xref: /freebsd/sys/netinet/tcp_subr.c (revision 3fe401a500cdfc73d8c066da3c577c4b9f0aa953)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/hhook.h>
45 #include <sys/kernel.h>
46 #include <sys/khelp.h>
47 #include <sys/sysctl.h>
48 #include <sys/jail.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #ifdef INET6
52 #include <sys/domain.h>
53 #endif
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/sdt.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/random.h>
61 
62 #include <vm/uma.h>
63 
64 #include <net/route.h>
65 #include <net/if.h>
66 #include <net/if_var.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/cc.h>
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/in_var.h>
75 #include <netinet/ip.h>
76 #include <netinet/ip_icmp.h>
77 #include <netinet/ip_var.h>
78 #ifdef INET6
79 #include <netinet/ip6.h>
80 #include <netinet6/in6_pcb.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #include <netinet6/nd6.h>
84 #endif
85 
86 #include <netinet/tcp_fsm.h>
87 #include <netinet/tcp_seq.h>
88 #include <netinet/tcp_timer.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/tcp_syncache.h>
91 #ifdef INET6
92 #include <netinet6/tcp6_var.h>
93 #endif
94 #include <netinet/tcpip.h>
95 #ifdef TCPDEBUG
96 #include <netinet/tcp_debug.h>
97 #endif
98 #ifdef INET6
99 #include <netinet6/ip6protosw.h>
100 #endif
101 #ifdef TCP_OFFLOAD
102 #include <netinet/tcp_offload.h>
103 #endif
104 
105 #ifdef IPSEC
106 #include <netipsec/ipsec.h>
107 #include <netipsec/xform.h>
108 #ifdef INET6
109 #include <netipsec/ipsec6.h>
110 #endif
111 #include <netipsec/key.h>
112 #include <sys/syslog.h>
113 #endif /*IPSEC*/
114 
115 #include <machine/in_cksum.h>
116 #include <sys/md5.h>
117 
118 #include <security/mac/mac_framework.h>
119 
120 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
121 #ifdef INET6
122 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
123 #endif
124 
125 static int
126 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
127 {
128 	int error, new;
129 
130 	new = V_tcp_mssdflt;
131 	error = sysctl_handle_int(oidp, &new, 0, req);
132 	if (error == 0 && req->newptr) {
133 		if (new < TCP_MINMSS)
134 			error = EINVAL;
135 		else
136 			V_tcp_mssdflt = new;
137 	}
138 	return (error);
139 }
140 
141 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
142     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
143     &sysctl_net_inet_tcp_mss_check, "I",
144     "Default TCP Maximum Segment Size");
145 
146 #ifdef INET6
147 static int
148 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
149 {
150 	int error, new;
151 
152 	new = V_tcp_v6mssdflt;
153 	error = sysctl_handle_int(oidp, &new, 0, req);
154 	if (error == 0 && req->newptr) {
155 		if (new < TCP_MINMSS)
156 			error = EINVAL;
157 		else
158 			V_tcp_v6mssdflt = new;
159 	}
160 	return (error);
161 }
162 
163 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
164     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
165     &sysctl_net_inet_tcp_mss_v6_check, "I",
166    "Default TCP Maximum Segment Size for IPv6");
167 #endif /* INET6 */
168 
169 /*
170  * Minimum MSS we accept and use. This prevents DoS attacks where
171  * we are forced to a ridiculous low MSS like 20 and send hundreds
172  * of packets instead of one. The effect scales with the available
173  * bandwidth and quickly saturates the CPU and network interface
174  * with packet generation and sending. Set to zero to disable MINMSS
175  * checking. This setting prevents us from sending too small packets.
176  */
177 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
179      &VNET_NAME(tcp_minmss), 0,
180     "Minimum TCP Maximum Segment Size");
181 
182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
184     &VNET_NAME(tcp_do_rfc1323), 0,
185     "Enable rfc1323 (high performance TCP) extensions");
186 
187 static int	tcp_log_debug = 0;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
189     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
190 
191 static int	tcp_tcbhashsize;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
193     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
194 
195 static int	do_tcpdrain = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
197     "Enable tcp_drain routine for extra help when low on mbufs");
198 
199 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
200     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
201 
202 static VNET_DEFINE(int, icmp_may_rst) = 1;
203 #define	V_icmp_may_rst			VNET(icmp_may_rst)
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
205     &VNET_NAME(icmp_may_rst), 0,
206     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
207 
208 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
209 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
211     &VNET_NAME(tcp_isn_reseed_interval), 0,
212     "Seconds between reseeding of ISN secret");
213 
214 static int	tcp_soreceive_stream;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
216     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
217 
218 #ifdef TCP_SIGNATURE
219 static int	tcp_sig_checksigs = 1;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
221     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
222 #endif
223 
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define	V_sack_hole_zone		VNET(sack_hole_zone)
226 
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
228 
229 static struct inpcb *tcp_notify(struct inpcb *, int);
230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
232 		    void *ip4hdr, const void *ip6hdr);
233 static void	tcp_timer_discard(struct tcpcb *, uint32_t);
234 
235 /*
236  * Target size of TCP PCB hash tables. Must be a power of two.
237  *
238  * Note that this can be overridden by the kernel environment
239  * variable net.inet.tcp.tcbhashsize
240  */
241 #ifndef TCBHASHSIZE
242 #define TCBHASHSIZE	0
243 #endif
244 
245 /*
246  * XXX
247  * Callouts should be moved into struct tcp directly.  They are currently
248  * separate because the tcpcb structure is exported to userland for sysctl
249  * parsing purposes, which do not know about callouts.
250  */
251 struct tcpcb_mem {
252 	struct	tcpcb		tcb;
253 	struct	tcp_timer	tt;
254 	struct	cc_var		ccv;
255 	struct	osd		osd;
256 };
257 
258 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
259 #define	V_tcpcb_zone			VNET(tcpcb_zone)
260 
261 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
262 static struct mtx isn_mtx;
263 
264 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
265 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
266 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
267 
268 /*
269  * TCP initialization.
270  */
271 static void
272 tcp_zone_change(void *tag)
273 {
274 
275 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
276 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
277 	tcp_tw_zone_change();
278 }
279 
280 static int
281 tcp_inpcb_init(void *mem, int size, int flags)
282 {
283 	struct inpcb *inp = mem;
284 
285 	INP_LOCK_INIT(inp, "inp", "tcpinp");
286 	return (0);
287 }
288 
289 /*
290  * Take a value and get the next power of 2 that doesn't overflow.
291  * Used to size the tcp_inpcb hash buckets.
292  */
293 static int
294 maketcp_hashsize(int size)
295 {
296 	int hashsize;
297 
298 	/*
299 	 * auto tune.
300 	 * get the next power of 2 higher than maxsockets.
301 	 */
302 	hashsize = 1 << fls(size);
303 	/* catch overflow, and just go one power of 2 smaller */
304 	if (hashsize < size) {
305 		hashsize = 1 << (fls(size) - 1);
306 	}
307 	return (hashsize);
308 }
309 
310 void
311 tcp_init(void)
312 {
313 	const char *tcbhash_tuneable;
314 	int hashsize;
315 
316 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
317 
318 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
319 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
320 		printf("%s: WARNING: unable to register helper hook\n", __func__);
321 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
322 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 		printf("%s: WARNING: unable to register helper hook\n", __func__);
324 
325 	hashsize = TCBHASHSIZE;
326 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
327 	if (hashsize == 0) {
328 		/*
329 		 * Auto tune the hash size based on maxsockets.
330 		 * A perfect hash would have a 1:1 mapping
331 		 * (hashsize = maxsockets) however it's been
332 		 * suggested that O(2) average is better.
333 		 */
334 		hashsize = maketcp_hashsize(maxsockets / 4);
335 		/*
336 		 * Our historical default is 512,
337 		 * do not autotune lower than this.
338 		 */
339 		if (hashsize < 512)
340 			hashsize = 512;
341 		if (bootverbose)
342 			printf("%s: %s auto tuned to %d\n", __func__,
343 			    tcbhash_tuneable, hashsize);
344 	}
345 	/*
346 	 * We require a hashsize to be a power of two.
347 	 * Previously if it was not a power of two we would just reset it
348 	 * back to 512, which could be a nasty surprise if you did not notice
349 	 * the error message.
350 	 * Instead what we do is clip it to the closest power of two lower
351 	 * than the specified hash value.
352 	 */
353 	if (!powerof2(hashsize)) {
354 		int oldhashsize = hashsize;
355 
356 		hashsize = maketcp_hashsize(hashsize);
357 		/* prevent absurdly low value */
358 		if (hashsize < 16)
359 			hashsize = 16;
360 		printf("%s: WARNING: TCB hash size not a power of 2, "
361 		    "clipped from %d to %d.\n", __func__, oldhashsize,
362 		    hashsize);
363 	}
364 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
365 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
366 	    IPI_HASHFIELDS_4TUPLE);
367 
368 	/*
369 	 * These have to be type stable for the benefit of the timers.
