xref: /freebsd/sys/netinet/tcp_subr.c (revision b2d48be1bc7df45ddd13b143a160d0acb5a383c5)
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 void tcp_mtudisc(struct inpcb *, int);
232 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
233 		    void *ip4hdr, const void *ip6hdr);
234 static void	tcp_timer_discard(struct tcpcb *, uint32_t);
235 
236 /*
237  * Target size of TCP PCB hash tables. Must be a power of two.
238  *
239  * Note that this can be overridden by the kernel environment
240  * variable net.inet.tcp.tcbhashsize
241  */
242 #ifndef TCBHASHSIZE
243 #define TCBHASHSIZE	0
244 #endif
245 
246 /*
247  * XXX
248  * Callouts should be moved into struct tcp directly.  They are currently
249  * separate because the tcpcb structure is exported to userland for sysctl
250  * parsing purposes, which do not know about callouts.
251  */
252 struct tcpcb_mem {
253 	struct	tcpcb		tcb;
254 	struct	tcp_timer	tt;
255 	struct	cc_var		ccv;
256 	struct	osd		osd;
257 };
258 
259 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
260 #define	V_tcpcb_zone			VNET(tcpcb_zone)
261 
262 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
263 static struct mtx isn_mtx;
264 
265 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
266 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
267 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
268 
269 /*
270  * TCP initialization.
271  */
272 static void
273 tcp_zone_change(void *tag)
274 {
275 
276 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
277 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
278 	tcp_tw_zone_change();
279 }
280 
281 static int
282 tcp_inpcb_init(void *mem, int size, int flags)
283 {
284 	struct inpcb *inp = mem;
285 
286 	INP_LOCK_INIT(inp, "inp", "tcpinp");
287 	return (0);
288 }
289 
290 /*
291  * Take a value and get the next power of 2 that doesn't overflow.
292  * Used to size the tcp_inpcb hash buckets.
293  */
294 static int
295 maketcp_hashsize(int size)
296 {
297 	int hashsize;
298 
299 	/*
300 	 * auto tune.
301 	 * get the next power of 2 higher than maxsockets.
302 	 */
303 	hashsize = 1 << fls(size);
304 	/* catch overflow, and just go one power of 2 smaller */
305 	if (hashsize < size) {
306 		hashsize = 1 << (fls(size) - 1);
307 	}
308 	return (hashsize);
309 }
310 
311 void
312 tcp_init(void)
313 {
314 	const char *tcbhash_tuneable;
315 	int hashsize;
316 
317 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
318 
319 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
320 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
321 		printf("%s: WARNING: unable to register helper hook\n", __func__);
322 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
323 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
324 		printf("%s: WARNING: unable to register helper hook\n", __func__);
325 
326 	hashsize = TCBHASHSIZE;
327 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
328 	if (hashsize == 0) {
329 		/*
330 		 * Auto tune the hash size based on maxsockets.
331 		 * A perfect hash would have a 1:1 mapping
332 		 * (hashsize = maxsockets) however it's been
333 		 * suggested that O(2) average is better.
334 		 */
335 		hashsize = maketcp_hashsize(maxsockets / 4);
336 		/*
337 		 * Our historical default is 512,
338 		 * do not autotune lower than this.
339 		 */
340 		if (hashsize < 512)
341 			hashsize = 512;
342 		if (bootverbose)
343 			printf("%s: %s auto tuned to %d\n", __func__,
344 			    tcbhash_tuneable, hashsize);
345 	}
346 	/*
347 	 * We require a hashsize to be a power of two.
348 	 * Previously if it was not a power of two we would just reset it
349 	 * back to 512, which could be a nasty surprise if you did not notice
350 	 * the error message.
351 	 * Instead what we do is clip it to the closest power of two lower
352 	 * than the specified hash value.
353 	 */
354 	if (!powerof2(hashsize)) {
355 		int oldhashsize = hashsize;
356 
357 		hashsize = maketcp_hashsize(hashsize);
358 		/* prevent absurdly low value */
359 		if (hashsize < 16)
360 			hashsize = 16;
361 		printf("%s: WARNING: TCB hash size not a power of 2, "
362 		    "clipped from %d to %d.\n", __func__, oldhashsize,
363 		    hashsize);
364 	}
365 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
366 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
367 	    IPI_HASHFIELDS_4TUPLE);
368 
369 	/*
370 	 * These have to be type stable for the benefit of the timers.
371 	 */
372 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
373 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
374 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
375 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
376 
377 	tcp_tw_init();
378 	syncache_init();
379 	tcp_hc_init();
380 
381 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
382 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
383 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384 
385 	/* Skip initialization of globals for non-default instances. */
386 	if (!IS_DEFAULT_VNET(curvnet))
387 		return;
388 
389 	tcp_reass_global_init();
390 
391 	/* XXX virtualize those bellow? */
392 	tcp_delacktime = TCPTV_DELACK;
393 	tcp_keepinit = TCPTV_KEEP_INIT;
394 	tcp_keepidle = TCPTV_KEEP_IDLE;
395 	tcp_keepintvl = TCPTV_KEEPINTVL;
396 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
397 	tcp_msl = TCPTV_MSL;
398 	tcp_rexmit_min = TCPTV_MIN;
399 	if (tcp_rexmit_min < 1)
400 		tcp_rexmit_min = 1;
401 	tcp_rexmit_slop = TCPTV_CPU_VAR;
402 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
403 	tcp_tcbhashsize = hashsize;
404 
405 	if (tcp_soreceive_stream) {
406 #ifdef INET
407 		tcp_usrreqs.pru_soreceive = soreceive_stream;
408 #endif
409 #ifdef INET6
410 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
411 #endif /* INET6 */
412 	}
413 
414 #ifdef INET6
415 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
416 #else /* INET6 */
417 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
418 #endif /* INET6 */
419 	if (max_protohdr < TCP_MINPROTOHDR)
420 		max_protohdr = TCP_MINPROTOHDR;
421 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
422 		panic("tcp_init");
423 #undef TCP_MINPROTOHDR
424 
425 	ISN_LOCK_INIT();
426 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
427 		SHUTDOWN_PRI_DEFAULT);
428 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
429 		EVENTHANDLER_PRI_ANY);
430 }
431 
432 #ifdef VIMAGE
433 void
434 tcp_destroy(void)
435 {
436 	int error;
437 
438 	tcp_hc_destroy();
439 	syncache_destroy();
440 	tcp_tw_destroy();
441 	in_pcbinfo_destroy(&V_tcbinfo);
442 	uma_zdestroy(V_sack_hole_zone);
443 	uma_zdestroy(V_tcpcb_zone);
444 
445 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
446 	if (error != 0) {
447 		printf("%s: WARNING: unable to deregister helper hook "
448 		    "type=%d, id=%d: error %d returned\n", __func__,
449 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
450 	}
451 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
452 	if (error != 0) {
453 		printf("%s: WARNING: unable to deregister helper hook "
454 		    "type=%d, id=%d: error %d returned\n", __func__,
455 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
456 	}
457 }
458 #endif
459 
460 void
461 tcp_fini(void *xtp)
462 {
463 
464 }
465 
466 /*
467  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
468  * tcp_template used to store this data in mbufs, but we now recopy it out
469  * of the tcpcb each time to conserve mbufs.
470  */
471 void
472 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
473 {
474 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
475 
476 	INP_WLOCK_ASSERT(inp);
477 
478 #ifdef INET6
479 	if ((inp->inp_vflag & INP_IPV6) != 0) {
480 		struct ip6_hdr *ip6;
481 
482 		ip6 = (struct ip6_hdr *)ip_ptr;
483 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
484 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
485 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
486 			(IPV6_VERSION & IPV6_VERSION_MASK);
487 		ip6->ip6_nxt = IPPROTO_TCP;
488 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
489 		ip6->ip6_src = inp->in6p_laddr;
490 		ip6->ip6_dst = inp->in6p_faddr;
491 	}
492 #endif /* INET6 */
493 #if defined(INET6) && defined(INET)
494 	else
495 #endif
496 #ifdef INET
497 	{
498 		struct ip *ip;
499 
500 		ip = (struct ip *)ip_ptr;
501 		ip->ip_v = IPVERSION;
502 		ip->ip_hl = 5;
503 		ip->ip_tos = inp->inp_ip_tos;
504 		ip->ip_len = 0;
505 		ip->ip_id = 0;
506 		ip->ip_off = 0;
507 		ip->ip_ttl = inp->inp_ip_ttl;
508 		ip->ip_sum = 0;
509 		ip->ip_p = IPPROTO_TCP;
510 		ip->ip_src = inp->inp_laddr;
511 		ip->ip_dst = inp->inp_faddr;
512 	}
513 #endif /* INET */
514 	th->th_sport = inp->inp_lport;
515 	th->th_dport = inp->inp_fport;
516 	th->th_seq = 0;
517 	th->th_ack = 0;
518 	th->th_x2 = 0;
519 	th->th_off = 5;
520 	th->th_flags = 0;
521 	th->th_win = 0;
522 	th->th_urp = 0;
523 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
524 }
525 
526 /*
527  * Create template to be used to send tcp packets on a connection.
