xref: /dflybsd-src/sys/netinet/in_pcb.c (revision fc7e83fa167ace550636f25b513c873f2f3135ff)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1991, 1993, 1995
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
63  * $FreeBSD: src/sys/netinet/in_pcb.c,v 1.59.2.27 2004/01/02 04:06:42 ambrisko Exp $
64  */
65 
66 #include "opt_ipsec.h"
67 #include "opt_inet6.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/domain.h>
74 #include <sys/protosw.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/proc.h>
78 #include <sys/priv.h>
79 #include <sys/jail.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 
83 #include <sys/thread2.h>
84 #include <sys/socketvar2.h>
85 #include <sys/msgport2.h>
86 
87 #include <machine/limits.h>
88 
89 #include <net/if.h>
90 #include <net/if_types.h>
91 #include <net/route.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/in_pcb.h>
95 #include <netinet/in_var.h>
96 #include <netinet/ip_var.h>
97 #ifdef INET6
98 #include <netinet/ip6.h>
99 #include <netinet6/ip6_var.h>
100 #endif /* INET6 */
101 
102 #ifdef IPSEC
103 #include <netinet6/ipsec.h>
104 #include <netproto/key/key.h>
105 #include <netproto/ipsec/esp_var.h>
106 #endif
107 
108 #ifdef FAST_IPSEC
109 #if defined(IPSEC) || defined(IPSEC_ESP)
110 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!"
111 #endif
112 
113 #include <netproto/ipsec/ipsec.h>
114 #include <netproto/ipsec/key.h>
115 #define	IPSEC
116 #endif /* FAST_IPSEC */
117 
118 #define INP_LOCALGROUP_SIZMIN	8
119 #define INP_LOCALGROUP_SIZMAX	256
120 
121 struct in_addr zeroin_addr;
122 
123 /*
124  * These configure the range of local port addresses assigned to
125  * "unspecified" outgoing connections/packets/whatever.
126  */
127 int ipport_lowfirstauto = IPPORT_RESERVED - 1;	/* 1023 */
128 int ipport_lowlastauto = IPPORT_RESERVEDSTART;	/* 600 */
129 
130 int ipport_firstauto = IPPORT_RESERVED;		/* 1024 */
131 int ipport_lastauto = IPPORT_USERRESERVED;	/* 5000 */
132 
133 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO;	/* 49152 */
134 int ipport_hilastauto = IPPORT_HILASTAUTO;	/* 65535 */
135 
136 #define RANGECHK(var, min, max) \
137 	if ((var) < (min)) { (var) = (min); } \
138 	else if ((var) > (max)) { (var) = (max); }
139 
140 int udpencap_enable = 1;	/* enabled by default */
141 int udpencap_port = 4500;	/* triggers decapsulation */
142 
143 static int
144 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
145 {
146 	int error;
147 
148 	error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
149 	if (!error) {
150 		RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
151 		RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
152 
153 		RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
154 		RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
155 
156 		RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
157 		RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
158 	}
159 	return (error);
160 }
161 
162 #undef RANGECHK
163 
164 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
165 
166 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
167 	   &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
168 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
169 	   &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
171 	   &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
173 	   &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
175 	   &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
177 	   &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
178 
179 /*
180  * in_pcb.c: manage the Protocol Control Blocks.
181  *
182  * NOTE: It is assumed that most of these functions will be called from
183  * a critical section.  XXX - There are, unfortunately, a few exceptions
184  * to this rule that should be fixed.
185  *
186  * NOTE: The caller should initialize the cpu field to the cpu running the
187  * protocol stack associated with this inpcbinfo.
188  */
189 
190 void
191 in_pcbinfo_init(struct inpcbinfo *pcbinfo)
192 {
193 	LIST_INIT(&pcbinfo->pcblisthead);
194 	pcbinfo->cpu = -1;
195 	pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
196 				    M_WAITOK | M_ZERO);
197 }
198 
199 struct baddynamicports baddynamicports;
200 
201 /*
202  * Check if the specified port is invalid for dynamic allocation.
203  */
204 int
205 in_baddynamic(u_int16_t port, u_int16_t proto)
206 {
207 	switch (proto) {
208 	case IPPROTO_TCP:
209 		return (DP_ISSET(baddynamicports.tcp, port));
210 	case IPPROTO_UDP:
211 #ifdef IPSEC
212 		/* Cannot preset this as it is a sysctl */
213 		if (port == udpencap_port)
214 			return (1);
215 #endif
216 		return (DP_ISSET(baddynamicports.udp, port));
217 	default:
218 		return (0);
219 	}
220 }
221 
222 
223 /*
224  * Allocate a PCB and associate it with the socket.
225  */
226 int
227 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
228 {
229 	struct inpcb *inp;
230 #ifdef IPSEC
231 	int error;
232 #endif
233 
234 	inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK);
235 	if (inp == NULL)
236 		return (ENOMEM);
237 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
238 	inp->inp_pcbinfo = inp->inp_cpcbinfo = pcbinfo;
239 	inp->inp_socket = so;
240 #ifdef IPSEC
241 	error = ipsec_init_policy(so, &inp->inp_sp);
242 	if (error != 0) {
243 		kfree(inp, M_PCB);
244 		return (error);
245 	}
246 #endif
247 #ifdef INET6
248 	if (INP_SOCKAF(so) == AF_INET6 && ip6_v6only)
249 		inp->inp_flags |= IN6P_IPV6_V6ONLY;
250 	if (ip6_auto_flowlabel)
251 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
252 #endif
253 	soreference(so);
254 	so->so_pcb = inp;
255 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
256 	pcbinfo->ipi_count++;
257 	return (0);
258 }
259 
260 /*
261  * Unlink a pcb with the intention of moving it to another cpu with a
262  * different pcbinfo.  While unlinked nothing should attempt to dereference
263  * inp_pcbinfo, NULL it out so we assert if it does.
264  */
265 void
266 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
267 {
268 	KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
269 	KASSERT(inp->inp_cpcbinfo == pcbinfo, ("cpcbinfo mismatch"));
270 	KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
271 	    ("already linked"));
272 
273 	LIST_REMOVE(inp, inp_list);
274 	pcbinfo->ipi_count--;
275 	inp->inp_pcbinfo = NULL;
276 	inp->inp_cpcbinfo = NULL;
277 }
278 
279 /*
280  * Relink a pcb into a new pcbinfo.
281  */
282 void
283 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
284 {
285 	KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
286 	KASSERT(inp->inp_cpcbinfo == NULL, ("has cpcbinfo"));
287 	KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
288 	    ("already linked"));
289 
290 	inp->inp_cpcbinfo = pcbinfo;
291 	inp->inp_pcbinfo = pcbinfo;
292 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
293 	pcbinfo->ipi_count++;
294 }
295 
296 static int
297 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
298 {
299 	struct inpcbportinfo *portinfo = inp->inp_pcbinfo->portinfo;
300 	u_short first, last, lport;
301 	u_short *lastport;
302 	int count, error;
303 
304 	inp->inp_flags |= INP_ANONPORT;
305 
306 	if (inp->inp_flags & INP_HIGHPORT) {
307 		first = ipport_hifirstauto;	/* sysctl */
308 		last  = ipport_hilastauto;
309 		lastport = &portinfo->lasthi;
310 	} else if (inp->inp_flags & INP_LOWPORT) {
311 		if (cred &&
312 		    (error =
313 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
314 			inp->inp_laddr.s_addr = INADDR_ANY;
315 			return error;
316 		}
317 		first = ipport_lowfirstauto;	/* 1023 */
318 		last  = ipport_lowlastauto;	/* 600 */
319 		lastport = &portinfo->lastlow;
320 	} else {
321 		first = ipport_firstauto;	/* sysctl */
322 		last  = ipport_lastauto;
323 		lastport = &portinfo->lastport;
324 	}
325 
326 	/*
327 	 * This has to be atomic.  If the porthash is shared across multiple
328 	 * protocol threads (aka tcp) then the token will be non-NULL.
329 	 */
330 	if (portinfo->porttoken)
331 		lwkt_gettoken(portinfo->porttoken);
332 
333 	/*
334 	 * Simple check to ensure all ports are not used up causing
335 	 * a deadlock here.
