xref: /freebsd/sys/netinet/tcp_subr.c (revision 9336e0699bda8a301cd2bfa37106b6ec5e32012e)
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_LOCK_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_LOCK_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 		m_freem(m->m_next);
475 		m->m_next = NULL;
476 		m->m_data = (caddr_t)ipgen;
477 		/* m_len is set later */
478 		tlen = 0;
479 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
480 #ifdef INET6
481 		if (isipv6) {
482 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
483 			nth = (struct tcphdr *)(ip6 + 1);
484 		} else
485 #endif /* INET6 */
486 	      {
487 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
488 		nth = (struct tcphdr *)(ip + 1);
489 	      }
490 		if (th != nth) {
491 			/*
492 			 * this is usually a case when an extension header
493 			 * exists between the IPv6 header and the
494 			 * TCP header.
495 			 */
496 			nth->th_sport = th->th_sport;
497 			nth->th_dport = th->th_dport;
498 		}
499 		xchg(nth->th_dport, nth->th_sport, n_short);
500 #undef xchg
501 	}
502 #ifdef INET6
503 	if (isipv6) {
504 		ip6->ip6_flow = 0;
505 		ip6->ip6_vfc = IPV6_VERSION;
506 		ip6->ip6_nxt = IPPROTO_TCP;
507 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
508 						tlen));
509 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
510 	} else
511 #endif
512 	{
513 		tlen += sizeof (struct tcpiphdr);
514 		ip->ip_len = tlen;
515 		ip->ip_ttl = ip_defttl;
516 		if (path_mtu_discovery)
517 			ip->ip_off |= IP_DF;
518 	}
519 	m->m_len = tlen;
520 	m->m_pkthdr.len = tlen;
521 	m->m_pkthdr.rcvif = NULL;
522 #ifdef MAC
523 	if (inp != NULL) {
524 		/*
525 		 * Packet is associated with a socket, so allow the
526 		 * label of the response to reflect the socket label.
527 		 */
528 		INP_LOCK_ASSERT(inp);
529 		mac_inpcb_create_mbuf(inp, m);
530 	} else {
531 		/*
532 		 * Packet is not associated with a socket, so possibly
533 		 * update the label in place.
534 		 */
535 		mac_netinet_tcp_reply(m);
536 	}
537 #endif
538 	nth->th_seq = htonl(seq);
539 	nth->th_ack = htonl(ack);
540 	nth->th_x2 = 0;
541 	nth->th_off = sizeof (struct tcphdr) >> 2;
542 	nth->th_flags = flags;
543 	if (tp != NULL)
544 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
545 	else
546 		nth->th_win = htons((u_short)win);
547 	nth->th_urp = 0;
548 #ifdef INET6
549 	if (isipv6) {
550 		nth->th_sum = 0;
551 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
552 					sizeof(struct ip6_hdr),
553 					tlen - sizeof(struct ip6_hdr));
554 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
555 		    NULL, NULL);
556 	} else
557 #endif /* INET6 */
558 	{
559 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
560 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
561 		m->m_pkthdr.csum_flags = CSUM_TCP;
562 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
563 	}
564 #ifdef TCPDEBUG
565 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
566 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
567 #endif
568 #ifdef INET6
569 	if (isipv6)
570 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
571 	else
572 #endif /* INET6 */
573 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
574 }
575 
576 /*
577  * Create a new TCP control block, making an
578  * empty reassembly queue and hooking it to the argument
579  * protocol control block.  The `inp' parameter must have
580  * come from the zone allocator set up in tcp_init().
581  */
582 struct tcpcb *
583 tcp_newtcpcb(struct inpcb *inp)
584 {
585 	struct tcpcb_mem *tm;
586 	struct tcpcb *tp;
587 #ifdef INET6
588 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
589 #endif /* INET6 */
590 
591 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
592 	if (tm == NULL)
593 		return (NULL);
594 	tp = &tm->tcb;
595 	tp->t_timers = &tm->tt;
596 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
597 	tp->t_maxseg = tp->t_maxopd =
598 #ifdef INET6
599 		isipv6 ? tcp_v6mssdflt :
600 #endif /* INET6 */
601 		tcp_mssdflt;
602 
603 	/* Set up our timeouts. */
604 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
605 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
606 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
607 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
608 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
609 
610 	if (tcp_do_rfc1323)
611 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
612 	if (tcp_do_sack)
613 		tp->t_flags |= TF_SACK_PERMIT;
614 	TAILQ_INIT(&tp->snd_holes);
615 	tp->t_inpcb = inp;	/* XXX */
616 	/*
617 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
618 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
619 	 * reasonable initial retransmit time.
620 	 */
621 	tp->t_srtt = TCPTV_SRTTBASE;
622 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
623 	tp->t_rttmin = tcp_rexmit_min;
624 	tp->t_rxtcur = TCPTV_RTOBASE;
625 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
626 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
627 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
628 	tp->t_rcvtime = ticks;
629 	tp->t_bw_rtttime = ticks;
630 	/*
631 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
632 	 * because the socket may be bound to an IPv6 wildcard address,
633 	 * which may match an IPv4-mapped IPv6 address.
634 	 */
635 	inp->inp_ip_ttl = ip_defttl;
636 	inp->inp_ppcb = tp;
637 	return (tp);		/* XXX */
638 }
639 
640 /*
641  * Drop a TCP connection, reporting
642  * the specified error.  If connection is synchronized,
643  * then send a RST to peer.
