xref: /freebsd/sys/netinet/tcp_subr.c (revision 33644623554bb0fc57ed3c7d874193a498679b22)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58 #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 <netinet6/ip6protosw.h>
98 
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
102 #ifdef INET6
103 #include <netipsec/ipsec6.h>
104 #endif
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
107 #endif /*IPSEC*/
108 
109 #include <machine/in_cksum.h>
110 #include <sys/md5.h>
111 
112 #include <security/mac/mac_framework.h>
113 
114 #ifdef VIMAGE_GLOBALS
115 int	tcp_mssdflt;
116 #ifdef INET6
117 int	tcp_v6mssdflt;
118 #endif
119 int	tcp_minmss;
120 int	tcp_do_rfc1323;
121 static int	icmp_may_rst;
122 static int	tcp_isn_reseed_interval;
123 static int	tcp_inflight_enable;
124 static int	tcp_inflight_rttthresh;
125 static int	tcp_inflight_min;
126 static int	tcp_inflight_max;
127 static int	tcp_inflight_stab;
128 #endif
129 
130 static int
131 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
132 {
133 	INIT_VNET_INET(curvnet);
134 	int error, new;
135 
136 	new = V_tcp_mssdflt;
137 	error = sysctl_handle_int(oidp, &new, 0, req);
138 	if (error == 0 && req->newptr) {
139 		if (new < TCP_MINMSS)
140 			error = EINVAL;
141 		else
142 			V_tcp_mssdflt = new;
143 	}
144 	return (error);
145 }
146 
147 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
148     CTLTYPE_INT|CTLFLAG_RW, tcp_mssdflt, 0,
149     &sysctl_net_inet_tcp_mss_check, "I",
150     "Default TCP Maximum Segment Size");
151 
152 #ifdef INET6
153 static int
154 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
155 {
156 	INIT_VNET_INET6(curvnet);
157 	int error, new;
158 
159 	new = V_tcp_v6mssdflt;
160 	error = sysctl_handle_int(oidp, &new, 0, req);
161 	if (error == 0 && req->newptr) {
162 		if (new < TCP_MINMSS)
163 			error = EINVAL;
164 		else
165 			V_tcp_v6mssdflt = new;
166 	}
167 	return (error);
168 }
169 
170 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
171     CTLTYPE_INT|CTLFLAG_RW, tcp_v6mssdflt, 0,
172     &sysctl_net_inet_tcp_mss_v6_check, "I",
173    "Default TCP Maximum Segment Size for IPv6");
174 #endif
175 
176 /*
177  * Minimum MSS we accept and use. This prevents DoS attacks where
178  * we are forced to a ridiculous low MSS like 20 and send hundreds
179  * of packets instead of one. The effect scales with the available
180  * bandwidth and quickly saturates the CPU and network interface
181  * with packet generation and sending. Set to zero to disable MINMSS
182  * checking. This setting prevents us from sending too small packets.
183  */
184 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss,
185     CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
186 
187 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
188     CTLFLAG_RW, tcp_do_rfc1323, 0,
189     "Enable rfc1323 (high performance TCP) extensions");
190 
191 static int	tcp_log_debug = 0;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
193     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
194 
195 static int	tcp_tcbhashsize = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
197     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
198 
199 static int	do_tcpdrain = 1;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
201     "Enable tcp_drain routine for extra help when low on mbufs");
202 
203 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount,
204     CTLFLAG_RD, V_tcbinfo.ipi_count, 0, "Number of active PCBs");
205 
206 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst,
207     CTLFLAG_RW, icmp_may_rst, 0,
208     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
209 
210 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval,
211     CTLFLAG_RW, tcp_isn_reseed_interval, 0,
212     "Seconds between reseeding of ISN secret");
213 
214 /*
215  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
216  * 1024 exists only for debugging.  A good production default would be
217  * something like 6100.
218  */
219 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
220     "TCP inflight data limiting");
221 
222 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable,
223     CTLFLAG_RW, tcp_inflight_enable, 0,
224     "Enable automatic TCP inflight data limiting");
225 
226 static int	tcp_inflight_debug = 0;
227 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
228     &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
229 
230 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, rttthresh,
231     CTLTYPE_INT|CTLFLAG_RW, tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks,
232     "I", "RTT threshold below which inflight will deactivate itself");
233 
234 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min,
235     CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
236 
237 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max,
238     CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
239 
240 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab,
241     CTLFLAG_RW, tcp_inflight_stab, 0,
242     "Inflight Algorithm Stabilization 20 = 2 packets");
243 
244 uma_zone_t sack_hole_zone;
245 
246 static struct inpcb *tcp_notify(struct inpcb *, int);
247 static void	tcp_isn_tick(void *);
248 
249 /*
250  * Target size of TCP PCB hash tables. Must be a power of two.
251  *
252  * Note that this can be overridden by the kernel environment
253  * variable net.inet.tcp.tcbhashsize
254  */
255 #ifndef TCBHASHSIZE
256 #define TCBHASHSIZE	512
257 #endif
258 
259 /*
260  * XXX
261  * Callouts should be moved into struct tcp directly.  They are currently
262  * separate because the tcpcb structure is exported to userland for sysctl
263  * parsing purposes, which do not know about callouts.
264  */
265 struct tcpcb_mem {
266 	struct	tcpcb		tcb;
267 	struct	tcp_timer	tt;
268 };
269 
270 static uma_zone_t tcpcb_zone;
271 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
272 struct callout isn_callout;
273 static struct mtx isn_mtx;
274 
275 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
276 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
277 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
278 
279 /*
280  * TCP initialization.
281  */
282 static void
283 tcp_zone_change(void *tag)
284 {
285 
286 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
287 	uma_zone_set_max(tcpcb_zone, maxsockets);
288 	tcp_tw_zone_change();
289 }
290 
291 static int
292 tcp_inpcb_init(void *mem, int size, int flags)
293 {
294 	struct inpcb *inp = mem;
295 
296 	INP_LOCK_INIT(inp, "inp", "tcpinp");
297 	return (0);
298 }
299 
300 void
301 tcp_init(void)
302 {
303 	INIT_VNET_INET(curvnet);
304 	int hashsize;
305 
306 	V_blackhole = 0;
307 	V_tcp_delack_enabled = 1;
308 	V_drop_synfin = 0;
309 	V_tcp_do_rfc3042 = 1;
310 	V_tcp_do_rfc3390 = 1;
311 	V_tcp_do_ecn = 0;
312 	V_tcp_ecn_maxretries = 1;
313 	V_tcp_insecure_rst = 0;
314 	V_tcp_do_autorcvbuf = 1;
315 	V_tcp_autorcvbuf_inc = 16*1024;
316 	V_tcp_autorcvbuf_max = 256*1024;
317 
318 	V_tcp_mssdflt = TCP_MSS;
319 #ifdef INET6
320 	V_tcp_v6mssdflt = TCP6_MSS;
321 #endif
322 	V_tcp_minmss = TCP_MINMSS;
323 	V_tcp_do_rfc1323 = 1;
324 	V_icmp_may_rst = 1;
325 	V_tcp_isn_reseed_interval = 0;
326 	V_tcp_inflight_enable = 1;
327 	V_tcp_inflight_min = 6144;
328 	V_tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
329 	V_tcp_inflight_stab = 20;
330 
331 	V_path_mtu_discovery = 1;
332 	V_ss_fltsz = 1;
333 	V_ss_fltsz_local = 4;
334 	V_tcp_do_newreno = 1;
335 	V_tcp_do_tso = 1;
336 	V_tcp_do_autosndbuf = 1;
337 	V_tcp_autosndbuf_inc = 8*1024;
338 	V_tcp_autosndbuf_max = 256*1024;
339 
340 	V_nolocaltimewait = 0;
341 
342 	V_tcp_do_sack = 1;
343 	V_tcp_sack_maxholes = 128;
344 	V_tcp_sack_globalmaxholes = 65536;
345 	V_tcp_sack_globalholes = 0;
346 
347 	tcp_delacktime = TCPTV_DELACK;
348 	tcp_keepinit = TCPTV_KEEP_INIT;
349 	tcp_keepidle = TCPTV_KEEP_IDLE;
350 	tcp_keepintvl = TCPTV_KEEPINTVL;
351 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
352 	tcp_msl = TCPTV_MSL;
353 	tcp_rexmit_min = TCPTV_MIN;
354 	if (tcp_rexmit_min < 1)
355 		tcp_rexmit_min = 1;
356 	tcp_rexmit_slop = TCPTV_CPU_VAR;
357 	V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
358 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
359 
360 	INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
361 	LIST_INIT(&V_tcb);
362 	V_tcbinfo.ipi_listhead = &V_tcb;
363 	hashsize = TCBHASHSIZE;
364 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
365 	if (!powerof2(hashsize)) {
366 		printf("WARNING: TCB hash size not a power of 2\n");
367 		hashsize = 512; /* safe default */
368 	}
369 	tcp_tcbhashsize = hashsize;
370 	V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
371 	    &V_tcbinfo.ipi_hashmask);
372 	V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
373 	    &V_tcbinfo.ipi_porthashmask);
374 	V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
375 	    NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
376 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
377 #ifdef INET6
378 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
379 #else /* INET6 */
380 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
381 #endif /* INET6 */
382 	if (max_protohdr < TCP_MINPROTOHDR)
383 		max_protohdr = TCP_MINPROTOHDR;
384 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
385 		panic("tcp_init");
386 #undef TCP_MINPROTOHDR
387 	/*
388 	 * These have to be type stable for the benefit of the timers.
