xref: /freebsd/sys/netinet/tcp_subr.c (revision 6356dba0b403daa023dec24559ab1f8e602e4f14)
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_mac.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.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 #include <sys/vimage.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/route.h>
63 #include <net/if.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 #endif /*IPSEC*/
107 
108 #include <machine/in_cksum.h>
109 #include <sys/md5.h>
110 
111 #include <security/mac/mac_framework.h>
112 
113 int	tcp_mssdflt = TCP_MSS;
114 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
115     &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
116 
117 #ifdef INET6
118 int	tcp_v6mssdflt = TCP6_MSS;
119 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
120     CTLFLAG_RW, &tcp_v6mssdflt , 0,
121     "Default TCP Maximum Segment Size for IPv6");
122 #endif
123 
124 /*
125  * Minimum MSS we accept and use. This prevents DoS attacks where
126  * we are forced to a ridiculous low MSS like 20 and send hundreds
127  * of packets instead of one. The effect scales with the available
128  * bandwidth and quickly saturates the CPU and network interface
129  * with packet generation and sending. Set to zero to disable MINMSS
130  * checking. This setting prevents us from sending too small packets.
131  */
132 int	tcp_minmss = TCP_MINMSS;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
134     &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
135 
136 int	tcp_do_rfc1323 = 1;
137 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
138     &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
139 
140 static int	tcp_log_debug = 0;
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
142     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
143 
144 static int	tcp_tcbhashsize = 0;
145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
146     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
147 
148 static int	do_tcpdrain = 1;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW,
150     &do_tcpdrain, 0,
151     "Enable tcp_drain routine for extra help when low on mbufs");
152 
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
154     &tcbinfo.ipi_count, 0, "Number of active PCBs");
155 
156 static int	icmp_may_rst = 1;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
158     &icmp_may_rst, 0,
159     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
160 
161 static int	tcp_isn_reseed_interval = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
163     &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
164 
165 /*
166  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
167  * 1024 exists only for debugging.  A good production default would be
168  * something like 6100.
169  */
170 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
171     "TCP inflight data limiting");
172 
173 static int	tcp_inflight_enable = 1;
174 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
175     &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
176 
177 static int	tcp_inflight_debug = 0;
178 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
179     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
180 
181 static int	tcp_inflight_rttthresh;
182 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
183     &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
184     "RTT threshold below which inflight will deactivate itself");
185 
186 static int	tcp_inflight_min = 6144;
187 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
188     &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
189 
190 static int	tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
191 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
192     &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
193 
194 static int	tcp_inflight_stab = 20;
195 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
196     &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
197 
198 uma_zone_t sack_hole_zone;
199 
200 static struct inpcb *tcp_notify(struct inpcb *, int);
201 static void	tcp_isn_tick(void *);
202 
203 /*
204  * Target size of TCP PCB hash tables. Must be a power of two.
205  *
206  * Note that this can be overridden by the kernel environment
207  * variable net.inet.tcp.tcbhashsize
208  */
209 #ifndef TCBHASHSIZE
210 #define TCBHASHSIZE	512
211 #endif
212 
213 /*
214  * XXX
215  * Callouts should be moved into struct tcp directly.  They are currently
216  * separate because the tcpcb structure is exported to userland for sysctl
217  * parsing purposes, which do not know about callouts.
218  */
219 struct tcpcb_mem {
220 	struct	tcpcb		tcb;
221 	struct	tcp_timer	tt;
222 };
223 
224 static uma_zone_t tcpcb_zone;
225 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
226 struct callout isn_callout;
227 static struct mtx isn_mtx;
228 
229 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
230 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
231 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
232 
233 /*
234  * TCP initialization.
235  */
236 static void
237 tcp_zone_change(void *tag)
238 {
239 
240 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
241 	uma_zone_set_max(tcpcb_zone, maxsockets);
242 	tcp_tw_zone_change();
243 }
244 
245 static int
246 tcp_inpcb_init(void *mem, int size, int flags)
247 {
248 	struct inpcb *inp = mem;
249 
250 	INP_LOCK_INIT(inp, "inp", "tcpinp");
251 	return (0);
252 }
253 
254 void
255 tcp_init(void)
256 {
257 
258 	int hashsize = TCBHASHSIZE;
259 	tcp_delacktime = TCPTV_DELACK;
260 	tcp_keepinit = TCPTV_KEEP_INIT;
261 	tcp_keepidle = TCPTV_KEEP_IDLE;
262 	tcp_keepintvl = TCPTV_KEEPINTVL;
263 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
264 	tcp_msl = TCPTV_MSL;
265 	tcp_rexmit_min = TCPTV_MIN;
266 	if (tcp_rexmit_min < 1)
267 		tcp_rexmit_min = 1;
268 	tcp_rexmit_slop = TCPTV_CPU_VAR;
269 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
270 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
271 
272 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
273 	LIST_INIT(&V_tcb);
274 	V_tcbinfo.ipi_listhead = &V_tcb;
275 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
276 	if (!powerof2(hashsize)) {
277 		printf("WARNING: TCB hash size not a power of 2\n");
278 		hashsize = 512; /* safe default */
279 	}
280 	tcp_tcbhashsize = hashsize;
281 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
282 	    &V_tcbinfo.ipi_hashmask);
283 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
284 	    &V_tcbinfo.ipi_porthashmask);
285 	V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
286 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
287 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
288 #ifdef INET6
289 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
290 #else /* INET6 */
291 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
292 #endif /* INET6 */
293 	if (max_protohdr < TCP_MINPROTOHDR)
294 		max_protohdr = TCP_MINPROTOHDR;
295 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
296 		panic("tcp_init");
297 #undef TCP_MINPROTOHDR
298 	/*
299 	 * These have to be type stable for the benefit of the timers.
300 	 */
301 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
302 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
303 	uma_zone_set_max(tcpcb_zone, maxsockets);
304 	tcp_tw_init();
305 	syncache_init();
306 	tcp_hc_init();
307 	tcp_reass_init();
308 	ISN_LOCK_INIT();
309 	callout_init(&isn_callout, CALLOUT_MPSAFE);
310 	tcp_isn_tick(NULL);
311 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
312 		SHUTDOWN_PRI_DEFAULT);
313 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
314 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
315 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
316 		EVENTHANDLER_PRI_ANY);
317 }
318 
319 void
320 tcp_fini(void *xtp)
321 {
322 
323 	callout_stop(&isn_callout);
324 }
325 
326 /*
327  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
328  * tcp_template used to store this data in mbufs, but we now recopy it out
329  * of the tcpcb each time to conserve mbufs.
330  */
331 void
332 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
333 {
334 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
335 
336 	INP_WLOCK_ASSERT(inp);
337 
338 #ifdef INET6
339 	if ((inp->inp_vflag & INP_IPV6) != 0) {
340 		struct ip6_hdr *ip6;
341 
342 		ip6 = (struct ip6_hdr *)ip_ptr;
343 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
344 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
345 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
346 			(IPV6_VERSION & IPV6_VERSION_MASK);
347 		ip6->ip6_nxt = IPPROTO_TCP;
348 		ip6->ip6_plen = sizeof(struct tcphdr);
349 		ip6->ip6_src = inp->in6p_laddr;
350 		ip6->ip6_dst = inp->in6p_faddr;
351 	} else
352 #endif
353 	{
354 		struct ip *ip;
355 
356 		ip = (struct ip *)ip_ptr;
357 		ip->ip_v = IPVERSION;
358 		ip->ip_hl = 5;
359 		ip->ip_tos = inp->inp_ip_tos;
360 		ip->ip_len = 0;
361 		ip->ip_id = 0;
362 		ip->ip_off = 0;
363 		ip->ip_ttl = inp->inp_ip_ttl;
364 		ip->ip_sum = 0;
365 		ip->ip_p = IPPROTO_TCP;
366 		ip->ip_src = inp->inp_laddr;
367 		ip->ip_dst = inp->inp_faddr;
368 	}
369 	th->th_sport = inp->inp_lport;
370 	th->th_dport = inp->inp_fport;
371 	th->th_seq = 0;
372 	th->th_ack = 0;
373 	th->th_x2 = 0;
374 	th->th_off = 5;
375 	th->th_flags = 0;
376 	th->th_win = 0;
377 	th->th_urp = 0;
378 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
379 }
380 
381 /*
382  * Create template to be used to send tcp packets on a connection.
383  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
384  * use for this function is in keepalives, which use tcp_respond.
