xref: /freebsd/sys/netinet/tcp_subr.c (revision 5861f9665471e98e544f6fa3ce73c4912229ff82)
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/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/jail.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 #include <sys/vimage.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/route.h>
63 #include <net/if.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet/vinet.h>
98 #include <netinet6/ip6protosw.h>
99 #include <netinet6/vinet6.h>
100 
101 #ifdef IPSEC
102 #include <netipsec/ipsec.h>
103 #include <netipsec/xform.h>
104 #ifdef INET6
105 #include <netipsec/ipsec6.h>
106 #endif
107 #include <netipsec/key.h>
108 #include <sys/syslog.h>
109 #endif /*IPSEC*/
110 
111 #include <machine/in_cksum.h>
112 #include <sys/md5.h>
113 
114 #include <security/mac/mac_framework.h>
115 
116 #ifdef VIMAGE_GLOBALS
117 int	tcp_mssdflt;
118 #ifdef INET6
119 int	tcp_v6mssdflt;
120 #endif
121 int	tcp_minmss;
122 int	tcp_do_rfc1323;
123 static int	icmp_may_rst;
124 static int	tcp_isn_reseed_interval;
125 static int	tcp_inflight_enable;
126 static int	tcp_inflight_rttthresh;
127 static int	tcp_inflight_min;
128 static int	tcp_inflight_max;
129 static int	tcp_inflight_stab;
130 #endif
131 
132 static int
133 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
134 {
135 	INIT_VNET_INET(TD_TO_VNET(req->td));
136 	int error, new;
137 
138 	new = V_tcp_mssdflt;
139 	error = sysctl_handle_int(oidp, &new, 0, req);
140 	if (error == 0 && req->newptr) {
141 		if (new < TCP_MINMSS)
142 			error = EINVAL;
143 		else
144 			V_tcp_mssdflt = new;
145 	}
146 	return (error);
147 }
148 
149 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
150     CTLTYPE_INT|CTLFLAG_RW, tcp_mssdflt, 0,
151     &sysctl_net_inet_tcp_mss_check, "I",
152     "Default TCP Maximum Segment Size");
153 
154 #ifdef INET6
155 static int
156 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
157 {
158 	INIT_VNET_INET(TD_TO_VNET(req->td));
159 	int error, new;
160 
161 	new = V_tcp_v6mssdflt;
162 	error = sysctl_handle_int(oidp, &new, 0, req);
163 	if (error == 0 && req->newptr) {
164 		if (new < TCP_MINMSS)
165 			error = EINVAL;
166 		else
167 			V_tcp_v6mssdflt = new;
168 	}
169 	return (error);
170 }
171 
172 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
173     CTLTYPE_INT|CTLFLAG_RW, tcp_v6mssdflt, 0,
174     &sysctl_net_inet_tcp_mss_v6_check, "I",
175    "Default TCP Maximum Segment Size for IPv6");
176 #endif
177 
178 /*
179  * Minimum MSS we accept and use. This prevents DoS attacks where
180  * we are forced to a ridiculous low MSS like 20 and send hundreds
181  * of packets instead of one. The effect scales with the available
182  * bandwidth and quickly saturates the CPU and network interface
183  * with packet generation and sending. Set to zero to disable MINMSS
184  * checking. This setting prevents us from sending too small packets.
185  */
186 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
187     CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
188 
189 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
190     CTLFLAG_RW, tcp_do_rfc1323, 0,
191     "Enable rfc1323 (high performance TCP) extensions");
192 
193 static int	tcp_log_debug = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
195     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
196 
197 static int	tcp_tcbhashsize = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
199     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
200 
201 static int	do_tcpdrain = 1;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
203     "Enable tcp_drain routine for extra help when low on mbufs");
204 
205 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
206     CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs");
207 
208 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
209     CTLFLAG_RW, icmp_may_rst, 0,
210     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
211 
212 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
213     CTLFLAG_RW, tcp_isn_reseed_interval, 0,
214     "Seconds between reseeding of ISN secret");
215 
216 /*
217  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
218  * 1024 exists only for debugging.  A good production default would be
219  * something like 6100.
220  */
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222     "TCP inflight data limiting");
223 
224 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
225     CTLFLAG_RW, tcp_inflight_enable, 0,
226     "Enable automatic TCP inflight data limiting");
227 
228 static int	tcp_inflight_debug = 0;
229 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
230     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
231 
232 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, rttthresh,
233     CTLTYPE_INT|CTLFLAG_RW, tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks,
234     "I", "RTT threshold below which inflight will deactivate itself");
235 
236 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
237     CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
238 
239 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
240     CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
241 
242 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
243     CTLFLAG_RW, tcp_inflight_stab, 0,
244     "Inflight Algorithm Stabilization 20 = 2 packets");
245 
246 #ifdef VIMAGE_GLOBALS
247 uma_zone_t sack_hole_zone;
248 #endif
249 
250 static struct inpcb *tcp_notify(struct inpcb *, int);
251 static void	tcp_isn_tick(void *);
252 
253 /*
254  * Target size of TCP PCB hash tables. Must be a power of two.
255  *
256  * Note that this can be overridden by the kernel environment
257  * variable net.inet.tcp.tcbhashsize
258  */
259 #ifndef TCBHASHSIZE
260 #define TCBHASHSIZE	512
261 #endif
262 
263 /*
264  * XXX
265  * Callouts should be moved into struct tcp directly.  They are currently
266  * separate because the tcpcb structure is exported to userland for sysctl
267  * parsing purposes, which do not know about callouts.
268  */
269 struct tcpcb_mem {
270 	struct	tcpcb		tcb;
271 	struct	tcp_timer	tt;
272 };
273 
274 #ifdef VIMAGE_GLOBALS
275 static uma_zone_t tcpcb_zone;
276 #endif
277 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
278 struct callout isn_callout;
279 static struct mtx isn_mtx;
280 
281 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
282 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
283 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
284 
285 /*
286  * TCP initialization.
287  */
288 static void
289 tcp_zone_change(void *tag)
290 {
291 	INIT_VNET_INET(curvnet);
292 
293 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
294 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
295 	tcp_tw_zone_change();
296 }
297 
298 static int
299 tcp_inpcb_init(void *mem, int size, int flags)
300 {
301 	struct inpcb *inp = mem;
302 
303 	INP_LOCK_INIT(inp, "inp", "tcpinp");
304 	return (0);
305 }
306 
307 void
308 tcp_init(void)
309 {
310 	INIT_VNET_INET(curvnet);
311 	int hashsize;
312 
313 	V_blackhole = 0;
314 	V_tcp_delack_enabled = 1;
315 	V_drop_synfin = 0;
316 	V_tcp_do_rfc3042 = 1;
317 	V_tcp_do_rfc3390 = 1;
318 	V_tcp_do_ecn = 0;
319 	V_tcp_ecn_maxretries = 1;
320 	V_tcp_insecure_rst = 0;
321 	V_tcp_do_autorcvbuf = 1;
322 	V_tcp_autorcvbuf_inc = 16*1024;
323 	V_tcp_autorcvbuf_max = 256*1024;
324 	V_tcp_do_rfc3465 = 1;
325 	V_tcp_abc_l_var = 2;
326 
327 	V_tcp_mssdflt = TCP_MSS;
328 #ifdef INET6
329 	V_tcp_v6mssdflt = TCP6_MSS;
330 #endif
331 	V_tcp_minmss = TCP_MINMSS;
332 	V_tcp_do_rfc1323 = 1;
333 	V_icmp_may_rst = 1;
334 	V_tcp_isn_reseed_interval = 0;
335 	V_tcp_inflight_enable = 1;
336 	V_tcp_inflight_min = 6144;
337 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
338 	V_tcp_inflight_stab = 20;
339 
340 	V_path_mtu_discovery = 1;
341 	V_ss_fltsz = 1;
342 	V_ss_fltsz_local = 4;
343 	V_tcp_do_newreno = 1;
344 	V_tcp_do_tso = 1;
345 	V_tcp_do_autosndbuf = 1;
346 	V_tcp_autosndbuf_inc = 8*1024;
347 	V_tcp_autosndbuf_max = 256*1024;
348 
349 	V_nolocaltimewait = 0;
350 
351 	V_tcp_do_sack = 1;
352 	V_tcp_sack_maxholes = 128;
353 	V_tcp_sack_globalmaxholes = 65536;
354 	V_tcp_sack_globalholes = 0;
355 
356 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
357 
358 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
359 
360 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
361 	LIST_INIT(&V_tcb);
362 #ifdef VIMAGE
363 	V_tcbinfo.ipi_vnet = curvnet;
364 #endif
365 	V_tcbinfo.ipi_listhead = &V_tcb;
366 	hashsize = TCBHASHSIZE;
367 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
368 	if (!powerof2(hashsize)) {
369 		printf("WARNING: TCB hash size not a power of 2\n");
370 		hashsize = 512; /* safe default */
371 	}
372 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
373 	    &V_tcbinfo.ipi_hashmask);
374 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
375 	    &V_tcbinfo.ipi_porthashmask);
376 	V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
377 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
378 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
379 	/*
380 	 * These have to be type stable for the benefit of the timers.