528  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
529  * use for this function is in keepalives, which use tcp_respond.
530  */
531 struct tcptemp *
532 tcpip_maketemplate(struct inpcb *inp)
533 {
534 	struct tcptemp *t;
535 
536 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
537 	if (t == NULL)
538 		return (NULL);
539 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
540 	return (t);
541 }
542 
543 /*
544  * Send a single message to the TCP at address specified by
545  * the given TCP/IP header.  If m == NULL, then we make a copy
546  * of the tcpiphdr at th and send directly to the addressed host.
547  * This is used to force keep alive messages out using the TCP
548  * template for a connection.  If flags are given then we send
549  * a message back to the TCP which originated the segment th,
550  * and discard the mbuf containing it and any other attached mbufs.
551  *
552  * In any case the ack and sequence number of the transmitted
553  * segment are as specified by the parameters.
554  *
555  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
556  */
557 void
558 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
559     tcp_seq ack, tcp_seq seq, int flags)
560 {
561 	int tlen;
562 	int win = 0;
563 	struct ip *ip;
564 	struct tcphdr *nth;
565 #ifdef INET6
566 	struct ip6_hdr *ip6;
567 	int isipv6;
568 #endif /* INET6 */
569 	int ipflags = 0;
570 	struct inpcb *inp;
571 
572 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
573 
574 #ifdef INET6
575 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
576 	ip6 = ipgen;
577 #endif /* INET6 */
578 	ip = ipgen;
579 
580 	if (tp != NULL) {
581 		inp = tp->t_inpcb;
582 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
583 		INP_WLOCK_ASSERT(inp);
584 	} else
585 		inp = NULL;
586 
587 	if (tp != NULL) {
588 		if (!(flags & TH_RST)) {
589 			win = sbspace(&inp->inp_socket->so_rcv);
590 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
591 				win = (long)TCP_MAXWIN << tp->rcv_scale;
592 		}
593 	}
594 	if (m == NULL) {
595 		m = m_gethdr(M_NOWAIT, MT_DATA);
596 		if (m == NULL)
597 			return;
598 		tlen = 0;
599 		m->m_data += max_linkhdr;
600 #ifdef INET6
601 		if (isipv6) {
602 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
603 			      sizeof(struct ip6_hdr));
604 			ip6 = mtod(m, struct ip6_hdr *);
605 			nth = (struct tcphdr *)(ip6 + 1);
606 		} else
607 #endif /* INET6 */
608 		{
609 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
610 			ip = mtod(m, struct ip *);
611 			nth = (struct tcphdr *)(ip + 1);
612 		}
613 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
614 		flags = TH_ACK;
615 	} else {
616 		/*
617 		 *  reuse the mbuf.
618 		 * XXX MRT We inherrit the FIB, which is lucky.
619 		 */
620 		m_freem(m->m_next);
621 		m->m_next = NULL;
622 		m->m_data = (caddr_t)ipgen;
623 		/* m_len is set later */
624 		tlen = 0;
625 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
626 #ifdef INET6
627 		if (isipv6) {
628 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
629 			nth = (struct tcphdr *)(ip6 + 1);
630 		} else
631 #endif /* INET6 */
632 		{
633 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
634 			nth = (struct tcphdr *)(ip + 1);
635 		}
636 		if (th != nth) {
637 			/*
638 			 * this is usually a case when an extension header
639 			 * exists between the IPv6 header and the
640 			 * TCP header.
641 			 */
642 			nth->th_sport = th->th_sport;
643 			nth->th_dport = th->th_dport;
644 		}
645 		xchg(nth->th_dport, nth->th_sport, uint16_t);
646 #undef xchg
647 	}
648 #ifdef INET6
649 	if (isipv6) {
650 		ip6->ip6_flow = 0;
651 		ip6->ip6_vfc = IPV6_VERSION;
652 		ip6->ip6_nxt = IPPROTO_TCP;
653 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
654 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
655 	}
656 #endif
657 #if defined(INET) && defined(INET6)
658 	else
659 #endif
660 #ifdef INET
661 	{
662 		tlen += sizeof (struct tcpiphdr);
663 		ip->ip_len = htons(tlen);
664 		ip->ip_ttl = V_ip_defttl;
665 		if (V_path_mtu_discovery)
666 			ip->ip_off |= htons(IP_DF);
667 	}
668 #endif
669 	m->m_len = tlen;
670 	m->m_pkthdr.len = tlen;
671 	m->m_pkthdr.rcvif = NULL;
672 #ifdef MAC
673 	if (inp != NULL) {
674 		/*
675 		 * Packet is associated with a socket, so allow the
676 		 * label of the response to reflect the socket label.
677 		 */
678 		INP_WLOCK_ASSERT(inp);
679 		mac_inpcb_create_mbuf(inp, m);
680 	} else {
681 		/*
682 		 * Packet is not associated with a socket, so possibly
683 		 * update the label in place.
684 		 */
685 		mac_netinet_tcp_reply(m);
686 	}
687 #endif
688 	nth->th_seq = htonl(seq);
689 	nth->th_ack = htonl(ack);
690 	nth->th_x2 = 0;
691 	nth->th_off = sizeof (struct tcphdr) >> 2;
692 	nth->th_flags = flags;
693 	if (tp != NULL)
694 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
695 	else
696 		nth->th_win = htons((u_short)win);
697 	nth->th_urp = 0;
698 
699 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
700 #ifdef INET6
701 	if (isipv6) {
702 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
703 		nth->th_sum = in6_cksum_pseudo(ip6,
704 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
705 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
706 		    NULL, NULL);
707 	}
708 #endif /* INET6 */
709 #if defined(INET6) && defined(INET)
710 	else
711 #endif
712 #ifdef INET
713 	{
714 		m->m_pkthdr.csum_flags = CSUM_TCP;
715 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
716 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
717 	}
718 #endif /* INET */
719 #ifdef TCPDEBUG
720 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
721 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
722 #endif
723 	TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
724 	if (flags & TH_RST)
725 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
726 		    tp, nth);
727 
728 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
729 #ifdef INET6
730 	if (isipv6)
731 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
732 #endif /* INET6 */
733 #if defined(INET) && defined(INET6)
734 	else
735 #endif
736 #ifdef INET
737 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
738 #endif
739 }
740 
741 /*
742  * Create a new TCP control block, making an
743  * empty reassembly queue and hooking it to the argument
744  * protocol control block.  The `inp' parameter must have
745  * come from the zone allocator set up in tcp_init().
746  */
747 struct tcpcb *
748 tcp_newtcpcb(struct inpcb *inp)
749 {
750 	struct tcpcb_mem *tm;
751 	struct tcpcb *tp;
752 #ifdef INET6
753 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
754 #endif /* INET6 */
755 
756 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
757 	if (tm == NULL)
758 		return (NULL);
759 	tp = &tm->tcb;
760 
761 	/* Initialise cc_var struct for this tcpcb. */
762 	tp->ccv = &tm->ccv;
763 	tp->ccv->type = IPPROTO_TCP;
764 	tp->ccv->ccvc.tcp = tp;
765 
766 	/*
767 	 * Use the current system default CC algorithm.
768 	 */
769 	CC_LIST_RLOCK();
770 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
771 	CC_ALGO(tp) = CC_DEFAULT();
772 	CC_LIST_RUNLOCK();
773 
774 	if (CC_ALGO(tp)->cb_init != NULL)
775 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
776 			uma_zfree(V_tcpcb_zone, tm);
777 			return (NULL);
778 		}
779 
780 	tp->osd = &tm->osd;
781 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
782 		uma_zfree(V_tcpcb_zone, tm);
783 		return (NULL);
784 	}
785 
786 #ifdef VIMAGE
787 	tp->t_vnet = inp->inp_vnet;
788 #endif
789 	tp->t_timers = &tm->tt;
790 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
791 	tp->t_maxseg = tp->t_maxopd =
792 #ifdef INET6
793 		isipv6 ? V_tcp_v6mssdflt :
794 #endif /* INET6 */
795 		V_tcp_mssdflt;
796 
797 	/* Set up our timeouts. */
798 	callout_init(&tp->t_timers->tt_rexmt, 1);
799 	callout_init(&tp->t_timers->tt_persist, 1);
800 	callout_init(&tp->t_timers->tt_keep, 1);
801 	callout_init(&tp->t_timers->tt_2msl, 1);
802 	callout_init(&tp->t_timers->tt_delack, 1);
803 
804 	if (V_tcp_do_rfc1323)
805 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
806 	if (V_tcp_do_sack)
807 		tp->t_flags |= TF_SACK_PERMIT;
808 	TAILQ_INIT(&tp->snd_holes);
809 	/*
810 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
811 	 * is called.
812 	 */
813 	in_pcbref(inp);	/* Reference for tcpcb */
814 	tp->t_inpcb = inp;
815 
816 	/*
817 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
818 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
819 	 * reasonable initial retransmit time.
