xref: /freebsd/sys/netinet/tcp_subr.c (revision ddd5b8e9b4d8957fce018c520657cdfa4ecffad3)
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/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/protosw.h>
59 #include <sys/random.h>
60 
61 #include <vm/uma.h>
62 
63 #include <net/route.h>
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/cc.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet/ip6.h>
77 #include <netinet6/in6_pcb.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #ifdef INET6
89 #include <netinet6/tcp6_var.h>
90 #endif
91 #include <netinet/tcpip.h>
92 #ifdef TCPDEBUG
93 #include <netinet/tcp_debug.h>
94 #endif
95 #ifdef INET6
96 #include <netinet6/ip6protosw.h>
97 #endif
98 #ifdef TCP_OFFLOAD
99 #include <netinet/tcp_offload.h>
100 #endif
101 
102 #ifdef IPSEC
103 #include <netipsec/ipsec.h>
104 #include <netipsec/xform.h>
105 #ifdef INET6
106 #include <netipsec/ipsec6.h>
107 #endif
108 #include <netipsec/key.h>
109 #include <sys/syslog.h>
110 #endif /*IPSEC*/
111 
112 #include <machine/in_cksum.h>
113 #include <sys/md5.h>
114 
115 #include <security/mac/mac_framework.h>
116 
117 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
118 #ifdef INET6
119 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
120 #endif
121 
122 static int
123 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
124 {
125 	int error, new;
126 
127 	new = V_tcp_mssdflt;
128 	error = sysctl_handle_int(oidp, &new, 0, req);
129 	if (error == 0 && req->newptr) {
130 		if (new < TCP_MINMSS)
131 			error = EINVAL;
132 		else
133 			V_tcp_mssdflt = new;
134 	}
135 	return (error);
136 }
137 
138 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
139     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
140     &sysctl_net_inet_tcp_mss_check, "I",
141     "Default TCP Maximum Segment Size");
142 
143 #ifdef INET6
144 static int
145 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
146 {
147 	int error, new;
148 
149 	new = V_tcp_v6mssdflt;
150 	error = sysctl_handle_int(oidp, &new, 0, req);
151 	if (error == 0 && req->newptr) {
152 		if (new < TCP_MINMSS)
153 			error = EINVAL;
154 		else
155 			V_tcp_v6mssdflt = new;
156 	}
157 	return (error);
158 }
159 
160 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
161     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
162     &sysctl_net_inet_tcp_mss_v6_check, "I",
163    "Default TCP Maximum Segment Size for IPv6");
164 #endif /* INET6 */
165 
166 /*
167  * Minimum MSS we accept and use. This prevents DoS attacks where
168  * we are forced to a ridiculous low MSS like 20 and send hundreds
169  * of packets instead of one. The effect scales with the available
170  * bandwidth and quickly saturates the CPU and network interface
171  * with packet generation and sending. Set to zero to disable MINMSS
172  * checking. This setting prevents us from sending too small packets.
173  */
174 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
176      &VNET_NAME(tcp_minmss), 0,
177     "Minmum TCP Maximum Segment Size");
178 
179 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
180 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
181     &VNET_NAME(tcp_do_rfc1323), 0,
182     "Enable rfc1323 (high performance TCP) extensions");
183 
184 static int	tcp_log_debug = 0;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
186     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
187 
188 static int	tcp_tcbhashsize = 0;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
190     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
191 
192 static int	do_tcpdrain = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
194     "Enable tcp_drain routine for extra help when low on mbufs");
195 
196 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
197     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
198 
199 static VNET_DEFINE(int, icmp_may_rst) = 1;
200 #define	V_icmp_may_rst			VNET(icmp_may_rst)
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
202     &VNET_NAME(icmp_may_rst), 0,
203     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
204 
205 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
206 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
207 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
208     &VNET_NAME(tcp_isn_reseed_interval), 0,
209     "Seconds between reseeding of ISN secret");
210 
211 static int	tcp_soreceive_stream = 0;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
213     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
214 
215 #ifdef TCP_SIGNATURE
216 static int	tcp_sig_checksigs = 1;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
218     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
219 #endif
220 
221 VNET_DEFINE(uma_zone_t, sack_hole_zone);
222 #define	V_sack_hole_zone		VNET(sack_hole_zone)
223 
224 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
225 
226 static struct inpcb *tcp_notify(struct inpcb *, int);
227 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
228 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
229 		    void *ip4hdr, const void *ip6hdr);
230 
231 /*
232  * Target size of TCP PCB hash tables. Must be a power of two.
233  *
234  * Note that this can be overridden by the kernel environment
235  * variable net.inet.tcp.tcbhashsize
236  */
237 #ifndef TCBHASHSIZE
238 #define TCBHASHSIZE	0
239 #endif
240 
241 /*
242  * XXX
243  * Callouts should be moved into struct tcp directly.  They are currently
244  * separate because the tcpcb structure is exported to userland for sysctl
245  * parsing purposes, which do not know about callouts.
246  */
247 struct tcpcb_mem {
248 	struct	tcpcb		tcb;
249 	struct	tcp_timer	tt;
250 	struct	cc_var		ccv;
251 	struct	osd		osd;
252 };
253 
254 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
255 #define	V_tcpcb_zone			VNET(tcpcb_zone)
256 
257 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
258 static struct mtx_padalign	isn_mtx;
259 
260 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
261 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
262 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
263 
264 /*
265  * TCP initialization.
266  */
267 static void
268 tcp_zone_change(void *tag)
269 {
270 
271 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
272 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
273 	tcp_tw_zone_change();
274 }
275 
276 static int
277 tcp_inpcb_init(void *mem, int size, int flags)
278 {
279 	struct inpcb *inp = mem;
280 
281 	INP_LOCK_INIT(inp, "inp", "tcpinp");
282 	return (0);
283 }
284 
285 /*
286  * Take a value and get the next power of 2 that doesn't overflow.
287  * Used to size the tcp_inpcb hash buckets.
288  */
289 static int
290 maketcp_hashsize(int size)
291 {
292 	int hashsize;
293 
294 	/*
295 	 * auto tune.
296 	 * get the next power of 2 higher than maxsockets.
297 	 */
298 	hashsize = 1 << fls(size);
299 	/* catch overflow, and just go one power of 2 smaller */
300 	if (hashsize < size) {
301 		hashsize = 1 << (fls(size) - 1);
302 	}
303 	return (hashsize);
304 }
305 
306 void
307 tcp_init(void)
308 {
309 	const char *tcbhash_tuneable;
310 	int hashsize;
311 
312 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
313 
314 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
315 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
316 		printf("%s: WARNING: unable to register helper hook\n", __func__);
317 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
318 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 		printf("%s: WARNING: unable to register helper hook\n", __func__);
320 
321 	hashsize = TCBHASHSIZE;
322 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
323 	if (hashsize == 0) {
324 		/*
325 		 * Auto tune the hash size based on maxsockets.
326 		 * A perfect hash would have a 1:1 mapping
327 		 * (hashsize = maxsockets) however it's been
328 		 * suggested that O(2) average is better.
329 		 */
330 		hashsize = maketcp_hashsize(maxsockets / 4);
331 		/*
332 		 * Our historical default is 512,
333 		 * do not autotune lower than this.
334 		 */
335 		if (hashsize < 512)
336 			hashsize = 512;
337 		if (bootverbose)
338 			printf("%s: %s auto tuned to %d\n", __func__,
339 			    tcbhash_tuneable, hashsize);
340 	}
341 	/*
342 	 * We require a hashsize to be a power of two.
343 	 * Previously if it was not a power of two we would just reset it
344 	 * back to 512, which could be a nasty surprise if you did not notice
345 	 * the error message.
346 	 * Instead what we do is clip it to the closest power of two lower
347 	 * than the specified hash value.
348 	 */
349 	if (!powerof2(hashsize)) {
350 		int oldhashsize = hashsize;
351 
352 		hashsize = maketcp_hashsize(hashsize);
353 		/* prevent absurdly low value */
354 		if (hashsize < 16)
355 			hashsize = 16;
356 		printf("%s: WARNING: TCB hash size not a power of 2, "
357 		    "clipped from %d to %d.\n", __func__, oldhashsize,
358 		    hashsize);
359 	}
360 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
361 	    "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
362 	    IPI_HASHFIELDS_4TUPLE);
363 
364 	/*
365 	 * These have to be type stable for the benefit of the timers.