644  */
645 struct tcpcb *
646 tcp_drop(struct tcpcb *tp, int errno)
647 {
648 	struct socket *so = tp->t_inpcb->inp_socket;
649 
650 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
651 	INP_LOCK_ASSERT(tp->t_inpcb);
652 
653 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
654 		tp->t_state = TCPS_CLOSED;
655 		(void) tcp_output_reset(tp);
656 		tcpstat.tcps_drops++;
657 	} else
658 		tcpstat.tcps_conndrops++;
659 	if (errno == ETIMEDOUT && tp->t_softerror)
660 		errno = tp->t_softerror;
661 	so->so_error = errno;
662 	return (tcp_close(tp));
663 }
664 
665 void
666 tcp_discardcb(struct tcpcb *tp)
667 {
668 	struct tseg_qent *q;
669 	struct inpcb *inp = tp->t_inpcb;
670 	struct socket *so = inp->inp_socket;
671 #ifdef INET6
672 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
673 #endif /* INET6 */
674 
675 	INP_LOCK_ASSERT(inp);
676 
677 	/*
678 	 * Make sure that all of our timers are stopped before we
679 	 * delete the PCB.
680 	 */
681 	callout_stop(&tp->t_timers->tt_rexmt);
682 	callout_stop(&tp->t_timers->tt_persist);
683 	callout_stop(&tp->t_timers->tt_keep);
684 	callout_stop(&tp->t_timers->tt_2msl);
685 	callout_stop(&tp->t_timers->tt_delack);
686 
687 	/*
688 	 * If we got enough samples through the srtt filter,
689 	 * save the rtt and rttvar in the routing entry.
690 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
691 	 * 4 samples is enough for the srtt filter to converge
692 	 * to within enough % of the correct value; fewer samples
693 	 * and we could save a bogus rtt. The danger is not high
694 	 * as tcp quickly recovers from everything.
695 	 * XXX: Works very well but needs some more statistics!
696 	 */
697 	if (tp->t_rttupdated >= 4) {
698 		struct hc_metrics_lite metrics;
699 		u_long ssthresh;
700 
701 		bzero(&metrics, sizeof(metrics));
702 		/*
703 		 * Update the ssthresh always when the conditions below
704 		 * are satisfied. This gives us better new start value
705 		 * for the congestion avoidance for new connections.
706 		 * ssthresh is only set if packet loss occured on a session.
707 		 *
708 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
709 		 * being torn down.  Ideally this code would not use 'so'.
710 		 */
711 		ssthresh = tp->snd_ssthresh;
712 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
713 			/*
714 			 * convert the limit from user data bytes to
715 			 * packets then to packet data bytes.
716 			 */
717 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
718 			if (ssthresh < 2)
719 				ssthresh = 2;
720 			ssthresh *= (u_long)(tp->t_maxseg +
721 #ifdef INET6
722 				      (isipv6 ? sizeof (struct ip6_hdr) +
723 					       sizeof (struct tcphdr) :
724 #endif
725 				       sizeof (struct tcpiphdr)
726 #ifdef INET6
727 				       )
728 #endif
729 				      );
730 		} else
731 			ssthresh = 0;
732 		metrics.rmx_ssthresh = ssthresh;
733 
734 		metrics.rmx_rtt = tp->t_srtt;
735 		metrics.rmx_rttvar = tp->t_rttvar;
736 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
737 		metrics.rmx_bandwidth = tp->snd_bandwidth;
738 		metrics.rmx_cwnd = tp->snd_cwnd;
739 		metrics.rmx_sendpipe = 0;
740 		metrics.rmx_recvpipe = 0;
741 
742 		tcp_hc_update(&inp->inp_inc, &metrics);
743 	}
744 
745 	/* free the reassembly queue, if any */
746 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
747 		LIST_REMOVE(q, tqe_q);
748 		m_freem(q->tqe_m);
749 		uma_zfree(tcp_reass_zone, q);
750 		tp->t_segqlen--;
751 		tcp_reass_qsize--;
752 	}
753 	/* Disconnect offload device, if any. */
754 	tcp_offload_detach(tp);
755 
756 	tcp_free_sackholes(tp);
757 	inp->inp_ppcb = NULL;
758 	tp->t_inpcb = NULL;
759 	uma_zfree(tcpcb_zone, tp);
760 }
761 
762 /*
763  * Attempt to close a TCP control block, marking it as dropped, and freeing
764  * the socket if we hold the only reference.
765  */
766 struct tcpcb *
767 tcp_close(struct tcpcb *tp)
768 {
769 	struct inpcb *inp = tp->t_inpcb;
770 	struct socket *so;
771 
772 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
773 	INP_LOCK_ASSERT(inp);
774 
775 	/* Notify any offload devices of listener close */
776 	if (tp->t_state == TCPS_LISTEN)
777 		tcp_offload_listen_close(tp);
778 	in_pcbdrop(inp);
779 	tcpstat.tcps_closed++;
780 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
781 	so = inp->inp_socket;
782 	soisdisconnected(so);
783 	if (inp->inp_vflag & INP_SOCKREF) {
784 		KASSERT(so->so_state & SS_PROTOREF,
785 		    ("tcp_close: !SS_PROTOREF"));
786 		inp->inp_vflag &= ~INP_SOCKREF;
787 		INP_UNLOCK(inp);
788 		ACCEPT_LOCK();
789 		SOCK_LOCK(so);
790 		so->so_state &= ~SS_PROTOREF;
791 		sofree(so);
792 		return (NULL);
793 	}
794 	return (tp);
795 }
796 
797 void
798 tcp_drain(void)
799 {
800 
801 	if (do_tcpdrain) {
802 		struct inpcb *inpb;
803 		struct tcpcb *tcpb;
804 		struct tseg_qent *te;
805 
806 	/*
807 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
808 	 * if there is one...