820 	 */
821 	tp->t_srtt = TCPTV_SRTTBASE;
822 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
823 	tp->t_rttmin = tcp_rexmit_min;
824 	tp->t_rxtcur = TCPTV_RTOBASE;
825 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
826 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
827 	tp->t_rcvtime = ticks;
828 	/*
829 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
830 	 * because the socket may be bound to an IPv6 wildcard address,
831 	 * which may match an IPv4-mapped IPv6 address.
832 	 */
833 	inp->inp_ip_ttl = V_ip_defttl;
834 	inp->inp_ppcb = tp;
835 	return (tp);		/* XXX */
836 }
837 
838 /*
839  * Switch the congestion control algorithm back to NewReno for any active
840  * control blocks using an algorithm which is about to go away.
841  * This ensures the CC framework can allow the unload to proceed without leaving
842  * any dangling pointers which would trigger a panic.
843  * Returning non-zero would inform the CC framework that something went wrong
844  * and it would be unsafe to allow the unload to proceed. However, there is no
845  * way for this to occur with this implementation so we always return zero.
846  */
847 int
848 tcp_ccalgounload(struct cc_algo *unload_algo)
849 {
850 	struct cc_algo *tmpalgo;
851 	struct inpcb *inp;
852 	struct tcpcb *tp;
853 	VNET_ITERATOR_DECL(vnet_iter);
854 
855 	/*
856 	 * Check all active control blocks across all network stacks and change
857 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
858 	 * requires cleanup code to be run, call it.
859 	 */
860 	VNET_LIST_RLOCK();
861 	VNET_FOREACH(vnet_iter) {
862 		CURVNET_SET(vnet_iter);
863 		INP_INFO_WLOCK(&V_tcbinfo);
864 		/*
865 		 * New connections already part way through being initialised
866 		 * with the CC algo we're removing will not race with this code
867 		 * because the INP_INFO_WLOCK is held during initialisation. We
868 		 * therefore don't enter the loop below until the connection
869 		 * list has stabilised.
870 		 */
871 		LIST_FOREACH(inp, &V_tcb, inp_list) {
872 			INP_WLOCK(inp);
873 			/* Important to skip tcptw structs. */
874 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
875 			    (tp = intotcpcb(inp)) != NULL) {
876 				/*
877 				 * By holding INP_WLOCK here, we are assured
878 				 * that the connection is not currently
879 				 * executing inside the CC module's functions
880 				 * i.e. it is safe to make the switch back to
881 				 * NewReno.
882 				 */
883 				if (CC_ALGO(tp) == unload_algo) {
884 					tmpalgo = CC_ALGO(tp);
885 					/* NewReno does not require any init. */
886 					CC_ALGO(tp) = &newreno_cc_algo;
887 					if (tmpalgo->cb_destroy != NULL)
888 						tmpalgo->cb_destroy(tp->ccv);
889 				}
890 			}
891 			INP_WUNLOCK(inp);
892 		}
893 		INP_INFO_WUNLOCK(&V_tcbinfo);
894 		CURVNET_RESTORE();
895 	}
896 	VNET_LIST_RUNLOCK();
897 
898 	return (0);
899 }
900 
901 /*
902  * Drop a TCP connection, reporting
903  * the specified error.  If connection is synchronized,
904  * then send a RST to peer.
905  */
906 struct tcpcb *
907 tcp_drop(struct tcpcb *tp, int errno)
908 {
909 	struct socket *so = tp->t_inpcb->inp_socket;
910 
911 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
912 	INP_WLOCK_ASSERT(tp->t_inpcb);
913 
914 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
915 		tcp_state_change(tp, TCPS_CLOSED);
916 		(void) tcp_output(tp);
917 		TCPSTAT_INC(tcps_drops);
918 	} else
919 		TCPSTAT_INC(tcps_conndrops);
920 	if (errno == ETIMEDOUT && tp->t_softerror)
921 		errno = tp->t_softerror;
922 	so->so_error = errno;
923 	return (tcp_close(tp));
924 }
925 
926 void
927 tcp_discardcb(struct tcpcb *tp)
928 {
929 	struct inpcb *inp = tp->t_inpcb;
930 	struct socket *so = inp->inp_socket;
931 #ifdef INET6
932 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
933 #endif /* INET6 */
934 	int released;
935 
936 	INP_WLOCK_ASSERT(inp);
937 
938 	/*
939 	 * Make sure that all of our timers are stopped before we delete the
940 	 * PCB.
941 	 *
942 	 * If stopping a timer fails, we schedule a discard function in same
943 	 * callout, and the last discard function called will take care of
944 	 * deleting the tcpcb.
945 	 */
946 	tcp_timer_stop(tp, TT_REXMT);
947 	tcp_timer_stop(tp, TT_PERSIST);
948 	tcp_timer_stop(tp, TT_KEEP);
949 	tcp_timer_stop(tp, TT_2MSL);
950 	tcp_timer_stop(tp, TT_DELACK);
951 
952 	/*
953 	 * If we got enough samples through the srtt filter,
954 	 * save the rtt and rttvar in the routing entry.
955 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
956 	 * 4 samples is enough for the srtt filter to converge
957 	 * to within enough % of the correct value; fewer samples
958 	 * and we could save a bogus rtt. The danger is not high
959 	 * as tcp quickly recovers from everything.
960 	 * XXX: Works very well but needs some more statistics!
961 	 */
962 	if (tp->t_rttupdated >= 4) {
963 		struct hc_metrics_lite metrics;
964 		u_long ssthresh;
965 
966 		bzero(&metrics, sizeof(metrics));
967 		/*
968 		 * Update the ssthresh always when the conditions below
969 		 * are satisfied. This gives us better new start value
970 		 * for the congestion avoidance for new connections.
971 		 * ssthresh is only set if packet loss occured on a session.
972 		 *
973 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
974 		 * being torn down.  Ideally this code would not use 'so'.
975 		 */
976 		ssthresh = tp->snd_ssthresh;
977 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
978 			/*
979 			 * convert the limit from user data bytes to
980 			 * packets then to packet data bytes.
981 			 */
982 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
983 			if (ssthresh < 2)
984 				ssthresh = 2;
985 			ssthresh *= (u_long)(tp->t_maxseg +
986 #ifdef INET6
987 			    (isipv6 ? sizeof (struct ip6_hdr) +
988 				sizeof (struct tcphdr) :
989 #endif
990 				sizeof (struct tcpiphdr)
991 #ifdef INET6
992 			    )
993 #endif
994 			    );
995 		} else
996 			ssthresh = 0;
997 		metrics.rmx_ssthresh = ssthresh;
998 
999 		metrics.rmx_rtt = tp->t_srtt;
1000 		metrics.rmx_rttvar = tp->t_rttvar;
1001 		metrics.rmx_cwnd = tp->snd_cwnd;
1002 		metrics.rmx_sendpipe = 0;
1003 		metrics.rmx_recvpipe = 0;
1004 
1005 		tcp_hc_update(&inp->inp_inc, &metrics);
1006 	}
1007 
1008 	/* free the reassembly queue, if any */
1009 	tcp_reass_flush(tp);
1010 
1011 #ifdef TCP_OFFLOAD
1012 	/* Disconnect offload device, if any. */
1013 	if (tp->t_flags & TF_TOE)
1014 		tcp_offload_detach(tp);
1015 #endif
1016 
1017 	tcp_free_sackholes(tp);
1018 
1019 	/* Allow the CC algorithm to clean up after itself. */
1020 	if (CC_ALGO(tp)->cb_destroy != NULL)
1021 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1022 
1023 	khelp_destroy_osd(tp->osd);
1024 
1025 	CC_ALGO(tp) = NULL;
1026 	inp->inp_ppcb = NULL;
1027 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1028 		/* We own the last reference on tcpcb, let's free it. */
1029 		tp->t_inpcb = NULL;
1030 		uma_zfree(V_tcpcb_zone, tp);
1031 		released = in_pcbrele_wlocked(inp);
1032 		KASSERT(!released, ("%s: inp %p should not have been released "
1033 			"here", __func__, inp));
1034 	}
1035 }
1036 
1037 void
1038 tcp_timer_2msl_discard(void *xtp)
1039 {
1040 
1041 	tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1042 }
1043 
1044 void
1045 tcp_timer_keep_discard(void *xtp)
1046 {
1047 
1048 	tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1049 }
1050 
1051 void
1052 tcp_timer_persist_discard(void *xtp)
1053 {
1054 
1055 	tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1056 }
1057 
1058 void
1059 tcp_timer_rexmt_discard(void *xtp)
1060 {
1061 
1062 	tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1063 }
1064 
1065 void
1066 tcp_timer_delack_discard(void *xtp)
1067 {
1068 
1069 	tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1070 }
1071 
1072 void
1073 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1074 {
1075 	struct inpcb *inp;
1076 
1077 	CURVNET_SET(tp->t_vnet);
1078 	INP_INFO_RLOCK(&V_tcbinfo);
1079 	inp = tp->t_inpcb;
1080 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1081 		__func__, tp));
1082 	INP_WLOCK(inp);
1083 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1084 		("%s: tcpcb has to be stopped here", __func__));
1085 	KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1086 		("%s: discard callout should be running", __func__));
1087 	tp->t_timers->tt_flags &= ~timer_type;
1088 	if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1089 		/* We own the last reference on this tcpcb, let's free it. */
1090 		tp->t_inpcb = NULL;
1091 		uma_zfree(V_tcpcb_zone, tp);
1092 		if (in_pcbrele_wlocked(inp)) {
1093 			INP_INFO_RUNLOCK(&V_tcbinfo);
1094 			CURVNET_RESTORE();
1095 			return;
1096 		}
1097 	}
1098 	INP_WUNLOCK(inp);
1099 	INP_INFO_RUNLOCK(&V_tcbinfo);
1100 	CURVNET_RESTORE();
1101 }
1102 
1103 /*
1104  * Attempt to close a TCP control block, marking it as dropped, and freeing
1105  * the socket if we hold the only reference.