366 	 */
367 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
368 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
369 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
370 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
371 
372 	tcp_tw_init();
373 	syncache_init();
374 	tcp_hc_init();
375 	tcp_reass_init();
376 
377 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
378 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
379 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
380 
381 	/* Skip initialization of globals for non-default instances. */
382 	if (!IS_DEFAULT_VNET(curvnet))
383 		return;
384 
385 	/* XXX virtualize those bellow? */
386 	tcp_delacktime = TCPTV_DELACK;
387 	tcp_keepinit = TCPTV_KEEP_INIT;
388 	tcp_keepidle = TCPTV_KEEP_IDLE;
389 	tcp_keepintvl = TCPTV_KEEPINTVL;
390 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
391 	tcp_msl = TCPTV_MSL;
392 	tcp_rexmit_min = TCPTV_MIN;
393 	if (tcp_rexmit_min < 1)
394 		tcp_rexmit_min = 1;
395 	tcp_rexmit_slop = TCPTV_CPU_VAR;
396 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
397 	tcp_tcbhashsize = hashsize;
398 
399 	TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
400 	if (tcp_soreceive_stream) {
401 #ifdef INET
402 		tcp_usrreqs.pru_soreceive = soreceive_stream;
403 #endif
404 #ifdef INET6
405 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
406 #endif /* INET6 */
407 	}
408 
409 #ifdef INET6
410 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
411 #else /* INET6 */
412 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
413 #endif /* INET6 */
414 	if (max_protohdr < TCP_MINPROTOHDR)
415 		max_protohdr = TCP_MINPROTOHDR;
416 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
417 		panic("tcp_init");
418 #undef TCP_MINPROTOHDR
419 
420 	ISN_LOCK_INIT();
421 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
422 		SHUTDOWN_PRI_DEFAULT);
423 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
424 		EVENTHANDLER_PRI_ANY);
425 }
426 
427 #ifdef VIMAGE
428 void
429 tcp_destroy(void)
430 {
431 
432 	tcp_reass_destroy();
433 	tcp_hc_destroy();
434 	syncache_destroy();
435 	tcp_tw_destroy();
436 	in_pcbinfo_destroy(&V_tcbinfo);
437 	uma_zdestroy(V_sack_hole_zone);
438 	uma_zdestroy(V_tcpcb_zone);
439 }
440 #endif
441 
442 void
443 tcp_fini(void *xtp)
444 {
445 
446 }
447 
448 /*
449  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
450  * tcp_template used to store this data in mbufs, but we now recopy it out
451  * of the tcpcb each time to conserve mbufs.
452  */
453 void
454 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
455 {
456 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
457 
458 	INP_WLOCK_ASSERT(inp);
459 
460 #ifdef INET6
461 	if ((inp->inp_vflag & INP_IPV6) != 0) {
462 		struct ip6_hdr *ip6;
463 
464 		ip6 = (struct ip6_hdr *)ip_ptr;
465 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
466 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
467 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
468 			(IPV6_VERSION & IPV6_VERSION_MASK);
469 		ip6->ip6_nxt = IPPROTO_TCP;
470 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
471 		ip6->ip6_src = inp->in6p_laddr;
472 		ip6->ip6_dst = inp->in6p_faddr;
473 	}
474 #endif /* INET6 */
475 #if defined(INET6) && defined(INET)
476 	else
477 #endif
478 #ifdef INET
479 	{
480 		struct ip *ip;
481 
482 		ip = (struct ip *)ip_ptr;
483 		ip->ip_v = IPVERSION;
484 		ip->ip_hl = 5;
485 		ip->ip_tos = inp->inp_ip_tos;
486 		ip->ip_len = 0;
487 		ip->ip_id = 0;
488 		ip->ip_off = 0;
489 		ip->ip_ttl = inp->inp_ip_ttl;
490 		ip->ip_sum = 0;
491 		ip->ip_p = IPPROTO_TCP;
492 		ip->ip_src = inp->inp_laddr;
493 		ip->ip_dst = inp->inp_faddr;
494 	}
495 #endif /* INET */
496 	th->th_sport = inp->inp_lport;
497 	th->th_dport = inp->inp_fport;
498 	th->th_seq = 0;
499 	th->th_ack = 0;
500 	th->th_x2 = 0;
501 	th->th_off = 5;
502 	th->th_flags = 0;
503 	th->th_win = 0;
504 	th->th_urp = 0;
505 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
506 }
507 
508 /*
509  * Create template to be used to send tcp packets on a connection.
510  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
511  * use for this function is in keepalives, which use tcp_respond.
512  */
513 struct tcptemp *
514 tcpip_maketemplate(struct inpcb *inp)
515 {
516 	struct tcptemp *t;
517 
518 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
519 	if (t == NULL)
520 		return (NULL);
521 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
522 	return (t);
523 }
524 
525 /*
526  * Send a single message to the TCP at address specified by
527  * the given TCP/IP header.  If m == NULL, then we make a copy
528  * of the tcpiphdr at ti and send directly to the addressed host.
529  * This is used to force keep alive messages out using the TCP
530  * template for a connection.  If flags are given then we send
531  * a message back to the TCP which originated the * segment ti,
532  * and discard the mbuf containing it and any other attached mbufs.
533  *
534  * In any case the ack and sequence number of the transmitted
535  * segment are as specified by the parameters.
536  *
537  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
538  */
539 void
540 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
541     tcp_seq ack, tcp_seq seq, int flags)
542 {
543 	int tlen;
544 	int win = 0;
545 	struct ip *ip;
546 	struct tcphdr *nth;
547 #ifdef INET6
548 	struct ip6_hdr *ip6;
549 	int isipv6;
550 #endif /* INET6 */
551 	int ipflags = 0;
552 	struct inpcb *inp;
553 
554 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
555 
556 #ifdef INET6
557 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
558 	ip6 = ipgen;
559 #endif /* INET6 */
560 	ip = ipgen;
561 
562 	if (tp != NULL) {
563 		inp = tp->t_inpcb;
564 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
565 		INP_WLOCK_ASSERT(inp);
566 	} else
567 		inp = NULL;
568 
569 	if (tp != NULL) {
570 		if (!(flags & TH_RST)) {
571 			win = sbspace(&inp->inp_socket->so_rcv);
572 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
573 				win = (long)TCP_MAXWIN << tp->rcv_scale;
574 		}
575 	}
576 	if (m == NULL) {
577 		m = m_gethdr(M_NOWAIT, MT_DATA);
578 		if (m == NULL)
579 			return;
580 		tlen = 0;
581 		m->m_data += max_linkhdr;
582 #ifdef INET6
583 		if (isipv6) {
584 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
585 			      sizeof(struct ip6_hdr));
586 			ip6 = mtod(m, struct ip6_hdr *);
587 			nth = (struct tcphdr *)(ip6 + 1);
588 		} else
589 #endif /* INET6 */
590 		{
591 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
592 			ip = mtod(m, struct ip *);
593 			nth = (struct tcphdr *)(ip + 1);
594 		}
595 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
596 		flags = TH_ACK;
597 	} else {
598 		/*
599 		 *  reuse the mbuf.
600 		 * XXX MRT We inherrit the FIB, which is lucky.
601 		 */
602 		m_freem(m->m_next);
603 		m->m_next = NULL;
604 		m->m_data = (caddr_t)ipgen;
605 		/* m_len is set later */
606 		tlen = 0;
607 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
608 #ifdef INET6
609 		if (isipv6) {
610 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
611 			nth = (struct tcphdr *)(ip6 + 1);
612 		} else
613 #endif /* INET6 */
614 		{
615 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
616 			nth = (struct tcphdr *)(ip + 1);
617 		}
618 		if (th != nth) {
619 			/*
620 			 * this is usually a case when an extension header
621 			 * exists between the IPv6 header and the
622 			 * TCP header.
623 			 */
624 			nth->th_sport = th->th_sport;
625 			nth->th_dport = th->th_dport;
626 		}
627 		xchg(nth->th_dport, nth->th_sport, uint16_t);
628 #undef xchg
629 	}
630 #ifdef INET6
631 	if (isipv6) {
632 		ip6->ip6_flow = 0;
633 		ip6->ip6_vfc = IPV6_VERSION;
634 		ip6->ip6_nxt = IPPROTO_TCP;
635 		ip6->ip6_plen = 0;		/* Set in ip6_output(). */
636 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
637 	}
638 #endif
639 #if defined(INET) && defined(INET6)
640 	else
641 #endif
642 #ifdef INET
643 	{
644 		tlen += sizeof (struct tcpiphdr);
645 		ip->ip_len = htons(tlen);
646 		ip->ip_ttl = V_ip_defttl;
647 		if (V_path_mtu_discovery)
648 			ip->ip_off |= htons(IP_DF);
649 	}
650 #endif
651 	m->m_len = tlen;
652 	m->m_pkthdr.len = tlen;
653 	m->m_pkthdr.rcvif = NULL;
654 #ifdef MAC
655 	if (inp != NULL) {
656 		/*
657 		 * Packet is associated with a socket, so allow the
658 		 * label of the response to reflect the socket label.
659 		 */
660 		INP_WLOCK_ASSERT(inp);
661 		mac_inpcb_create_mbuf(inp, m);
662 	} else {
663 		/*
664 		 * Packet is not associated with a socket, so possibly
665 		 * update the label in place.