370 	 */
371 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
372 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
373 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
374 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
375 
376 	tcp_tw_init();
377 	syncache_init();
378 	tcp_hc_init();
379 
380 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
381 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
382 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
383 
384 	/* Skip initialization of globals for non-default instances. */
385 	if (!IS_DEFAULT_VNET(curvnet))
386 		return;
387 
388 	tcp_reass_global_init();
389 
390 	/* XXX virtualize those bellow? */
391 	tcp_delacktime = TCPTV_DELACK;
392 	tcp_keepinit = TCPTV_KEEP_INIT;
393 	tcp_keepidle = TCPTV_KEEP_IDLE;
394 	tcp_keepintvl = TCPTV_KEEPINTVL;
395 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
396 	tcp_msl = TCPTV_MSL;
397 	tcp_rexmit_min = TCPTV_MIN;
398 	if (tcp_rexmit_min < 1)
399 		tcp_rexmit_min = 1;
400 	tcp_rexmit_slop = TCPTV_CPU_VAR;
401 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
402 	tcp_tcbhashsize = hashsize;
403 
404 	if (tcp_soreceive_stream) {
405 #ifdef INET
406 		tcp_usrreqs.pru_soreceive = soreceive_stream;
407 #endif
408 #ifdef INET6
409 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
410 #endif /* INET6 */
411 	}
412 
413 #ifdef INET6
414 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
415 #else /* INET6 */
416 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
417 #endif /* INET6 */
418 	if (max_protohdr < TCP_MINPROTOHDR)
419 		max_protohdr = TCP_MINPROTOHDR;
420 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
421 		panic("tcp_init");
422 #undef TCP_MINPROTOHDR
423 
424 	ISN_LOCK_INIT();
425 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
426 		SHUTDOWN_PRI_DEFAULT);
427 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
428 		EVENTHANDLER_PRI_ANY);
429 }
430 
431 #ifdef VIMAGE
432 void
433 tcp_destroy(void)
434 {
435 	int error;
436 
437 	tcp_hc_destroy();
438 	syncache_destroy();
439 	tcp_tw_destroy();
440 	in_pcbinfo_destroy(&V_tcbinfo);
441 	uma_zdestroy(V_sack_hole_zone);
442 	uma_zdestroy(V_tcpcb_zone);
443 
444 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
445 	if (error != 0) {
446 		printf("%s: WARNING: unable to deregister helper hook "
447 		    "type=%d, id=%d: error %d returned\n", __func__,
448 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
449 	}
450 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
451 	if (error != 0) {
452 		printf("%s: WARNING: unable to deregister helper hook "
453 		    "type=%d, id=%d: error %d returned\n", __func__,
454 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
455 	}
456 }
457 #endif
458 
459 void
460 tcp_fini(void *xtp)
461 {
462 
463 }
464 
465 /*
466  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
467  * tcp_template used to store this data in mbufs, but we now recopy it out
468  * of the tcpcb each time to conserve mbufs.
469  */
470 void
471 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
472 {
473 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
474 
475 	INP_WLOCK_ASSERT(inp);
476 
477 #ifdef INET6
478 	if ((inp->inp_vflag & INP_IPV6) != 0) {
479 		struct ip6_hdr *ip6;
480 
481 		ip6 = (struct ip6_hdr *)ip_ptr;
482 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
483 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
484 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
485 			(IPV6_VERSION & IPV6_VERSION_MASK);
486 		ip6->ip6_nxt = IPPROTO_TCP;
487 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
488 		ip6->ip6_src = inp->in6p_laddr;
489 		ip6->ip6_dst = inp->in6p_faddr;
490 	}
491 #endif /* INET6 */
492 #if defined(INET6) && defined(INET)
493 	else
494 #endif
495 #ifdef INET
496 	{
497 		struct ip *ip;
498 
499 		ip = (struct ip *)ip_ptr;
500 		ip->ip_v = IPVERSION;
501 		ip->ip_hl = 5;
502 		ip->ip_tos = inp->inp_ip_tos;
503 		ip->ip_len = 0;
504 		ip->ip_id = 0;
505 		ip->ip_off = 0;
506 		ip->ip_ttl = inp->inp_ip_ttl;
507 		ip->ip_sum = 0;
508 		ip->ip_p = IPPROTO_TCP;
509 		ip->ip_src = inp->inp_laddr;
510 		ip->ip_dst = inp->inp_faddr;
511 	}
512 #endif /* INET */
513 	th->th_sport = inp->inp_lport;
514 	th->th_dport = inp->inp_fport;
515 	th->th_seq = 0;
516 	th->th_ack = 0;
517 	th->th_x2 = 0;
518 	th->th_off = 5;
519 	th->th_flags = 0;
520 	th->th_win = 0;
521 	th->th_urp = 0;
522 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
523 }
524 
525 /*
526  * Create template to be used to send tcp packets on a connection.
527  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
528  * use for this function is in keepalives, which use tcp_respond.
529  */
530 struct tcptemp *
531 tcpip_maketemplate(struct inpcb *inp)
532 {
533 	struct tcptemp *t;
534 
535 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
536 	if (t == NULL)
537 		return (NULL);
538 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
539 	return (t);
540 }
541 
542 /*
543  * Send a single message to the TCP at address specified by
544  * the given TCP/IP header.  If m == NULL, then we make a copy
545  * of the tcpiphdr at th and send directly to the addressed host.
546  * This is used to force keep alive messages out using the TCP
547  * template for a connection.  If flags are given then we send
548  * a message back to the TCP which originated the segment th,
549  * and discard the mbuf containing it and any other attached mbufs.
550  *
551  * In any case the ack and sequence number of the transmitted
552  * segment are as specified by the parameters.
553  *
554  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
555  */
556 void
557 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
558     tcp_seq ack, tcp_seq seq, int flags)
559 {
560 	int tlen;
561 	int win = 0;
562 	struct ip *ip;
563 	struct tcphdr *nth;
564 #ifdef INET6
565 	struct ip6_hdr *ip6;
566 	int isipv6;
567 #endif /* INET6 */
568 	int ipflags = 0;
569 	struct inpcb *inp;
570 
571 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
572 
573 #ifdef INET6
574 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
575 	ip6 = ipgen;
576 #endif /* INET6 */
577 	ip = ipgen;
578 
579 	if (tp != NULL) {
580 		inp = tp->t_inpcb;
581 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
582 		INP_WLOCK_ASSERT(inp);
583 	} else
584 		inp = NULL;
585 
586 	if (tp != NULL) {
587 		if (!(flags & TH_RST)) {
588 			win = sbspace(&inp->inp_socket->so_rcv);
589 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
590 				win = (long)TCP_MAXWIN << tp->rcv_scale;
591 		}
592 	}
593 	if (m == NULL) {
594 		m = m_gethdr(M_NOWAIT, MT_DATA);
595 		if (m == NULL)
596 			return;
597 		tlen = 0;
598 		m->m_data += max_linkhdr;
599 #ifdef INET6
600 		if (isipv6) {
601 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
602 			      sizeof(struct ip6_hdr));
603 			ip6 = mtod(m, struct ip6_hdr *);
604 			nth = (struct tcphdr *)(ip6 + 1);
605 		} else
606 #endif /* INET6 */
607 		{
608 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
609 			ip = mtod(m, struct ip *);
610 			nth = (struct tcphdr *)(ip + 1);
611 		}
612 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
613 		flags = TH_ACK;
614 	} else {
615 		/*
616 		 *  reuse the mbuf.
617 		 * XXX MRT We inherrit the FIB, which is lucky.
618 		 */
619 		m_freem(m->m_next);
620 		m->m_next = NULL;
621 		m->m_data = (caddr_t)ipgen;
622 		/* m_len is set later */
623 		tlen = 0;
624 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
625 #ifdef INET6
626 		if (isipv6) {
627 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
628 			nth = (struct tcphdr *)(ip6 + 1);
629 		} else
630 #endif /* INET6 */
631 		{
632 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
633 			nth = (struct tcphdr *)(ip + 1);
634 		}
635 		if (th != nth) {
636 			/*
637 			 * this is usually a case when an extension header
638 			 * exists between the IPv6 header and the
639 			 * TCP header.
640 			 */
641 			nth->th_sport = th->th_sport;
642 			nth->th_dport = th->th_dport;
643 		}
644 		xchg(nth->th_dport, nth->th_sport, uint16_t);
645 #undef xchg
646 	}
647 #ifdef INET6
648 	if (isipv6) {
649 		ip6->ip6_flow = 0;
650 		ip6->ip6_vfc = IPV6_VERSION;
651 		ip6->ip6_nxt = IPPROTO_TCP;
652 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
653 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
654 	}
655 #endif
656 #if defined(INET) && defined(INET6)
657 	else
658 #endif
659 #ifdef INET
660 	{
661 		tlen += sizeof (struct tcpiphdr);
662 		ip->ip_len = htons(tlen);
663 		ip->ip_ttl = V_ip_defttl;
664 		if (V_path_mtu_discovery)
665 			ip->ip_off |= htons(IP_DF);
666 	}
667 #endif
668 	m->m_len = tlen;
669 	m->m_pkthdr.len = tlen;
670 	m->m_pkthdr.rcvif = NULL;
671 #ifdef MAC
672 	if (inp != NULL) {
673 		/*
674 		 * Packet is associated with a socket, so allow the
675 		 * label of the response to reflect the socket label.
676 		 */
677 		INP_WLOCK_ASSERT(inp);
678 		mac_inpcb_create_mbuf(inp, m);
679 	} else {
680 		/*
681 		 * Packet is not associated with a socket, so possibly
682 		 * update the label in place.
