xref: /freebsd/sys/netinet/tcp_subr.c (revision b2db760808f74bb53c232900091c9da801ebbfcc)
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/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/jail.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/route.h>
62 #include <net/if.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet6/ip6protosw.h>
98 
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
102 #ifdef INET6
103 #include <netipsec/ipsec6.h>
104 #endif
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
107 #endif /*IPSEC*/
108 
109 #include <machine/in_cksum.h>
110 #include <sys/md5.h>
111 
112 #include <security/mac/mac_framework.h>
113 
114 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
115 #ifdef INET6
116 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
117 #endif
118 
119 static int
120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
121 {
122 	int error, new;
123 
124 	new = V_tcp_mssdflt;
125 	error = sysctl_handle_int(oidp, &new, 0, req);
126 	if (error == 0 && req->newptr) {
127 		if (new < TCP_MINMSS)
128 			error = EINVAL;
129 		else
130 			V_tcp_mssdflt = new;
131 	}
132 	return (error);
133 }
134 
135 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
136     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
137     &sysctl_net_inet_tcp_mss_check, "I",
138     "Default TCP Maximum Segment Size");
139 
140 #ifdef INET6
141 static int
142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
143 {
144 	int error, new;
145 
146 	new = V_tcp_v6mssdflt;
147 	error = sysctl_handle_int(oidp, &new, 0, req);
148 	if (error == 0 && req->newptr) {
149 		if (new < TCP_MINMSS)
150 			error = EINVAL;
151 		else
152 			V_tcp_v6mssdflt = new;
153 	}
154 	return (error);
155 }
156 
157 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
158     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
159     &sysctl_net_inet_tcp_mss_v6_check, "I",
160    "Default TCP Maximum Segment Size for IPv6");
161 #endif
162 
163 static int
164 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
165 {
166 
167 	VNET_SYSCTL_ARG(req, arg1);
168 	return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
169 }
170 
171 /*
172  * Minimum MSS we accept and use. This prevents DoS attacks where
173  * we are forced to a ridiculous low MSS like 20 and send hundreds
174  * of packets instead of one. The effect scales with the available
175  * bandwidth and quickly saturates the CPU and network interface
176  * with packet generation and sending. Set to zero to disable MINMSS
177  * checking. This setting prevents us from sending too small packets.
178  */
179 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
180 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
181      &VNET_NAME(tcp_minmss), 0,
182     "Minmum TCP Maximum Segment Size");
183 
184 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
186     &VNET_NAME(tcp_do_rfc1323), 0,
187     "Enable rfc1323 (high performance TCP) extensions");
188 
189 static int	tcp_log_debug = 0;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
191     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
192 
193 static int	tcp_tcbhashsize = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
195     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
196 
197 static int	do_tcpdrain = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
199     "Enable tcp_drain routine for extra help when low on mbufs");
200 
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
202     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
203 
204 static VNET_DEFINE(int, icmp_may_rst) = 1;
205 #define	V_icmp_may_rst			VNET(icmp_may_rst)
206 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
207     &VNET_NAME(icmp_may_rst), 0,
208     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
209 
210 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
211 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
212 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
213     &VNET_NAME(tcp_isn_reseed_interval), 0,
214     "Seconds between reseeding of ISN secret");
215 
216 /*
217  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
218  * 1024 exists only for debugging.  A good production default would be
219  * something like 6100.
220  */
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222     "TCP inflight data limiting");
223 
224 static VNET_DEFINE(int, tcp_inflight_enable) = 0;
225 #define	V_tcp_inflight_enable		VNET(tcp_inflight_enable)
226 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
227     &VNET_NAME(tcp_inflight_enable), 0,
228     "Enable automatic TCP inflight data limiting");
229 
230 static int	tcp_inflight_debug = 0;
231 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
232     &tcp_inflight_debug, 0,
233     "Debug TCP inflight calculations");
234 
235 static VNET_DEFINE(int, tcp_inflight_rttthresh);
236 #define	V_tcp_inflight_rttthresh	VNET(tcp_inflight_rttthresh)
237 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
238     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
239     vnet_sysctl_msec_to_ticks, "I",
240     "RTT threshold below which inflight will deactivate itself");
241 
242 static VNET_DEFINE(int, tcp_inflight_min) = 6144;
243 #define	V_tcp_inflight_min		VNET(tcp_inflight_min)
244 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
245     &VNET_NAME(tcp_inflight_min), 0,
246     "Lower-bound for TCP inflight window");
247 
248 static VNET_DEFINE(int, tcp_inflight_max) = TCP_MAXWIN << TCP_MAX_WINSHIFT;
249 #define	V_tcp_inflight_max		VNET(tcp_inflight_max)
250 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
251     &VNET_NAME(tcp_inflight_max), 0,
252     "Upper-bound for TCP inflight window");
253 
254 static VNET_DEFINE(int, tcp_inflight_stab) = 20;
255 #define	V_tcp_inflight_stab		VNET(tcp_inflight_stab)
256 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
257     &VNET_NAME(tcp_inflight_stab), 0,
258     "Inflight Algorithm Stabilization 20 = 2 packets");
259 
260 #ifdef TCP_SORECEIVE_STREAM
261 static int	tcp_soreceive_stream = 0;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
263     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
264 #endif
265 
266 VNET_DEFINE(uma_zone_t, sack_hole_zone);
267 #define	V_sack_hole_zone		VNET(sack_hole_zone)
268 
269 static struct inpcb *tcp_notify(struct inpcb *, int);
270 static void	tcp_isn_tick(void *);
271 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
272 		    void *ip4hdr, const void *ip6hdr);
273 
274 /*
275  * Target size of TCP PCB hash tables. Must be a power of two.
276  *
277  * Note that this can be overridden by the kernel environment
278  * variable net.inet.tcp.tcbhashsize
279  */
280 #ifndef TCBHASHSIZE
281 #define TCBHASHSIZE	512
282 #endif
283 
284 /*
285  * XXX
286  * Callouts should be moved into struct tcp directly.  They are currently
287  * separate because the tcpcb structure is exported to userland for sysctl
288  * parsing purposes, which do not know about callouts.
289  */
290 struct tcpcb_mem {
291 	struct	tcpcb		tcb;
292 	struct	tcp_timer	tt;
293 };
294 
295 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
296 #define	V_tcpcb_zone			VNET(tcpcb_zone)
297 
298 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
299 struct callout isn_callout;
300 static struct mtx isn_mtx;
301 
302 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
303 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
304 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
305 
306 /*
307  * TCP initialization.
308  */
309 static void
310 tcp_zone_change(void *tag)
311 {
312 
313 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
314 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
315 	tcp_tw_zone_change();
316 }
317 
318 static int
319 tcp_inpcb_init(void *mem, int size, int flags)
320 {
321 	struct inpcb *inp = mem;
322 
323 	INP_LOCK_INIT(inp, "inp", "tcpinp");
324 	return (0);
325 }
326 
327 void
328 tcp_init(void)
329 {
330 	int hashsize;
331 
332 	hashsize = TCBHASHSIZE;
333 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
334 	if (!powerof2(hashsize)) {
335 		printf("WARNING: TCB hash size not a power of 2\n");
336 		hashsize = 512; /* safe default */
337 	}
338 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
339 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE);
340 
341 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
342 
343 	/*
344 	 * These have to be type stable for the benefit of the timers.
345 	 */
346 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
347 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
348 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
349 
350 	tcp_tw_init();
351 	syncache_init();
352 	tcp_hc_init();
353 	tcp_reass_init();
354 
355 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
356 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
357 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
358 
359 	/* Skip initialization of globals for non-default instances. */
360 	if (!IS_DEFAULT_VNET(curvnet))
361 		return;
362 
363 	/* XXX virtualize those bellow? */
364 	tcp_delacktime = TCPTV_DELACK;
365 	tcp_keepinit = TCPTV_KEEP_INIT;
366 	tcp_keepidle = TCPTV_KEEP_IDLE;
367 	tcp_keepintvl = TCPTV_KEEPINTVL;
368 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
369 	tcp_msl = TCPTV_MSL;
370 	tcp_rexmit_min = TCPTV_MIN;
371 	if (tcp_rexmit_min < 1)
372 		tcp_rexmit_min = 1;
373 	tcp_rexmit_slop = TCPTV_CPU_VAR;
374 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
375 	tcp_tcbhashsize = hashsize;
376 
377 #ifdef TCP_SORECEIVE_STREAM
378 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
379 	if (tcp_soreceive_stream) {
380 		tcp_usrreqs.pru_soreceive = soreceive_stream;
381 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
382 	}
383 #endif
384 
385 #ifdef INET6
386 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
387 #else /* INET6 */
388 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
389 #endif /* INET6 */
390 	if (max_protohdr < TCP_MINPROTOHDR)
391 		max_protohdr = TCP_MINPROTOHDR;
392 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
393 		panic("tcp_init");
394 #undef TCP_MINPROTOHDR
395 
396 	ISN_LOCK_INIT();
397 	callout_init(&isn_callout, CALLOUT_MPSAFE);
398 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
399 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
400 		SHUTDOWN_PRI_DEFAULT);
401 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
402 		EVENTHANDLER_PRI_ANY);
403 }
404 
405 #ifdef VIMAGE
406 void
407 tcp_destroy(void)
408 {
409 
410 	tcp_reass_destroy();
411 	tcp_hc_destroy();
412 	syncache_destroy();
413 	tcp_tw_destroy();
414 	in_pcbinfo_destroy(&V_tcbinfo);
415 	uma_zdestroy(V_sack_hole_zone);
416 	uma_zdestroy(V_tcpcb_zone);
417 }
418 #endif
419 
420 void
421 tcp_fini(void *xtp)
422 {
423 
424 	callout_stop(&isn_callout);
425 }
426 
427 /*
428  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
429  * tcp_template used to store this data in mbufs, but we now recopy it out
430  * of the tcpcb each time to conserve mbufs.