666 		 */
667 		mac_netinet_tcp_reply(m);
668 	}
669 #endif
670 	nth->th_seq = htonl(seq);
671 	nth->th_ack = htonl(ack);
672 	nth->th_x2 = 0;
673 	nth->th_off = sizeof (struct tcphdr) >> 2;
674 	nth->th_flags = flags;
675 	if (tp != NULL)
676 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
677 	else
678 		nth->th_win = htons((u_short)win);
679 	nth->th_urp = 0;
680 
681 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
682 #ifdef INET6
683 	if (isipv6) {
684 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
685 		nth->th_sum = in6_cksum_pseudo(ip6,
686 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
687 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
688 		    NULL, NULL);
689 	}
690 #endif /* INET6 */
691 #if defined(INET6) && defined(INET)
692 	else
693 #endif
694 #ifdef INET
695 	{
696 		m->m_pkthdr.csum_flags = CSUM_TCP;
697 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
698 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
699 	}
700 #endif /* INET */
701 #ifdef TCPDEBUG
702 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
703 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
704 #endif
705 #ifdef INET6
706 	if (isipv6)
707 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
708 #endif /* INET6 */
709 #if defined(INET) && defined(INET6)
710 	else
711 #endif
712 #ifdef INET
713 		(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
714 #endif
715 }
716 
717 /*
718  * Create a new TCP control block, making an
719  * empty reassembly queue and hooking it to the argument
720  * protocol control block.  The `inp' parameter must have
721  * come from the zone allocator set up in tcp_init().
722  */
723 struct tcpcb *
724 tcp_newtcpcb(struct inpcb *inp)
725 {
726 	struct tcpcb_mem *tm;
727 	struct tcpcb *tp;
728 #ifdef INET6
729 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
730 #endif /* INET6 */
731 
732 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
733 	if (tm == NULL)
734 		return (NULL);
735 	tp = &tm->tcb;
736 
737 	/* Initialise cc_var struct for this tcpcb. */
738 	tp->ccv = &tm->ccv;
739 	tp->ccv->type = IPPROTO_TCP;
740 	tp->ccv->ccvc.tcp = tp;
741 
742 	/*
743 	 * Use the current system default CC algorithm.
744 	 */
745 	CC_LIST_RLOCK();
746 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
747 	CC_ALGO(tp) = CC_DEFAULT();
748 	CC_LIST_RUNLOCK();
749 
750 	if (CC_ALGO(tp)->cb_init != NULL)
751 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
752 			uma_zfree(V_tcpcb_zone, tm);
753 			return (NULL);
754 		}
755 
756 	tp->osd = &tm->osd;
757 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
758 		uma_zfree(V_tcpcb_zone, tm);
759 		return (NULL);
760 	}
761 
762 #ifdef VIMAGE
763 	tp->t_vnet = inp->inp_vnet;
764 #endif
765 	tp->t_timers = &tm->tt;
766 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
767 	tp->t_maxseg = tp->t_maxopd =
768 #ifdef INET6
769 		isipv6 ? V_tcp_v6mssdflt :
770 #endif /* INET6 */
771 		V_tcp_mssdflt;
772 
773 	/* Set up our timeouts. */
774 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
775 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
776 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
777 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
778 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
779 
780 	if (V_tcp_do_rfc1323)
781 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
782 	if (V_tcp_do_sack)
783 		tp->t_flags |= TF_SACK_PERMIT;
784 	TAILQ_INIT(&tp->snd_holes);
785 	tp->t_inpcb = inp;	/* XXX */
786 	/*
787 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
788 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
789 	 * reasonable initial retransmit time.
790 	 */
791 	tp->t_srtt = TCPTV_SRTTBASE;
792 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
793 	tp->t_rttmin = tcp_rexmit_min;
794 	tp->t_rxtcur = TCPTV_RTOBASE;
795 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
796 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
797 	tp->t_rcvtime = ticks;
798 	/*
799 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
800 	 * because the socket may be bound to an IPv6 wildcard address,
801 	 * which may match an IPv4-mapped IPv6 address.
802 	 */
803 	inp->inp_ip_ttl = V_ip_defttl;
804 	inp->inp_ppcb = tp;
805 	return (tp);		/* XXX */
806 }
807 
808 /*
809  * Switch the congestion control algorithm back to NewReno for any active
810  * control blocks using an algorithm which is about to go away.
811  * This ensures the CC framework can allow the unload to proceed without leaving
812  * any dangling pointers which would trigger a panic.
813  * Returning non-zero would inform the CC framework that something went wrong
814  * and it would be unsafe to allow the unload to proceed. However, there is no
815  * way for this to occur with this implementation so we always return zero.
816  */
817 int
818 tcp_ccalgounload(struct cc_algo *unload_algo)
819 {
820 	struct cc_algo *tmpalgo;
821 	struct inpcb *inp;
822 	struct tcpcb *tp;
823 	VNET_ITERATOR_DECL(vnet_iter);
824 
825 	/*
826 	 * Check all active control blocks across all network stacks and change
827 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
828 	 * requires cleanup code to be run, call it.
829 	 */
830 	VNET_LIST_RLOCK();
831 	VNET_FOREACH(vnet_iter) {
832 		CURVNET_SET(vnet_iter);
833 		INP_INFO_RLOCK(&V_tcbinfo);
834 		/*
835 		 * New connections already part way through being initialised
836 		 * with the CC algo we're removing will not race with this code
837 		 * because the INP_INFO_WLOCK is held during initialisation. We
838 		 * therefore don't enter the loop below until the connection
839 		 * list has stabilised.
840 		 */
841 		LIST_FOREACH(inp, &V_tcb, inp_list) {
842 			INP_WLOCK(inp);
843 			/* Important to skip tcptw structs. */
844 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
845 			    (tp = intotcpcb(inp)) != NULL) {
846 				/*
847 				 * By holding INP_WLOCK here, we are assured
848 				 * that the connection is not currently
849 				 * executing inside the CC module's functions
850 				 * i.e. it is safe to make the switch back to
851 				 * NewReno.
852 				 */
853 				if (CC_ALGO(tp) == unload_algo) {
854 					tmpalgo = CC_ALGO(tp);
855 					/* NewReno does not require any init. */
856 					CC_ALGO(tp) = &newreno_cc_algo;
857 					if (tmpalgo->cb_destroy != NULL)
858 						tmpalgo->cb_destroy(tp->ccv);
859 				}
860 			}
861 			INP_WUNLOCK(inp);
862 		}
863 		INP_INFO_RUNLOCK(&V_tcbinfo);
864 		CURVNET_RESTORE();
865 	}
866 	VNET_LIST_RUNLOCK();
867 
868 	return (0);
869 }
870 
871 /*
872  * Drop a TCP connection, reporting
873  * the specified error.  If connection is synchronized,
874  * then send a RST to peer.
875  */
876 struct tcpcb *
877 tcp_drop(struct tcpcb *tp, int errno)
878 {
879 	struct socket *so = tp->t_inpcb->inp_socket;
880 
881 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
882 	INP_WLOCK_ASSERT(tp->t_inpcb);
883 
884 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
885 		tp->t_state = TCPS_CLOSED;
886 		(void) tcp_output(tp);
887 		TCPSTAT_INC(tcps_drops);
888 	} else
889 		TCPSTAT_INC(tcps_conndrops);
890 	if (errno == ETIMEDOUT && tp->t_softerror)
891 		errno = tp->t_softerror;
892 	so->so_error = errno;
893 	return (tcp_close(tp));
894 }
895 
896 void
897 tcp_discardcb(struct tcpcb *tp)
898 {
899 	struct inpcb *inp = tp->t_inpcb;
900 	struct socket *so = inp->inp_socket;
901 #ifdef INET6
902 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
903 #endif /* INET6 */
904 
905 	INP_WLOCK_ASSERT(inp);
906 
907 	/*
908 	 * Make sure that all of our timers are stopped before we delete the
909 	 * PCB.
910 	 *
911 	 * XXXRW: Really, we would like to use callout_drain() here in order
912 	 * to avoid races experienced in tcp_timer.c where a timer is already
913 	 * executing at this point.  However, we can't, both because we're
914 	 * running in a context where we can't sleep, and also because we
915 	 * hold locks required by the timers.  What we instead need to do is
916 	 * test to see if callout_drain() is required, and if so, defer some
917 	 * portion of the remainder of tcp_discardcb() to an asynchronous
918 	 * context that can callout_drain() and then continue.  Some care
919 	 * will be required to ensure that no further processing takes place
920 	 * on the tcpcb, even though it hasn't been freed (a flag?).
921 	 */
922 	callout_stop(&tp->t_timers->tt_rexmt);
923 	callout_stop(&tp->t_timers->tt_persist);
924 	callout_stop(&tp->t_timers->tt_keep);
925 	callout_stop(&tp->t_timers->tt_2msl);
926 	callout_stop(&tp->t_timers->tt_delack);
927 
928 	/*
929 	 * If we got enough samples through the srtt filter,
930 	 * save the rtt and rttvar in the routing entry.
931 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
932 	 * 4 samples is enough for the srtt filter to converge
933 	 * to within enough % of the correct value; fewer samples
934 	 * and we could save a bogus rtt. The danger is not high
935 	 * as tcp quickly recovers from everything.
936 	 * XXX: Works very well but needs some more statistics!
937 	 */
938 	if (tp->t_rttupdated >= 4) {
939 		struct hc_metrics_lite metrics;
940 		u_long ssthresh;
941 
942 		bzero(&metrics, sizeof(metrics));
943 		/*
944 		 * Update the ssthresh always when the conditions below
945 		 * are satisfied. This gives us better new start value
946 		 * for the congestion avoidance for new connections.