336 	 *
337 	 * We split the two cases (up and down) so that the direction
338 	 * is not being tested on each round of the loop.
339 	 */
340 	if (first > last) {
341 		/*
342 		 * counting down
343 		 */
344 		count = first - last;
345 
346 		do {
347 			if (count-- < 0) {	/* completely used? */
348 				inp->inp_laddr.s_addr = INADDR_ANY;
349 				error = EADDRNOTAVAIL;
350 				goto done;
351 			}
352 			--*lastport;
353 			if (*lastport > first || *lastport < last)
354 				*lastport = first;
355 			lport = htons(*lastport);
356 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
357 		    wild, cred));
358 	} else {
359 		/*
360 		 * counting up
361 		 */
362 		count = last - first;
363 
364 		do {
365 			if (count-- < 0) {	/* completely used? */
366 				inp->inp_laddr.s_addr = INADDR_ANY;
367 				error = EADDRNOTAVAIL;
368 				goto done;
369 			}
370 			++*lastport;
371 			if (*lastport < first || *lastport > last)
372 				*lastport = first;
373 			lport = htons(*lastport);
374 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
375 		    wild, cred));
376 	}
377 	inp->inp_lport = lport;
378 	in_pcbinsporthash(portinfo, inp);
379 	error = 0;
380 done:
381 	if (portinfo->porttoken)
382 		lwkt_reltoken(portinfo->porttoken);
383 	return error;
384 }
385 
386 int
387 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
388 {
389 	struct socket *so = inp->inp_socket;
390 	struct sockaddr_in jsin;
391 	struct ucred *cred = NULL;
392 	int wild = 0;
393 
394 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
395 		return (EADDRNOTAVAIL);
396 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
397 		return (EINVAL);	/* already bound */
398 
399 	if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
400 		wild = 1;    /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
401 	if (td->td_proc)
402 		cred = td->td_proc->p_ucred;
403 
404 	if (nam != NULL) {
405 		struct sockaddr_in *sin = (struct sockaddr_in *)nam;
406 		struct inpcbportinfo *portinfo;
407 		struct inpcb *t;
408 		u_short lport;
409 		int reuseport = (so->so_options & SO_REUSEPORT);
410 		int error;
411 
412 		if (nam->sa_len != sizeof *sin)
413 			return (EINVAL);
414 #ifdef notdef
415 		/*
416 		 * We should check the family, but old programs
417 		 * incorrectly fail to initialize it.
418 		 */
419 		if (sin->sin_family != AF_INET)
420 			return (EAFNOSUPPORT);
421 #endif
422 		if (!prison_replace_wildcards(td, nam))
423 			return (EINVAL);
424 
425 		lport = sin->sin_port;
426 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
427 			/*
428 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
429 			 * allow complete duplication of binding if
430 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
431 			 * and a multicast address is bound on both
432 			 * new and duplicated sockets.
433 			 */
434 			if (so->so_options & SO_REUSEADDR)
435 				reuseport = SO_REUSEADDR | SO_REUSEPORT;
436 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
437 			sin->sin_port = 0;		/* yech... */
438 			bzero(&sin->sin_zero, sizeof sin->sin_zero);
439 			if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
440 				return (EADDRNOTAVAIL);
441 		}
442 
443 		inp->inp_laddr = sin->sin_addr;
444 
445 		jsin.sin_family = AF_INET;
446 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
447 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
448 			inp->inp_laddr.s_addr = INADDR_ANY;
449 			return (EINVAL);
450 		}
451 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
452 
453 		if (lport == 0) {
454 			/* Auto-select local port */
455 			return in_pcbsetlport(inp, wild, cred);
456 		}
457 
458 		/* GROSS */
459 		if (ntohs(lport) < IPPORT_RESERVED && cred &&
460 		    (error =
461 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
462 			inp->inp_laddr.s_addr = INADDR_ANY;
463 			return (error);
464 		}
465 
466 		portinfo = inp->inp_pcbinfo->portinfo;
467 
468 		/*
469 		 * This has to be atomic.  If the porthash is shared across
470 		 * multiple protocol threads (aka tcp) then the token will
471 		 * be non-NULL.
472 		 */
473 		if (portinfo->porttoken)
474 			lwkt_gettoken(portinfo->porttoken);
475 
476 		if (so->so_cred->cr_uid != 0 &&
477 		    !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
478 			t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
479 			    INPLOOKUP_WILDCARD, cred);
480 			if (t &&
481 			    (!in_nullhost(sin->sin_addr) ||
482 			     !in_nullhost(t->inp_laddr) ||
483 			     (t->inp_socket->so_options & SO_REUSEPORT) == 0) &&
484 			    (so->so_cred->cr_uid !=
485 			     t->inp_socket->so_cred->cr_uid)) {
486 #ifdef INET6
487 				if (!in_nullhost(sin->sin_addr) ||
488 				    !in_nullhost(t->inp_laddr) ||
489 				    INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
490 #endif
491 				{
492 					inp->inp_laddr.s_addr = INADDR_ANY;
493 					error = EADDRINUSE;
494 					goto done;
495 				}
496 			}
497 		}
498 		if (cred && !prison_replace_wildcards(td, nam)) {
499 			inp->inp_laddr.s_addr = INADDR_ANY;
500 			error = EADDRNOTAVAIL;
501 			goto done;
502 		}
503 		t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
504 		    wild, cred);
505 		if (t && !(reuseport & t->inp_socket->so_options)) {
506 #ifdef INET6
507 			if (!in_nullhost(sin->sin_addr) ||
508 			    !in_nullhost(t->inp_laddr) ||
509 			    INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
510 #endif
511 			{
512 				inp->inp_laddr.s_addr = INADDR_ANY;
513 				error = EADDRINUSE;
514 				goto done;
515 			}
516 		}
517 		inp->inp_lport = lport;
518 		in_pcbinsporthash(portinfo, inp);
519 		error = 0;
520 done:
521 		if (portinfo->porttoken)
522 			lwkt_reltoken(portinfo->porttoken);
523 		return (error);
524 	} else {
525 		jsin.sin_family = AF_INET;
526 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
527 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
528 			inp->inp_laddr.s_addr = INADDR_ANY;
529 			return (EINVAL);
530 		}
531 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
532 
533 		return in_pcbsetlport(inp, wild, cred);
534 	}
535 }
536 
537 static struct inpcb *
538 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr,
539     u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
540 {
541 	struct inpcb *inp;
542 	struct inpcbporthead *porthash;
543 	struct inpcbport *phd;
544 	struct inpcb *match = NULL;
545 
546 	/*
547 	 * If the porthashbase is shared across several cpus, it must
548 	 * have been locked.
549 	 */
550 	if (portinfo->porttoken)
551 		ASSERT_LWKT_TOKEN_HELD(portinfo->porttoken);
552 
553 	/*
554 	 * Best fit PCB lookup.
555 	 *
556 	 * First see if this local port is in use by looking on the
557 	 * port hash list.