809 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
810 	 *      reassembly queue should be flushed, but in a situation
811 	 *	where we're really low on mbufs, this is potentially
812 	 *	usefull.
813 	 */
814 		INP_INFO_RLOCK(&tcbinfo);
815 		LIST_FOREACH(inpb, tcbinfo.ipi_listhead, inp_list) {
816 			if (inpb->inp_vflag & INP_TIMEWAIT)
817 				continue;
818 			INP_LOCK(inpb);
819 			if ((tcpb = intotcpcb(inpb)) != NULL) {
820 				while ((te = LIST_FIRST(&tcpb->t_segq))
821 			            != NULL) {
822 					LIST_REMOVE(te, tqe_q);
823 					m_freem(te->tqe_m);
824 					uma_zfree(tcp_reass_zone, te);
825 					tcpb->t_segqlen--;
826 					tcp_reass_qsize--;
827 				}
828 				tcp_clean_sackreport(tcpb);
829 			}
830 			INP_UNLOCK(inpb);
831 		}
832 		INP_INFO_RUNLOCK(&tcbinfo);
833 	}
834 }
835 
836 /*
837  * Notify a tcp user of an asynchronous error;
838  * store error as soft error, but wake up user
839  * (for now, won't do anything until can select for soft error).
840  *
841  * Do not wake up user since there currently is no mechanism for
842  * reporting soft errors (yet - a kqueue filter may be added).
843  */
844 static struct inpcb *
845 tcp_notify(struct inpcb *inp, int error)
846 {
847 	struct tcpcb *tp;
848 
849 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
850 	INP_LOCK_ASSERT(inp);
851 
852 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
853 	    (inp->inp_vflag & INP_DROPPED))
854 		return (inp);
855 
856 	tp = intotcpcb(inp);
857 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
858 
859 	/*
860 	 * Ignore some errors if we are hooked up.
861 	 * If connection hasn't completed, has retransmitted several times,
862 	 * and receives a second error, give up now.  This is better
863 	 * than waiting a long time to establish a connection that
864 	 * can never complete.
865 	 */
866 	if (tp->t_state == TCPS_ESTABLISHED &&
867 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
868 	     error == EHOSTDOWN)) {
869 		return (inp);
870 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
871 	    tp->t_softerror) {
872 		tp = tcp_drop(tp, error);
873 		if (tp != NULL)
874 			return (inp);
875 		else
876 			return (NULL);
877 	} else {
878 		tp->t_softerror = error;
879 		return (inp);
880 	}
881 #if 0
882 	wakeup( &so->so_timeo);
883 	sorwakeup(so);
884 	sowwakeup(so);
885 #endif
886 }
887 
888 static int
889 tcp_pcblist(SYSCTL_HANDLER_ARGS)
890 {
891 	int error, i, m, n, pcb_count;
892 	struct inpcb *inp, **inp_list;
893 	inp_gen_t gencnt;
894 	struct xinpgen xig;
895 
896 	/*
897 	 * The process of preparing the TCB list is too time-consuming and
898 	 * resource-intensive to repeat twice on every request.
899 	 */
900 	if (req->oldptr == NULL) {
901 		m = syncache_pcbcount();
902 		n = tcbinfo.ipi_count;
903 		req->oldidx = 2 * (sizeof xig)
904 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
905 		return (0);
906 	}
907 
908 	if (req->newptr != NULL)
909 		return (EPERM);
910 
911 	/*
912 	 * OK, now we're committed to doing something.
913 	 */
914 	INP_INFO_RLOCK(&tcbinfo);
915 	gencnt = tcbinfo.ipi_gencnt;
916 	n = tcbinfo.ipi_count;
917 	INP_INFO_RUNLOCK(&tcbinfo);
918 
919 	m = syncache_pcbcount();
920 
921 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
922 		+ (n + m) * sizeof(struct xtcpcb));
923 	if (error != 0)
924 		return (error);
925 
926 	xig.xig_len = sizeof xig;
927 	xig.xig_count = n + m;
928 	xig.xig_gen = gencnt;
929 	xig.xig_sogen = so_gencnt;
930 	error = SYSCTL_OUT(req, &xig, sizeof xig);
931 	if (error)
932 		return (error);
933 
934 	error = syncache_pcblist(req, m, &pcb_count);
935 	if (error)
936 		return (error);
937 
938 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
939 	if (inp_list == NULL)
940 		return (ENOMEM);
941 
942 	INP_INFO_RLOCK(&tcbinfo);
943 	for (inp = LIST_FIRST(tcbinfo.ipi_listhead), i = 0; inp != NULL && i
944 	    < n; inp = LIST_NEXT(inp, inp_list)) {
945 		INP_LOCK(inp);
946 		if (inp->inp_gencnt <= gencnt) {
947 			/*
948 			 * XXX: This use of cr_cansee(), introduced with
949 			 * TCP state changes, is not quite right, but for
950 			 * now, better than nothing.