381 	 */
382 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
383 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
385 	tcp_tw_init();
386 	syncache_init();
387 	tcp_hc_init();
388 	tcp_reass_init();
389 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
390 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
391 
392 	/* Skip initialization of globals for non-default instances. */
393 	if (!IS_DEFAULT_VNET(curvnet))
394 		return;
395 
396 	/* XXX virtualize those bellow? */
397 	tcp_delacktime = TCPTV_DELACK;
398 	tcp_keepinit = TCPTV_KEEP_INIT;
399 	tcp_keepidle = TCPTV_KEEP_IDLE;
400 	tcp_keepintvl = TCPTV_KEEPINTVL;
401 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
402 	tcp_msl = TCPTV_MSL;
403 	tcp_rexmit_min = TCPTV_MIN;
404 	if (tcp_rexmit_min < 1)
405 		tcp_rexmit_min = 1;
406 	tcp_rexmit_slop = TCPTV_CPU_VAR;
407 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
408 	tcp_tcbhashsize = hashsize;
409 
410 #ifdef INET6
411 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
412 #else /* INET6 */
413 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
414 #endif /* INET6 */
415 	if (max_protohdr < TCP_MINPROTOHDR)
416 		max_protohdr = TCP_MINPROTOHDR;
417 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
418 		panic("tcp_init");
419 #undef TCP_MINPROTOHDR
420 
421 	ISN_LOCK_INIT();
422 	callout_init(&isn_callout, CALLOUT_MPSAFE);
423 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
424 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
425 		SHUTDOWN_PRI_DEFAULT);
426 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
427 		EVENTHANDLER_PRI_ANY);
428 }
429 
430 #ifdef VIMAGE
431 void
432 tcp_destroy(void)
433 {
434 	INIT_VNET_INET(curvnet);
435 
436 	tcp_tw_destroy();
437 	tcp_hc_destroy();
438 	syncache_destroy();
439 
440 	/* XXX check that hashes are empty! */
441 	hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
442 	    V_tcbinfo.ipi_hashmask);
443 	hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
444 	    V_tcbinfo.ipi_porthashmask);
445 	INP_INFO_LOCK_DESTROY(&V_tcbinfo);
446 }
447 #endif
448 
449 void
450 tcp_fini(void *xtp)
451 {
452 
453 	callout_stop(&isn_callout);
454 }
455 
456 /*
457  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
458  * tcp_template used to store this data in mbufs, but we now recopy it out
459  * of the tcpcb each time to conserve mbufs.
460  */
461 void
462 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
463 {
464 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
465 
466 	INP_WLOCK_ASSERT(inp);
467 
468 #ifdef INET6
469 	if ((inp->inp_vflag & INP_IPV6) != 0) {
470 		struct ip6_hdr *ip6;
471 
472 		ip6 = (struct ip6_hdr *)ip_ptr;
473 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
474 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
475 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
476 			(IPV6_VERSION & IPV6_VERSION_MASK);
477 		ip6->ip6_nxt = IPPROTO_TCP;
478 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
479 		ip6->ip6_src = inp->in6p_laddr;
480 		ip6->ip6_dst = inp->in6p_faddr;
481 	} else
482 #endif
483 	{
484 		struct ip *ip;
485 
486 		ip = (struct ip *)ip_ptr;
487 		ip->ip_v = IPVERSION;
488 		ip->ip_hl = 5;
489 		ip->ip_tos = inp->inp_ip_tos;
490 		ip->ip_len = 0;
491 		ip->ip_id = 0;
492 		ip->ip_off = 0;
493 		ip->ip_ttl = inp->inp_ip_ttl;
494 		ip->ip_sum = 0;
495 		ip->ip_p = IPPROTO_TCP;
496 		ip->ip_src = inp->inp_laddr;
497 		ip->ip_dst = inp->inp_faddr;
498 	}
499 	th->th_sport = inp->inp_lport;
500 	th->th_dport = inp->inp_fport;
501 	th->th_seq = 0;
502 	th->th_ack = 0;
503 	th->th_x2 = 0;
504 	th->th_off = 5;
505 	th->th_flags = 0;
506 	th->th_win = 0;
507 	th->th_urp = 0;
508 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
509 }
510 
511 /*
512  * Create template to be used to send tcp packets on a connection.
513  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
514  * use for this function is in keepalives, which use tcp_respond.
515  */
516 struct tcptemp *
517 tcpip_maketemplate(struct inpcb *inp)
518 {
519 	struct tcptemp *t;
520 
521 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
522 	if (t == NULL)
523 		return (NULL);
524 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
525 	return (t);
526 }
527 
528 /*
529  * Send a single message to the TCP at address specified by
530  * the given TCP/IP header.  If m == NULL, then we make a copy
531  * of the tcpiphdr at ti and send directly to the addressed host.
532  * This is used to force keep alive messages out using the TCP
533  * template for a connection.  If flags are given then we send
534  * a message back to the TCP which originated the * segment ti,
535  * and discard the mbuf containing it and any other attached mbufs.
536  *
537  * In any case the ack and sequence number of the transmitted
538  * segment are as specified by the parameters.
539  *
540  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
541  */
542 void
543 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
544     tcp_seq ack, tcp_seq seq, int flags)
545 {
546 	INIT_VNET_INET(curvnet);
547 	int tlen;
548 	int win = 0;
549 	struct ip *ip;
550 	struct tcphdr *nth;
551 #ifdef INET6
552 	struct ip6_hdr *ip6;
553 	int isipv6;
554 #endif /* INET6 */
555 	int ipflags = 0;
556 	struct inpcb *inp;
557 
558 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
559 
560 #ifdef INET6
561 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
562 	ip6 = ipgen;
563 #endif /* INET6 */
564 	ip = ipgen;
565 
566 	if (tp != NULL) {
567 		inp = tp->t_inpcb;
568 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
569 		INP_WLOCK_ASSERT(inp);
570 	} else
571 		inp = NULL;
572 
573 	if (tp != NULL) {
574 		if (!(flags & TH_RST)) {
575 			win = sbspace(&inp->inp_socket->so_rcv);
576 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
577 				win = (long)TCP_MAXWIN << tp->rcv_scale;
578 		}
579 	}
580 	if (m == NULL) {
581 		m = m_gethdr(M_DONTWAIT, MT_DATA);
582 		if (m == NULL)
583 			return;
584 		tlen = 0;
585 		m->m_data += max_linkhdr;
586 #ifdef INET6
587 		if (isipv6) {
588 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
589 			      sizeof(struct ip6_hdr));
590 			ip6 = mtod(m, struct ip6_hdr *);
591 			nth = (struct tcphdr *)(ip6 + 1);
592 		} else
593 #endif /* INET6 */
594 	      {
595 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
596 		ip = mtod(m, struct ip *);
597 		nth = (struct tcphdr *)(ip + 1);
598 	      }
599 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
600 		flags = TH_ACK;
601 	} else {
602 		/*
603 		 *  reuse the mbuf.
604 		 * XXX MRT We inherrit the FIB, which is lucky.
605 		 */
606 		m_freem(m->m_next);
607 		m->m_next = NULL;
608 		m->m_data = (caddr_t)ipgen;
609 		/* m_len is set later */
610 		tlen = 0;
611 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
612 #ifdef INET6
613 		if (isipv6) {
614 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
615 			nth = (struct tcphdr *)(ip6 + 1);
616 		} else
617 #endif /* INET6 */
618 	      {
619 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
620 		nth = (struct tcphdr *)(ip + 1);
621 	      }
622 		if (th != nth) {
623 			/*
624 			 * this is usually a case when an extension header
625 			 * exists between the IPv6 header and the
626 			 * TCP header.
627 			 */
628 			nth->th_sport = th->th_sport;
629 			nth->th_dport = th->th_dport;
630 		}
631 		xchg(nth->th_dport, nth->th_sport, uint16_t);
632 #undef xchg
633 	}
634 #ifdef INET6
635 	if (isipv6) {
636 		ip6->ip6_flow = 0;
637 		ip6->ip6_vfc = IPV6_VERSION;
638 		ip6->ip6_nxt = IPPROTO_TCP;
639 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
640 						tlen));
641 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
642 	} else
643 #endif
644 	{
645 		tlen += sizeof (struct tcpiphdr);
646 		ip->ip_len = tlen;
647 		ip->ip_ttl = V_ip_defttl;
648 		if (V_path_mtu_discovery)
649 			ip->ip_off |= IP_DF;
650 	}
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 #ifdef INET6
681 	if (isipv6) {
682 		nth->th_sum = 0;
683 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
684 					sizeof(struct ip6_hdr),
685 					tlen - sizeof(struct ip6_hdr));
686 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
687 		    NULL, NULL);
688 	} else
689 #endif /* INET6 */
690 	{
691 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
692 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
693 		m->m_pkthdr.csum_flags = CSUM_TCP;
694 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
695 	}
696 #ifdef TCPDEBUG
697 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
698 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
699 #endif
700 #ifdef INET6
701 	if (isipv6)
702 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
703 	else
704 #endif /* INET6 */
705 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
706 }
707 
708 /*
709  * Create a new TCP control block, making an
710  * empty reassembly queue and hooking it to the argument
711  * protocol control block.  The `inp' parameter must have
712  * come from the zone allocator set up in tcp_init().