558 	 */
559 	porthash = &portinfo->porthashbase[
560 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
561 	LIST_FOREACH(phd, porthash, phd_hash) {
562 		if (phd->phd_port == lport)
563 			break;
564 	}
565 	if (phd != NULL) {
566 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
567 #ifdef INET6
568 			if ((inp->inp_vflag & INP_IPV4) == 0)
569 				continue;
570 #endif
571 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
572 			    inp->inp_laddr.s_addr != laddr.s_addr)
573 				continue;
574 
575 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
576 			    inp->inp_faddr.s_addr != faddr.s_addr)
577 				continue;
578 
579 			if (inp->inp_fport != 0 && inp->inp_fport != fport)
580 				continue;
581 
582 			if (cred == NULL ||
583 			    cred->cr_prison ==
584 			    inp->inp_socket->so_cred->cr_prison) {
585 				match = inp;
586 				break;
587 			}
588 		}
589 	}
590 	return (match);
591 }
592 
593 int
594 in_pcbsetlport_remote(struct inpcb *inp, const struct sockaddr *remote,
595     struct thread *td)
596 {
597 	struct proc *p = td->td_proc;
598 	unsigned short *lastport;
599 	const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
600 	struct sockaddr_in jsin;
601 	struct inpcbportinfo *portinfo = inp->inp_pcbinfo->portinfo;
602 	struct ucred *cred = NULL;
603 	u_short lport = 0;
604 	ushort first, last;
605 	int count, error, dup = 0;
606 
607 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
608 		return (EADDRNOTAVAIL);
609 
610 	KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
611 	if (inp->inp_lport != 0)
612 		return (EINVAL);	/* already bound */
613 
614 	KKASSERT(p);
615 	cred = p->p_ucred;
616 
617 	jsin.sin_family = AF_INET;
618 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
619 	if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
620 		inp->inp_laddr.s_addr = INADDR_ANY;
621 		return (EINVAL);
622 	}
623 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
624 
625 	inp->inp_flags |= INP_ANONPORT;
626 
627 	if (inp->inp_flags & INP_HIGHPORT) {
628 		first = ipport_hifirstauto;	/* sysctl */
629 		last  = ipport_hilastauto;
630 		lastport = &portinfo->lasthi;
631 	} else if (inp->inp_flags & INP_LOWPORT) {
632 		if (cred &&
633 		    (error =
634 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
635 			inp->inp_laddr.s_addr = INADDR_ANY;
636 			return (error);
637 		}
638 		first = ipport_lowfirstauto;	/* 1023 */
639 		last  = ipport_lowlastauto;	/* 600 */
640 		lastport = &portinfo->lastlow;
641 	} else {
642 		first = ipport_firstauto;	/* sysctl */
643 		last  = ipport_lastauto;
644 		lastport = &portinfo->lastport;
645 	}
646 
647 	/*
648 	 * This has to be atomic.  If the porthash is shared across multiple
649 	 * protocol threads (aka tcp) then the token will be non-NULL.
650 	 */
651 	if (portinfo->porttoken)
652 		lwkt_gettoken(portinfo->porttoken);
653 
654 again:
655 	/*
656 	 * Simple check to ensure all ports are not used up causing
657 	 * a deadlock here.
658 	 *
659 	 * We split the two cases (up and down) so that the direction
660 	 * is not being tested on each round of the loop.
661 	 */
662 	if (first > last) {
663 		/*
664 		 * counting down
665 		 */
666 		count = first - last;
667 
668 		do {
669 			if (count-- < 0) {	/* completely used? */
670 				inp->inp_laddr.s_addr = INADDR_ANY;
671 				error = EADDRNOTAVAIL;
672 				goto done;
673 			}
674 			--*lastport;
675 			if (*lastport > first || *lastport < last)
676 				*lastport = first;
677 			lport = htons(*lastport);
678 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
679 		    lport, sin->sin_addr, sin->sin_port, cred));
680 	} else {
681 		/*
682 		 * counting up
683 		 */
684 		count = last - first;
685 
686 		do {
687 			if (count-- < 0) {	/* completely used? */
688 				inp->inp_laddr.s_addr = INADDR_ANY;
689 				error = EADDRNOTAVAIL;
690 				goto done;
691 			}
692 			++*lastport;
693 			if (*lastport < first || *lastport > last)
694 				*lastport = first;
695 			lport = htons(*lastport);
696 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
697 		    lport, sin->sin_addr, sin->sin_port, cred));
698 	}
699 
700 	/* This could happen on loopback interface */
701 	if (sin->sin_port == lport &&
702 	    sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
703 		if (dup) {
704 			/*
705 			 * Duplicate again; give up
706 			 */
707 			inp->inp_laddr.s_addr = INADDR_ANY;
708 			error = EADDRNOTAVAIL;
709 			goto done;
710 		}
711 		dup = 1;
712 		goto again;
713 	}
714 	inp->inp_lport = lport;
715 	in_pcbinsporthash(portinfo, inp);
716 	error = 0;
717 done:
718 	if (portinfo->porttoken)
719 		lwkt_reltoken(portinfo->porttoken);
720 	return error;
721 }
722 
723 /*
724  *   Transform old in_pcbconnect() into an inner subroutine for new
725  *   in_pcbconnect(): Do some validity-checking on the remote
726  *   address (in mbuf 'nam') and then determine local host address
727  *   (i.e., which interface) to use to access that remote host.
728  *
729  *   This preserves definition of in_pcbconnect(), while supporting a
730  *   slightly different version for T/TCP.  (This is more than
731  *   a bit of a kludge, but cleaning up the internal interfaces would
732  *   have forced minor changes in every protocol).
733  */
734 int
735 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
736     struct sockaddr_in **plocal_sin, struct thread *td, int find)
737 {
738 	struct in_ifaddr *ia;
739 	struct ucred *cred = NULL;
740 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
741 	struct sockaddr *jsin;
742 	int jailed = 0, alloc_route = 0;
743 
744 	if (nam->sa_len != sizeof *sin)
745 		return (EINVAL);
746 	if (sin->sin_family != AF_INET)
747 		return (EAFNOSUPPORT);
748 	if (sin->sin_port == 0)
749 		return (EADDRNOTAVAIL);
750 	if (td && td->td_proc && td->td_proc->p_ucred)
751 		cred = td->td_proc->p_ucred;
752 	if (cred && cred->cr_prison)
753 		jailed = 1;
754 	if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
755 		ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
756 		/*
757 		 * If the destination address is INADDR_ANY,
758 		 * use the primary local address.
759 		 * If the supplied address is INADDR_BROADCAST,
760 		 * and the primary interface supports broadcast,
761 		 * choose the broadcast address for that interface.
762 		 */
763 		if (sin->sin_addr.s_addr == INADDR_ANY)
764 			sin->sin_addr = IA_SIN(ia)->sin_addr;
765 		else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
766 		    (ia->ia_ifp->if_flags & IFF_BROADCAST))
767 			sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
768 	}
769 	if (find) {
770 		struct route *ro;
771 
772 		ia = NULL;
773 		/*
774 		 * If route is known or can be allocated now,
775 		 * our src addr is taken from the i/f, else punt.
776 		 * Note that we should check the address family of the cached
777 		 * destination, in case of sharing the cache with IPv6.
778 		 */
779 		ro = &inp->inp_route;
780 		if (ro->ro_rt &&
781 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
782 		     ro->ro_dst.sa_family != AF_INET ||
783 		     satosin(&ro->ro_dst)->sin_addr.s_addr !=
784 				      sin->sin_addr.s_addr ||
785 		     inp->inp_socket->so_options & SO_DONTROUTE)) {
786 			RTFREE(ro->ro_rt);
787 			ro->ro_rt = NULL;
788 		}
789 		if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
790 		    (ro->ro_rt == NULL ||
791 		    ro->ro_rt->rt_ifp == NULL)) {
792 			/* No route yet, so try to acquire one */
793 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
794 			ro->ro_dst.sa_family = AF_INET;
795 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
796 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
797 				sin->sin_addr;
798 			rtalloc(ro);
799 			alloc_route = 1;
800 		}
801 		/*
802 		 * If we found a route, use the address
803 		 * corresponding to the outgoing interface
804 		 * unless it is the loopback (in case a route
805 		 * to our address on another net goes to loopback).
806 		 */
807 		if (ro->ro_rt && !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
808 			if (jailed) {
809 				if (jailed_ip(cred->cr_prison,
810 				    ro->ro_rt->rt_ifa->ifa_addr)) {
811 					ia = ifatoia(ro->ro_rt->rt_ifa);
812 				}
813 			} else {
814 				ia = ifatoia(ro->ro_rt->rt_ifa);
815 			}
816 		}
817 		if (ia == NULL) {
818 			u_short fport = sin->sin_port;
819 
820 			sin->sin_port = 0;
821 			ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
822 			if (ia && jailed && !jailed_ip(cred->cr_prison,
823 			    sintosa(&ia->ia_addr)))
824 				ia = NULL;
825 			if (ia == NULL)
826 				ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
827 			if (ia && jailed && !jailed_ip(cred->cr_prison,
828 			    sintosa(&ia->ia_addr)))
829 				ia = NULL;
830 			sin->sin_port = fport;
831 			if (ia == NULL &&
832 			    !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
833 				ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
834 			if (ia && jailed && !jailed_ip(cred->cr_prison,
835 			    sintosa(&ia->ia_addr)))
836 				ia = NULL;
837 
838 			if (!jailed && ia == NULL)
839 				goto fail;
840 		}
841 		/*
842 		 * If the destination address is multicast and an outgoing
843 		 * interface has been set as a multicast option, use the
844 		 * address of that interface as our source address.
