xref: /freebsd/sys/netinet/tcp_subr.c (revision 1593c65a5bc643d8afee4e9c88d81e9bb052fe91)
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 
59 #include <vm/uma.h>
60 
61 #include <net/route.h>
62 #include <net/if.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #include <netinet6/nd6.h>
80 #endif
81 #include <netinet/ip_icmp.h>
82 #include <netinet/tcp.h>
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #include <netinet/tcp_offload.h>
89 #ifdef INET6
90 #include <netinet6/tcp6_var.h>
91 #endif
92 #include <netinet/tcpip.h>
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #include <netinet6/ip6protosw.h>
97 
98 #ifdef IPSEC
99 #include <netipsec/ipsec.h>
100 #include <netipsec/xform.h>
101 #ifdef INET6
102 #include <netipsec/ipsec6.h>
103 #endif
104 #include <netipsec/key.h>
105 #endif /*IPSEC*/
106 
107 #include <machine/in_cksum.h>
108 #include <sys/md5.h>
109 
110 #include <security/mac/mac_framework.h>
111 
112 int	tcp_mssdflt = TCP_MSS;
113 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
114     &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
115 
116 #ifdef INET6
117 int	tcp_v6mssdflt = TCP6_MSS;
118 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
119     CTLFLAG_RW, &tcp_v6mssdflt , 0,
120     "Default TCP Maximum Segment Size for IPv6");
121 #endif
122 
123 /*
124  * Minimum MSS we accept and use. This prevents DoS attacks where
125  * we are forced to a ridiculous low MSS like 20 and send hundreds
126  * of packets instead of one. The effect scales with the available
127  * bandwidth and quickly saturates the CPU and network interface
128  * with packet generation and sending. Set to zero to disable MINMSS
129  * checking. This setting prevents us from sending too small packets.
130  */
131 int	tcp_minmss = TCP_MINMSS;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
133     &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
134 
135 int	tcp_do_rfc1323 = 1;
136 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
137     &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
138 
139 static int	tcp_log_debug = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
141     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
142 
143 static int	tcp_tcbhashsize = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
145     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
146 
147 static int	do_tcpdrain = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW,
149     &do_tcpdrain, 0,
150     "Enable tcp_drain routine for extra help when low on mbufs");
151 
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
153     &tcbinfo.ipi_count, 0, "Number of active PCBs");
154 
155 static int	icmp_may_rst = 1;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
157     &icmp_may_rst, 0,
158     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
159 
160 static int	tcp_isn_reseed_interval = 0;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
162     &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
163 
164 /*
165  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
166  * 1024 exists only for debugging.  A good production default would be
167  * something like 6100.
168  */
169 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
170     "TCP inflight data limiting");
171 
172 static int	tcp_inflight_enable = 1;
173 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
174     &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
175 
176 static int	tcp_inflight_debug = 0;
177 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
178     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
179 
180 static int	tcp_inflight_rttthresh;
181 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
182     &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
183     "RTT threshold below which inflight will deactivate itself");
184 
185 static int	tcp_inflight_min = 6144;
186 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
187     &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
188 
189 static int	tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
190 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
191     &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
192 
193 static int	tcp_inflight_stab = 20;
194 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
195     &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
196 
197 uma_zone_t sack_hole_zone;
198 
199 static struct inpcb *tcp_notify(struct inpcb *, int);
200 static void	tcp_isn_tick(void *);
201 
202 /*
203  * Target size of TCP PCB hash tables. Must be a power of two.
204  *
205  * Note that this can be overridden by the kernel environment
206  * variable net.inet.tcp.tcbhashsize
207  */
208 #ifndef TCBHASHSIZE
209 #define TCBHASHSIZE	512
210 #endif
211 
212 /*
213  * XXX
214  * Callouts should be moved into struct tcp directly.  They are currently
215  * separate because the tcpcb structure is exported to userland for sysctl
216  * parsing purposes, which do not know about callouts.
217  */
218 struct tcpcb_mem {
219 	struct	tcpcb		tcb;
220 	struct	tcp_timer	tt;
221 };
222 
223 static uma_zone_t tcpcb_zone;
224 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
225 struct callout isn_callout;
226 static struct mtx isn_mtx;
227 
228 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
229 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
230 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
231 
232 /*
233  * TCP initialization.
234  */
235 static void
236 tcp_zone_change(void *tag)
237 {
238 
239 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
240 	uma_zone_set_max(tcpcb_zone, maxsockets);
241 	tcp_tw_zone_change();
242 }
243 
244 static int
245 tcp_inpcb_init(void *mem, int size, int flags)
246 {
247 	struct inpcb *inp = mem;
248 
249 	INP_LOCK_INIT(inp, "inp", "tcpinp");
250 	return (0);
251 }
252 
253 void
254 tcp_init(void)
255 {
256 
257 	int hashsize = TCBHASHSIZE;
258 	tcp_delacktime = TCPTV_DELACK;
259 	tcp_keepinit = TCPTV_KEEP_INIT;
260 	tcp_keepidle = TCPTV_KEEP_IDLE;
261 	tcp_keepintvl = TCPTV_KEEPINTVL;
262 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
263 	tcp_msl = TCPTV_MSL;
264 	tcp_rexmit_min = TCPTV_MIN;
265 	if (tcp_rexmit_min < 1)
266 		tcp_rexmit_min = 1;
267 	tcp_rexmit_slop = TCPTV_CPU_VAR;
268 	tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
269 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
270 
271 	INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
272 	LIST_INIT(&tcb);
273 	tcbinfo.ipi_listhead = &tcb;
274 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
275 	if (!powerof2(hashsize)) {
276 		printf("WARNING: TCB hash size not a power of 2\n");
277 		hashsize = 512; /* safe default */
278 	}
279 	tcp_tcbhashsize = hashsize;
280 	tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
281 	    &tcbinfo.ipi_hashmask);
282 	tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
283 	    &tcbinfo.ipi_porthashmask);
284 	tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
285 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
286 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
287 #ifdef INET6
288 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
289 #else /* INET6 */
290 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
291 #endif /* INET6 */
292 	if (max_protohdr < TCP_MINPROTOHDR)
293 		max_protohdr = TCP_MINPROTOHDR;
294 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
295 		panic("tcp_init");
296 #undef TCP_MINPROTOHDR
297 	/*
298 	 * These have to be type stable for the benefit of the timers.
299 	 */
300 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
301 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
302 	uma_zone_set_max(tcpcb_zone, maxsockets);
303 	tcp_tw_init();
304 	syncache_init();
305 	tcp_hc_init();
306 	tcp_reass_init();
307 	ISN_LOCK_INIT();
308 	callout_init(&isn_callout, CALLOUT_MPSAFE);
309 	tcp_isn_tick(NULL);
310 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
311 		SHUTDOWN_PRI_DEFAULT);
312 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
313 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
315 		EVENTHANDLER_PRI_ANY);
316 }
317 
318 void
319 tcp_fini(void *xtp)
320 {
321 
322 	callout_stop(&isn_callout);
323 }
324 
325 /*
326  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
327  * tcp_template used to store this data in mbufs, but we now recopy it out
328  * of the tcpcb each time to conserve mbufs.
329  */
330 void
331 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
332 {
333 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
334 
335 	INP_WLOCK_ASSERT(inp);
336 
337 #ifdef INET6
338 	if ((inp->inp_vflag & INP_IPV6) != 0) {
339 		struct ip6_hdr *ip6;
340 
341 		ip6 = (struct ip6_hdr *)ip_ptr;
342 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
343 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
344 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
345 			(IPV6_VERSION & IPV6_VERSION_MASK);
346 		ip6->ip6_nxt = IPPROTO_TCP;
347 		ip6->ip6_plen = sizeof(struct tcphdr);
348 		ip6->ip6_src = inp->in6p_laddr;
349 		ip6->ip6_dst = inp->in6p_faddr;
350 	} else
351 #endif
352 	{
353 		struct ip *ip;
354 
355 		ip = (struct ip *)ip_ptr;
356 		ip->ip_v = IPVERSION;
357 		ip->ip_hl = 5;
358 		ip->ip_tos = inp->inp_ip_tos;
359 		ip->ip_len = 0;
360 		ip->ip_id = 0;
361 		ip->ip_off = 0;
362 		ip->ip_ttl = inp->inp_ip_ttl;
363 		ip->ip_sum = 0;
364 		ip->ip_p = IPPROTO_TCP;
365 		ip->ip_src = inp->inp_laddr;
366 		ip->ip_dst = inp->inp_faddr;
367 	}
368 	th->th_sport = inp->inp_lport;
369 	th->th_dport = inp->inp_fport;
370 	th->th_seq = 0;
371 	th->th_ack = 0;
372 	th->th_x2 = 0;
373 	th->th_off = 5;
374 	th->th_flags = 0;
375 	th->th_win = 0;
376 	th->th_urp = 0;
377 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
378 }
379 
380 /*
381  * Create template to be used to send tcp packets on a connection.