947 		 * ssthresh is only set if packet loss occured on a session.
948 		 *
949 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
950 		 * being torn down.  Ideally this code would not use 'so'.
951 		 */
952 		ssthresh = tp->snd_ssthresh;
953 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
954 			/*
955 			 * convert the limit from user data bytes to
956 			 * packets then to packet data bytes.
957 			 */
958 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
959 			if (ssthresh < 2)
960 				ssthresh = 2;
961 			ssthresh *= (u_long)(tp->t_maxseg +
962 #ifdef INET6
963 			    (isipv6 ? sizeof (struct ip6_hdr) +
964 				sizeof (struct tcphdr) :
965 #endif
966 				sizeof (struct tcpiphdr)
967 #ifdef INET6
968 			    )
969 #endif
970 			    );
971 		} else
972 			ssthresh = 0;
973 		metrics.rmx_ssthresh = ssthresh;
974 
975 		metrics.rmx_rtt = tp->t_srtt;
976 		metrics.rmx_rttvar = tp->t_rttvar;
977 		metrics.rmx_cwnd = tp->snd_cwnd;
978 		metrics.rmx_sendpipe = 0;
979 		metrics.rmx_recvpipe = 0;
980 
981 		tcp_hc_update(&inp->inp_inc, &metrics);
982 	}
983 
984 	/* free the reassembly queue, if any */
985 	tcp_reass_flush(tp);
986 
987 #ifdef TCP_OFFLOAD
988 	/* Disconnect offload device, if any. */
989 	if (tp->t_flags & TF_TOE)
990 		tcp_offload_detach(tp);
991 #endif
992 
993 	tcp_free_sackholes(tp);
994 
995 	/* Allow the CC algorithm to clean up after itself. */
996 	if (CC_ALGO(tp)->cb_destroy != NULL)
997 		CC_ALGO(tp)->cb_destroy(tp->ccv);
998 
999 	khelp_destroy_osd(tp->osd);
1000 
1001 	CC_ALGO(tp) = NULL;
1002 	inp->inp_ppcb = NULL;
1003 	tp->t_inpcb = NULL;
1004 	uma_zfree(V_tcpcb_zone, tp);
1005 }
1006 
1007 /*
1008  * Attempt to close a TCP control block, marking it as dropped, and freeing
1009  * the socket if we hold the only reference.
1010  */
1011 struct tcpcb *
1012 tcp_close(struct tcpcb *tp)
1013 {
1014 	struct inpcb *inp = tp->t_inpcb;
1015 	struct socket *so;
1016 
1017 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1018 	INP_WLOCK_ASSERT(inp);
1019 
1020 #ifdef TCP_OFFLOAD
1021 	if (tp->t_state == TCPS_LISTEN)
1022 		tcp_offload_listen_stop(tp);
1023 #endif
1024 	in_pcbdrop(inp);
1025 	TCPSTAT_INC(tcps_closed);
1026 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1027 	so = inp->inp_socket;
1028 	soisdisconnected(so);
1029 	if (inp->inp_flags & INP_SOCKREF) {
1030 		KASSERT(so->so_state & SS_PROTOREF,
1031 		    ("tcp_close: !SS_PROTOREF"));
1032 		inp->inp_flags &= ~INP_SOCKREF;
1033 		INP_WUNLOCK(inp);
1034 		ACCEPT_LOCK();
1035 		SOCK_LOCK(so);
1036 		so->so_state &= ~SS_PROTOREF;
1037 		sofree(so);
1038 		return (NULL);
1039 	}
1040 	return (tp);
1041 }
1042 
1043 void
1044 tcp_drain(void)
1045 {
1046 	VNET_ITERATOR_DECL(vnet_iter);
1047 
1048 	if (!do_tcpdrain)
1049 		return;
1050 
1051 	VNET_LIST_RLOCK_NOSLEEP();
1052 	VNET_FOREACH(vnet_iter) {
1053 		CURVNET_SET(vnet_iter);
1054 		struct inpcb *inpb;
1055 		struct tcpcb *tcpb;
1056 
1057 	/*
1058 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1059 	 * if there is one...
1060 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1061 	 *      reassembly queue should be flushed, but in a situation
1062 	 *	where we're really low on mbufs, this is potentially
1063 	 *	usefull.
1064 	 */
1065 		INP_INFO_RLOCK(&V_tcbinfo);
1066 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1067 			if (inpb->inp_flags & INP_TIMEWAIT)
1068 				continue;
1069 			INP_WLOCK(inpb);
1070 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1071 				tcp_reass_flush(tcpb);
1072 				tcp_clean_sackreport(tcpb);
1073 			}
1074 			INP_WUNLOCK(inpb);
1075 		}
1076 		INP_INFO_RUNLOCK(&V_tcbinfo);
1077 		CURVNET_RESTORE();
1078 	}
1079 	VNET_LIST_RUNLOCK_NOSLEEP();
1080 }
1081 
1082 /*
1083  * Notify a tcp user of an asynchronous error;
1084  * store error as soft error, but wake up user
1085  * (for now, won't do anything until can select for soft error).
1086  *
1087  * Do not wake up user since there currently is no mechanism for
1088  * reporting soft errors (yet - a kqueue filter may be added).
1089  */
1090 static struct inpcb *
1091 tcp_notify(struct inpcb *inp, int error)
1092 {
1093 	struct tcpcb *tp;
1094 
1095 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1096 	INP_WLOCK_ASSERT(inp);
1097 
1098 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1099 	    (inp->inp_flags & INP_DROPPED))
1100 		return (inp);
1101 
1102 	tp = intotcpcb(inp);
1103 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1104 
1105 	/*
1106 	 * Ignore some errors if we are hooked up.
1107 	 * If connection hasn't completed, has retransmitted several times,
1108 	 * and receives a second error, give up now.  This is better
1109 	 * than waiting a long time to establish a connection that
1110 	 * can never complete.
1111 	 */
1112 	if (tp->t_state == TCPS_ESTABLISHED &&
1113 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1114 	     error == EHOSTDOWN)) {
1115 		return (inp);
1116 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1117 	    tp->t_softerror) {
1118 		tp = tcp_drop(tp, error);
1119 		if (tp != NULL)
1120 			return (inp);
1121 		else
1122 			return (NULL);
1123 	} else {
1124 		tp->t_softerror = error;
1125 		return (inp);
1126 	}
1127 #if 0
1128 	wakeup( &so->so_timeo);
1129 	sorwakeup(so);
1130 	sowwakeup(so);
1131 #endif
1132 }
1133 
1134 static int
1135 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1136 {
1137 	int error, i, m, n, pcb_count;
1138 	struct inpcb *inp, **inp_list;
1139 	inp_gen_t gencnt;
1140 	struct xinpgen xig;
1141 
1142 	/*
1143 	 * The process of preparing the TCB list is too time-consuming and
1144 	 * resource-intensive to repeat twice on every request.
1145 	 */
1146 	if (req->oldptr == NULL) {
1147 		n = V_tcbinfo.ipi_count + syncache_pcbcount();
1148 		n += imax(n / 8, 10);
1149 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1150 		return (0);
1151 	}
1152 
1153 	if (req->newptr != NULL)
1154 		return (EPERM);
1155 
1156 	/*
1157 	 * OK, now we're committed to doing something.
1158 	 */
1159 	INP_INFO_RLOCK(&V_tcbinfo);
1160 	gencnt = V_tcbinfo.ipi_gencnt;
1161 	n = V_tcbinfo.ipi_count;
1162 	INP_INFO_RUNLOCK(&V_tcbinfo);
1163 
1164 	m = syncache_pcbcount();
1165 
1166 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1167 		+ (n + m) * sizeof(struct xtcpcb));
1168 	if (error != 0)
1169 		return (error);
1170 
1171 	xig.xig_len = sizeof xig;
1172 	xig.xig_count = n + m;
1173 	xig.xig_gen = gencnt;
1174 	xig.xig_sogen = so_gencnt;
1175 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1176 	if (error)
1177 		return (error);
1178 
1179 	error = syncache_pcblist(req, m, &pcb_count);
1180 	if (error)
1181 		return (error);
1182 
1183 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1184 	if (inp_list == NULL)
1185 		return (ENOMEM);
1186 
1187 	INP_INFO_RLOCK(&V_tcbinfo);
1188 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1189 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1190 		INP_WLOCK(inp);
1191 		if (inp->inp_gencnt <= gencnt) {
1192 			/*
1193 			 * XXX: This use of cr_cansee(), introduced with
1194 			 * TCP state changes, is not quite right, but for
1195 			 * now, better than nothing.