385  */
386 struct tcptemp *
387 tcpip_maketemplate(struct inpcb *inp)
388 {
389 	struct tcptemp *t;
390 
391 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
392 	if (t == NULL)
393 		return (NULL);
394 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
395 	return (t);
396 }
397 
398 /*
399  * Send a single message to the TCP at address specified by
400  * the given TCP/IP header.  If m == NULL, then we make a copy
401  * of the tcpiphdr at ti and send directly to the addressed host.
402  * This is used to force keep alive messages out using the TCP
403  * template for a connection.  If flags are given then we send
404  * a message back to the TCP which originated the * segment ti,
405  * and discard the mbuf containing it and any other attached mbufs.
406  *
407  * In any case the ack and sequence number of the transmitted
408  * segment are as specified by the parameters.
409  *
410  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
411  */
412 void
413 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
414     tcp_seq ack, tcp_seq seq, int flags)
415 {
416 	int tlen;
417 	int win = 0;
418 	struct ip *ip;
419 	struct tcphdr *nth;
420 #ifdef INET6
421 	struct ip6_hdr *ip6;
422 	int isipv6;
423 #endif /* INET6 */
424 	int ipflags = 0;
425 	struct inpcb *inp;
426 
427 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
428 
429 #ifdef INET6
430 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
431 	ip6 = ipgen;
432 #endif /* INET6 */
433 	ip = ipgen;
434 
435 	if (tp != NULL) {
436 		inp = tp->t_inpcb;
437 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
438 		INP_WLOCK_ASSERT(inp);
439 	} else
440 		inp = NULL;
441 
442 	if (tp != NULL) {
443 		if (!(flags & TH_RST)) {
444 			win = sbspace(&inp->inp_socket->so_rcv);
445 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
446 				win = (long)TCP_MAXWIN << tp->rcv_scale;
447 		}
448 	}
449 	if (m == NULL) {
450 		m = m_gethdr(M_DONTWAIT, MT_DATA);
451 		if (m == NULL)
452 			return;
453 		tlen = 0;
454 		m->m_data += max_linkhdr;
455 #ifdef INET6
456 		if (isipv6) {
457 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
458 			      sizeof(struct ip6_hdr));
459 			ip6 = mtod(m, struct ip6_hdr *);
460 			nth = (struct tcphdr *)(ip6 + 1);
461 		} else
462 #endif /* INET6 */
463 	      {
464 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
465 		ip = mtod(m, struct ip *);
466 		nth = (struct tcphdr *)(ip + 1);
467 	      }
468 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
469 		flags = TH_ACK;
470 	} else {
471 		/*
472 		 *  reuse the mbuf.
473 		 * XXX MRT We inherrit the FIB, which is lucky.
474 		 */
475 		m_freem(m->m_next);
476 		m->m_next = NULL;
477 		m->m_data = (caddr_t)ipgen;
478 		/* m_len is set later */
479 		tlen = 0;
480 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
481 #ifdef INET6
482 		if (isipv6) {
483 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
484 			nth = (struct tcphdr *)(ip6 + 1);
485 		} else
486 #endif /* INET6 */
487 	      {
488 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
489 		nth = (struct tcphdr *)(ip + 1);
490 	      }
491 		if (th != nth) {
492 			/*
493 			 * this is usually a case when an extension header
494 			 * exists between the IPv6 header and the
495 			 * TCP header.
496 			 */
497 			nth->th_sport = th->th_sport;
498 			nth->th_dport = th->th_dport;
499 		}
500 		xchg(nth->th_dport, nth->th_sport, n_short);
501 #undef xchg
502 	}
503 #ifdef INET6
504 	if (isipv6) {
505 		ip6->ip6_flow = 0;
506 		ip6->ip6_vfc = IPV6_VERSION;
507 		ip6->ip6_nxt = IPPROTO_TCP;
508 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
509 						tlen));
510 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
511 	} else
512 #endif
513 	{
514 		tlen += sizeof (struct tcpiphdr);
515 		ip->ip_len = tlen;
516 		ip->ip_ttl = V_ip_defttl;
517 		if (V_path_mtu_discovery)
518 			ip->ip_off |= IP_DF;
519 	}
520 	m->m_len = tlen;
521 	m->m_pkthdr.len = tlen;
522 	m->m_pkthdr.rcvif = NULL;
523 #ifdef MAC
524 	if (inp != NULL) {
525 		/*
526 		 * Packet is associated with a socket, so allow the
527 		 * label of the response to reflect the socket label.
528 		 */
529 		INP_WLOCK_ASSERT(inp);
530 		mac_inpcb_create_mbuf(inp, m);
531 	} else {
532 		/*
533 		 * Packet is not associated with a socket, so possibly
534 		 * update the label in place.
535 		 */
536 		mac_netinet_tcp_reply(m);
537 	}
538 #endif
539 	nth->th_seq = htonl(seq);
540 	nth->th_ack = htonl(ack);
541 	nth->th_x2 = 0;
542 	nth->th_off = sizeof (struct tcphdr) >> 2;
543 	nth->th_flags = flags;
544 	if (tp != NULL)
545 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
546 	else
547 		nth->th_win = htons((u_short)win);
548 	nth->th_urp = 0;
549 #ifdef INET6
550 	if (isipv6) {
551 		nth->th_sum = 0;
552 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
553 					sizeof(struct ip6_hdr),
554 					tlen - sizeof(struct ip6_hdr));
555 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
556 		    NULL, NULL);
557 	} else
558 #endif /* INET6 */
559 	{
560 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
561 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
562 		m->m_pkthdr.csum_flags = CSUM_TCP;
563 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
564 	}
565 #ifdef TCPDEBUG
566 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
567 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
568 #endif
569 #ifdef INET6
570 	if (isipv6)
571 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
572 	else
573 #endif /* INET6 */
574 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
575 }
576 
577 /*
578  * Create a new TCP control block, making an
579  * empty reassembly queue and hooking it to the argument
580  * protocol control block.  The `inp' parameter must have
581  * come from the zone allocator set up in tcp_init().
582  */
583 struct tcpcb *
584 tcp_newtcpcb(struct inpcb *inp)
585 {
586 	struct tcpcb_mem *tm;
587 	struct tcpcb *tp;
588 #ifdef INET6
589 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
590 #endif /* INET6 */
591 
592 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
593 	if (tm == NULL)
594 		return (NULL);
595 	tp = &tm->tcb;
596 	tp->t_timers = &tm->tt;
597 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
598 	tp->t_maxseg = tp->t_maxopd =
599 #ifdef INET6
600 		isipv6 ? V_tcp_v6mssdflt :
601 #endif /* INET6 */
602 		V_tcp_mssdflt;
603 
604 	/* Set up our timeouts. */
605 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
606 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
607 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
608 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
609 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
610 
611 	if (V_tcp_do_rfc1323)
612 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
613 	if (V_tcp_do_sack)
614 		tp->t_flags |= TF_SACK_PERMIT;
615 	TAILQ_INIT(&tp->snd_holes);
616 	tp->t_inpcb = inp;	/* XXX */
617 	/*
618 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
619 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
620 	 * reasonable initial retransmit time.
621 	 */
622 	tp->t_srtt = TCPTV_SRTTBASE;
623 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
624 	tp->t_rttmin = tcp_rexmit_min;
625 	tp->t_rxtcur = TCPTV_RTOBASE;
626 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
627 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
628 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
629 	tp->t_rcvtime = ticks;
630 	tp->t_bw_rtttime = ticks;
631 	/*
632 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
633 	 * because the socket may be bound to an IPv6 wildcard address,
634 	 * which may match an IPv4-mapped IPv6 address.
635 	 */
636 	inp->inp_ip_ttl = V_ip_defttl;
637 	inp->inp_ppcb = tp;
638 	return (tp);		/* XXX */
639 }
640 
641 /*
642  * Drop a TCP connection, reporting
643  * the specified error.  If connection is synchronized,
644  * then send a RST to peer.
645  */
646 struct tcpcb *
647 tcp_drop(struct tcpcb *tp, int errno)
648 {
649 	struct socket *so = tp->t_inpcb->inp_socket;
650 
651 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
652 	INP_WLOCK_ASSERT(tp->t_inpcb);
653 
654 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
655 		tp->t_state = TCPS_CLOSED;
656 		(void) tcp_output_reset(tp);
657 		V_tcpstat.tcps_drops++;
658 	} else
659 		V_tcpstat.tcps_conndrops++;
660 	if (errno == ETIMEDOUT && tp->t_softerror)
661 		errno = tp->t_softerror;
662 	so->so_error = errno;
663 	return (tcp_close(tp));
664 }
665 
666 void
667 tcp_discardcb(struct tcpcb *tp)
668 {
669 	struct tseg_qent *q;
670 	struct inpcb *inp = tp->t_inpcb;
671 	struct socket *so = inp->inp_socket;
672 #ifdef INET6
673 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
674 #endif /* INET6 */
675 
676 	INP_WLOCK_ASSERT(inp);
677 
678 	/*
679 	 * Make sure that all of our timers are stopped before we
680 	 * delete the PCB.