951 			 */
952 			if (inp->inp_vflag & INP_TIMEWAIT) {
953 				if (intotw(inp) != NULL)
954 					error = cr_cansee(req->td->td_ucred,
955 					    intotw(inp)->tw_cred);
956 				else
957 					error = EINVAL;	/* Skip this inp. */
958 			} else
959 				error = cr_canseesocket(req->td->td_ucred,
960 				    inp->inp_socket);
961 			if (error == 0)
962 				inp_list[i++] = inp;
963 		}
964 		INP_UNLOCK(inp);
965 	}
966 	INP_INFO_RUNLOCK(&tcbinfo);
967 	n = i;
968 
969 	error = 0;
970 	for (i = 0; i < n; i++) {
971 		inp = inp_list[i];
972 		INP_LOCK(inp);
973 		if (inp->inp_gencnt <= gencnt) {
974 			struct xtcpcb xt;
975 			void *inp_ppcb;
976 
977 			bzero(&xt, sizeof(xt));
978 			xt.xt_len = sizeof xt;
979 			/* XXX should avoid extra copy */
980 			bcopy(inp, &xt.xt_inp, sizeof *inp);
981 			inp_ppcb = inp->inp_ppcb;
982 			if (inp_ppcb == NULL)
983 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
984 			else if (inp->inp_vflag & INP_TIMEWAIT) {
985 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
986 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
987 			} else
988 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
989 			if (inp->inp_socket != NULL)
990 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
991 			else {
992 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
993 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
994 			}
995 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
996 			INP_UNLOCK(inp);
997 			error = SYSCTL_OUT(req, &xt, sizeof xt);
998 		} else
999 			INP_UNLOCK(inp);
1000 
1001 	}
1002 	if (!error) {
1003 		/*
1004 		 * Give the user an updated idea of our state.
1005 		 * If the generation differs from what we told
1006 		 * her before, she knows that something happened
1007 		 * while we were processing this request, and it
1008 		 * might be necessary to retry.
1009 		 */
1010 		INP_INFO_RLOCK(&tcbinfo);
1011 		xig.xig_gen = tcbinfo.ipi_gencnt;
1012 		xig.xig_sogen = so_gencnt;
1013 		xig.xig_count = tcbinfo.ipi_count + pcb_count;
1014 		INP_INFO_RUNLOCK(&tcbinfo);
1015 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1016 	}
1017 	free(inp_list, M_TEMP);
1018 	return (error);
1019 }
1020 
1021 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1022     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1023 
1024 static int
1025 tcp_getcred(SYSCTL_HANDLER_ARGS)
1026 {
1027 	struct xucred xuc;
1028 	struct sockaddr_in addrs[2];
1029 	struct inpcb *inp;
1030 	int error;
1031 
1032 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1033 	if (error)
1034 		return (error);
1035 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1036 	if (error)
1037 		return (error);
1038 	INP_INFO_RLOCK(&tcbinfo);
1039 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1040 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1041 	if (inp == NULL) {
1042 		error = ENOENT;
1043 		goto outunlocked;
1044 	}
1045 	INP_LOCK(inp);
1046 	if (inp->inp_socket == NULL) {
1047 		error = ENOENT;
1048 		goto out;
1049 	}
1050 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1051 	if (error)
1052 		goto out;
1053 	cru2x(inp->inp_socket->so_cred, &xuc);
1054 out:
1055 	INP_UNLOCK(inp);
1056 outunlocked:
1057 	INP_INFO_RUNLOCK(&tcbinfo);
1058 	if (error == 0)
1059 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1060 	return (error);
1061 }
1062 
1063 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1064     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1065     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1066 
1067 #ifdef INET6
1068 static int
1069 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1070 {
1071 	struct xucred xuc;
1072 	struct sockaddr_in6 addrs[2];
1073 	struct inpcb *inp;
1074 	int error, mapped = 0;
1075 
1076 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1077 	if (error)
1078 		return (error);
1079 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1080 	if (error)
1081 		return (error);
1082 	if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1083 	    (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1084 		return (error);
1085 	}
1086 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1087 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1088 			mapped = 1;
1089 		else
1090 			return (EINVAL);
1091 	}
1092 
1093 	INP_INFO_RLOCK(&tcbinfo);
1094 	if (mapped == 1)
1095 		inp = in_pcblookup_hash(&tcbinfo,
1096 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1097 			addrs[1].sin6_port,
1098 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1099 			addrs[0].sin6_port,
1100 			0, NULL);
1101 	else
1102 		inp = in6_pcblookup_hash(&tcbinfo,
1103 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1104 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1105 	if (inp == NULL) {
1106 		error = ENOENT;
1107 		goto outunlocked;
1108 	}
1109 	INP_LOCK(inp);
1110 	if (inp->inp_socket == NULL) {
1111 		error = ENOENT;
1112 		goto out;
1113 	}
1114 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1115 	if (error)
1116 		goto out;
1117 	cru2x(inp->inp_socket->so_cred, &xuc);
1118 out:
1119 	INP_UNLOCK(inp);
1120 outunlocked:
1121 	INP_INFO_RUNLOCK(&tcbinfo);
1122 	if (error == 0)
1123 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1124 	return (error);
1125 }
1126 
1127 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1128     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1129     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1130 #endif
1131 
1132 
1133 void
1134 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1135 {
1136 	struct ip *ip = vip;
1137 	struct tcphdr *th;
1138 	struct in_addr faddr;
1139 	struct inpcb *inp;
1140 	struct tcpcb *tp;
1141 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1142 	struct icmp *icp;
1143 	struct in_conninfo inc;
1144 	tcp_seq icmp_tcp_seq;
1145 	int mtu;
1146 
1147 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1148 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1149 		return;
1150 
1151 	if (cmd == PRC_MSGSIZE)
1152 		notify = tcp_mtudisc;
1153 	else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1154 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1155 		notify = tcp_drop_syn_sent;
1156 	/*
1157 	 * Redirects don't need to be handled up here.
1158 	 */
1159 	else if (PRC_IS_REDIRECT(cmd))
1160 		return;
1161 	/*
1162 	 * Source quench is depreciated.