845 		 */
846 		if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
847 		    inp->inp_moptions != NULL) {
848 			struct ip_moptions *imo;
849 			struct ifnet *ifp;
850 
851 			imo = inp->inp_moptions;
852 			if (imo->imo_multicast_ifp != NULL) {
853 				struct in_ifaddr_container *iac;
854 
855 				ifp = imo->imo_multicast_ifp;
856 				ia = NULL;
857 				TAILQ_FOREACH(iac,
858 				&in_ifaddrheads[mycpuid], ia_link) {
859 					if (iac->ia->ia_ifp == ifp) {
860 						ia = iac->ia;
861 						break;
862 					}
863 				}
864 				if (ia == NULL)
865 					goto fail;
866 			}
867 		}
868 		/*
869 		 * Don't do pcblookup call here; return interface in plocal_sin
870 		 * and exit to caller, that will do the lookup.
871 		 */
872 		if (ia == NULL && jailed) {
873 			if ((jsin = prison_get_nonlocal(cred->cr_prison, AF_INET, NULL)) != NULL ||
874 			    (jsin = prison_get_local(cred->cr_prison, AF_INET, NULL)) != NULL) {
875 				*plocal_sin = satosin(jsin);
876 			} else {
877 				/* IPv6 only Jail */
878 				goto fail;
879 			}
880 		} else {
881 			*plocal_sin = &ia->ia_addr;
882 		}
883 	}
884 	return (0);
885 fail:
886 	if (alloc_route) {
887 		struct route *ro = &inp->inp_route;
888 
889 		if (ro->ro_rt != NULL)
890 			RTFREE(ro->ro_rt);
891 		bzero(ro, sizeof(*ro));
892 	}
893 	return (EADDRNOTAVAIL);
894 }
895 
896 int
897 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
898     struct sockaddr_in **plocal_sin, struct thread *td)
899 {
900 	return in_pcbladdr_find(inp, nam, plocal_sin, td,
901 	    (inp->inp_laddr.s_addr == INADDR_ANY));
902 }
903 
904 /*
905  * Outer subroutine:
906  * Connect from a socket to a specified address.
907  * Both address and port must be specified in argument sin.
908  * If don't have a local address for this socket yet,
909  * then pick one.
910  */
911 int
912 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
913 {
914 	struct sockaddr_in *if_sin;
915 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
916 	int error;
917 
918 	/* Call inner routine to assign local interface address. */
919 	if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
920 		return (error);
921 
922 	if (in_pcblookup_hash(inp->inp_cpcbinfo, sin->sin_addr, sin->sin_port,
923 			      inp->inp_laddr.s_addr ?
924 				inp->inp_laddr : if_sin->sin_addr,
925 			      inp->inp_lport, FALSE, NULL) != NULL) {
926 		return (EADDRINUSE);
927 	}
928 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
929 		if (inp->inp_lport == 0) {
930 			error = in_pcbbind(inp, NULL, td);
931 			if (error)
932 				return (error);
933 		}
934 		inp->inp_laddr = if_sin->sin_addr;
935 	}
936 	inp->inp_faddr = sin->sin_addr;
937 	inp->inp_fport = sin->sin_port;
938 	in_pcbinsconnhash(inp);
939 	return (0);
940 }
941 
942 void
943 in_pcbdisconnect(struct inpcb *inp)
944 {
945 
946 	inp->inp_faddr.s_addr = INADDR_ANY;
947 	inp->inp_fport = 0;
948 	in_pcbremconnhash(inp);
949 	if (inp->inp_socket->so_state & SS_NOFDREF)
950 		in_pcbdetach(inp);
951 }
952 
953 void
954 in_pcbdetach(struct inpcb *inp)
955 {
956 	struct socket *so = inp->inp_socket;
957 	struct inpcbinfo *ipi = inp->inp_pcbinfo;
958 
959 #ifdef IPSEC
960 	ipsec4_delete_pcbpolicy(inp);
961 #endif /*IPSEC*/
962 	inp->inp_gencnt = ++ipi->ipi_gencnt;
963 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
964 	in_pcbremlists(inp);
965 	so->so_pcb = NULL;
966 	sofree(so);			/* remove pcb ref */
967 	if (inp->inp_options)
968 		m_free(inp->inp_options);
969 	if (inp->inp_route.ro_rt)
970 		rtfree(inp->inp_route.ro_rt);
971 	ip_freemoptions(inp->inp_moptions);
972 	inp->inp_vflag = 0;
973 	kfree(inp, M_PCB);
974 }
975 
976 /*
977  * The calling convention of in_setsockaddr() and in_setpeeraddr() was
978  * modified to match the pru_sockaddr() and pru_peeraddr() entry points
979  * in struct pr_usrreqs, so that protocols can just reference then directly
980  * without the need for a wrapper function.  The socket must have a valid
981  * (i.e., non-nil) PCB, but it should be impossible to get an invalid one
982  * except through a kernel programming error, so it is acceptable to panic
983  * (or in this case trap) if the PCB is invalid.  (Actually, we don't trap
984  * because there actually /is/ a programming error somewhere... XXX)
985  */
986 int
987 in_setsockaddr(struct socket *so, struct sockaddr **nam)
988 {
989 	struct inpcb *inp;
990 	struct sockaddr_in *sin;
991 
992 	/*
993 	 * Do the malloc first in case it blocks.
994 	 */
995 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
996 	sin->sin_family = AF_INET;
997 	sin->sin_len = sizeof *sin;
998 
999 	crit_enter();
1000 	inp = so->so_pcb;
1001 	if (!inp) {
1002 		crit_exit();
1003 		kfree(sin, M_SONAME);
1004 		return (ECONNRESET);
1005 	}
1006 	sin->sin_port = inp->inp_lport;
1007 	sin->sin_addr = inp->inp_laddr;
1008 	crit_exit();
1009 
1010 	*nam = (struct sockaddr *)sin;
1011 	return (0);
1012 }
1013 
1014 void
1015 in_setsockaddr_dispatch(netmsg_t msg)
1016 {
1017 	int error;
1018 
1019 	error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1020 	lwkt_replymsg(&msg->lmsg, error);
1021 }
1022 
1023 int
1024 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1025 {
1026 	struct inpcb *inp;
1027 	struct sockaddr_in *sin;
1028 
1029 	/*
1030 	 * Do the malloc first in case it blocks.
1031 	 */
1032 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1033 	sin->sin_family = AF_INET;
1034 	sin->sin_len = sizeof *sin;
1035 
1036 	crit_enter();
1037 	inp = so->so_pcb;
1038 	if (!inp) {
1039 		crit_exit();
1040 		kfree(sin, M_SONAME);
1041 		return (ECONNRESET);
1042 	}
1043 	sin->sin_port = inp->inp_fport;
1044 	sin->sin_addr = inp->inp_faddr;
1045 	crit_exit();
1046 
1047 	*nam = (struct sockaddr *)sin;
1048 	return (0);
1049 }
1050 
1051 void
1052 in_setpeeraddr_dispatch(netmsg_t msg)
1053 {
1054 	int error;
1055 
1056 	error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1057 	lwkt_replymsg(&msg->lmsg, error);
1058 }
1059 
1060 void
1061 in_pcbnotifyall(struct inpcbhead *head, struct in_addr faddr, int err,
1062 		void (*notify)(struct inpcb *, int))
1063 {
1064 	struct inpcb *inp, *ninp;
1065 
1066 	/*
1067 	 * note: if INP_PLACEMARKER is set we must ignore the rest of
1068 	 * the structure and skip it.