681 	 */
682 	callout_stop(&tp->t_timers->tt_rexmt);
683 	callout_stop(&tp->t_timers->tt_persist);
684 	callout_stop(&tp->t_timers->tt_keep);
685 	callout_stop(&tp->t_timers->tt_2msl);
686 	callout_stop(&tp->t_timers->tt_delack);
687 
688 	/*
689 	 * If we got enough samples through the srtt filter,
690 	 * save the rtt and rttvar in the routing entry.
691 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
692 	 * 4 samples is enough for the srtt filter to converge
693 	 * to within enough % of the correct value; fewer samples
694 	 * and we could save a bogus rtt. The danger is not high
695 	 * as tcp quickly recovers from everything.
696 	 * XXX: Works very well but needs some more statistics!
697 	 */
698 	if (tp->t_rttupdated >= 4) {
699 		struct hc_metrics_lite metrics;
700 		u_long ssthresh;
701 
702 		bzero(&metrics, sizeof(metrics));
703 		/*
704 		 * Update the ssthresh always when the conditions below
705 		 * are satisfied. This gives us better new start value
706 		 * for the congestion avoidance for new connections.
707 		 * ssthresh is only set if packet loss occured on a session.
708 		 *
709 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
710 		 * being torn down.  Ideally this code would not use 'so'.
711 		 */
712 		ssthresh = tp->snd_ssthresh;
713 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
714 			/*
715 			 * convert the limit from user data bytes to
716 			 * packets then to packet data bytes.
717 			 */
718 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
719 			if (ssthresh < 2)
720 				ssthresh = 2;
721 			ssthresh *= (u_long)(tp->t_maxseg +
722 #ifdef INET6
723 				      (isipv6 ? sizeof (struct ip6_hdr) +
724 					       sizeof (struct tcphdr) :
725 #endif
726 				       sizeof (struct tcpiphdr)
727 #ifdef INET6
728 				       )
729 #endif
730 				      );
731 		} else
732 			ssthresh = 0;
733 		metrics.rmx_ssthresh = ssthresh;
734 
735 		metrics.rmx_rtt = tp->t_srtt;
736 		metrics.rmx_rttvar = tp->t_rttvar;
737 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
738 		metrics.rmx_bandwidth = tp->snd_bandwidth;
739 		metrics.rmx_cwnd = tp->snd_cwnd;
740 		metrics.rmx_sendpipe = 0;
741 		metrics.rmx_recvpipe = 0;
742 
743 		tcp_hc_update(&inp->inp_inc, &metrics);
744 	}
745 
746 	/* free the reassembly queue, if any */
747 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
748 		LIST_REMOVE(q, tqe_q);
749 		m_freem(q->tqe_m);
750 		uma_zfree(tcp_reass_zone, q);
751 		tp->t_segqlen--;
752 		V_tcp_reass_qsize--;
753 	}
754 	/* Disconnect offload device, if any. */
755 	tcp_offload_detach(tp);
756 
757 	tcp_free_sackholes(tp);
758 	inp->inp_ppcb = NULL;
759 	tp->t_inpcb = NULL;
760 	uma_zfree(tcpcb_zone, tp);
761 }
762 
763 /*
764  * Attempt to close a TCP control block, marking it as dropped, and freeing
765  * the socket if we hold the only reference.
766  */
767 struct tcpcb *
768 tcp_close(struct tcpcb *tp)
769 {
770 	struct inpcb *inp = tp->t_inpcb;
771 	struct socket *so;
772 
773 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
774 	INP_WLOCK_ASSERT(inp);
775 
776 	/* Notify any offload devices of listener close */
777 	if (tp->t_state == TCPS_LISTEN)
778 		tcp_offload_listen_close(tp);
779 	in_pcbdrop(inp);
780 	V_tcpstat.tcps_closed++;
781 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
782 	so = inp->inp_socket;
783 	soisdisconnected(so);
784 	if (inp->inp_vflag & INP_SOCKREF) {
785 		KASSERT(so->so_state & SS_PROTOREF,
786 		    ("tcp_close: !SS_PROTOREF"));
787 		inp->inp_vflag &= ~INP_SOCKREF;
788 		INP_WUNLOCK(inp);
789 		ACCEPT_LOCK();
790 		SOCK_LOCK(so);
791 		so->so_state &= ~SS_PROTOREF;
792 		sofree(so);
793 		return (NULL);
794 	}
795 	return (tp);
796 }
797 
798 void
799 tcp_drain(void)
800 {
801 
802 	if (do_tcpdrain) {
803 		struct inpcb *inpb;
804 		struct tcpcb *tcpb;
805 		struct tseg_qent *te;
806 
807 	/*
808 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
809 	 * if there is one...
810 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
811 	 *      reassembly queue should be flushed, but in a situation
812 	 *	where we're really low on mbufs, this is potentially
813 	 *	usefull.
814 	 */
815 		INP_INFO_RLOCK(&V_tcbinfo);
816 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
817 			if (inpb->inp_vflag & INP_TIMEWAIT)
818 				continue;
819 			INP_WLOCK(inpb);
820 			if ((tcpb = intotcpcb(inpb)) != NULL) {
821 				while ((te = LIST_FIRST(&tcpb->t_segq))
822 			            != NULL) {
823 					LIST_REMOVE(te, tqe_q);
824 					m_freem(te->tqe_m);
825 					uma_zfree(tcp_reass_zone, te);
826 					tcpb->t_segqlen--;
827 					V_tcp_reass_qsize--;
828 				}
829 				tcp_clean_sackreport(tcpb);
830 			}
831 			INP_WUNLOCK(inpb);
832 		}
833 		INP_INFO_RUNLOCK(&V_tcbinfo);
834 	}
835 }
836 
837 /*
838  * Notify a tcp user of an asynchronous error;
839  * store error as soft error, but wake up user
840  * (for now, won't do anything until can select for soft error).
841  *
842  * Do not wake up user since there currently is no mechanism for
843  * reporting soft errors (yet - a kqueue filter may be added).
844  */
845 static struct inpcb *
846 tcp_notify(struct inpcb *inp, int error)
847 {
848 	struct tcpcb *tp;
849 
850 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
851 	INP_WLOCK_ASSERT(inp);
852 
853 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
854 	    (inp->inp_vflag & INP_DROPPED))
855 		return (inp);
856 
857 	tp = intotcpcb(inp);
858 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
859 
860 	/*
861 	 * Ignore some errors if we are hooked up.
862 	 * If connection hasn't completed, has retransmitted several times,
863 	 * and receives a second error, give up now.  This is better
864 	 * than waiting a long time to establish a connection that
865 	 * can never complete.
866 	 */
867 	if (tp->t_state == TCPS_ESTABLISHED &&
868 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
869 	     error == EHOSTDOWN)) {
870 		return (inp);
871 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
872 	    tp->t_softerror) {
873 		tp = tcp_drop(tp, error);
874 		if (tp != NULL)
875 			return (inp);
876 		else
877 			return (NULL);
878 	} else {
879 		tp->t_softerror = error;
880 		return (inp);
881 	}
882 #if 0
883 	wakeup( &so->so_timeo);
884 	sorwakeup(so);
885 	sowwakeup(so);
886 #endif
887 }
888 
889 static int
890 tcp_pcblist(SYSCTL_HANDLER_ARGS)
891 {
892 	int error, i, m, n, pcb_count;
893 	struct inpcb *inp, **inp_list;
894 	inp_gen_t gencnt;
895 	struct xinpgen xig;
896 
897 	/*
898 	 * The process of preparing the TCB list is too time-consuming and
899 	 * resource-intensive to repeat twice on every request.
900 	 */
901 	if (req->oldptr == NULL) {
902 		m = syncache_pcbcount();
903 		n = V_tcbinfo.ipi_count;
904 		req->oldidx = 2 * (sizeof xig)
905 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
906 		return (0);
907 	}
908 
909 	if (req->newptr != NULL)
910 		return (EPERM);
911 
912 	/*
913 	 * OK, now we're committed to doing something.