382  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
383  * use for this function is in keepalives, which use tcp_respond.
384  */
385 struct tcptemp *
386 tcpip_maketemplate(struct inpcb *inp)
387 {
388 	struct mbuf *m;
389 	struct tcptemp *n;
390 
391 	m = m_get(M_DONTWAIT, MT_DATA);
392 	if (m == NULL)
393 		return (0);
394 	m->m_len = sizeof(struct tcptemp);
395 	n = mtod(m, struct tcptemp *);
396 
397 	tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
398 	return (n);
399 }
400 
401 /*
402  * Send a single message to the TCP at address specified by
403  * the given TCP/IP header.  If m == NULL, then we make a copy
404  * of the tcpiphdr at ti and send directly to the addressed host.
405  * This is used to force keep alive messages out using the TCP
406  * template for a connection.  If flags are given then we send
407  * a message back to the TCP which originated the * segment ti,
408  * and discard the mbuf containing it and any other attached mbufs.
409  *
410  * In any case the ack and sequence number of the transmitted
411  * segment are as specified by the parameters.
412  *
413  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
414  */
415 void
416 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
417     tcp_seq ack, tcp_seq seq, int flags)
418 {
419 	int tlen;
420 	int win = 0;
421 	struct ip *ip;
422 	struct tcphdr *nth;
423 #ifdef INET6
424 	struct ip6_hdr *ip6;
425 	int isipv6;
426 #endif /* INET6 */
427 	int ipflags = 0;
428 	struct inpcb *inp;
429 
430 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
431 
432 #ifdef INET6
433 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
434 	ip6 = ipgen;
435 #endif /* INET6 */
436 	ip = ipgen;
437 
438 	if (tp != NULL) {
439 		inp = tp->t_inpcb;
440 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
441 		INP_WLOCK_ASSERT(inp);
442 	} else
443 		inp = NULL;
444 
445 	if (tp != NULL) {
446 		if (!(flags & TH_RST)) {
447 			win = sbspace(&inp->inp_socket->so_rcv);
448 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
449 				win = (long)TCP_MAXWIN << tp->rcv_scale;
450 		}
451 	}
452 	if (m == NULL) {
453 		m = m_gethdr(M_DONTWAIT, MT_DATA);
454 		if (m == NULL)
455 			return;
456 		tlen = 0;
457 		m->m_data += max_linkhdr;
458 #ifdef INET6
459 		if (isipv6) {
460 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
461 			      sizeof(struct ip6_hdr));
462 			ip6 = mtod(m, struct ip6_hdr *);
463 			nth = (struct tcphdr *)(ip6 + 1);
464 		} else
465 #endif /* INET6 */
466 	      {
467 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
468 		ip = mtod(m, struct ip *);
469 		nth = (struct tcphdr *)(ip + 1);
470 	      }
471 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
472 		flags = TH_ACK;
473 	} else {
474 		/*
475 		 *  reuse the mbuf.
476 		 * XXX MRT We inherrit the FIB, which is lucky.
477 		 */
478 		m_freem(m->m_next);
479 		m->m_next = NULL;
480 		m->m_data = (caddr_t)ipgen;
481 		/* m_len is set later */
482 		tlen = 0;
483 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
484 #ifdef INET6
485 		if (isipv6) {
486 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
487 			nth = (struct tcphdr *)(ip6 + 1);
488 		} else
489 #endif /* INET6 */
490 	      {
491 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
492 		nth = (struct tcphdr *)(ip + 1);
493 	      }
494 		if (th != nth) {
495 			/*
496 			 * this is usually a case when an extension header
497 			 * exists between the IPv6 header and the
498 			 * TCP header.
499 			 */
500 			nth->th_sport = th->th_sport;
501 			nth->th_dport = th->th_dport;
502 		}
503 		xchg(nth->th_dport, nth->th_sport, n_short);
504 #undef xchg
505 	}
506 #ifdef INET6
507 	if (isipv6) {
508 		ip6->ip6_flow = 0;
509 		ip6->ip6_vfc = IPV6_VERSION;
510 		ip6->ip6_nxt = IPPROTO_TCP;
511 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
512 						tlen));
513 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
514 	} else
515 #endif
516 	{
517 		tlen += sizeof (struct tcpiphdr);
518 		ip->ip_len = tlen;
519 		ip->ip_ttl = ip_defttl;
520 		if (path_mtu_discovery)
521 			ip->ip_off |= IP_DF;
522 	}
523 	m->m_len = tlen;
524 	m->m_pkthdr.len = tlen;
525 	m->m_pkthdr.rcvif = NULL;
526 #ifdef MAC
527 	if (inp != NULL) {
528 		/*
529 		 * Packet is associated with a socket, so allow the
530 		 * label of the response to reflect the socket label.
531 		 */
532 		INP_WLOCK_ASSERT(inp);
533 		mac_inpcb_create_mbuf(inp, m);
534 	} else {
535 		/*
536 		 * Packet is not associated with a socket, so possibly
537 		 * update the label in place.
538 		 */
539 		mac_netinet_tcp_reply(m);
540 	}
541 #endif
542 	nth->th_seq = htonl(seq);
543 	nth->th_ack = htonl(ack);
544 	nth->th_x2 = 0;
545 	nth->th_off = sizeof (struct tcphdr) >> 2;
546 	nth->th_flags = flags;
547 	if (tp != NULL)
548 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
549 	else
550 		nth->th_win = htons((u_short)win);
551 	nth->th_urp = 0;
552 #ifdef INET6
553 	if (isipv6) {
554 		nth->th_sum = 0;
555 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
556 					sizeof(struct ip6_hdr),
557 					tlen - sizeof(struct ip6_hdr));
558 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
559 		    NULL, NULL);
560 	} else
561 #endif /* INET6 */
562 	{
563 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
564 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
565 		m->m_pkthdr.csum_flags = CSUM_TCP;
566 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
567 	}
568 #ifdef TCPDEBUG
569 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
570 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
571 #endif
572 #ifdef INET6
573 	if (isipv6)
574 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
575 	else
576 #endif /* INET6 */
577 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
578 }
579 
580 /*
581  * Create a new TCP control block, making an
582  * empty reassembly queue and hooking it to the argument
583  * protocol control block.  The `inp' parameter must have
584  * come from the zone allocator set up in tcp_init().
585  */
586 struct tcpcb *
587 tcp_newtcpcb(struct inpcb *inp)
588 {
589 	struct tcpcb_mem *tm;
590 	struct tcpcb *tp;
591 #ifdef INET6
592 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
593 #endif /* INET6 */
594 
595 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
596 	if (tm == NULL)
597 		return (NULL);
598 	tp = &tm->tcb;
599 	tp->t_timers = &tm->tt;
600 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
601 	tp->t_maxseg = tp->t_maxopd =
602 #ifdef INET6
603 		isipv6 ? tcp_v6mssdflt :
604 #endif /* INET6 */
605 		tcp_mssdflt;
606 
607 	/* Set up our timeouts. */
608 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
609 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
610 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
611 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
612 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
613 
614 	if (tcp_do_rfc1323)
615 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
616 	if (tcp_do_sack)
617 		tp->t_flags |= TF_SACK_PERMIT;
618 	TAILQ_INIT(&tp->snd_holes);
619 	tp->t_inpcb = inp;	/* XXX */
620 	/*
621 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
622 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
623 	 * reasonable initial retransmit time.
624 	 */
625 	tp->t_srtt = TCPTV_SRTTBASE;
626 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
627 	tp->t_rttmin = tcp_rexmit_min;
628 	tp->t_rxtcur = TCPTV_RTOBASE;
629 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
630 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
631 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
632 	tp->t_rcvtime = ticks;
633 	tp->t_bw_rtttime = ticks;
634 	/*
635 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
636 	 * because the socket may be bound to an IPv6 wildcard address,
637 	 * which may match an IPv4-mapped IPv6 address.
638 	 */
639 	inp->inp_ip_ttl = ip_defttl;
640 	inp->inp_ppcb = tp;
641 	return (tp);		/* XXX */
642 }
643 
644 /*
645  * Drop a TCP connection, reporting
646  * the specified error.  If connection is synchronized,
647  * then send a RST to peer.