683 		 */
684 		mac_netinet_tcp_reply(m);
685 	}
686 #endif
687 	nth->th_seq = htonl(seq);
688 	nth->th_ack = htonl(ack);
689 	nth->th_x2 = 0;
690 	nth->th_off = sizeof (struct tcphdr) >> 2;
691 	nth->th_flags = flags;
692 	if (tp != NULL)
693 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
694 	else
695 		nth->th_win = htons((u_short)win);
696 	nth->th_urp = 0;
697 
698 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
699 #ifdef INET6
700 	if (isipv6) {
701 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
702 		nth->th_sum = in6_cksum_pseudo(ip6,
703 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
704 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
705 		    NULL, NULL);
706 	}
707 #endif /* INET6 */
708 #if defined(INET6) && defined(INET)
709 	else
710 #endif
711 #ifdef INET
712 	{
713 		m->m_pkthdr.csum_flags = CSUM_TCP;
714 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
715 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
716 	}
717 #endif /* INET */
718 #ifdef TCPDEBUG
719 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
720 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
721 #endif
722 	if (flags & TH_RST)
723 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
724 		    tp, nth);
725 
726 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
727 #ifdef INET6
728 	if (isipv6)
729 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
730 #endif /* INET6 */
731 #if defined(INET) && defined(INET6)
732 	else
733 #endif
734 #ifdef INET
735 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
736 #endif
737 }
738 
739 /*
740  * Create a new TCP control block, making an
741  * empty reassembly queue and hooking it to the argument
742  * protocol control block.  The `inp' parameter must have
743  * come from the zone allocator set up in tcp_init().
744  */
745 struct tcpcb *
746 tcp_newtcpcb(struct inpcb *inp)
747 {
748 	struct tcpcb_mem *tm;
749 	struct tcpcb *tp;
750 #ifdef INET6
751 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
752 #endif /* INET6 */
753 
754 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
755 	if (tm == NULL)
756 		return (NULL);
757 	tp = &tm->tcb;
758 
759 	/* Initialise cc_var struct for this tcpcb. */
760 	tp->ccv = &tm->ccv;
761 	tp->ccv->type = IPPROTO_TCP;
762 	tp->ccv->ccvc.tcp = tp;
763 
764 	/*
765 	 * Use the current system default CC algorithm.
766 	 */
767 	CC_LIST_RLOCK();
768 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
769 	CC_ALGO(tp) = CC_DEFAULT();
770 	CC_LIST_RUNLOCK();
771 
772 	if (CC_ALGO(tp)->cb_init != NULL)
773 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
774 			uma_zfree(V_tcpcb_zone, tm);
775 			return (NULL);
776 		}
777 
778 	tp->osd = &tm->osd;
779 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
780 		uma_zfree(V_tcpcb_zone, tm);
781 		return (NULL);
782 	}
783 
784 #ifdef VIMAGE
785 	tp->t_vnet = inp->inp_vnet;
786 #endif
787 	tp->t_timers = &tm->tt;
788 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
789 	tp->t_maxseg = tp->t_maxopd =
790 #ifdef INET6
791 		isipv6 ? V_tcp_v6mssdflt :
792 #endif /* INET6 */
793 		V_tcp_mssdflt;
794 
795 	/* Set up our timeouts. */
796 	callout_init(&tp->t_timers->tt_rexmt, 1);
797 	callout_init(&tp->t_timers->tt_persist, 1);
798 	callout_init(&tp->t_timers->tt_keep, 1);
799 	callout_init(&tp->t_timers->tt_2msl, 1);
800 	callout_init(&tp->t_timers->tt_delack, 1);
801 
802 	if (V_tcp_do_rfc1323)
803 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
804 	if (V_tcp_do_sack)
805 		tp->t_flags |= TF_SACK_PERMIT;
806 	TAILQ_INIT(&tp->snd_holes);
807 	/*
808 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
809 	 * is called.
810 	 */
811 	in_pcbref(inp);	/* Reference for tcpcb */
812 	tp->t_inpcb = inp;
813 
814 	/*
815 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
816 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
817 	 * reasonable initial retransmit time.
818 	 */
819 	tp->t_srtt = TCPTV_SRTTBASE;
820 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
821 	tp->t_rttmin = tcp_rexmit_min;
822 	tp->t_rxtcur = TCPTV_RTOBASE;
823 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
824 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
825 	tp->t_rcvtime = ticks;
826 	/*
827 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
828 	 * because the socket may be bound to an IPv6 wildcard address,
829 	 * which may match an IPv4-mapped IPv6 address.
830 	 */
831 	inp->inp_ip_ttl = V_ip_defttl;
832 	inp->inp_ppcb = tp;
833 	return (tp);		/* XXX */
834 }
835 
836 /*
837  * Switch the congestion control algorithm back to NewReno for any active
838  * control blocks using an algorithm which is about to go away.
839  * This ensures the CC framework can allow the unload to proceed without leaving
840  * any dangling pointers which would trigger a panic.
841  * Returning non-zero would inform the CC framework that something went wrong
842  * and it would be unsafe to allow the unload to proceed. However, there is no
843  * way for this to occur with this implementation so we always return zero.
844  */
845 int
846 tcp_ccalgounload(struct cc_algo *unload_algo)
847 {
848 	struct cc_algo *tmpalgo;
849 	struct inpcb *inp;
850 	struct tcpcb *tp;
851 	VNET_ITERATOR_DECL(vnet_iter);
852 
853 	/*
854 	 * Check all active control blocks across all network stacks and change
855 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
856 	 * requires cleanup code to be run, call it.
857 	 */
858 	VNET_LIST_RLOCK();
859 	VNET_FOREACH(vnet_iter) {
860 		CURVNET_SET(vnet_iter);
861 		INP_INFO_RLOCK(&V_tcbinfo);
862 		/*
863 		 * New connections already part way through being initialised
864 		 * with the CC algo we're removing will not race with this code
865 		 * because the INP_INFO_WLOCK is held during initialisation. We
866 		 * therefore don't enter the loop below until the connection
867 		 * list has stabilised.
868 		 */
869 		LIST_FOREACH(inp, &V_tcb, inp_list) {
870 			INP_WLOCK(inp);
871 			/* Important to skip tcptw structs. */
872 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
873 			    (tp = intotcpcb(inp)) != NULL) {
874 				/*
875 				 * By holding INP_WLOCK here, we are assured
876 				 * that the connection is not currently
877 				 * executing inside the CC module's functions
878 				 * i.e. it is safe to make the switch back to
879 				 * NewReno.
880 				 */
881 				if (CC_ALGO(tp) == unload_algo) {
882 					tmpalgo = CC_ALGO(tp);
883 					/* NewReno does not require any init. */
884 					CC_ALGO(tp) = &newreno_cc_algo;
885 					if (tmpalgo->cb_destroy != NULL)
886 						tmpalgo->cb_destroy(tp->ccv);
887 				}
888 			}
889 			INP_WUNLOCK(inp);
890 		}
891 		INP_INFO_RUNLOCK(&V_tcbinfo);
892 		CURVNET_RESTORE();
893 	}
894 	VNET_LIST_RUNLOCK();
895 
896 	return (0);
897 }
898 
899 /*
900  * Drop a TCP connection, reporting
901  * the specified error.  If connection is synchronized,
902  * then send a RST to peer.
903  */
904 struct tcpcb *
905 tcp_drop(struct tcpcb *tp, int errno)
906 {
907 	struct socket *so = tp->t_inpcb->inp_socket;
908 
909 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
910 	INP_WLOCK_ASSERT(tp->t_inpcb);
911 
912 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
913 		tcp_state_change(tp, TCPS_CLOSED);
914 		(void) tcp_output(tp);
915 		TCPSTAT_INC(tcps_drops);
916 	} else
917 		TCPSTAT_INC(tcps_conndrops);
918 	if (errno == ETIMEDOUT && tp->t_softerror)
919 		errno = tp->t_softerror;
920 	so->so_error = errno;
921 	return (tcp_close(tp));
922 }
923 
924 void
925 tcp_discardcb(struct tcpcb *tp)
926 {
927 	struct inpcb *inp = tp->t_inpcb;
928 	struct socket *so = inp->inp_socket;
929 #ifdef INET6
930 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
931 #endif /* INET6 */
932 	int released;
933 
934 	INP_WLOCK_ASSERT(inp);
935 
936 	/*
937 	 * Make sure that all of our timers are stopped before we delete the
938 	 * PCB.
939 	 *
940 	 * If stopping a timer fails, we schedule a discard function in same
941 	 * callout, and the last discard function called will take care of
942 	 * deleting the tcpcb.
943 	 */
944 	tcp_timer_stop(tp, TT_REXMT);
945 	tcp_timer_stop(tp, TT_PERSIST);
946 	tcp_timer_stop(tp, TT_KEEP);
947 	tcp_timer_stop(tp, TT_2MSL);
948 	tcp_timer_stop(tp, TT_DELACK);
949 
950 	/*
951 	 * If we got enough samples through the srtt filter,
952 	 * save the rtt and rttvar in the routing entry.
953 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
954 	 * 4 samples is enough for the srtt filter to converge
955 	 * to within enough % of the correct value; fewer samples
956 	 * and we could save a bogus rtt. The danger is not high
957 	 * as tcp quickly recovers from everything.
958 	 * XXX: Works very well but needs some more statistics!
959 	 */
960 	if (tp->t_rttupdated >= 4) {
961 		struct hc_metrics_lite metrics;
962 		u_long ssthresh;
963 
964 		bzero(&metrics, sizeof(metrics));
965 		/*
966 		 * Update the ssthresh always when the conditions below
967 		 * are satisfied. This gives us better new start value
968 		 * for the congestion avoidance for new connections.
969 		 * ssthresh is only set if packet loss occured on a session.
970 		 *
971 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
972 		 * being torn down.  Ideally this code would not use 'so'.
973 		 */
974 		ssthresh = tp->snd_ssthresh;
975 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
976 			/*
977 			 * convert the limit from user data bytes to
978 			 * packets then to packet data bytes.
