp_twclose(tw, 0);
2241 			else
2242 				INP_WUNLOCK(inp);
2243 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2244 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2245 			tp = intotcpcb(inp);
2246 			tp = tcp_drop(tp, ECONNABORTED);
2247 			if (tp != NULL)
2248 				INP_WUNLOCK(inp);
2249 		} else
2250 			INP_WUNLOCK(inp);
2251 	} else
2252 		error = ESRCH;
2253 	INP_INFO_WUNLOCK(&V_tcbinfo);
2254 	return (error);
2255 }
2256 
2257 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2258     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2259     0, sysctl_drop, "", "Drop TCP connection");
2260 
2261 /*
2262  * Generate a standardized TCP log line for use throughout the
2263  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2264  * allow use in the interrupt context.
2265  *
2266  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2267  * NB: The function may return NULL if memory allocation failed.
2268  *
2269  * Due to header inclusion and ordering limitations the struct ip
2270  * and ip6_hdr pointers have to be passed as void pointers.
2271  */
2272 char *
2273 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2274     const void *ip6hdr)
2275 {
2276 
2277 	/* Is logging enabled? */
2278 	if (tcp_log_in_vain == 0)
2279 		return (NULL);
2280 
2281 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2282 }
2283 
2284 char *
2285 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2286     const void *ip6hdr)
2287 {
2288 
2289 	/* Is logging enabled? */
2290 	if (tcp_log_debug == 0)
2291 		return (NULL);
2292 
2293 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2294 }
2295 
2296 static char *
2297 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2298     const void *ip6hdr)
2299 {
2300 	char *s, *sp;
2301 	size_t size;
2302 	struct ip *ip;
2303 #ifdef INET6
2304 	const struct ip6_hdr *ip6;
2305 
2306 	ip6 = (const struct ip6_hdr *)ip6hdr;
2307 #endif /* INET6 */
2308 	ip = (struct ip *)ip4hdr;
2309 
2310 	/*
2311 	 * The log line looks like this:
2312 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2313 	 */
2314 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2315 	    sizeof(PRINT_TH_FLAGS) + 1 +
2316 #ifdef INET6
2317 	    2 * INET6_ADDRSTRLEN;
2318 #else
2319 	    2 * INET_ADDRSTRLEN;
2320 #endif /* INET6 */
2321 
2322 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2323 	if (s == NULL)
2324 		return (NULL);
2325 
2326 	strcat(s, "TCP: [");
2327 	sp = s + strlen(s);
2328 
2329 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2330 		inet_ntoa_r(inc->inc_faddr, sp);
2331 		sp = s + strlen(s);
2332 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2333 		sp = s + strlen(s);
2334 		inet_ntoa_r(inc->inc_laddr, sp);
2335 		sp = s + strlen(s);
2336 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2337 #ifdef INET6
2338 	} else if (inc) {
2339 		ip6_sprintf(sp, &inc->inc6_faddr);
2340 		sp = s + strlen(s);
2341 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2342 		sp = s + strlen(s);
2343 		ip6_sprintf(sp, &inc->inc6_laddr);
2344 		sp = s + strlen(s);
2345 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2346 	} else if (ip6 && th) {
2347 		ip6_sprintf(sp, &ip6->ip6_src);
2348 		sp = s + strlen(s);
2349 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2350 		sp = s + strlen(s);
2351 		ip6_sprintf(sp, &ip6->ip6_dst);
2352 		sp = s + strlen(s);
2353 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2354 #endif /* INET6 */
2355 #ifdef INET
2356 	} else if (ip && th) {
2357 		inet_ntoa_r(ip->ip_src, sp);
2358 		sp = s + strlen(s);
2359 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2360 		sp = s + strlen(s);
2361 		inet_ntoa_r(ip->ip_dst, sp);
2362 		sp = s + strlen(s);
2363 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2364 #endif /* INET */
2365 	} else {
2366 		free(s, M_TCPLOG);
2367 		return (NULL);
2368 	}
2369 	sp = s + strlen(s);
2370 	if (th)
2371 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2372 	if (*(s + size - 1) != '\0')
2373 		panic("%s: string too long", __func__);
2374 	return (s);
2375 }
2376