648  */
649 struct tcpcb *
650 tcp_drop(struct tcpcb *tp, int errno)
651 {
652 	struct socket *so = tp->t_inpcb->inp_socket;
653 
654 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
655 	INP_WLOCK_ASSERT(tp->t_inpcb);
656 
657 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
658 		tp->t_state = TCPS_CLOSED;
659 		(void) tcp_output_reset(tp);
660 		tcpstat.tcps_drops++;
661 	} else
662 		tcpstat.tcps_conndrops++;
663 	if (errno == ETIMEDOUT && tp->t_softerror)
664 		errno = tp->t_softerror;
665 	so->so_error = errno;
666 	return (tcp_close(tp));
667 }
668 
669 void
670 tcp_discardcb(struct tcpcb *tp)
671 {
672 	struct tseg_qent *q;
673 	struct inpcb *inp = tp->t_inpcb;
674 	struct socket *so = inp->inp_socket;
675 #ifdef INET6
676 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
677 #endif /* INET6 */
678 
679 	INP_WLOCK_ASSERT(inp);
680 
681 	/*
682 	 * Make sure that all of our timers are stopped before we
683 	 * delete the PCB.
684 	 */
685 	callout_stop(&tp->t_timers->tt_rexmt);
686 	callout_stop(&tp->t_timers->tt_persist);
687 	callout_stop(&tp->t_timers->tt_keep);
688 	callout_stop(&tp->t_timers->tt_2msl);
689 	callout_stop(&tp->t_timers->tt_delack);
690 
691 	/*
692 	 * If we got enough samples through the srtt filter,
693 	 * save the rtt and rttvar in the routing entry.
694 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
695 	 * 4 samples is enough for the srtt filter to converge
696 	 * to within enough % of the correct value; fewer samples
697 	 * and we could save a bogus rtt. The danger is not high
698 	 * as tcp quickly recovers from everything.
699 	 * XXX: Works very well but needs some more statistics!
700 	 */
701 	if (tp->t_rttupdated >= 4) {
702 		struct hc_metrics_lite metrics;
703 		u_long ssthresh;
704 
705 		bzero(&metrics, sizeof(metrics));
706 		/*
707 		 * Update the ssthresh always when the conditions below
708 		 * are satisfied. This gives us better new start value
709 		 * for the congestion avoidance for new connections.
710 		 * ssthresh is only set if packet loss occured on a session.
711 		 *
712 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
713 		 * being torn down.  Ideally this code would not use 'so'.
714 		 */
715 		ssthresh = tp->snd_ssthresh;
716 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
717 			/*
718 			 * convert the limit from user data bytes to
719 			 * packets then to packet data bytes.
720 			 */
721 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
722 			if (ssthresh < 2)
723 				ssthresh = 2;
724 			ssthresh *= (u_long)(tp->t_maxseg +
725 #ifdef INET6
726 				      (isipv6 ? sizeof (struct ip6_hdr) +
727 					       sizeof (struct tcphdr) :
728 #endif
729 				       sizeof (struct tcpiphdr)
730 #ifdef INET6
731 				       )
732 #endif
733 				      );
734 		} else
735 			ssthresh = 0;
736 		metrics.rmx_ssthresh = ssthresh;
737 
738 		metrics.rmx_rtt = tp->t_srtt;
739 		metrics.rmx_rttvar = tp->t_rttvar;
740 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
741 		metrics.rmx_bandwidth = tp->snd_bandwidth;
742 		metrics.rmx_cwnd = tp->snd_cwnd;
743 		metrics.rmx_sendpipe = 0;
744 		metrics.rmx_recvpipe = 0;
745 
746 		tcp_hc_update(&inp->inp_inc, &metrics);
747 	}
748 
749 	/* free the reassembly queue, if any */
750 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
751 		LIST_REMOVE(q, tqe_q);
752 		m_freem(q->tqe_m);
753 		uma_zfree(tcp_reass_zone, q);
754 		tp->t_segqlen--;
755 		tcp_reass_qsize--;
756 	}
757 	/* Disconnect offload device, if any. */
758 	tcp_offload_detach(tp);
759 
760 	tcp_free_sackholes(tp);
761 	inp->inp_ppcb = NULL;
762 	tp->t_inpcb = NULL;
763 	uma_zfree(tcpcb_zone, tp);
764 }
765 
766 /*
767  * Attempt to close a TCP control block, marking it as dropped, and freeing
768  * the socket if we hold the only reference.
769  */
770 struct tcpcb *
771 tcp_close(struct tcpcb *tp)
772 {
773 	struct inpcb *inp = tp->t_inpcb;
774 	struct socket *so;
775 
776 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
777 	INP_WLOCK_ASSERT(inp);
778 
779 	/* Notify any offload devices of listener close */
780 	if (tp->t_state == TCPS_LISTEN)
781 		tcp_offload_listen_close(tp);
782 	in_pcbdrop(inp);
783 	tcpstat.tcps_closed++;
784 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
785 	so = inp->inp_socket;
786 	soisdisconnected(so);
787 	if (inp->inp_vflag & INP_SOCKREF) {
788 		KASSERT(so->so_state & SS_PROTOREF,
789 		    ("tcp_close: !SS_PROTOREF"));
790 		inp->inp_vflag &= ~INP_SOCKREF;
791 		INP_WUNLOCK(inp);
792 		ACCEPT_LOCK();
793 		SOCK_LOCK(so);
794 		so->so_state &= ~SS_PROTOREF;
795 		sofree(so);
796 		return (NULL);
797 	}
798 	return (tp);
799 }
800 
801 void
802 tcp_drain(void)
803 {
804 
805 	if (do_tcpdrain) {
806 		struct inpcb *inpb;
807 		struct tcpcb *tcpb;
808 		struct tseg_qent *te;
809 
810 	/*
811 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
812 	 * if there is one...
813 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
814 	 *      reassembly queue should be flushed, but in a situation
815 	 *	where we're really low on mbufs, this is potentially
816 	 *	usefull.
817 	 */
818 		INP_INFO_RLOCK(&tcbinfo);
819 		LIST_FOREACH(inpb, tcbinfo.ipi_listhead, inp_list) {
820 			if (inpb->inp_vflag & INP_TIMEWAIT)
821 				continue;
822 			INP_WLOCK(inpb);
823 			if ((tcpb = intotcpcb(inpb)) != NULL) {
824 				while ((te = LIST_FIRST(&tcpb->t_segq))
825 			            != NULL) {
826 					LIST_REMOVE(te, tqe_q);
827 					m_freem(te->tqe_m);
828 					uma_zfree(tcp_reass_zone, te);
829 					tcpb->t_segqlen--;
830 					tcp_reass_qsize--;
831 				}
832 				tcp_clean_sackreport(tcpb);
833 			}
834 			INP_WUNLOCK(inpb);
835 		}
836 		INP_INFO_RUNLOCK(&tcbinfo);
837 	}
838 }
839 
840 /*
841  * Notify a tcp user of an asynchronous error;
842  * store error as soft error, but wake up user
843  * (for now, won't do anything until can select for soft error).
844  *
845  * Do not wake up user since there currently is no mechanism for
846  * reporting soft errors (yet - a kqueue filter may be added).
847  */
848 static struct inpcb *
849 tcp_notify(struct inpcb *inp, int error)
850 {
851 	struct tcpcb *tp;
852 
853 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
854 	INP_WLOCK_ASSERT(inp);
855 
856 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
857 	    (inp->inp_vflag & INP_DROPPED))
858 		return (inp);
859 
860 	tp = intotcpcb(inp);
861 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
862 
863 	/*
864 	 * Ignore some errors if we are hooked up.
865 	 * If connection hasn't completed, has retransmitted several times,
866 	 * and receives a second error, give up now.  This is better
867 	 * than waiting a long time to establish a connection that
868 	 * can never complete.
869 	 */
870 	if (tp->t_state == TCPS_ESTABLISHED &&
871 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
872 	     error == EHOSTDOWN)) {
873 		return (inp);
874 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
875 	    tp->t_softerror) {
876 		tp = tcp_drop(tp, error);
877 		if (tp != NULL)
878 			return (inp);
879 		else
880 			return (NULL);
881 	} else {
882 		tp->t_softerror = error;
883 		return (inp);
884 	}
885 #if 0
886 	wakeup( &so->so_timeo);
887 	sorwakeup(so);
888 	sowwakeup(so);
889 #endif
890 }
891 
892 static int
893 tcp_pcblist(SYSCTL_HANDLER_ARGS)
894 {
895 	int error, i, m, n, pcb_count;
896 	struct inpcb *inp, **inp_list;
897 	inp_gen_t gencnt;
898 	struct xinpgen xig;
899 
900 	/*
901 	 * The process of preparing the TCB list is too time-consuming and
902 	 * resource-intensive to repeat twice on every request.