431  */
432 void
433 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
434 {
435 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
436 
437 	INP_WLOCK_ASSERT(inp);
438 
439 #ifdef INET6
440 	if ((inp->inp_vflag & INP_IPV6) != 0) {
441 		struct ip6_hdr *ip6;
442 
443 		ip6 = (struct ip6_hdr *)ip_ptr;
444 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
445 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
446 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
447 			(IPV6_VERSION & IPV6_VERSION_MASK);
448 		ip6->ip6_nxt = IPPROTO_TCP;
449 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
450 		ip6->ip6_src = inp->in6p_laddr;
451 		ip6->ip6_dst = inp->in6p_faddr;
452 	} else
453 #endif
454 	{
455 		struct ip *ip;
456 
457 		ip = (struct ip *)ip_ptr;
458 		ip->ip_v = IPVERSION;
459 		ip->ip_hl = 5;
460 		ip->ip_tos = inp->inp_ip_tos;
461 		ip->ip_len = 0;
462 		ip->ip_id = 0;
463 		ip->ip_off = 0;
464 		ip->ip_ttl = inp->inp_ip_ttl;
465 		ip->ip_sum = 0;
466 		ip->ip_p = IPPROTO_TCP;
467 		ip->ip_src = inp->inp_laddr;
468 		ip->ip_dst = inp->inp_faddr;
469 	}
470 	th->th_sport = inp->inp_lport;
471 	th->th_dport = inp->inp_fport;
472 	th->th_seq = 0;
473 	th->th_ack = 0;
474 	th->th_x2 = 0;
475 	th->th_off = 5;
476 	th->th_flags = 0;
477 	th->th_win = 0;
478 	th->th_urp = 0;
479 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
480 }
481 
482 /*
483  * Create template to be used to send tcp packets on a connection.
484  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
485  * use for this function is in keepalives, which use tcp_respond.
486  */
487 struct tcptemp *
488 tcpip_maketemplate(struct inpcb *inp)
489 {
490 	struct tcptemp *t;
491 
492 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
493 	if (t == NULL)
494 		return (NULL);
495 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
496 	return (t);
497 }
498 
499 /*
500  * Send a single message to the TCP at address specified by
501  * the given TCP/IP header.  If m == NULL, then we make a copy
502  * of the tcpiphdr at ti and send directly to the addressed host.
503  * This is used to force keep alive messages out using the TCP
504  * template for a connection.  If flags are given then we send
505  * a message back to the TCP which originated the * segment ti,
506  * and discard the mbuf containing it and any other attached mbufs.
507  *
508  * In any case the ack and sequence number of the transmitted
509  * segment are as specified by the parameters.
510  *
511  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
512  */
513 void
514 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
515     tcp_seq ack, tcp_seq seq, int flags)
516 {
517 	int tlen;
518 	int win = 0;
519 	struct ip *ip;
520 	struct tcphdr *nth;
521 #ifdef INET6
522 	struct ip6_hdr *ip6;
523 	int isipv6;
524 #endif /* INET6 */
525 	int ipflags = 0;
526 	struct inpcb *inp;
527 
528 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
529 
530 #ifdef INET6
531 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
532 	ip6 = ipgen;
533 #endif /* INET6 */
534 	ip = ipgen;
535 
536 	if (tp != NULL) {
537 		inp = tp->t_inpcb;
538 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
539 		INP_WLOCK_ASSERT(inp);
540 	} else
541 		inp = NULL;
542 
543 	if (tp != NULL) {
544 		if (!(flags & TH_RST)) {
545 			win = sbspace(&inp->inp_socket->so_rcv);
546 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
547 				win = (long)TCP_MAXWIN << tp->rcv_scale;
548 		}
549 	}
550 	if (m == NULL) {
551 		m = m_gethdr(M_DONTWAIT, MT_DATA);
552 		if (m == NULL)
553 			return;
554 		tlen = 0;
555 		m->m_data += max_linkhdr;
556 #ifdef INET6
557 		if (isipv6) {
558 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
559 			      sizeof(struct ip6_hdr));
560 			ip6 = mtod(m, struct ip6_hdr *);
561 			nth = (struct tcphdr *)(ip6 + 1);
562 		} else
563 #endif /* INET6 */
564 	      {
565 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
566 		ip = mtod(m, struct ip *);
567 		nth = (struct tcphdr *)(ip + 1);
568 	      }
569 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
570 		flags = TH_ACK;
571 	} else {
572 		/*
573 		 *  reuse the mbuf.
574 		 * XXX MRT We inherrit the FIB, which is lucky.
575 		 */
576 		m_freem(m->m_next);
577 		m->m_next = NULL;
578 		m->m_data = (caddr_t)ipgen;
579 		/* m_len is set later */
580 		tlen = 0;
581 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
582 #ifdef INET6
583 		if (isipv6) {
584 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
585 			nth = (struct tcphdr *)(ip6 + 1);
586 		} else
587 #endif /* INET6 */
588 	      {
589 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
590 		nth = (struct tcphdr *)(ip + 1);
591 	      }
592 		if (th != nth) {
593 			/*
594 			 * this is usually a case when an extension header
595 			 * exists between the IPv6 header and the
596 			 * TCP header.
597 			 */
598 			nth->th_sport = th->th_sport;
599 			nth->th_dport = th->th_dport;
600 		}
601 		xchg(nth->th_dport, nth->th_sport, uint16_t);
602 #undef xchg
603 	}
604 #ifdef INET6
605 	if (isipv6) {
606 		ip6->ip6_flow = 0;
607 		ip6->ip6_vfc = IPV6_VERSION;
608 		ip6->ip6_nxt = IPPROTO_TCP;
609 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
610 						tlen));
611 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
612 	} else
613 #endif
614 	{
615 		tlen += sizeof (struct tcpiphdr);
616 		ip->ip_len = tlen;
617 		ip->ip_ttl = V_ip_defttl;
618 		if (V_path_mtu_discovery)
619 			ip->ip_off |= IP_DF;
620 	}
621 	m->m_len = tlen;
622 	m->m_pkthdr.len = tlen;
623 	m->m_pkthdr.rcvif = NULL;
624 #ifdef MAC
625 	if (inp != NULL) {
626 		/*
627 		 * Packet is associated with a socket, so allow the
628 		 * label of the response to reflect the socket label.
629 		 */
630 		INP_WLOCK_ASSERT(inp);
631 		mac_inpcb_create_mbuf(inp, m);
632 	} else {
633 		/*
634 		 * Packet is not associated with a socket, so possibly
635 		 * update the label in place.
636 		 */
637 		mac_netinet_tcp_reply(m);
638 	}
639 #endif
640 	nth->th_seq = htonl(seq);
641 	nth->th_ack = htonl(ack);
642 	nth->th_x2 = 0;
643 	nth->th_off = sizeof (struct tcphdr) >> 2;
644 	nth->th_flags = flags;
645 	if (tp != NULL)
646 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
647 	else
648 		nth->th_win = htons((u_short)win);
649 	nth->th_urp = 0;
650 #ifdef INET6
651 	if (isipv6) {
652 		nth->th_sum = 0;
653 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
654 					sizeof(struct ip6_hdr),
655 					tlen - sizeof(struct ip6_hdr));
656 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
657 		    NULL, NULL);
658 	} else
659 #endif /* INET6 */
660 	{
661 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
662 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
663 		m->m_pkthdr.csum_flags = CSUM_TCP;
664 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
665 	}
666 #ifdef TCPDEBUG
667 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
668 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
669 #endif
670 #ifdef INET6
671 	if (isipv6)
672 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
673 	else
674 #endif /* INET6 */
675 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
676 }
677 
678 /*
679  * Create a new TCP control block, making an
680  * empty reassembly queue and hooking it to the argument
681  * protocol control block.  The `inp' parameter must have
682  * come from the zone allocator set up in tcp_init().