713  */
714 struct tcpcb *
715 tcp_newtcpcb(struct inpcb *inp)
716 {
717 	INIT_VNET_INET(inp->inp_vnet);
718 	struct tcpcb_mem *tm;
719 	struct tcpcb *tp;
720 #ifdef INET6
721 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
722 #endif /* INET6 */
723 
724 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
725 	if (tm == NULL)
726 		return (NULL);
727 	tp = &tm->tcb;
728 #ifdef VIMAGE
729 	tp->t_vnet = inp->inp_vnet;
730 #endif
731 	tp->t_timers = &tm->tt;
732 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
733 	tp->t_maxseg = tp->t_maxopd =
734 #ifdef INET6
735 		isipv6 ? V_tcp_v6mssdflt :
736 #endif /* INET6 */
737 		V_tcp_mssdflt;
738 
739 	/* Set up our timeouts. */
740 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
741 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
742 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
743 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
744 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
745 
746 	if (V_tcp_do_rfc1323)
747 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
748 	if (V_tcp_do_sack)
749 		tp->t_flags |= TF_SACK_PERMIT;
750 	TAILQ_INIT(&tp->snd_holes);
751 	tp->t_inpcb = inp;	/* XXX */
752 	/*
753 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
754 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
755 	 * reasonable initial retransmit time.
756 	 */
757 	tp->t_srtt = TCPTV_SRTTBASE;
758 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
759 	tp->t_rttmin = tcp_rexmit_min;
760 	tp->t_rxtcur = TCPTV_RTOBASE;
761 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
762 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
763 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
764 	tp->t_rcvtime = ticks;
765 	tp->t_bw_rtttime = ticks;
766 	/*
767 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
768 	 * because the socket may be bound to an IPv6 wildcard address,
769 	 * which may match an IPv4-mapped IPv6 address.
770 	 */
771 	inp->inp_ip_ttl = V_ip_defttl;
772 	inp->inp_ppcb = tp;
773 	return (tp);		/* XXX */
774 }
775 
776 /*
777  * Drop a TCP connection, reporting
778  * the specified error.  If connection is synchronized,
779  * then send a RST to peer.
780  */
781 struct tcpcb *
782 tcp_drop(struct tcpcb *tp, int errno)
783 {
784 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
785 	struct socket *so = tp->t_inpcb->inp_socket;
786 
787 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
788 	INP_WLOCK_ASSERT(tp->t_inpcb);
789 
790 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
791 		tp->t_state = TCPS_CLOSED;
792 		(void) tcp_output_reset(tp);
793 		TCPSTAT_INC(tcps_drops);
794 	} else
795 		TCPSTAT_INC(tcps_conndrops);
796 	if (errno == ETIMEDOUT && tp->t_softerror)
797 		errno = tp->t_softerror;
798 	so->so_error = errno;
799 	return (tcp_close(tp));
800 }
801 
802 void
803 tcp_discardcb(struct tcpcb *tp)
804 {
805 	INIT_VNET_INET(tp->t_vnet);
806 	struct tseg_qent *q;
807 	struct inpcb *inp = tp->t_inpcb;
808 	struct socket *so = inp->inp_socket;
809 #ifdef INET6
810 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
811 #endif /* INET6 */
812 
813 	INP_WLOCK_ASSERT(inp);
814 
815 	/*
816 	 * Make sure that all of our timers are stopped before we
817 	 * delete the PCB.
818 	 */
819 	callout_stop(&tp->t_timers->tt_rexmt);
820 	callout_stop(&tp->t_timers->tt_persist);
821 	callout_stop(&tp->t_timers->tt_keep);
822 	callout_stop(&tp->t_timers->tt_2msl);
823 	callout_stop(&tp->t_timers->tt_delack);
824 
825 	/*
826 	 * If we got enough samples through the srtt filter,
827 	 * save the rtt and rttvar in the routing entry.
828 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
829 	 * 4 samples is enough for the srtt filter to converge
830 	 * to within enough % of the correct value; fewer samples
831 	 * and we could save a bogus rtt. The danger is not high
832 	 * as tcp quickly recovers from everything.
833 	 * XXX: Works very well but needs some more statistics!
834 	 */
835 	if (tp->t_rttupdated >= 4) {
836 		struct hc_metrics_lite metrics;
837 		u_long ssthresh;
838 
839 		bzero(&metrics, sizeof(metrics));
840 		/*
841 		 * Update the ssthresh always when the conditions below
842 		 * are satisfied. This gives us better new start value
843 		 * for the congestion avoidance for new connections.
844 		 * ssthresh is only set if packet loss occured on a session.
845 		 *
846 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
847 		 * being torn down.  Ideally this code would not use 'so'.
848 		 */
849 		ssthresh = tp->snd_ssthresh;
850 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
851 			/*
852 			 * convert the limit from user data bytes to
853 			 * packets then to packet data bytes.
854 			 */
855 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
856 			if (ssthresh < 2)
857 				ssthresh = 2;
858 			ssthresh *= (u_long)(tp->t_maxseg +
859 #ifdef INET6
860 				      (isipv6 ? sizeof (struct ip6_hdr) +
861 					       sizeof (struct tcphdr) :
862 #endif
863 				       sizeof (struct tcpiphdr)
864 #ifdef INET6
865 				       )
866 #endif
867 				      );
868 		} else
869 			ssthresh = 0;
870 		metrics.rmx_ssthresh = ssthresh;
871 
872 		metrics.rmx_rtt = tp->t_srtt;
873 		metrics.rmx_rttvar = tp->t_rttvar;
874 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
875 		metrics.rmx_bandwidth = tp->snd_bandwidth;
876 		metrics.rmx_cwnd = tp->snd_cwnd;
877 		metrics.rmx_sendpipe = 0;
878 		metrics.rmx_recvpipe = 0;
879 
880 		tcp_hc_update(&inp->inp_inc, &metrics);
881 	}
882 
883 	/* free the reassembly queue, if any */
884 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
885 		LIST_REMOVE(q, tqe_q);
886 		m_freem(q->tqe_m);
887 		uma_zfree(V_tcp_reass_zone, q);
888 		tp->t_segqlen--;
889 		V_tcp_reass_qsize--;
890 	}
891 	/* Disconnect offload device, if any. */
892 	tcp_offload_detach(tp);
893 
894 	tcp_free_sackholes(tp);
895 	inp->inp_ppcb = NULL;
896 	tp->t_inpcb = NULL;
897 	uma_zfree(V_tcpcb_zone, tp);
898 }
899 
900 /*
901  * Attempt to close a TCP control block, marking it as dropped, and freeing
902  * the socket if we hold the only reference.
903  */
904 struct tcpcb *
905 tcp_close(struct tcpcb *tp)
906 {
907 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
908 	struct inpcb *inp = tp->t_inpcb;
909 	struct socket *so;
910 
911 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
912 	INP_WLOCK_ASSERT(inp);
913 
914 	/* Notify any offload devices of listener close */
915 	if (tp->t_state == TCPS_LISTEN)
916 		tcp_offload_listen_close(tp);
917 	in_pcbdrop(inp);
918 	TCPSTAT_INC(tcps_closed);
919 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
920 	so = inp->inp_socket;
921 	soisdisconnected(so);
922 	if (inp->inp_flags & INP_SOCKREF) {
923 		KASSERT(so->so_state & SS_PROTOREF,
924 		    ("tcp_close: !SS_PROTOREF"));
925 		inp->inp_flags &= ~INP_SOCKREF;
926 		INP_WUNLOCK(inp);
927 		ACCEPT_LOCK();
928 		SOCK_LOCK(so);
929 		so->so_state &= ~SS_PROTOREF;
930 		sofree(so);
931 		return (NULL);
932 	}
933 	return (tp);
934 }
935 
936 void
937 tcp_drain(void)
938 {
939 	VNET_ITERATOR_DECL(vnet_iter);
940 
941 	if (!do_tcpdrain)
942 		return;
943 
944 	VNET_LIST_RLOCK();
945 	VNET_FOREACH(vnet_iter) {
946 		CURVNET_SET(vnet_iter);
947 		INIT_VNET_INET(vnet_iter);
948 		struct inpcb *inpb;
949 		struct tcpcb *tcpb;
950 		struct tseg_qent *te;
951 
952 	/*
953 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
954 	 * if there is one...
955 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
956 	 *      reassembly queue should be flushed, but in a situation
957 	 *	where we're really low on mbufs, this is potentially
958 	 *	usefull.