903 	 */
904 	if (req->oldptr == NULL) {
905 		m = syncache_pcbcount();
906 		n = tcbinfo.ipi_count;
907 		req->oldidx = 2 * (sizeof xig)
908 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
909 		return (0);
910 	}
911 
912 	if (req->newptr != NULL)
913 		return (EPERM);
914 
915 	/*
916 	 * OK, now we're committed to doing something.
917 	 */
918 	INP_INFO_RLOCK(&tcbinfo);
919 	gencnt = tcbinfo.ipi_gencnt;
920 	n = tcbinfo.ipi_count;
921 	INP_INFO_RUNLOCK(&tcbinfo);
922 
923 	m = syncache_pcbcount();
924 
925 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
926 		+ (n + m) * sizeof(struct xtcpcb));
927 	if (error != 0)
928 		return (error);
929 
930 	xig.xig_len = sizeof xig;
931 	xig.xig_count = n + m;
932 	xig.xig_gen = gencnt;
933 	xig.xig_sogen = so_gencnt;
934 	error = SYSCTL_OUT(req, &xig, sizeof xig);
935 	if (error)
936 		return (error);
937 
938 	error = syncache_pcblist(req, m, &pcb_count);
939 	if (error)
940 		return (error);
941 
942 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
943 	if (inp_list == NULL)
944 		return (ENOMEM);
945 
946 	INP_INFO_RLOCK(&tcbinfo);
947 	for (inp = LIST_FIRST(tcbinfo.ipi_listhead), i = 0; inp != NULL && i
948 	    < n; inp = LIST_NEXT(inp, inp_list)) {
949 		INP_WLOCK(inp);
950 		if (inp->inp_gencnt <= gencnt) {
951 			/*
952 			 * XXX: This use of cr_cansee(), introduced with
953 			 * TCP state changes, is not quite right, but for
954 			 * now, better than nothing.
955 			 */
956 			if (inp->inp_vflag & INP_TIMEWAIT) {
957 				if (intotw(inp) != NULL)
958 					error = cr_cansee(req->td->td_ucred,
959 					    intotw(inp)->tw_cred);
960 				else
961 					error = EINVAL;	/* Skip this inp. */
962 			} else
963 				error = cr_canseesocket(req->td->td_ucred,
964 				    inp->inp_socket);
965 			if (error == 0)
966 				inp_list[i++] = inp;
967 		}
968 		INP_WUNLOCK(inp);
969 	}
970 	INP_INFO_RUNLOCK(&tcbinfo);
971 	n = i;
972 
973 	error = 0;
974 	for (i = 0; i < n; i++) {
975 		inp = inp_list[i];
976 		INP_WLOCK(inp);
977 		if (inp->inp_gencnt <= gencnt) {
978 			struct xtcpcb xt;
979 			void *inp_ppcb;
980 
981 			bzero(&xt, sizeof(xt));
982 			xt.xt_len = sizeof xt;
983 			/* XXX should avoid extra copy */
984 			bcopy(inp, &xt.xt_inp, sizeof *inp);
985 			inp_ppcb = inp->inp_ppcb;
986 			if (inp_ppcb == NULL)
987 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
988 			else if (inp->inp_vflag & INP_TIMEWAIT) {
989 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
990 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
991 			} else
992 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
993 			if (inp->inp_socket != NULL)
994 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
995 			else {
996 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
997 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
998 			}
999 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1000 			INP_WUNLOCK(inp);
1001 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1002 		} else
1003 			INP_WUNLOCK(inp);
1004 
1005 	}
1006 	if (!error) {
1007 		/*
1008 		 * Give the user an updated idea of our state.
1009 		 * If the generation differs from what we told
1010 		 * her before, she knows that something happened
1011 		 * while we were processing this request, and it
1012 		 * might be necessary to retry.
1013 		 */
1014 		INP_INFO_RLOCK(&tcbinfo);
1015 		xig.xig_gen = tcbinfo.ipi_gencnt;
1016 		xig.xig_sogen = so_gencnt;
1017 		xig.xig_count = tcbinfo.ipi_count + pcb_count;
1018 		INP_INFO_RUNLOCK(&tcbinfo);
1019 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1020 	}
1021 	free(inp_list, M_TEMP);
1022 	return (error);
1023 }
1024 
1025 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1026     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1027 
1028 static int
1029 tcp_getcred(SYSCTL_HANDLER_ARGS)
1030 {
1031 	struct xucred xuc;
1032 	struct sockaddr_in addrs[2];
1033 	struct inpcb *inp;
1034 	int error;
1035 
1036 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1037 	if (error)
1038 		return (error);
1039 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1040 	if (error)
1041 		return (error);
1042 	INP_INFO_RLOCK(&tcbinfo);
1043 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1044 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1045 	if (inp == NULL) {
1046 		error = ENOENT;
1047 		goto outunlocked;
1048 	}
1049 	INP_WLOCK(inp);
1050 	if (inp->inp_socket == NULL) {
1051 		error = ENOENT;
1052 		goto out;
1053 	}
1054 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1055 	if (error)
1056 		goto out;
1057 	cru2x(inp->inp_socket->so_cred, &xuc);
1058 out:
1059 	INP_WUNLOCK(inp);
1060 outunlocked:
1061 	INP_INFO_RUNLOCK(&tcbinfo);
1062 	if (error == 0)
1063 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1064 	return (error);
1065 }
1066 
1067 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1068     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1069     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1070 
1071 #ifdef INET6
1072 static int
1073 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1074 {
1075 	struct xucred xuc;
1076 	struct sockaddr_in6 addrs[2];
1077 	struct inpcb *inp;
1078 	int error, mapped = 0;
1079 
1080 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1081 	if (error)
1082 		return (error);
1083 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1084 	if (error)
1085 		return (error);
1086 	if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1087 	    (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1088 		return (error);
1089 	}
1090 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1091 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1092 			mapped = 1;
1093 		else
1094 			return (EINVAL);
1095 	}
1096 
1097 	INP_INFO_RLOCK(&tcbinfo);
1098 	if (mapped == 1)
1099 		inp = in_pcblookup_hash(&tcbinfo,
1100 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1101 			addrs[1].sin6_port,
1102 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1103 			addrs[0].sin6_port,
1104 			0, NULL);
1105 	else
1106 		inp = in6_pcblookup_hash(&tcbinfo,
1107 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1108 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1109 	if (inp == NULL) {
1110 		error = ENOENT;
1111 		goto outunlocked;
1112 	}
1113 	INP_WLOCK(inp);
1114 	if (inp->inp_socket == NULL) {
1115 		error = ENOENT;
1116 		goto out;
1117 	}
1118 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1119 	if (error)
1120 		goto out;
1121 	cru2x(inp->inp_socket->so_cred, &xuc);
1122 out:
1123 	INP_WUNLOCK(inp);
1124 outunlocked:
1125 	INP_INFO_RUNLOCK(&tcbinfo);
1126 	if (error == 0)
1127 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1128 	return (error);
1129 }
1130 
1131 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1132     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1133     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1134 #endif
1135 
1136 
1137 void
1138 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1139 {
1140 	struct ip *ip = vip;
1141 	struct tcphdr *th;
1142 	struct in_addr faddr;
1143 	struct inpcb *inp;
1144 	struct tcpcb *tp;
1145 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1146 	struct icmp *icp;
1147 	struct in_conninfo inc;
1148 	tcp_seq icmp_tcp_seq;
1149 	int mtu;
1150 
1151 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1152 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1153 		return;
1154 
1155 	if (cmd == PRC_MSGSIZE)
1156 		notify = tcp_mtudisc;
1157 	else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1158 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1159 		notify = tcp_drop_syn_sent;
1160 	/*
1161 	 * Redirects don't need to be handled up here.
1162 	 */
1163 	else if (PRC_IS_REDIRECT(cmd))
1164 		return;
1165 	/*
1166 	 * Source quench is depreciated.
1167 	 */
1168 	else if (cmd == PRC_QUENCH)
1169 		return;
1170 	/*
1171 	 * Hostdead is ugly because it goes linearly through all PCBs.
1172 	 * XXX: We never get this from ICMP, otherwise it makes an
1173 	 * excellent DoS attack on machines with many connections.