914 	 */
915 	INP_INFO_RLOCK(&V_tcbinfo);
916 	gencnt = V_tcbinfo.ipi_gencnt;
917 	n = V_tcbinfo.ipi_count;
918 	INP_INFO_RUNLOCK(&V_tcbinfo);
919 
920 	m = syncache_pcbcount();
921 
922 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
923 		+ (n + m) * sizeof(struct xtcpcb));
924 	if (error != 0)
925 		return (error);
926 
927 	xig.xig_len = sizeof xig;
928 	xig.xig_count = n + m;
929 	xig.xig_gen = gencnt;
930 	xig.xig_sogen = so_gencnt;
931 	error = SYSCTL_OUT(req, &xig, sizeof xig);
932 	if (error)
933 		return (error);
934 
935 	error = syncache_pcblist(req, m, &pcb_count);
936 	if (error)
937 		return (error);
938 
939 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
940 	if (inp_list == NULL)
941 		return (ENOMEM);
942 
943 	INP_INFO_RLOCK(&V_tcbinfo);
944 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
945 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
946 		INP_RLOCK(inp);
947 		if (inp->inp_gencnt <= gencnt) {
948 			/*
949 			 * XXX: This use of cr_cansee(), introduced with
950 			 * TCP state changes, is not quite right, but for
951 			 * now, better than nothing.
952 			 */
953 			if (inp->inp_vflag & INP_TIMEWAIT) {
954 				if (intotw(inp) != NULL)
955 					error = cr_cansee(req->td->td_ucred,
956 					    intotw(inp)->tw_cred);
957 				else
958 					error = EINVAL;	/* Skip this inp. */
959 			} else
960 				error = cr_canseesocket(req->td->td_ucred,
961 				    inp->inp_socket);
962 			if (error == 0)
963 				inp_list[i++] = inp;
964 		}
965 		INP_RUNLOCK(inp);
966 	}
967 	INP_INFO_RUNLOCK(&V_tcbinfo);
968 	n = i;
969 
970 	error = 0;
971 	for (i = 0; i < n; i++) {
972 		inp = inp_list[i];
973 		INP_RLOCK(inp);
974 		if (inp->inp_gencnt <= gencnt) {
975 			struct xtcpcb xt;
976 			void *inp_ppcb;
977 
978 			bzero(&xt, sizeof(xt));
979 			xt.xt_len = sizeof xt;
980 			/* XXX should avoid extra copy */
981 			bcopy(inp, &xt.xt_inp, sizeof *inp);
982 			inp_ppcb = inp->inp_ppcb;
983 			if (inp_ppcb == NULL)
984 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
985 			else if (inp->inp_vflag & INP_TIMEWAIT) {
986 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
987 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
988 			} else
989 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
990 			if (inp->inp_socket != NULL)
991 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
992 			else {
993 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
994 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
995 			}
996 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
997 			INP_RUNLOCK(inp);
998 			error = SYSCTL_OUT(req, &xt, sizeof xt);
999 		} else
1000 			INP_RUNLOCK(inp);
1001 
1002 	}
1003 	if (!error) {
1004 		/*
1005 		 * Give the user an updated idea of our state.
1006 		 * If the generation differs from what we told
1007 		 * her before, she knows that something happened
1008 		 * while we were processing this request, and it
1009 		 * might be necessary to retry.
1010 		 */
1011 		INP_INFO_RLOCK(&V_tcbinfo);
1012 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1013 		xig.xig_sogen = so_gencnt;
1014 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1015 		INP_INFO_RUNLOCK(&V_tcbinfo);
1016 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1017 	}
1018 	free(inp_list, M_TEMP);
1019 	return (error);
1020 }
1021 
1022 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1023     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1024 
1025 static int
1026 tcp_getcred(SYSCTL_HANDLER_ARGS)
1027 {
1028 	struct xucred xuc;
1029 	struct sockaddr_in addrs[2];
1030 	struct inpcb *inp;
1031 	int error;
1032 
1033 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1034 	if (error)
1035 		return (error);
1036 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1037 	if (error)
1038 		return (error);
1039 	INP_INFO_RLOCK(&V_tcbinfo);
1040 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1041 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1042 	if (inp != NULL) {
1043 		INP_RLOCK(inp);
1044 		INP_INFO_RUNLOCK(&V_tcbinfo);
1045 		if (inp->inp_socket == NULL)
1046 			error = ENOENT;
1047 		if (error == 0)
1048 			error = cr_canseesocket(req->td->td_ucred,
1049 			    inp->inp_socket);
1050 		if (error == 0)
1051 			cru2x(inp->inp_socket->so_cred, &xuc);
1052 		INP_RUNLOCK(inp);
1053 	} else {
1054 		INP_INFO_RUNLOCK(&V_tcbinfo);
1055 		error = ENOENT;
1056 	}
1057 	if (error == 0)
1058 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1059 	return (error);
1060 }
1061 
1062 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1063     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1064     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1065 
1066 #ifdef INET6
1067 static int
1068 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1069 {
1070 	struct xucred xuc;
1071 	struct sockaddr_in6 addrs[2];
1072 	struct inpcb *inp;
1073 	int error, mapped = 0;
1074 
1075 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1076 	if (error)
1077 		return (error);
1078 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1079 	if (error)
1080 		return (error);
1081 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1082 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1083 		return (error);
1084 	}
1085 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1086 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1087 			mapped = 1;
1088 		else
1089 			return (EINVAL);
1090 	}
1091 
1092 	INP_INFO_RLOCK(&V_tcbinfo);
1093 	if (mapped == 1)
1094 		inp = in_pcblookup_hash(&V_tcbinfo,
1095 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1096 			addrs[1].sin6_port,
1097 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1098 			addrs[0].sin6_port,
1099 			0, NULL);
1100 	else
1101 		inp = in6_pcblookup_hash(&V_tcbinfo,
1102 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1103 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1104 	if (inp != NULL) {
1105 		INP_RLOCK(inp);
1106 		INP_INFO_RUNLOCK(&V_tcbinfo);
1107 		if (inp->inp_socket == NULL)
1108 			error = ENOENT;
1109 		if (error == 0)
1110 			error = cr_canseesocket(req->td->td_ucred,
1111 			    inp->inp_socket);
1112 		if (error == 0)
1113 			cru2x(inp->inp_socket->so_cred, &xuc);
1114 		INP_RUNLOCK(inp);
1115 	} else {
1116 		INP_INFO_RUNLOCK(&V_tcbinfo);
1117 		error = ENOENT;
1118 	}
1119 	if (error == 0)
1120 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1121 	return (error);
1122 }
1123 
1124 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1125     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1126     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1127 #endif
1128 
1129 
1130 void
1131 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1132 {
1133 	struct ip *ip = vip;
1134 	struct tcphdr *th;
1135 	struct in_addr faddr;
1136 	struct inpcb *inp;
1137 	struct tcpcb *tp;
1138 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1139 	struct icmp *icp;
1140 	struct in_conninfo inc;
1141 	tcp_seq icmp_tcp_seq;
1142 	int mtu;
1143 
1144 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1145 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1146 		return;
1147 
1148 	if (cmd == PRC_MSGSIZE)
1149 		notify = tcp_mtudisc;
1150 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1151 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1152 		notify = tcp_drop_syn_sent;
1153 	/*
1154 	 * Redirects don't need to be handled up here.
1155 	 */
1156 	else if (PRC_IS_REDIRECT(cmd))
1157 		return;
1158 	/*
1159 	 * Source quench is depreciated.
1160 	 */
1161 	else if (cmd == PRC_QUENCH)
1162 		return;
1163 	/*
1164 	 * Hostdead is ugly because it goes linearly through all PCBs.
1165 	 * XXX: We never get this from ICMP, otherwise it makes an
1166 	 * excellent DoS attack on machines with many connections.
1167 	 */
1168 	else if (cmd == PRC_HOSTDEAD)
1169 		ip = NULL;
1170 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1171 		return;
1172 	if (ip != NULL) {
1173 		icp = (struct icmp *)((caddr_t)ip
1174 				      - offsetof(struct icmp, icmp_ip));
1175 		th = (struct tcphdr *)((caddr_t)ip
1176 				       + (ip->ip_hl << 2));
1177 		INP_INFO_WLOCK(&V_tcbinfo);
1178 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1179 		    ip->ip_src, th->th_sport, 0, NULL);
1180 		if (inp != NULL)  {
1181 			INP_WLOCK(inp);
1182 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1183 			    !(inp->inp_vflag & INP_DROPPED) &&
1184 			    !(inp->inp_socket == NULL)) {
1185 				icmp_tcp_seq = htonl(th->th_seq);
1186 				tp = intotcpcb(inp);
1187 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1188 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1189 					if (cmd == PRC_MSGSIZE) {
1190 					    /*
1191 					     * MTU discovery:
1192 					     * If we got a needfrag set the MTU
1193 					     * in the route to the suggested new
1194 					     * value (if given) and then notify.