959 	 */
960 		INP_INFO_RLOCK(&V_tcbinfo);
961 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
962 			if (inpb->inp_flags & INP_TIMEWAIT)
963 				continue;
964 			INP_WLOCK(inpb);
965 			if ((tcpb = intotcpcb(inpb)) != NULL) {
966 				while ((te = LIST_FIRST(&tcpb->t_segq))
967 			            != NULL) {
968 					LIST_REMOVE(te, tqe_q);
969 					m_freem(te->tqe_m);
970 					uma_zfree(V_tcp_reass_zone, te);
971 					tcpb->t_segqlen--;
972 					V_tcp_reass_qsize--;
973 				}
974 				tcp_clean_sackreport(tcpb);
975 			}
976 			INP_WUNLOCK(inpb);
977 		}
978 		INP_INFO_RUNLOCK(&V_tcbinfo);
979 		CURVNET_RESTORE();
980 	}
981 	VNET_LIST_RUNLOCK();
982 }
983 
984 /*
985  * Notify a tcp user of an asynchronous error;
986  * store error as soft error, but wake up user
987  * (for now, won't do anything until can select for soft error).
988  *
989  * Do not wake up user since there currently is no mechanism for
990  * reporting soft errors (yet - a kqueue filter may be added).
991  */
992 static struct inpcb *
993 tcp_notify(struct inpcb *inp, int error)
994 {
995 	struct tcpcb *tp;
996 #ifdef INVARIANTS
997 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
998 #endif
999 
1000 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1001 	INP_WLOCK_ASSERT(inp);
1002 
1003 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1004 	    (inp->inp_flags & INP_DROPPED))
1005 		return (inp);
1006 
1007 	tp = intotcpcb(inp);
1008 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1009 
1010 	/*
1011 	 * Ignore some errors if we are hooked up.
1012 	 * If connection hasn't completed, has retransmitted several times,
1013 	 * and receives a second error, give up now.  This is better
1014 	 * than waiting a long time to establish a connection that
1015 	 * can never complete.
1016 	 */
1017 	if (tp->t_state == TCPS_ESTABLISHED &&
1018 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1019 	     error == EHOSTDOWN)) {
1020 		return (inp);
1021 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1022 	    tp->t_softerror) {
1023 		tp = tcp_drop(tp, error);
1024 		if (tp != NULL)
1025 			return (inp);
1026 		else
1027 			return (NULL);
1028 	} else {
1029 		tp->t_softerror = error;
1030 		return (inp);
1031 	}
1032 #if 0
1033 	wakeup( &so->so_timeo);
1034 	sorwakeup(so);
1035 	sowwakeup(so);
1036 #endif
1037 }
1038 
1039 static int
1040 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1041 {
1042 	INIT_VNET_INET(curvnet);
1043 	int error, i, m, n, pcb_count;
1044 	struct inpcb *inp, **inp_list;
1045 	inp_gen_t gencnt;
1046 	struct xinpgen xig;
1047 
1048 	/*
1049 	 * The process of preparing the TCB list is too time-consuming and
1050 	 * resource-intensive to repeat twice on every request.
1051 	 */
1052 	if (req->oldptr == NULL) {
1053 		m = syncache_pcbcount();
1054 		n = V_tcbinfo.ipi_count;
1055 		req->oldidx = 2 * (sizeof xig)
1056 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1057 		return (0);
1058 	}
1059 
1060 	if (req->newptr != NULL)
1061 		return (EPERM);
1062 
1063 	/*
1064 	 * OK, now we're committed to doing something.
1065 	 */
1066 	INP_INFO_RLOCK(&V_tcbinfo);
1067 	gencnt = V_tcbinfo.ipi_gencnt;
1068 	n = V_tcbinfo.ipi_count;
1069 	INP_INFO_RUNLOCK(&V_tcbinfo);
1070 
1071 	m = syncache_pcbcount();
1072 
1073 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1074 		+ (n + m) * sizeof(struct xtcpcb));
1075 	if (error != 0)
1076 		return (error);
1077 
1078 	xig.xig_len = sizeof xig;
1079 	xig.xig_count = n + m;
1080 	xig.xig_gen = gencnt;
1081 	xig.xig_sogen = so_gencnt;
1082 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1083 	if (error)
1084 		return (error);
1085 
1086 	error = syncache_pcblist(req, m, &pcb_count);
1087 	if (error)
1088 		return (error);
1089 
1090 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1091 	if (inp_list == NULL)
1092 		return (ENOMEM);
1093 
1094 	INP_INFO_RLOCK(&V_tcbinfo);
1095 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1096 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1097 		INP_RLOCK(inp);
1098 		if (inp->inp_gencnt <= gencnt) {
1099 			/*
1100 			 * XXX: This use of cr_cansee(), introduced with
1101 			 * TCP state changes, is not quite right, but for
1102 			 * now, better than nothing.
1103 			 */
1104 			if (inp->inp_flags & INP_TIMEWAIT) {
1105 				if (intotw(inp) != NULL)
1106 					error = cr_cansee(req->td->td_ucred,
1107 					    intotw(inp)->tw_cred);
1108 				else
1109 					error = EINVAL;	/* Skip this inp. */
1110 			} else
1111 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1112 			if (error == 0)
1113 				inp_list[i++] = inp;
1114 		}
1115 		INP_RUNLOCK(inp);
1116 	}
1117 	INP_INFO_RUNLOCK(&V_tcbinfo);
1118 	n = i;
1119 
1120 	error = 0;
1121 	for (i = 0; i < n; i++) {
1122 		inp = inp_list[i];
1123 		INP_RLOCK(inp);
1124 		if (inp->inp_gencnt <= gencnt) {
1125 			struct xtcpcb xt;
1126 			void *inp_ppcb;
1127 
1128 			bzero(&xt, sizeof(xt));
1129 			xt.xt_len = sizeof xt;
1130 			/* XXX should avoid extra copy */
1131 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1132 			inp_ppcb = inp->inp_ppcb;
1133 			if (inp_ppcb == NULL)
1134 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1135 			else if (inp->inp_flags & INP_TIMEWAIT) {
1136 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1137 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1138 			} else
1139 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1140 			if (inp->inp_socket != NULL)
1141 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1142 			else {
1143 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1144 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1145 			}
1146 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1147 			INP_RUNLOCK(inp);
1148 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1149 		} else
1150 			INP_RUNLOCK(inp);
1151 
1152 	}
1153 	if (!error) {
1154 		/*
1155 		 * Give the user an updated idea of our state.
1156 		 * If the generation differs from what we told
1157 		 * her before, she knows that something happened
1158 		 * while we were processing this request, and it
1159 		 * might be necessary to retry.
1160 		 */
1161 		INP_INFO_RLOCK(&V_tcbinfo);
1162 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1163 		xig.xig_sogen = so_gencnt;
1164 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1165 		INP_INFO_RUNLOCK(&V_tcbinfo);
1166 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1167 	}
1168 	free(inp_list, M_TEMP);
1169 	return (error);
1170 }
1171 
1172 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1173     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1174 
1175 static int
1176 tcp_getcred(SYSCTL_HANDLER_ARGS)
1177 {
1178 	INIT_VNET_INET(curvnet);
1179 	struct xucred xuc;
1180 	struct sockaddr_in addrs[2];
1181 	struct inpcb *inp;
1182 	int error;
1183 
1184 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1185 	if (error)
1186 		return (error);
1187 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1188 	if (error)
1189 		return (error);
1190 	INP_INFO_RLOCK(&V_tcbinfo);
1191 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1192 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1193 	if (inp != NULL) {
1194 		INP_RLOCK(inp);
1195 		INP_INFO_RUNLOCK(&V_tcbinfo);
1196 		if (inp->inp_socket == NULL)
1197 			error = ENOENT;
1198 		if (error == 0)
1199 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1200 		if (error == 0)
1201 			cru2x(inp->inp_cred, &xuc);
1202 		INP_RUNLOCK(inp);
1203 	} else {
1204 		INP_INFO_RUNLOCK(&V_tcbinfo);
1205 		error = ENOENT;
1206 	}
1207 	if (error == 0)
1208 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1209 	return (error);
1210 }
1211 
1212 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1213     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1214     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1215 
1216 #ifdef INET6
1217 static int
1218 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1219 {
1220 	INIT_VNET_INET(curvnet);
1221 	INIT_VNET_INET6(curvnet);
1222 	struct xucred xuc;
1223 	struct sockaddr_in6 addrs[2];
1224 	struct inpcb *inp;
1225 	int error, mapped = 0;
1226 
1227 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1228 	if (error)
1229 		return (error);
1230 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1231 	if (error)
1232 		return (error);
1233 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1234 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1235 		return (error);
1236 	}
1237 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1238 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1239 			mapped = 1;
1240 		else
1241 			return (EINVAL);
1242 	}
1243 
1244 	INP_INFO_RLOCK(&V_tcbinfo);
1245 	if (mapped == 1)
1246 		inp = in_pcblookup_hash(&V_tcbinfo,
1247 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1248 			addrs[1].sin6_port,
1249 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1250 			addrs[0].sin6_port,
1251 			0, NULL);
1252 	else
1253 		inp = in6_pcblookup_hash(&V_tcbinfo,
1254 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1255 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1256 	if (inp != NULL) {
1257 		INP_RLOCK(inp);
1258 		INP_INFO_RUNLOCK(&V_tcbinfo);
1259 		if (inp->inp_socket == NULL)
1260 			error = ENOENT;
1261 		if (error == 0)
1262 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1263 		if (error == 0)
1264 			cru2x(inp->inp_cred, &xuc);
1265 		INP_RUNLOCK(inp);
1266 	} else {
1267 		INP_INFO_RUNLOCK(&V_tcbinfo);
1268 		error = ENOENT;
1269 	}
1270 	if (error == 0)
1271 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1272 	return (error);
1273 }
1274 
1275 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1276     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1277     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1278 #endif
1279 
1280 
1281 void
1282 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1283 {
1284 	INIT_VNET_INET(curvnet);
1285 	struct ip *ip = vip;
1286 	struct tcphdr *th;
1287 	struct in_addr faddr;
1288 	struct inpcb *inp;
1289 	struct tcpcb *tp;
1290 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1291 	struct icmp *icp;
1292 	struct in_conninfo inc;
1293 	tcp_seq icmp_tcp_seq;
1294 	int mtu;
1295 
1296 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1297 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1298 		return;
1299 
1300 	if (cmd == PRC_MSGSIZE)
1301 		notify = tcp_mtudisc;
1302 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1303 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1304 		notify = tcp_drop_syn_sent;
1305 	/*
1306 	 * Redirects don't need to be handled up here.