1174 	 */
1175 	else if (cmd == PRC_HOSTDEAD)
1176 		ip = NULL;
1177 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1178 		return;
1179 	if (ip != NULL) {
1180 		icp = (struct icmp *)((caddr_t)ip
1181 				      - offsetof(struct icmp, icmp_ip));
1182 		th = (struct tcphdr *)((caddr_t)ip
1183 				       + (ip->ip_hl << 2));
1184 		INP_INFO_WLOCK(&tcbinfo);
1185 		inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1186 		    ip->ip_src, th->th_sport, 0, NULL);
1187 		if (inp != NULL)  {
1188 			INP_WLOCK(inp);
1189 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1190 			    !(inp->inp_vflag & INP_DROPPED) &&
1191 			    !(inp->inp_socket == NULL)) {
1192 				icmp_tcp_seq = htonl(th->th_seq);
1193 				tp = intotcpcb(inp);
1194 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1195 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1196 					if (cmd == PRC_MSGSIZE) {
1197 					    /*
1198 					     * MTU discovery:
1199 					     * If we got a needfrag set the MTU
1200 					     * in the route to the suggested new
1201 					     * value (if given) and then notify.
1202 					     */
1203 					    bzero(&inc, sizeof(inc));
1204 					    inc.inc_flags = 0;	/* IPv4 */
1205 					    inc.inc_faddr = faddr;
1206 					    inc.inc_fibnum =
1207 						inp->inp_inc.inc_fibnum;
1208 
1209 					    mtu = ntohs(icp->icmp_nextmtu);
1210 					    /*
1211 					     * If no alternative MTU was
1212 					     * proposed, try the next smaller
1213 					     * one.  ip->ip_len has already
1214 					     * been swapped in icmp_input().
1215 					     */
1216 					    if (!mtu)
1217 						mtu = ip_next_mtu(ip->ip_len,
1218 						 1);
1219 					    if (mtu < max(296, (tcp_minmss)
1220 						 + sizeof(struct tcpiphdr)))
1221 						mtu = 0;
1222 					    if (!mtu)
1223 						mtu = tcp_mssdflt
1224 						 + sizeof(struct tcpiphdr);
1225 					    /*
1226 					     * Only cache the the MTU if it
1227 					     * is smaller than the interface
1228 					     * or route MTU.  tcp_mtudisc()
1229 					     * will do right thing by itself.
1230 					     */
1231 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1232 						tcp_hc_updatemtu(&inc, mtu);
1233 					}
1234 
1235 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1236 				}
1237 			}
1238 			if (inp != NULL)
1239 				INP_WUNLOCK(inp);
1240 		} else {
1241 			inc.inc_fport = th->th_dport;
1242 			inc.inc_lport = th->th_sport;
1243 			inc.inc_faddr = faddr;
1244 			inc.inc_laddr = ip->ip_src;
1245 #ifdef INET6
1246 			inc.inc_isipv6 = 0;
1247 #endif
1248 			syncache_unreach(&inc, th);
1249 		}
1250 		INP_INFO_WUNLOCK(&tcbinfo);
1251 	} else
1252 		in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1253 }
1254 
1255 #ifdef INET6
1256 void
1257 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1258 {
1259 	struct tcphdr th;
1260 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1261 	struct ip6_hdr *ip6;
1262 	struct mbuf *m;
1263 	struct ip6ctlparam *ip6cp = NULL;
1264 	const struct sockaddr_in6 *sa6_src = NULL;
1265 	int off;
1266 	struct tcp_portonly {
1267 		u_int16_t th_sport;
1268 		u_int16_t th_dport;
1269 	} *thp;
1270 
1271 	if (sa->sa_family != AF_INET6 ||
1272 	    sa->sa_len != sizeof(struct sockaddr_in6))
1273 		return;
1274 
1275 	if (cmd == PRC_MSGSIZE)
1276 		notify = tcp_mtudisc;
1277 	else if (!PRC_IS_REDIRECT(cmd) &&
1278 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1279 		return;
1280 	/* Source quench is depreciated. */
1281 	else if (cmd == PRC_QUENCH)
1282 		return;
1283 
1284 	/* if the parameter is from icmp6, decode it. */
1285 	if (d != NULL) {
1286 		ip6cp = (struct ip6ctlparam *)d;
1287 		m = ip6cp->ip6c_m;
1288 		ip6 = ip6cp->ip6c_ip6;
1289 		off = ip6cp->ip6c_off;
1290 		sa6_src = ip6cp->ip6c_src;
1291 	} else {
1292 		m = NULL;
1293 		ip6 = NULL;
1294 		off = 0;	/* fool gcc */
1295 		sa6_src = &sa6_any;
1296 	}
1297 
1298 	if (ip6 != NULL) {
1299 		struct in_conninfo inc;
1300 		/*
1301 		 * XXX: We assume that when IPV6 is non NULL,
1302 		 * M and OFF are valid.
1303 		 */
1304 
1305 		/* check if we can safely examine src and dst ports */
1306 		if (m->m_pkthdr.len < off + sizeof(*thp))
1307 			return;
1308 
1309 		bzero(&th, sizeof(th));
1310 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1311 
1312 		in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1313 		    (struct sockaddr *)ip6cp->ip6c_src,
1314 		    th.th_sport, cmd, NULL, notify);
1315 
1316 		inc.inc_fport = th.th_dport;
1317 		inc.inc_lport = th.th_sport;
1318 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1319 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1320 		inc.inc_isipv6 = 1;
1321 		INP_INFO_WLOCK(&tcbinfo);
1322 		syncache_unreach(&inc, &th);
1323 		INP_INFO_WUNLOCK(&tcbinfo);
1324 	} else
1325 		in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1326 			      0, cmd, NULL, notify);
1327 }
1328 #endif /* INET6 */
1329 
1330 
1331 /*
1332  * Following is where TCP initial sequence number generation occurs.
1333  *
1334  * There are two places where we must use initial sequence numbers:
1335  * 1.  In SYN-ACK packets.
1336  * 2.  In SYN packets.
1337  *
1338  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1339  * tcp_syncache.c for details.
1340  *
1341  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1342  * depends on this property.  In addition, these ISNs should be
1343  * unguessable so as to prevent connection hijacking.  To satisfy
1344  * the requirements of this situation, the algorithm outlined in
1345  * RFC 1948 is used, with only small modifications.
1346  *
1347  * Implementation details:
1348  *
1349  * Time is based off the system timer, and is corrected so that it
1350  * increases by one megabyte per second.  This allows for proper
1351  * recycling on high speed LANs while still leaving over an hour
1352  * before rollover.
1353  *
1354  * As reading the *exact* system time is too expensive to be done
1355  * whenever setting up a TCP connection, we increment the time
1356  * offset in two ways.  First, a small random positive increment
1357  * is added to isn_offset for each connection that is set up.
1358  * Second, the function tcp_isn_tick fires once per clock tick
1359  * and increments isn_offset as necessary so that sequence numbers
1360  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1361  * random positive increments serve only to ensure that the same
1362  * exact sequence number is never sent out twice (as could otherwise
1363  * happen when a port is recycled in less than the system tick
1364  * interval.)
1365  *
1366  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1367  * between seeding of isn_secret.  This is normally set to zero,
1368  * as reseeding should not be necessary.
1369  *
1370  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1371  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1372  * general, this means holding an exclusive (write) lock.
1373  */
1374 
1375 #define ISN_BYTES_PER_SECOND 1048576
1376 #define ISN_STATIC_INCREMENT 4096
1377 #define ISN_RANDOM_INCREMENT (4096 - 1)
1378 
1379 static u_char isn_secret[32];
1380 static int isn_last_reseed;
1381 static u_int32_t isn_offset, isn_offset_old;
1382 static MD5_CTX isn_ctx;
1383 
1384 tcp_seq
1385 tcp_new_isn(struct tcpcb *tp)
1386 {
1387 	u_int32_t md5_buffer[4];
1388 	tcp_seq new_isn;
1389 
1390 	INP_WLOCK_ASSERT(tp->t_inpcb);
1391 
1392 	ISN_LOCK();
1393 	/* Seed if this is the first use, reseed if requested. */
1394 	if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1395 	     (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1396 		< (u_int)ticks))) {
1397 		read_random(&isn_secret, sizeof(isn_secret));
1398 		isn_last_reseed = ticks;
1399 	}
1400 
1401 	/* Compute the md5 hash and return the ISN. */
1402 	MD5Init(&isn_ctx);
1403 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1404 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1405 #ifdef INET6
1406 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1407 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1408 			  sizeof(struct in6_addr));
1409 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1410 			  sizeof(struct in6_addr));
1411 	} else
1412 #endif
1413 	{
1414 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1415 			  sizeof(struct in_addr));
1416 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1417 			  sizeof(struct in_addr));
1418 	}
1419 	MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1420 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1421 	new_isn = (tcp_seq) md5_buffer[0];
1422 	isn_offset += ISN_STATIC_INCREMENT +
1423 		(arc4random() & ISN_RANDOM_INCREMENT);
1424 	new_isn += isn_offset;
1425 	ISN_UNLOCK();
1426 	return (new_isn);
1427 }
1428 
1429 /*
1430  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1431  * to keep time flowing at a relatively constant rate.  If the random
1432  * increments have already pushed us past the projected offset, do nothing.