1106  */
1107 struct tcpcb *
1108 tcp_close(struct tcpcb *tp)
1109 {
1110 	struct inpcb *inp = tp->t_inpcb;
1111 	struct socket *so;
1112 
1113 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1114 	INP_WLOCK_ASSERT(inp);
1115 
1116 #ifdef TCP_OFFLOAD
1117 	if (tp->t_state == TCPS_LISTEN)
1118 		tcp_offload_listen_stop(tp);
1119 #endif
1120 	in_pcbdrop(inp);
1121 	TCPSTAT_INC(tcps_closed);
1122 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1123 	so = inp->inp_socket;
1124 	soisdisconnected(so);
1125 	if (inp->inp_flags & INP_SOCKREF) {
1126 		KASSERT(so->so_state & SS_PROTOREF,
1127 		    ("tcp_close: !SS_PROTOREF"));
1128 		inp->inp_flags &= ~INP_SOCKREF;
1129 		INP_WUNLOCK(inp);
1130 		ACCEPT_LOCK();
1131 		SOCK_LOCK(so);
1132 		so->so_state &= ~SS_PROTOREF;
1133 		sofree(so);
1134 		return (NULL);
1135 	}
1136 	return (tp);
1137 }
1138 
1139 void
1140 tcp_drain(void)
1141 {
1142 	VNET_ITERATOR_DECL(vnet_iter);
1143 
1144 	if (!do_tcpdrain)
1145 		return;
1146 
1147 	VNET_LIST_RLOCK_NOSLEEP();
1148 	VNET_FOREACH(vnet_iter) {
1149 		CURVNET_SET(vnet_iter);
1150 		struct inpcb *inpb;
1151 		struct tcpcb *tcpb;
1152 
1153 	/*
1154 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1155 	 * if there is one...
1156 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1157 	 *      reassembly queue should be flushed, but in a situation
1158 	 *	where we're really low on mbufs, this is potentially
1159 	 *	useful.
1160 	 */
1161 		INP_INFO_WLOCK(&V_tcbinfo);
1162 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1163 			if (inpb->inp_flags & INP_TIMEWAIT)
1164 				continue;
1165 			INP_WLOCK(inpb);
1166 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1167 				tcp_reass_flush(tcpb);
1168 				tcp_clean_sackreport(tcpb);
1169 			}
1170 			INP_WUNLOCK(inpb);
1171 		}
1172 		INP_INFO_WUNLOCK(&V_tcbinfo);
1173 		CURVNET_RESTORE();
1174 	}
1175 	VNET_LIST_RUNLOCK_NOSLEEP();
1176 }
1177 
1178 /*
1179  * Notify a tcp user of an asynchronous error;
1180  * store error as soft error, but wake up user
1181  * (for now, won't do anything until can select for soft error).
1182  *
1183  * Do not wake up user since there currently is no mechanism for
1184  * reporting soft errors (yet - a kqueue filter may be added).
1185  */
1186 static struct inpcb *
1187 tcp_notify(struct inpcb *inp, int error)
1188 {
1189 	struct tcpcb *tp;
1190 
1191 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1192 	INP_WLOCK_ASSERT(inp);
1193 
1194 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1195 	    (inp->inp_flags & INP_DROPPED))
1196 		return (inp);
1197 
1198 	tp = intotcpcb(inp);
1199 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1200 
1201 	/*
1202 	 * Ignore some errors if we are hooked up.
1203 	 * If connection hasn't completed, has retransmitted several times,
1204 	 * and receives a second error, give up now.  This is better
1205 	 * than waiting a long time to establish a connection that
1206 	 * can never complete.
1207 	 */
1208 	if (tp->t_state == TCPS_ESTABLISHED &&
1209 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1210 	     error == EHOSTDOWN)) {
1211 		return (inp);
1212 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1213 	    tp->t_softerror) {
1214 		tp = tcp_drop(tp, error);
1215 		if (tp != NULL)
1216 			return (inp);
1217 		else
1218 			return (NULL);
1219 	} else {
1220 		tp->t_softerror = error;
1221 		return (inp);
1222 	}
1223 #if 0
1224 	wakeup( &so->so_timeo);
1225 	sorwakeup(so);
1226 	sowwakeup(so);
1227 #endif
1228 }
1229 
1230 static int
1231 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1232 {
1233 	int error, i, m, n, pcb_count;
1234 	struct inpcb *inp, **inp_list;
1235 	inp_gen_t gencnt;
1236 	struct xinpgen xig;
1237 
1238 	/*
1239 	 * The process of preparing the TCB list is too time-consuming and
1240 	 * resource-intensive to repeat twice on every request.
1241 	 */
1242 	if (req->oldptr == NULL) {
1243 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1244 		n += imax(n / 8, 10);
1245 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1246 		return (0);
1247 	}
1248 
1249 	if (req->newptr != NULL)
1250 		return (EPERM);
1251 
1252 	/*
1253 	 * OK, now we're committed to doing something.
1254 	 */
1255 	INP_LIST_RLOCK(&V_tcbinfo);
1256 	gencnt = V_tcbinfo.ipi_gencnt;
1257 	n = V_tcbinfo.ipi_count;
1258 	INP_LIST_RUNLOCK(&V_tcbinfo);
1259 
1260 	m = syncache_pcbcount();
1261 
1262 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1263 		+ (n + m) * sizeof(struct xtcpcb));
1264 	if (error != 0)
1265 		return (error);
1266 
1267 	xig.xig_len = sizeof xig;
1268 	xig.xig_count = n + m;
1269 	xig.xig_gen = gencnt;
1270 	xig.xig_sogen = so_gencnt;
1271 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1272 	if (error)
1273 		return (error);
1274 
1275 	error = syncache_pcblist(req, m, &pcb_count);
1276 	if (error)
1277 		return (error);
1278 
1279 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1280 	if (inp_list == NULL)
1281 		return (ENOMEM);
1282 
1283 	INP_INFO_WLOCK(&V_tcbinfo);
1284 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1285 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1286 		INP_WLOCK(inp);
1287 		if (inp->inp_gencnt <= gencnt) {
1288 			/*
1289 			 * XXX: This use of cr_cansee(), introduced with
1290 			 * TCP state changes, is not quite right, but for
1291 			 * now, better than nothing.
1292 			 */
1293 			if (inp->inp_flags & INP_TIMEWAIT) {
1294 				if (intotw(inp) != NULL)
1295 					error = cr_cansee(req->td->td_ucred,
1296 					    intotw(inp)->tw_cred);
1297 				else
1298 					error = EINVAL;	/* Skip this inp. */
1299 			} else
1300 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1301 			if (error == 0) {
1302 				in_pcbref(inp);
1303 				inp_list[i++] = inp;
1304 			}
1305 		}
1306 		INP_WUNLOCK(inp);
1307 	}
1308 	INP_INFO_WUNLOCK(&V_tcbinfo);
1309 	n = i;
1310 
1311 	error = 0;
1312 	for (i = 0; i < n; i++) {
1313 		inp = inp_list[i];
1314 		INP_RLOCK(inp);
1315 		if (inp->inp_gencnt <= gencnt) {
1316 			struct xtcpcb xt;
1317 			void *inp_ppcb;
1318 
1319 			bzero(&xt, sizeof(xt));
1320 			xt.xt_len = sizeof xt;
1321 			/* XXX should avoid extra copy */
1322 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1323 			inp_ppcb = inp->inp_ppcb;
1324 			if (inp_ppcb == NULL)
1325 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1326 			else if (inp->inp_flags & INP_TIMEWAIT) {
1327 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1328 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1329 			} else {
1330 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1331 				if (xt.xt_tp.t_timers)
1332 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1333 			}
1334 			if (inp->inp_socket != NULL)
1335 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1336 			else {
1337 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1338 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1339 			}
1340 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1341 			INP_RUNLOCK(inp);
1342 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1343 		} else
1344 			INP_RUNLOCK(inp);
1345 	}
1346 	INP_INFO_RLOCK(&V_tcbinfo);
1347 	for (i = 0; i < n; i++) {
1348 		inp = inp_list[i];
1349 		INP_RLOCK(inp);
1350 		if (!in_pcbrele_rlocked(inp))
1351 			INP_RUNLOCK(inp);
1352 	}
1353 	INP_INFO_RUNLOCK(&V_tcbinfo);
1354 
1355 	if (!error) {
1356 		/*
1357 		 * Give the user an updated idea of our state.