1307 	 */
1308 	else if (PRC_IS_REDIRECT(cmd))
1309 		return;
1310 	/*
1311 	 * Source quench is depreciated.
1312 	 */
1313 	else if (cmd == PRC_QUENCH)
1314 		return;
1315 	/*
1316 	 * Hostdead is ugly because it goes linearly through all PCBs.
1317 	 * XXX: We never get this from ICMP, otherwise it makes an
1318 	 * excellent DoS attack on machines with many connections.
1319 	 */
1320 	else if (cmd == PRC_HOSTDEAD)
1321 		ip = NULL;
1322 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1323 		return;
1324 	if (ip != NULL) {
1325 		icp = (struct icmp *)((caddr_t)ip
1326 				      - offsetof(struct icmp, icmp_ip));
1327 		th = (struct tcphdr *)((caddr_t)ip
1328 				       + (ip->ip_hl << 2));
1329 		INP_INFO_WLOCK(&V_tcbinfo);
1330 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1331 		    ip->ip_src, th->th_sport, 0, NULL);
1332 		if (inp != NULL)  {
1333 			INP_WLOCK(inp);
1334 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1335 			    !(inp->inp_flags & INP_DROPPED) &&
1336 			    !(inp->inp_socket == NULL)) {
1337 				icmp_tcp_seq = htonl(th->th_seq);
1338 				tp = intotcpcb(inp);
1339 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1340 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1341 					if (cmd == PRC_MSGSIZE) {
1342 					    /*
1343 					     * MTU discovery:
1344 					     * If we got a needfrag set the MTU
1345 					     * in the route to the suggested new
1346 					     * value (if given) and then notify.
1347 					     */
1348 					    bzero(&inc, sizeof(inc));
1349 					    inc.inc_faddr = faddr;
1350 					    inc.inc_fibnum =
1351 						inp->inp_inc.inc_fibnum;
1352 
1353 					    mtu = ntohs(icp->icmp_nextmtu);
1354 					    /*
1355 					     * If no alternative MTU was
1356 					     * proposed, try the next smaller
1357 					     * one.  ip->ip_len has already
1358 					     * been swapped in icmp_input().
1359 					     */
1360 					    if (!mtu)
1361 						mtu = ip_next_mtu(ip->ip_len,
1362 						 1);
1363 					    if (mtu < max(296, V_tcp_minmss
1364 						 + sizeof(struct tcpiphdr)))
1365 						mtu = 0;
1366 					    if (!mtu)
1367 						mtu = V_tcp_mssdflt
1368 						 + sizeof(struct tcpiphdr);
1369 					    /*
1370 					     * Only cache the the MTU if it
1371 					     * is smaller than the interface
1372 					     * or route MTU.  tcp_mtudisc()
1373 					     * will do right thing by itself.
1374 					     */
1375 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1376 						tcp_hc_updatemtu(&inc, mtu);
1377 					}
1378 
1379 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1380 				}
1381 			}
1382 			if (inp != NULL)
1383 				INP_WUNLOCK(inp);
1384 		} else {
1385 			bzero(&inc, sizeof(inc));
1386 			inc.inc_fport = th->th_dport;
1387 			inc.inc_lport = th->th_sport;
1388 			inc.inc_faddr = faddr;
1389 			inc.inc_laddr = ip->ip_src;
1390 			syncache_unreach(&inc, th);
1391 		}
1392 		INP_INFO_WUNLOCK(&V_tcbinfo);
1393 	} else
1394 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1395 }
1396 
1397 #ifdef INET6
1398 void
1399 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1400 {
1401 	INIT_VNET_INET(curvnet);
1402 	struct tcphdr th;
1403 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1404 	struct ip6_hdr *ip6;
1405 	struct mbuf *m;
1406 	struct ip6ctlparam *ip6cp = NULL;
1407 	const struct sockaddr_in6 *sa6_src = NULL;
1408 	int off;
1409 	struct tcp_portonly {
1410 		u_int16_t th_sport;
1411 		u_int16_t th_dport;
1412 	} *thp;
1413 
1414 	if (sa->sa_family != AF_INET6 ||
1415 	    sa->sa_len != sizeof(struct sockaddr_in6))
1416 		return;
1417 
1418 	if (cmd == PRC_MSGSIZE)
1419 		notify = tcp_mtudisc;
1420 	else if (!PRC_IS_REDIRECT(cmd) &&
1421 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1422 		return;
1423 	/* Source quench is depreciated. */
1424 	else if (cmd == PRC_QUENCH)
1425 		return;
1426 
1427 	/* if the parameter is from icmp6, decode it. */
1428 	if (d != NULL) {
1429 		ip6cp = (struct ip6ctlparam *)d;
1430 		m = ip6cp->ip6c_m;
1431 		ip6 = ip6cp->ip6c_ip6;
1432 		off = ip6cp->ip6c_off;
1433 		sa6_src = ip6cp->ip6c_src;
1434 	} else {
1435 		m = NULL;
1436 		ip6 = NULL;
1437 		off = 0;	/* fool gcc */
1438 		sa6_src = &sa6_any;
1439 	}
1440 
1441 	if (ip6 != NULL) {
1442 		struct in_conninfo inc;
1443 		/*
1444 		 * XXX: We assume that when IPV6 is non NULL,
1445 		 * M and OFF are valid.
1446 		 */
1447 
1448 		/* check if we can safely examine src and dst ports */
1449 		if (m->m_pkthdr.len < off + sizeof(*thp))
1450 			return;
1451 
1452 		bzero(&th, sizeof(th));
1453 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1454 
1455 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1456 		    (struct sockaddr *)ip6cp->ip6c_src,
1457 		    th.th_sport, cmd, NULL, notify);
1458 
1459 		bzero(&inc, sizeof(inc));
1460 		inc.inc_fport = th.th_dport;
1461 		inc.inc_lport = th.th_sport;
1462 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1463 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1464 		inc.inc_flags |= INC_ISIPV6;
1465 		INP_INFO_WLOCK(&V_tcbinfo);
1466 		syncache_unreach(&inc, &th);
1467 		INP_INFO_WUNLOCK(&V_tcbinfo);
1468 	} else
1469 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1470 			      0, cmd, NULL, notify);
1471 }
1472 #endif /* INET6 */
1473 
1474 
1475 /*
1476  * Following is where TCP initial sequence number generation occurs.
1477  *
1478  * There are two places where we must use initial sequence numbers:
1479  * 1.  In SYN-ACK packets.
1480  * 2.  In SYN packets.
1481  *
1482  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1483  * tcp_syncache.c for details.
1484  *
1485  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1486  * depends on this property.  In addition, these ISNs should be
1487  * unguessable so as to prevent connection hijacking.  To satisfy
1488  * the requirements of this situation, the algorithm outlined in
1489  * RFC 1948 is used, with only small modifications.
1490  *
1491  * Implementation details:
1492  *
1493  * Time is based off the system timer, and is corrected so that it
1494  * increases by one megabyte per second.  This allows for proper
1495  * recycling on high speed LANs while still leaving over an hour
1496  * before rollover.
1497  *
1498  * As reading the *exact* system time is too expensive to be done
1499  * whenever setting up a TCP connection, we increment the time
1500  * offset in two ways.  First, a small random positive increment
1501  * is added to isn_offset for each connection that is set up.
1502  * Second, the function tcp_isn_tick fires once per clock tick
1503  * and increments isn_offset as necessary so that sequence numbers
1504  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1505  * random positive increments serve only to ensure that the same
1506  * exact sequence number is never sent out twice (as could otherwise
1507  * happen when a port is recycled in less than the system tick
1508  * interval.)
1509  *
1510  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1511  * between seeding of isn_secret.  This is normally set to zero,
1512  * as reseeding should not be necessary.
1513  *
1514  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1515  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1516  * general, this means holding an exclusive (write) lock.