1433  */
1434 static void
1435 tcp_isn_tick(void *xtp)
1436 {
1437 	u_int32_t projected_offset;
1438 
1439 	ISN_LOCK();
1440 	projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1441 
1442 	if (SEQ_GT(projected_offset, isn_offset))
1443 		isn_offset = projected_offset;
1444 
1445 	isn_offset_old = isn_offset;
1446 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1447 	ISN_UNLOCK();
1448 }
1449 
1450 /*
1451  * When a specific ICMP unreachable message is received and the
1452  * connection state is SYN-SENT, drop the connection.  This behavior
1453  * is controlled by the icmp_may_rst sysctl.
1454  */
1455 struct inpcb *
1456 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1457 {
1458 	struct tcpcb *tp;
1459 
1460 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
1461 	INP_WLOCK_ASSERT(inp);
1462 
1463 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1464 	    (inp->inp_vflag & INP_DROPPED))
1465 		return (inp);
1466 
1467 	tp = intotcpcb(inp);
1468 	if (tp->t_state != TCPS_SYN_SENT)
1469 		return (inp);
1470 
1471 	tp = tcp_drop(tp, errno);
1472 	if (tp != NULL)
1473 		return (inp);
1474 	else
1475 		return (NULL);
1476 }
1477 
1478 /*
1479  * When `need fragmentation' ICMP is received, update our idea of the MSS
1480  * based on the new value in the route.  Also nudge TCP to send something,
1481  * since we know the packet we just sent was dropped.
1482  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1483  */
1484 struct inpcb *
1485 tcp_mtudisc(struct inpcb *inp, int errno)
1486 {
1487 	struct tcpcb *tp;
1488 	struct socket *so = inp->inp_socket;
1489 	u_int maxmtu;
1490 	u_int romtu;
1491 	int mss;
1492 #ifdef INET6
1493 	int isipv6;
1494 #endif /* INET6 */
1495 
1496 	INP_WLOCK_ASSERT(inp);
1497 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1498 	    (inp->inp_vflag & INP_DROPPED))
1499 		return (inp);
1500 
1501 	tp = intotcpcb(inp);
1502 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1503 
1504 #ifdef INET6
1505 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1506 #endif
1507 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1508 	romtu =
1509 #ifdef INET6
1510 	    isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
1511 #endif /* INET6 */
1512 	    tcp_maxmtu(&inp->inp_inc, NULL);
1513 	if (!maxmtu)
1514 		maxmtu = romtu;
1515 	else
1516 		maxmtu = min(maxmtu, romtu);
1517 	if (!maxmtu) {
1518 		tp->t_maxopd = tp->t_maxseg =
1519 #ifdef INET6
1520 			isipv6 ? tcp_v6mssdflt :
1521 #endif /* INET6 */
1522 			tcp_mssdflt;
1523 		return (inp);
1524 	}
1525 	mss = maxmtu -
1526 #ifdef INET6
1527 		(isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1528 #endif /* INET6 */
1529 		 sizeof(struct tcpiphdr)
1530 #ifdef INET6
1531 		 )
1532 #endif /* INET6 */
1533 		;
1534 
1535 	/*
1536 	 * XXX - The above conditional probably violates the TCP
1537 	 * spec.  The problem is that, since we don't know the
1538 	 * other end's MSS, we are supposed to use a conservative
1539 	 * default.  But, if we do that, then MTU discovery will
1540 	 * never actually take place, because the conservative
1541 	 * default is much less than the MTUs typically seen
1542 	 * on the Internet today.  For the moment, we'll sweep
1543 	 * this under the carpet.
1544 	 *
1545 	 * The conservative default might not actually be a problem
1546 	 * if the only case this occurs is when sending an initial
1547 	 * SYN with options and data to a host we've never talked
1548 	 * to before.  Then, they will reply with an MSS value which
1549 	 * will get recorded and the new parameters should get
1550 	 * recomputed.  For Further Study.
1551 	 */
1552 	if (tp->t_maxopd <= mss)
1553 		return (inp);
1554 	tp->t_maxopd = mss;
1555 
1556 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1557 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1558 		mss -= TCPOLEN_TSTAMP_APPA;
1559 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1560 	if (mss > MCLBYTES)
1561 		mss &= ~(MCLBYTES-1);
1562 #else
1563 	if (mss > MCLBYTES)
1564 		mss = mss / MCLBYTES * MCLBYTES;
1565 #endif
1566 	if (so->so_snd.sb_hiwat < mss)
1567 		mss = so->so_snd.sb_hiwat;
1568 
1569 	tp->t_maxseg = mss;
1570 
1571 	tcpstat.tcps_mturesent++;
1572 	tp->t_rtttime = 0;
1573 	tp->snd_nxt = tp->snd_una;
1574 	tcp_free_sackholes(tp);
1575 	tp->snd_recover = tp->snd_max;
1576 	if (tp->t_flags & TF_SACK_PERMIT)
1577 		EXIT_FASTRECOVERY(tp);
1578 	tcp_output_send(tp);
1579 	return (inp);
1580 }
1581 
1582 /*
1583  * Look-up the routing entry to the peer of this inpcb.  If no route
1584  * is found and it cannot be allocated, then return NULL.  This routine
1585  * is called by TCP routines that access the rmx structure and by tcp_mss
1586  * to get the interface MTU.
1587  */
1588 u_long
1589 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1590 {
1591 	struct route sro;
1592 	struct sockaddr_in *dst;
1593 	struct ifnet *ifp;
1594 	u_long maxmtu = 0;
1595 
1596 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1597 
1598 	bzero(&sro, sizeof(sro));
1599 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1600 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1601 		dst->sin_family = AF_INET;
1602 		dst->sin_len = sizeof(*dst);
1603 		dst->sin_addr = inc->inc_faddr;
1604 		in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1605 	}
1606 	if (sro.ro_rt != NULL) {
1607 		ifp = sro.ro_rt->rt_ifp;
1608 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1609 			maxmtu = ifp->if_mtu;
1610 		else
1611 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1612 
1613 		/* Report additional interface capabilities. */
1614 		if (flags != NULL) {
1615 			if (ifp->if_capenable & IFCAP_TSO4 &&
1616 			    ifp->if_hwassist & CSUM_TSO)
1617 				*flags |= CSUM_TSO;
1618 		}
1619 		RTFREE(sro.ro_rt);
1620 	}
1621 	return (maxmtu);
1622 }
1623 
1624 #ifdef INET6
1625 u_long
1626 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1627 {
1628 	struct route_in6 sro6;
1629 	struct ifnet *ifp;
1630 	u_long maxmtu = 0;
1631 
1632 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1633 
1634 	bzero(&sro6, sizeof(sro6));
1635 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1636 		sro6.ro_dst.sin6_family = AF_INET6;
1637 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1638 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1639 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1640 	}
1641 	if (sro6.ro_rt != NULL) {
1642 		ifp = sro6.ro_rt->rt_ifp;
1643 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1644 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1645 		else
1646 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1647 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1648 
1649 		/* Report additional interface capabilities. */
1650 		if (flags != NULL) {
1651 			if (ifp->if_capenable & IFCAP_TSO6 &&
1652 			    ifp->if_hwassist & CSUM_TSO)
1653 				*flags |= CSUM_TSO;
1654 		}
1655 		RTFREE(sro6.ro_rt);
1656 	}
1657 
1658 	return (maxmtu);
1659 }
1660 #endif /* INET6 */
1661 
1662 #ifdef IPSEC
1663 /* compute ESP/AH header size for TCP, including outer IP header. */
1664 size_t
1665 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1666 {
1667 	struct inpcb *inp;
1668 	struct mbuf *m;
1669 	size_t hdrsiz;
1670 	struct ip *ip;
1671 #ifdef INET6
1672 	struct ip6_hdr *ip6;
1673 #endif
1674 	struct tcphdr *th;
1675 
1676 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1677 		return (0);
1678 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1679 	if (!m)
1680 		return (0);
1681 
1682 #ifdef INET6
1683 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1684 		ip6 = mtod(m, struct ip6_hdr *);
1685 		th = (struct tcphdr *)(ip6 + 1);
1686 		m->m_pkthdr.len = m->m_len =
1687 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1688 		tcpip_fillheaders(inp, ip6, th);
1689 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1690 	} else
1691 #endif /* INET6 */
1692 	{
1693 		ip = mtod(m, struct ip *);
1694 		th = (struct tcphdr *)(ip + 1);
1695 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1696 		tcpip_fillheaders(inp, ip, th);
1697 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1698 	}
1699 
1700 	m_free(m);
1701 	return (hdrsiz);
1702 }
1703 #endif /* IPSEC */
1704 
1705 /*
1706  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1707  *
1708  * This code attempts to calculate the bandwidth-delay product as a
1709  * means of determining the optimal window size to maximize bandwidth,
1710  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1711  * routers.  This code also does a fairly good job keeping RTTs in check
1712  * across slow links like modems.  We implement an algorithm which is very
1713  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1714  * transmitter side of a TCP connection and so only effects the transmit
1715  * side of the connection.