1358 		 * If the generation differs from what we told
1359 		 * her before, she knows that something happened
1360 		 * while we were processing this request, and it
1361 		 * might be necessary to retry.
1362 		 */
1363 		INP_LIST_RLOCK(&V_tcbinfo);
1364 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1365 		xig.xig_sogen = so_gencnt;
1366 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1367 		INP_LIST_RUNLOCK(&V_tcbinfo);
1368 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1369 	}
1370 	free(inp_list, M_TEMP);
1371 	return (error);
1372 }
1373 
1374 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1375     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1376     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1377 
1378 #ifdef INET
1379 static int
1380 tcp_getcred(SYSCTL_HANDLER_ARGS)
1381 {
1382 	struct xucred xuc;
1383 	struct sockaddr_in addrs[2];
1384 	struct inpcb *inp;
1385 	int error;
1386 
1387 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1388 	if (error)
1389 		return (error);
1390 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1391 	if (error)
1392 		return (error);
1393 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1394 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1395 	if (inp != NULL) {
1396 		if (inp->inp_socket == NULL)
1397 			error = ENOENT;
1398 		if (error == 0)
1399 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1400 		if (error == 0)
1401 			cru2x(inp->inp_cred, &xuc);
1402 		INP_RUNLOCK(inp);
1403 	} else
1404 		error = ENOENT;
1405 	if (error == 0)
1406 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1407 	return (error);
1408 }
1409 
1410 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1411     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1412     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1413 #endif /* INET */
1414 
1415 #ifdef INET6
1416 static int
1417 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1418 {
1419 	struct xucred xuc;
1420 	struct sockaddr_in6 addrs[2];
1421 	struct inpcb *inp;
1422 	int error;
1423 #ifdef INET
1424 	int mapped = 0;
1425 #endif
1426 
1427 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1428 	if (error)
1429 		return (error);
1430 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1431 	if (error)
1432 		return (error);
1433 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1434 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1435 		return (error);
1436 	}
1437 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1438 #ifdef INET
1439 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1440 			mapped = 1;
1441 		else
1442 #endif
1443 			return (EINVAL);
1444 	}
1445 
1446 #ifdef INET
1447 	if (mapped == 1)
1448 		inp = in_pcblookup(&V_tcbinfo,
1449 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1450 			addrs[1].sin6_port,
1451 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1452 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1453 	else
1454 #endif
1455 		inp = in6_pcblookup(&V_tcbinfo,
1456 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1457 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1458 			INPLOOKUP_RLOCKPCB, NULL);
1459 	if (inp != NULL) {
1460 		if (inp->inp_socket == NULL)
1461 			error = ENOENT;
1462 		if (error == 0)
1463 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1464 		if (error == 0)
1465 			cru2x(inp->inp_cred, &xuc);
1466 		INP_RUNLOCK(inp);
1467 	} else
1468 		error = ENOENT;
1469 	if (error == 0)
1470 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1471 	return (error);
1472 }
1473 
1474 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1475     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1476     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1477 #endif /* INET6 */
1478 
1479 
1480 #ifdef INET
1481 void
1482 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1483 {
1484 	struct ip *ip = vip;
1485 	struct tcphdr *th;
1486 	struct in_addr faddr;
1487 	struct inpcb *inp;
1488 	struct tcpcb *tp;
1489 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1490 	struct icmp *icp;
1491 	struct in_conninfo inc;
1492 	tcp_seq icmp_tcp_seq;
1493 	int mtu;
1494 
1495 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1496 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1497 		return;
1498 
1499 	if (cmd == PRC_MSGSIZE)
1500 		notify = tcp_mtudisc_notify;
1501 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1502 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1503 		notify = tcp_drop_syn_sent;
1504 	/*
1505 	 * Redirects don't need to be handled up here.
1506 	 */
1507 	else if (PRC_IS_REDIRECT(cmd))
1508 		return;
1509 	/*
1510 	 * Hostdead is ugly because it goes linearly through all PCBs.
1511 	 * XXX: We never get this from ICMP, otherwise it makes an
1512 	 * excellent DoS attack on machines with many connections.
1513 	 */
1514 	else if (cmd == PRC_HOSTDEAD)
1515 		ip = NULL;
1516 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1517 		return;
1518 
1519 	if (ip == NULL) {
1520 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1521 		return;
1522 	}
1523 
1524 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1525 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1526 	INP_INFO_RLOCK(&V_tcbinfo);
1527 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1528 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1529 	if (inp != NULL)  {
1530 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
1531 		    !(inp->inp_flags & INP_DROPPED) &&
1532 		    !(inp->inp_socket == NULL)) {
1533 			icmp_tcp_seq = ntohl(th->th_seq);
1534 			tp = intotcpcb(inp);
1535 			if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1536 			    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1537 				if (cmd == PRC_MSGSIZE) {
1538 					/*
1539 					 * MTU discovery:
1540 					 * If we got a needfrag set the MTU
1541 					 * in the route to the suggested new
1542 					 * value (if given) and then notify.
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 process the offered MTU if it
1559 					 * is smaller than the current one.
1560 					 */
1561 					if (mtu < tp->t_maxopd +
1562 					    sizeof(struct tcpiphdr)) {
1563 						bzero(&inc, sizeof(inc));
1564 						inc.inc_faddr = faddr;
1565 						inc.inc_fibnum =
1566 						    inp->inp_inc.inc_fibnum;
1567 						tcp_hc_updatemtu(&inc, mtu);
1568 						tcp_mtudisc(inp, mtu);
1569 					}
1570 				} else
1571 					inp = (*notify)(inp,
1572 					    inetctlerrmap[cmd]);
1573 			}
1574 		}
1575 		if (inp != NULL)
1576 			INP_WUNLOCK(inp);
1577 	} else {
1578 		bzero(&inc, sizeof(inc));
1579 		inc.inc_fport = th->th_dport;
1580 		inc.inc_lport = th->th_sport;
1581 		inc.inc_faddr = faddr;
1582 		inc.inc_laddr = ip->ip_src;
1583 		syncache_unreach(&inc, th);
1584 	}
1585 	INP_INFO_RUNLOCK(&V_tcbinfo);
1586 }
1587 #endif /* INET */
1588 
1589 #ifdef INET6
1590 void
1591 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1592 {
1593 	struct tcphdr th;
1594 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1595 	struct ip6_hdr *ip6;
1596 	struct mbuf *m;
1597 	struct ip6ctlparam *ip6cp = NULL;
1598 	const struct sockaddr_in6 *sa6_src = NULL;
1599 	int off;
1600 	struct tcp_portonly {
1601 		u_int16_t th_sport;
1602 		u_int16_t th_dport;
1603 	} *thp;
1604 
1605 	if (sa->sa_family != AF_INET6 ||
1606 	    sa->sa_len != sizeof(struct sockaddr_in6))
1607 		return;
1608 
1609 	if (cmd == PRC_MSGSIZE)
1610 		notify = tcp_mtudisc_notify;
1611 	else if (!PRC_IS_REDIRECT(cmd) &&
1612 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1613 		return;
1614 
1615 	/* if the parameter is from icmp6, decode it. */
1616 	if (d != NULL) {
1617 		ip6cp = (struct ip6ctlparam *)d;
1618 		m = ip6cp->ip6c_m;
1619 		ip6 = ip6cp->ip6c_ip6;
1620 		off = ip6cp->ip6c_off;
1621 		sa6_src = ip6cp->ip6c_src;
1622 	} else {
1623 		m = NULL;
1624 		ip6 = NULL;
1625 		off = 0;	/* fool gcc */
1626 		sa6_src = &sa6_any;
1627 	}
1628 
1629 	if (ip6 != NULL) {
1630 		struct in_conninfo inc;
1631 		/*
1632 		 * XXX: We assume that when IPV6 is non NULL,
1633 		 * M and OFF are valid.
1634 		 */
1635 
1636 		/* check if we can safely examine src and dst ports */
1637 		if (m->m_pkthdr.len < off + sizeof(*thp))
1638 			return;
1639 
1640 		bzero(&th, sizeof(th));
1641 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1642 
1643 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1644 		    (struct sockaddr *)ip6cp->ip6c_src,
1645 		    th.th_sport, cmd, NULL, notify);
1646 
1647 		bzero(&inc, sizeof(inc));
1648 		inc.inc_fport = th.th_dport;
1649 		inc.inc_lport = th.th_sport;
1650 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1651 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1652 		inc.inc_flags |= INC_ISIPV6;
1653 		INP_INFO_RLOCK(&V_tcbinfo);
1654 		syncache_unreach(&inc, &th);
1655 		INP_INFO_RUNLOCK(&V_tcbinfo);
1656 	} else
1657 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1658 			      0, cmd, NULL, notify);
1659 }
1660 #endif /* INET6 */
1661 
1662 
1663 /*
1664  * Following is where TCP initial sequence number generation occurs.