1196 			 */
1197 			if (inp->inp_flags & INP_TIMEWAIT) {
1198 				if (intotw(inp) != NULL)
1199 					error = cr_cansee(req->td->td_ucred,
1200 					    intotw(inp)->tw_cred);
1201 				else
1202 					error = EINVAL;	/* Skip this inp. */
1203 			} else
1204 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1205 			if (error == 0) {
1206 				in_pcbref(inp);
1207 				inp_list[i++] = inp;
1208 			}
1209 		}
1210 		INP_WUNLOCK(inp);
1211 	}
1212 	INP_INFO_RUNLOCK(&V_tcbinfo);
1213 	n = i;
1214 
1215 	error = 0;
1216 	for (i = 0; i < n; i++) {
1217 		inp = inp_list[i];
1218 		INP_RLOCK(inp);
1219 		if (inp->inp_gencnt <= gencnt) {
1220 			struct xtcpcb xt;
1221 			void *inp_ppcb;
1222 
1223 			bzero(&xt, sizeof(xt));
1224 			xt.xt_len = sizeof xt;
1225 			/* XXX should avoid extra copy */
1226 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1227 			inp_ppcb = inp->inp_ppcb;
1228 			if (inp_ppcb == NULL)
1229 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1230 			else if (inp->inp_flags & INP_TIMEWAIT) {
1231 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1232 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1233 			} else {
1234 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1235 				if (xt.xt_tp.t_timers)
1236 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1237 			}
1238 			if (inp->inp_socket != NULL)
1239 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1240 			else {
1241 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1242 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1243 			}
1244 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1245 			INP_RUNLOCK(inp);
1246 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1247 		} else
1248 			INP_RUNLOCK(inp);
1249 	}
1250 	INP_INFO_WLOCK(&V_tcbinfo);
1251 	for (i = 0; i < n; i++) {
1252 		inp = inp_list[i];
1253 		INP_RLOCK(inp);
1254 		if (!in_pcbrele_rlocked(inp))
1255 			INP_RUNLOCK(inp);
1256 	}
1257 	INP_INFO_WUNLOCK(&V_tcbinfo);
1258 
1259 	if (!error) {
1260 		/*
1261 		 * Give the user an updated idea of our state.
1262 		 * If the generation differs from what we told
1263 		 * her before, she knows that something happened
1264 		 * while we were processing this request, and it
1265 		 * might be necessary to retry.
1266 		 */
1267 		INP_INFO_RLOCK(&V_tcbinfo);
1268 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1269 		xig.xig_sogen = so_gencnt;
1270 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1271 		INP_INFO_RUNLOCK(&V_tcbinfo);
1272 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1273 	}
1274 	free(inp_list, M_TEMP);
1275 	return (error);
1276 }
1277 
1278 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1279     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1280     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1281 
1282 #ifdef INET
1283 static int
1284 tcp_getcred(SYSCTL_HANDLER_ARGS)
1285 {
1286 	struct xucred xuc;
1287 	struct sockaddr_in addrs[2];
1288 	struct inpcb *inp;
1289 	int error;
1290 
1291 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1292 	if (error)
1293 		return (error);
1294 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1295 	if (error)
1296 		return (error);
1297 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1298 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1299 	if (inp != NULL) {
1300 		if (inp->inp_socket == NULL)
1301 			error = ENOENT;
1302 		if (error == 0)
1303 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1304 		if (error == 0)
1305 			cru2x(inp->inp_cred, &xuc);
1306 		INP_RUNLOCK(inp);
1307 	} else
1308 		error = ENOENT;
1309 	if (error == 0)
1310 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1311 	return (error);
1312 }
1313 
1314 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1315     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1316     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1317 #endif /* INET */
1318 
1319 #ifdef INET6
1320 static int
1321 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1322 {
1323 	struct xucred xuc;
1324 	struct sockaddr_in6 addrs[2];
1325 	struct inpcb *inp;
1326 	int error;
1327 #ifdef INET
1328 	int mapped = 0;
1329 #endif
1330 
1331 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1332 	if (error)
1333 		return (error);
1334 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1335 	if (error)
1336 		return (error);
1337 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1338 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1339 		return (error);
1340 	}
1341 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1342 #ifdef INET
1343 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1344 			mapped = 1;
1345 		else
1346 #endif
1347 			return (EINVAL);
1348 	}
1349 
1350 #ifdef INET
1351 	if (mapped == 1)
1352 		inp = in_pcblookup(&V_tcbinfo,
1353 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1354 			addrs[1].sin6_port,
1355 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1356 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1357 	else
1358 #endif
1359 		inp = in6_pcblookup(&V_tcbinfo,
1360 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1361 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1362 			INPLOOKUP_RLOCKPCB, NULL);
1363 	if (inp != NULL) {
1364 		if (inp->inp_socket == NULL)
1365 			error = ENOENT;
1366 		if (error == 0)
1367 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1368 		if (error == 0)
1369 			cru2x(inp->inp_cred, &xuc);
1370 		INP_RUNLOCK(inp);
1371 	} else
1372 		error = ENOENT;
1373 	if (error == 0)
1374 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1375 	return (error);
1376 }
1377 
1378 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1379     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1380     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1381 #endif /* INET6 */
1382 
1383 
1384 #ifdef INET
1385 void
1386 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1387 {
1388 	struct ip *ip = vip;
1389 	struct tcphdr *th;
1390 	struct in_addr faddr;
1391 	struct inpcb *inp;
1392 	struct tcpcb *tp;
1393 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1394 	struct icmp *icp;
1395 	struct in_conninfo inc;
1396 	tcp_seq icmp_tcp_seq;
1397 	int mtu;
1398 
1399 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1400 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1401 		return;
1402 
1403 	if (cmd == PRC_MSGSIZE)
1404 		notify = tcp_mtudisc_notify;
1405 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1406 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1407 		notify = tcp_drop_syn_sent;
1408 	/*
1409 	 * Redirects don't need to be handled up here.
1410 	 */
1411 	else if (PRC_IS_REDIRECT(cmd))
1412 		return;
1413 	/*
1414 	 * Source quench is depreciated.
1415 	 */
1416 	else if (cmd == PRC_QUENCH)
1417 		return;
1418 	/*
1419 	 * Hostdead is ugly because it goes linearly through all PCBs.
1420 	 * XXX: We never get this from ICMP, otherwise it makes an
1421 	 * excellent DoS attack on machines with many connections.
1422 	 */
1423 	else if (cmd == PRC_HOSTDEAD)
1424 		ip = NULL;
1425 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1426 		return;
1427 	if (ip != NULL) {
1428 		icp = (struct icmp *)((caddr_t)ip
1429 				      - offsetof(struct icmp, icmp_ip));
1430 		th = (struct tcphdr *)((caddr_t)ip
1431 				       + (ip->ip_hl << 2));
1432 		INP_INFO_WLOCK(&V_tcbinfo);
1433 		inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1434 		    ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1435 		if (inp != NULL)  {
1436 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1437 			    !(inp->inp_flags & INP_DROPPED) &&
1438 			    !(inp->inp_socket == NULL)) {
1439 				icmp_tcp_seq = htonl(th->th_seq);
1440 				tp = intotcpcb(inp);
1441 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1442 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1443 					if (cmd == PRC_MSGSIZE) {
1444 					    /*
1445 					     * MTU discovery:
1446 					     * If we got a needfrag set the MTU
1447 					     * in the route to the suggested new
1448 					     * value (if given) and then notify.
1449 					     */
1450 					    bzero(&inc, sizeof(inc));
1451 					    inc.inc_faddr = faddr;
1452 					    inc.inc_fibnum =
1453 						inp->inp_inc.inc_fibnum;
1454 
1455 					    mtu = ntohs(icp->icmp_nextmtu);
1456 					    /*
1457 					     * If no alternative MTU was
1458 					     * proposed, try the next smaller
1459 					     * one.
1460 					     */
1461 					    if (!mtu)
1462 						mtu = ip_next_mtu(
1463 						 ntohs(ip->ip_len), 1);
1464 					    if (mtu < V_tcp_minmss
1465 						 + sizeof(struct tcpiphdr))
1466 						mtu = V_tcp_minmss
1467 						 + sizeof(struct tcpiphdr);
1468 					    /*
1469 					     * Only cache the MTU if it
1470 					     * is smaller than the interface
1471 					     * or route MTU.  tcp_mtudisc()
1472 					     * will do right thing by itself.