683  */
684 struct tcpcb *
685 tcp_newtcpcb(struct inpcb *inp)
686 {
687 	struct tcpcb_mem *tm;
688 	struct tcpcb *tp;
689 #ifdef INET6
690 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
691 #endif /* INET6 */
692 
693 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
694 	if (tm == NULL)
695 		return (NULL);
696 	tp = &tm->tcb;
697 #ifdef VIMAGE
698 	tp->t_vnet = inp->inp_vnet;
699 #endif
700 	tp->t_timers = &tm->tt;
701 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
702 	tp->t_maxseg = tp->t_maxopd =
703 #ifdef INET6
704 		isipv6 ? V_tcp_v6mssdflt :
705 #endif /* INET6 */
706 		V_tcp_mssdflt;
707 
708 	/* Set up our timeouts. */
709 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
710 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
711 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
712 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
713 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
714 
715 	if (V_tcp_do_rfc1323)
716 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
717 	if (V_tcp_do_sack)
718 		tp->t_flags |= TF_SACK_PERMIT;
719 	TAILQ_INIT(&tp->snd_holes);
720 	tp->t_inpcb = inp;	/* XXX */
721 	/*
722 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
723 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
724 	 * reasonable initial retransmit time.
725 	 */
726 	tp->t_srtt = TCPTV_SRTTBASE;
727 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
728 	tp->t_rttmin = tcp_rexmit_min;
729 	tp->t_rxtcur = TCPTV_RTOBASE;
730 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
731 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
732 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
733 	tp->t_rcvtime = ticks;
734 	tp->t_bw_rtttime = ticks;
735 	/*
736 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
737 	 * because the socket may be bound to an IPv6 wildcard address,
738 	 * which may match an IPv4-mapped IPv6 address.
739 	 */
740 	inp->inp_ip_ttl = V_ip_defttl;
741 	inp->inp_ppcb = tp;
742 	return (tp);		/* XXX */
743 }
744 
745 /*
746  * Drop a TCP connection, reporting
747  * the specified error.  If connection is synchronized,
748  * then send a RST to peer.
749  */
750 struct tcpcb *
751 tcp_drop(struct tcpcb *tp, int errno)
752 {
753 	struct socket *so = tp->t_inpcb->inp_socket;
754 
755 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
756 	INP_WLOCK_ASSERT(tp->t_inpcb);
757 
758 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
759 		tp->t_state = TCPS_CLOSED;
760 		(void) tcp_output_reset(tp);
761 		TCPSTAT_INC(tcps_drops);
762 	} else
763 		TCPSTAT_INC(tcps_conndrops);
764 	if (errno == ETIMEDOUT && tp->t_softerror)
765 		errno = tp->t_softerror;
766 	so->so_error = errno;
767 	return (tcp_close(tp));
768 }
769 
770 void
771 tcp_discardcb(struct tcpcb *tp)
772 {
773 	struct tseg_qent *q;
774 	struct inpcb *inp = tp->t_inpcb;
775 	struct socket *so = inp->inp_socket;
776 #ifdef INET6
777 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
778 #endif /* INET6 */
779 
780 	INP_WLOCK_ASSERT(inp);
781 
782 	/*
783 	 * Make sure that all of our timers are stopped before we delete the
784 	 * PCB.
785 	 *
786 	 * XXXRW: Really, we would like to use callout_drain() here in order
787 	 * to avoid races experienced in tcp_timer.c where a timer is already
788 	 * executing at this point.  However, we can't, both because we're
789 	 * running in a context where we can't sleep, and also because we
790 	 * hold locks required by the timers.  What we instead need to do is
791 	 * test to see if callout_drain() is required, and if so, defer some
792 	 * portion of the remainder of tcp_discardcb() to an asynchronous
793 	 * context that can callout_drain() and then continue.  Some care
794 	 * will be required to ensure that no further processing takes place
795 	 * on the tcpcb, even though it hasn't been freed (a flag?).
796 	 */
797 	callout_stop(&tp->t_timers->tt_rexmt);
798 	callout_stop(&tp->t_timers->tt_persist);
799 	callout_stop(&tp->t_timers->tt_keep);
800 	callout_stop(&tp->t_timers->tt_2msl);
801 	callout_stop(&tp->t_timers->tt_delack);
802 
803 	/*
804 	 * If we got enough samples through the srtt filter,
805 	 * save the rtt and rttvar in the routing entry.
806 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
807 	 * 4 samples is enough for the srtt filter to converge
808 	 * to within enough % of the correct value; fewer samples
809 	 * and we could save a bogus rtt. The danger is not high
810 	 * as tcp quickly recovers from everything.
811 	 * XXX: Works very well but needs some more statistics!
812 	 */
813 	if (tp->t_rttupdated >= 4) {
814 		struct hc_metrics_lite metrics;
815 		u_long ssthresh;
816 
817 		bzero(&metrics, sizeof(metrics));
818 		/*
819 		 * Update the ssthresh always when the conditions below
820 		 * are satisfied. This gives us better new start value
821 		 * for the congestion avoidance for new connections.
822 		 * ssthresh is only set if packet loss occured on a session.
823 		 *
824 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
825 		 * being torn down.  Ideally this code would not use 'so'.
826 		 */
827 		ssthresh = tp->snd_ssthresh;
828 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
829 			/*
830 			 * convert the limit from user data bytes to
831 			 * packets then to packet data bytes.
832 			 */
833 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
834 			if (ssthresh < 2)
835 				ssthresh = 2;
836 			ssthresh *= (u_long)(tp->t_maxseg +
837 #ifdef INET6
838 				      (isipv6 ? sizeof (struct ip6_hdr) +
839 					       sizeof (struct tcphdr) :
840 #endif
841 				       sizeof (struct tcpiphdr)
842 #ifdef INET6
843 				       )
844 #endif
845 				      );
846 		} else
847 			ssthresh = 0;
848 		metrics.rmx_ssthresh = ssthresh;
849 
850 		metrics.rmx_rtt = tp->t_srtt;
851 		metrics.rmx_rttvar = tp->t_rttvar;
852 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
853 		metrics.rmx_bandwidth = tp->snd_bandwidth;
854 		metrics.rmx_cwnd = tp->snd_cwnd;
855 		metrics.rmx_sendpipe = 0;
856 		metrics.rmx_recvpipe = 0;
857 
858 		tcp_hc_update(&inp->inp_inc, &metrics);
859 	}
860 
861 	/* free the reassembly queue, if any */
862 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
863 		LIST_REMOVE(q, tqe_q);
864 		m_freem(q->tqe_m);
865 		uma_zfree(V_tcp_reass_zone, q);
866 		tp->t_segqlen--;
867 		V_tcp_reass_qsize--;
868 	}
869 	/* Disconnect offload device, if any. */
870 	tcp_offload_detach(tp);
871 
872 	tcp_free_sackholes(tp);
873 	inp->inp_ppcb = NULL;
874 	tp->t_inpcb = NULL;
875 	uma_zfree(V_tcpcb_zone, tp);
876 }
877 
878 /*
879  * Attempt to close a TCP control block, marking it as dropped, and freeing
880  * the socket if we hold the only reference.
881  */
882 struct tcpcb *
883 tcp_close(struct tcpcb *tp)
884 {
885 	struct inpcb *inp = tp->t_inpcb;
886 	struct socket *so;
887 
888 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
889 	INP_WLOCK_ASSERT(inp);
890 
891 	/* Notify any offload devices of listener close */
892 	if (tp->t_state == TCPS_LISTEN)
893 		tcp_offload_listen_close(tp);
894 	in_pcbdrop(inp);
895 	TCPSTAT_INC(tcps_closed);
896 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
897 	so = inp->inp_socket;
898 	soisdisconnected(so);
899 	if (inp->inp_flags & INP_SOCKREF) {
900 		KASSERT(so->so_state & SS_PROTOREF,
901 		    ("tcp_close: !SS_PROTOREF"));
902 		inp->inp_flags &= ~INP_SOCKREF;
903 		INP_WUNLOCK(inp);
904 		ACCEPT_LOCK();
905 		SOCK_LOCK(so);
906 		so->so_state &= ~SS_PROTOREF;
907 		sofree(so);
908 		return (NULL);
909 	}
910 	return (tp);
911 }
912 
913 void
914 tcp_drain(void)
915 {
916 	VNET_ITERATOR_DECL(vnet_iter);
917 
918 	if (!do_tcpdrain)
919 		return;
920 
921 	VNET_LIST_RLOCK_NOSLEEP();
922 	VNET_FOREACH(vnet_iter) {
923 		CURVNET_SET(vnet_iter);
924 		struct inpcb *inpb;
925 		struct tcpcb *tcpb;
926 		struct tseg_qent *te;
927 
928 	/*
929 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
930 	 * if there is one...
931 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
932 	 *      reassembly queue should be flushed, but in a situation
933 	 *	where we're really low on mbufs, this is potentially
934 	 *	usefull.
935 	 */
936 		INP_INFO_RLOCK(&V_tcbinfo);
937 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
938 			if (inpb->inp_flags & INP_TIMEWAIT)
939 				continue;
940 			INP_WLOCK(inpb);
941 			if ((tcpb = intotcpcb(inpb)) != NULL) {
942 				while ((te = LIST_FIRST(&tcpb->t_segq))
943 			            != NULL) {
944 					LIST_REMOVE(te, tqe_q);
945 					m_freem(te->tqe_m);
946 					uma_zfree(V_tcp_reass_zone, te);
947 					tcpb->t_segqlen--;
948 					V_tcp_reass_qsize--;
949 				}
950 				tcp_clean_sackreport(tcpb);
951 			}
952 			INP_WUNLOCK(inpb);
953 		}
954 		INP_INFO_RUNLOCK(&V_tcbinfo);
955 		CURVNET_RESTORE();
956 	}
957 	VNET_LIST_RUNLOCK_NOSLEEP();
958 }
959 
960 /*
961  * Notify a tcp user of an asynchronous error;
962  * store error as soft error, but wake up user
963  * (for now, won't do anything until can select for soft error).