389 	 */
390 	tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
391 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
392 	uma_zone_set_max(tcpcb_zone, maxsockets);
393 	tcp_tw_init();
394 	syncache_init();
395 	tcp_hc_init();
396 	tcp_reass_init();
397 	ISN_LOCK_INIT();
398 	callout_init(&isn_callout, CALLOUT_MPSAFE);
399 	tcp_isn_tick(NULL);
400 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
401 		SHUTDOWN_PRI_DEFAULT);
402 	sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
403 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
404 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
405 		EVENTHANDLER_PRI_ANY);
406 }
407 
408 void
409 tcp_fini(void *xtp)
410 {
411 
412 	callout_stop(&isn_callout);
413 }
414 
415 /*
416  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
417  * tcp_template used to store this data in mbufs, but we now recopy it out
418  * of the tcpcb each time to conserve mbufs.
419  */
420 void
421 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
422 {
423 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
424 
425 	INP_WLOCK_ASSERT(inp);
426 
427 #ifdef INET6
428 	if ((inp->inp_vflag & INP_IPV6) != 0) {
429 		struct ip6_hdr *ip6;
430 
431 		ip6 = (struct ip6_hdr *)ip_ptr;
432 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
433 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
434 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
435 			(IPV6_VERSION & IPV6_VERSION_MASK);
436 		ip6->ip6_nxt = IPPROTO_TCP;
437 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
438 		ip6->ip6_src = inp->in6p_laddr;
439 		ip6->ip6_dst = inp->in6p_faddr;
440 	} else
441 #endif
442 	{
443 		struct ip *ip;
444 
445 		ip = (struct ip *)ip_ptr;
446 		ip->ip_v = IPVERSION;
447 		ip->ip_hl = 5;
448 		ip->ip_tos = inp->inp_ip_tos;
449 		ip->ip_len = 0;
450 		ip->ip_id = 0;
451 		ip->ip_off = 0;
452 		ip->ip_ttl = inp->inp_ip_ttl;
453 		ip->ip_sum = 0;
454 		ip->ip_p = IPPROTO_TCP;
455 		ip->ip_src = inp->inp_laddr;
456 		ip->ip_dst = inp->inp_faddr;
457 	}
458 	th->th_sport = inp->inp_lport;
459 	th->th_dport = inp->inp_fport;
460 	th->th_seq = 0;
461 	th->th_ack = 0;
462 	th->th_x2 = 0;
463 	th->th_off = 5;
464 	th->th_flags = 0;
465 	th->th_win = 0;
466 	th->th_urp = 0;
467 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
468 }
469 
470 /*
471  * Create template to be used to send tcp packets on a connection.
472  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
473  * use for this function is in keepalives, which use tcp_respond.
474  */
475 struct tcptemp *
476 tcpip_maketemplate(struct inpcb *inp)
477 {
478 	struct tcptemp *t;
479 
480 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
481 	if (t == NULL)
482 		return (NULL);
483 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
484 	return (t);
485 }
486 
487 /*
488  * Send a single message to the TCP at address specified by
489  * the given TCP/IP header.  If m == NULL, then we make a copy
490  * of the tcpiphdr at ti and send directly to the addressed host.
491  * This is used to force keep alive messages out using the TCP
492  * template for a connection.  If flags are given then we send
493  * a message back to the TCP which originated the * segment ti,
494  * and discard the mbuf containing it and any other attached mbufs.
495  *
496  * In any case the ack and sequence number of the transmitted
497  * segment are as specified by the parameters.
498  *
499  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
500  */
501 void
502 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
503     tcp_seq ack, tcp_seq seq, int flags)
504 {
505 	INIT_VNET_INET(curvnet);
506 	int tlen;
507 	int win = 0;
508 	struct ip *ip;
509 	struct tcphdr *nth;
510 #ifdef INET6
511 	struct ip6_hdr *ip6;
512 	int isipv6;
513 #endif /* INET6 */
514 	int ipflags = 0;
515 	struct inpcb *inp;
516 
517 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
518 
519 #ifdef INET6
520 	isipv6 = ((struct ip *)ipgen)->ip_v == 6;
521 	ip6 = ipgen;
522 #endif /* INET6 */
523 	ip = ipgen;
524 
525 	if (tp != NULL) {
526 		inp = tp->t_inpcb;
527 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
528 		INP_WLOCK_ASSERT(inp);
529 	} else
530 		inp = NULL;
531 
532 	if (tp != NULL) {
533 		if (!(flags & TH_RST)) {
534 			win = sbspace(&inp->inp_socket->so_rcv);
535 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
536 				win = (long)TCP_MAXWIN << tp->rcv_scale;
537 		}
538 	}
539 	if (m == NULL) {
540 		m = m_gethdr(M_DONTWAIT, MT_DATA);
541 		if (m == NULL)
542 			return;
543 		tlen = 0;
544 		m->m_data += max_linkhdr;
545 #ifdef INET6
546 		if (isipv6) {
547 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
548 			      sizeof(struct ip6_hdr));
549 			ip6 = mtod(m, struct ip6_hdr *);
550 			nth = (struct tcphdr *)(ip6 + 1);
551 		} else
552 #endif /* INET6 */
553 	      {
554 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
555 		ip = mtod(m, struct ip *);
556 		nth = (struct tcphdr *)(ip + 1);
557 	      }
558 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
559 		flags = TH_ACK;
560 	} else {
561 		/*
562 		 *  reuse the mbuf.
563 		 * XXX MRT We inherrit the FIB, which is lucky.
564 		 */
565 		m_freem(m->m_next);
566 		m->m_next = NULL;
567 		m->m_data = (caddr_t)ipgen;
568 		/* m_len is set later */
569 		tlen = 0;
570 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
571 #ifdef INET6
572 		if (isipv6) {
573 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
574 			nth = (struct tcphdr *)(ip6 + 1);
575 		} else
576 #endif /* INET6 */
577 	      {
578 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
579 		nth = (struct tcphdr *)(ip + 1);
580 	      }
581 		if (th != nth) {
582 			/*
583 			 * this is usually a case when an extension header
584 			 * exists between the IPv6 header and the
585 			 * TCP header.
586 			 */
587 			nth->th_sport = th->th_sport;
588 			nth->th_dport = th->th_dport;
589 		}
590 		xchg(nth->th_dport, nth->th_sport, n_short);
591 #undef xchg
592 	}
593 #ifdef INET6
594 	if (isipv6) {
595 		ip6->ip6_flow = 0;
596 		ip6->ip6_vfc = IPV6_VERSION;
597 		ip6->ip6_nxt = IPPROTO_TCP;
598 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
599 						tlen));
600 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
601 	} else
602 #endif
603 	{
604 		tlen += sizeof (struct tcpiphdr);
605 		ip->ip_len = tlen;
606 		ip->ip_ttl = V_ip_defttl;
607 		if (V_path_mtu_discovery)
608 			ip->ip_off |= IP_DF;
609 	}
610 	m->m_len = tlen;
611 	m->m_pkthdr.len = tlen;
612 	m->m_pkthdr.rcvif = NULL;
613 #ifdef MAC
614 	if (inp != NULL) {
615 		/*
616 		 * Packet is associated with a socket, so allow the
617 		 * label of the response to reflect the socket label.
618 		 */
619 		INP_WLOCK_ASSERT(inp);
620 		mac_inpcb_create_mbuf(inp, m);
621 	} else {
622 		/*
623 		 * Packet is not associated with a socket, so possibly
624 		 * update the label in place.
625 		 */
626 		mac_netinet_tcp_reply(m);
627 	}
628 #endif
629 	nth->th_seq = htonl(seq);
630 	nth->th_ack = htonl(ack);
631 	nth->th_x2 = 0;
632 	nth->th_off = sizeof (struct tcphdr) >> 2;
633 	nth->th_flags = flags;
634 	if (tp != NULL)
635 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
636 	else
637 		nth->th_win = htons((u_short)win);
638 	nth->th_urp = 0;
639 #ifdef INET6
640 	if (isipv6) {
641 		nth->th_sum = 0;
642 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
643 					sizeof(struct ip6_hdr),
644 					tlen - sizeof(struct ip6_hdr));
645 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
646 		    NULL, NULL);
647 	} else
648 #endif /* INET6 */
649 	{
650 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
651 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
652 		m->m_pkthdr.csum_flags = CSUM_TCP;
653 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
654 	}
655 #ifdef TCPDEBUG
656 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
657 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
658 #endif
659 #ifdef INET6
660 	if (isipv6)
661 		(void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
662 	else
663 #endif /* INET6 */
664 	(void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
665 }
666 
667 /*
668  * Create a new TCP control block, making an
669  * empty reassembly queue and hooking it to the argument
670  * protocol control block.  The `inp' parameter must have
671  * come from the zone allocator set up in tcp_init().