1473 					     */
1474 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1475 						tcp_hc_updatemtu(&inc, mtu);
1476 					    tcp_mtudisc(inp, mtu);
1477 					} else
1478 						inp = (*notify)(inp,
1479 						    inetctlerrmap[cmd]);
1480 				}
1481 			}
1482 			if (inp != NULL)
1483 				INP_WUNLOCK(inp);
1484 		} else {
1485 			bzero(&inc, sizeof(inc));
1486 			inc.inc_fport = th->th_dport;
1487 			inc.inc_lport = th->th_sport;
1488 			inc.inc_faddr = faddr;
1489 			inc.inc_laddr = ip->ip_src;
1490 			syncache_unreach(&inc, th);
1491 		}
1492 		INP_INFO_WUNLOCK(&V_tcbinfo);
1493 	} else
1494 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1495 }
1496 #endif /* INET */
1497 
1498 #ifdef INET6
1499 void
1500 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1501 {
1502 	struct tcphdr th;
1503 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1504 	struct ip6_hdr *ip6;
1505 	struct mbuf *m;
1506 	struct ip6ctlparam *ip6cp = NULL;
1507 	const struct sockaddr_in6 *sa6_src = NULL;
1508 	int off;
1509 	struct tcp_portonly {
1510 		u_int16_t th_sport;
1511 		u_int16_t th_dport;
1512 	} *thp;
1513 
1514 	if (sa->sa_family != AF_INET6 ||
1515 	    sa->sa_len != sizeof(struct sockaddr_in6))
1516 		return;
1517 
1518 	if (cmd == PRC_MSGSIZE)
1519 		notify = tcp_mtudisc_notify;
1520 	else if (!PRC_IS_REDIRECT(cmd) &&
1521 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1522 		return;
1523 	/* Source quench is depreciated. */
1524 	else if (cmd == PRC_QUENCH)
1525 		return;
1526 
1527 	/* if the parameter is from icmp6, decode it. */
1528 	if (d != NULL) {
1529 		ip6cp = (struct ip6ctlparam *)d;
1530 		m = ip6cp->ip6c_m;
1531 		ip6 = ip6cp->ip6c_ip6;
1532 		off = ip6cp->ip6c_off;
1533 		sa6_src = ip6cp->ip6c_src;
1534 	} else {
1535 		m = NULL;
1536 		ip6 = NULL;
1537 		off = 0;	/* fool gcc */
1538 		sa6_src = &sa6_any;
1539 	}
1540 
1541 	if (ip6 != NULL) {
1542 		struct in_conninfo inc;
1543 		/*
1544 		 * XXX: We assume that when IPV6 is non NULL,
1545 		 * M and OFF are valid.
1546 		 */
1547 
1548 		/* check if we can safely examine src and dst ports */
1549 		if (m->m_pkthdr.len < off + sizeof(*thp))
1550 			return;
1551 
1552 		bzero(&th, sizeof(th));
1553 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1554 
1555 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1556 		    (struct sockaddr *)ip6cp->ip6c_src,
1557 		    th.th_sport, cmd, NULL, notify);
1558 
1559 		bzero(&inc, sizeof(inc));
1560 		inc.inc_fport = th.th_dport;
1561 		inc.inc_lport = th.th_sport;
1562 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1563 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1564 		inc.inc_flags |= INC_ISIPV6;
1565 		INP_INFO_WLOCK(&V_tcbinfo);
1566 		syncache_unreach(&inc, &th);
1567 		INP_INFO_WUNLOCK(&V_tcbinfo);
1568 	} else
1569 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1570 			      0, cmd, NULL, notify);
1571 }
1572 #endif /* INET6 */
1573 
1574 
1575 /*
1576  * Following is where TCP initial sequence number generation occurs.
1577  *
1578  * There are two places where we must use initial sequence numbers:
1579  * 1.  In SYN-ACK packets.
1580  * 2.  In SYN packets.
1581  *
1582  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1583  * tcp_syncache.c for details.
1584  *
1585  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1586  * depends on this property.  In addition, these ISNs should be
1587  * unguessable so as to prevent connection hijacking.  To satisfy
1588  * the requirements of this situation, the algorithm outlined in
1589  * RFC 1948 is used, with only small modifications.
1590  *
1591  * Implementation details:
1592  *
1593  * Time is based off the system timer, and is corrected so that it
1594  * increases by one megabyte per second.  This allows for proper
1595  * recycling on high speed LANs while still leaving over an hour
1596  * before rollover.
1597  *
1598  * As reading the *exact* system time is too expensive to be done
1599  * whenever setting up a TCP connection, we increment the time
1600  * offset in two ways.  First, a small random positive increment
1601  * is added to isn_offset for each connection that is set up.
1602  * Second, the function tcp_isn_tick fires once per clock tick
1603  * and increments isn_offset as necessary so that sequence numbers
1604  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1605  * random positive increments serve only to ensure that the same
1606  * exact sequence number is never sent out twice (as could otherwise
1607  * happen when a port is recycled in less than the system tick
1608  * interval.)
1609  *
1610  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1611  * between seeding of isn_secret.  This is normally set to zero,
1612  * as reseeding should not be necessary.
1613  *
1614  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1615  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1616  * general, this means holding an exclusive (write) lock.
1617  */
1618 
1619 #define ISN_BYTES_PER_SECOND 1048576
1620 #define ISN_STATIC_INCREMENT 4096
1621 #define ISN_RANDOM_INCREMENT (4096 - 1)
1622 
1623 static VNET_DEFINE(u_char, isn_secret[32]);
1624 static VNET_DEFINE(int, isn_last);
1625 static VNET_DEFINE(int, isn_last_reseed);
1626 static VNET_DEFINE(u_int32_t, isn_offset);
1627 static VNET_DEFINE(u_int32_t, isn_offset_old);
1628 
1629 #define	V_isn_secret			VNET(isn_secret)
1630 #define	V_isn_last			VNET(isn_last)
1631 #define	V_isn_last_reseed		VNET(isn_last_reseed)
1632 #define	V_isn_offset			VNET(isn_offset)
1633 #define	V_isn_offset_old		VNET(isn_offset_old)
1634 
1635 tcp_seq
1636 tcp_new_isn(struct tcpcb *tp)
1637 {
1638 	MD5_CTX isn_ctx;
1639 	u_int32_t md5_buffer[4];
1640 	tcp_seq new_isn;
1641 	u_int32_t projected_offset;
1642 
1643 	INP_WLOCK_ASSERT(tp->t_inpcb);
1644 
1645 	ISN_LOCK();
1646 	/* Seed if this is the first use, reseed if requested. */
1647 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1648 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1649 		< (u_int)ticks))) {
1650 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1651 		V_isn_last_reseed = ticks;
1652 	}
1653 
1654 	/* Compute the md5 hash and return the ISN. */
1655 	MD5Init(&isn_ctx);
1656 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1657 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1658 #ifdef INET6
1659 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1660 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1661 			  sizeof(struct in6_addr));
1662 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1663 			  sizeof(struct in6_addr));
1664 	} else
1665 #endif
1666 	{
1667 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1668 			  sizeof(struct in_addr));
1669 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1670 			  sizeof(struct in_addr));
1671 	}
1672 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1673 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1674 	new_isn = (tcp_seq) md5_buffer[0];
1675 	V_isn_offset += ISN_STATIC_INCREMENT +
1676 		(arc4random() & ISN_RANDOM_INCREMENT);
1677 	if (ticks != V_isn_last) {
1678 		projected_offset = V_isn_offset_old +
1679 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1680 		if (SEQ_GT(projected_offset, V_isn_offset))
1681 			V_isn_offset = projected_offset;
1682 		V_isn_offset_old = V_isn_offset;
1683 		V_isn_last = ticks;
1684 	}
1685 	new_isn += V_isn_offset;
1686 	ISN_UNLOCK();
1687 	return (new_isn);
1688 }
1689 
1690 /*
1691  * When a specific ICMP unreachable message is received and the
1692  * connection state is SYN-SENT, drop the connection.  This behavior
1693  * is controlled by the icmp_may_rst sysctl.
1694  */
1695 struct inpcb *
1696 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1697 {
1698 	struct tcpcb *tp;
1699 
1700 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1701 	INP_WLOCK_ASSERT(inp);
1702 
1703 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1704 	    (inp->inp_flags & INP_DROPPED))
1705 		return (inp);
1706 
1707 	tp = intotcpcb(inp);
1708 	if (tp->t_state != TCPS_SYN_SENT)
1709 		return (inp);
1710 
1711 	tp = tcp_drop(tp, errno);
1712 	if (tp != NULL)
1713 		return (inp);
1714 	else
1715 		return (NULL);
1716 }
1717 
1718 /*
1719  * When `need fragmentation' ICMP is received, update our idea of the MSS
1720  * based on the new value. Also nudge TCP to send something, since we
1721  * know the packet we just sent was dropped.
1722  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1723  */
1724 static struct inpcb *
1725 tcp_mtudisc_notify(struct inpcb *inp, int error)
1726 {
1727 
1728 	return (tcp_mtudisc(inp, -1));
1729 }
1730 
1731 struct inpcb *
1732 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1733 {
1734 	struct tcpcb *tp;
1735 	struct socket *so;
1736 
1737 	INP_WLOCK_ASSERT(inp);
1738 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1739 	    (inp->inp_flags & INP_DROPPED))
1740 		return (inp);
1741 
1742 	tp = intotcpcb(inp);
1743 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1744 
1745 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1746 
1747 	so = inp->inp_socket;
1748 	SOCKBUF_LOCK(&so->so_snd);
1749 	/* If the mss is larger than the socket buffer, decrease the mss. */
1750 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1751 		tp->t_maxseg = so->so_snd.sb_hiwat;
1752 	SOCKBUF_UNLOCK(&so->so_snd);
1753 
1754 	TCPSTAT_INC(tcps_mturesent);
1755 	tp->t_rtttime = 0;
1756 	tp->snd_nxt = tp->snd_una;
1757 	tcp_free_sackholes(tp);
1758 	tp->snd_recover = tp->snd_max;
1759 	if (tp->t_flags & TF_SACK_PERMIT)
1760 		EXIT_FASTRECOVERY(tp->t_flags);
1761 	tcp_output(tp);
1762 	return (inp);
1763 }
1764 
1765 #ifdef INET
1766 /*
1767  * Look-up the routing entry to the peer of this inpcb.  If no route
1768  * is found and it cannot be allocated, then return 0.  This routine
1769  * is called by TCP routines that access the rmx structure and by
1770  * tcp_mss_update to get the peer/interface MTU.