1069 	 */
1070 	crit_enter();
1071 	LIST_FOREACH_MUTABLE(inp, head, inp_list, ninp) {
1072 		if (inp->inp_flags & INP_PLACEMARKER)
1073 			continue;
1074 #ifdef INET6
1075 		if (!(inp->inp_vflag & INP_IPV4))
1076 			continue;
1077 #endif
1078 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1079 		    inp->inp_socket == NULL)
1080 			continue;
1081 		(*notify)(inp, err);		/* can remove inp from list! */
1082 	}
1083 	crit_exit();
1084 }
1085 
1086 void
1087 in_pcbpurgeif0(struct inpcb *head, struct ifnet *ifp)
1088 {
1089 	struct inpcb *inp;
1090 	struct ip_moptions *imo;
1091 	int i, gap;
1092 
1093 	for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) {
1094 		if (inp->inp_flags & INP_PLACEMARKER)
1095 			continue;
1096 		imo = inp->inp_moptions;
1097 		if ((inp->inp_vflag & INP_IPV4) && imo != NULL) {
1098 			/*
1099 			 * Unselect the outgoing interface if it is being
1100 			 * detached.
1101 			 */
1102 			if (imo->imo_multicast_ifp == ifp)
1103 				imo->imo_multicast_ifp = NULL;
1104 
1105 			/*
1106 			 * Drop multicast group membership if we joined
1107 			 * through the interface being detached.
1108 			 */
1109 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1110 			    i++) {
1111 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1112 					in_delmulti(imo->imo_membership[i]);
1113 					gap++;
1114 				} else if (gap != 0)
1115 					imo->imo_membership[i - gap] =
1116 					    imo->imo_membership[i];
1117 			}
1118 			imo->imo_num_memberships -= gap;
1119 		}
1120 	}
1121 }
1122 
1123 /*
1124  * Check for alternatives when higher level complains
1125  * about service problems.  For now, invalidate cached
1126  * routing information.  If the route was created dynamically
1127  * (by a redirect), time to try a default gateway again.
1128  */
1129 void
1130 in_losing(struct inpcb *inp)
1131 {
1132 	struct rtentry *rt;
1133 	struct rt_addrinfo rtinfo;
1134 
1135 	if ((rt = inp->inp_route.ro_rt)) {
1136 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
1137 		rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1138 		rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1139 		rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1140 		rtinfo.rti_flags = rt->rt_flags;
1141 		rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1142 		if (rt->rt_flags & RTF_DYNAMIC) {
1143 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1144 			    rt_mask(rt), rt->rt_flags, NULL);
1145 		}
1146 		inp->inp_route.ro_rt = NULL;
1147 		rtfree(rt);
1148 		/*
1149 		 * A new route can be allocated
1150 		 * the next time output is attempted.
1151 		 */
1152 	}
1153 }
1154 
1155 /*
1156  * After a routing change, flush old routing
1157  * and allocate a (hopefully) better one.
1158  */
1159 void
1160 in_rtchange(struct inpcb *inp, int err)
1161 {
1162 	if (inp->inp_route.ro_rt) {
1163 		rtfree(inp->inp_route.ro_rt);
1164 		inp->inp_route.ro_rt = NULL;
1165 		/*
1166 		 * A new route can be allocated the next time
1167 		 * output is attempted.
1168 		 */
1169 	}
1170 }
1171 
1172 /*
1173  * Lookup a PCB based on the local address and port.
1174  */
1175 struct inpcb *
1176 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr,
1177 		   u_int lport_arg, int wild_okay, struct ucred *cred)
1178 {
1179 	struct inpcb *inp;
1180 	int matchwild = 3, wildcard;
1181 	u_short lport = lport_arg;
1182 	struct inpcbporthead *porthash;
1183 	struct inpcbport *phd;
1184 	struct inpcb *match = NULL;
1185 
1186 	/*
1187 	 * If the porthashbase is shared across several cpus, it must
1188 	 * have been locked.
1189 	 */
1190 	if (portinfo->porttoken)
1191 		ASSERT_LWKT_TOKEN_HELD(portinfo->porttoken);
1192 
1193 	/*
1194 	 * Best fit PCB lookup.
1195 	 *
1196 	 * First see if this local port is in use by looking on the
1197 	 * port hash list.
1198 	 */
1199 	porthash = &portinfo->porthashbase[
1200 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
1201 	LIST_FOREACH(phd, porthash, phd_hash) {
1202 		if (phd->phd_port == lport)
1203 			break;
1204 	}
1205 	if (phd != NULL) {
1206 		/*
1207 		 * Port is in use by one or more PCBs. Look for best
1208 		 * fit.
1209 		 */
1210 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1211 			wildcard = 0;
1212 #ifdef INET6
1213 			if ((inp->inp_vflag & INP_IPV4) == 0)
1214 				continue;
1215 #endif
1216 			if (inp->inp_faddr.s_addr != INADDR_ANY)
1217 				wildcard++;
1218 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
1219 				if (laddr.s_addr == INADDR_ANY)
1220 					wildcard++;
1221 				else if (inp->inp_laddr.s_addr != laddr.s_addr)
1222 					continue;
1223 			} else {
1224 				if (laddr.s_addr != INADDR_ANY)
1225 					wildcard++;
1226 			}
1227 			if (wildcard && !wild_okay)
1228 				continue;
1229 			if (wildcard < matchwild &&
1230 			    (cred == NULL ||
1231 			     cred->cr_prison ==
1232 					inp->inp_socket->so_cred->cr_prison)) {
1233 				match = inp;
1234 				matchwild = wildcard;
1235 				if (matchwild == 0) {
1236 					break;
1237 				}
1238 			}
1239 		}
1240 	}
1241 	return (match);
1242 }
1243 
1244 struct inpcb *
1245 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1246     const struct inpcb *inp)
1247 {
1248 	const struct inp_localgrphead *hdr;
1249 	const struct inp_localgroup *grp;
1250 	int i;
1251 
1252 	if (pcbinfo->localgrphashbase == NULL)
1253 		return NULL;
1254 
1255 	hdr = &pcbinfo->localgrphashbase[
1256 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1257 
1258 	LIST_FOREACH(grp, hdr, il_list) {
1259 		if (grp->il_vflag == inp->inp_vflag &&
1260 		    grp->il_lport == inp->inp_lport &&
1261 		    memcmp(&grp->il_dependladdr,
1262 			&inp->inp_inc.inc_ie.ie_dependladdr,
1263 			sizeof(grp->il_dependladdr)) == 0) {
1264 			break;
1265 		}
1266 	}
1267 	if (grp == NULL || grp->il_inpcnt == 1)
1268 		return NULL;
1269 
1270 	KASSERT(grp->il_inpcnt >= 2,
1271 	    ("invalid localgroup inp count %d", grp->il_inpcnt));
1272 	for (i = 0; i < grp->il_inpcnt; ++i) {
1273 		if (grp->il_inp[i] == inp) {
1274 			int last = grp->il_inpcnt - 1;
1275 
1276 			if (i == last)
1277 				last = grp->il_inpcnt - 2;
1278 			return grp->il_inp[last];
1279 		}
1280 	}
1281 	return NULL;
1282 }
1283 
1284 static struct inpcb *
1285 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1286     struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1287 {
1288 	struct inpcb *local_wild = NULL;
1289 	const struct inp_localgrphead *hdr;
1290 	const struct inp_localgroup *grp;
1291 
1292 	hdr = &pcbinfo->localgrphashbase[
1293 	    INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1294 #ifdef INP_LOCALGROUP_HASHTHR
1295 	pkt_hash >>= ncpus2_shift;
1296 #endif
1297 
1298 	/*
1299 	 * Order of socket selection:
1300 	 * 1. non-wild.
1301 	 * 2. wild.
1302 	 *
1303 	 * NOTE:
1304 	 * - Local group does not contain jailed sockets
1305 	 * - Local group does not contain IPv4 mapped INET6 wild sockets
1306 	 */
1307 	LIST_FOREACH(grp, hdr, il_list) {
1308 #ifdef INET6
1309 		if (!(grp->il_vflag & INP_IPV4))
1310 			continue;
1311 #endif
1312 		if (grp->il_lport == lport) {
1313 			int idx;
1314 
1315 #ifdef INP_LOCALGROUP_HASHTHR
1316 			idx = pkt_hash / grp->il_factor;
1317 			KASSERT(idx < grp->il_inpcnt && idx >= 0,
1318 			    ("invalid hash %04x, cnt %d or fact %d",
1319 			     pkt_hash, grp->il_inpcnt, grp->il_factor));
1320 #else
1321 			/*
1322 			 * Modulo-N is used here, which greatly reduces
1323 			 * completion queue token contention, thus more
1324 			 * cpu time is saved.