979 			 */
980 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
981 			if (ssthresh < 2)
982 				ssthresh = 2;
983 			ssthresh *= (u_long)(tp->t_maxseg +
984 #ifdef INET6
985 			    (isipv6 ? sizeof (struct ip6_hdr) +
986 				sizeof (struct tcphdr) :
987 #endif
988 				sizeof (struct tcpiphdr)
989 #ifdef INET6
990 			    )
991 #endif
992 			    );
993 		} else
994 			ssthresh = 0;
995 		metrics.rmx_ssthresh = ssthresh;
996 
997 		metrics.rmx_rtt = tp->t_srtt;
998 		metrics.rmx_rttvar = tp->t_rttvar;
999 		metrics.rmx_cwnd = tp->snd_cwnd;
1000 		metrics.rmx_sendpipe = 0;
1001 		metrics.rmx_recvpipe = 0;
1002 
1003 		tcp_hc_update(&inp->inp_inc, &metrics);
1004 	}
1005 
1006 	/* free the reassembly queue, if any */
1007 	tcp_reass_flush(tp);
1008 
1009 #ifdef TCP_OFFLOAD
1010 	/* Disconnect offload device, if any. */
1011 	if (tp->t_flags & TF_TOE)
1012 		tcp_offload_detach(tp);
1013 #endif
1014 
1015 	tcp_free_sackholes(tp);
1016 
1017 	/* Allow the CC algorithm to clean up after itself. */
1018 	if (CC_ALGO(tp)->cb_destroy != NULL)
1019 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1020 
1021 	khelp_destroy_osd(tp->osd);
1022 
1023 	CC_ALGO(tp) = NULL;
1024 	inp->inp_ppcb = NULL;
1025 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1026 		/* We own the last reference on tcpcb, let's free it. */
1027 		tp->t_inpcb = NULL;
1028 		uma_zfree(V_tcpcb_zone, tp);
1029 		released = in_pcbrele_wlocked(inp);
1030 		KASSERT(!released, ("%s: inp %p should not have been released "
1031 			"here", __func__, inp));
1032 	}
1033 }
1034 
1035 void
1036 tcp_timer_2msl_discard(void *xtp)
1037 {
1038 
1039 	tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1040 }
1041 
1042 void
1043 tcp_timer_keep_discard(void *xtp)
1044 {
1045 
1046 	tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1047 }
1048 
1049 void
1050 tcp_timer_persist_discard(void *xtp)
1051 {
1052 
1053 	tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1054 }
1055 
1056 void
1057 tcp_timer_rexmt_discard(void *xtp)
1058 {
1059 
1060 	tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1061 }
1062 
1063 void
1064 tcp_timer_delack_discard(void *xtp)
1065 {
1066 
1067 	tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1068 }
1069 
1070 void
1071 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1072 {
1073 	struct inpcb *inp;
1074 
1075 	CURVNET_SET(tp->t_vnet);
1076 	INP_INFO_WLOCK(&V_tcbinfo);
1077 	inp = tp->t_inpcb;
1078 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1079 		__func__, tp));
1080 	INP_WLOCK(inp);
1081 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1082 		("%s: tcpcb has to be stopped here", __func__));
1083 	KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1084 		("%s: discard callout should be running", __func__));
1085 	tp->t_timers->tt_flags &= ~timer_type;
1086 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1087 		/* We own the last reference on this tcpcb, let's free it. */
1088 		tp->t_inpcb = NULL;
1089 		uma_zfree(V_tcpcb_zone, tp);
1090 		if (in_pcbrele_wlocked(inp)) {
1091 			INP_INFO_WUNLOCK(&V_tcbinfo);
1092 			CURVNET_RESTORE();
1093 			return;
1094 		}
1095 	}
1096 	INP_WUNLOCK(inp);
1097 	INP_INFO_WUNLOCK(&V_tcbinfo);
1098 	CURVNET_RESTORE();
1099 }
1100 
1101 /*
1102  * Attempt to close a TCP control block, marking it as dropped, and freeing
1103  * the socket if we hold the only reference.
1104  */
1105 struct tcpcb *
1106 tcp_close(struct tcpcb *tp)
1107 {
1108 	struct inpcb *inp = tp->t_inpcb;
1109 	struct socket *so;
1110 
1111 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1112 	INP_WLOCK_ASSERT(inp);
1113 
1114 #ifdef TCP_OFFLOAD
1115 	if (tp->t_state == TCPS_LISTEN)
1116 		tcp_offload_listen_stop(tp);
1117 #endif
1118 	in_pcbdrop(inp);
1119 	TCPSTAT_INC(tcps_closed);
1120 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1121 	so = inp->inp_socket;
1122 	soisdisconnected(so);
1123 	if (inp->inp_flags & INP_SOCKREF) {
1124 		KASSERT(so->so_state & SS_PROTOREF,
1125 		    ("tcp_close: !SS_PROTOREF"));
1126 		inp->inp_flags &= ~INP_SOCKREF;
1127 		INP_WUNLOCK(inp);
1128 		ACCEPT_LOCK();
1129 		SOCK_LOCK(so);
1130 		so->so_state &= ~SS_PROTOREF;
1131 		sofree(so);
1132 		return (NULL);
1133 	}
1134 	return (tp);
1135 }
1136 
1137 void
1138 tcp_drain(void)
1139 {
1140 	VNET_ITERATOR_DECL(vnet_iter);
1141 
1142 	if (!do_tcpdrain)
1143 		return;
1144 
1145 	VNET_LIST_RLOCK_NOSLEEP();
1146 	VNET_FOREACH(vnet_iter) {
1147 		CURVNET_SET(vnet_iter);
1148 		struct inpcb *inpb;
1149 		struct tcpcb *tcpb;
1150 
1151 	/*
1152 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1153 	 * if there is one...
1154 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1155 	 *      reassembly queue should be flushed, but in a situation
1156 	 *	where we're really low on mbufs, this is potentially
1157 	 *	useful.
1158 	 */
1159 		INP_INFO_RLOCK(&V_tcbinfo);
1160 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1161 			if (inpb->inp_flags & INP_TIMEWAIT)
1162 				continue;
1163 			INP_WLOCK(inpb);
1164 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1165 				tcp_reass_flush(tcpb);
1166 				tcp_clean_sackreport(tcpb);
1167 			}
1168 			INP_WUNLOCK(inpb);
1169 		}
1170 		INP_INFO_RUNLOCK(&V_tcbinfo);
1171 		CURVNET_RESTORE();
1172 	}
1173 	VNET_LIST_RUNLOCK_NOSLEEP();
1174 }
1175 
1176 /*
1177  * Notify a tcp user of an asynchronous error;
1178  * store error as soft error, but wake up user
1179  * (for now, won't do anything until can select for soft error).
1180  *
1181  * Do not wake up user since there currently is no mechanism for
1182  * reporting soft errors (yet - a kqueue filter may be added).
1183  */
1184 static struct inpcb *
1185 tcp_notify(struct inpcb *inp, int error)
1186 {
1187 	struct tcpcb *tp;
1188 
1189 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1190 	INP_WLOCK_ASSERT(inp);
1191 
1192 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1193 	    (inp->inp_flags & INP_DROPPED))
1194 		return (inp);
1195 
1196 	tp = intotcpcb(inp);
1197 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1198 
1199 	/*
1200 	 * Ignore some errors if we are hooked up.
1201 	 * If connection hasn't completed, has retransmitted several times,
1202 	 * and receives a second error, give up now.  This is better
1203 	 * than waiting a long time to establish a connection that
1204 	 * can never complete.
1205 	 */
1206 	if (tp->t_state == TCPS_ESTABLISHED &&
1207 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1208 	     error == EHOSTDOWN)) {
1209 		return (inp);
1210 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1211 	    tp->t_softerror) {
1212 		tp = tcp_drop(tp, error);
1213 		if (tp != NULL)
1214 			return (inp);
1215 		else
1216 			return (NULL);
1217 	} else {
1218 		tp->t_softerror = error;
1219 		return (inp);
1220 	}
1221 #if 0
1222 	wakeup( &so->so_timeo);
1223 	sorwakeup(so);
1224 	sowwakeup(so);
1225 #endif
1226 }
1227 
1228 static int
1229 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1230 {
1231 	int error, i, m, n, pcb_count;
1232 	struct inpcb *inp, **inp_list;
1233 	inp_gen_t gencnt;
1234 	struct xinpgen xig;
1235 
1236 	/*
1237 	 * The process of preparing the TCB list is too time-consuming and
1238 	 * resource-intensive to repeat twice on every request.
1239 	 */
1240 	if (req->oldptr == NULL) {
1241 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1242 		n += imax(n / 8, 10);
1243 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1244 		return (0);
1245 	}
1246 
1247 	if (req->newptr != NULL)
1248 		return (EPERM);
1249 
1250 	/*
1251 	 * OK, now we're committed to doing something.
1252 	 */
1253 	INP_INFO_RLOCK(&V_tcbinfo);
1254 	gencnt = V_tcbinfo.ipi_gencnt;
1255 	n = V_tcbinfo.ipi_count;
1256 	INP_INFO_RUNLOCK(&V_tcbinfo);
1257 
1258 	m = syncache_pcbcount();
1259 
1260 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1261 		+ (n + m) * sizeof(struct xtcpcb));
1262 	if (error != 0)
1263 		return (error);
1264 
1265 	xig.xig_len = sizeof xig;
1266 	xig.xig_count = n + m;
1267 	xig.xig_gen = gencnt;
1268 	xig.xig_sogen = so_gencnt;
1269 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1270 	if (error)
1271 		return (error);
1272 
1273 	error = syncache_pcblist(req, m, &pcb_count);
1274 	if (error)
1275 		return (error);
1276 
1277 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1278 	if (inp_list == NULL)
1279 		return (ENOMEM);
1280 
1281 	INP_INFO_RLOCK(&V_tcbinfo);
1282 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1283 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1284 		INP_WLOCK(inp);
1285 		if (inp->inp_gencnt <= gencnt) {
1286 			/*
1287 			 * XXX: This use of cr_cansee(), introduced with
1288 			 * TCP state changes, is not quite right, but for
1289 			 * now, better than nothing.