1163 	 */
1164 	else if (cmd == PRC_QUENCH)
1165 		return;
1166 	/*
1167 	 * Hostdead is ugly because it goes linearly through all PCBs.
1168 	 * XXX: We never get this from ICMP, otherwise it makes an
1169 	 * excellent DoS attack on machines with many connections.
1170 	 */
1171 	else if (cmd == PRC_HOSTDEAD)
1172 		ip = NULL;
1173 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1174 		return;
1175 	if (ip != NULL) {
1176 		icp = (struct icmp *)((caddr_t)ip
1177 				      - offsetof(struct icmp, icmp_ip));
1178 		th = (struct tcphdr *)((caddr_t)ip
1179 				       + (ip->ip_hl << 2));
1180 		INP_INFO_WLOCK(&tcbinfo);
1181 		inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1182 		    ip->ip_src, th->th_sport, 0, NULL);
1183 		if (inp != NULL)  {
1184 			INP_LOCK(inp);
1185 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1186 			    !(inp->inp_vflag & INP_DROPPED) &&
1187 			    !(inp->inp_socket == NULL)) {
1188 				icmp_tcp_seq = htonl(th->th_seq);
1189 				tp = intotcpcb(inp);
1190 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1191 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1192 					if (cmd == PRC_MSGSIZE) {
1193 					    /*
1194 					     * MTU discovery:
1195 					     * If we got a needfrag set the MTU
1196 					     * in the route to the suggested new
1197 					     * value (if given) and then notify.
1198 					     */
1199 					    bzero(&inc, sizeof(inc));
1200 					    inc.inc_flags = 0;	/* IPv4 */
1201 					    inc.inc_faddr = faddr;
1202 
1203 					    mtu = ntohs(icp->icmp_nextmtu);
1204 					    /*
1205 					     * If no alternative MTU was
1206 					     * proposed, try the next smaller
1207 					     * one.  ip->ip_len has already
1208 					     * been swapped in icmp_input().
1209 					     */
1210 					    if (!mtu)
1211 						mtu = ip_next_mtu(ip->ip_len,
1212 						 1);
1213 					    if (mtu < max(296, (tcp_minmss)
1214 						 + sizeof(struct tcpiphdr)))
1215 						mtu = 0;
1216 					    if (!mtu)
1217 						mtu = tcp_mssdflt
1218 						 + sizeof(struct tcpiphdr);
1219 					    /*
1220 					     * Only cache the the MTU if it
1221 					     * is smaller than the interface
1222 					     * or route MTU.  tcp_mtudisc()
1223 					     * will do right thing by itself.
1224 					     */
1225 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1226 						tcp_hc_updatemtu(&inc, mtu);
1227 					}
1228 
1229 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1230 				}
1231 			}
1232 			if (inp != NULL)
1233 				INP_UNLOCK(inp);
1234 		} else {
1235 			inc.inc_fport = th->th_dport;
1236 			inc.inc_lport = th->th_sport;
1237 			inc.inc_faddr = faddr;
1238 			inc.inc_laddr = ip->ip_src;
1239 #ifdef INET6
1240 			inc.inc_isipv6 = 0;
1241 #endif
1242 			syncache_unreach(&inc, th);
1243 		}
1244 		INP_INFO_WUNLOCK(&tcbinfo);
1245 	} else
1246 		in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1247 }
1248 
1249 #ifdef INET6
1250 void
1251 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1252 {
1253 	struct tcphdr th;
1254 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1255 	struct ip6_hdr *ip6;
1256 	struct mbuf *m;
1257 	struct ip6ctlparam *ip6cp = NULL;
1258 	const struct sockaddr_in6 *sa6_src = NULL;
1259 	int off;
1260 	struct tcp_portonly {
1261 		u_int16_t th_sport;
1262 		u_int16_t th_dport;
1263 	} *thp;
1264 
1265 	if (sa->sa_family != AF_INET6 ||
1266 	    sa->sa_len != sizeof(struct sockaddr_in6))
1267 		return;
1268 
1269 	if (cmd == PRC_MSGSIZE)
1270 		notify = tcp_mtudisc;
1271 	else if (!PRC_IS_REDIRECT(cmd) &&
1272 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1273 		return;
1274 	/* Source quench is depreciated. */
1275 	else if (cmd == PRC_QUENCH)
1276 		return;
1277 
1278 	/* if the parameter is from icmp6, decode it. */
1279 	if (d != NULL) {
1280 		ip6cp = (struct ip6ctlparam *)d;
1281 		m = ip6cp->ip6c_m;
1282 		ip6 = ip6cp->ip6c_ip6;
1283 		off = ip6cp->ip6c_off;
1284 		sa6_src = ip6cp->ip6c_src;
1285 	} else {
1286 		m = NULL;
1287 		ip6 = NULL;
1288 		off = 0;	/* fool gcc */
1289 		sa6_src = &sa6_any;
1290 	}
1291 
1292 	if (ip6 != NULL) {
1293 		struct in_conninfo inc;
1294 		/*
1295 		 * XXX: We assume that when IPV6 is non NULL,
1296 		 * M and OFF are valid.