964  *
965  * Do not wake up user since there currently is no mechanism for
966  * reporting soft errors (yet - a kqueue filter may be added).
967  */
968 static struct inpcb *
969 tcp_notify(struct inpcb *inp, int error)
970 {
971 	struct tcpcb *tp;
972 
973 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
974 	INP_WLOCK_ASSERT(inp);
975 
976 	if ((inp->inp_flags & INP_TIMEWAIT) ||
977 	    (inp->inp_flags & INP_DROPPED))
978 		return (inp);
979 
980 	tp = intotcpcb(inp);
981 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
982 
983 	/*
984 	 * Ignore some errors if we are hooked up.
985 	 * If connection hasn't completed, has retransmitted several times,
986 	 * and receives a second error, give up now.  This is better
987 	 * than waiting a long time to establish a connection that
988 	 * can never complete.
989 	 */
990 	if (tp->t_state == TCPS_ESTABLISHED &&
991 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
992 	     error == EHOSTDOWN)) {
993 		return (inp);
994 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
995 	    tp->t_softerror) {
996 		tp = tcp_drop(tp, error);
997 		if (tp != NULL)
998 			return (inp);
999 		else
1000 			return (NULL);
1001 	} else {
1002 		tp->t_softerror = error;
1003 		return (inp);
1004 	}
1005 #if 0
1006 	wakeup( &so->so_timeo);
1007 	sorwakeup(so);
1008 	sowwakeup(so);
1009 #endif
1010 }
1011 
1012 static int
1013 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1014 {
1015 	int error, i, m, n, pcb_count;
1016 	struct inpcb *inp, **inp_list;
1017 	inp_gen_t gencnt;
1018 	struct xinpgen xig;
1019 
1020 	/*
1021 	 * The process of preparing the TCB list is too time-consuming and
1022 	 * resource-intensive to repeat twice on every request.
1023 	 */
1024 	if (req->oldptr == NULL) {
1025 		m = syncache_pcbcount();
1026 		n = V_tcbinfo.ipi_count;
1027 		n += imax((m + n) / 8, 10);
1028 		req->oldidx = 2 * (sizeof xig) +
1029 		    (m + n) * sizeof(struct xtcpcb);
1030 		return (0);
1031 	}
1032 
1033 	if (req->newptr != NULL)
1034 		return (EPERM);
1035 
1036 	/*
1037 	 * OK, now we're committed to doing something.
1038 	 */
1039 	INP_INFO_RLOCK(&V_tcbinfo);
1040 	gencnt = V_tcbinfo.ipi_gencnt;
1041 	n = V_tcbinfo.ipi_count;
1042 	INP_INFO_RUNLOCK(&V_tcbinfo);
1043 
1044 	m = syncache_pcbcount();
1045 
1046 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1047 		+ (n + m) * sizeof(struct xtcpcb));
1048 	if (error != 0)
1049 		return (error);
1050 
1051 	xig.xig_len = sizeof xig;
1052 	xig.xig_count = n + m;
1053 	xig.xig_gen = gencnt;
1054 	xig.xig_sogen = so_gencnt;
1055 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1056 	if (error)
1057 		return (error);
1058 
1059 	error = syncache_pcblist(req, m, &pcb_count);
1060 	if (error)
1061 		return (error);
1062 
1063 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1064 	if (inp_list == NULL)
1065 		return (ENOMEM);
1066 
1067 	INP_INFO_RLOCK(&V_tcbinfo);
1068 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1069 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1070 		INP_WLOCK(inp);
1071 		if (inp->inp_gencnt <= gencnt) {
1072 			/*
1073 			 * XXX: This use of cr_cansee(), introduced with
1074 			 * TCP state changes, is not quite right, but for
1075 			 * now, better than nothing.
1076 			 */
1077 			if (inp->inp_flags & INP_TIMEWAIT) {
1078 				if (intotw(inp) != NULL)
1079 					error = cr_cansee(req->td->td_ucred,
1080 					    intotw(inp)->tw_cred);
1081 				else
1082 					error = EINVAL;	/* Skip this inp. */
1083 			} else
1084 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1085 			if (error == 0) {
1086 				in_pcbref(inp);
1087 				inp_list[i++] = inp;
1088 			}
1089 		}
1090 		INP_WUNLOCK(inp);
1091 	}
1092 	INP_INFO_RUNLOCK(&V_tcbinfo);
1093 	n = i;
1094 
1095 	error = 0;
1096 	for (i = 0; i < n; i++) {
1097 		inp = inp_list[i];
1098 		INP_RLOCK(inp);
1099 		if (inp->inp_gencnt <= gencnt) {
1100 			struct xtcpcb xt;
1101 			void *inp_ppcb;
1102 
1103 			bzero(&xt, sizeof(xt));
1104 			xt.xt_len = sizeof xt;
1105 			/* XXX should avoid extra copy */
1106 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1107 			inp_ppcb = inp->inp_ppcb;
1108 			if (inp_ppcb == NULL)
1109 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1110 			else if (inp->inp_flags & INP_TIMEWAIT) {
1111 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1112 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1113 			} else {
1114 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1115 				if (xt.xt_tp.t_timers)
1116 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1117 			}
1118 			if (inp->inp_socket != NULL)
1119 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1120 			else {
1121 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1122 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1123 			}
1124 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1125 			INP_RUNLOCK(inp);
1126 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1127 		} else
1128 			INP_RUNLOCK(inp);
1129 	}
1130 	INP_INFO_WLOCK(&V_tcbinfo);
1131 	for (i = 0; i < n; i++) {
1132 		inp = inp_list[i];
1133 		INP_WLOCK(inp);
1134 		if (!in_pcbrele(inp))
1135 			INP_WUNLOCK(inp);
1136 	}
1137 	INP_INFO_WUNLOCK(&V_tcbinfo);
1138 
1139 	if (!error) {
1140 		/*
1141 		 * Give the user an updated idea of our state.
1142 		 * If the generation differs from what we told
1143 		 * her before, she knows that something happened
1144 		 * while we were processing this request, and it
1145 		 * might be necessary to retry.
1146 		 */
1147 		INP_INFO_RLOCK(&V_tcbinfo);
1148 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1149 		xig.xig_sogen = so_gencnt;
1150 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1151 		INP_INFO_RUNLOCK(&V_tcbinfo);
1152 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1153 	}
1154 	free(inp_list, M_TEMP);
1155 	return (error);
1156 }
1157 
1158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1159     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1160 
1161 static int
1162 tcp_getcred(SYSCTL_HANDLER_ARGS)
1163 {
1164 	struct xucred xuc;
1165 	struct sockaddr_in addrs[2];
1166 	struct inpcb *inp;
1167 	int error;
1168 
1169 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1170 	if (error)
1171 		return (error);
1172 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1173 	if (error)
1174 		return (error);
1175 	INP_INFO_RLOCK(&V_tcbinfo);
1176 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1177 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1178 	if (inp != NULL) {
1179 		INP_RLOCK(inp);
1180 		INP_INFO_RUNLOCK(&V_tcbinfo);
1181 		if (inp->inp_socket == NULL)
1182 			error = ENOENT;
1183 		if (error == 0)
1184 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1185 		if (error == 0)
1186 			cru2x(inp->inp_cred, &xuc);
1187 		INP_RUNLOCK(inp);
1188 	} else {
1189 		INP_INFO_RUNLOCK(&V_tcbinfo);
1190 		error = ENOENT;
1191 	}
1192 	if (error == 0)
1193 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1194 	return (error);
1195 }
1196 
1197 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1198     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1199     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1200 
1201 #ifdef INET6
1202 static int
1203 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1204 {
1205 	struct xucred xuc;
1206 	struct sockaddr_in6 addrs[2];
1207 	struct inpcb *inp;
1208 	int error, mapped = 0;
1209 
1210 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1211 	if (error)
1212 		return (error);
1213 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1214 	if (error)
1215 		return (error);
1216 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1217 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1218 		return (error);
1219 	}
1220 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1221 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1222 			mapped = 1;
1223 		else
1224 			return (EINVAL);
1225 	}
1226 
1227 	INP_INFO_RLOCK(&V_tcbinfo);
1228 	if (mapped == 1)
1229 		inp = in_pcblookup_hash(&V_tcbinfo,
1230 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1231 			addrs[1].sin6_port,
1232 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1233 			addrs[0].sin6_port,
1234 			0, NULL);
1235 	else
1236 		inp = in6_pcblookup_hash(&V_tcbinfo,
1237 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1238 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1239 	if (inp != NULL) {
1240 		INP_RLOCK(inp);
1241 		INP_INFO_RUNLOCK(&V_tcbinfo);
1242 		if (inp->inp_socket == NULL)
1243 			error = ENOENT;
1244 		if (error == 0)
1245 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1246 		if (error == 0)
1247 			cru2x(inp->inp_cred, &xuc);
1248 		INP_RUNLOCK(inp);
1249 	} else {
1250 		INP_INFO_RUNLOCK(&V_tcbinfo);
1251 		error = ENOENT;
1252 	}
1253 	if (error == 0)
1254 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1255 	return (error);
1256 }
1257 
1258 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1259     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1260     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1261 #endif
1262 
1263 
1264 void
1265 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1266 {
1267 	struct ip *ip = vip;
1268 	struct tcphdr *th;
1269 	struct in_addr faddr;
1270 	struct inpcb *inp;
1271 	struct tcpcb *tp;
1272 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1273 	struct icmp *icp;
1274 	struct in_conninfo inc;
1275 	tcp_seq icmp_tcp_seq;
1276 	int mtu;
1277 
1278 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1279 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1280 		return;
1281 
1282 	if (cmd == PRC_MSGSIZE)
1283 		notify = tcp_mtudisc;
1284 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1285 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1286 		notify = tcp_drop_syn_sent;
1287 	/*
1288 	 * Redirects don't need to be handled up here.