1290 			 */
1291 			if (inp->inp_flags & INP_TIMEWAIT) {
1292 				if (intotw(inp) != NULL)
1293 					error = cr_cansee(req->td->td_ucred,
1294 					    intotw(inp)->tw_cred);
1295 				else
1296 					error = EINVAL;	/* Skip this inp. */
1297 			} else
1298 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1299 			if (error == 0) {
1300 				in_pcbref(inp);
1301 				inp_list[i++] = inp;
1302 			}
1303 		}
1304 		INP_WUNLOCK(inp);
1305 	}
1306 	INP_INFO_RUNLOCK(&V_tcbinfo);
1307 	n = i;
1308 
1309 	error = 0;
1310 	for (i = 0; i < n; i++) {
1311 		inp = inp_list[i];
1312 		INP_RLOCK(inp);
1313 		if (inp->inp_gencnt <= gencnt) {
1314 			struct xtcpcb xt;
1315 			void *inp_ppcb;
1316 
1317 			bzero(&xt, sizeof(xt));
1318 			xt.xt_len = sizeof xt;
1319 			/* XXX should avoid extra copy */
1320 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1321 			inp_ppcb = inp->inp_ppcb;
1322 			if (inp_ppcb == NULL)
1323 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1324 			else if (inp->inp_flags & INP_TIMEWAIT) {
1325 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1326 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1327 			} else {
1328 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1329 				if (xt.xt_tp.t_timers)
1330 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1331 			}
1332 			if (inp->inp_socket != NULL)
1333 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1334 			else {
1335 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1336 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1337 			}
1338 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1339 			INP_RUNLOCK(inp);
1340 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1341 		} else
1342 			INP_RUNLOCK(inp);
1343 	}
1344 	INP_INFO_WLOCK(&V_tcbinfo);
1345 	for (i = 0; i < n; i++) {
1346 		inp = inp_list[i];
1347 		INP_RLOCK(inp);
1348 		if (!in_pcbrele_rlocked(inp))
1349 			INP_RUNLOCK(inp);
1350 	}
1351 	INP_INFO_WUNLOCK(&V_tcbinfo);
1352 
1353 	if (!error) {
1354 		/*
1355 		 * Give the user an updated idea of our state.
1356 		 * If the generation differs from what we told
1357 		 * her before, she knows that something happened
1358 		 * while we were processing this request, and it
1359 		 * might be necessary to retry.
1360 		 */
1361 		INP_INFO_RLOCK(&V_tcbinfo);
1362 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1363 		xig.xig_sogen = so_gencnt;
1364 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1365 		INP_INFO_RUNLOCK(&V_tcbinfo);
1366 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1367 	}
1368 	free(inp_list, M_TEMP);
1369 	return (error);
1370 }
1371 
1372 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1373     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1374     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1375 
1376 #ifdef INET
1377 static int
1378 tcp_getcred(SYSCTL_HANDLER_ARGS)
1379 {
1380 	struct xucred xuc;
1381 	struct sockaddr_in addrs[2];
1382 	struct inpcb *inp;
1383 	int error;
1384 
1385 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1386 	if (error)
1387 		return (error);
1388 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1389 	if (error)
1390 		return (error);
1391 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1392 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1393 	if (inp != NULL) {
1394 		if (inp->inp_socket == NULL)
1395 			error = ENOENT;
1396 		if (error == 0)
1397 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1398 		if (error == 0)
1399 			cru2x(inp->inp_cred, &xuc);
1400 		INP_RUNLOCK(inp);
1401 	} else
1402 		error = ENOENT;
1403 	if (error == 0)
1404 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1405 	return (error);
1406 }
1407 
1408 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1409     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1410     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1411 #endif /* INET */
1412 
1413 #ifdef INET6
1414 static int
1415 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1416 {
1417 	struct xucred xuc;
1418 	struct sockaddr_in6 addrs[2];
1419 	struct inpcb *inp;
1420 	int error;
1421 #ifdef INET
1422 	int mapped = 0;
1423 #endif
1424 
1425 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1426 	if (error)
1427 		return (error);
1428 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1429 	if (error)
1430 		return (error);
1431 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1432 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1433 		return (error);
1434 	}
1435 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1436 #ifdef INET
1437 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1438 			mapped = 1;
1439 		else
1440 #endif
1441 			return (EINVAL);
1442 	}
1443 
1444 #ifdef INET
1445 	if (mapped == 1)
1446 		inp = in_pcblookup(&V_tcbinfo,
1447 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1448 			addrs[1].sin6_port,
1449 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1450 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1451 	else
1452 #endif
1453 		inp = in6_pcblookup(&V_tcbinfo,
1454 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1455 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1456 			INPLOOKUP_RLOCKPCB, NULL);
1457 	if (inp != NULL) {
1458 		if (inp->inp_socket == NULL)
1459 			error = ENOENT;
1460 		if (error == 0)
1461 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1462 		if (error == 0)
1463 			cru2x(inp->inp_cred, &xuc);
1464 		INP_RUNLOCK(inp);
1465 	} else
1466 		error = ENOENT;
1467 	if (error == 0)
1468 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1469 	return (error);
1470 }
1471 
1472 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1473     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1474     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1475 #endif /* INET6 */
1476 
1477 
1478 #ifdef INET
1479 void
1480 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1481 {
1482 	struct ip *ip = vip;
1483 	struct tcphdr *th;
1484 	struct in_addr faddr;
1485 	struct inpcb *inp;
1486 	struct tcpcb *tp;
1487 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1488 	struct icmp *icp;
1489 	struct in_conninfo inc;
1490 	tcp_seq icmp_tcp_seq;
1491 	int mtu;
1492 
1493 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1494 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1495 		return;
1496 
1497 	if (cmd == PRC_MSGSIZE)
1498 		notify = tcp_mtudisc_notify;
1499 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1500 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1501 		notify = tcp_drop_syn_sent;
1502 	/*
1503 	 * Redirects don't need to be handled up here.
1504 	 */
1505 	else if (PRC_IS_REDIRECT(cmd))
1506 		return;
1507 	/*
1508 	 * Hostdead is ugly because it goes linearly through all PCBs.
1509 	 * XXX: We never get this from ICMP, otherwise it makes an
1510 	 * excellent DoS attack on machines with many connections.
1511 	 */
1512 	else if (cmd == PRC_HOSTDEAD)
1513 		ip = NULL;
1514 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1515 		return;
1516 	if (ip != NULL) {
1517 		icp = (struct icmp *)((caddr_t)ip
1518 				      - offsetof(struct icmp, icmp_ip));
1519 		th = (struct tcphdr *)((caddr_t)ip
1520 				       + (ip->ip_hl << 2));
1521 		INP_INFO_WLOCK(&V_tcbinfo);
1522 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1523 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1524 		if (inp != NULL)  {
1525 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1526 			    !(inp->inp_flags & INP_DROPPED) &&
1527 			    !(inp->inp_socket == NULL)) {
1528 				icmp_tcp_seq = htonl(th->th_seq);
1529 				tp = intotcpcb(inp);
1530 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1531 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1532 					if (cmd == PRC_MSGSIZE) {
1533 					    /*
1534 					     * MTU discovery:
1535 					     * If we got a needfrag set the MTU
1536 					     * in the route to the suggested new
1537 					     * value (if given) and then notify.
1538 					     */
1539 					    bzero(&inc, sizeof(inc));
1540 					    inc.inc_faddr = faddr;
1541 					    inc.inc_fibnum =
1542 						inp->inp_inc.inc_fibnum;
1543 
1544 					    mtu = ntohs(icp->icmp_nextmtu);
1545 					    /*
1546 					     * If no alternative MTU was
1547 					     * proposed, try the next smaller
1548 					     * one.
1549 					     */
1550 					    if (!mtu)
1551 						mtu = ip_next_mtu(
1552 						 ntohs(ip->ip_len), 1);
1553 					    if (mtu < V_tcp_minmss
1554 						 + sizeof(struct tcpiphdr))
1555 						mtu = V_tcp_minmss
1556 						 + sizeof(struct tcpiphdr);
1557 					    /*
1558 					     * Only cache the MTU if it
1559 					     * is smaller than the interface
1560 					     * or route MTU.  tcp_mtudisc()
1561 					     * will do right thing by itself.