1325 			 */
1326 			idx = pkt_hash % grp->il_inpcnt;
1327 #endif
1328 
1329 			if (grp->il_laddr.s_addr == laddr.s_addr)
1330 				return grp->il_inp[idx];
1331 			else if (grp->il_laddr.s_addr == INADDR_ANY)
1332 				local_wild = grp->il_inp[idx];
1333 		}
1334 	}
1335 	if (local_wild != NULL)
1336 		return local_wild;
1337 	return NULL;
1338 }
1339 
1340 /*
1341  * Lookup PCB in hash list.
1342  */
1343 struct inpcb *
1344 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1345     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1346     boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1347 {
1348 	struct inpcbhead *head;
1349 	struct inpcb *inp, *jinp=NULL;
1350 	u_short fport = fport_arg, lport = lport_arg;
1351 
1352 	/*
1353 	 * First look for an exact match.
1354 	 */
1355 	head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1356 	    laddr.s_addr, lport, pcbinfo->hashmask)];
1357 	LIST_FOREACH(inp, head, inp_hash) {
1358 #ifdef INET6
1359 		if (!(inp->inp_vflag & INP_IPV4))
1360 			continue;
1361 #endif
1362 		if (in_hosteq(inp->inp_faddr, faddr) &&
1363 		    in_hosteq(inp->inp_laddr, laddr) &&
1364 		    inp->inp_fport == fport && inp->inp_lport == lport) {
1365 			/* found */
1366 			if (inp->inp_socket == NULL ||
1367 			    inp->inp_socket->so_cred->cr_prison == NULL) {
1368 				return (inp);
1369 			} else {
1370 				if  (jinp == NULL)
1371 					jinp = inp;
1372 			}
1373 		}
1374 	}
1375 	if (jinp != NULL)
1376 		return (jinp);
1377 	if (wildcard) {
1378 		struct inpcb *local_wild = NULL;
1379 		struct inpcb *jinp_wild = NULL;
1380 #ifdef INET6
1381 		struct inpcb *local_wild_mapped = NULL;
1382 #endif
1383 		struct inpcontainer *ic;
1384 		struct inpcontainerhead *chead;
1385 		struct sockaddr_in jsin;
1386 		struct ucred *cred;
1387 
1388 		/*
1389 		 * Check local group first
1390 		 */
1391 		if (pcbinfo->localgrphashbase != NULL &&
1392 		    m != NULL && (m->m_flags & M_HASH) &&
1393 		    !(ifp && ifp->if_type == IFT_FAITH)) {
1394 			inp = inp_localgroup_lookup(pcbinfo,
1395 			    laddr, lport, m->m_pkthdr.hash);
1396 			if (inp != NULL)
1397 				return inp;
1398 		}
1399 
1400 		/*
1401 		 * Order of socket selection:
1402 		 * 1. non-jailed, non-wild.
1403 		 * 2. non-jailed, wild.
1404 		 * 3. jailed, non-wild.
1405 		 * 4. jailed, wild.
1406 		 */
1407 		jsin.sin_family = AF_INET;
1408 		chead = &pcbinfo->wildcardhashbase[
1409 		    INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1410 		LIST_FOREACH(ic, chead, ic_list) {
1411 			inp = ic->ic_inp;
1412 			jsin.sin_addr.s_addr = laddr.s_addr;
1413 #ifdef INET6
1414 			if (!(inp->inp_vflag & INP_IPV4))
1415 				continue;
1416 #endif
1417 			if (inp->inp_socket != NULL)
1418 				cred = inp->inp_socket->so_cred;
1419 			else
1420 				cred = NULL;
1421 			if (cred != NULL && jailed(cred)) {
1422 				if (jinp != NULL)
1423 					continue;
1424 				else
1425 					if (!jailed_ip(cred->cr_prison,
1426 					    (struct sockaddr *)&jsin))
1427 						continue;
1428 			}
1429 			if (inp->inp_lport == lport) {
1430 				if (ifp && ifp->if_type == IFT_FAITH &&
1431 				    !(inp->inp_flags & INP_FAITH))
1432 					continue;
1433 				if (inp->inp_laddr.s_addr == laddr.s_addr) {
1434 					if (cred != NULL && jailed(cred))
1435 						jinp = inp;
1436 					else
1437 						return (inp);
1438 				}
1439 				if (inp->inp_laddr.s_addr == INADDR_ANY) {
1440 #ifdef INET6
1441 					if (INP_CHECK_SOCKAF(inp->inp_socket,
1442 							     AF_INET6))
1443 						local_wild_mapped = inp;
1444 					else
1445 #endif
1446 						if (cred != NULL &&
1447 						    jailed(cred))
1448 							jinp_wild = inp;
1449 						else
1450 							local_wild = inp;
1451 				}
1452 			}
1453 		}
1454 		if (local_wild != NULL)
1455 			return (local_wild);
1456 #ifdef INET6
1457 		if (local_wild_mapped != NULL)
1458 			return (local_wild_mapped);
1459 #endif
1460 		if (jinp != NULL)
1461 			return (jinp);
1462 		return (jinp_wild);
1463 	}
1464 
1465 	/*
1466 	 * Not found.
1467 	 */
1468 	return (NULL);
1469 }
1470 
1471 struct inpcb *
1472 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1473     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1474     boolean_t wildcard, struct ifnet *ifp)
1475 {
1476 	return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1477 	    laddr, lport_arg, wildcard, ifp, NULL);
1478 }
1479 
1480 /*
1481  * Insert PCB into connection hash table.
1482  */
1483 void
1484 in_pcbinsconnhash(struct inpcb *inp)
1485 {
1486 	struct inpcbinfo *pcbinfo = inp->inp_cpcbinfo;
1487 	struct inpcbhead *bucket;
1488 	u_int32_t hashkey_faddr, hashkey_laddr;
1489 
1490 #ifdef INET6
1491 	if (inp->inp_vflag & INP_IPV6) {
1492 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1493 		hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1494 	} else {
1495 #endif
1496 		hashkey_faddr = inp->inp_faddr.s_addr;
1497 		hashkey_laddr = inp->inp_laddr.s_addr;
1498 #ifdef INET6
1499 	}
1500 #endif
1501 
1502 	KASSERT(!(inp->inp_flags & INP_WILDCARD),
1503 		("already on wildcardhash"));
1504 	KASSERT(!(inp->inp_flags & INP_CONNECTED),
1505 		("already on connhash"));
1506 	inp->inp_flags |= INP_CONNECTED;
1507 
1508 	/*
1509 	 * Insert into the connection hash table.
1510 	 */
1511 	bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1512 	    inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1513 	LIST_INSERT_HEAD(bucket, inp, inp_hash);
1514 }
1515 
1516 /*
1517  * Remove PCB from connection hash table.
1518  */
1519 void
1520 in_pcbremconnhash(struct inpcb *inp)
1521 {
1522 	KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1523 	LIST_REMOVE(inp, inp_hash);
1524 	inp->inp_flags &= ~INP_CONNECTED;
1525 }
1526 
1527 /*
1528  * Insert PCB into port hash table.
1529  */
1530 void
1531 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp)
1532 {
1533 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1534 	struct inpcbporthead *pcbporthash;
1535 	struct inpcbport *phd;
1536 
1537 	/*
1538 	 * If the porthashbase is shared across several cpus, it must
1539 	 * have been locked.
1540 	 */
1541 	if (portinfo->porttoken)
1542 		ASSERT_LWKT_TOKEN_HELD(portinfo->porttoken);
1543 
1544 	/*
1545 	 * Insert into the port hash table.