1517  */
1518 
1519 #define ISN_BYTES_PER_SECOND 1048576
1520 #define ISN_STATIC_INCREMENT 4096
1521 #define ISN_RANDOM_INCREMENT (4096 - 1)
1522 
1523 #ifdef VIMAGE_GLOBALS
1524 static u_char isn_secret[32];
1525 static int isn_last_reseed;
1526 static u_int32_t isn_offset, isn_offset_old;
1527 #endif
1528 
1529 tcp_seq
1530 tcp_new_isn(struct tcpcb *tp)
1531 {
1532 	INIT_VNET_INET(tp->t_vnet);
1533 	MD5_CTX isn_ctx;
1534 	u_int32_t md5_buffer[4];
1535 	tcp_seq new_isn;
1536 
1537 	INP_WLOCK_ASSERT(tp->t_inpcb);
1538 
1539 	ISN_LOCK();
1540 	/* Seed if this is the first use, reseed if requested. */
1541 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1542 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1543 		< (u_int)ticks))) {
1544 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1545 		V_isn_last_reseed = ticks;
1546 	}
1547 
1548 	/* Compute the md5 hash and return the ISN. */
1549 	MD5Init(&isn_ctx);
1550 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1551 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1552 #ifdef INET6
1553 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1554 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1555 			  sizeof(struct in6_addr));
1556 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1557 			  sizeof(struct in6_addr));
1558 	} else
1559 #endif
1560 	{
1561 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1562 			  sizeof(struct in_addr));
1563 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1564 			  sizeof(struct in_addr));
1565 	}
1566 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1567 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1568 	new_isn = (tcp_seq) md5_buffer[0];
1569 	V_isn_offset += ISN_STATIC_INCREMENT +
1570 		(arc4random() & ISN_RANDOM_INCREMENT);
1571 	new_isn += V_isn_offset;
1572 	ISN_UNLOCK();
1573 	return (new_isn);
1574 }
1575 
1576 /*
1577  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1578  * to keep time flowing at a relatively constant rate.  If the random
1579  * increments have already pushed us past the projected offset, do nothing.
1580  */
1581 static void
1582 tcp_isn_tick(void *xtp)
1583 {
1584 	VNET_ITERATOR_DECL(vnet_iter);
1585 	u_int32_t projected_offset;
1586 
1587 	VNET_LIST_RLOCK();
1588 	ISN_LOCK();
1589 	VNET_FOREACH(vnet_iter) {
1590 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1591 		INIT_VNET_INET(curvnet);
1592 		projected_offset =
1593 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1594 
1595 		if (SEQ_GT(projected_offset, V_isn_offset))
1596 			V_isn_offset = projected_offset;
1597 
1598 		V_isn_offset_old = V_isn_offset;
1599 		CURVNET_RESTORE();
1600 	}
1601 	ISN_UNLOCK();
1602 	VNET_LIST_RUNLOCK();
1603 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1604 }
1605 
1606 /*
1607  * When a specific ICMP unreachable message is received and the
1608  * connection state is SYN-SENT, drop the connection.  This behavior
1609  * is controlled by the icmp_may_rst sysctl.
1610  */
1611 struct inpcb *
1612 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1613 {
1614 #ifdef INVARIANTS
1615 	INIT_VNET_INET(inp->inp_vnet);
1616 #endif
1617 	struct tcpcb *tp;
1618 
1619 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1620 	INP_WLOCK_ASSERT(inp);
1621 
1622 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1623 	    (inp->inp_flags & INP_DROPPED))
1624 		return (inp);
1625 
1626 	tp = intotcpcb(inp);
1627 	if (tp->t_state != TCPS_SYN_SENT)
1628 		return (inp);
1629 
1630 	tp = tcp_drop(tp, errno);
1631 	if (tp != NULL)
1632 		return (inp);
1633 	else
1634 		return (NULL);
1635 }
1636 
1637 /*
1638  * When `need fragmentation' ICMP is received, update our idea of the MSS
1639  * based on the new value in the route.  Also nudge TCP to send something,
1640  * since we know the packet we just sent was dropped.
1641  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1642  */
1643 struct inpcb *
1644 tcp_mtudisc(struct inpcb *inp, int errno)
1645 {
1646 	INIT_VNET_INET(inp->inp_vnet);
1647 	struct tcpcb *tp;
1648 	struct socket *so;
1649 
1650 	INP_WLOCK_ASSERT(inp);
1651 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1652 	    (inp->inp_flags & INP_DROPPED))
1653 		return (inp);
1654 
1655 	tp = intotcpcb(inp);
1656 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1657 
1658 	tcp_mss_update(tp, -1, NULL, NULL);
1659 
1660 	so = inp->inp_socket;
1661 	SOCKBUF_LOCK(&so->so_snd);
1662 	/* If the mss is larger than the socket buffer, decrease the mss. */
1663 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1664 		tp->t_maxseg = so->so_snd.sb_hiwat;
1665 	SOCKBUF_UNLOCK(&so->so_snd);
1666 
1667 	TCPSTAT_INC(tcps_mturesent);
1668 	tp->t_rtttime = 0;
1669 	tp->snd_nxt = tp->snd_una;
1670 	tcp_free_sackholes(tp);
1671 	tp->snd_recover = tp->snd_max;
1672 	if (tp->t_flags & TF_SACK_PERMIT)
1673 		EXIT_FASTRECOVERY(tp);
1674 	tcp_output_send(tp);
1675 	return (inp);
1676 }
1677 
1678 /*
1679  * Look-up the routing entry to the peer of this inpcb.  If no route
1680  * is found and it cannot be allocated, then return 0.  This routine
1681  * is called by TCP routines that access the rmx structure and by
1682  * tcp_mss_update to get the peer/interface MTU.
1683  */
1684 u_long
1685 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1686 {
1687 	struct route sro;
1688 	struct sockaddr_in *dst;
1689 	struct ifnet *ifp;
1690 	u_long maxmtu = 0;
1691 
1692 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1693 
1694 	bzero(&sro, sizeof(sro));
1695 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1696 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1697 		dst->sin_family = AF_INET;
1698 		dst->sin_len = sizeof(*dst);
1699 		dst->sin_addr = inc->inc_faddr;
1700 		in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1701 	}
1702 	if (sro.ro_rt != NULL) {
1703 		ifp = sro.ro_rt->rt_ifp;
1704 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1705 			maxmtu = ifp->if_mtu;
1706 		else
1707 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1708 
1709 		/* Report additional interface capabilities. */
1710 		if (flags != NULL) {
1711 			if (ifp->if_capenable & IFCAP_TSO4 &&
1712 			    ifp->if_hwassist & CSUM_TSO)
1713 				*flags |= CSUM_TSO;
1714 		}
1715 		RTFREE(sro.ro_rt);
1716 	}
1717 	return (maxmtu);
1718 }
1719 
1720 #ifdef INET6
1721 u_long
1722 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1723 {
1724 	struct route_in6 sro6;
1725 	struct ifnet *ifp;
1726 	u_long maxmtu = 0;
1727 
1728 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1729 
1730 	bzero(&sro6, sizeof(sro6));
1731 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1732 		sro6.ro_dst.sin6_family = AF_INET6;
1733 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1734 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1735 		rtalloc_ign((struct route *)&sro6, 0);
1736 	}
1737 	if (sro6.ro_rt != NULL) {
1738 		ifp = sro6.ro_rt->rt_ifp;
1739 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1740 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1741 		else
1742 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1743 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1744 
1745 		/* Report additional interface capabilities. */
1746 		if (flags != NULL) {
1747 			if (ifp->if_capenable & IFCAP_TSO6 &&
1748 			    ifp->if_hwassist & CSUM_TSO)
1749 				*flags |= CSUM_TSO;
1750 		}
1751 		RTFREE(sro6.ro_rt);
1752 	}
1753 
1754 	return (maxmtu);
1755 }
1756 #endif /* INET6 */
1757 
1758 #ifdef IPSEC
1759 /* compute ESP/AH header size for TCP, including outer IP header. */
1760 size_t
1761 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1762 {
1763 	struct inpcb *inp;
1764 	struct mbuf *m;
1765 	size_t hdrsiz;
1766 	struct ip *ip;
1767 #ifdef INET6
1768 	struct ip6_hdr *ip6;
1769 #endif
1770 	struct tcphdr *th;
1771 
1772 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1773 		return (0);
1774 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1775 	if (!m)
1776 		return (0);
1777 
1778 #ifdef INET6
1779 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1780 		ip6 = mtod(m, struct ip6_hdr *);
1781 		th = (struct tcphdr *)(ip6 + 1);
1782 		m->m_pkthdr.len = m->m_len =
1783 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1784 		tcpip_fillheaders(inp, ip6, th);
1785 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1786 	} else
1787 #endif /* INET6 */
1788 	{
1789 		ip = mtod(m, struct ip *);
1790 		th = (struct tcphdr *)(ip + 1);
1791 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1792 		tcpip_fillheaders(inp, ip, th);
1793 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1794 	}
1795 
1796 	m_free(m);
1797 	return (hdrsiz);
1798 }
1799 #endif /* IPSEC */
1800 
1801 /*
1802  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1803  *
1804  * This code attempts to calculate the bandwidth-delay product as a
1805  * means of determining the optimal window size to maximize bandwidth,
1806  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1807  * routers.  This code also does a fairly good job keeping RTTs in check
1808  * across slow links like modems.  We implement an algorithm which is very
1809  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1810  * transmitter side of a TCP connection and so only effects the transmit
1811  * side of the connection.