1716  *
1717  * BACKGROUND:  TCP makes no provision for the management of buffer space
1718  * at the end points or at the intermediate routers and switches.  A TCP
1719  * stream, whether using NewReno or not, will eventually buffer as
1720  * many packets as it is able and the only reason this typically works is
1721  * due to the fairly small default buffers made available for a connection
1722  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1723  * scaling it is now fairly easy for even a single TCP connection to blow-out
1724  * all available buffer space not only on the local interface, but on
1725  * intermediate routers and switches as well.  NewReno makes a misguided
1726  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1727  * then backing off, then steadily increasing the window again until another
1728  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1729  * is only made worse as network loads increase and the idea of intentionally
1730  * blowing out network buffers is, frankly, a terrible way to manage network
1731  * resources.
1732  *
1733  * It is far better to limit the transmit window prior to the failure
1734  * condition being achieved.  There are two general ways to do this:  First
1735  * you can 'scan' through different transmit window sizes and locate the
1736  * point where the RTT stops increasing, indicating that you have filled the
1737  * pipe, then scan backwards until you note that RTT stops decreasing, then
1738  * repeat ad-infinitum.  This method works in principle but has severe
1739  * implementation issues due to RTT variances, timer granularity, and
1740  * instability in the algorithm which can lead to many false positives and
1741  * create oscillations as well as interact badly with other TCP streams
1742  * implementing the same algorithm.
1743  *
1744  * The second method is to limit the window to the bandwidth delay product
1745  * of the link.  This is the method we implement.  RTT variances and our
1746  * own manipulation of the congestion window, bwnd, can potentially
1747  * destabilize the algorithm.  For this reason we have to stabilize the
1748  * elements used to calculate the window.  We do this by using the minimum
1749  * observed RTT, the long term average of the observed bandwidth, and
1750  * by adding two segments worth of slop.  It isn't perfect but it is able
1751  * to react to changing conditions and gives us a very stable basis on
1752  * which to extend the algorithm.
1753  */
1754 void
1755 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1756 {
1757 	u_long bw;
1758 	u_long bwnd;
1759 	int save_ticks;
1760 
1761 	INP_WLOCK_ASSERT(tp->t_inpcb);
1762 
1763 	/*
1764 	 * If inflight_enable is disabled in the middle of a tcp connection,
1765 	 * make sure snd_bwnd is effectively disabled.
1766 	 */
1767 	if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
1768 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1769 		tp->snd_bandwidth = 0;
1770 		return;
1771 	}
1772 
1773 	/*
1774 	 * Figure out the bandwidth.  Due to the tick granularity this
1775 	 * is a very rough number and it MUST be averaged over a fairly
1776 	 * long period of time.  XXX we need to take into account a link
1777 	 * that is not using all available bandwidth, but for now our
1778 	 * slop will ramp us up if this case occurs and the bandwidth later
1779 	 * increases.
1780 	 *
1781 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1782 	 * effectively reset t_bw_rtttime for this case.
1783 	 */
1784 	save_ticks = ticks;
1785 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1786 		return;
1787 
1788 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1789 	    (save_ticks - tp->t_bw_rtttime);
1790 	tp->t_bw_rtttime = save_ticks;
1791 	tp->t_bw_rtseq = ack_seq;
1792 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1793 		return;
1794 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1795 
1796 	tp->snd_bandwidth = bw;
1797 
1798 	/*
1799 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1800 	 * segments.  The additional slop puts us squarely in the sweet
1801 	 * spot and also handles the bandwidth run-up case and stabilization.
1802 	 * Without the slop we could be locking ourselves into a lower
1803 	 * bandwidth.
1804 	 *
1805 	 * Situations Handled:
1806 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1807 	 *	    high speed LANs, allowing larger TCP buffers to be
1808 	 *	    specified, and also does a good job preventing
1809 	 *	    over-queueing of packets over choke points like modems
1810 	 *	    (at least for the transmit side).
1811 	 *
1812 	 *	(2) Is able to handle changing network loads (bandwidth
1813 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1814 	 *	    increases).
1815 	 *
1816 	 *	(3) Theoretically should stabilize in the face of multiple
1817 	 *	    connections implementing the same algorithm (this may need
1818 	 *	    a little work).
1819 	 *
1820 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1821 	 *	    be adjusted with a sysctl but typically only needs to be
1822 	 *	    on very slow connections.  A value no smaller then 5
1823 	 *	    should be used, but only reduce this default if you have
1824 	 *	    no other choice.
1825 	 */
1826 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1827 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
1828 #undef USERTT
1829 
1830 	if (tcp_inflight_debug > 0) {
1831 		static int ltime;
1832 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1833 			ltime = ticks;
1834 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1835 			    tp,
1836 			    bw,
1837 			    tp->t_rttbest,
1838 			    tp->t_srtt,
1839 			    bwnd
1840 			);
1841 		}
1842 	}
1843 	if ((long)bwnd < tcp_inflight_min)
1844 		bwnd = tcp_inflight_min;
1845 	if (bwnd > tcp_inflight_max)
1846 		bwnd = tcp_inflight_max;
1847 	if ((long)bwnd < tp->t_maxseg * 2)
1848 		bwnd = tp->t_maxseg * 2;
1849 	tp->snd_bwnd = bwnd;
1850 }
1851 
1852 #ifdef TCP_SIGNATURE
1853 /*
1854  * Callback function invoked by m_apply() to digest TCP segment data
1855  * contained within an mbuf chain.
1856  */
1857 static int
1858 tcp_signature_apply(void *fstate, void *data, u_int len)
1859 {
1860 
1861 	MD5Update(fstate, (u_char *)data, len);
1862 	return (0);
1863 }
1864 
1865 /*
1866  * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
1867  *
1868  * Parameters:
1869  * m		pointer to head of mbuf chain
1870  * off0		offset to TCP header within the mbuf chain
1871  * len		length of TCP segment data, excluding options
1872  * optlen	length of TCP segment options
1873  * buf		pointer to storage for computed MD5 digest
1874  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1875  *
1876  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1877  * When called from tcp_input(), we can be sure that th_sum has been
1878  * zeroed out and verified already.
1879  *
1880  * This function is for IPv4 use only. Calling this function with an
1881  * IPv6 packet in the mbuf chain will yield undefined results.
1882  *
1883  * Return 0 if successful, otherwise return -1.
1884  *
1885  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1886  * search with the destination IP address, and a 'magic SPI' to be
1887  * determined by the application. This is hardcoded elsewhere to 1179
1888  * right now. Another branch of this code exists which uses the SPD to
1889  * specify per-application flows but it is unstable.
1890  */
1891 int
1892 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
1893     u_char *buf, u_int direction)
1894 {
1895 	union sockaddr_union dst;
1896 	struct ippseudo ippseudo;
1897 	MD5_CTX ctx;
1898 	int doff;
1899 	struct ip *ip;
1900 	struct ipovly *ipovly;
1901 	struct secasvar *sav;
1902 	struct tcphdr *th;
1903 	u_short savecsum;
1904 
1905 	KASSERT(m != NULL, ("NULL mbuf chain"));
1906 	KASSERT(buf != NULL, ("NULL signature pointer"));
1907 
1908 	/* Extract the destination from the IP header in the mbuf. */
1909 	ip = mtod(m, struct ip *);
1910 	bzero(&dst, sizeof(union sockaddr_union));
1911 	dst.sa.sa_len = sizeof(struct sockaddr_in);
1912 	dst.sa.sa_family = AF_INET;
1913 	dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1914 	    ip->ip_src : ip->ip_dst;
1915 
1916 	/* Look up an SADB entry which matches the address of the peer. */
1917 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1918 	if (sav == NULL) {
1919 		printf("%s: SADB lookup failed for %s\n", __func__,
1920 		    inet_ntoa(dst.sin.sin_addr));
1921 		return (EINVAL);
1922 	}
1923 
1924 	MD5Init(&ctx);
1925 	ipovly = (struct ipovly *)ip;
1926 	th = (struct tcphdr *)((u_char *)ip + off0);
1927 	doff = off0 + sizeof(struct tcphdr) + optlen;
1928 
1929 	/*
1930 	 * Step 1: Update MD5 hash with IP pseudo-header.