672  */
673 struct tcpcb *
674 tcp_newtcpcb(struct inpcb *inp)
675 {
676 	INIT_VNET_INET(inp->inp_vnet);
677 	struct tcpcb_mem *tm;
678 	struct tcpcb *tp;
679 #ifdef INET6
680 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
681 #endif /* INET6 */
682 
683 	tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
684 	if (tm == NULL)
685 		return (NULL);
686 	tp = &tm->tcb;
687 	tp->t_timers = &tm->tt;
688 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
689 	tp->t_maxseg = tp->t_maxopd =
690 #ifdef INET6
691 		isipv6 ? V_tcp_v6mssdflt :
692 #endif /* INET6 */
693 		V_tcp_mssdflt;
694 
695 	/* Set up our timeouts. */
696 	callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
697 	callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
698 	callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
699 	callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
700 	callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
701 
702 	if (V_tcp_do_rfc1323)
703 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
704 	if (V_tcp_do_sack)
705 		tp->t_flags |= TF_SACK_PERMIT;
706 	TAILQ_INIT(&tp->snd_holes);
707 	tp->t_inpcb = inp;	/* XXX */
708 	/*
709 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
710 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
711 	 * reasonable initial retransmit time.
712 	 */
713 	tp->t_srtt = TCPTV_SRTTBASE;
714 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
715 	tp->t_rttmin = tcp_rexmit_min;
716 	tp->t_rxtcur = TCPTV_RTOBASE;
717 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
718 	tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
719 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
720 	tp->t_rcvtime = ticks;
721 	tp->t_bw_rtttime = ticks;
722 	/*
723 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
724 	 * because the socket may be bound to an IPv6 wildcard address,
725 	 * which may match an IPv4-mapped IPv6 address.
726 	 */
727 	inp->inp_ip_ttl = V_ip_defttl;
728 	inp->inp_ppcb = tp;
729 	return (tp);		/* XXX */
730 }
731 
732 /*
733  * Drop a TCP connection, reporting
734  * the specified error.  If connection is synchronized,
735  * then send a RST to peer.
736  */
737 struct tcpcb *
738 tcp_drop(struct tcpcb *tp, int errno)
739 {
740 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
741 	struct socket *so = tp->t_inpcb->inp_socket;
742 
743 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
744 	INP_WLOCK_ASSERT(tp->t_inpcb);
745 
746 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
747 		tp->t_state = TCPS_CLOSED;
748 		(void) tcp_output_reset(tp);
749 		V_tcpstat.tcps_drops++;
750 	} else
751 		V_tcpstat.tcps_conndrops++;
752 	if (errno == ETIMEDOUT && tp->t_softerror)
753 		errno = tp->t_softerror;
754 	so->so_error = errno;
755 	return (tcp_close(tp));
756 }
757 
758 void
759 tcp_discardcb(struct tcpcb *tp)
760 {
761 	INIT_VNET_INET(tp->t_vnet);
762 	struct tseg_qent *q;
763 	struct inpcb *inp = tp->t_inpcb;
764 	struct socket *so = inp->inp_socket;
765 #ifdef INET6
766 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
767 #endif /* INET6 */
768 
769 	INP_WLOCK_ASSERT(inp);
770 
771 	/*
772 	 * Make sure that all of our timers are stopped before we
773 	 * delete the PCB.
774 	 */
775 	callout_stop(&tp->t_timers->tt_rexmt);
776 	callout_stop(&tp->t_timers->tt_persist);
777 	callout_stop(&tp->t_timers->tt_keep);
778 	callout_stop(&tp->t_timers->tt_2msl);
779 	callout_stop(&tp->t_timers->tt_delack);
780 
781 	/*
782 	 * If we got enough samples through the srtt filter,
783 	 * save the rtt and rttvar in the routing entry.
784 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
785 	 * 4 samples is enough for the srtt filter to converge
786 	 * to within enough % of the correct value; fewer samples
787 	 * and we could save a bogus rtt. The danger is not high
788 	 * as tcp quickly recovers from everything.
789 	 * XXX: Works very well but needs some more statistics!
790 	 */
791 	if (tp->t_rttupdated >= 4) {
792 		struct hc_metrics_lite metrics;
793 		u_long ssthresh;
794 
795 		bzero(&metrics, sizeof(metrics));
796 		/*
797 		 * Update the ssthresh always when the conditions below
798 		 * are satisfied. This gives us better new start value
799 		 * for the congestion avoidance for new connections.
800 		 * ssthresh is only set if packet loss occured on a session.
801 		 *
802 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
803 		 * being torn down.  Ideally this code would not use 'so'.
804 		 */
805 		ssthresh = tp->snd_ssthresh;
806 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
807 			/*
808 			 * convert the limit from user data bytes to
809 			 * packets then to packet data bytes.
810 			 */
811 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
812 			if (ssthresh < 2)
813 				ssthresh = 2;
814 			ssthresh *= (u_long)(tp->t_maxseg +
815 #ifdef INET6
816 				      (isipv6 ? sizeof (struct ip6_hdr) +
817 					       sizeof (struct tcphdr) :
818 #endif
819 				       sizeof (struct tcpiphdr)
820 #ifdef INET6
821 				       )
822 #endif
823 				      );
824 		} else
825 			ssthresh = 0;
826 		metrics.rmx_ssthresh = ssthresh;
827 
828 		metrics.rmx_rtt = tp->t_srtt;
829 		metrics.rmx_rttvar = tp->t_rttvar;
830 		/* XXX: This wraps if the pipe is more than 4 Gbit per second */
831 		metrics.rmx_bandwidth = tp->snd_bandwidth;
832 		metrics.rmx_cwnd = tp->snd_cwnd;
833 		metrics.rmx_sendpipe = 0;
834 		metrics.rmx_recvpipe = 0;
835 
836 		tcp_hc_update(&inp->inp_inc, &metrics);
837 	}
838 
839 	/* free the reassembly queue, if any */
840 	while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
841 		LIST_REMOVE(q, tqe_q);
842 		m_freem(q->tqe_m);
843 		uma_zfree(tcp_reass_zone, q);
844 		tp->t_segqlen--;
845 		V_tcp_reass_qsize--;
846 	}
847 	/* Disconnect offload device, if any. */
848 	tcp_offload_detach(tp);
849 
850 	tcp_free_sackholes(tp);
851 	inp->inp_ppcb = NULL;
852 	tp->t_inpcb = NULL;
853 	uma_zfree(tcpcb_zone, tp);
854 }
855 
856 /*
857  * Attempt to close a TCP control block, marking it as dropped, and freeing
858  * the socket if we hold the only reference.
859  */
860 struct tcpcb *
861 tcp_close(struct tcpcb *tp)
862 {
863 	INIT_VNET_INET(tp->t_inpcb->inp_vnet);
864 	struct inpcb *inp = tp->t_inpcb;
865 	struct socket *so;
866 
867 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
868 	INP_WLOCK_ASSERT(inp);
869 
870 	/* Notify any offload devices of listener close */
871 	if (tp->t_state == TCPS_LISTEN)
872 		tcp_offload_listen_close(tp);
873 	in_pcbdrop(inp);
874 	V_tcpstat.tcps_closed++;
875 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
876 	so = inp->inp_socket;
877 	soisdisconnected(so);
878 	if (inp->inp_vflag & INP_SOCKREF) {
879 		KASSERT(so->so_state & SS_PROTOREF,
880 		    ("tcp_close: !SS_PROTOREF"));
881 		inp->inp_vflag &= ~INP_SOCKREF;
882 		INP_WUNLOCK(inp);
883 		ACCEPT_LOCK();
884 		SOCK_LOCK(so);
885 		so->so_state &= ~SS_PROTOREF;
886 		sofree(so);
887 		return (NULL);
888 	}
889 	return (tp);
890 }
891 
892 void
893 tcp_drain(void)
894 {
895 	VNET_ITERATOR_DECL(vnet_iter);
896 
897 	if (!do_tcpdrain)
898 		return;
899 
900 	VNET_LIST_RLOCK();
901 	VNET_FOREACH(vnet_iter) {
902 		CURVNET_SET(vnet_iter);
903 		INIT_VNET_INET(vnet_iter);
904 		struct inpcb *inpb;
905 		struct tcpcb *tcpb;
906 		struct tseg_qent *te;
907 
908 	/*
909 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
910 	 * if there is one...
911 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
912 	 *      reassembly queue should be flushed, but in a situation
913 	 *	where we're really low on mbufs, this is potentially
914 	 *	usefull.
915 	 */
916 		INP_INFO_RLOCK(&V_tcbinfo);
917 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
918 			if (inpb->inp_vflag & INP_TIMEWAIT)
919 				continue;
920 			INP_WLOCK(inpb);
921 			if ((tcpb = intotcpcb(inpb)) != NULL) {
922 				while ((te = LIST_FIRST(&tcpb->t_segq))
923 			            != NULL) {
924 					LIST_REMOVE(te, tqe_q);
925 					m_freem(te->tqe_m);
926 					uma_zfree(tcp_reass_zone, te);
927 					tcpb->t_segqlen--;
928 					V_tcp_reass_qsize--;
929 				}
930 				tcp_clean_sackreport(tcpb);
931 			}
932 			INP_WUNLOCK(inpb);
933 		}
934 		INP_INFO_RUNLOCK(&V_tcbinfo);
935 		CURVNET_RESTORE();
936 	}
937 	VNET_LIST_RUNLOCK();
938 }
939 
940 /*
941  * Notify a tcp user of an asynchronous error;
942  * store error as soft error, but wake up user
943  * (for now, won't do anything until can select for soft error).