1771  */
1772 u_long
1773 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1774 {
1775 	struct route sro;
1776 	struct sockaddr_in *dst;
1777 	struct ifnet *ifp;
1778 	u_long maxmtu = 0;
1779 
1780 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1781 
1782 	bzero(&sro, sizeof(sro));
1783 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1784 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1785 		dst->sin_family = AF_INET;
1786 		dst->sin_len = sizeof(*dst);
1787 		dst->sin_addr = inc->inc_faddr;
1788 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1789 	}
1790 	if (sro.ro_rt != NULL) {
1791 		ifp = sro.ro_rt->rt_ifp;
1792 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1793 			maxmtu = ifp->if_mtu;
1794 		else
1795 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1796 
1797 		/* Report additional interface capabilities. */
1798 		if (flags != NULL) {
1799 			if (ifp->if_capenable & IFCAP_TSO4 &&
1800 			    ifp->if_hwassist & CSUM_TSO)
1801 				*flags |= CSUM_TSO;
1802 		}
1803 		RTFREE(sro.ro_rt);
1804 	}
1805 	return (maxmtu);
1806 }
1807 #endif /* INET */
1808 
1809 #ifdef INET6
1810 u_long
1811 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1812 {
1813 	struct route_in6 sro6;
1814 	struct ifnet *ifp;
1815 	u_long maxmtu = 0;
1816 
1817 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1818 
1819 	bzero(&sro6, sizeof(sro6));
1820 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1821 		sro6.ro_dst.sin6_family = AF_INET6;
1822 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1823 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1824 		in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1825 	}
1826 	if (sro6.ro_rt != NULL) {
1827 		ifp = sro6.ro_rt->rt_ifp;
1828 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1829 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1830 		else
1831 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1832 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1833 
1834 		/* Report additional interface capabilities. */
1835 		if (flags != NULL) {
1836 			if (ifp->if_capenable & IFCAP_TSO6 &&
1837 			    ifp->if_hwassist & CSUM_TSO)
1838 				*flags |= CSUM_TSO;
1839 		}
1840 		RTFREE(sro6.ro_rt);
1841 	}
1842 
1843 	return (maxmtu);
1844 }
1845 #endif /* INET6 */
1846 
1847 #ifdef IPSEC
1848 /* compute ESP/AH header size for TCP, including outer IP header. */
1849 size_t
1850 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1851 {
1852 	struct inpcb *inp;
1853 	struct mbuf *m;
1854 	size_t hdrsiz;
1855 	struct ip *ip;
1856 #ifdef INET6
1857 	struct ip6_hdr *ip6;
1858 #endif
1859 	struct tcphdr *th;
1860 
1861 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1862 		return (0);
1863 	m = m_gethdr(M_NOWAIT, MT_DATA);
1864 	if (!m)
1865 		return (0);
1866 
1867 #ifdef INET6
1868 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1869 		ip6 = mtod(m, struct ip6_hdr *);
1870 		th = (struct tcphdr *)(ip6 + 1);
1871 		m->m_pkthdr.len = m->m_len =
1872 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1873 		tcpip_fillheaders(inp, ip6, th);
1874 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1875 	} else
1876 #endif /* INET6 */
1877 	{
1878 		ip = mtod(m, struct ip *);
1879 		th = (struct tcphdr *)(ip + 1);
1880 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1881 		tcpip_fillheaders(inp, ip, th);
1882 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1883 	}
1884 
1885 	m_free(m);
1886 	return (hdrsiz);
1887 }
1888 #endif /* IPSEC */
1889 
1890 #ifdef TCP_SIGNATURE
1891 /*
1892  * Callback function invoked by m_apply() to digest TCP segment data
1893  * contained within an mbuf chain.
1894  */
1895 static int
1896 tcp_signature_apply(void *fstate, void *data, u_int len)
1897 {
1898 
1899 	MD5Update(fstate, (u_char *)data, len);
1900 	return (0);
1901 }
1902 
1903 /*
1904  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1905  *
1906  * Parameters:
1907  * m		pointer to head of mbuf chain
1908  * _unused
1909  * len		length of TCP segment data, excluding options
1910  * optlen	length of TCP segment options
1911  * buf		pointer to storage for computed MD5 digest
1912  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1913  *
1914  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1915  * When called from tcp_input(), we can be sure that th_sum has been
1916  * zeroed out and verified already.
1917  *
1918  * Return 0 if successful, otherwise return -1.
1919  *
1920  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1921  * search with the destination IP address, and a 'magic SPI' to be
1922  * determined by the application. This is hardcoded elsewhere to 1179
1923  * right now. Another branch of this code exists which uses the SPD to
1924  * specify per-application flows but it is unstable.
1925  */
1926 int
1927 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1928     u_char *buf, u_int direction)
1929 {
1930 	union sockaddr_union dst;
1931 #ifdef INET
1932 	struct ippseudo ippseudo;
1933 #endif
1934 	MD5_CTX ctx;
1935 	int doff;
1936 	struct ip *ip;
1937 #ifdef INET
1938 	struct ipovly *ipovly;
1939 #endif
1940 	struct secasvar *sav;
1941 	struct tcphdr *th;
1942 #ifdef INET6
1943 	struct ip6_hdr *ip6;
1944 	struct in6_addr in6;
1945 	char ip6buf[INET6_ADDRSTRLEN];
1946 	uint32_t plen;
1947 	uint16_t nhdr;
1948 #endif
1949 	u_short savecsum;
1950 
1951 	KASSERT(m != NULL, ("NULL mbuf chain"));
1952 	KASSERT(buf != NULL, ("NULL signature pointer"));
1953 
1954 	/* Extract the destination from the IP header in the mbuf. */
1955 	bzero(&dst, sizeof(union sockaddr_union));
1956 	ip = mtod(m, struct ip *);
1957 #ifdef INET6
1958 	ip6 = NULL;	/* Make the compiler happy. */
1959 #endif
1960 	switch (ip->ip_v) {
1961 #ifdef INET
1962 	case IPVERSION:
1963 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1964 		dst.sa.sa_family = AF_INET;
1965 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1966 		    ip->ip_src : ip->ip_dst;
1967 		break;
1968 #endif
1969 #ifdef INET6
1970 	case (IPV6_VERSION >> 4):
1971 		ip6 = mtod(m, struct ip6_hdr *);
1972 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1973 		dst.sa.sa_family = AF_INET6;
1974 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1975 		    ip6->ip6_src : ip6->ip6_dst;
1976 		break;
1977 #endif
1978 	default:
1979 		return (EINVAL);
1980 		/* NOTREACHED */
1981 		break;
1982 	}
1983 
1984 	/* Look up an SADB entry which matches the address of the peer. */
1985 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1986 	if (sav == NULL) {
1987 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1988 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1989 #ifdef INET6
1990 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
1991 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1992 #endif
1993 			"(unsupported)"));
1994 		return (EINVAL);
1995 	}
1996 
1997 	MD5Init(&ctx);
1998 	/*
1999 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2000 	 *
2001 	 * XXX The ippseudo header MUST be digested in network byte order,
2002 	 * or else we'll fail the regression test. Assume all fields we've
2003 	 * been doing arithmetic on have been in host byte order.
2004 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2005 	 * tcp_output(), the underlying ip_len member has not yet been set.
2006 	 */
2007 	switch (ip->ip_v) {
2008 #ifdef INET
2009 	case IPVERSION:
2010 		ipovly = (struct ipovly *)ip;
2011 		ippseudo.ippseudo_src = ipovly->ih_src;
2012 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2013 		ippseudo.ippseudo_pad = 0;
2014 		ippseudo.ippseudo_p = IPPROTO_TCP;
2015 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2016 		    optlen);
2017 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2018 
2019 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2020 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2021 		break;
2022 #endif
2023 #ifdef INET6
2024 	/*
2025 	 * RFC 2385, 2.0  Proposal
2026 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2027 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2028 	 * extended next header value (to form 32 bits), and 32-bit segment
2029 	 * length.
2030 	 * Note: Upper-Layer Packet Length comes before Next Header.
2031 	 */
2032 	case (IPV6_VERSION >> 4):
2033 		in6 = ip6->ip6_src;
2034 		in6_clearscope(&in6);
2035 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2036 		in6 = ip6->ip6_dst;
2037 		in6_clearscope(&in6);
2038 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2039 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2040 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2041 		nhdr = 0;
2042 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2043 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2044 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2045 		nhdr = IPPROTO_TCP;
2046 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2047 
2048 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2049 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2050 		break;
2051 #endif
2052 	default:
2053 		return (EINVAL);
2054 		/* NOTREACHED */
2055 		break;
2056 	}
2057 
2058 
2059 	/*
2060 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2061 	 * The TCP checksum must be set to zero.