1812  *
1813  * BACKGROUND:  TCP makes no provision for the management of buffer space
1814  * at the end points or at the intermediate routers and switches.  A TCP
1815  * stream, whether using NewReno or not, will eventually buffer as
1816  * many packets as it is able and the only reason this typically works is
1817  * due to the fairly small default buffers made available for a connection
1818  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1819  * scaling it is now fairly easy for even a single TCP connection to blow-out
1820  * all available buffer space not only on the local interface, but on
1821  * intermediate routers and switches as well.  NewReno makes a misguided
1822  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1823  * then backing off, then steadily increasing the window again until another
1824  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1825  * is only made worse as network loads increase and the idea of intentionally
1826  * blowing out network buffers is, frankly, a terrible way to manage network
1827  * resources.
1828  *
1829  * It is far better to limit the transmit window prior to the failure
1830  * condition being achieved.  There are two general ways to do this:  First
1831  * you can 'scan' through different transmit window sizes and locate the
1832  * point where the RTT stops increasing, indicating that you have filled the
1833  * pipe, then scan backwards until you note that RTT stops decreasing, then
1834  * repeat ad-infinitum.  This method works in principle but has severe
1835  * implementation issues due to RTT variances, timer granularity, and
1836  * instability in the algorithm which can lead to many false positives and
1837  * create oscillations as well as interact badly with other TCP streams
1838  * implementing the same algorithm.
1839  *
1840  * The second method is to limit the window to the bandwidth delay product
1841  * of the link.  This is the method we implement.  RTT variances and our
1842  * own manipulation of the congestion window, bwnd, can potentially
1843  * destabilize the algorithm.  For this reason we have to stabilize the
1844  * elements used to calculate the window.  We do this by using the minimum
1845  * observed RTT, the long term average of the observed bandwidth, and
1846  * by adding two segments worth of slop.  It isn't perfect but it is able
1847  * to react to changing conditions and gives us a very stable basis on
1848  * which to extend the algorithm.
1849  */
1850 void
1851 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1852 {
1853 	INIT_VNET_INET(tp->t_vnet);
1854 	u_long bw;
1855 	u_long bwnd;
1856 	int save_ticks;
1857 
1858 	INP_WLOCK_ASSERT(tp->t_inpcb);
1859 
1860 	/*
1861 	 * If inflight_enable is disabled in the middle of a tcp connection,
1862 	 * make sure snd_bwnd is effectively disabled.
1863 	 */
1864 	if (V_tcp_inflight_enable == 0 ||
1865 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1866 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1867 		tp->snd_bandwidth = 0;
1868 		return;
1869 	}
1870 
1871 	/*
1872 	 * Figure out the bandwidth.  Due to the tick granularity this
1873 	 * is a very rough number and it MUST be averaged over a fairly
1874 	 * long period of time.  XXX we need to take into account a link
1875 	 * that is not using all available bandwidth, but for now our
1876 	 * slop will ramp us up if this case occurs and the bandwidth later
1877 	 * increases.
1878 	 *
1879 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1880 	 * effectively reset t_bw_rtttime for this case.
1881 	 */
1882 	save_ticks = ticks;
1883 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1884 		return;
1885 
1886 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1887 	    (save_ticks - tp->t_bw_rtttime);
1888 	tp->t_bw_rtttime = save_ticks;
1889 	tp->t_bw_rtseq = ack_seq;
1890 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1891 		return;
1892 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1893 
1894 	tp->snd_bandwidth = bw;
1895 
1896 	/*
1897 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1898 	 * segments.  The additional slop puts us squarely in the sweet
1899 	 * spot and also handles the bandwidth run-up case and stabilization.
1900 	 * Without the slop we could be locking ourselves into a lower
1901 	 * bandwidth.
1902 	 *
1903 	 * Situations Handled:
1904 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1905 	 *	    high speed LANs, allowing larger TCP buffers to be
1906 	 *	    specified, and also does a good job preventing
1907 	 *	    over-queueing of packets over choke points like modems
1908 	 *	    (at least for the transmit side).
1909 	 *
1910 	 *	(2) Is able to handle changing network loads (bandwidth
1911 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1912 	 *	    increases).
1913 	 *
1914 	 *	(3) Theoretically should stabilize in the face of multiple
1915 	 *	    connections implementing the same algorithm (this may need
1916 	 *	    a little work).
1917 	 *
1918 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1919 	 *	    be adjusted with a sysctl but typically only needs to be
1920 	 *	    on very slow connections.  A value no smaller then 5
1921 	 *	    should be used, but only reduce this default if you have
1922 	 *	    no other choice.
1923 	 */
1924 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1925 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1926 #undef USERTT
1927 
1928 	if (tcp_inflight_debug > 0) {
1929 		static int ltime;
1930 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1931 			ltime = ticks;
1932 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1933 			    tp,
1934 			    bw,
1935 			    tp->t_rttbest,
1936 			    tp->t_srtt,
1937 			    bwnd
1938 			);
1939 		}
1940 	}
1941 	if ((long)bwnd < V_tcp_inflight_min)
1942 		bwnd = V_tcp_inflight_min;
1943 	if (bwnd > V_tcp_inflight_max)
1944 		bwnd = V_tcp_inflight_max;
1945 	if ((long)bwnd < tp->t_maxseg * 2)
1946 		bwnd = tp->t_maxseg * 2;
1947 	tp->snd_bwnd = bwnd;
1948 }
1949 
1950 #ifdef TCP_SIGNATURE
1951 /*
1952  * Callback function invoked by m_apply() to digest TCP segment data
1953  * contained within an mbuf chain.
1954  */
1955 static int
1956 tcp_signature_apply(void *fstate, void *data, u_int len)
1957 {
1958 
1959 	MD5Update(fstate, (u_char *)data, len);
1960 	return (0);
1961 }
1962 
1963 /*
1964  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1965  *
1966  * Parameters:
1967  * m		pointer to head of mbuf chain
1968  * _unused
1969  * len		length of TCP segment data, excluding options
1970  * optlen	length of TCP segment options
1971  * buf		pointer to storage for computed MD5 digest
1972  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1973  *
1974  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1975  * When called from tcp_input(), we can be sure that th_sum has been
1976  * zeroed out and verified already.
1977  *
1978  * Return 0 if successful, otherwise return -1.
1979  *
1980  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1981  * search with the destination IP address, and a 'magic SPI' to be
1982  * determined by the application. This is hardcoded elsewhere to 1179
1983  * right now. Another branch of this code exists which uses the SPD to
1984  * specify per-application flows but it is unstable.
1985  */
1986 int
1987 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1988     u_char *buf, u_int direction)
1989 {
1990 	INIT_VNET_IPSEC(curvnet);
1991 	union sockaddr_union dst;
1992 	struct ippseudo ippseudo;
1993 	MD5_CTX ctx;
1994 	int doff;
1995 	struct ip *ip;
1996 	struct ipovly *ipovly;
1997 	struct secasvar *sav;
1998 	struct tcphdr *th;
1999 #ifdef INET6
2000 	struct ip6_hdr *ip6;
2001 	struct in6_addr in6;
2002 	char ip6buf[INET6_ADDRSTRLEN];
2003 	uint32_t plen;
2004 	uint16_t nhdr;
2005 #endif
2006 	u_short savecsum;
2007 
2008 	KASSERT(m != NULL, ("NULL mbuf chain"));
2009 	KASSERT(buf != NULL, ("NULL signature pointer"));
2010 
2011 	/* Extract the destination from the IP header in the mbuf. */
2012 	bzero(&dst, sizeof(union sockaddr_union));
2013 	ip = mtod(m, struct ip *);
2014 #ifdef INET6
2015 	ip6 = NULL;	/* Make the compiler happy. */
2016 #endif
2017 	switch (ip->ip_v) {
2018 	case IPVERSION:
2019 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2020 		dst.sa.sa_family = AF_INET;
2021 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2022 		    ip->ip_src : ip->ip_dst;
2023 		break;
2024 #ifdef INET6
2025 	case (IPV6_VERSION >> 4):
2026 		ip6 = mtod(m, struct ip6_hdr *);
2027 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2028 		dst.sa.sa_family = AF_INET6;
2029 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2030 		    ip6->ip6_src : ip6->ip6_dst;
2031 		break;
2032 #endif
2033 	default:
2034 		return (EINVAL);
2035 		/* NOTREACHED */
2036 		break;
2037 	}
2038 
2039 	/* Look up an SADB entry which matches the address of the peer. */
2040 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2041 	if (sav == NULL) {
2042 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2043 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2044 #ifdef INET6
2045 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2046 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2047 #endif
2048 			"(unsupported)"));
2049 		return (EINVAL);
2050 	}
2051 
2052 	MD5Init(&ctx);
2053 	/*
2054 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2055 	 *
2056 	 * XXX The ippseudo header MUST be digested in network byte order,
2057 	 * or else we'll fail the regression test. Assume all fields we've
2058 	 * been doing arithmetic on have been in host byte order.