944  *
945  * Do not wake up user since there currently is no mechanism for
946  * reporting soft errors (yet - a kqueue filter may be added).
947  */
948 static struct inpcb *
949 tcp_notify(struct inpcb *inp, int error)
950 {
951 	struct tcpcb *tp;
952 #ifdef INVARIANTS
953 	INIT_VNET_INET(inp->inp_vnet); /* V_tcbinfo WLOCK ASSERT */
954 #endif
955 
956 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
957 	INP_WLOCK_ASSERT(inp);
958 
959 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
960 	    (inp->inp_vflag & INP_DROPPED))
961 		return (inp);
962 
963 	tp = intotcpcb(inp);
964 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
965 
966 	/*
967 	 * Ignore some errors if we are hooked up.
968 	 * If connection hasn't completed, has retransmitted several times,
969 	 * and receives a second error, give up now.  This is better
970 	 * than waiting a long time to establish a connection that
971 	 * can never complete.
972 	 */
973 	if (tp->t_state == TCPS_ESTABLISHED &&
974 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
975 	     error == EHOSTDOWN)) {
976 		return (inp);
977 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
978 	    tp->t_softerror) {
979 		tp = tcp_drop(tp, error);
980 		if (tp != NULL)
981 			return (inp);
982 		else
983 			return (NULL);
984 	} else {
985 		tp->t_softerror = error;
986 		return (inp);
987 	}
988 #if 0
989 	wakeup( &so->so_timeo);
990 	sorwakeup(so);
991 	sowwakeup(so);
992 #endif
993 }
994 
995 static int
996 tcp_pcblist(SYSCTL_HANDLER_ARGS)
997 {
998 	INIT_VNET_INET(curvnet);
999 	int error, i, m, n, pcb_count;
1000 	struct inpcb *inp, **inp_list;
1001 	inp_gen_t gencnt;
1002 	struct xinpgen xig;
1003 
1004 	/*
1005 	 * The process of preparing the TCB list is too time-consuming and
1006 	 * resource-intensive to repeat twice on every request.
1007 	 */
1008 	if (req->oldptr == NULL) {
1009 		m = syncache_pcbcount();
1010 		n = V_tcbinfo.ipi_count;
1011 		req->oldidx = 2 * (sizeof xig)
1012 			+ ((m + n) + n/8) * sizeof(struct xtcpcb);
1013 		return (0);
1014 	}
1015 
1016 	if (req->newptr != NULL)
1017 		return (EPERM);
1018 
1019 	/*
1020 	 * OK, now we're committed to doing something.
1021 	 */
1022 	INP_INFO_RLOCK(&V_tcbinfo);
1023 	gencnt = V_tcbinfo.ipi_gencnt;
1024 	n = V_tcbinfo.ipi_count;
1025 	INP_INFO_RUNLOCK(&V_tcbinfo);
1026 
1027 	m = syncache_pcbcount();
1028 
1029 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1030 		+ (n + m) * sizeof(struct xtcpcb));
1031 	if (error != 0)
1032 		return (error);
1033 
1034 	xig.xig_len = sizeof xig;
1035 	xig.xig_count = n + m;
1036 	xig.xig_gen = gencnt;
1037 	xig.xig_sogen = so_gencnt;
1038 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1039 	if (error)
1040 		return (error);
1041 
1042 	error = syncache_pcblist(req, m, &pcb_count);
1043 	if (error)
1044 		return (error);
1045 
1046 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1047 	if (inp_list == NULL)
1048 		return (ENOMEM);
1049 
1050 	INP_INFO_RLOCK(&V_tcbinfo);
1051 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1052 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1053 		INP_RLOCK(inp);
1054 		if (inp->inp_gencnt <= gencnt) {
1055 			/*
1056 			 * XXX: This use of cr_cansee(), introduced with
1057 			 * TCP state changes, is not quite right, but for
1058 			 * now, better than nothing.
1059 			 */
1060 			if (inp->inp_vflag & INP_TIMEWAIT) {
1061 				if (intotw(inp) != NULL)
1062 					error = cr_cansee(req->td->td_ucred,
1063 					    intotw(inp)->tw_cred);
1064 				else
1065 					error = EINVAL;	/* Skip this inp. */
1066 			} else
1067 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1068 			if (error == 0)
1069 				inp_list[i++] = inp;
1070 		}
1071 		INP_RUNLOCK(inp);
1072 	}
1073 	INP_INFO_RUNLOCK(&V_tcbinfo);
1074 	n = i;
1075 
1076 	error = 0;
1077 	for (i = 0; i < n; i++) {
1078 		inp = inp_list[i];
1079 		INP_RLOCK(inp);
1080 		if (inp->inp_gencnt <= gencnt) {
1081 			struct xtcpcb xt;
1082 			void *inp_ppcb;
1083 
1084 			bzero(&xt, sizeof(xt));
1085 			xt.xt_len = sizeof xt;
1086 			/* XXX should avoid extra copy */
1087 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1088 			inp_ppcb = inp->inp_ppcb;
1089 			if (inp_ppcb == NULL)
1090 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1091 			else if (inp->inp_vflag & INP_TIMEWAIT) {
1092 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1093 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1094 			} else
1095 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1096 			if (inp->inp_socket != NULL)
1097 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1098 			else {
1099 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1100 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1101 			}
1102 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1103 			INP_RUNLOCK(inp);
1104 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1105 		} else
1106 			INP_RUNLOCK(inp);
1107 
1108 	}
1109 	if (!error) {
1110 		/*
1111 		 * Give the user an updated idea of our state.
1112 		 * If the generation differs from what we told
1113 		 * her before, she knows that something happened
1114 		 * while we were processing this request, and it
1115 		 * might be necessary to retry.
1116 		 */
1117 		INP_INFO_RLOCK(&V_tcbinfo);
1118 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1119 		xig.xig_sogen = so_gencnt;
1120 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1121 		INP_INFO_RUNLOCK(&V_tcbinfo);
1122 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1123 	}
1124 	free(inp_list, M_TEMP);
1125 	return (error);
1126 }
1127 
1128 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1129     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1130 
1131 static int
1132 tcp_getcred(SYSCTL_HANDLER_ARGS)
1133 {
1134 	INIT_VNET_INET(curvnet);
1135 	struct xucred xuc;
1136 	struct sockaddr_in addrs[2];
1137 	struct inpcb *inp;
1138 	int error;
1139 
1140 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1141 	if (error)
1142 		return (error);
1143 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1144 	if (error)
1145 		return (error);
1146 	INP_INFO_RLOCK(&V_tcbinfo);
1147 	inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1148 	    addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1149 	if (inp != NULL) {
1150 		INP_RLOCK(inp);
1151 		INP_INFO_RUNLOCK(&V_tcbinfo);
1152 		if (inp->inp_socket == NULL)
1153 			error = ENOENT;
1154 		if (error == 0)
1155 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1156 		if (error == 0)
1157 			cru2x(inp->inp_cred, &xuc);
1158 		INP_RUNLOCK(inp);
1159 	} else {
1160 		INP_INFO_RUNLOCK(&V_tcbinfo);
1161 		error = ENOENT;
1162 	}
1163 	if (error == 0)
1164 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1165 	return (error);
1166 }
1167 
1168 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1169     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1170     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1171 
1172 #ifdef INET6
1173 static int
1174 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1175 {
1176 	INIT_VNET_INET(curvnet);
1177 	INIT_VNET_INET6(curvnet);
1178 	struct xucred xuc;
1179 	struct sockaddr_in6 addrs[2];
1180 	struct inpcb *inp;
1181 	int error, mapped = 0;
1182 
1183 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1184 	if (error)
1185 		return (error);
1186 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1187 	if (error)
1188 		return (error);
1189 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1190 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1191 		return (error);
1192 	}
1193 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1194 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1195 			mapped = 1;
1196 		else
1197 			return (EINVAL);
1198 	}
1199 
1200 	INP_INFO_RLOCK(&V_tcbinfo);
1201 	if (mapped == 1)
1202 		inp = in_pcblookup_hash(&V_tcbinfo,
1203 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1204 			addrs[1].sin6_port,
1205 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1206 			addrs[0].sin6_port,
1207 			0, NULL);
1208 	else
1209 		inp = in6_pcblookup_hash(&V_tcbinfo,
1210 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1211 			&addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1212 	if (inp != NULL) {
1213 		INP_RLOCK(inp);
1214 		INP_INFO_RUNLOCK(&V_tcbinfo);
1215 		if (inp->inp_socket == NULL)
1216 			error = ENOENT;
1217 		if (error == 0)
1218 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1219 		if (error == 0)
1220 			cru2x(inp->inp_cred, &xuc);
1221 		INP_RUNLOCK(inp);
1222 	} else {
1223 		INP_INFO_RUNLOCK(&V_tcbinfo);
1224 		error = ENOENT;
1225 	}
1226 	if (error == 0)
1227 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1228 	return (error);
1229 }
1230 
1231 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1232     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1233     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1234 #endif
1235 
1236 
1237 void
1238 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1239 {
1240 	INIT_VNET_INET(curvnet);
1241 	struct ip *ip = vip;
1242 	struct tcphdr *th;
1243 	struct in_addr faddr;
1244 	struct inpcb *inp;
1245 	struct tcpcb *tp;
1246 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1247 	struct icmp *icp;
1248 	struct in_conninfo inc;
1249 	tcp_seq icmp_tcp_seq;
1250 	int mtu;
1251 
1252 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1253 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1254 		return;
1255 
1256 	if (cmd == PRC_MSGSIZE)
1257 		notify = tcp_mtudisc;
1258 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1259 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1260 		notify = tcp_drop_syn_sent;
1261 	/*
1262 	 * Redirects don't need to be handled up here.