1562 					     */
1563 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1564 						tcp_hc_updatemtu(&inc, mtu);
1565 					    tcp_mtudisc(inp, mtu);
1566 					} else
1567 						inp = (*notify)(inp,
1568 						    inetctlerrmap[cmd]);
1569 				}
1570 			}
1571 			if (inp != NULL)
1572 				INP_WUNLOCK(inp);
1573 		} else {
1574 			bzero(&inc, sizeof(inc));
1575 			inc.inc_fport = th->th_dport;
1576 			inc.inc_lport = th->th_sport;
1577 			inc.inc_faddr = faddr;
1578 			inc.inc_laddr = ip->ip_src;
1579 			syncache_unreach(&inc, th);
1580 		}
1581 		INP_INFO_WUNLOCK(&V_tcbinfo);
1582 	} else
1583 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1584 }
1585 #endif /* INET */
1586 
1587 #ifdef INET6
1588 void
1589 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1590 {
1591 	struct tcphdr th;
1592 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1593 	struct ip6_hdr *ip6;
1594 	struct mbuf *m;
1595 	struct ip6ctlparam *ip6cp = NULL;
1596 	const struct sockaddr_in6 *sa6_src = NULL;
1597 	int off;
1598 	struct tcp_portonly {
1599 		u_int16_t th_sport;
1600 		u_int16_t th_dport;
1601 	} *thp;
1602 
1603 	if (sa->sa_family != AF_INET6 ||
1604 	    sa->sa_len != sizeof(struct sockaddr_in6))
1605 		return;
1606 
1607 	if (cmd == PRC_MSGSIZE)
1608 		notify = tcp_mtudisc_notify;
1609 	else if (!PRC_IS_REDIRECT(cmd) &&
1610 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1611 		return;
1612 
1613 	/* if the parameter is from icmp6, decode it. */
1614 	if (d != NULL) {
1615 		ip6cp = (struct ip6ctlparam *)d;
1616 		m = ip6cp->ip6c_m;
1617 		ip6 = ip6cp->ip6c_ip6;
1618 		off = ip6cp->ip6c_off;
1619 		sa6_src = ip6cp->ip6c_src;
1620 	} else {
1621 		m = NULL;
1622 		ip6 = NULL;
1623 		off = 0;	/* fool gcc */
1624 		sa6_src = &sa6_any;
1625 	}
1626 
1627 	if (ip6 != NULL) {
1628 		struct in_conninfo inc;
1629 		/*
1630 		 * XXX: We assume that when IPV6 is non NULL,
1631 		 * M and OFF are valid.
1632 		 */
1633 
1634 		/* check if we can safely examine src and dst ports */
1635 		if (m->m_pkthdr.len < off + sizeof(*thp))
1636 			return;
1637 
1638 		bzero(&th, sizeof(th));
1639 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1640 
1641 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1642 		    (struct sockaddr *)ip6cp->ip6c_src,
1643 		    th.th_sport, cmd, NULL, notify);
1644 
1645 		bzero(&inc, sizeof(inc));
1646 		inc.inc_fport = th.th_dport;
1647 		inc.inc_lport = th.th_sport;
1648 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1649 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1650 		inc.inc_flags |= INC_ISIPV6;
1651 		INP_INFO_WLOCK(&V_tcbinfo);
1652 		syncache_unreach(&inc, &th);
1653 		INP_INFO_WUNLOCK(&V_tcbinfo);
1654 	} else
1655 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1656 			      0, cmd, NULL, notify);
1657 }
1658 #endif /* INET6 */
1659 
1660 
1661 /*
1662  * Following is where TCP initial sequence number generation occurs.
1663  *
1664  * There are two places where we must use initial sequence numbers:
1665  * 1.  In SYN-ACK packets.
1666  * 2.  In SYN packets.
1667  *
1668  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1669  * tcp_syncache.c for details.
1670  *
1671  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1672  * depends on this property.  In addition, these ISNs should be
1673  * unguessable so as to prevent connection hijacking.  To satisfy
1674  * the requirements of this situation, the algorithm outlined in
1675  * RFC 1948 is used, with only small modifications.
1676  *
1677  * Implementation details:
1678  *
1679  * Time is based off the system timer, and is corrected so that it
1680  * increases by one megabyte per second.  This allows for proper
1681  * recycling on high speed LANs while still leaving over an hour
1682  * before rollover.
1683  *
1684  * As reading the *exact* system time is too expensive to be done
1685  * whenever setting up a TCP connection, we increment the time
1686  * offset in two ways.  First, a small random positive increment
1687  * is added to isn_offset for each connection that is set up.
1688  * Second, the function tcp_isn_tick fires once per clock tick
1689  * and increments isn_offset as necessary so that sequence numbers
1690  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1691  * random positive increments serve only to ensure that the same
1692  * exact sequence number is never sent out twice (as could otherwise
1693  * happen when a port is recycled in less than the system tick
1694  * interval.)
1695  *
1696  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1697  * between seeding of isn_secret.  This is normally set to zero,
1698  * as reseeding should not be necessary.
1699  *
1700  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1701  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1702  * general, this means holding an exclusive (write) lock.
1703  */
1704 
1705 #define ISN_BYTES_PER_SECOND 1048576
1706 #define ISN_STATIC_INCREMENT 4096
1707 #define ISN_RANDOM_INCREMENT (4096 - 1)
1708 
1709 static VNET_DEFINE(u_char, isn_secret[32]);
1710 static VNET_DEFINE(int, isn_last);
1711 static VNET_DEFINE(int, isn_last_reseed);
1712 static VNET_DEFINE(u_int32_t, isn_offset);
1713 static VNET_DEFINE(u_int32_t, isn_offset_old);
1714 
1715 #define	V_isn_secret			VNET(isn_secret)
1716 #define	V_isn_last			VNET(isn_last)
1717 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1718 #define	V_isn_offset			VNET(isn_offset)
1719 #define	V_isn_offset_old		VNET(isn_offset_old)
1720 
1721 tcp_seq
1722 tcp_new_isn(struct tcpcb *tp)
1723 {
1724 	MD5_CTX isn_ctx;
1725 	u_int32_t md5_buffer[4];
1726 	tcp_seq new_isn;
1727 	u_int32_t projected_offset;
1728 
1729 	INP_WLOCK_ASSERT(tp->t_inpcb);
1730 
1731 	ISN_LOCK();
1732 	/* Seed if this is the first use, reseed if requested. */
1733 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1734 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1735 		< (u_int)ticks))) {
1736 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1737 		V_isn_last_reseed = ticks;
1738 	}
1739 
1740 	/* Compute the md5 hash and return the ISN. */
1741 	MD5Init(&isn_ctx);
1742 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1743 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1744 #ifdef INET6
1745 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1746 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1747 			  sizeof(struct in6_addr));
1748 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1749 			  sizeof(struct in6_addr));
1750 	} else
1751 #endif
1752 	{
1753 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1754 			  sizeof(struct in_addr));
1755 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1756 			  sizeof(struct in_addr));
1757 	}
1758 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1759 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1760 	new_isn = (tcp_seq) md5_buffer[0];
1761 	V_isn_offset += ISN_STATIC_INCREMENT +
1762 		(arc4random() & ISN_RANDOM_INCREMENT);
1763 	if (ticks != V_isn_last) {
1764 		projected_offset = V_isn_offset_old +
1765 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1766 		if (SEQ_GT(projected_offset, V_isn_offset))
1767 			V_isn_offset = projected_offset;
1768 		V_isn_offset_old = V_isn_offset;
1769 		V_isn_last = ticks;
1770 	}
1771 	new_isn += V_isn_offset;
1772 	ISN_UNLOCK();
1773 	return (new_isn);
1774 }
1775 
1776 /*
1777  * When a specific ICMP unreachable message is received and the
1778  * connection state is SYN-SENT, drop the connection.  This behavior
1779  * is controlled by the icmp_may_rst sysctl.
1780  */
1781 struct inpcb *
1782 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1783 {
1784 	struct tcpcb *tp;
1785 
1786 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1787 	INP_WLOCK_ASSERT(inp);
1788 
1789 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1790 	    (inp->inp_flags & INP_DROPPED))
1791 		return (inp);
1792 
1793 	tp = intotcpcb(inp);
1794 	if (tp->t_state != TCPS_SYN_SENT)
1795 		return (inp);
1796 
1797 	tp = tcp_drop(tp, errno);
1798 	if (tp != NULL)
1799 		return (inp);
1800 	else
1801 		return (NULL);
1802 }
1803 
1804 /*
1805  * When `need fragmentation' ICMP is received, update our idea of the MSS
1806  * based on the new value. Also nudge TCP to send something, since we
1807  * know the packet we just sent was dropped.
1808  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1809  */
1810 static struct inpcb *
1811 tcp_mtudisc_notify(struct inpcb *inp, int error)
1812 {
1813 
1814 	return (tcp_mtudisc(inp, -1));
1815 }
1816 
1817 struct inpcb *
1818 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1819 {
1820 	struct tcpcb *tp;
1821 	struct socket *so;
1822 
1823 	INP_WLOCK_ASSERT(inp);
1824 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1825 	    (inp->inp_flags & INP_DROPPED))
1826 		return (inp);
1827 
1828 	tp = intotcpcb(inp);
1829 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1830 
1831 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1832 
1833 	so = inp->inp_socket;
1834 	SOCKBUF_LOCK(&so->so_snd);
1835 	/* If the mss is larger than the socket buffer, decrease the mss. */
1836 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1837 		tp->t_maxseg = so->so_snd.sb_hiwat;
1838 	SOCKBUF_UNLOCK(&so->so_snd);
1839 
1840 	TCPSTAT_INC(tcps_mturesent);
1841 	tp->t_rtttime = 0;
1842 	tp->snd_nxt = tp->snd_una;
1843 	tcp_free_sackholes(tp);
1844 	tp->snd_recover = tp->snd_max;
1845 	if (tp->t_flags & TF_SACK_PERMIT)
1846 		EXIT_FASTRECOVERY(tp->t_flags);
1847 	tcp_output(tp);
1848 	return (inp);
1849 }
1850 
1851 #ifdef INET
1852 /*
1853  * Look-up the routing entry to the peer of this inpcb.  If no route
1854  * is found and it cannot be allocated, then return 0.  This routine
1855  * is called by TCP routines that access the rmx structure and by
1856  * tcp_mss_update to get the peer/interface MTU.