1289 	 */
1290 	else if (PRC_IS_REDIRECT(cmd))
1291 		return;
1292 	/*
1293 	 * Source quench is depreciated.
1294 	 */
1295 	else if (cmd == PRC_QUENCH)
1296 		return;
1297 	/*
1298 	 * Hostdead is ugly because it goes linearly through all PCBs.
1299 	 * XXX: We never get this from ICMP, otherwise it makes an
1300 	 * excellent DoS attack on machines with many connections.
1301 	 */
1302 	else if (cmd == PRC_HOSTDEAD)
1303 		ip = NULL;
1304 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1305 		return;
1306 	if (ip != NULL) {
1307 		icp = (struct icmp *)((caddr_t)ip
1308 				      - offsetof(struct icmp, icmp_ip));
1309 		th = (struct tcphdr *)((caddr_t)ip
1310 				       + (ip->ip_hl << 2));
1311 		INP_INFO_WLOCK(&V_tcbinfo);
1312 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1313 		    ip->ip_src, th->th_sport, 0, NULL);
1314 		if (inp != NULL)  {
1315 			INP_WLOCK(inp);
1316 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1317 			    !(inp->inp_flags & INP_DROPPED) &&
1318 			    !(inp->inp_socket == NULL)) {
1319 				icmp_tcp_seq = htonl(th->th_seq);
1320 				tp = intotcpcb(inp);
1321 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1322 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1323 					if (cmd == PRC_MSGSIZE) {
1324 					    /*
1325 					     * MTU discovery:
1326 					     * If we got a needfrag set the MTU
1327 					     * in the route to the suggested new
1328 					     * value (if given) and then notify.
1329 					     */
1330 					    bzero(&inc, sizeof(inc));
1331 					    inc.inc_faddr = faddr;
1332 					    inc.inc_fibnum =
1333 						inp->inp_inc.inc_fibnum;
1334 
1335 					    mtu = ntohs(icp->icmp_nextmtu);
1336 					    /*
1337 					     * If no alternative MTU was
1338 					     * proposed, try the next smaller
1339 					     * one.  ip->ip_len has already
1340 					     * been swapped in icmp_input().
1341 					     */
1342 					    if (!mtu)
1343 						mtu = ip_next_mtu(ip->ip_len,
1344 						 1);
1345 					    if (mtu < V_tcp_minmss
1346 						 + sizeof(struct tcpiphdr))
1347 						mtu = V_tcp_minmss
1348 						 + sizeof(struct tcpiphdr);
1349 					    /*
1350 					     * Only cache the the MTU if it
1351 					     * is smaller than the interface
1352 					     * or route MTU.  tcp_mtudisc()
1353 					     * will do right thing by itself.
1354 					     */
1355 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1356 						tcp_hc_updatemtu(&inc, mtu);
1357 					}
1358 
1359 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1360 				}
1361 			}
1362 			if (inp != NULL)
1363 				INP_WUNLOCK(inp);
1364 		} else {
1365 			bzero(&inc, sizeof(inc));
1366 			inc.inc_fport = th->th_dport;
1367 			inc.inc_lport = th->th_sport;
1368 			inc.inc_faddr = faddr;
1369 			inc.inc_laddr = ip->ip_src;
1370 			syncache_unreach(&inc, th);
1371 		}
1372 		INP_INFO_WUNLOCK(&V_tcbinfo);
1373 	} else
1374 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1375 }
1376 
1377 #ifdef INET6
1378 void
1379 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1380 {
1381 	struct tcphdr th;
1382 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1383 	struct ip6_hdr *ip6;
1384 	struct mbuf *m;
1385 	struct ip6ctlparam *ip6cp = NULL;
1386 	const struct sockaddr_in6 *sa6_src = NULL;
1387 	int off;
1388 	struct tcp_portonly {
1389 		u_int16_t th_sport;
1390 		u_int16_t th_dport;
1391 	} *thp;
1392 
1393 	if (sa->sa_family != AF_INET6 ||
1394 	    sa->sa_len != sizeof(struct sockaddr_in6))
1395 		return;
1396 
1397 	if (cmd == PRC_MSGSIZE)
1398 		notify = tcp_mtudisc;
1399 	else if (!PRC_IS_REDIRECT(cmd) &&
1400 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1401 		return;
1402 	/* Source quench is depreciated. */
1403 	else if (cmd == PRC_QUENCH)
1404 		return;
1405 
1406 	/* if the parameter is from icmp6, decode it. */
1407 	if (d != NULL) {
1408 		ip6cp = (struct ip6ctlparam *)d;
1409 		m = ip6cp->ip6c_m;
1410 		ip6 = ip6cp->ip6c_ip6;
1411 		off = ip6cp->ip6c_off;
1412 		sa6_src = ip6cp->ip6c_src;
1413 	} else {
1414 		m = NULL;
1415 		ip6 = NULL;
1416 		off = 0;	/* fool gcc */
1417 		sa6_src = &sa6_any;
1418 	}
1419 
1420 	if (ip6 != NULL) {
1421 		struct in_conninfo inc;
1422 		/*
1423 		 * XXX: We assume that when IPV6 is non NULL,
1424 		 * M and OFF are valid.
1425 		 */
1426 
1427 		/* check if we can safely examine src and dst ports */
1428 		if (m->m_pkthdr.len < off + sizeof(*thp))
1429 			return;
1430 
1431 		bzero(&th, sizeof(th));
1432 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1433 
1434 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1435 		    (struct sockaddr *)ip6cp->ip6c_src,
1436 		    th.th_sport, cmd, NULL, notify);
1437 
1438 		bzero(&inc, sizeof(inc));
1439 		inc.inc_fport = th.th_dport;
1440 		inc.inc_lport = th.th_sport;
1441 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1442 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1443 		inc.inc_flags |= INC_ISIPV6;
1444 		INP_INFO_WLOCK(&V_tcbinfo);
1445 		syncache_unreach(&inc, &th);
1446 		INP_INFO_WUNLOCK(&V_tcbinfo);
1447 	} else
1448 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1449 			      0, cmd, NULL, notify);
1450 }
1451 #endif /* INET6 */
1452 
1453 
1454 /*
1455  * Following is where TCP initial sequence number generation occurs.
1456  *
1457  * There are two places where we must use initial sequence numbers:
1458  * 1.  In SYN-ACK packets.
1459  * 2.  In SYN packets.
1460  *
1461  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1462  * tcp_syncache.c for details.
1463  *
1464  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1465  * depends on this property.  In addition, these ISNs should be
1466  * unguessable so as to prevent connection hijacking.  To satisfy
1467  * the requirements of this situation, the algorithm outlined in
1468  * RFC 1948 is used, with only small modifications.
1469  *
1470  * Implementation details:
1471  *
1472  * Time is based off the system timer, and is corrected so that it
1473  * increases by one megabyte per second.  This allows for proper
1474  * recycling on high speed LANs while still leaving over an hour
1475  * before rollover.
1476  *
1477  * As reading the *exact* system time is too expensive to be done
1478  * whenever setting up a TCP connection, we increment the time
1479  * offset in two ways.  First, a small random positive increment
1480  * is added to isn_offset for each connection that is set up.
1481  * Second, the function tcp_isn_tick fires once per clock tick
1482  * and increments isn_offset as necessary so that sequence numbers
1483  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1484  * random positive increments serve only to ensure that the same
1485  * exact sequence number is never sent out twice (as could otherwise
1486  * happen when a port is recycled in less than the system tick
1487  * interval.)
1488  *
1489  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1490  * between seeding of isn_secret.  This is normally set to zero,
1491  * as reseeding should not be necessary.
1492  *
1493  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1494  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1495  * general, this means holding an exclusive (write) lock.