1297 		 */
1298 
1299 		/* check if we can safely examine src and dst ports */
1300 		if (m->m_pkthdr.len < off + sizeof(*thp))
1301 			return;
1302 
1303 		bzero(&th, sizeof(th));
1304 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1305 
1306 		in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1307 		    (struct sockaddr *)ip6cp->ip6c_src,
1308 		    th.th_sport, cmd, NULL, notify);
1309 
1310 		inc.inc_fport = th.th_dport;
1311 		inc.inc_lport = th.th_sport;
1312 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1313 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1314 		inc.inc_isipv6 = 1;
1315 		INP_INFO_WLOCK(&tcbinfo);
1316 		syncache_unreach(&inc, &th);
1317 		INP_INFO_WUNLOCK(&tcbinfo);
1318 	} else
1319 		in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1320 			      0, cmd, NULL, notify);
1321 }
1322 #endif /* INET6 */
1323 
1324 
1325 /*
1326  * Following is where TCP initial sequence number generation occurs.
1327  *
1328  * There are two places where we must use initial sequence numbers:
1329  * 1.  In SYN-ACK packets.
1330  * 2.  In SYN packets.
1331  *
1332  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1333  * tcp_syncache.c for details.
1334  *
1335  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1336  * depends on this property.  In addition, these ISNs should be
1337  * unguessable so as to prevent connection hijacking.  To satisfy
1338  * the requirements of this situation, the algorithm outlined in
1339  * RFC 1948 is used, with only small modifications.
1340  *
1341  * Implementation details:
1342  *
1343  * Time is based off the system timer, and is corrected so that it
1344  * increases by one megabyte per second.  This allows for proper
1345  * recycling on high speed LANs while still leaving over an hour
1346  * before rollover.
1347  *
1348  * As reading the *exact* system time is too expensive to be done
1349  * whenever setting up a TCP connection, we increment the time
1350  * offset in two ways.  First, a small random positive increment
1351  * is added to isn_offset for each connection that is set up.
1352  * Second, the function tcp_isn_tick fires once per clock tick
1353  * and increments isn_offset as necessary so that sequence numbers
1354  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1355  * random positive increments serve only to ensure that the same
1356  * exact sequence number is never sent out twice (as could otherwise
1357  * happen when a port is recycled in less than the system tick
1358  * interval.)
1359  *
1360  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1361  * between seeding of isn_secret.  This is normally set to zero,
1362  * as reseeding should not be necessary.
1363  *
1364  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1365  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1366  * general, this means holding an exclusive (write) lock.
1367  */
1368 
1369 #define ISN_BYTES_PER_SECOND 1048576
1370 #define ISN_STATIC_INCREMENT 4096
1371 #define ISN_RANDOM_INCREMENT (4096 - 1)
1372 
1373 static u_char isn_secret[32];
1374 static int isn_last_reseed;
1375 static u_int32_t isn_offset, isn_offset_old;
1376 static MD5_CTX isn_ctx;
1377 
1378 tcp_seq
1379 tcp_new_isn(struct tcpcb *tp)
1380 {
1381 	u_int32_t md5_buffer[4];
1382 	tcp_seq new_isn;
1383 
1384 	INP_LOCK_ASSERT(tp->t_inpcb);
1385 
1386 	ISN_LOCK();
1387 	/* Seed if this is the first use, reseed if requested. */
1388 	if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1389 	     (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1390 		< (u_int)ticks))) {
1391 		read_random(&isn_secret, sizeof(isn_secret));
1392 		isn_last_reseed = ticks;
1393 	}
1394 
1395 	/* Compute the md5 hash and return the ISN. */
1396 	MD5Init(&isn_ctx);
1397 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1398 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1399 #ifdef INET6
1400 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1401 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1402 			  sizeof(struct in6_addr));
1403 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1404 			  sizeof(struct in6_addr));
1405 	} else
1406 #endif
1407 	{
1408 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1409 			  sizeof(struct in_addr));
1410 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1411 			  sizeof(struct in_addr));
1412 	}
1413 	MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1414 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1415 	new_isn = (tcp_seq) md5_buffer[0];
1416 	isn_offset += ISN_STATIC_INCREMENT +
1417 		(arc4random() & ISN_RANDOM_INCREMENT);
1418 	new_isn += isn_offset;
1419 	ISN_UNLOCK();
1420 	return (new_isn);
1421 }
1422 
1423 /*
1424  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1425  * to keep time flowing at a relatively constant rate.  If the random
1426  * increments have already pushed us past the projected offset, do nothing.
1427  */
1428 static void
1429 tcp_isn_tick(void *xtp)
1430 {
1431 	u_int32_t projected_offset;
1432 
1433 	ISN_LOCK();
1434 	projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1435 
1436 	if (SEQ_GT(projected_offset, isn_offset))
1437 		isn_offset = projected_offset;
1438 
1439 	isn_offset_old = isn_offset;
1440 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1441 	ISN_UNLOCK();
1442 }
1443 
1444 /*
1445  * When a specific ICMP unreachable message is received and the
1446  * connection state is SYN-SENT, drop the connection.  This behavior
1447  * is controlled by the icmp_may_rst sysctl.
1448  */
1449 struct inpcb *
1450 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1451 {
1452 	struct tcpcb *tp;
1453 
1454 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
1455 	INP_LOCK_ASSERT(inp);
1456 
1457 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1458 	    (inp->inp_vflag & INP_DROPPED))
1459 		return (inp);
1460 
1461 	tp = intotcpcb(inp);
1462 	if (tp->t_state != TCPS_SYN_SENT)
1463 		return (inp);
1464 
1465 	tp = tcp_drop(tp, errno);
1466 	if (tp != NULL)
1467 		return (inp);
1468 	else
1469 		return (NULL);
1470 }
1471 
1472 /*
1473  * When `need fragmentation' ICMP is received, update our idea of the MSS
1474  * based on the new value in the route.  Also nudge TCP to send something,
1475  * since we know the packet we just sent was dropped.