1546 	 */
1547 	pcbporthash = &portinfo->porthashbase[
1548 	    INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)];
1549 
1550 	/* Go through port list and look for a head for this lport. */
1551 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
1552 		if (phd->phd_port == inp->inp_lport)
1553 			break;
1554 	}
1555 
1556 	/* If none exists, malloc one and tack it on. */
1557 	if (phd == NULL) {
1558 		KKASSERT(pcbinfo->portsave != NULL);
1559 		phd = pcbinfo->portsave;
1560 		pcbinfo->portsave = NULL;
1561 		phd->phd_port = inp->inp_lport;
1562 		LIST_INIT(&phd->phd_pcblist);
1563 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1564 	}
1565 
1566 	inp->inp_portinfo = portinfo;
1567 	inp->inp_phd = phd;
1568 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1569 
1570 	if (pcbinfo->portsave == NULL) {
1571 		pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1572 					    M_PCB, M_INTWAIT | M_ZERO);
1573 	}
1574 }
1575 
1576 void
1577 in_pcbinsporthash_lport(struct inpcb *inp)
1578 {
1579 	struct inpcbportinfo *portinfo = inp->inp_pcbinfo->portinfo;
1580 
1581 	if (portinfo->porttoken)
1582 		lwkt_gettoken(portinfo->porttoken);
1583 
1584 	in_pcbinsporthash(portinfo, inp);
1585 
1586 	if (portinfo->porttoken)
1587 		lwkt_reltoken(portinfo->porttoken);
1588 }
1589 
1590 static struct inp_localgroup *
1591 inp_localgroup_alloc(struct inp_localgrphead *hdr, u_char vflag,
1592     uint16_t port, const union in_dependaddr *addr, int size)
1593 {
1594 	struct inp_localgroup *grp;
1595 
1596 	grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1597 	    M_TEMP, M_INTWAIT | M_ZERO);
1598 	grp->il_vflag = vflag;
1599 	grp->il_lport = port;
1600 	grp->il_dependladdr = *addr;
1601 	grp->il_inpsiz = size;
1602 
1603 	LIST_INSERT_HEAD(hdr, grp, il_list);
1604 
1605 	return grp;
1606 }
1607 
1608 static void
1609 inp_localgroup_free(struct inp_localgroup *grp)
1610 {
1611 	LIST_REMOVE(grp, il_list);
1612 	kfree(grp, M_TEMP);
1613 }
1614 
1615 static struct inp_localgroup *
1616 inp_localgroup_resize(struct inp_localgrphead *hdr,
1617     struct inp_localgroup *old_grp, int size)
1618 {
1619 	struct inp_localgroup *grp;
1620 	int i;
1621 
1622 	grp = inp_localgroup_alloc(hdr, old_grp->il_vflag,
1623 	    old_grp->il_lport, &old_grp->il_dependladdr, size);
1624 
1625 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1626 	    ("invalid new local group size %d and old local group count %d",
1627 	     grp->il_inpsiz, old_grp->il_inpcnt));
1628 	for (i = 0; i < old_grp->il_inpcnt; ++i)
1629 		grp->il_inp[i] = old_grp->il_inp[i];
1630 	grp->il_inpcnt = old_grp->il_inpcnt;
1631 	grp->il_factor = old_grp->il_factor;
1632 
1633 	inp_localgroup_free(old_grp);
1634 
1635 	return grp;
1636 }
1637 
1638 static void
1639 inp_localgroup_factor(struct inp_localgroup *grp)
1640 {
1641 	grp->il_factor =
1642 	    ((uint32_t)(0xffff >> ncpus2_shift) / grp->il_inpcnt) + 1;
1643 	KASSERT(grp->il_factor != 0, ("invalid local group factor, "
1644 	    "ncpus2_shift %d, inpcnt %d", ncpus2_shift, grp->il_inpcnt));
1645 }
1646 
1647 static void
1648 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1649 {
1650 	struct inp_localgrphead *hdr;
1651 	struct inp_localgroup *grp;
1652 	struct ucred *cred;
1653 
1654 	if (pcbinfo->localgrphashbase == NULL)
1655 		return;
1656 
1657 	/*
1658 	 * XXX don't allow jailed socket to join local group
1659 	 */
1660 	if (inp->inp_socket != NULL)
1661 		cred = inp->inp_socket->so_cred;
1662 	else
1663 		cred = NULL;
1664 	if (cred != NULL && jailed(cred))
1665 		return;
1666 
1667 #ifdef INET6
1668 	/*
1669 	 * XXX don't allow IPv4 mapped INET6 wild socket
1670 	 */
1671 	if ((inp->inp_vflag & INP_IPV4) &&
1672 	    inp->inp_laddr.s_addr == INADDR_ANY &&
1673 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6))
1674 		return;
1675 #endif
1676 
1677 	hdr = &pcbinfo->localgrphashbase[
1678 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1679 
1680 	LIST_FOREACH(grp, hdr, il_list) {
1681 		if (grp->il_vflag == inp->inp_vflag &&
1682 		    grp->il_lport == inp->inp_lport &&
1683 		    memcmp(&grp->il_dependladdr,
1684 		        &inp->inp_inc.inc_ie.ie_dependladdr,
1685 		        sizeof(grp->il_dependladdr)) == 0) {
1686 			break;
1687 		}
1688 	}
1689 	if (grp == NULL) {
1690 		/* Create new local group */
1691 		grp = inp_localgroup_alloc(hdr, inp->inp_vflag,
1692 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1693 		    INP_LOCALGROUP_SIZMIN);
1694 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
1695 		if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1696 			static int limit_logged = 0;
1697 
1698 			if (!limit_logged) {
1699 				limit_logged = 1;
1700 				kprintf("local group port %d, "
1701 				    "limit reached\n", ntohs(grp->il_lport));
1702 			}
1703 			return;
1704 		}
1705 
1706 		/* Expand this local group */
1707 		grp = inp_localgroup_resize(hdr, grp, grp->il_inpsiz * 2);
1708 	}
1709 
1710 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
1711 	    ("invalid local group size %d and count %d",
1712 	     grp->il_inpsiz, grp->il_inpcnt));
1713 	grp->il_inp[grp->il_inpcnt] = inp;
1714 	grp->il_inpcnt++;
1715 	inp_localgroup_factor(grp);
1716 }
1717 
1718 void
1719 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1720 {
1721 	struct inpcontainer *ic;
1722 	struct inpcontainerhead *bucket;
1723 
1724 	in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
1725 
1726 	bucket = &pcbinfo->wildcardhashbase[
1727 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1728 
1729 	ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
1730 	ic->ic_inp = inp;
1731 	LIST_INSERT_HEAD(bucket, ic, ic_list);
1732 }
1733 
1734 /*
1735  * Insert PCB into wildcard hash table.
1736  */
1737 void
1738 in_pcbinswildcardhash(struct inpcb *inp)
1739 {
1740 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1741 
1742 	KASSERT(!(inp->inp_flags & INP_CONNECTED),
1743 		("already on connhash"));
1744 	KASSERT(!(inp->inp_flags & INP_WILDCARD),
1745 		("already on wildcardhash"));
1746 	inp->inp_flags |= INP_WILDCARD;
1747 
1748 	in_pcbinswildcardhash_oncpu(inp, pcbinfo);
1749 }
1750 
1751 static void
1752 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1753 {
1754 	struct inp_localgrphead *hdr;
1755 	struct inp_localgroup *grp;
1756 
1757 	if (pcbinfo->localgrphashbase == NULL)
1758 		return;
1759 
1760 	hdr = &pcbinfo->localgrphashbase[
1761 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1762 
1763 	LIST_FOREACH(grp, hdr, il_list) {
1764 		int i;
1765 
1766 		for (i = 0; i < grp->il_inpcnt; ++i) {
1767 			if (grp->il_inp[i] != inp)
1768 				continue;
1769 
1770 			if (grp->il_inpcnt == 1) {
1771 				/* Free this local group */
1772 				inp_localgroup_free(grp);
1773 			} else {
1774 				/* Pull up inpcbs */
1775 				for (; i + 1 < grp->il_inpcnt; ++i)
1776 					grp->il_inp[i] = grp->il_inp[i + 1];
1777 				grp->il_inpcnt--;
1778 				inp_localgroup_factor(grp);
1779 
1780 				if (grp->il_inpsiz > INP_LOCALGROUP_SIZMIN &&
1781 				    grp->il_inpcnt <= (grp->il_inpsiz / 4)) {
1782 					/* Shrink this local group */
1783 					grp = inp_localgroup_resize(hdr, grp,
1784 					    grp->il_inpsiz / 2);
1785 				}
1786 			}
1787 			return;
1788 		}
1789 	}
1790 }
1791 
1792 void
1793 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1794 {
1795 	struct inpcontainer *ic;
1796 	struct inpcontainerhead *head;
1797 
1798 	in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
1799 
1800 	/* find bucket */
1801 	head = &pcbinfo->wildcardhashbase[
1802 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1803 
1804 	LIST_FOREACH(ic, head, ic_list) {
1805 		if (ic->ic_inp == inp)
1806 			goto found;
1807 	}
1808 	return;			/* not found! */
1809 
1810 found:
1811 	LIST_REMOVE(ic, ic_list);	/* remove container from bucket chain */
1812 	kfree(ic, M_TEMP);		/* deallocate container */
1813 }
1814 
1815 /*
1816  * Remove PCB from wildcard hash table.