1195 					     */
1196 					    bzero(&inc, sizeof(inc));
1197 					    inc.inc_flags = 0;	/* IPv4 */
1198 					    inc.inc_faddr = faddr;
1199 					    inc.inc_fibnum =
1200 						inp->inp_inc.inc_fibnum;
1201 
1202 					    mtu = ntohs(icp->icmp_nextmtu);
1203 					    /*
1204 					     * If no alternative MTU was
1205 					     * proposed, try the next smaller
1206 					     * one.  ip->ip_len has already
1207 					     * been swapped in icmp_input().
1208 					     */
1209 					    if (!mtu)
1210 						mtu = ip_next_mtu(ip->ip_len,
1211 						 1);
1212 					    if (mtu < max(296, V_tcp_minmss
1213 						 + sizeof(struct tcpiphdr)))
1214 						mtu = 0;
1215 					    if (!mtu)
1216 						mtu = V_tcp_mssdflt
1217 						 + sizeof(struct tcpiphdr);
1218 					    /*
1219 					     * Only cache the the MTU if it
1220 					     * is smaller than the interface
1221 					     * or route MTU.  tcp_mtudisc()
1222 					     * will do right thing by itself.
1223 					     */
1224 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1225 						tcp_hc_updatemtu(&inc, mtu);
1226 					}
1227 
1228 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1229 				}
1230 			}
1231 			if (inp != NULL)
1232 				INP_WUNLOCK(inp);
1233 		} else {
1234 			inc.inc_fport = th->th_dport;
1235 			inc.inc_lport = th->th_sport;
1236 			inc.inc_faddr = faddr;
1237 			inc.inc_laddr = ip->ip_src;
1238 #ifdef INET6
1239 			inc.inc_isipv6 = 0;
1240 #endif
1241 			syncache_unreach(&inc, th);
1242 		}
1243 		INP_INFO_WUNLOCK(&V_tcbinfo);
1244 	} else
1245 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1246 }
1247 
1248 #ifdef INET6
1249 void
1250 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1251 {
1252 	struct tcphdr th;
1253 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1254 	struct ip6_hdr *ip6;
1255 	struct mbuf *m;
1256 	struct ip6ctlparam *ip6cp = NULL;
1257 	const struct sockaddr_in6 *sa6_src = NULL;
1258 	int off;
1259 	struct tcp_portonly {
1260 		u_int16_t th_sport;
1261 		u_int16_t th_dport;
1262 	} *thp;
1263 
1264 	if (sa->sa_family != AF_INET6 ||
1265 	    sa->sa_len != sizeof(struct sockaddr_in6))
1266 		return;
1267 
1268 	if (cmd == PRC_MSGSIZE)
1269 		notify = tcp_mtudisc;
1270 	else if (!PRC_IS_REDIRECT(cmd) &&
1271 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1272 		return;
1273 	/* Source quench is depreciated. */
1274 	else if (cmd == PRC_QUENCH)
1275 		return;
1276 
1277 	/* if the parameter is from icmp6, decode it. */
1278 	if (d != NULL) {
1279 		ip6cp = (struct ip6ctlparam *)d;
1280 		m = ip6cp->ip6c_m;
1281 		ip6 = ip6cp->ip6c_ip6;
1282 		off = ip6cp->ip6c_off;
1283 		sa6_src = ip6cp->ip6c_src;
1284 	} else {
1285 		m = NULL;
1286 		ip6 = NULL;
1287 		off = 0;	/* fool gcc */
1288 		sa6_src = &sa6_any;
1289 	}
1290 
1291 	if (ip6 != NULL) {
1292 		struct in_conninfo inc;
1293 		/*
1294 		 * XXX: We assume that when IPV6 is non NULL,
1295 		 * M and OFF are valid.
1296 		 */
1297 
1298 		/* check if we can safely examine src and dst ports */
1299 		if (m->m_pkthdr.len < off + sizeof(*thp))
1300 			return;
1301 
1302 		bzero(&th, sizeof(th));
1303 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1304 
1305 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1306 		    (struct sockaddr *)ip6cp->ip6c_src,
1307 		    th.th_sport, cmd, NULL, notify);
1308 
1309 		inc.inc_fport = th.th_dport;
1310 		inc.inc_lport = th.th_sport;
1311 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1312 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1313 		inc.inc_isipv6 = 1;
1314 		INP_INFO_WLOCK(&V_tcbinfo);
1315 		syncache_unreach(&inc, &th);
1316 		INP_INFO_WUNLOCK(&V_tcbinfo);
1317 	} else
1318 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1319 			      0, cmd, NULL, notify);
1320 }
1321 #endif /* INET6 */
1322 
1323 
1324 /*
1325  * Following is where TCP initial sequence number generation occurs.
1326  *
1327  * There are two places where we must use initial sequence numbers:
1328  * 1.  In SYN-ACK packets.
1329  * 2.  In SYN packets.
1330  *
1331  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1332  * tcp_syncache.c for details.
1333  *
1334  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1335  * depends on this property.  In addition, these ISNs should be
1336  * unguessable so as to prevent connection hijacking.  To satisfy
1337  * the requirements of this situation, the algorithm outlined in
1338  * RFC 1948 is used, with only small modifications.
1339  *
1340  * Implementation details:
1341  *
1342  * Time is based off the system timer, and is corrected so that it
1343  * increases by one megabyte per second.  This allows for proper
1344  * recycling on high speed LANs while still leaving over an hour
1345  * before rollover.
1346  *
1347  * As reading the *exact* system time is too expensive to be done
1348  * whenever setting up a TCP connection, we increment the time
1349  * offset in two ways.  First, a small random positive increment
1350  * is added to isn_offset for each connection that is set up.
1351  * Second, the function tcp_isn_tick fires once per clock tick
1352  * and increments isn_offset as necessary so that sequence numbers
1353  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1354  * random positive increments serve only to ensure that the same
1355  * exact sequence number is never sent out twice (as could otherwise
1356  * happen when a port is recycled in less than the system tick
1357  * interval.)
1358  *
1359  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1360  * between seeding of isn_secret.  This is normally set to zero,
1361  * as reseeding should not be necessary.
1362  *
1363  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1364  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1365  * general, this means holding an exclusive (write) lock.
1366  */
1367 
1368 #define ISN_BYTES_PER_SECOND 1048576
1369 #define ISN_STATIC_INCREMENT 4096
1370 #define ISN_RANDOM_INCREMENT (4096 - 1)
1371 
1372 static u_char isn_secret[32];
1373 static int isn_last_reseed;
1374 static u_int32_t isn_offset, isn_offset_old;
1375 static MD5_CTX isn_ctx;
1376 
1377 tcp_seq
1378 tcp_new_isn(struct tcpcb *tp)
1379 {
1380 	u_int32_t md5_buffer[4];
1381 	tcp_seq new_isn;
1382 
1383 	INP_WLOCK_ASSERT(tp->t_inpcb);
1384 
1385 	ISN_LOCK();
1386 	/* Seed if this is the first use, reseed if requested. */
1387 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1388 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1389 		< (u_int)ticks))) {
1390 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1391 		V_isn_last_reseed = ticks;
1392 	}
1393 
1394 	/* Compute the md5 hash and return the ISN. */
1395 	MD5Init(&V_isn_ctx);
1396 	MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1397 	MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1398 #ifdef INET6
1399 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1400 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1401 			  sizeof(struct in6_addr));
1402 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1403 			  sizeof(struct in6_addr));
1404 	} else
1405 #endif
1406 	{
1407 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1408 			  sizeof(struct in_addr));
1409 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1410 			  sizeof(struct in_addr));
1411 	}
1412 	MD5Update(&V_isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1413 	MD5Final((u_char *) &md5_buffer, &V_isn_ctx);
1414 	new_isn = (tcp_seq) md5_buffer[0];
1415 	V_isn_offset += ISN_STATIC_INCREMENT +
1416 		(arc4random() & ISN_RANDOM_INCREMENT);
1417 	new_isn += V_isn_offset;
1418 	ISN_UNLOCK();
1419 	return (new_isn);
1420 }
1421 
1422 /*
1423  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1424  * to keep time flowing at a relatively constant rate.  If the random
1425  * increments have already pushed us past the projected offset, do nothing.