1263 	 */
1264 	else if (PRC_IS_REDIRECT(cmd))
1265 		return;
1266 	/*
1267 	 * Source quench is depreciated.
1268 	 */
1269 	else if (cmd == PRC_QUENCH)
1270 		return;
1271 	/*
1272 	 * Hostdead is ugly because it goes linearly through all PCBs.
1273 	 * XXX: We never get this from ICMP, otherwise it makes an
1274 	 * excellent DoS attack on machines with many connections.
1275 	 */
1276 	else if (cmd == PRC_HOSTDEAD)
1277 		ip = NULL;
1278 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1279 		return;
1280 	if (ip != NULL) {
1281 		icp = (struct icmp *)((caddr_t)ip
1282 				      - offsetof(struct icmp, icmp_ip));
1283 		th = (struct tcphdr *)((caddr_t)ip
1284 				       + (ip->ip_hl << 2));
1285 		INP_INFO_WLOCK(&V_tcbinfo);
1286 		inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1287 		    ip->ip_src, th->th_sport, 0, NULL);
1288 		if (inp != NULL)  {
1289 			INP_WLOCK(inp);
1290 			if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1291 			    !(inp->inp_vflag & INP_DROPPED) &&
1292 			    !(inp->inp_socket == NULL)) {
1293 				icmp_tcp_seq = htonl(th->th_seq);
1294 				tp = intotcpcb(inp);
1295 				if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1296 				    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1297 					if (cmd == PRC_MSGSIZE) {
1298 					    /*
1299 					     * MTU discovery:
1300 					     * If we got a needfrag set the MTU
1301 					     * in the route to the suggested new
1302 					     * value (if given) and then notify.
1303 					     */
1304 					    bzero(&inc, sizeof(inc));
1305 					    inc.inc_flags = 0;	/* IPv4 */
1306 					    inc.inc_faddr = faddr;
1307 					    inc.inc_fibnum =
1308 						inp->inp_inc.inc_fibnum;
1309 
1310 					    mtu = ntohs(icp->icmp_nextmtu);
1311 					    /*
1312 					     * If no alternative MTU was
1313 					     * proposed, try the next smaller
1314 					     * one.  ip->ip_len has already
1315 					     * been swapped in icmp_input().
1316 					     */
1317 					    if (!mtu)
1318 						mtu = ip_next_mtu(ip->ip_len,
1319 						 1);
1320 					    if (mtu < max(296, V_tcp_minmss
1321 						 + sizeof(struct tcpiphdr)))
1322 						mtu = 0;
1323 					    if (!mtu)
1324 						mtu = V_tcp_mssdflt
1325 						 + sizeof(struct tcpiphdr);
1326 					    /*
1327 					     * Only cache the the MTU if it
1328 					     * is smaller than the interface
1329 					     * or route MTU.  tcp_mtudisc()
1330 					     * will do right thing by itself.
1331 					     */
1332 					    if (mtu <= tcp_maxmtu(&inc, NULL))
1333 						tcp_hc_updatemtu(&inc, mtu);
1334 					}
1335 
1336 					inp = (*notify)(inp, inetctlerrmap[cmd]);
1337 				}
1338 			}
1339 			if (inp != NULL)
1340 				INP_WUNLOCK(inp);
1341 		} else {
1342 			inc.inc_fport = th->th_dport;
1343 			inc.inc_lport = th->th_sport;
1344 			inc.inc_faddr = faddr;
1345 			inc.inc_laddr = ip->ip_src;
1346 #ifdef INET6
1347 			inc.inc_isipv6 = 0;
1348 #endif
1349 			syncache_unreach(&inc, th);
1350 		}
1351 		INP_INFO_WUNLOCK(&V_tcbinfo);
1352 	} else
1353 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1354 }
1355 
1356 #ifdef INET6
1357 void
1358 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1359 {
1360 	INIT_VNET_INET(curvnet);
1361 	struct tcphdr th;
1362 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1363 	struct ip6_hdr *ip6;
1364 	struct mbuf *m;
1365 	struct ip6ctlparam *ip6cp = NULL;
1366 	const struct sockaddr_in6 *sa6_src = NULL;
1367 	int off;
1368 	struct tcp_portonly {
1369 		u_int16_t th_sport;
1370 		u_int16_t th_dport;
1371 	} *thp;
1372 
1373 	if (sa->sa_family != AF_INET6 ||
1374 	    sa->sa_len != sizeof(struct sockaddr_in6))
1375 		return;
1376 
1377 	if (cmd == PRC_MSGSIZE)
1378 		notify = tcp_mtudisc;
1379 	else if (!PRC_IS_REDIRECT(cmd) &&
1380 		 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1381 		return;
1382 	/* Source quench is depreciated. */
1383 	else if (cmd == PRC_QUENCH)
1384 		return;
1385 
1386 	/* if the parameter is from icmp6, decode it. */
1387 	if (d != NULL) {
1388 		ip6cp = (struct ip6ctlparam *)d;
1389 		m = ip6cp->ip6c_m;
1390 		ip6 = ip6cp->ip6c_ip6;
1391 		off = ip6cp->ip6c_off;
1392 		sa6_src = ip6cp->ip6c_src;
1393 	} else {
1394 		m = NULL;
1395 		ip6 = NULL;
1396 		off = 0;	/* fool gcc */
1397 		sa6_src = &sa6_any;
1398 	}
1399 
1400 	if (ip6 != NULL) {
1401 		struct in_conninfo inc;
1402 		/*
1403 		 * XXX: We assume that when IPV6 is non NULL,
1404 		 * M and OFF are valid.
1405 		 */
1406 
1407 		/* check if we can safely examine src and dst ports */
1408 		if (m->m_pkthdr.len < off + sizeof(*thp))
1409 			return;
1410 
1411 		bzero(&th, sizeof(th));
1412 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1413 
1414 		in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1415 		    (struct sockaddr *)ip6cp->ip6c_src,
1416 		    th.th_sport, cmd, NULL, notify);
1417 
1418 		inc.inc_fport = th.th_dport;
1419 		inc.inc_lport = th.th_sport;
1420 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1421 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1422 		inc.inc_isipv6 = 1;
1423 		INP_INFO_WLOCK(&V_tcbinfo);
1424 		syncache_unreach(&inc, &th);
1425 		INP_INFO_WUNLOCK(&V_tcbinfo);
1426 	} else
1427 		in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1428 			      0, cmd, NULL, notify);
1429 }
1430 #endif /* INET6 */
1431 
1432 
1433 /*
1434  * Following is where TCP initial sequence number generation occurs.
1435  *
1436  * There are two places where we must use initial sequence numbers:
1437  * 1.  In SYN-ACK packets.
1438  * 2.  In SYN packets.
1439  *
1440  * All ISNs for SYN-ACK packets are generated by the syncache.  See
1441  * tcp_syncache.c for details.
1442  *
1443  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1444  * depends on this property.  In addition, these ISNs should be
1445  * unguessable so as to prevent connection hijacking.  To satisfy
1446  * the requirements of this situation, the algorithm outlined in
1447  * RFC 1948 is used, with only small modifications.
1448  *
1449  * Implementation details:
1450  *
1451  * Time is based off the system timer, and is corrected so that it
1452  * increases by one megabyte per second.  This allows for proper
1453  * recycling on high speed LANs while still leaving over an hour
1454  * before rollover.
1455  *
1456  * As reading the *exact* system time is too expensive to be done
1457  * whenever setting up a TCP connection, we increment the time
1458  * offset in two ways.  First, a small random positive increment
1459  * is added to isn_offset for each connection that is set up.
1460  * Second, the function tcp_isn_tick fires once per clock tick
1461  * and increments isn_offset as necessary so that sequence numbers
1462  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
1463  * random positive increments serve only to ensure that the same
1464  * exact sequence number is never sent out twice (as could otherwise
1465  * happen when a port is recycled in less than the system tick
1466  * interval.)
1467  *
1468  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1469  * between seeding of isn_secret.  This is normally set to zero,
1470  * as reseeding should not be necessary.
1471  *
1472  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1473  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
1474  * general, this means holding an exclusive (write) lock.