1496  */
1497 
1498 #define ISN_BYTES_PER_SECOND 1048576
1499 #define ISN_STATIC_INCREMENT 4096
1500 #define ISN_RANDOM_INCREMENT (4096 - 1)
1501 
1502 static VNET_DEFINE(u_char, isn_secret[32]);
1503 static VNET_DEFINE(int, isn_last_reseed);
1504 static VNET_DEFINE(u_int32_t, isn_offset);
1505 static VNET_DEFINE(u_int32_t, isn_offset_old);
1506 
1507 #define	V_isn_secret			VNET(isn_secret)
1508 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1509 #define	V_isn_offset			VNET(isn_offset)
1510 #define	V_isn_offset_old		VNET(isn_offset_old)
1511 
1512 tcp_seq
1513 tcp_new_isn(struct tcpcb *tp)
1514 {
1515 	MD5_CTX isn_ctx;
1516 	u_int32_t md5_buffer[4];
1517 	tcp_seq new_isn;
1518 
1519 	INP_WLOCK_ASSERT(tp->t_inpcb);
1520 
1521 	ISN_LOCK();
1522 	/* Seed if this is the first use, reseed if requested. */
1523 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1524 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1525 		< (u_int)ticks))) {
1526 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1527 		V_isn_last_reseed = ticks;
1528 	}
1529 
1530 	/* Compute the md5 hash and return the ISN. */
1531 	MD5Init(&isn_ctx);
1532 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1533 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1534 #ifdef INET6
1535 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1536 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1537 			  sizeof(struct in6_addr));
1538 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1539 			  sizeof(struct in6_addr));
1540 	} else
1541 #endif
1542 	{
1543 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1544 			  sizeof(struct in_addr));
1545 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1546 			  sizeof(struct in_addr));
1547 	}
1548 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1549 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1550 	new_isn = (tcp_seq) md5_buffer[0];
1551 	V_isn_offset += ISN_STATIC_INCREMENT +
1552 		(arc4random() & ISN_RANDOM_INCREMENT);
1553 	new_isn += V_isn_offset;
1554 	ISN_UNLOCK();
1555 	return (new_isn);
1556 }
1557 
1558 /*
1559  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1560  * to keep time flowing at a relatively constant rate.  If the random
1561  * increments have already pushed us past the projected offset, do nothing.
1562  */
1563 static void
1564 tcp_isn_tick(void *xtp)
1565 {
1566 	VNET_ITERATOR_DECL(vnet_iter);
1567 	u_int32_t projected_offset;
1568 
1569 	VNET_LIST_RLOCK_NOSLEEP();
1570 	ISN_LOCK();
1571 	VNET_FOREACH(vnet_iter) {
1572 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1573 		projected_offset =
1574 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1575 
1576 		if (SEQ_GT(projected_offset, V_isn_offset))
1577 			V_isn_offset = projected_offset;
1578 
1579 		V_isn_offset_old = V_isn_offset;
1580 		CURVNET_RESTORE();
1581 	}
1582 	ISN_UNLOCK();
1583 	VNET_LIST_RUNLOCK_NOSLEEP();
1584 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1585 }
1586 
1587 /*
1588  * When a specific ICMP unreachable message is received and the
1589  * connection state is SYN-SENT, drop the connection.  This behavior
1590  * is controlled by the icmp_may_rst sysctl.
1591  */
1592 struct inpcb *
1593 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1594 {
1595 	struct tcpcb *tp;
1596 
1597 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1598 	INP_WLOCK_ASSERT(inp);
1599 
1600 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1601 	    (inp->inp_flags & INP_DROPPED))
1602 		return (inp);
1603 
1604 	tp = intotcpcb(inp);
1605 	if (tp->t_state != TCPS_SYN_SENT)
1606 		return (inp);
1607 
1608 	tp = tcp_drop(tp, errno);
1609 	if (tp != NULL)
1610 		return (inp);
1611 	else
1612 		return (NULL);
1613 }
1614 
1615 /*
1616  * When `need fragmentation' ICMP is received, update our idea of the MSS
1617  * based on the new value in the route.  Also nudge TCP to send something,
1618  * since we know the packet we just sent was dropped.
1619  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1620  */
1621 struct inpcb *
1622 tcp_mtudisc(struct inpcb *inp, int errno)
1623 {
1624 	struct tcpcb *tp;
1625 	struct socket *so;
1626 
1627 	INP_WLOCK_ASSERT(inp);
1628 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1629 	    (inp->inp_flags & INP_DROPPED))
1630 		return (inp);
1631 
1632 	tp = intotcpcb(inp);
1633 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1634 
1635 	tcp_mss_update(tp, -1, NULL, NULL);
1636 
1637 	so = inp->inp_socket;
1638 	SOCKBUF_LOCK(&so->so_snd);
1639 	/* If the mss is larger than the socket buffer, decrease the mss. */
1640 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1641 		tp->t_maxseg = so->so_snd.sb_hiwat;
1642 	SOCKBUF_UNLOCK(&so->so_snd);
1643 
1644 	TCPSTAT_INC(tcps_mturesent);
1645 	tp->t_rtttime = 0;
1646 	tp->snd_nxt = tp->snd_una;
1647 	tcp_free_sackholes(tp);
1648 	tp->snd_recover = tp->snd_max;
1649 	if (tp->t_flags & TF_SACK_PERMIT)
1650 		EXIT_FASTRECOVERY(tp);
1651 	tcp_output_send(tp);
1652 	return (inp);
1653 }
1654 
1655 /*
1656  * Look-up the routing entry to the peer of this inpcb.  If no route
1657  * is found and it cannot be allocated, then return 0.  This routine
1658  * is called by TCP routines that access the rmx structure and by
1659  * tcp_mss_update to get the peer/interface MTU.
1660  */
1661 u_long
1662 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1663 {
1664 	struct route sro;
1665 	struct sockaddr_in *dst;
1666 	struct ifnet *ifp;
1667 	u_long maxmtu = 0;
1668 
1669 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1670 
1671 	bzero(&sro, sizeof(sro));
1672 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1673 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1674 		dst->sin_family = AF_INET;
1675 		dst->sin_len = sizeof(*dst);
1676 		dst->sin_addr = inc->inc_faddr;
1677 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1678 	}
1679 	if (sro.ro_rt != NULL) {
1680 		ifp = sro.ro_rt->rt_ifp;
1681 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1682 			maxmtu = ifp->if_mtu;
1683 		else
1684 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1685 
1686 		/* Report additional interface capabilities. */
1687 		if (flags != NULL) {
1688 			if (ifp->if_capenable & IFCAP_TSO4 &&
1689 			    ifp->if_hwassist & CSUM_TSO)
1690 				*flags |= CSUM_TSO;
1691 		}
1692 		RTFREE(sro.ro_rt);
1693 	}
1694 	return (maxmtu);
1695 }
1696 
1697 #ifdef INET6
1698 u_long
1699 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1700 {
1701 	struct route_in6 sro6;
1702 	struct ifnet *ifp;
1703 	u_long maxmtu = 0;
1704 
1705 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1706 
1707 	bzero(&sro6, sizeof(sro6));
1708 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1709 		sro6.ro_dst.sin6_family = AF_INET6;
1710 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1711 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1712 		rtalloc_ign((struct route *)&sro6, 0);
1713 	}
1714 	if (sro6.ro_rt != NULL) {
1715 		ifp = sro6.ro_rt->rt_ifp;
1716 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1717 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1718 		else
1719 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1720 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1721 
1722 		/* Report additional interface capabilities. */
1723 		if (flags != NULL) {
1724 			if (ifp->if_capenable & IFCAP_TSO6 &&
1725 			    ifp->if_hwassist & CSUM_TSO)
1726 				*flags |= CSUM_TSO;
1727 		}
1728 		RTFREE(sro6.ro_rt);
1729 	}
1730 
1731 	return (maxmtu);
1732 }
1733 #endif /* INET6 */
1734 
1735 #ifdef IPSEC
1736 /* compute ESP/AH header size for TCP, including outer IP header. */
1737 size_t
1738 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1739 {
1740 	struct inpcb *inp;
1741 	struct mbuf *m;
1742 	size_t hdrsiz;
1743 	struct ip *ip;
1744 #ifdef INET6
1745 	struct ip6_hdr *ip6;
1746 #endif
1747 	struct tcphdr *th;
1748 
1749 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1750 		return (0);
1751 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1752 	if (!m)
1753 		return (0);
1754 
1755 #ifdef INET6
1756 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1757 		ip6 = mtod(m, struct ip6_hdr *);
1758 		th = (struct tcphdr *)(ip6 + 1);
1759 		m->m_pkthdr.len = m->m_len =
1760 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1761 		tcpip_fillheaders(inp, ip6, th);
1762 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1763 	} else
1764 #endif /* INET6 */
1765 	{
1766 		ip = mtod(m, struct ip *);
1767 		th = (struct tcphdr *)(ip + 1);
1768 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1769 		tcpip_fillheaders(inp, ip, th);
1770 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1771 	}
1772 
1773 	m_free(m);
1774 	return (hdrsiz);
1775 }
1776 #endif /* IPSEC */
1777 
1778 /*
1779  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1780  *
1781  * This code attempts to calculate the bandwidth-delay product as a
1782  * means of determining the optimal window size to maximize bandwidth,
1783  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1784  * routers.  This code also does a fairly good job keeping RTTs in check
1785  * across slow links like modems.  We implement an algorithm which is very
1786  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1787  * transmitter side of a TCP connection and so only effects the transmit
1788  * side of the connection.