1426  */
1427 static void
1428 tcp_isn_tick(void *xtp)
1429 {
1430 	u_int32_t projected_offset;
1431 
1432 	ISN_LOCK();
1433 	projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1434 
1435 	if (SEQ_GT(projected_offset, V_isn_offset))
1436 		V_isn_offset = projected_offset;
1437 
1438 	V_isn_offset_old = V_isn_offset;
1439 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1440 	ISN_UNLOCK();
1441 }
1442 
1443 /*
1444  * When a specific ICMP unreachable message is received and the
1445  * connection state is SYN-SENT, drop the connection.  This behavior
1446  * is controlled by the icmp_may_rst sysctl.
1447  */
1448 struct inpcb *
1449 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1450 {
1451 	struct tcpcb *tp;
1452 
1453 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1454 	INP_WLOCK_ASSERT(inp);
1455 
1456 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1457 	    (inp->inp_vflag & INP_DROPPED))
1458 		return (inp);
1459 
1460 	tp = intotcpcb(inp);
1461 	if (tp->t_state != TCPS_SYN_SENT)
1462 		return (inp);
1463 
1464 	tp = tcp_drop(tp, errno);
1465 	if (tp != NULL)
1466 		return (inp);
1467 	else
1468 		return (NULL);
1469 }
1470 
1471 /*
1472  * When `need fragmentation' ICMP is received, update our idea of the MSS
1473  * based on the new value in the route.  Also nudge TCP to send something,
1474  * since we know the packet we just sent was dropped.
1475  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1476  */
1477 struct inpcb *
1478 tcp_mtudisc(struct inpcb *inp, int errno)
1479 {
1480 	struct tcpcb *tp;
1481 	struct socket *so = inp->inp_socket;
1482 	u_int maxmtu;
1483 	u_int romtu;
1484 	int mss;
1485 #ifdef INET6
1486 	int isipv6;
1487 #endif /* INET6 */
1488 
1489 	INP_WLOCK_ASSERT(inp);
1490 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1491 	    (inp->inp_vflag & INP_DROPPED))
1492 		return (inp);
1493 
1494 	tp = intotcpcb(inp);
1495 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1496 
1497 #ifdef INET6
1498 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1499 #endif
1500 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1501 	romtu =
1502 #ifdef INET6
1503 	    isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
1504 #endif /* INET6 */
1505 	    tcp_maxmtu(&inp->inp_inc, NULL);
1506 	if (!maxmtu)
1507 		maxmtu = romtu;
1508 	else
1509 		maxmtu = min(maxmtu, romtu);
1510 	if (!maxmtu) {
1511 		tp->t_maxopd = tp->t_maxseg =
1512 #ifdef INET6
1513 			isipv6 ? V_tcp_v6mssdflt :
1514 #endif /* INET6 */
1515 			V_tcp_mssdflt;
1516 		return (inp);
1517 	}
1518 	mss = maxmtu -
1519 #ifdef INET6
1520 		(isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1521 #endif /* INET6 */
1522 		 sizeof(struct tcpiphdr)
1523 #ifdef INET6
1524 		 )
1525 #endif /* INET6 */
1526 		;
1527 
1528 	/*
1529 	 * XXX - The above conditional probably violates the TCP
1530 	 * spec.  The problem is that, since we don't know the
1531 	 * other end's MSS, we are supposed to use a conservative
1532 	 * default.  But, if we do that, then MTU discovery will
1533 	 * never actually take place, because the conservative
1534 	 * default is much less than the MTUs typically seen
1535 	 * on the Internet today.  For the moment, we'll sweep
1536 	 * this under the carpet.
1537 	 *
1538 	 * The conservative default might not actually be a problem
1539 	 * if the only case this occurs is when sending an initial
1540 	 * SYN with options and data to a host we've never talked
1541 	 * to before.  Then, they will reply with an MSS value which
1542 	 * will get recorded and the new parameters should get
1543 	 * recomputed.  For Further Study.
1544 	 */
1545 	if (tp->t_maxopd <= mss)
1546 		return (inp);
1547 	tp->t_maxopd = mss;
1548 
1549 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1550 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1551 		mss -= TCPOLEN_TSTAMP_APPA;
1552 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1553 	if (mss > MCLBYTES)
1554 		mss &= ~(MCLBYTES-1);
1555 #else
1556 	if (mss > MCLBYTES)
1557 		mss = mss / MCLBYTES * MCLBYTES;
1558 #endif
1559 	if (so->so_snd.sb_hiwat < mss)
1560 		mss = so->so_snd.sb_hiwat;
1561 
1562 	tp->t_maxseg = mss;
1563 
1564 	V_tcpstat.tcps_mturesent++;
1565 	tp->t_rtttime = 0;
1566 	tp->snd_nxt = tp->snd_una;
1567 	tcp_free_sackholes(tp);
1568 	tp->snd_recover = tp->snd_max;
1569 	if (tp->t_flags & TF_SACK_PERMIT)
1570 		EXIT_FASTRECOVERY(tp);
1571 	tcp_output_send(tp);
1572 	return (inp);
1573 }
1574 
1575 /*
1576  * Look-up the routing entry to the peer of this inpcb.  If no route
1577  * is found and it cannot be allocated, then return NULL.  This routine
1578  * is called by TCP routines that access the rmx structure and by tcp_mss
1579  * to get the interface MTU.
1580  */
1581 u_long
1582 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1583 {
1584 	struct route sro;
1585 	struct sockaddr_in *dst;
1586 	struct ifnet *ifp;
1587 	u_long maxmtu = 0;
1588 
1589 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1590 
1591 	bzero(&sro, sizeof(sro));
1592 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1593 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1594 		dst->sin_family = AF_INET;
1595 		dst->sin_len = sizeof(*dst);
1596 		dst->sin_addr = inc->inc_faddr;
1597 		in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1598 	}
1599 	if (sro.ro_rt != NULL) {
1600 		ifp = sro.ro_rt->rt_ifp;
1601 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1602 			maxmtu = ifp->if_mtu;
1603 		else
1604 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1605 
1606 		/* Report additional interface capabilities. */
1607 		if (flags != NULL) {
1608 			if (ifp->if_capenable & IFCAP_TSO4 &&
1609 			    ifp->if_hwassist & CSUM_TSO)
1610 				*flags |= CSUM_TSO;
1611 		}
1612 		RTFREE(sro.ro_rt);
1613 	}
1614 	return (maxmtu);
1615 }
1616 
1617 #ifdef INET6
1618 u_long
1619 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1620 {
1621 	struct route_in6 sro6;
1622 	struct ifnet *ifp;
1623 	u_long maxmtu = 0;
1624 
1625 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1626 
1627 	bzero(&sro6, sizeof(sro6));
1628 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1629 		sro6.ro_dst.sin6_family = AF_INET6;
1630 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1631 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1632 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1633 	}
1634 	if (sro6.ro_rt != NULL) {
1635 		ifp = sro6.ro_rt->rt_ifp;
1636 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1637 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1638 		else
1639 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1640 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1641 
1642 		/* Report additional interface capabilities. */
1643 		if (flags != NULL) {
1644 			if (ifp->if_capenable & IFCAP_TSO6 &&
1645 			    ifp->if_hwassist & CSUM_TSO)
1646 				*flags |= CSUM_TSO;
1647 		}
1648 		RTFREE(sro6.ro_rt);
1649 	}
1650 
1651 	return (maxmtu);
1652 }
1653 #endif /* INET6 */
1654 
1655 #ifdef IPSEC
1656 /* compute ESP/AH header size for TCP, including outer IP header. */
1657 size_t
1658 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1659 {
1660 	struct inpcb *inp;
1661 	struct mbuf *m;
1662 	size_t hdrsiz;
1663 	struct ip *ip;
1664 #ifdef INET6
1665 	struct ip6_hdr *ip6;
1666 #endif
1667 	struct tcphdr *th;
1668 
1669 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1670 		return (0);
1671 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1672 	if (!m)
1673 		return (0);
1674 
1675 #ifdef INET6
1676 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1677 		ip6 = mtod(m, struct ip6_hdr *);
1678 		th = (struct tcphdr *)(ip6 + 1);
1679 		m->m_pkthdr.len = m->m_len =
1680 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1681 		tcpip_fillheaders(inp, ip6, th);
1682 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1683 	} else
1684 #endif /* INET6 */
1685 	{
1686 		ip = mtod(m, struct ip *);
1687 		th = (struct tcphdr *)(ip + 1);
1688 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1689 		tcpip_fillheaders(inp, ip, th);
1690 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1691 	}
1692 
1693 	m_free(m);
1694 	return (hdrsiz);
1695 }
1696 #endif /* IPSEC */
1697 
1698 /*
1699  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1700  *
1701  * This code attempts to calculate the bandwidth-delay product as a
1702  * means of determining the optimal window size to maximize bandwidth,
1703  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1704  * routers.  This code also does a fairly good job keeping RTTs in check
1705  * across slow links like modems.  We implement an algorithm which is very
1706  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1707  * transmitter side of a TCP connection and so only effects the transmit
1708  * side of the connection.