1817  */
1818 void
1819 in_pcbremwildcardhash(struct inpcb *inp)
1820 {
1821 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1822 
1823 	KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
1824 	in_pcbremwildcardhash_oncpu(inp, pcbinfo);
1825 	inp->inp_flags &= ~INP_WILDCARD;
1826 }
1827 
1828 /*
1829  * Remove PCB from various lists.
1830  */
1831 void
1832 in_pcbremlists(struct inpcb *inp)
1833 {
1834 	if (inp->inp_lport) {
1835 		struct inpcbportinfo *portinfo;
1836 		struct inpcbport *phd;
1837 
1838 		/*
1839 		 * NOTE:
1840 		 * inp->inp_portinfo is _not_ necessary same as
1841 		 * inp->inp_pcbinfo->portinfo.
1842 		 */
1843 		portinfo = inp->inp_portinfo;
1844 		if (portinfo->porttoken)
1845 			lwkt_gettoken(portinfo->porttoken);
1846 
1847 		phd = inp->inp_phd;
1848 		LIST_REMOVE(inp, inp_portlist);
1849 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1850 			LIST_REMOVE(phd, phd_hash);
1851 			kfree(phd, M_PCB);
1852 		}
1853 		if (portinfo->porttoken)
1854 			lwkt_reltoken(portinfo->porttoken);
1855 	}
1856 	if (inp->inp_flags & INP_WILDCARD) {
1857 		in_pcbremwildcardhash(inp);
1858 	} else if (inp->inp_flags & INP_CONNECTED) {
1859 		in_pcbremconnhash(inp);
1860 	}
1861 	LIST_REMOVE(inp, inp_list);
1862 	inp->inp_pcbinfo->ipi_count--;
1863 }
1864 
1865 int
1866 prison_xinpcb(struct thread *td, struct inpcb *inp)
1867 {
1868 	struct ucred *cr;
1869 
1870 	if (td->td_proc == NULL)
1871 		return (0);
1872 	cr = td->td_proc->p_ucred;
1873 	if (cr->cr_prison == NULL)
1874 		return (0);
1875 	if (inp->inp_socket && inp->inp_socket->so_cred &&
1876 	    inp->inp_socket->so_cred->cr_prison &&
1877 	    cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
1878 		return (0);
1879 	return (1);
1880 }
1881 
1882 int
1883 in_pcblist_global(SYSCTL_HANDLER_ARGS)
1884 {
1885 	struct inpcbinfo *pcbinfo = arg1;
1886 	struct inpcb *inp, *marker;
1887 	struct xinpcb xi;
1888 	int error, i, n;
1889 
1890 	/*
1891 	 * The process of preparing the TCB list is too time-consuming and
1892 	 * resource-intensive to repeat twice on every request.
1893 	 */
1894 	if (req->oldptr == NULL) {
1895 		n = pcbinfo->ipi_count;
1896 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
1897 		return 0;
1898 	}
1899 
1900 	if (req->newptr != NULL)
1901 		return EPERM;
1902 
1903 	/*
1904 	 * OK, now we're committed to doing something.  Re-fetch ipi_count
1905 	 * after obtaining the generation count.
1906 	 */
1907 	n = pcbinfo->ipi_count;
1908 
1909 	marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
1910 	marker->inp_flags |= INP_PLACEMARKER;
1911 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1912 
1913 	i = 0;
1914 	error = 0;
1915 
1916 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
1917 		LIST_REMOVE(marker, inp_list);
1918 		LIST_INSERT_AFTER(inp, marker, inp_list);
1919 
1920 		if (inp->inp_flags & INP_PLACEMARKER)
1921 			continue;
1922 		if (prison_xinpcb(req->td, inp))
1923 			continue;
1924 		bzero(&xi, sizeof xi);
1925 		xi.xi_len = sizeof xi;
1926 		bcopy(inp, &xi.xi_inp, sizeof *inp);
1927 		if (inp->inp_socket)
1928 			sotoxsocket(inp->inp_socket, &xi.xi_socket);
1929 		if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
1930 			break;
1931 		++i;
1932 	}
1933 	LIST_REMOVE(marker, inp_list);
1934 	if (error == 0 && i < n) {
1935 		bzero(&xi, sizeof xi);
1936 		xi.xi_len = sizeof xi;
1937 		while (i < n) {
1938 			error = SYSCTL_OUT(req, &xi, sizeof xi);
1939 			++i;
1940 		}
1941 	}
1942 	kfree(marker, M_TEMP);
1943 	return(error);
1944 }
1945 
1946 int
1947 in_pcblist_global_nomarker(SYSCTL_HANDLER_ARGS, struct xinpcb **xi0, int *nxi0)
1948 {
1949 	struct inpcbinfo *pcbinfo = arg1;
1950 	struct inpcb *inp;
1951 	struct xinpcb *xi;
1952 	int nxi;
1953 
1954 	*nxi0 = 0;
1955 	*xi0 = NULL;
1956 
1957 	/*
1958 	 * The process of preparing the PCB list is too time-consuming and
1959 	 * resource-intensive to repeat twice on every request.
1960 	 */
1961 	if (req->oldptr == NULL) {
1962 		int n = pcbinfo->ipi_count;
1963 
1964 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
1965 		return 0;
1966 	}
1967 
1968 	if (req->newptr != NULL)
1969 		return EPERM;
1970 
1971 	if (pcbinfo->ipi_count == 0)
1972 		return 0;
1973 
1974 	nxi = 0;
1975 	xi = kmalloc(pcbinfo->ipi_count * sizeof(*xi), M_TEMP,
1976 		     M_WAITOK | M_ZERO | M_NULLOK);
1977 	if (xi == NULL)
1978 		return ENOMEM;
1979 
1980 	LIST_FOREACH(inp, &pcbinfo->pcblisthead, inp_list) {
1981 		struct xinpcb *xi_ptr = &xi[nxi];
1982 
1983 		if (prison_xinpcb(req->td, inp))
1984 			continue;
1985 
1986 		xi_ptr->xi_len = sizeof(*xi_ptr);
1987 		bcopy(inp, &xi_ptr->xi_inp, sizeof(*inp));
1988 		if (inp->inp_socket)
1989 			sotoxsocket(inp->inp_socket, &xi_ptr->xi_socket);
1990 		++nxi;
1991 	}
1992 
1993 	if (nxi == 0) {
1994 		kfree(xi, M_TEMP);
1995 		return 0;
1996 	}
1997 
1998 	*nxi0 = nxi;
1999 	*xi0 = xi;
2000 
2001 	return 0;
2002 }
2003 
2004 void
2005 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2006 {
2007 	struct sockaddr_in *sin;
2008 
2009 	KASSERT(faddr->sa_family == AF_INET,
2010 	    ("not AF_INET faddr %d", faddr->sa_family));
2011 
2012 	sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2013 	sin->sin_family = AF_INET;
2014 	sin->sin_len = sizeof(*sin);
2015 	sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2016 	sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2017 
2018 	so->so_faddr = (struct sockaddr *)sin;
2019 }
2020 
2021 void
2022 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2023     boolean_t shared)
2024 {
2025 	memset(portinfo, 0, sizeof(*portinfo));
2026 	portinfo->porthashbase = hashinit(hashsize, M_PCB,
2027 	    &portinfo->porthashmask);
2028 	if (shared) {
2029 		portinfo->porttoken = kmalloc(sizeof(struct lwkt_token),
2030 		    M_PCB, M_WAITOK);
2031 		lwkt_token_init(portinfo->porttoken, "porttoken");
2032 	}
2033 }
2034