1475  */
1476 
1477 #define ISN_BYTES_PER_SECOND 1048576
1478 #define ISN_STATIC_INCREMENT 4096
1479 #define ISN_RANDOM_INCREMENT (4096 - 1)
1480 
1481 #ifdef VIMAGE_GLOBALS
1482 static u_char isn_secret[32];
1483 static int isn_last_reseed;
1484 static u_int32_t isn_offset, isn_offset_old;
1485 static MD5_CTX isn_ctx;
1486 #endif
1487 
1488 tcp_seq
1489 tcp_new_isn(struct tcpcb *tp)
1490 {
1491 	INIT_VNET_INET(tp->t_vnet);
1492 	u_int32_t md5_buffer[4];
1493 	tcp_seq new_isn;
1494 
1495 	INP_WLOCK_ASSERT(tp->t_inpcb);
1496 
1497 	ISN_LOCK();
1498 	/* Seed if this is the first use, reseed if requested. */
1499 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1500 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1501 		< (u_int)ticks))) {
1502 		read_random(&V_isn_secret, sizeof(V_isn_secret));
1503 		V_isn_last_reseed = ticks;
1504 	}
1505 
1506 	/* Compute the md5 hash and return the ISN. */
1507 	MD5Init(&V_isn_ctx);
1508 	MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1509 	MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1510 #ifdef INET6
1511 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1512 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1513 			  sizeof(struct in6_addr));
1514 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1515 			  sizeof(struct in6_addr));
1516 	} else
1517 #endif
1518 	{
1519 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1520 			  sizeof(struct in_addr));
1521 		MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1522 			  sizeof(struct in_addr));
1523 	}
1524 	MD5Update(&V_isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1525 	MD5Final((u_char *) &md5_buffer, &V_isn_ctx);
1526 	new_isn = (tcp_seq) md5_buffer[0];
1527 	V_isn_offset += ISN_STATIC_INCREMENT +
1528 		(arc4random() & ISN_RANDOM_INCREMENT);
1529 	new_isn += V_isn_offset;
1530 	ISN_UNLOCK();
1531 	return (new_isn);
1532 }
1533 
1534 /*
1535  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1536  * to keep time flowing at a relatively constant rate.  If the random
1537  * increments have already pushed us past the projected offset, do nothing.
1538  */
1539 static void
1540 tcp_isn_tick(void *xtp)
1541 {
1542 	VNET_ITERATOR_DECL(vnet_iter);
1543 	u_int32_t projected_offset;
1544 
1545 	ISN_LOCK();
1546 	VNET_LIST_RLOCK();
1547 	VNET_FOREACH(vnet_iter) {
1548 		CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1549 		INIT_VNET_INET(curvnet);
1550 		projected_offset =
1551 		    V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1552 
1553 		if (SEQ_GT(projected_offset, V_isn_offset))
1554 			V_isn_offset = projected_offset;
1555 
1556 		V_isn_offset_old = V_isn_offset;
1557 		CURVNET_RESTORE();
1558 	}
1559 	VNET_LIST_RUNLOCK();
1560 	callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1561 	ISN_UNLOCK();
1562 }
1563 
1564 /*
1565  * When a specific ICMP unreachable message is received and the
1566  * connection state is SYN-SENT, drop the connection.  This behavior
1567  * is controlled by the icmp_may_rst sysctl.
1568  */
1569 struct inpcb *
1570 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1571 {
1572 #ifdef INVARIANTS
1573 	INIT_VNET_INET(inp->inp_vnet);
1574 #endif
1575 	struct tcpcb *tp;
1576 
1577 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1578 	INP_WLOCK_ASSERT(inp);
1579 
1580 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1581 	    (inp->inp_vflag & INP_DROPPED))
1582 		return (inp);
1583 
1584 	tp = intotcpcb(inp);
1585 	if (tp->t_state != TCPS_SYN_SENT)
1586 		return (inp);
1587 
1588 	tp = tcp_drop(tp, errno);
1589 	if (tp != NULL)
1590 		return (inp);
1591 	else
1592 		return (NULL);
1593 }
1594 
1595 /*
1596  * When `need fragmentation' ICMP is received, update our idea of the MSS
1597  * based on the new value in the route.  Also nudge TCP to send something,
1598  * since we know the packet we just sent was dropped.
1599  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1600  */
1601 struct inpcb *
1602 tcp_mtudisc(struct inpcb *inp, int errno)
1603 {
1604 	INIT_VNET_INET(inp->inp_vnet);
1605 	struct tcpcb *tp;
1606 	struct socket *so;
1607 
1608 	INP_WLOCK_ASSERT(inp);
1609 	if ((inp->inp_vflag & INP_TIMEWAIT) ||
1610 	    (inp->inp_vflag & INP_DROPPED))
1611 		return (inp);
1612 
1613 	tp = intotcpcb(inp);
1614 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1615 
1616 	tcp_mss_update(tp, -1, NULL, NULL);
1617 
1618 	so = inp->inp_socket;
1619 	SOCKBUF_LOCK(&so->so_snd);
1620 	/* If the mss is larger than the socket buffer, decrease the mss. */
1621 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
1622 		tp->t_maxseg = so->so_snd.sb_hiwat;
1623 	SOCKBUF_UNLOCK(&so->so_snd);
1624 
1625 	V_tcpstat.tcps_mturesent++;
1626 	tp->t_rtttime = 0;
1627 	tp->snd_nxt = tp->snd_una;
1628 	tcp_free_sackholes(tp);
1629 	tp->snd_recover = tp->snd_max;
1630 	if (tp->t_flags & TF_SACK_PERMIT)
1631 		EXIT_FASTRECOVERY(tp);
1632 	tcp_output_send(tp);
1633 	return (inp);
1634 }
1635 
1636 /*
1637  * Look-up the routing entry to the peer of this inpcb.  If no route
1638  * is found and it cannot be allocated, then return 0.  This routine
1639  * is called by TCP routines that access the rmx structure and by
1640  * tcp_mss_update to get the peer/interface MTU.
1641  */
1642 u_long
1643 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1644 {
1645 	struct route sro;
1646 	struct sockaddr_in *dst;
1647 	struct ifnet *ifp;
1648 	u_long maxmtu = 0;
1649 
1650 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1651 
1652 	bzero(&sro, sizeof(sro));
1653 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
1654 	        dst = (struct sockaddr_in *)&sro.ro_dst;
1655 		dst->sin_family = AF_INET;
1656 		dst->sin_len = sizeof(*dst);
1657 		dst->sin_addr = inc->inc_faddr;
1658 		in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1659 	}
1660 	if (sro.ro_rt != NULL) {
1661 		ifp = sro.ro_rt->rt_ifp;
1662 		if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1663 			maxmtu = ifp->if_mtu;
1664 		else
1665 			maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1666 
1667 		/* Report additional interface capabilities. */
1668 		if (flags != NULL) {
1669 			if (ifp->if_capenable & IFCAP_TSO4 &&
1670 			    ifp->if_hwassist & CSUM_TSO)
1671 				*flags |= CSUM_TSO;
1672 		}
1673 		RTFREE(sro.ro_rt);
1674 	}
1675 	return (maxmtu);
1676 }
1677 
1678 #ifdef INET6
1679 u_long
1680 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1681 {
1682 	struct route_in6 sro6;
1683 	struct ifnet *ifp;
1684 	u_long maxmtu = 0;
1685 
1686 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1687 
1688 	bzero(&sro6, sizeof(sro6));
1689 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1690 		sro6.ro_dst.sin6_family = AF_INET6;
1691 		sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1692 		sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1693 		rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1694 	}
1695 	if (sro6.ro_rt != NULL) {
1696 		ifp = sro6.ro_rt->rt_ifp;
1697 		if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1698 			maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1699 		else
1700 			maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1701 				     IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1702 
1703 		/* Report additional interface capabilities. */
1704 		if (flags != NULL) {
1705 			if (ifp->if_capenable & IFCAP_TSO6 &&
1706 			    ifp->if_hwassist & CSUM_TSO)
1707 				*flags |= CSUM_TSO;
1708 		}
1709 		RTFREE(sro6.ro_rt);
1710 	}
1711 
1712 	return (maxmtu);
1713 }
1714 #endif /* INET6 */
1715 
1716 #ifdef IPSEC
1717 /* compute ESP/AH header size for TCP, including outer IP header. */
1718 size_t
1719 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1720 {
1721 	struct inpcb *inp;
1722 	struct mbuf *m;
1723 	size_t hdrsiz;
1724 	struct ip *ip;
1725 #ifdef INET6
1726 	struct ip6_hdr *ip6;
1727 #endif
1728 	struct tcphdr *th;
1729 
1730 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1731 		return (0);
1732 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1733 	if (!m)
1734 		return (0);
1735 
1736 #ifdef INET6
1737 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1738 		ip6 = mtod(m, struct ip6_hdr *);
1739 		th = (struct tcphdr *)(ip6 + 1);
1740 		m->m_pkthdr.len = m->m_len =
1741 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1742 		tcpip_fillheaders(inp, ip6, th);
1743 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1744 	} else
1745 #endif /* INET6 */
1746 	{
1747 		ip = mtod(m, struct ip *);
1748 		th = (struct tcphdr *)(ip + 1);
1749 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1750 		tcpip_fillheaders(inp, ip, th);
1751 		hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1752 	}
1753 
1754 	m_free(m);
1755 	return (hdrsiz);
1756 }
1757 #endif /* IPSEC */
1758 
1759 /*
1760  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1761  *
1762  * This code attempts to calculate the bandwidth-delay product as a
1763  * means of determining the optimal window size to maximize bandwidth,
1764  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1765  * routers.  This code also does a fairly good job keeping RTTs in check
1766  * across slow links like modems.  We implement an algorithm which is very
1767  * similar (but not meant to be) TCP/Vegas.  The code operates on the
1768  * transmitter side of a TCP connection and so only effects the transmit
1769  * side of the connection.