1709  *
1710  * BACKGROUND:  TCP makes no provision for the management of buffer space
1711  * at the end points or at the intermediate routers and switches.  A TCP
1712  * stream, whether using NewReno or not, will eventually buffer as
1713  * many packets as it is able and the only reason this typically works is
1714  * due to the fairly small default buffers made available for a connection
1715  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1716  * scaling it is now fairly easy for even a single TCP connection to blow-out
1717  * all available buffer space not only on the local interface, but on
1718  * intermediate routers and switches as well.  NewReno makes a misguided
1719  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1720  * then backing off, then steadily increasing the window again until another
1721  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1722  * is only made worse as network loads increase and the idea of intentionally
1723  * blowing out network buffers is, frankly, a terrible way to manage network
1724  * resources.
1725  *
1726  * It is far better to limit the transmit window prior to the failure
1727  * condition being achieved.  There are two general ways to do this:  First
1728  * you can 'scan' through different transmit window sizes and locate the
1729  * point where the RTT stops increasing, indicating that you have filled the
1730  * pipe, then scan backwards until you note that RTT stops decreasing, then
1731  * repeat ad-infinitum.  This method works in principle but has severe
1732  * implementation issues due to RTT variances, timer granularity, and
1733  * instability in the algorithm which can lead to many false positives and
1734  * create oscillations as well as interact badly with other TCP streams
1735  * implementing the same algorithm.
1736  *
1737  * The second method is to limit the window to the bandwidth delay product
1738  * of the link.  This is the method we implement.  RTT variances and our
1739  * own manipulation of the congestion window, bwnd, can potentially
1740  * destabilize the algorithm.  For this reason we have to stabilize the
1741  * elements used to calculate the window.  We do this by using the minimum
1742  * observed RTT, the long term average of the observed bandwidth, and
1743  * by adding two segments worth of slop.  It isn't perfect but it is able
1744  * to react to changing conditions and gives us a very stable basis on
1745  * which to extend the algorithm.
1746  */
1747 void
1748 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1749 {
1750 	u_long bw;
1751 	u_long bwnd;
1752 	int save_ticks;
1753 
1754 	INP_WLOCK_ASSERT(tp->t_inpcb);
1755 
1756 	/*
1757 	 * If inflight_enable is disabled in the middle of a tcp connection,
1758 	 * make sure snd_bwnd is effectively disabled.
1759 	 */
1760 	if (V_tcp_inflight_enable == 0 ||
1761 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1762 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1763 		tp->snd_bandwidth = 0;
1764 		return;
1765 	}
1766 
1767 	/*
1768 	 * Figure out the bandwidth.  Due to the tick granularity this
1769 	 * is a very rough number and it MUST be averaged over a fairly
1770 	 * long period of time.  XXX we need to take into account a link
1771 	 * that is not using all available bandwidth, but for now our
1772 	 * slop will ramp us up if this case occurs and the bandwidth later
1773 	 * increases.
1774 	 *
1775 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1776 	 * effectively reset t_bw_rtttime for this case.
1777 	 */
1778 	save_ticks = ticks;
1779 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1780 		return;
1781 
1782 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1783 	    (save_ticks - tp->t_bw_rtttime);
1784 	tp->t_bw_rtttime = save_ticks;
1785 	tp->t_bw_rtseq = ack_seq;
1786 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1787 		return;
1788 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1789 
1790 	tp->snd_bandwidth = bw;
1791 
1792 	/*
1793 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1794 	 * segments.  The additional slop puts us squarely in the sweet
1795 	 * spot and also handles the bandwidth run-up case and stabilization.
1796 	 * Without the slop we could be locking ourselves into a lower
1797 	 * bandwidth.
1798 	 *
1799 	 * Situations Handled:
1800 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1801 	 *	    high speed LANs, allowing larger TCP buffers to be
1802 	 *	    specified, and also does a good job preventing
1803 	 *	    over-queueing of packets over choke points like modems
1804 	 *	    (at least for the transmit side).
1805 	 *
1806 	 *	(2) Is able to handle changing network loads (bandwidth
1807 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1808 	 *	    increases).
1809 	 *
1810 	 *	(3) Theoretically should stabilize in the face of multiple
1811 	 *	    connections implementing the same algorithm (this may need
1812 	 *	    a little work).
1813 	 *
1814 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1815 	 *	    be adjusted with a sysctl but typically only needs to be
1816 	 *	    on very slow connections.  A value no smaller then 5
1817 	 *	    should be used, but only reduce this default if you have
1818 	 *	    no other choice.
1819 	 */
1820 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1821 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1822 #undef USERTT
1823 
1824 	if (tcp_inflight_debug > 0) {
1825 		static int ltime;
1826 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1827 			ltime = ticks;
1828 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1829 			    tp,
1830 			    bw,
1831 			    tp->t_rttbest,
1832 			    tp->t_srtt,
1833 			    bwnd
1834 			);
1835 		}
1836 	}
1837 	if ((long)bwnd < V_tcp_inflight_min)
1838 		bwnd = V_tcp_inflight_min;
1839 	if (bwnd > V_tcp_inflight_max)
1840 		bwnd = V_tcp_inflight_max;
1841 	if ((long)bwnd < tp->t_maxseg * 2)
1842 		bwnd = tp->t_maxseg * 2;
1843 	tp->snd_bwnd = bwnd;
1844 }
1845 
1846 #ifdef TCP_SIGNATURE
1847 /*
1848  * Callback function invoked by m_apply() to digest TCP segment data
1849  * contained within an mbuf chain.
1850  */
1851 static int
1852 tcp_signature_apply(void *fstate, void *data, u_int len)
1853 {
1854 
1855 	MD5Update(fstate, (u_char *)data, len);
1856 	return (0);
1857 }
1858 
1859 /*
1860  * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
1861  *
1862  * Parameters:
1863  * m		pointer to head of mbuf chain
1864  * off0		offset to TCP header within the mbuf chain
1865  * len		length of TCP segment data, excluding options
1866  * optlen	length of TCP segment options
1867  * buf		pointer to storage for computed MD5 digest
1868  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1869  *
1870  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1871  * When called from tcp_input(), we can be sure that th_sum has been
1872  * zeroed out and verified already.
1873  *
1874  * This function is for IPv4 use only. Calling this function with an
1875  * IPv6 packet in the mbuf chain will yield undefined results.
1876  *
1877  * Return 0 if successful, otherwise return -1.
1878  *
1879  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1880  * search with the destination IP address, and a 'magic SPI' to be
1881  * determined by the application. This is hardcoded elsewhere to 1179
1882  * right now. Another branch of this code exists which uses the SPD to
1883  * specify per-application flows but it is unstable.
1884  */
1885 int
1886 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
1887     u_char *buf, u_int direction)
1888 {
1889 	union sockaddr_union dst;
1890 	struct ippseudo ippseudo;
1891 	MD5_CTX ctx;
1892 	int doff;
1893 	struct ip *ip;
1894 	struct ipovly *ipovly;
1895 	struct secasvar *sav;
1896 	struct tcphdr *th;
1897 	u_short savecsum;
1898 
1899 	KASSERT(m != NULL, ("NULL mbuf chain"));
1900 	KASSERT(buf != NULL, ("NULL signature pointer"));
1901 
1902 	/* Extract the destination from the IP header in the mbuf. */
1903 	ip = mtod(m, struct ip *);
1904 	bzero(&dst, sizeof(union sockaddr_union));
1905 	dst.sa.sa_len = sizeof(struct sockaddr_in);
1906 	dst.sa.sa_family = AF_INET;
1907 	dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1908 	    ip->ip_src : ip->ip_dst;
1909 
1910 	/* Look up an SADB entry which matches the address of the peer. */
1911 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1912 	if (sav == NULL) {
1913 		printf("%s: SADB lookup failed for %s\n", __func__,
1914 		    inet_ntoa(dst.sin.sin_addr));
1915 		return (EINVAL);
1916 	}
1917 
1918 	MD5Init(&ctx);
1919 	ipovly = (struct ipovly *)ip;
1920 	th = (struct tcphdr *)((u_char *)ip + off0);
1921 	doff = off0 + sizeof(struct tcphdr) + optlen;
1922 
1923 	/*
1924 	 * Step 1: Update MD5 hash with IP pseudo-header.