1665  *
1666  * There are two places where we must use initial sequence numbers:
1667  * 1.  In SYN-ACK packets.
1668  * 2.  In SYN packets.
1669  *
1670  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1671  * tcp_syncache.c for details.
1672  *
1673  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1674  * depends on this property.  In addition, these ISNs should be
1675  * unguessable so as to prevent connection hijacking.  To satisfy
1676  * the requirements of this situation, the algorithm outlined in
1677  * RFC 1948 is used, with only small modifications.
1678  *
1679  * Implementation details:
1680  *
1681  * Time is based off the system timer, and is corrected so that it
1682  * increases by one megabyte per second.  This allows for proper
1683  * recycling on high speed LANs while still leaving over an hour
1684  * before rollover.
1685  *
1686  * As reading the *exact* system time is too expensive to be done
1687  * whenever setting up a TCP connection, we increment the time
1688  * offset in two ways.  First, a small random positive increment
1689  * is added to isn_offset for each connection that is set up.
1690  * Second, the function tcp_isn_tick fires once per clock tick
1691  * and increments isn_offset as necessary so that sequence numbers
1692  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1693  * random positive increments serve only to ensure that the same
1694  * exact sequence number is never sent out twice (as could otherwise
1695  * happen when a port is recycled in less than the system tick
1696  * interval.)
1697  *
1698  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1699  * between seeding of isn_secret.  This is normally set to zero,
1700  * as reseeding should not be necessary.
1701  *
1702  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1703  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1704  * general, this means holding an exclusive (write) lock.
1705  */
1706 
1707 #define ISN_BYTES_PER_SECOND 1048576
1708 #define ISN_STATIC_INCREMENT 4096
1709 #define ISN_RANDOM_INCREMENT (4096 - 1)
1710 
1711 static VNET_DEFINE(u_char, isn_secret[32]);
1712 static VNET_DEFINE(int, isn_last);
1713 static VNET_DEFINE(int, isn_last_reseed);
1714 static VNET_DEFINE(u_int32_t, isn_offset);
1715 static VNET_DEFINE(u_int32_t, isn_offset_old);
1716 
1717 #define	V_isn_secret			VNET(isn_secret)
1718 #define	V_isn_last			VNET(isn_last)
1719 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1720 #define	V_isn_offset			VNET(isn_offset)
1721 #define	V_isn_offset_old		VNET(isn_offset_old)
1722 
1723 tcp_seq
1724 tcp_new_isn(struct tcpcb *tp)
1725 {
1726 	MD5_CTX isn_ctx;
1727 	u_int32_t md5_buffer[4];
1728 	tcp_seq new_isn;
1729 	u_int32_t projected_offset;
1730 
1731 	INP_WLOCK_ASSERT(tp->t_inpcb);
1732 
1733 	ISN_LOCK();
1734 	/* Seed if this is the first use, reseed if requested. */
1735 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1736 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1737 		< (u_int)ticks))) {
1738 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1739 		V_isn_last_reseed = ticks;
1740 	}
1741 
1742 	/* Compute the md5 hash and return the ISN. */
1743 	MD5Init(&isn_ctx);
1744 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1745 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1746 #ifdef INET6
1747 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1748 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1749 			  sizeof(struct in6_addr));
1750 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1751 			  sizeof(struct in6_addr));
1752 	} else
1753 #endif
1754 	{
1755 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1756 			  sizeof(struct in_addr));
1757 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1758 			  sizeof(struct in_addr));
1759 	}
1760 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1761 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1762 	new_isn = (tcp_seq) md5_buffer[0];
1763 	V_isn_offset += ISN_STATIC_INCREMENT +
1764 		(arc4random() & ISN_RANDOM_INCREMENT);
1765 	if (ticks != V_isn_last) {
1766 		projected_offset = V_isn_offset_old +
1767 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1768 		if (SEQ_GT(projected_offset, V_isn_offset))
1769 			V_isn_offset = projected_offset;
1770 		V_isn_offset_old = V_isn_offset;
1771 		V_isn_last = ticks;
1772 	}
1773 	new_isn += V_isn_offset;
1774 	ISN_UNLOCK();
1775 	return (new_isn);
1776 }
1777 
1778 /*
1779  * When a specific ICMP unreachable message is received and the
1780  * connection state is SYN-SENT, drop the connection.  This behavior
1781  * is controlled by the icmp_may_rst sysctl.
1782  */
1783 struct inpcb *
1784 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1785 {
1786 	struct tcpcb *tp;
1787 
1788 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1789 	INP_WLOCK_ASSERT(inp);
1790 
1791 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1792 	    (inp->inp_flags & INP_DROPPED))
1793 		return (inp);
1794 
1795 	tp = intotcpcb(inp);
1796 	if (tp->t_state != TCPS_SYN_SENT)
1797 		return (inp);
1798 
1799 	tp = tcp_drop(tp, errno);
1800 	if (tp != NULL)
1801 		return (inp);
1802 	else
1803 		return (NULL);
1804 }
1805 
1806 /*
1807  * When `need fragmentation' ICMP is received, update our idea of the MSS
1808  * based on the new value. Also nudge TCP to send something, since we
1809  * know the packet we just sent was dropped.
1810  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1811  */
1812 static struct inpcb *
1813 tcp_mtudisc_notify(struct inpcb *inp, int error)
1814 {
1815 
1816 	tcp_mtudisc(inp, -1);
1817 	return (inp);
1818 }
1819 
1820 static void
1821 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1822 {
1823 	struct tcpcb *tp;
1824 	struct socket *so;
1825 
1826 	INP_WLOCK_ASSERT(inp);
1827 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1828 	    (inp->inp_flags & INP_DROPPED))
1829 		return;
1830 
1831 	tp = intotcpcb(inp);
1832 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1833 
1834 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1835 
1836 	so = inp->inp_socket;
1837 	SOCKBUF_LOCK(&so->so_snd);
1838 	/* If the mss is larger than the socket buffer, decrease the mss. */
1839 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1840 		tp->t_maxseg = so->so_snd.sb_hiwat;
1841 	SOCKBUF_UNLOCK(&so->so_snd);
1842 
1843 	TCPSTAT_INC(tcps_mturesent);
1844 	tp->t_rtttime = 0;
1845 	tp->snd_nxt = tp->snd_una;
1846 	tcp_free_sackholes(tp);
1847 	tp->snd_recover = tp->snd_max;
1848 	if (tp->t_flags & TF_SACK_PERMIT)
1849 		EXIT_FASTRECOVERY(tp->t_flags);
1850 	tcp_output(tp);
1851 }
1852 
1853 #ifdef INET
1854 /*
1855  * Look-up the routing entry to the peer of this inpcb.  If no route
1856  * is found and it cannot be allocated, then return 0.  This routine
1857  * is called by TCP routines that access the rmx structure and by
1858  * tcp_mss_update to get the peer/interface MTU.