1857  */
1858 u_long
1859 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1860 {
1861 	struct route sro;
1862 	struct sockaddr_in *dst;
1863 	struct ifnet *ifp;
1864 	u_long maxmtu = 0;
1865 
1866 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1867 
1868 	bzero(&sro, sizeof(sro));
1869 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1870 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1871 		dst->sin_family = AF_INET;
1872 		dst->sin_len = sizeof(*dst);
1873 		dst->sin_addr = inc->inc_faddr;
1874 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1875 	}
1876 	if (sro.ro_rt != NULL) {
1877 		ifp = sro.ro_rt->rt_ifp;
1878 		if (sro.ro_rt->rt_mtu == 0)
1879 			maxmtu = ifp->if_mtu;
1880 		else
1881 			maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1882 
1883 		/* Report additional interface capabilities. */
1884 		if (cap != NULL) {
1885 			if (ifp->if_capenable & IFCAP_TSO4 &&
1886 			    ifp->if_hwassist & CSUM_TSO) {
1887 				cap->ifcap |= CSUM_TSO;
1888 				cap->tsomax = ifp->if_hw_tsomax;
1889 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1890 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1891 			}
1892 		}
1893 		RTFREE(sro.ro_rt);
1894 	}
1895 	return (maxmtu);
1896 }
1897 #endif /* INET */
1898 
1899 #ifdef INET6
1900 u_long
1901 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1902 {
1903 	struct route_in6 sro6;
1904 	struct ifnet *ifp;
1905 	u_long maxmtu = 0;
1906 
1907 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1908 
1909 	bzero(&sro6, sizeof(sro6));
1910 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1911 		sro6.ro_dst.sin6_family = AF_INET6;
1912 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1913 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1914 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1915 	}
1916 	if (sro6.ro_rt != NULL) {
1917 		ifp = sro6.ro_rt->rt_ifp;
1918 		if (sro6.ro_rt->rt_mtu == 0)
1919 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1920 		else
1921 			maxmtu = min(sro6.ro_rt->rt_mtu,
1922 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1923 
1924 		/* Report additional interface capabilities. */
1925 		if (cap != NULL) {
1926 			if (ifp->if_capenable & IFCAP_TSO6 &&
1927 			    ifp->if_hwassist & CSUM_TSO) {
1928 				cap->ifcap |= CSUM_TSO;
1929 				cap->tsomax = ifp->if_hw_tsomax;
1930 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1931 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1932 			}
1933 		}
1934 		RTFREE(sro6.ro_rt);
1935 	}
1936 
1937 	return (maxmtu);
1938 }
1939 #endif /* INET6 */
1940 
1941 #ifdef IPSEC
1942 /* compute ESP/AH header size for TCP, including outer IP header. */
1943 size_t
1944 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1945 {
1946 	struct inpcb *inp;
1947 	struct mbuf *m;
1948 	size_t hdrsiz;
1949 	struct ip *ip;
1950 #ifdef INET6
1951 	struct ip6_hdr *ip6;
1952 #endif
1953 	struct tcphdr *th;
1954 
1955 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1956 		return (0);
1957 	m = m_gethdr(M_NOWAIT, MT_DATA);
1958 	if (!m)
1959 		return (0);
1960 
1961 #ifdef INET6
1962 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1963 		ip6 = mtod(m, struct ip6_hdr *);
1964 		th = (struct tcphdr *)(ip6 + 1);
1965 		m->m_pkthdr.len = m->m_len =
1966 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1967 		tcpip_fillheaders(inp, ip6, th);
1968 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1969 	} else
1970 #endif /* INET6 */
1971 	{
1972 		ip = mtod(m, struct ip *);
1973 		th = (struct tcphdr *)(ip + 1);
1974 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1975 		tcpip_fillheaders(inp, ip, th);
1976 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1977 	}
1978 
1979 	m_free(m);
1980 	return (hdrsiz);
1981 }
1982 #endif /* IPSEC */
1983 
1984 #ifdef TCP_SIGNATURE
1985 /*
1986  * Callback function invoked by m_apply() to digest TCP segment data
1987  * contained within an mbuf chain.
1988  */
1989 static int
1990 tcp_signature_apply(void *fstate, void *data, u_int len)
1991 {
1992 
1993 	MD5Update(fstate, (u_char *)data, len);
1994 	return (0);
1995 }
1996 
1997 /*
1998  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1999  * search with the destination IP address, and a 'magic SPI' to be
2000  * determined by the application. This is hardcoded elsewhere to 1179
2001 */
2002 struct secasvar *
2003 tcp_get_sav(struct mbuf *m, u_int direction)
2004 {
2005 	union sockaddr_union dst;
2006 	struct secasvar *sav;
2007 	struct ip *ip;
2008 #ifdef INET6
2009 	struct ip6_hdr *ip6;
2010 	char ip6buf[INET6_ADDRSTRLEN];
2011 #endif
2012 
2013 	/* Extract the destination from the IP header in the mbuf. */
2014 	bzero(&dst, sizeof(union sockaddr_union));
2015 	ip = mtod(m, struct ip *);
2016 #ifdef INET6
2017 	ip6 = NULL;	/* Make the compiler happy. */
2018 #endif
2019 	switch (ip->ip_v) {
2020 #ifdef INET
2021 	case IPVERSION:
2022 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2023 		dst.sa.sa_family = AF_INET;
2024 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2025 		    ip->ip_src : ip->ip_dst;
2026 		break;
2027 #endif
2028 #ifdef INET6
2029 	case (IPV6_VERSION >> 4):
2030 		ip6 = mtod(m, struct ip6_hdr *);
2031 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2032 		dst.sa.sa_family = AF_INET6;
2033 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2034 		    ip6->ip6_src : ip6->ip6_dst;
2035 		break;
2036 #endif
2037 	default:
2038 		return (NULL);
2039 		/* NOTREACHED */
2040 		break;
2041 	}
2042 
2043 	/* Look up an SADB entry which matches the address of the peer. */
2044 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2045 	if (sav == NULL) {
2046 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2047 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2048 #ifdef INET6
2049 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2050 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2051 #endif
2052 			"(unsupported)"));
2053 	}
2054 
2055 	return (sav);
2056 }
2057 
2058 /*
2059  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2060  *
2061  * Parameters:
2062  * m		pointer to head of mbuf chain
2063  * len		length of TCP segment data, excluding options
2064  * optlen	length of TCP segment options
2065  * buf		pointer to storage for computed MD5 digest
2066  * sav		pointer to security assosiation
2067  *
2068  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2069  * When called from tcp_input(), we can be sure that th_sum has been
2070  * zeroed out and verified already.
2071  *
2072  * Releases reference to SADB key before return.
2073  *
2074  * Return 0 if successful, otherwise return -1.
2075  *
2076  */
2077 int
2078 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2079     u_char *buf, struct secasvar *sav)
2080 {
2081 #ifdef INET
2082 	struct ippseudo ippseudo;
2083 #endif
2084 	MD5_CTX ctx;
2085 	int doff;
2086 	struct ip *ip;
2087 #ifdef INET
2088 	struct ipovly *ipovly;
2089 #endif
2090 	struct tcphdr *th;
2091 #ifdef INET6
2092 	struct ip6_hdr *ip6;
2093 	struct in6_addr in6;
2094 	uint32_t plen;
2095 	uint16_t nhdr;
2096 #endif
2097 	u_short savecsum;
2098 
2099 	KASSERT(m != NULL, ("NULL mbuf chain"));
2100 	KASSERT(buf != NULL, ("NULL signature pointer"));
2101 
2102 	/* Extract the destination from the IP header in the mbuf. */
2103 	ip = mtod(m, struct ip *);
2104 #ifdef INET6
2105 	ip6 = NULL;	/* Make the compiler happy. */
2106 #endif
2107 
2108 	MD5Init(&ctx);
2109 	/*
2110 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2111 	 *
2112 	 * XXX The ippseudo header MUST be digested in network byte order,
2113 	 * or else we'll fail the regression test. Assume all fields we've
2114 	 * been doing arithmetic on have been in host byte order.
2115 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2116 	 * tcp_output(), the underlying ip_len member has not yet been set.
2117 	 */
2118 	switch (ip->ip_v) {
2119 #ifdef INET
2120 	case IPVERSION:
2121 		ipovly = (struct ipovly *)ip;
2122 		ippseudo.ippseudo_src = ipovly->ih_src;
2123 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2124 		ippseudo.ippseudo_pad = 0;
2125 		ippseudo.ippseudo_p = IPPROTO_TCP;
2126 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2127 		    optlen);
2128 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2129 
2130 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2131 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2132 		break;
2133 #endif
2134 #ifdef INET6
2135 	/*
2136 	 * RFC 2385, 2.0  Proposal
2137 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2138 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2139 	 * extended next header value (to form 32 bits), and 32-bit segment
2140 	 * length.
2141 	 * Note: Upper-Layer Packet Length comes before Next Header.
2142 	 */
2143 	case (IPV6_VERSION >> 4):
2144 		in6 = ip6->ip6_src;
2145 		in6_clearscope(&in6);
2146 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2147 		in6 = ip6->ip6_dst;
2148 		in6_clearscope(&in6);
2149 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2150 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2151 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2152 		nhdr = 0;
2153 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2154 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2155 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2156 		nhdr = IPPROTO_TCP;
2157 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2158 
2159 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2160 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2161 		break;
2162 #endif
2163 	default:
2164 		KEY_FREESAV(&sav);
2165 		return (-1);
2166 		/* NOTREACHED */
2167 		break;
2168 	}
2169 
2170 
2171 	/*
2172 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2173 	 * The TCP checksum must be set to zero.
2174 	 */
2175 	savecsum = th->th_sum;
2176 	th->th_sum = 0;
2177 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2178 	th->th_sum = savecsum;
2179 
2180 	/*
2181 	 * Step 3: Update MD5 hash with TCP segment data.
2182 	 *         Use m_apply() to avoid an early m_pullup().