1789  *
1790  * BACKGROUND:  TCP makes no provision for the management of buffer space
1791  * at the end points or at the intermediate routers and switches.  A TCP
1792  * stream, whether using NewReno or not, will eventually buffer as
1793  * many packets as it is able and the only reason this typically works is
1794  * due to the fairly small default buffers made available for a connection
1795  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1796  * scaling it is now fairly easy for even a single TCP connection to blow-out
1797  * all available buffer space not only on the local interface, but on
1798  * intermediate routers and switches as well.  NewReno makes a misguided
1799  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1800  * then backing off, then steadily increasing the window again until another
1801  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1802  * is only made worse as network loads increase and the idea of intentionally
1803  * blowing out network buffers is, frankly, a terrible way to manage network
1804  * resources.
1805  *
1806  * It is far better to limit the transmit window prior to the failure
1807  * condition being achieved.  There are two general ways to do this:  First
1808  * you can 'scan' through different transmit window sizes and locate the
1809  * point where the RTT stops increasing, indicating that you have filled the
1810  * pipe, then scan backwards until you note that RTT stops decreasing, then
1811  * repeat ad-infinitum.  This method works in principle but has severe
1812  * implementation issues due to RTT variances, timer granularity, and
1813  * instability in the algorithm which can lead to many false positives and
1814  * create oscillations as well as interact badly with other TCP streams
1815  * implementing the same algorithm.
1816  *
1817  * The second method is to limit the window to the bandwidth delay product
1818  * of the link.  This is the method we implement.  RTT variances and our
1819  * own manipulation of the congestion window, bwnd, can potentially
1820  * destabilize the algorithm.  For this reason we have to stabilize the
1821  * elements used to calculate the window.  We do this by using the minimum
1822  * observed RTT, the long term average of the observed bandwidth, and
1823  * by adding two segments worth of slop.  It isn't perfect but it is able
1824  * to react to changing conditions and gives us a very stable basis on
1825  * which to extend the algorithm.
1826  */
1827 void
1828 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1829 {
1830 	u_long bw;
1831 	u_long bwnd;
1832 	int save_ticks;
1833 
1834 	INP_WLOCK_ASSERT(tp->t_inpcb);
1835 
1836 	/*
1837 	 * If inflight_enable is disabled in the middle of a tcp connection,
1838 	 * make sure snd_bwnd is effectively disabled.
1839 	 */
1840 	if (V_tcp_inflight_enable == 0 ||
1841 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1842 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1843 		tp->snd_bandwidth = 0;
1844 		return;
1845 	}
1846 
1847 	/*
1848 	 * Figure out the bandwidth.  Due to the tick granularity this
1849 	 * is a very rough number and it MUST be averaged over a fairly
1850 	 * long period of time.  XXX we need to take into account a link
1851 	 * that is not using all available bandwidth, but for now our
1852 	 * slop will ramp us up if this case occurs and the bandwidth later
1853 	 * increases.
1854 	 *
1855 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1856 	 * effectively reset t_bw_rtttime for this case.
1857 	 */
1858 	save_ticks = ticks;
1859 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1860 		return;
1861 
1862 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1863 	    (save_ticks - tp->t_bw_rtttime);
1864 	tp->t_bw_rtttime = save_ticks;
1865 	tp->t_bw_rtseq = ack_seq;
1866 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1867 		return;
1868 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1869 
1870 	tp->snd_bandwidth = bw;
1871 
1872 	/*
1873 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1874 	 * segments.  The additional slop puts us squarely in the sweet
1875 	 * spot and also handles the bandwidth run-up case and stabilization.
1876 	 * Without the slop we could be locking ourselves into a lower
1877 	 * bandwidth.
1878 	 *
1879 	 * Situations Handled:
1880 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1881 	 *	    high speed LANs, allowing larger TCP buffers to be
1882 	 *	    specified, and also does a good job preventing
1883 	 *	    over-queueing of packets over choke points like modems
1884 	 *	    (at least for the transmit side).
1885 	 *
1886 	 *	(2) Is able to handle changing network loads (bandwidth
1887 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1888 	 *	    increases).
1889 	 *
1890 	 *	(3) Theoretically should stabilize in the face of multiple
1891 	 *	    connections implementing the same algorithm (this may need
1892 	 *	    a little work).
1893 	 *
1894 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1895 	 *	    be adjusted with a sysctl but typically only needs to be
1896 	 *	    on very slow connections.  A value no smaller then 5
1897 	 *	    should be used, but only reduce this default if you have
1898 	 *	    no other choice.
1899 	 */
1900 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1901 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1902 #undef USERTT
1903 
1904 	if (tcp_inflight_debug > 0) {
1905 		static int ltime;
1906 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1907 			ltime = ticks;
1908 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1909 			    tp,
1910 			    bw,
1911 			    tp->t_rttbest,
1912 			    tp->t_srtt,
1913 			    bwnd
1914 			);
1915 		}
1916 	}
1917 	if ((long)bwnd < V_tcp_inflight_min)
1918 		bwnd = V_tcp_inflight_min;
1919 	if (bwnd > V_tcp_inflight_max)
1920 		bwnd = V_tcp_inflight_max;
1921 	if ((long)bwnd < tp->t_maxseg * 2)
1922 		bwnd = tp->t_maxseg * 2;
1923 	tp->snd_bwnd = bwnd;
1924 }
1925 
1926 #ifdef TCP_SIGNATURE
1927 /*
1928  * Callback function invoked by m_apply() to digest TCP segment data
1929  * contained within an mbuf chain.
1930  */
1931 static int
1932 tcp_signature_apply(void *fstate, void *data, u_int len)
1933 {
1934 
1935 	MD5Update(fstate, (u_char *)data, len);
1936 	return (0);
1937 }
1938 
1939 /*
1940  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1941  *
1942  * Parameters:
1943  * m		pointer to head of mbuf chain
1944  * _unused
1945  * len		length of TCP segment data, excluding options
1946  * optlen	length of TCP segment options
1947  * buf		pointer to storage for computed MD5 digest
1948  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1949  *
1950  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1951  * When called from tcp_input(), we can be sure that th_sum has been
1952  * zeroed out and verified already.
1953  *
1954  * Return 0 if successful, otherwise return -1.
1955  *
1956  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1957  * search with the destination IP address, and a 'magic SPI' to be
1958  * determined by the application. This is hardcoded elsewhere to 1179
1959  * right now. Another branch of this code exists which uses the SPD to
1960  * specify per-application flows but it is unstable.
1961  */
1962 int
1963 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1964     u_char *buf, u_int direction)
1965 {
1966 	union sockaddr_union dst;
1967 	struct ippseudo ippseudo;
1968 	MD5_CTX ctx;
1969 	int doff;
1970 	struct ip *ip;
1971 	struct ipovly *ipovly;
1972 	struct secasvar *sav;
1973 	struct tcphdr *th;
1974 #ifdef INET6
1975 	struct ip6_hdr *ip6;
1976 	struct in6_addr in6;
1977 	char ip6buf[INET6_ADDRSTRLEN];
1978 	uint32_t plen;
1979 	uint16_t nhdr;
1980 #endif
1981 	u_short savecsum;
1982 
1983 	KASSERT(m != NULL, ("NULL mbuf chain"));
1984 	KASSERT(buf != NULL, ("NULL signature pointer"));
1985 
1986 	/* Extract the destination from the IP header in the mbuf. */
1987 	bzero(&dst, sizeof(union sockaddr_union));
1988 	ip = mtod(m, struct ip *);
1989 #ifdef INET6
1990 	ip6 = NULL;	/* Make the compiler happy. */
1991 #endif
1992 	switch (ip->ip_v) {
1993 	case IPVERSION:
1994 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1995 		dst.sa.sa_family = AF_INET;
1996 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1997 		    ip->ip_src : ip->ip_dst;
1998 		break;
1999 #ifdef INET6
2000 	case (IPV6_VERSION >> 4):
2001 		ip6 = mtod(m, struct ip6_hdr *);
2002 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2003 		dst.sa.sa_family = AF_INET6;
2004 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2005 		    ip6->ip6_src : ip6->ip6_dst;
2006 		break;
2007 #endif
2008 	default:
2009 		return (EINVAL);
2010 		/* NOTREACHED */
2011 		break;
2012 	}
2013 
2014 	/* Look up an SADB entry which matches the address of the peer. */
2015 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2016 	if (sav == NULL) {
2017 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2018 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2019 #ifdef INET6
2020 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2021 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2022 #endif
2023 			"(unsupported)"));
2024 		return (EINVAL);
2025 	}
2026 
2027 	MD5Init(&ctx);
2028 	/*
2029 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2030 	 *
2031 	 * XXX The ippseudo header MUST be digested in network byte order,
2032 	 * or else we'll fail the regression test. Assume all fields we've
2033 	 * been doing arithmetic on have been in host byte order.