1859  */
1860 u_long
1861 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1862 {
1863 	struct route sro;
1864 	struct sockaddr_in *dst;
1865 	struct ifnet *ifp;
1866 	u_long maxmtu = 0;
1867 
1868 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1869 
1870 	bzero(&sro, sizeof(sro));
1871 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1872 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1873 		dst->sin_family = AF_INET;
1874 		dst->sin_len = sizeof(*dst);
1875 		dst->sin_addr = inc->inc_faddr;
1876 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1877 	}
1878 	if (sro.ro_rt != NULL) {
1879 		ifp = sro.ro_rt->rt_ifp;
1880 		if (sro.ro_rt->rt_mtu == 0)
1881 			maxmtu = ifp->if_mtu;
1882 		else
1883 			maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1884 
1885 		/* Report additional interface capabilities. */
1886 		if (cap != NULL) {
1887 			if (ifp->if_capenable & IFCAP_TSO4 &&
1888 			    ifp->if_hwassist & CSUM_TSO) {
1889 				cap->ifcap |= CSUM_TSO;
1890 				cap->tsomax = ifp->if_hw_tsomax;
1891 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1892 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1893 			}
1894 		}
1895 		RTFREE(sro.ro_rt);
1896 	}
1897 	return (maxmtu);
1898 }
1899 #endif /* INET */
1900 
1901 #ifdef INET6
1902 u_long
1903 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1904 {
1905 	struct route_in6 sro6;
1906 	struct ifnet *ifp;
1907 	u_long maxmtu = 0;
1908 
1909 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1910 
1911 	bzero(&sro6, sizeof(sro6));
1912 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1913 		sro6.ro_dst.sin6_family = AF_INET6;
1914 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1915 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1916 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1917 	}
1918 	if (sro6.ro_rt != NULL) {
1919 		ifp = sro6.ro_rt->rt_ifp;
1920 		if (sro6.ro_rt->rt_mtu == 0)
1921 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1922 		else
1923 			maxmtu = min(sro6.ro_rt->rt_mtu,
1924 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1925 
1926 		/* Report additional interface capabilities. */
1927 		if (cap != NULL) {
1928 			if (ifp->if_capenable & IFCAP_TSO6 &&
1929 			    ifp->if_hwassist & CSUM_TSO) {
1930 				cap->ifcap |= CSUM_TSO;
1931 				cap->tsomax = ifp->if_hw_tsomax;
1932 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1933 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1934 			}
1935 		}
1936 		RTFREE(sro6.ro_rt);
1937 	}
1938 
1939 	return (maxmtu);
1940 }
1941 #endif /* INET6 */
1942 
1943 #ifdef IPSEC
1944 /* compute ESP/AH header size for TCP, including outer IP header. */
1945 size_t
1946 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1947 {
1948 	struct inpcb *inp;
1949 	struct mbuf *m;
1950 	size_t hdrsiz;
1951 	struct ip *ip;
1952 #ifdef INET6
1953 	struct ip6_hdr *ip6;
1954 #endif
1955 	struct tcphdr *th;
1956 
1957 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1958 		return (0);
1959 	m = m_gethdr(M_NOWAIT, MT_DATA);
1960 	if (!m)
1961 		return (0);
1962 
1963 #ifdef INET6
1964 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1965 		ip6 = mtod(m, struct ip6_hdr *);
1966 		th = (struct tcphdr *)(ip6 + 1);
1967 		m->m_pkthdr.len = m->m_len =
1968 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1969 		tcpip_fillheaders(inp, ip6, th);
1970 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1971 	} else
1972 #endif /* INET6 */
1973 	{
1974 		ip = mtod(m, struct ip *);
1975 		th = (struct tcphdr *)(ip + 1);
1976 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1977 		tcpip_fillheaders(inp, ip, th);
1978 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1979 	}
1980 
1981 	m_free(m);
1982 	return (hdrsiz);
1983 }
1984 #endif /* IPSEC */
1985 
1986 #ifdef TCP_SIGNATURE
1987 /*
1988  * Callback function invoked by m_apply() to digest TCP segment data
1989  * contained within an mbuf chain.
1990  */
1991 static int
1992 tcp_signature_apply(void *fstate, void *data, u_int len)
1993 {
1994 
1995 	MD5Update(fstate, (u_char *)data, len);
1996 	return (0);
1997 }
1998 
1999 /*
2000  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2001  * search with the destination IP address, and a 'magic SPI' to be
2002  * determined by the application. This is hardcoded elsewhere to 1179
2003 */
2004 struct secasvar *
2005 tcp_get_sav(struct mbuf *m, u_int direction)
2006 {
2007 	union sockaddr_union dst;
2008 	struct secasvar *sav;
2009 	struct ip *ip;
2010 #ifdef INET6
2011 	struct ip6_hdr *ip6;
2012 	char ip6buf[INET6_ADDRSTRLEN];
2013 #endif
2014 
2015 	/* Extract the destination from the IP header in the mbuf. */
2016 	bzero(&dst, sizeof(union sockaddr_union));
2017 	ip = mtod(m, struct ip *);
2018 #ifdef INET6
2019 	ip6 = NULL;	/* Make the compiler happy. */
2020 #endif
2021 	switch (ip->ip_v) {
2022 #ifdef INET
2023 	case IPVERSION:
2024 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2025 		dst.sa.sa_family = AF_INET;
2026 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2027 		    ip->ip_src : ip->ip_dst;
2028 		break;
2029 #endif
2030 #ifdef INET6
2031 	case (IPV6_VERSION >> 4):
2032 		ip6 = mtod(m, struct ip6_hdr *);
2033 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2034 		dst.sa.sa_family = AF_INET6;
2035 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2036 		    ip6->ip6_src : ip6->ip6_dst;
2037 		break;
2038 #endif
2039 	default:
2040 		return (NULL);
2041 		/* NOTREACHED */
2042 		break;
2043 	}
2044 
2045 	/* Look up an SADB entry which matches the address of the peer. */
2046 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2047 	if (sav == NULL) {
2048 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2049 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2050 #ifdef INET6
2051 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2052 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2053 #endif
2054 			"(unsupported)"));
2055 	}
2056 
2057 	return (sav);
2058 }
2059 
2060 /*
2061  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2062  *
2063  * Parameters:
2064  * m		pointer to head of mbuf chain
2065  * len		length of TCP segment data, excluding options
2066  * optlen	length of TCP segment options
2067  * buf		pointer to storage for computed MD5 digest
2068  * sav		pointer to security assosiation
2069  *
2070  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2071  * When called from tcp_input(), we can be sure that th_sum has been
2072  * zeroed out and verified already.
2073  *
2074  * Releases reference to SADB key before return.
2075  *
2076  * Return 0 if successful, otherwise return -1.
2077  *
2078  */
2079 int
2080 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2081     u_char *buf, struct secasvar *sav)
2082 {
2083 #ifdef INET
2084 	struct ippseudo ippseudo;
2085 #endif
2086 	MD5_CTX ctx;
2087 	int doff;
2088 	struct ip *ip;
2089 #ifdef INET
2090 	struct ipovly *ipovly;
2091 #endif
2092 	struct tcphdr *th;
2093 #ifdef INET6
2094 	struct ip6_hdr *ip6;
2095 	struct in6_addr in6;
2096 	uint32_t plen;
2097 	uint16_t nhdr;
2098 #endif
2099 	u_short savecsum;
2100 
2101 	KASSERT(m != NULL, ("NULL mbuf chain"));
2102 	KASSERT(buf != NULL, ("NULL signature pointer"));
2103 
2104 	/* Extract the destination from the IP header in the mbuf. */
2105 	ip = mtod(m, struct ip *);
2106 #ifdef INET6
2107 	ip6 = NULL;	/* Make the compiler happy. */
2108 #endif
2109 
2110 	MD5Init(&ctx);
2111 	/*
2112 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2113 	 *
2114 	 * XXX The ippseudo header MUST be digested in network byte order,
2115 	 * or else we'll fail the regression test. Assume all fields we've
2116 	 * been doing arithmetic on have been in host byte order.
2117 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2118 	 * tcp_output(), the underlying ip_len member has not yet been set.
2119 	 */
2120 	switch (ip->ip_v) {
2121 #ifdef INET
2122 	case IPVERSION:
2123 		ipovly = (struct ipovly *)ip;
2124 		ippseudo.ippseudo_src = ipovly->ih_src;
2125 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2126 		ippseudo.ippseudo_pad = 0;
2127 		ippseudo.ippseudo_p = IPPROTO_TCP;
2128 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2129 		    optlen);
2130 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2131 
2132 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2133 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2134 		break;
2135 #endif
2136 #ifdef INET6
2137 	/*
2138 	 * RFC 2385, 2.0  Proposal
2139 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2140 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2141 	 * extended next header value (to form 32 bits), and 32-bit segment
2142 	 * length.
2143 	 * Note: Upper-Layer Packet Length comes before Next Header.
2144 	 */
2145 	case (IPV6_VERSION >> 4):
2146 		in6 = ip6->ip6_src;
2147 		in6_clearscope(&in6);
2148 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2149 		in6 = ip6->ip6_dst;
2150 		in6_clearscope(&in6);
2151 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2152 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2153 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2154 		nhdr = 0;
2155 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2156 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2157 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2158 		nhdr = IPPROTO_TCP;
2159 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2160 
2161 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2162 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2163 		break;
2164 #endif
2165 	default:
2166 		KEY_FREESAV(&sav);
2167 		return (-1);
2168 		/* NOTREACHED */
2169 		break;
2170 	}
2171 
2172 
2173 	/*
2174 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2175 	 * The TCP checksum must be set to zero.
2176 	 */
2177 	savecsum = th->th_sum;
2178 	th->th_sum = 0;
2179 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2180 	th->th_sum = savecsum;
2181 
2182 	/*
2183 	 * Step 3: Update MD5 hash with TCP segment data.
2184 	 *         Use m_apply() to avoid an early m_pullup().
2185 	 */
2186 	if (len > 0)
2187 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2188 
2189 	/*
2190 	 * Step 4: Update MD5 hash with shared secret.
2191 	 */
2192 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2193 	MD5Final(buf, &ctx);
2194 
2195 	key_sa_recordxfer(sav, m);
2196 	KEY_FREESAV(&sav);
2197 	return (0);
2198 }
2199 
2200 /*
2201  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2202  *
2203  * Return 0 if successful, otherwise return -1.