1770  *
1771  * BACKGROUND:  TCP makes no provision for the management of buffer space
1772  * at the end points or at the intermediate routers and switches.  A TCP
1773  * stream, whether using NewReno or not, will eventually buffer as
1774  * many packets as it is able and the only reason this typically works is
1775  * due to the fairly small default buffers made available for a connection
1776  * (typicaly 16K or 32K).  As machines use larger windows and/or window
1777  * scaling it is now fairly easy for even a single TCP connection to blow-out
1778  * all available buffer space not only on the local interface, but on
1779  * intermediate routers and switches as well.  NewReno makes a misguided
1780  * attempt to 'solve' this problem by waiting for an actual failure to occur,
1781  * then backing off, then steadily increasing the window again until another
1782  * failure occurs, ad-infinitum.  This results in terrible oscillation that
1783  * is only made worse as network loads increase and the idea of intentionally
1784  * blowing out network buffers is, frankly, a terrible way to manage network
1785  * resources.
1786  *
1787  * It is far better to limit the transmit window prior to the failure
1788  * condition being achieved.  There are two general ways to do this:  First
1789  * you can 'scan' through different transmit window sizes and locate the
1790  * point where the RTT stops increasing, indicating that you have filled the
1791  * pipe, then scan backwards until you note that RTT stops decreasing, then
1792  * repeat ad-infinitum.  This method works in principle but has severe
1793  * implementation issues due to RTT variances, timer granularity, and
1794  * instability in the algorithm which can lead to many false positives and
1795  * create oscillations as well as interact badly with other TCP streams
1796  * implementing the same algorithm.
1797  *
1798  * The second method is to limit the window to the bandwidth delay product
1799  * of the link.  This is the method we implement.  RTT variances and our
1800  * own manipulation of the congestion window, bwnd, can potentially
1801  * destabilize the algorithm.  For this reason we have to stabilize the
1802  * elements used to calculate the window.  We do this by using the minimum
1803  * observed RTT, the long term average of the observed bandwidth, and
1804  * by adding two segments worth of slop.  It isn't perfect but it is able
1805  * to react to changing conditions and gives us a very stable basis on
1806  * which to extend the algorithm.
1807  */
1808 void
1809 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1810 {
1811 	INIT_VNET_INET(tp->t_vnet);
1812 	u_long bw;
1813 	u_long bwnd;
1814 	int save_ticks;
1815 
1816 	INP_WLOCK_ASSERT(tp->t_inpcb);
1817 
1818 	/*
1819 	 * If inflight_enable is disabled in the middle of a tcp connection,
1820 	 * make sure snd_bwnd is effectively disabled.
1821 	 */
1822 	if (V_tcp_inflight_enable == 0 ||
1823 	    tp->t_rttlow < V_tcp_inflight_rttthresh) {
1824 		tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1825 		tp->snd_bandwidth = 0;
1826 		return;
1827 	}
1828 
1829 	/*
1830 	 * Figure out the bandwidth.  Due to the tick granularity this
1831 	 * is a very rough number and it MUST be averaged over a fairly
1832 	 * long period of time.  XXX we need to take into account a link
1833 	 * that is not using all available bandwidth, but for now our
1834 	 * slop will ramp us up if this case occurs and the bandwidth later
1835 	 * increases.
1836 	 *
1837 	 * Note: if ticks rollover 'bw' may wind up negative.  We must
1838 	 * effectively reset t_bw_rtttime for this case.
1839 	 */
1840 	save_ticks = ticks;
1841 	if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1842 		return;
1843 
1844 	bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1845 	    (save_ticks - tp->t_bw_rtttime);
1846 	tp->t_bw_rtttime = save_ticks;
1847 	tp->t_bw_rtseq = ack_seq;
1848 	if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1849 		return;
1850 	bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1851 
1852 	tp->snd_bandwidth = bw;
1853 
1854 	/*
1855 	 * Calculate the semi-static bandwidth delay product, plus two maximal
1856 	 * segments.  The additional slop puts us squarely in the sweet
1857 	 * spot and also handles the bandwidth run-up case and stabilization.
1858 	 * Without the slop we could be locking ourselves into a lower
1859 	 * bandwidth.
1860 	 *
1861 	 * Situations Handled:
1862 	 *	(1) Prevents over-queueing of packets on LANs, especially on
1863 	 *	    high speed LANs, allowing larger TCP buffers to be
1864 	 *	    specified, and also does a good job preventing
1865 	 *	    over-queueing of packets over choke points like modems
1866 	 *	    (at least for the transmit side).
1867 	 *
1868 	 *	(2) Is able to handle changing network loads (bandwidth
1869 	 *	    drops so bwnd drops, bandwidth increases so bwnd
1870 	 *	    increases).
1871 	 *
1872 	 *	(3) Theoretically should stabilize in the face of multiple
1873 	 *	    connections implementing the same algorithm (this may need
1874 	 *	    a little work).
1875 	 *
1876 	 *	(4) Stability value (defaults to 20 = 2 maximal packets) can
1877 	 *	    be adjusted with a sysctl but typically only needs to be
1878 	 *	    on very slow connections.  A value no smaller then 5
1879 	 *	    should be used, but only reduce this default if you have
1880 	 *	    no other choice.
1881 	 */
1882 #define USERTT	((tp->t_srtt + tp->t_rttbest) / 2)
1883 	bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1884 #undef USERTT
1885 
1886 	if (tcp_inflight_debug > 0) {
1887 		static int ltime;
1888 		if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1889 			ltime = ticks;
1890 			printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1891 			    tp,
1892 			    bw,
1893 			    tp->t_rttbest,
1894 			    tp->t_srtt,
1895 			    bwnd
1896 			);
1897 		}
1898 	}
1899 	if ((long)bwnd < V_tcp_inflight_min)
1900 		bwnd = V_tcp_inflight_min;
1901 	if (bwnd > V_tcp_inflight_max)
1902 		bwnd = V_tcp_inflight_max;
1903 	if ((long)bwnd < tp->t_maxseg * 2)
1904 		bwnd = tp->t_maxseg * 2;
1905 	tp->snd_bwnd = bwnd;
1906 }
1907 
1908 #ifdef TCP_SIGNATURE
1909 /*
1910  * Callback function invoked by m_apply() to digest TCP segment data
1911  * contained within an mbuf chain.
1912  */
1913 static int
1914 tcp_signature_apply(void *fstate, void *data, u_int len)
1915 {
1916 
1917 	MD5Update(fstate, (u_char *)data, len);
1918 	return (0);
1919 }
1920 
1921 /*
1922  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1923  *
1924  * Parameters:
1925  * m		pointer to head of mbuf chain
1926  * _unused
1927  * len		length of TCP segment data, excluding options
1928  * optlen	length of TCP segment options
1929  * buf		pointer to storage for computed MD5 digest
1930  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1931  *
1932  * We do this over ip, tcphdr, segment data, and the key in the SADB.
1933  * When called from tcp_input(), we can be sure that th_sum has been
1934  * zeroed out and verified already.
1935  *
1936  * Return 0 if successful, otherwise return -1.
1937  *
1938  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1939  * search with the destination IP address, and a 'magic SPI' to be
1940  * determined by the application. This is hardcoded elsewhere to 1179
1941  * right now. Another branch of this code exists which uses the SPD to
1942  * specify per-application flows but it is unstable.
1943  */
1944 int
1945 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1946     u_char *buf, u_int direction)
1947 {
1948 	INIT_VNET_IPSEC(curvnet);
1949 	union sockaddr_union dst;
1950 	struct ippseudo ippseudo;
1951 	MD5_CTX ctx;
1952 	int doff;
1953 	struct ip *ip;
1954 	struct ipovly *ipovly;
1955 	struct secasvar *sav;
1956 	struct tcphdr *th;
1957 #ifdef INET6
1958 	struct ip6_hdr *ip6;
1959 	struct in6_addr in6;
1960 	char ip6buf[INET6_ADDRSTRLEN];
1961 	uint32_t plen;
1962 	uint16_t nhdr;
1963 #endif
1964 	u_short savecsum;
1965 
1966 	KASSERT(m != NULL, ("NULL mbuf chain"));
1967 	KASSERT(buf != NULL, ("NULL signature pointer"));
1968 
1969 	/* Extract the destination from the IP header in the mbuf. */
1970 	bzero(&dst, sizeof(union sockaddr_union));
1971 	ip = mtod(m, struct ip *);
1972 #ifdef INET6
1973 	ip6 = NULL;	/* Make the compiler happy. */
1974 #endif
1975 	switch (ip->ip_v) {
1976 	case IPVERSION:
1977 		dst.sa.sa_len = sizeof(struct sockaddr_in);
1978 		dst.sa.sa_family = AF_INET;
1979 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1980 		    ip->ip_src : ip->ip_dst;
1981 		break;
1982 #ifdef INET6
1983 	case (IPV6_VERSION >> 4):
1984 		ip6 = mtod(m, struct ip6_hdr *);
1985 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
1986 		dst.sa.sa_family = AF_INET6;
1987 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1988 		    ip6->ip6_src : ip6->ip6_dst;
1989 		break;
1990 #endif
1991 	default:
1992 		return (EINVAL);
1993 		/* NOTREACHED */
1994 		break;
1995 	}
1996 
1997 	/* Look up an SADB entry which matches the address of the peer. */
1998 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1999 	if (sav == NULL) {
2000 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2001 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2002 #ifdef INET6
2003 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2004 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2005 #endif
2006 			"(unsupported)"));
2007 		return (EINVAL);
2008 	}
2009 
2010 	MD5Init(&ctx);
2011 	/*
2012 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2013 	 *
2014 	 * XXX The ippseudo header MUST be digested in network byte order,
2015 	 * or else we'll fail the regression test. Assume all fields we've
2016 	 * been doing arithmetic on have been in host byte order.