1476  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1477  */
1478 struct inpcb *
1479 tcp_mtudisc(struct inpcb *inp, int errno)
1480 {
1481 	struct tcpcb *tp;
1482 	struct socket *so = inp->inp_socket;
1483 	u_int maxmtu;
1484 	u_int romtu;
1485 	int mss;
1486 #ifdef INET6
1487 	int isipv6;
1488 #endif /* INET6 */
1489 
1490 	INP_LOCK_ASSERT(inp);
1491 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1492 	    (inp->inp_vflag & INP_DROPPED))
1493 		return (inp);
1494 
1495 	tp = intotcpcb(inp);
1496 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1497 
1498 #ifdef INET6
1499 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1500 #endif
1501 	maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1502 	romtu =
1503 #ifdef INET6
1504 	    isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
1505 #endif /* INET6 */
1506 	    tcp_maxmtu(&inp->inp_inc, NULL);
1507 	if (!maxmtu)
1508 		maxmtu = romtu;
1509 	else
1510 		maxmtu = min(maxmtu, romtu);
1511 	if (!maxmtu) {
1512 		tp->t_maxopd = tp->t_maxseg =
1513 #ifdef INET6
1514 			isipv6 ? tcp_v6mssdflt :
1515 #endif /* INET6 */
1516 			tcp_mssdflt;
1517 		return (inp);
1518 	}
1519 	mss = maxmtu -
1520 #ifdef INET6
1521 		(isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1522 #endif /* INET6 */
1523 		 sizeof(struct tcpiphdr)
1524 #ifdef INET6
1525 		 )
1526 #endif /* INET6 */
1527 		;
1528 
1529 	/*
1530 	 * XXX - The above conditional probably violates the TCP
1531 	 * spec.  The problem is that, since we don't know the
1532 	 * other end's MSS, we are supposed to use a conservative
1533 	 * default.  But, if we do that, then MTU discovery will
1534 	 * never actually take place, because the conservative
1535 	 * default is much less than the MTUs typically seen
1536 	 * on the Internet today.  For the moment, we'll sweep
1537 	 * this under the carpet.
1538 	 *
1539 	 * The conservative default might not actually be a problem
1540 	 * if the only case this occurs is when sending an initial
1541 	 * SYN with options and data to a host we've never talked
1542 	 * to before.  Then, they will reply with an MSS value which
1543 	 * will get recorded and the new parameters should get
1544 	 * recomputed.  For Further Study.
1545 	 */
1546 	if (tp->t_maxopd <= mss)
1547 		return (inp);
1548 	tp->t_maxopd = mss;
1549 
1550 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1551 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1552 		mss -= TCPOLEN_TSTAMP_APPA;
1553 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1554 	if (mss > MCLBYTES)
1555 		mss &= ~(MCLBYTES-1);
1556 #else
1557 	if (mss > MCLBYTES)
1558 		mss = mss / MCLBYTES * MCLBYTES;
1559 #endif
1560 	if (so->so_snd.sb_hiwat < mss)
1561 		mss = so->so_snd.sb_hiwat;
1562 
1563 	tp->t_maxseg = mss;
1564 
1565 	tcpstat.tcps_mturesent++;
1566 	tp->t_rtttime = 0;
1567 	tp->snd_nxt = tp->snd_una;
1568 	tcp_free_sackholes(tp);
1569 	tp->snd_recover = tp->snd_max;
1570 	if (tp->t_flags & TF_SACK_PERMIT)
1571 		EXIT_FASTRECOVERY(tp);
1572 	tcp_output_send(tp);
1573 	return (inp);
1574 }
1575 
1576 /*
1577  * Look-up the routing entry to the peer of this inpcb.  If no route
1578  * is found and it cannot be allocated, then return NULL.  This routine
1579  * is called by TCP routines that access the rmx structure and by tcp_mss
1580  * to get the interface MTU.
1581  */
1582 u_long
1583 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1584 {
1585 	struct route sro;
1586 	struct sockaddr_in *dst;
1587 	struct ifnet *ifp;
1588 	u_long maxmtu = 0;
1589 
1590 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1591 
1592 	bzero(&sro, sizeof(sro));
1593 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1594 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1595 		dst->sin_family = AF_INET;
1596 		dst->sin_len = sizeof(*dst);
1597 		dst->sin_addr = inc->inc_faddr;
1598 		rtalloc_ign(&sro, RTF_CLONING);
1599 	}
1600 	if (sro.ro_rt != NULL) {
1601 		ifp = sro.ro_rt->rt_ifp;
1602 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1603 			maxmtu = ifp->if_mtu;
1604 		else
1605 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1606 
1607 		/* Report additional interface capabilities. */
1608 		if (flags != NULL) {
1609 			if (ifp->if_capenable & IFCAP_TSO4 &&
1610 			    ifp->if_hwassist & CSUM_TSO)
1611 				*flags |= CSUM_TSO;
1612 		}
1613 		RTFREE(sro.ro_rt);
1614 	}
1615 	return (maxmtu);
1616 }
1617 
1618 #ifdef INET6
1619 u_long
1620 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1621 {
1622 	struct route_in6 sro6;
1623 	struct ifnet *ifp;
1624 	u_long maxmtu = 0;
1625 
1626 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1627 
1628 	bzero(&sro6, sizeof(sro6));
1629 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1630 		sro6.ro_dst.sin6_family = AF_INET6;
1631 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1632 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1633 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1634 	}
1635 	if (sro6.ro_rt != NULL) {
1636 		ifp = sro6.ro_rt->rt_ifp;
1637 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1638 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1639 		else
1640 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1641 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1642 
1643 		/* Report additional interface capabilities. */
1644 		if (flags != NULL) {
1645 			if (ifp->if_capenable & IFCAP_TSO6 &&
1646 			    ifp->if_hwassist & CSUM_TSO)
1647 				*flags |= CSUM_TSO;
1648 		}
1649 		RTFREE(sro6.ro_rt);
1650 	}
1651 
1652 	return (maxmtu);
1653 }
1654 #endif /* INET6 */
1655 
1656 #ifdef IPSEC
1657 /* compute ESP/AH header size for TCP, including outer IP header. */