2034 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2035 	 * tcp_output(), the underlying ip_len member has not yet been set.
2036 	 */
2037 	switch (ip->ip_v) {
2038 	case IPVERSION:
2039 		ipovly = (struct ipovly *)ip;
2040 		ippseudo.ippseudo_src = ipovly->ih_src;
2041 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2042 		ippseudo.ippseudo_pad = 0;
2043 		ippseudo.ippseudo_p = IPPROTO_TCP;
2044 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2045 		    optlen);
2046 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2047 
2048 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2049 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2050 		break;
2051 #ifdef INET6
2052 	/*
2053 	 * RFC 2385, 2.0  Proposal
2054 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2055 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2056 	 * extended next header value (to form 32 bits), and 32-bit segment
2057 	 * length.
2058 	 * Note: Upper-Layer Packet Length comes before Next Header.
2059 	 */
2060 	case (IPV6_VERSION >> 4):
2061 		in6 = ip6->ip6_src;
2062 		in6_clearscope(&in6);
2063 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2064 		in6 = ip6->ip6_dst;
2065 		in6_clearscope(&in6);
2066 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2067 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2068 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2069 		nhdr = 0;
2070 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2072 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2073 		nhdr = IPPROTO_TCP;
2074 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2075 
2076 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2077 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2078 		break;
2079 #endif
2080 	default:
2081 		return (EINVAL);
2082 		/* NOTREACHED */
2083 		break;
2084 	}
2085 
2086 
2087 	/*
2088 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2089 	 * The TCP checksum must be set to zero.
2090 	 */
2091 	savecsum = th->th_sum;
2092 	th->th_sum = 0;
2093 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2094 	th->th_sum = savecsum;
2095 
2096 	/*
2097 	 * Step 3: Update MD5 hash with TCP segment data.
2098 	 *         Use m_apply() to avoid an early m_pullup().
2099 	 */
2100 	if (len > 0)
2101 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2102 
2103 	/*
2104 	 * Step 4: Update MD5 hash with shared secret.
2105 	 */
2106 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2107 	MD5Final(buf, &ctx);
2108 
2109 	key_sa_recordxfer(sav, m);
2110 	KEY_FREESAV(&sav);
2111 	return (0);
2112 }
2113 #endif /* TCP_SIGNATURE */
2114 
2115 static int
2116 sysctl_drop(SYSCTL_HANDLER_ARGS)
2117 {
2118 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2119 	struct sockaddr_storage addrs[2];
2120 	struct inpcb *inp;
2121 	struct tcpcb *tp;
2122 	struct tcptw *tw;
2123 	struct sockaddr_in *fin, *lin;
2124 #ifdef INET6
2125 	struct sockaddr_in6 *fin6, *lin6;
2126 #endif
2127 	int error;
2128 
2129 	inp = NULL;
2130 	fin = lin = NULL;
2131 #ifdef INET6
2132 	fin6 = lin6 = NULL;
2133 #endif
2134 	error = 0;
2135 
2136 	if (req->oldptr != NULL || req->oldlen != 0)
2137 		return (EINVAL);
2138 	if (req->newptr == NULL)
2139 		return (EPERM);
2140 	if (req->newlen < sizeof(addrs))
2141 		return (ENOMEM);
2142 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2143 	if (error)
2144 		return (error);
2145 
2146 	switch (addrs[0].ss_family) {
2147 #ifdef INET6
2148 	case AF_INET6:
2149 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2150 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2151 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2152 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2153 			return (EINVAL);
2154 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2155 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2156 				return (EINVAL);
2157 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2158 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2159 			fin = (struct sockaddr_in *)&addrs[0];
2160 			lin = (struct sockaddr_in *)&addrs[1];
2161 			break;
2162 		}
2163 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2164 		if (error)
2165 			return (error);
2166 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2167 		if (error)
2168 			return (error);
2169 		break;
2170 #endif
2171 	case AF_INET:
2172 		fin = (struct sockaddr_in *)&addrs[0];
2173 		lin = (struct sockaddr_in *)&addrs[1];
2174 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2175 		    lin->sin_len != sizeof(struct sockaddr_in))
2176 			return (EINVAL);
2177 		break;
2178 	default:
2179 		return (EINVAL);
2180 	}
2181 	INP_INFO_WLOCK(&V_tcbinfo);
2182 	switch (addrs[0].ss_family) {
2183 #ifdef INET6
2184 	case AF_INET6:
2185 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2186 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2187 		    NULL);
2188 		break;
2189 #endif
2190 	case AF_INET:
2191 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2192 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2193 		break;
2194 	}
2195 	if (inp != NULL) {
2196 		INP_WLOCK(inp);
2197 		if (inp->inp_flags & INP_TIMEWAIT) {
2198 			/*
2199 			 * XXXRW: There currently exists a state where an
2200 			 * inpcb is present, but its timewait state has been
2201 			 * discarded.  For now, don't allow dropping of this
2202 			 * type of inpcb.
2203 			 */
2204 			tw = intotw(inp);
2205 			if (tw != NULL)
2206 				tcp_twclose(tw, 0);
2207 			else
2208 				INP_WUNLOCK(inp);
2209 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2210 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2211 			tp = intotcpcb(inp);
2212 			tp = tcp_drop(tp, ECONNABORTED);
2213 			if (tp != NULL)
2214 				INP_WUNLOCK(inp);
2215 		} else
2216 			INP_WUNLOCK(inp);
2217 	} else
2218 		error = ESRCH;
2219 	INP_INFO_WUNLOCK(&V_tcbinfo);
2220 	return (error);
2221 }
2222 
2223 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2224     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2225     0, sysctl_drop, "", "Drop TCP connection");
2226 
2227 /*
2228  * Generate a standardized TCP log line for use throughout the
2229  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2230  * allow use in the interrupt context.
2231  *
2232  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2233  * NB: The function may return NULL if memory allocation failed.
2234  *
2235  * Due to header inclusion and ordering limitations the struct ip
2236  * and ip6_hdr pointers have to be passed as void pointers.
2237  */
2238 char *
2239 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2240     const void *ip6hdr)
2241 {
2242 
2243 	/* Is logging enabled? */
2244 	if (tcp_log_in_vain == 0)
2245 		return (NULL);
2246 
2247 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2248 }
2249 
2250 char *
2251 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2252     const void *ip6hdr)
2253 {
2254 
2255 	/* Is logging enabled? */
2256 	if (tcp_log_debug == 0)
2257 		return (NULL);
2258 
2259 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2260 }
2261 
2262 static char *
2263 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2264     const void *ip6hdr)
2265 {
2266 	char *s, *sp;
2267 	size_t size;
2268 	struct ip *ip;
2269 #ifdef INET6
2270 	const struct ip6_hdr *ip6;
2271 
2272 	ip6 = (const struct ip6_hdr *)ip6hdr;
2273 #endif /* INET6 */
2274 	ip = (struct ip *)ip4hdr;
2275 
2276 	/*
2277 	 * The log line looks like this:
2278 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2279 	 */
2280 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2281 	    sizeof(PRINT_TH_FLAGS) + 1 +
2282 #ifdef INET6
2283 	    2 * INET6_ADDRSTRLEN;
2284 #else
2285 	    2 * INET_ADDRSTRLEN;
2286 #endif /* INET6 */
2287 
2288 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2289 	if (s == NULL)
2290 		return (NULL);
2291 
2292 	strcat(s, "TCP: [");
2293 	sp = s + strlen(s);
2294 
2295 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2296 		inet_ntoa_r(inc->inc_faddr, sp);
2297 		sp = s + strlen(s);
2298 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2299 		sp = s + strlen(s);
2300 		inet_ntoa_r(inc->inc_laddr, sp);
2301 		sp = s + strlen(s);
2302 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2303 #ifdef INET6
2304 	} else if (inc) {
2305 		ip6_sprintf(sp, &inc->inc6_faddr);
2306 		sp = s + strlen(s);
2307 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2308 		sp = s + strlen(s);
2309 		ip6_sprintf(sp, &inc->inc6_laddr);
2310 		sp = s + strlen(s);
2311 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2312 	} else if (ip6 && th) {
2313 		ip6_sprintf(sp, &ip6->ip6_src);
2314 		sp = s + strlen(s);
2315 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2316 		sp = s + strlen(s);
2317 		ip6_sprintf(sp, &ip6->ip6_dst);
2318 		sp = s + strlen(s);
2319 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2320 #endif /* INET6 */
2321 	} else if (ip && th) {
2322 		inet_ntoa_r(ip->ip_src, sp);
2323 		sp = s + strlen(s);
2324 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2325 		sp = s + strlen(s);
2326 		inet_ntoa_r(ip->ip_dst, sp);
2327 		sp = s + strlen(s);
2328 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2329 	} else {
2330 		free(s, M_TCPLOG);
2331 		return (NULL);
2332 	}
2333 	sp = s + strlen(s);
2334 	if (th)
2335 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2336 	if (*(s + size - 1) != '\0')
2337 		panic("%s: string too long", __func__);
2338 	return (s);
2339 }
2340