2017 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2018 	 * tcp_output(), the underlying ip_len member has not yet been set.
2019 	 */
2020 	switch (ip->ip_v) {
2021 	case IPVERSION:
2022 		ipovly = (struct ipovly *)ip;
2023 		ippseudo.ippseudo_src = ipovly->ih_src;
2024 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2025 		ippseudo.ippseudo_pad = 0;
2026 		ippseudo.ippseudo_p = IPPROTO_TCP;
2027 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2028 		    optlen);
2029 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2030 
2031 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2032 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2033 		break;
2034 #ifdef INET6
2035 	/*
2036 	 * RFC 2385, 2.0  Proposal
2037 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2038 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2039 	 * extended next header value (to form 32 bits), and 32-bit segment
2040 	 * length.
2041 	 * Note: Upper-Layer Packet Length comes before Next Header.
2042 	 */
2043 	case (IPV6_VERSION >> 4):
2044 		in6 = ip6->ip6_src;
2045 		in6_clearscope(&in6);
2046 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2047 		in6 = ip6->ip6_dst;
2048 		in6_clearscope(&in6);
2049 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2050 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2051 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2052 		nhdr = 0;
2053 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2054 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2055 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2056 		nhdr = IPPROTO_TCP;
2057 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2058 
2059 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2060 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2061 		break;
2062 #endif
2063 	default:
2064 		return (EINVAL);
2065 		/* NOTREACHED */
2066 		break;
2067 	}
2068 
2069 
2070 	/*
2071 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2072 	 * The TCP checksum must be set to zero.
2073 	 */
2074 	savecsum = th->th_sum;
2075 	th->th_sum = 0;
2076 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2077 	th->th_sum = savecsum;
2078 
2079 	/*
2080 	 * Step 3: Update MD5 hash with TCP segment data.
2081 	 *         Use m_apply() to avoid an early m_pullup().
2082 	 */
2083 	if (len > 0)
2084 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2085 
2086 	/*
2087 	 * Step 4: Update MD5 hash with shared secret.
2088 	 */
2089 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2090 	MD5Final(buf, &ctx);
2091 
2092 	key_sa_recordxfer(sav, m);
2093 	KEY_FREESAV(&sav);
2094 	return (0);
2095 }
2096 #endif /* TCP_SIGNATURE */
2097 
2098 static int
2099 sysctl_drop(SYSCTL_HANDLER_ARGS)
2100 {
2101 	INIT_VNET_INET(curvnet);
2102 #ifdef INET6
2103 	INIT_VNET_INET6(curvnet);
2104 #endif
2105 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2106 	struct sockaddr_storage addrs[2];
2107 	struct inpcb *inp;
2108 	struct tcpcb *tp;
2109 	struct tcptw *tw;
2110 	struct sockaddr_in *fin, *lin;
2111 #ifdef INET6
2112 	struct sockaddr_in6 *fin6, *lin6;
2113 	struct in6_addr f6, l6;
2114 #endif
2115 	int error;
2116 
2117 	inp = NULL;
2118 	fin = lin = NULL;
2119 #ifdef INET6
2120 	fin6 = lin6 = NULL;
2121 #endif
2122 	error = 0;
2123 
2124 	if (req->oldptr != NULL || req->oldlen != 0)
2125 		return (EINVAL);
2126 	if (req->newptr == NULL)
2127 		return (EPERM);
2128 	if (req->newlen < sizeof(addrs))
2129 		return (ENOMEM);
2130 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2131 	if (error)
2132 		return (error);
2133 
2134 	switch (addrs[0].ss_family) {
2135 #ifdef INET6
2136 	case AF_INET6:
2137 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2138 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2139 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2140 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2141 			return (EINVAL);
2142 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2143 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2144 				return (EINVAL);
2145 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2146 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2147 			fin = (struct sockaddr_in *)&addrs[0];
2148 			lin = (struct sockaddr_in *)&addrs[1];
2149 			break;
2150 		}
2151 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2152 		if (error)
2153 			return (error);
2154 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2155 		if (error)
2156 			return (error);
2157 		break;
2158 #endif
2159 	case AF_INET:
2160 		fin = (struct sockaddr_in *)&addrs[0];
2161 		lin = (struct sockaddr_in *)&addrs[1];
2162 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2163 		    lin->sin_len != sizeof(struct sockaddr_in))
2164 			return (EINVAL);
2165 		break;
2166 	default:
2167 		return (EINVAL);
2168 	}
2169 	INP_INFO_WLOCK(&V_tcbinfo);
2170 	switch (addrs[0].ss_family) {
2171 #ifdef INET6
2172 	case AF_INET6:
2173 		inp = in6_pcblookup_hash(&V_tcbinfo, &f6, fin6->sin6_port,
2174 		    &l6, lin6->sin6_port, 0, NULL);
2175 		break;
2176 #endif
2177 	case AF_INET:
2178 		inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2179 		    fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2180 		break;
2181 	}
2182 	if (inp != NULL) {
2183 		INP_WLOCK(inp);
2184 		if (inp->inp_vflag & INP_TIMEWAIT) {
2185 			/*
2186 			 * XXXRW: There currently exists a state where an
2187 			 * inpcb is present, but its timewait state has been
2188 			 * discarded.  For now, don't allow dropping of this
2189 			 * type of inpcb.
2190 			 */
2191 			tw = intotw(inp);
2192 			if (tw != NULL)
2193 				tcp_twclose(tw, 0);
2194 			else
2195 				INP_WUNLOCK(inp);
2196 		} else if (!(inp->inp_vflag & INP_DROPPED) &&
2197 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2198 			tp = intotcpcb(inp);
2199 			tp = tcp_drop(tp, ECONNABORTED);
2200 			if (tp != NULL)
2201 				INP_WUNLOCK(inp);
2202 		} else
2203 			INP_WUNLOCK(inp);
2204 	} else
2205 		error = ESRCH;
2206 	INP_INFO_WUNLOCK(&V_tcbinfo);
2207 	return (error);
2208 }
2209 
2210 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2211     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2212     0, sysctl_drop, "", "Drop TCP connection");
2213 
2214 /*
2215  * Generate a standardized TCP log line for use throughout the
2216  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2217  * allow use in the interrupt context.
2218  *
2219  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2220  * NB: The function may return NULL if memory allocation failed.
2221  *
2222  * Due to header inclusion and ordering limitations the struct ip
2223  * and ip6_hdr pointers have to be passed as void pointers.
2224  */
2225 char *
2226 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2227     const void *ip6hdr)
2228 {
2229 	char *s, *sp;
2230 	size_t size;
2231 	struct ip *ip;
2232 #ifdef INET6
2233 	const struct ip6_hdr *ip6;
2234 
2235 	ip6 = (const struct ip6_hdr *)ip6hdr;
2236 #endif /* INET6 */
2237 	ip = (struct ip *)ip4hdr;
2238 
2239 	/*
2240 	 * The log line looks like this:
2241 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2242 	 */
2243 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2244 	    sizeof(PRINT_TH_FLAGS) + 1 +
2245 #ifdef INET6
2246 	    2 * INET6_ADDRSTRLEN;
2247 #else
2248 	    2 * INET_ADDRSTRLEN;
2249 #endif /* INET6 */
2250 
2251 	/* Is logging enabled? */
2252 	if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2253 		return (NULL);
2254 
2255 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2256 	if (s == NULL)
2257 		return (NULL);
2258 
2259 	strcat(s, "TCP: [");
2260 	sp = s + strlen(s);
2261 
2262 	if (inc && inc->inc_isipv6 == 0) {
2263 		inet_ntoa_r(inc->inc_faddr, sp);
2264 		sp = s + strlen(s);
2265 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2266 		sp = s + strlen(s);
2267 		inet_ntoa_r(inc->inc_laddr, sp);
2268 		sp = s + strlen(s);
2269 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2270 #ifdef INET6
2271 	} else if (inc) {
2272 		ip6_sprintf(sp, &inc->inc6_faddr);
2273 		sp = s + strlen(s);
2274 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2275 		sp = s + strlen(s);
2276 		ip6_sprintf(sp, &inc->inc6_laddr);
2277 		sp = s + strlen(s);
2278 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2279 	} else if (ip6 && th) {
2280 		ip6_sprintf(sp, &ip6->ip6_src);
2281 		sp = s + strlen(s);
2282 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2283 		sp = s + strlen(s);
2284 		ip6_sprintf(sp, &ip6->ip6_dst);
2285 		sp = s + strlen(s);
2286 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2287 #endif /* INET6 */
2288 	} else if (ip && th) {
2289 		inet_ntoa_r(ip->ip_src, sp);
2290 		sp = s + strlen(s);
2291 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2292 		sp = s + strlen(s);
2293 		inet_ntoa_r(ip->ip_dst, sp);
2294 		sp = s + strlen(s);
2295 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2296 	} else {
2297 		free(s, M_TCPLOG);
2298 		return (NULL);
2299 	}
2300 	sp = s + strlen(s);
2301 	if (th)
2302 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2303 	if (*(s + size - 1) != '\0')
2304 		panic("%s: string too long", __func__);
2305 	return (s);
2306 }
2307