1658 size_t
1659 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1660 {
1661 	struct inpcb *inp;
1662 	struct mbuf *m;
1663 	size_t hdrsiz;
1664 	struct ip *ip;
1665 #ifdef INET6
1666 	struct ip6_hdr *ip6;
1667 #endif
1668 	struct tcphdr *th;
1669 
1670 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1671 		return (0);
1672 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1673 	if (!m)
1674 		return (0);
1675 
1676 #ifdef INET6
1677 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1678 		ip6 = mtod(m, struct ip6_hdr *);
1679 		th = (struct tcphdr *)(ip6 + 1);
1680 		m->m_pkthdr.len = m->m_len =
1681 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1682 		tcpip_fillheaders(inp, ip6, th);
1683 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1684 	} else
1685 #endif /* INET6 */
1686 	{
1687 		ip = mtod(m, struct ip *);
1688 		th = (struct tcphdr *)(ip + 1);
1689 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1690 		tcpip_fillheaders(inp, ip, th);
1691 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1692 	}
1693 
1694 	m_free(m);
1695 	return (hdrsiz);
1696 }
1697 #endif /* IPSEC */
1698 
1699 /*
1700  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1701  *
1702  * This code attempts to calculate the bandwidth-delay product as a
1703  * means of determining the optimal window size to maximize bandwidth,
1704  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1705  * routers.  This code also does a fairly good job keeping RTTs in check
1706  * across slow links like modems.  We implement an algorithm which is very
1707  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1708  * transmitter side of a TCP connection and so only effects the transmit
1709  * side of the connection.
1710  *
1711  * BACKGROUND:  TCP makes no provision for the management of buffer space
1712  * at the end points or at the intermediate routers and switches.  A TCP
1713  * stream, whether using NewReno or not, will eventually buffer as
1714  * many packets as it is able and the only reason this typically works is
1715  * due to the fairly small default buffers made available for a connection
1716  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1717  * scaling it is now fairly easy for even a single TCP connection to blow-out
1718  * all available buffer space not only on the local interface, but on
1719  * intermediate routers and switches as well.  NewReno makes a misguided
1720  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1721  * then backing off, then steadily increasing the window again until another
1722  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1723  * is only made worse as network loads increase and the idea of intentionally
1724  * blowing out network buffers is, frankly, a terrible way to manage network
1725  * resources.
1726  *
1727  * It is far better to limit the transmit window prior to the failure
1728  * condition being achieved.  There are two general ways to do this:  First
1729  * you can 'scan' through different transmit window sizes and locate the
1730  * point where the RTT stops increasing, indicating that you have filled the
1731  * pipe, then scan backwards until you note that RTT stops decreasing, then
1732  * repeat ad-infinitum.  This method works in principle but has severe
1733  * implementation issues due to RTT variances, timer granularity, and
1734  * instability in the algorithm which can lead to many false positives and
1735  * create oscillations as well as interact badly with other TCP streams
1736  * implementing the same algorithm.
1737  *
1738  * The second method is to limit the window to the bandwidth delay product
1739  * of the link.  This is the method we implement.  RTT variances and our
1740  * own manipulation of the congestion window, bwnd, can potentially
1741  * destabilize the algorithm.  For this reason we have to stabilize the
1742  * elements used to calculate the window.  We do this by using the minimum
1743  * observed RTT, the long term average of the observed bandwidth, and
1744  * by adding two segments worth of slop.  It isn't perfect but it is able
1745  * to react to changing conditions and gives us a very stable basis on
1746  * which to extend the algorithm.
1747  */
1748 void
1749 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1750 {
1751 	u_long bw;
1752 	u_long bwnd;
1753 	int save_ticks;
1754 
1755 	INP_LOCK_ASSERT(tp->t_inpcb);
1756 
1757 	/*
1758 	 * If inflight_enable is disabled in the middle of a tcp connection,
1759 	 * make sure snd_bwnd is effectively disabled.
1760 	 */
1761 	if (tcp_inflight_enable == 0 || tp->t_rttlow < 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) + 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 < tcp_inflight_min)
1838 		bwnd = tcp_inflight_min;
1839 	if (bwnd > tcp_inflight_max)
1840 		bwnd = 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, ip6_use_defzone);
2017 		if (error)
2018 			return (error);
2019 		error = sa6_embedscope(lin6, 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(&tcbinfo);
2035 	switch (addrs[0].ss_family) {
2036 #ifdef INET6
2037 	case AF_INET6:
2038 		inp = in6_pcblookup_hash(&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(&tcbinfo, fin->sin_addr, fin->sin_port,
2044 		    lin->sin_addr, lin->sin_port, 0, NULL);
2045 		break;
2046 	}
2047 	if (inp != NULL) {
2048 		INP_LOCK(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_UNLOCK(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_UNLOCK(inp);
2067 		} else
2068 			INP_UNLOCK(inp);
2069 	} else
2070 		error = ESRCH;
2071 	INP_INFO_WUNLOCK(&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