2204  */
2205 int
2206 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2207     u_char *buf, u_int direction)
2208 {
2209 	struct secasvar *sav;
2210 
2211 	if ((sav = tcp_get_sav(m, direction)) == NULL)
2212 		return (-1);
2213 
2214 	return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2215 }
2216 
2217 /*
2218  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2219  *
2220  * Parameters:
2221  * m		pointer to head of mbuf chain
2222  * len		length of TCP segment data, excluding options
2223  * optlen	length of TCP segment options
2224  * buf		pointer to storage for computed MD5 digest
2225  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2226  *
2227  * Return 1 if successful, otherwise return 0.
2228  */
2229 int
2230 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2231     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2232 {
2233 	char tmpdigest[TCP_SIGLEN];
2234 
2235 	if (tcp_sig_checksigs == 0)
2236 		return (1);
2237 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2238 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2239 
2240 			/*
2241 			 * If this socket is not expecting signature but
2242 			 * the segment contains signature just fail.
2243 			 */
2244 			TCPSTAT_INC(tcps_sig_err_sigopt);
2245 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2246 			return (0);
2247 		}
2248 
2249 		/* Signature is not expected, and not present in segment. */
2250 		return (1);
2251 	}
2252 
2253 	/*
2254 	 * If this socket is expecting signature but the segment does not
2255 	 * contain any just fail.
2256 	 */
2257 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2258 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2259 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2260 		return (0);
2261 	}
2262 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2263 	    IPSEC_DIR_INBOUND) == -1) {
2264 		TCPSTAT_INC(tcps_sig_err_buildsig);
2265 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2266 		return (0);
2267 	}
2268 
2269 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2270 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2271 		return (0);
2272 	}
2273 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2274 	return (1);
2275 }
2276 #endif /* TCP_SIGNATURE */
2277 
2278 static int
2279 sysctl_drop(SYSCTL_HANDLER_ARGS)
2280 {
2281 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2282 	struct sockaddr_storage addrs[2];
2283 	struct inpcb *inp;
2284 	struct tcpcb *tp;
2285 	struct tcptw *tw;
2286 	struct sockaddr_in *fin, *lin;
2287 #ifdef INET6
2288 	struct sockaddr_in6 *fin6, *lin6;
2289 #endif
2290 	int error;
2291 
2292 	inp = NULL;
2293 	fin = lin = NULL;
2294 #ifdef INET6
2295 	fin6 = lin6 = NULL;
2296 #endif
2297 	error = 0;
2298 
2299 	if (req->oldptr != NULL || req->oldlen != 0)
2300 		return (EINVAL);
2301 	if (req->newptr == NULL)
2302 		return (EPERM);
2303 	if (req->newlen < sizeof(addrs))
2304 		return (ENOMEM);
2305 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2306 	if (error)
2307 		return (error);
2308 
2309 	switch (addrs[0].ss_family) {
2310 #ifdef INET6
2311 	case AF_INET6:
2312 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2313 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2314 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2315 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2316 			return (EINVAL);
2317 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2318 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2319 				return (EINVAL);
2320 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2321 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2322 			fin = (struct sockaddr_in *)&addrs[0];
2323 			lin = (struct sockaddr_in *)&addrs[1];
2324 			break;
2325 		}
2326 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2327 		if (error)
2328 			return (error);
2329 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2330 		if (error)
2331 			return (error);
2332 		break;
2333 #endif
2334 #ifdef INET
2335 	case AF_INET:
2336 		fin = (struct sockaddr_in *)&addrs[0];
2337 		lin = (struct sockaddr_in *)&addrs[1];
2338 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2339 		    lin->sin_len != sizeof(struct sockaddr_in))
2340 			return (EINVAL);
2341 		break;
2342 #endif
2343 	default:
2344 		return (EINVAL);
2345 	}
2346 	INP_INFO_RLOCK(&V_tcbinfo);
2347 	switch (addrs[0].ss_family) {
2348 #ifdef INET6
2349 	case AF_INET6:
2350 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2351 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2352 		    INPLOOKUP_WLOCKPCB, NULL);
2353 		break;
2354 #endif
2355 #ifdef INET
2356 	case AF_INET:
2357 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2358 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2359 		break;
2360 #endif
2361 	}
2362 	if (inp != NULL) {
2363 		if (inp->inp_flags & INP_TIMEWAIT) {
2364 			/*
2365 			 * XXXRW: There currently exists a state where an
2366 			 * inpcb is present, but its timewait state has been
2367 			 * discarded.  For now, don't allow dropping of this
2368 			 * type of inpcb.
2369 			 */
2370 			tw = intotw(inp);
2371 			if (tw != NULL)
2372 				tcp_twclose(tw, 0);
2373 			else
2374 				INP_WUNLOCK(inp);
2375 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2376 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2377 			tp = intotcpcb(inp);
2378 			tp = tcp_drop(tp, ECONNABORTED);
2379 			if (tp != NULL)
2380 				INP_WUNLOCK(inp);
2381 		} else
2382 			INP_WUNLOCK(inp);
2383 	} else
2384 		error = ESRCH;
2385 	INP_INFO_RUNLOCK(&V_tcbinfo);
2386 	return (error);
2387 }
2388 
2389 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2390     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2391     0, sysctl_drop, "", "Drop TCP connection");
2392 
2393 /*
2394  * Generate a standardized TCP log line for use throughout the
2395  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2396  * allow use in the interrupt context.
2397  *
2398  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2399  * NB: The function may return NULL if memory allocation failed.
2400  *
2401  * Due to header inclusion and ordering limitations the struct ip
2402  * and ip6_hdr pointers have to be passed as void pointers.
2403  */
2404 char *
2405 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2406     const void *ip6hdr)
2407 {
2408 
2409 	/* Is logging enabled? */
2410 	if (tcp_log_in_vain == 0)
2411 		return (NULL);
2412 
2413 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2414 }
2415 
2416 char *
2417 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2418     const void *ip6hdr)
2419 {
2420 
2421 	/* Is logging enabled? */
2422 	if (tcp_log_debug == 0)
2423 		return (NULL);
2424 
2425 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2426 }
2427 
2428 static char *
2429 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2430     const void *ip6hdr)
2431 {
2432 	char *s, *sp;
2433 	size_t size;
2434 	struct ip *ip;
2435 #ifdef INET6
2436 	const struct ip6_hdr *ip6;
2437 
2438 	ip6 = (const struct ip6_hdr *)ip6hdr;
2439 #endif /* INET6 */
2440 	ip = (struct ip *)ip4hdr;
2441 
2442 	/*
2443 	 * The log line looks like this:
2444 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2445 	 */
2446 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2447 	    sizeof(PRINT_TH_FLAGS) + 1 +
2448 #ifdef INET6
2449 	    2 * INET6_ADDRSTRLEN;
2450 #else
2451 	    2 * INET_ADDRSTRLEN;
2452 #endif /* INET6 */
2453 
2454 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2455 	if (s == NULL)
2456 		return (NULL);
2457 
2458 	strcat(s, "TCP: [");
2459 	sp = s + strlen(s);
2460 
2461 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2462 		inet_ntoa_r(inc->inc_faddr, sp);
2463 		sp = s + strlen(s);
2464 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2465 		sp = s + strlen(s);
2466 		inet_ntoa_r(inc->inc_laddr, sp);
2467 		sp = s + strlen(s);
2468 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2469 #ifdef INET6
2470 	} else if (inc) {
2471 		ip6_sprintf(sp, &inc->inc6_faddr);
2472 		sp = s + strlen(s);
2473 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2474 		sp = s + strlen(s);
2475 		ip6_sprintf(sp, &inc->inc6_laddr);
2476 		sp = s + strlen(s);
2477 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2478 	} else if (ip6 && th) {
2479 		ip6_sprintf(sp, &ip6->ip6_src);
2480 		sp = s + strlen(s);
2481 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2482 		sp = s + strlen(s);
2483 		ip6_sprintf(sp, &ip6->ip6_dst);
2484 		sp = s + strlen(s);
2485 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2486 #endif /* INET6 */
2487 #ifdef INET
2488 	} else if (ip && th) {
2489 		inet_ntoa_r(ip->ip_src, sp);
2490 		sp = s + strlen(s);
2491 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2492 		sp = s + strlen(s);
2493 		inet_ntoa_r(ip->ip_dst, sp);
2494 		sp = s + strlen(s);
2495 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2496 #endif /* INET */
2497 	} else {
2498 		free(s, M_TCPLOG);
2499 		return (NULL);
2500 	}
2501 	sp = s + strlen(s);
2502 	if (th)
2503 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2504 	if (*(s + size - 1) != '\0')
2505 		panic("%s: string too long", __func__);
2506 	return (s);
2507 }
2508 
2509 /*
2510  * A subroutine which makes it easy to track TCP state changes with DTrace.
2511  * This function shouldn't be called for t_state initializations that don't
2512  * correspond to actual TCP state transitions.
2513  */
2514 void
2515 tcp_state_change(struct tcpcb *tp, int newstate)
2516 {
2517 #if defined(KDTRACE_HOOKS)
2518 	int pstate = tp->t_state;
2519 #endif
2520 
2521 	tp->t_state = newstate;
2522 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2523 }
2524