2062 	 */
2063 	savecsum = th->th_sum;
2064 	th->th_sum = 0;
2065 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2066 	th->th_sum = savecsum;
2067 
2068 	/*
2069 	 * Step 3: Update MD5 hash with TCP segment data.
2070 	 *         Use m_apply() to avoid an early m_pullup().
2071 	 */
2072 	if (len > 0)
2073 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2074 
2075 	/*
2076 	 * Step 4: Update MD5 hash with shared secret.
2077 	 */
2078 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2079 	MD5Final(buf, &ctx);
2080 
2081 	key_sa_recordxfer(sav, m);
2082 	KEY_FREESAV(&sav);
2083 	return (0);
2084 }
2085 
2086 /*
2087  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2088  *
2089  * Parameters:
2090  * m		pointer to head of mbuf chain
2091  * len		length of TCP segment data, excluding options
2092  * optlen	length of TCP segment options
2093  * buf		pointer to storage for computed MD5 digest
2094  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2095  *
2096  * Return 1 if successful, otherwise return 0.
2097  */
2098 int
2099 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2100     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2101 {
2102 	char tmpdigest[TCP_SIGLEN];
2103 
2104 	if (tcp_sig_checksigs == 0)
2105 		return (1);
2106 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2107 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2108 
2109 			/*
2110 			 * If this socket is not expecting signature but
2111 			 * the segment contains signature just fail.
2112 			 */
2113 			TCPSTAT_INC(tcps_sig_err_sigopt);
2114 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2115 			return (0);
2116 		}
2117 
2118 		/* Signature is not expected, and not present in segment. */
2119 		return (1);
2120 	}
2121 
2122 	/*
2123 	 * If this socket is expecting signature but the segment does not
2124 	 * contain any just fail.
2125 	 */
2126 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2127 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2128 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2129 		return (0);
2130 	}
2131 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2132 	    IPSEC_DIR_INBOUND) == -1) {
2133 		TCPSTAT_INC(tcps_sig_err_buildsig);
2134 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2135 		return (0);
2136 	}
2137 
2138 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2139 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2140 		return (0);
2141 	}
2142 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2143 	return (1);
2144 }
2145 #endif /* TCP_SIGNATURE */
2146 
2147 static int
2148 sysctl_drop(SYSCTL_HANDLER_ARGS)
2149 {
2150 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2151 	struct sockaddr_storage addrs[2];
2152 	struct inpcb *inp;
2153 	struct tcpcb *tp;
2154 	struct tcptw *tw;
2155 	struct sockaddr_in *fin, *lin;
2156 #ifdef INET6
2157 	struct sockaddr_in6 *fin6, *lin6;
2158 #endif
2159 	int error;
2160 
2161 	inp = NULL;
2162 	fin = lin = NULL;
2163 #ifdef INET6
2164 	fin6 = lin6 = NULL;
2165 #endif
2166 	error = 0;
2167 
2168 	if (req->oldptr != NULL || req->oldlen != 0)
2169 		return (EINVAL);
2170 	if (req->newptr == NULL)
2171 		return (EPERM);
2172 	if (req->newlen < sizeof(addrs))
2173 		return (ENOMEM);
2174 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2175 	if (error)
2176 		return (error);
2177 
2178 	switch (addrs[0].ss_family) {
2179 #ifdef INET6
2180 	case AF_INET6:
2181 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2182 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2183 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2184 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2185 			return (EINVAL);
2186 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2187 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2188 				return (EINVAL);
2189 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2190 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2191 			fin = (struct sockaddr_in *)&addrs[0];
2192 			lin = (struct sockaddr_in *)&addrs[1];
2193 			break;
2194 		}
2195 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2196 		if (error)
2197 			return (error);
2198 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2199 		if (error)
2200 			return (error);
2201 		break;
2202 #endif
2203 #ifdef INET
2204 	case AF_INET:
2205 		fin = (struct sockaddr_in *)&addrs[0];
2206 		lin = (struct sockaddr_in *)&addrs[1];
2207 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2208 		    lin->sin_len != sizeof(struct sockaddr_in))
2209 			return (EINVAL);
2210 		break;
2211 #endif
2212 	default:
2213 		return (EINVAL);
2214 	}
2215 	INP_INFO_WLOCK(&V_tcbinfo);
2216 	switch (addrs[0].ss_family) {
2217 #ifdef INET6
2218 	case AF_INET6:
2219 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2220 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2221 		    INPLOOKUP_WLOCKPCB, NULL);
2222 		break;
2223 #endif
2224 #ifdef INET
2225 	case AF_INET:
2226 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2227 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2228 		break;
2229 #endif
2230 	}
2231 	if (inp != NULL) {
2232 		if (inp->inp_flags & INP_TIMEWAIT) {
2233 			/*
2234 			 * XXXRW: There currently exists a state where an
2235 			 * inpcb is present, but its timewait state has been
2236 			 * discarded.  For now, don't allow dropping of this
2237 			 * type of inpcb.
2238 			 */
2239 			tw = intotw(inp);
2240 			if (tw != NULL)
2241 				tcp_twclose(tw, 0);
2242 			else
2243 				INP_WUNLOCK(inp);
2244 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2245 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2246 			tp = intotcpcb(inp);
2247 			tp = tcp_drop(tp, ECONNABORTED);
2248 			if (tp != NULL)
2249 				INP_WUNLOCK(inp);
2250 		} else
2251 			INP_WUNLOCK(inp);
2252 	} else
2253 		error = ESRCH;
2254 	INP_INFO_WUNLOCK(&V_tcbinfo);
2255 	return (error);
2256 }
2257 
2258 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2259     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2260     0, sysctl_drop, "", "Drop TCP connection");
2261 
2262 /*
2263  * Generate a standardized TCP log line for use throughout the
2264  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2265  * allow use in the interrupt context.
2266  *
2267  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2268  * NB: The function may return NULL if memory allocation failed.
2269  *
2270  * Due to header inclusion and ordering limitations the struct ip
2271  * and ip6_hdr pointers have to be passed as void pointers.
2272  */
2273 char *
2274 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2275     const void *ip6hdr)
2276 {
2277 
2278 	/* Is logging enabled? */
2279 	if (tcp_log_in_vain == 0)
2280 		return (NULL);
2281 
2282 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2283 }
2284 
2285 char *
2286 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2287     const void *ip6hdr)
2288 {
2289 
2290 	/* Is logging enabled? */
2291 	if (tcp_log_debug == 0)
2292 		return (NULL);
2293 
2294 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2295 }
2296 
2297 static char *
2298 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2299     const void *ip6hdr)
2300 {
2301 	char *s, *sp;
2302 	size_t size;
2303 	struct ip *ip;
2304 #ifdef INET6
2305 	const struct ip6_hdr *ip6;
2306 
2307 	ip6 = (const struct ip6_hdr *)ip6hdr;
2308 #endif /* INET6 */
2309 	ip = (struct ip *)ip4hdr;
2310 
2311 	/*
2312 	 * The log line looks like this:
2313 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2314 	 */
2315 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2316 	    sizeof(PRINT_TH_FLAGS) + 1 +
2317 #ifdef INET6
2318 	    2 * INET6_ADDRSTRLEN;
2319 #else
2320 	    2 * INET_ADDRSTRLEN;
2321 #endif /* INET6 */
2322 
2323 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2324 	if (s == NULL)
2325 		return (NULL);
2326 
2327 	strcat(s, "TCP: [");
2328 	sp = s + strlen(s);
2329 
2330 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2331 		inet_ntoa_r(inc->inc_faddr, sp);
2332 		sp = s + strlen(s);
2333 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2334 		sp = s + strlen(s);
2335 		inet_ntoa_r(inc->inc_laddr, sp);
2336 		sp = s + strlen(s);
2337 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2338 #ifdef INET6
2339 	} else if (inc) {
2340 		ip6_sprintf(sp, &inc->inc6_faddr);
2341 		sp = s + strlen(s);
2342 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2343 		sp = s + strlen(s);
2344 		ip6_sprintf(sp, &inc->inc6_laddr);
2345 		sp = s + strlen(s);
2346 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2347 	} else if (ip6 && th) {
2348 		ip6_sprintf(sp, &ip6->ip6_src);
2349 		sp = s + strlen(s);
2350 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2351 		sp = s + strlen(s);
2352 		ip6_sprintf(sp, &ip6->ip6_dst);
2353 		sp = s + strlen(s);
2354 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2355 #endif /* INET6 */
2356 #ifdef INET
2357 	} else if (ip && th) {
2358 		inet_ntoa_r(ip->ip_src, sp);
2359 		sp = s + strlen(s);
2360 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2361 		sp = s + strlen(s);
2362 		inet_ntoa_r(ip->ip_dst, sp);
2363 		sp = s + strlen(s);
2364 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2365 #endif /* INET */
2366 	} else {
2367 		free(s, M_TCPLOG);
2368 		return (NULL);
2369 	}
2370 	sp = s + strlen(s);
2371 	if (th)
2372 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2373 	if (*(s + size - 1) != '\0')
2374 		panic("%s: string too long", __func__);
2375 	return (s);
2376 }
2377