1925 	 *
1926 	 * XXX The ippseudo header MUST be digested in network byte order,
1927 	 * or else we'll fail the regression test. Assume all fields we've
1928 	 * been doing arithmetic on have been in host byte order.
1929 	 * XXX One cannot depend on ipovly->ih_len here. When called from
1930 	 * tcp_output(), the underlying ip_len member has not yet been set.
1931 	 */
1932 	ippseudo.ippseudo_src = ipovly->ih_src;
1933 	ippseudo.ippseudo_dst = ipovly->ih_dst;
1934 	ippseudo.ippseudo_pad = 0;
1935 	ippseudo.ippseudo_p = IPPROTO_TCP;
1936 	ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
1937 	MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1938 
1939 	/*
1940 	 * Step 2: Update MD5 hash with TCP header, excluding options.
1941 	 * The TCP checksum must be set to zero.
1942 	 */
1943 	savecsum = th->th_sum;
1944 	th->th_sum = 0;
1945 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1946 	th->th_sum = savecsum;
1947 
1948 	/*
1949 	 * Step 3: Update MD5 hash with TCP segment data.
1950 	 *         Use m_apply() to avoid an early m_pullup().
1951 	 */
1952 	if (len > 0)
1953 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
1954 
1955 	/*
1956 	 * Step 4: Update MD5 hash with shared secret.
1957 	 */
1958 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1959 	MD5Final(buf, &ctx);
1960 
1961 	key_sa_recordxfer(sav, m);
1962 	KEY_FREESAV(&sav);
1963 	return (0);
1964 }
1965 #endif /* TCP_SIGNATURE */
1966 
1967 static int
1968 sysctl_drop(SYSCTL_HANDLER_ARGS)
1969 {
1970 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
1971 	struct sockaddr_storage addrs[2];
1972 	struct inpcb *inp;
1973 	struct tcpcb *tp;
1974 	struct tcptw *tw;
1975 	struct sockaddr_in *fin, *lin;
1976 #ifdef INET6
1977 	struct sockaddr_in6 *fin6, *lin6;
1978 	struct in6_addr f6, l6;
1979 #endif
1980 	int error;
1981 
1982 	inp = NULL;
1983 	fin = lin = NULL;
1984 #ifdef INET6
1985 	fin6 = lin6 = NULL;
1986 #endif
1987 	error = 0;
1988 
1989 	if (req->oldptr != NULL || req->oldlen != 0)
1990 		return (EINVAL);
1991 	if (req->newptr == NULL)
1992 		return (EPERM);
1993 	if (req->newlen < sizeof(addrs))
1994 		return (ENOMEM);
1995 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
1996 	if (error)
1997 		return (error);
1998 
1999 	switch (addrs[0].ss_family) {
2000 #ifdef INET6
2001 	case AF_INET6:
2002 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2003 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2004 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2005 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2006 			return (EINVAL);
2007 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2008 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2009 				return (EINVAL);
2010 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2011 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2012 			fin = (struct sockaddr_in *)&addrs[0];
2013 			lin = (struct sockaddr_in *)&addrs[1];
2014 			break;
2015 		}
2016 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2017 		if (error)
2018 			return (error);
2019 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2020 		if (error)
2021 			return (error);
2022 		break;
2023 #endif
2024 	case AF_INET:
2025 		fin = (struct sockaddr_in *)&addrs[0];
2026 		lin = (struct sockaddr_in *)&addrs[1];
2027 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2028 		    lin->sin_len != sizeof(struct sockaddr_in))
2029 			return (EINVAL);
2030 		break;
2031 	default:
2032 		return (EINVAL);
2033 	}
2034 	INP_INFO_WLOCK(&V_tcbinfo);
2035 	switch (addrs[0].ss_family) {
2036 #ifdef INET6
2037 	case AF_INET6:
2038 		inp = in6_pcblookup_hash(&V_tcbinfo, &f6, fin6->sin6_port,
2039 		    &l6, lin6->sin6_port, 0, NULL);
2040 		break;
2041 #endif
2042 	case AF_INET:
2043 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2044 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2045 		break;
2046 	}
2047 	if (inp != NULL) {
2048 		INP_WLOCK(inp);
2049 		if (inp->inp_vflag & INP_TIMEWAIT) {
2050 			/*
2051 			 * XXXRW: There currently exists a state where an
2052 			 * inpcb is present, but its timewait state has been
2053 			 * discarded.  For now, don't allow dropping of this
2054 			 * type of inpcb.
2055 			 */
2056 			tw = intotw(inp);
2057 			if (tw != NULL)
2058 				tcp_twclose(tw, 0);
2059 			else
2060 				INP_WUNLOCK(inp);
2061 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2062 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2063 			tp = intotcpcb(inp);
2064 			tp = tcp_drop(tp, ECONNABORTED);
2065 			if (tp != NULL)
2066 				INP_WUNLOCK(inp);
2067 		} else
2068 			INP_WUNLOCK(inp);
2069 	} else
2070 		error = ESRCH;
2071 	INP_INFO_WUNLOCK(&V_tcbinfo);
2072 	return (error);
2073 }
2074 
2075 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2076     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2077     0, sysctl_drop, "", "Drop TCP connection");
2078 
2079 /*
2080  * Generate a standardized TCP log line for use throughout the
2081  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2082  * allow use in the interrupt context.
2083  *
2084  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2085  * NB: The function may return NULL if memory allocation failed.
2086  *
2087  * Due to header inclusion and ordering limitations the struct ip
2088  * and ip6_hdr pointers have to be passed as void pointers.
2089  */
2090 char *
2091 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2092     const void *ip6hdr)
2093 {
2094 	char *s, *sp;
2095 	size_t size;
2096 	struct ip *ip;
2097 #ifdef INET6
2098 	const struct ip6_hdr *ip6;
2099 
2100 	ip6 = (const struct ip6_hdr *)ip6hdr;
2101 #endif /* INET6 */
2102 	ip = (struct ip *)ip4hdr;
2103 
2104 	/*
2105 	 * The log line looks like this:
2106 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2107 	 */
2108 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2109 	    sizeof(PRINT_TH_FLAGS) + 1 +
2110 #ifdef INET6
2111 	    2 * INET6_ADDRSTRLEN;
2112 #else
2113 	    2 * INET_ADDRSTRLEN;
2114 #endif /* INET6 */
2115 
2116 	/* Is logging enabled? */
2117 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2118 		return (NULL);
2119 
2120 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2121 	if (s == NULL)
2122 		return (NULL);
2123 
2124 	strcat(s, "TCP: [");
2125 	sp = s + strlen(s);
2126 
2127 	if (inc && inc->inc_isipv6 == 0) {
2128 		inet_ntoa_r(inc->inc_faddr, sp);
2129 		sp = s + strlen(s);
2130 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2131 		sp = s + strlen(s);
2132 		inet_ntoa_r(inc->inc_laddr, sp);
2133 		sp = s + strlen(s);
2134 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2135 #ifdef INET6
2136 	} else if (inc) {
2137 		ip6_sprintf(sp, &inc->inc6_faddr);
2138 		sp = s + strlen(s);
2139 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2140 		sp = s + strlen(s);
2141 		ip6_sprintf(sp, &inc->inc6_laddr);
2142 		sp = s + strlen(s);
2143 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2144 	} else if (ip6 && th) {
2145 		ip6_sprintf(sp, &ip6->ip6_src);
2146 		sp = s + strlen(s);
2147 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2148 		sp = s + strlen(s);
2149 		ip6_sprintf(sp, &ip6->ip6_dst);
2150 		sp = s + strlen(s);
2151 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2152 #endif /* INET6 */
2153 	} else if (ip && th) {
2154 		inet_ntoa_r(ip->ip_src, sp);
2155 		sp = s + strlen(s);
2156 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2157 		sp = s + strlen(s);
2158 		inet_ntoa_r(ip->ip_dst, sp);
2159 		sp = s + strlen(s);
2160 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2161 	} else {
2162 		free(s, M_TCPLOG);
2163 		return (NULL);
2164 	}
2165 	sp = s + strlen(s);
2166 	if (th)
2167 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2168 	if (*(s + size - 1) != '\0')
2169 		panic("%s: string too long", __func__);
2170 	return (s);
2171 }
2172