2059 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2060 	 * tcp_output(), the underlying ip_len member has not yet been set.
2061 	 */
2062 	switch (ip->ip_v) {
2063 	case IPVERSION:
2064 		ipovly = (struct ipovly *)ip;
2065 		ippseudo.ippseudo_src = ipovly->ih_src;
2066 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2067 		ippseudo.ippseudo_pad = 0;
2068 		ippseudo.ippseudo_p = IPPROTO_TCP;
2069 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2070 		    optlen);
2071 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2072 
2073 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2074 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2075 		break;
2076 #ifdef INET6
2077 	/*
2078 	 * RFC 2385, 2.0  Proposal
2079 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2080 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2081 	 * extended next header value (to form 32 bits), and 32-bit segment
2082 	 * length.
2083 	 * Note: Upper-Layer Packet Length comes before Next Header.
2084 	 */
2085 	case (IPV6_VERSION >> 4):
2086 		in6 = ip6->ip6_src;
2087 		in6_clearscope(&in6);
2088 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2089 		in6 = ip6->ip6_dst;
2090 		in6_clearscope(&in6);
2091 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2092 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2093 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2094 		nhdr = 0;
2095 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2096 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2097 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2098 		nhdr = IPPROTO_TCP;
2099 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2100 
2101 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2102 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2103 		break;
2104 #endif
2105 	default:
2106 		return (EINVAL);
2107 		/* NOTREACHED */
2108 		break;
2109 	}
2110 
2111 
2112 	/*
2113 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2114 	 * The TCP checksum must be set to zero.
2115 	 */
2116 	savecsum = th->th_sum;
2117 	th->th_sum = 0;
2118 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2119 	th->th_sum = savecsum;
2120 
2121 	/*
2122 	 * Step 3: Update MD5 hash with TCP segment data.
2123 	 *         Use m_apply() to avoid an early m_pullup().
2124 	 */
2125 	if (len > 0)
2126 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2127 
2128 	/*
2129 	 * Step 4: Update MD5 hash with shared secret.
2130 	 */
2131 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2132 	MD5Final(buf, &ctx);
2133 
2134 	key_sa_recordxfer(sav, m);
2135 	KEY_FREESAV(&sav);
2136 	return (0);
2137 }
2138 #endif /* TCP_SIGNATURE */
2139 
2140 static int
2141 sysctl_drop(SYSCTL_HANDLER_ARGS)
2142 {
2143 	INIT_VNET_INET(curvnet);
2144 #ifdef INET6
2145 	INIT_VNET_INET6(curvnet);
2146 #endif
2147 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2148 	struct sockaddr_storage addrs[2];
2149 	struct inpcb *inp;
2150 	struct tcpcb *tp;
2151 	struct tcptw *tw;
2152 	struct sockaddr_in *fin, *lin;
2153 #ifdef INET6
2154 	struct sockaddr_in6 *fin6, *lin6;
2155 #endif
2156 	int error;
2157 
2158 	inp = NULL;
2159 	fin = lin = NULL;
2160 #ifdef INET6
2161 	fin6 = lin6 = NULL;
2162 #endif
2163 	error = 0;
2164 
2165 	if (req->oldptr != NULL || req->oldlen != 0)
2166 		return (EINVAL);
2167 	if (req->newptr == NULL)
2168 		return (EPERM);
2169 	if (req->newlen < sizeof(addrs))
2170 		return (ENOMEM);
2171 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2172 	if (error)
2173 		return (error);
2174 
2175 	switch (addrs[0].ss_family) {
2176 #ifdef INET6
2177 	case AF_INET6:
2178 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2179 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2180 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2181 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2182 			return (EINVAL);
2183 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2184 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2185 				return (EINVAL);
2186 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2187 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2188 			fin = (struct sockaddr_in *)&addrs[0];
2189 			lin = (struct sockaddr_in *)&addrs[1];
2190 			break;
2191 		}
2192 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2193 		if (error)
2194 			return (error);
2195 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2196 		if (error)
2197 			return (error);
2198 		break;
2199 #endif
2200 	case AF_INET:
2201 		fin = (struct sockaddr_in *)&addrs[0];
2202 		lin = (struct sockaddr_in *)&addrs[1];
2203 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2204 		    lin->sin_len != sizeof(struct sockaddr_in))
2205 			return (EINVAL);
2206 		break;
2207 	default:
2208 		return (EINVAL);
2209 	}
2210 	INP_INFO_WLOCK(&V_tcbinfo);
2211 	switch (addrs[0].ss_family) {
2212 #ifdef INET6
2213 	case AF_INET6:
2214 		inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2215 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2216 		    NULL);
2217 		break;
2218 #endif
2219 	case AF_INET:
2220 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2221 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2222 		break;
2223 	}
2224 	if (inp != NULL) {
2225 		INP_WLOCK(inp);
2226 		if (inp->inp_flags & INP_TIMEWAIT) {
2227 			/*
2228 			 * XXXRW: There currently exists a state where an
2229 			 * inpcb is present, but its timewait state has been
2230 			 * discarded.  For now, don't allow dropping of this
2231 			 * type of inpcb.
2232 			 */
2233 			tw = intotw(inp);
2234 			if (tw != NULL)
2235 				tcp_twclose(tw, 0);
2236 			else
2237 				INP_WUNLOCK(inp);
2238 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2239 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2240 			tp = intotcpcb(inp);
2241 			tp = tcp_drop(tp, ECONNABORTED);
2242 			if (tp != NULL)
2243 				INP_WUNLOCK(inp);
2244 		} else
2245 			INP_WUNLOCK(inp);
2246 	} else
2247 		error = ESRCH;
2248 	INP_INFO_WUNLOCK(&V_tcbinfo);
2249 	return (error);
2250 }
2251 
2252 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2253     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2254     0, sysctl_drop, "", "Drop TCP connection");
2255 
2256 /*
2257  * Generate a standardized TCP log line for use throughout the
2258  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2259  * allow use in the interrupt context.
2260  *
2261  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2262  * NB: The function may return NULL if memory allocation failed.
2263  *
2264  * Due to header inclusion and ordering limitations the struct ip
2265  * and ip6_hdr pointers have to be passed as void pointers.
2266  */
2267 char *
2268 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2269     const void *ip6hdr)
2270 {
2271 	char *s, *sp;
2272 	size_t size;
2273 	struct ip *ip;
2274 #ifdef INET6
2275 	const struct ip6_hdr *ip6;
2276 
2277 	ip6 = (const struct ip6_hdr *)ip6hdr;
2278 #endif /* INET6 */
2279 	ip = (struct ip *)ip4hdr;
2280 
2281 	/*
2282 	 * The log line looks like this:
2283 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2284 	 */
2285 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2286 	    sizeof(PRINT_TH_FLAGS) + 1 +
2287 #ifdef INET6
2288 	    2 * INET6_ADDRSTRLEN;
2289 #else
2290 	    2 * INET_ADDRSTRLEN;
2291 #endif /* INET6 */
2292 
2293 	/* Is logging enabled? */
2294 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2295 		return (NULL);
2296 
2297 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2298 	if (s == NULL)
2299 		return (NULL);
2300 
2301 	strcat(s, "TCP: [");
2302 	sp = s + strlen(s);
2303 
2304 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2305 		inet_ntoa_r(inc->inc_faddr, sp);
2306 		sp = s + strlen(s);
2307 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2308 		sp = s + strlen(s);
2309 		inet_ntoa_r(inc->inc_laddr, sp);
2310 		sp = s + strlen(s);
2311 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2312 #ifdef INET6
2313 	} else if (inc) {
2314 		ip6_sprintf(sp, &inc->inc6_faddr);
2315 		sp = s + strlen(s);
2316 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2317 		sp = s + strlen(s);
2318 		ip6_sprintf(sp, &inc->inc6_laddr);
2319 		sp = s + strlen(s);
2320 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2321 	} else if (ip6 && th) {
2322 		ip6_sprintf(sp, &ip6->ip6_src);
2323 		sp = s + strlen(s);
2324 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2325 		sp = s + strlen(s);
2326 		ip6_sprintf(sp, &ip6->ip6_dst);
2327 		sp = s + strlen(s);
2328 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2329 #endif /* INET6 */
2330 	} else if (ip && th) {
2331 		inet_ntoa_r(ip->ip_src, sp);
2332 		sp = s + strlen(s);
2333 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2334 		sp = s + strlen(s);
2335 		inet_ntoa_r(ip->ip_dst, sp);
2336 		sp = s + strlen(s);
2337 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2338 	} else {
2339 		free(s, M_TCPLOG);
2340 		return (NULL);
2341 	}
2342 	sp = s + strlen(s);
2343 	if (th)
2344 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2345 	if (*(s + size - 1) != '\0')
2346 		panic("%s: string too long", __func__);
2347 	return (s);
2348 }
2349