1931 	 *
1932 	 * XXX The ippseudo header MUST be digested in network byte order,
1933 	 * or else we'll fail the regression test. Assume all fields we've
1934 	 * been doing arithmetic on have been in host byte order.
1935 	 * XXX One cannot depend on ipovly->ih_len here. When called from
1936 	 * tcp_output(), the underlying ip_len member has not yet been set.
1937 	 */
1938 	ippseudo.ippseudo_src = ipovly->ih_src;
1939 	ippseudo.ippseudo_dst = ipovly->ih_dst;
1940 	ippseudo.ippseudo_pad = 0;
1941 	ippseudo.ippseudo_p = IPPROTO_TCP;
1942 	ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
1943 	MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1944 
1945 	/*
1946 	 * Step 2: Update MD5 hash with TCP header, excluding options.
1947 	 * The TCP checksum must be set to zero.
1948 	 */
1949 	savecsum = th->th_sum;
1950 	th->th_sum = 0;
1951 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1952 	th->th_sum = savecsum;
1953 
1954 	/*
1955 	 * Step 3: Update MD5 hash with TCP segment data.
1956 	 *         Use m_apply() to avoid an early m_pullup().
1957 	 */
1958 	if (len > 0)
1959 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
1960 
1961 	/*
1962 	 * Step 4: Update MD5 hash with shared secret.
1963 	 */
1964 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1965 	MD5Final(buf, &ctx);
1966 
1967 	key_sa_recordxfer(sav, m);
1968 	KEY_FREESAV(&sav);
1969 	return (0);
1970 }
1971 #endif /* TCP_SIGNATURE */
1972 
1973 static int
1974 sysctl_drop(SYSCTL_HANDLER_ARGS)
1975 {
1976 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
1977 	struct sockaddr_storage addrs[2];
1978 	struct inpcb *inp;
1979 	struct tcpcb *tp;
1980 	struct tcptw *tw;
1981 	struct sockaddr_in *fin, *lin;
1982 #ifdef INET6
1983 	struct sockaddr_in6 *fin6, *lin6;
1984 	struct in6_addr f6, l6;
1985 #endif
1986 	int error;
1987 
1988 	inp = NULL;
1989 	fin = lin = NULL;
1990 #ifdef INET6
1991 	fin6 = lin6 = NULL;
1992 #endif
1993 	error = 0;
1994 
1995 	if (req->oldptr != NULL || req->oldlen != 0)
1996 		return (EINVAL);
1997 	if (req->newptr == NULL)
1998 		return (EPERM);
1999 	if (req->newlen < sizeof(addrs))
2000 		return (ENOMEM);
2001 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2002 	if (error)
2003 		return (error);
2004 
2005 	switch (addrs[0].ss_family) {
2006 #ifdef INET6
2007 	case AF_INET6:
2008 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2009 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2010 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2011 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2012 			return (EINVAL);
2013 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2014 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2015 				return (EINVAL);
2016 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2017 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2018 			fin = (struct sockaddr_in *)&addrs[0];
2019 			lin = (struct sockaddr_in *)&addrs[1];
2020 			break;
2021 		}
2022 		error = sa6_embedscope(fin6, ip6_use_defzone);
2023 		if (error)
2024 			return (error);
2025 		error = sa6_embedscope(lin6, ip6_use_defzone);
2026 		if (error)
2027 			return (error);
2028 		break;
2029 #endif
2030 	case AF_INET:
2031 		fin = (struct sockaddr_in *)&addrs[0];
2032 		lin = (struct sockaddr_in *)&addrs[1];
2033 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2034 		    lin->sin_len != sizeof(struct sockaddr_in))
2035 			return (EINVAL);
2036 		break;
2037 	default:
2038 		return (EINVAL);
2039 	}
2040 	INP_INFO_WLOCK(&tcbinfo);
2041 	switch (addrs[0].ss_family) {
2042 #ifdef INET6
2043 	case AF_INET6:
2044 		inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2045 		    &l6, lin6->sin6_port, 0, NULL);
2046 		break;
2047 #endif
2048 	case AF_INET:
2049 		inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2050 		    lin->sin_addr, lin->sin_port, 0, NULL);
2051 		break;
2052 	}
2053 	if (inp != NULL) {
2054 		INP_WLOCK(inp);
2055 		if (inp->inp_vflag & INP_TIMEWAIT) {
2056 			/*
2057 			 * XXXRW: There currently exists a state where an
2058 			 * inpcb is present, but its timewait state has been
2059 			 * discarded.  For now, don't allow dropping of this
2060 			 * type of inpcb.
2061 			 */
2062 			tw = intotw(inp);
2063 			if (tw != NULL)
2064 				tcp_twclose(tw, 0);
2065 			else
2066 				INP_WUNLOCK(inp);
2067 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2068 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2069 			tp = intotcpcb(inp);
2070 			tp = tcp_drop(tp, ECONNABORTED);
2071 			if (tp != NULL)
2072 				INP_WUNLOCK(inp);
2073 		} else
2074 			INP_WUNLOCK(inp);
2075 	} else
2076 		error = ESRCH;
2077 	INP_INFO_WUNLOCK(&tcbinfo);
2078 	return (error);
2079 }
2080 
2081 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2082     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2083     0, sysctl_drop, "", "Drop TCP connection");
2084 
2085 /*
2086  * Generate a standardized TCP log line for use throughout the
2087  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2088  * allow use in the interrupt context.
2089  *
2090  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2091  * NB: The function may return NULL if memory allocation failed.
2092  *
2093  * Due to header inclusion and ordering limitations the struct ip
2094  * and ip6_hdr pointers have to be passed as void pointers.
2095  */
2096 char *
2097 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2098     const void *ip6hdr)
2099 {
2100 	char *s, *sp;
2101 	size_t size;
2102 	struct ip *ip;
2103 #ifdef INET6
2104 	const struct ip6_hdr *ip6;
2105 
2106 	ip6 = (const struct ip6_hdr *)ip6hdr;
2107 #endif /* INET6 */
2108 	ip = (struct ip *)ip4hdr;
2109 
2110 	/*
2111 	 * The log line looks like this:
2112 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2113 	 */
2114 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2115 	    sizeof(PRINT_TH_FLAGS) + 1 +
2116 #ifdef INET6
2117 	    2 * INET6_ADDRSTRLEN;
2118 #else
2119 	    2 * INET_ADDRSTRLEN;
2120 #endif /* INET6 */
2121 
2122 	/* Is logging enabled? */
2123 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2124 		return (NULL);
2125 
2126 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2127 	if (s == NULL)
2128 		return (NULL);
2129 
2130 	strcat(s, "TCP: [");
2131 	sp = s + strlen(s);
2132 
2133 	if (inc && inc->inc_isipv6 == 0) {
2134 		inet_ntoa_r(inc->inc_faddr, sp);
2135 		sp = s + strlen(s);
2136 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2137 		sp = s + strlen(s);
2138 		inet_ntoa_r(inc->inc_laddr, sp);
2139 		sp = s + strlen(s);
2140 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2141 #ifdef INET6
2142 	} else if (inc) {
2143 		ip6_sprintf(sp, &inc->inc6_faddr);
2144 		sp = s + strlen(s);
2145 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2146 		sp = s + strlen(s);
2147 		ip6_sprintf(sp, &inc->inc6_laddr);
2148 		sp = s + strlen(s);
2149 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2150 	} else if (ip6 && th) {
2151 		ip6_sprintf(sp, &ip6->ip6_src);
2152 		sp = s + strlen(s);
2153 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2154 		sp = s + strlen(s);
2155 		ip6_sprintf(sp, &ip6->ip6_dst);
2156 		sp = s + strlen(s);
2157 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2158 #endif /* INET6 */
2159 	} else if (ip && th) {
2160 		inet_ntoa_r(ip->ip_src, sp);
2161 		sp = s + strlen(s);
2162 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2163 		sp = s + strlen(s);
2164 		inet_ntoa_r(ip->ip_dst, sp);
2165 		sp = s + strlen(s);
2166 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2167 	} else {
2168 		free(s, M_TCPLOG);
2169 		return (NULL);
2170 	}
2171 	sp = s + strlen(s);
2172 	if (th)
2173 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2174 	if (*(s + size - 1) != '\0')
2175 		panic("%s: string too long", __func__);
2176 	return (s);
2177 }
2178