2183 	 */
2184 	if (len > 0)
2185 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2186 
2187 	/*
2188 	 * Step 4: Update MD5 hash with shared secret.
2189 	 */
2190 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2191 	MD5Final(buf, &ctx);
2192 
2193 	key_sa_recordxfer(sav, m);
2194 	KEY_FREESAV(&sav);
2195 	return (0);
2196 }
2197 
2198 /*
2199  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2200  *
2201  * Return 0 if successful, otherwise return -1.
2202  */
2203 int
2204 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2205     u_char *buf, u_int direction)
2206 {
2207 	struct secasvar *sav;
2208 
2209 	if ((sav = tcp_get_sav(m, direction)) == NULL)
2210 		return (-1);
2211 
2212 	return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2213 }
2214 
2215 /*
2216  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2217  *
2218  * Parameters:
2219  * m		pointer to head of mbuf chain
2220  * len		length of TCP segment data, excluding options
2221  * optlen	length of TCP segment options
2222  * buf		pointer to storage for computed MD5 digest
2223  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2224  *
2225  * Return 1 if successful, otherwise return 0.
2226  */
2227 int
2228 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2229     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2230 {
2231 	char tmpdigest[TCP_SIGLEN];
2232 
2233 	if (tcp_sig_checksigs == 0)
2234 		return (1);
2235 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2236 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2237 
2238 			/*
2239 			 * If this socket is not expecting signature but
2240 			 * the segment contains signature just fail.
2241 			 */
2242 			TCPSTAT_INC(tcps_sig_err_sigopt);
2243 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2244 			return (0);
2245 		}
2246 
2247 		/* Signature is not expected, and not present in segment. */
2248 		return (1);
2249 	}
2250 
2251 	/*
2252 	 * If this socket is expecting signature but the segment does not
2253 	 * contain any just fail.
2254 	 */
2255 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2256 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2257 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2258 		return (0);
2259 	}
2260 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2261 	    IPSEC_DIR_INBOUND) == -1) {
2262 		TCPSTAT_INC(tcps_sig_err_buildsig);
2263 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2264 		return (0);
2265 	}
2266 
2267 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2268 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2269 		return (0);
2270 	}
2271 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2272 	return (1);
2273 }
2274 #endif /* TCP_SIGNATURE */
2275 
2276 static int
2277 sysctl_drop(SYSCTL_HANDLER_ARGS)
2278 {
2279 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2280 	struct sockaddr_storage addrs[2];
2281 	struct inpcb *inp;
2282 	struct tcpcb *tp;
2283 	struct tcptw *tw;
2284 	struct sockaddr_in *fin, *lin;
2285 #ifdef INET6
2286 	struct sockaddr_in6 *fin6, *lin6;
2287 #endif
2288 	int error;
2289 
2290 	inp = NULL;
2291 	fin = lin = NULL;
2292 #ifdef INET6
2293 	fin6 = lin6 = NULL;
2294 #endif
2295 	error = 0;
2296 
2297 	if (req->oldptr != NULL || req->oldlen != 0)
2298 		return (EINVAL);
2299 	if (req->newptr == NULL)
2300 		return (EPERM);
2301 	if (req->newlen < sizeof(addrs))
2302 		return (ENOMEM);
2303 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2304 	if (error)
2305 		return (error);
2306 
2307 	switch (addrs[0].ss_family) {
2308 #ifdef INET6
2309 	case AF_INET6:
2310 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2311 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2312 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2313 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2314 			return (EINVAL);
2315 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2316 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2317 				return (EINVAL);
2318 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2319 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2320 			fin = (struct sockaddr_in *)&addrs[0];
2321 			lin = (struct sockaddr_in *)&addrs[1];
2322 			break;
2323 		}
2324 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2325 		if (error)
2326 			return (error);
2327 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2328 		if (error)
2329 			return (error);
2330 		break;
2331 #endif
2332 #ifdef INET
2333 	case AF_INET:
2334 		fin = (struct sockaddr_in *)&addrs[0];
2335 		lin = (struct sockaddr_in *)&addrs[1];
2336 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2337 		    lin->sin_len != sizeof(struct sockaddr_in))
2338 			return (EINVAL);
2339 		break;
2340 #endif
2341 	default:
2342 		return (EINVAL);
2343 	}
2344 	INP_INFO_WLOCK(&V_tcbinfo);
2345 	switch (addrs[0].ss_family) {
2346 #ifdef INET6
2347 	case AF_INET6:
2348 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2349 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2350 		    INPLOOKUP_WLOCKPCB, NULL);
2351 		break;
2352 #endif
2353 #ifdef INET
2354 	case AF_INET:
2355 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2356 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2357 		break;
2358 #endif
2359 	}
2360 	if (inp != NULL) {
2361 		if (inp->inp_flags & INP_TIMEWAIT) {
2362 			/*
2363 			 * XXXRW: There currently exists a state where an
2364 			 * inpcb is present, but its timewait state has been
2365 			 * discarded.  For now, don't allow dropping of this
2366 			 * type of inpcb.
2367 			 */
2368 			tw = intotw(inp);
2369 			if (tw != NULL)
2370 				tcp_twclose(tw, 0);
2371 			else
2372 				INP_WUNLOCK(inp);
2373 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2374 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2375 			tp = intotcpcb(inp);
2376 			tp = tcp_drop(tp, ECONNABORTED);
2377 			if (tp != NULL)
2378 				INP_WUNLOCK(inp);
2379 		} else
2380 			INP_WUNLOCK(inp);
2381 	} else
2382 		error = ESRCH;
2383 	INP_INFO_WUNLOCK(&V_tcbinfo);
2384 	return (error);
2385 }
2386 
2387 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2388     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2389     0, sysctl_drop, "", "Drop TCP connection");
2390 
2391 /*
2392  * Generate a standardized TCP log line for use throughout the
2393  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2394  * allow use in the interrupt context.
2395  *
2396  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2397  * NB: The function may return NULL if memory allocation failed.
2398  *
2399  * Due to header inclusion and ordering limitations the struct ip
2400  * and ip6_hdr pointers have to be passed as void pointers.
2401  */
2402 char *
2403 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2404     const void *ip6hdr)
2405 {
2406 
2407 	/* Is logging enabled? */
2408 	if (tcp_log_in_vain == 0)
2409 		return (NULL);
2410 
2411 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2412 }
2413 
2414 char *
2415 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2416     const void *ip6hdr)
2417 {
2418 
2419 	/* Is logging enabled? */
2420 	if (tcp_log_debug == 0)
2421 		return (NULL);
2422 
2423 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2424 }
2425 
2426 static char *
2427 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2428     const void *ip6hdr)
2429 {
2430 	char *s, *sp;
2431 	size_t size;
2432 	struct ip *ip;
2433 #ifdef INET6
2434 	const struct ip6_hdr *ip6;
2435 
2436 	ip6 = (const struct ip6_hdr *)ip6hdr;
2437 #endif /* INET6 */
2438 	ip = (struct ip *)ip4hdr;
2439 
2440 	/*
2441 	 * The log line looks like this:
2442 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2443 	 */
2444 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2445 	    sizeof(PRINT_TH_FLAGS) + 1 +
2446 #ifdef INET6
2447 	    2 * INET6_ADDRSTRLEN;
2448 #else
2449 	    2 * INET_ADDRSTRLEN;
2450 #endif /* INET6 */
2451 
2452 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2453 	if (s == NULL)
2454 		return (NULL);
2455 
2456 	strcat(s, "TCP: [");
2457 	sp = s + strlen(s);
2458 
2459 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2460 		inet_ntoa_r(inc->inc_faddr, sp);
2461 		sp = s + strlen(s);
2462 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2463 		sp = s + strlen(s);
2464 		inet_ntoa_r(inc->inc_laddr, sp);
2465 		sp = s + strlen(s);
2466 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2467 #ifdef INET6
2468 	} else if (inc) {
2469 		ip6_sprintf(sp, &inc->inc6_faddr);
2470 		sp = s + strlen(s);
2471 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2472 		sp = s + strlen(s);
2473 		ip6_sprintf(sp, &inc->inc6_laddr);
2474 		sp = s + strlen(s);
2475 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2476 	} else if (ip6 && th) {
2477 		ip6_sprintf(sp, &ip6->ip6_src);
2478 		sp = s + strlen(s);
2479 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2480 		sp = s + strlen(s);
2481 		ip6_sprintf(sp, &ip6->ip6_dst);
2482 		sp = s + strlen(s);
2483 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2484 #endif /* INET6 */
2485 #ifdef INET
2486 	} else if (ip && th) {
2487 		inet_ntoa_r(ip->ip_src, sp);
2488 		sp = s + strlen(s);
2489 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2490 		sp = s + strlen(s);
2491 		inet_ntoa_r(ip->ip_dst, sp);
2492 		sp = s + strlen(s);
2493 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2494 #endif /* INET */
2495 	} else {
2496 		free(s, M_TCPLOG);
2497 		return (NULL);
2498 	}
2499 	sp = s + strlen(s);
2500 	if (th)
2501 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2502 	if (*(s + size - 1) != '\0')
2503 		panic("%s: string too long", __func__);
2504 	return (s);
2505 }
2506 
2507 /*
2508  * A subroutine which makes it easy to track TCP state changes with DTrace.
2509  * This function shouldn't be called for t_state initializations that don't
2510  * correspond to actual TCP state transitions.
2511  */
2512 void
2513 tcp_state_change(struct tcpcb *tp, int newstate)
2514 {
2515 #if defined(KDTRACE_HOOKS)
2516 	int pstate = tp->t_state;
2517 #endif
2518 
2519 	tp->t_state = newstate;
2520 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2521 }
2522