xref: /dflybsd-src/sys/netinet/in_pcb.c (revision d435571a44e4312d9f604571232dad19c5cfe9f7)
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. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
67  * $FreeBSD: src/sys/netinet/in_pcb.c,v 1.59.2.27 2004/01/02 04:06:42 ambrisko Exp $
68  * $DragonFly: src/sys/netinet/in_pcb.c,v 1.48 2008/11/08 03:38:23 sephe Exp $
69  */
70 
71 #include "opt_ipsec.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/domain.h>
79 #include <sys/protosw.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/proc.h>
83 #include <sys/priv.h>
84 #include <sys/jail.h>
85 #include <sys/kernel.h>
86 #include <sys/sysctl.h>
87 
88 #include <sys/thread2.h>
89 #include <sys/socketvar2.h>
90 #include <sys/msgport2.h>
91 
92 #include <machine/limits.h>
93 
94 #include <net/if.h>
95 #include <net/if_types.h>
96 #include <net/route.h>
97 
98 #include <netinet/in.h>
99 #include <netinet/in_pcb.h>
100 #include <netinet/in_var.h>
101 #include <netinet/ip_var.h>
102 #ifdef INET6
103 #include <netinet/ip6.h>
104 #include <netinet6/ip6_var.h>
105 #endif /* INET6 */
106 
107 #ifdef IPSEC
108 #include <netinet6/ipsec.h>
109 #include <netproto/key/key.h>
110 #include <netproto/ipsec/esp_var.h>
111 #endif
112 
113 #ifdef FAST_IPSEC
114 #if defined(IPSEC) || defined(IPSEC_ESP)
115 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!"
116 #endif
117 
118 #include <netproto/ipsec/ipsec.h>
119 #include <netproto/ipsec/key.h>
120 #define	IPSEC
121 #endif /* FAST_IPSEC */
122 
123 struct in_addr zeroin_addr;
124 
125 /*
126  * These configure the range of local port addresses assigned to
127  * "unspecified" outgoing connections/packets/whatever.
128  */
129 int ipport_lowfirstauto = IPPORT_RESERVED - 1;	/* 1023 */
130 int ipport_lowlastauto = IPPORT_RESERVEDSTART;	/* 600 */
131 
132 int ipport_firstauto = IPPORT_RESERVED;		/* 1024 */
133 int ipport_lastauto = IPPORT_USERRESERVED;	/* 5000 */
134 
135 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO;	/* 49152 */
136 int ipport_hilastauto = IPPORT_HILASTAUTO;	/* 65535 */
137 
138 #define RANGECHK(var, min, max) \
139 	if ((var) < (min)) { (var) = (min); } \
140 	else if ((var) > (max)) { (var) = (max); }
141 
142 int udpencap_enable = 1;	/* enabled by default */
143 int udpencap_port = 4500;	/* triggers decapsulation */
144 
145 static int
146 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
147 {
148 	int error;
149 
150 	error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
151 	if (!error) {
152 		RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
153 		RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
154 
155 		RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
156 		RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
157 
158 		RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
159 		RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
160 	}
161 	return (error);
162 }
163 
164 #undef RANGECHK
165 
166 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
167 
168 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
169 	   &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
171 	   &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
173 	   &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
175 	   &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
177 	   &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
179 	   &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
180 
181 /*
182  * in_pcb.c: manage the Protocol Control Blocks.
183  *
184  * NOTE: It is assumed that most of these functions will be called from
185  * a critical section.  XXX - There are, unfortunately, a few exceptions
186  * to this rule that should be fixed.
187  *
188  * NOTE: The caller should initialize the cpu field to the cpu running the
189  * protocol stack associated with this inpcbinfo.
190  */
191 
192 void
193 in_pcbinfo_init(struct inpcbinfo *pcbinfo)
194 {
195 	LIST_INIT(&pcbinfo->pcblisthead);
196 	pcbinfo->cpu = -1;
197 	pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
198 				    M_WAITOK | M_ZERO);
199 }
200 
201 struct baddynamicports baddynamicports;
202 
203 /*
204  * Check if the specified port is invalid for dynamic allocation.
205  */
206 int
207 in_baddynamic(u_int16_t port, u_int16_t proto)
208 {
209 	switch (proto) {
210 	case IPPROTO_TCP:
211 		return (DP_ISSET(baddynamicports.tcp, port));
212 	case IPPROTO_UDP:
213 #ifdef IPSEC
214 		/* Cannot preset this as it is a sysctl */
215 		if (port == udpencap_port)
216 			return (1);
217 #endif
218 		return (DP_ISSET(baddynamicports.udp, port));
219 	default:
220 		return (0);
221 	}
222 }
223 
224 
225 /*
226  * Allocate a PCB and associate it with the socket.
227  */
228 int
229 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
230 {
231 	struct inpcb *inp;
232 #ifdef IPSEC
233 	int error;
234 #endif
235 
236 	inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO);
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 	KKASSERT(inp->inp_pcbinfo == pcbinfo);
269 
270 	LIST_REMOVE(inp, inp_list);
271 	pcbinfo->ipi_count--;
272 	inp->inp_pcbinfo = NULL;
273 }
274 
275 /*
276  * Relink a pcb into a new pcbinfo.
277  */
278 void
279 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
280 {
281 	KKASSERT(inp->inp_pcbinfo == NULL);
282 	inp->inp_pcbinfo = pcbinfo;
283 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
284 	pcbinfo->ipi_count++;
285 }
286 
287 int
288 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
289 {
290 	struct socket *so = inp->inp_socket;
291 	struct proc *p = td->td_proc;
292 	unsigned short *lastport;
293 	struct sockaddr_in *sin;
294 	struct sockaddr_in jsin;
295 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
296 	struct ucred *cred = NULL;
297 	u_short lport = 0;
298 	int wild = 0, reuseport = (so->so_options & SO_REUSEPORT);
299 	int error;
300 
301 	KKASSERT(p);
302 
303 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
304 		return (EADDRNOTAVAIL);
305 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
306 		return (EINVAL);	/* already bound */
307 
308 	if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
309 		wild = 1;    /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
310 	if (p)
311 		cred = p->p_ucred;
312 
313 	/*
314 	 * This has to be atomic.  If the porthash is shared across multiple
315 	 * protocol threads (aka tcp) then the token will be non-NULL.
316 	 */
317 	if (pcbinfo->porttoken)
318 		lwkt_gettoken(pcbinfo->porttoken);
319 
320 	if (nam != NULL) {
321 		sin = (struct sockaddr_in *)nam;
322 		if (nam->sa_len != sizeof *sin) {
323 			error = EINVAL;
324 			goto done;
325 		}
326 #ifdef notdef
327 		/*
328 		 * We should check the family, but old programs
329 		 * incorrectly fail to initialize it.
330 		 */
331 		if (sin->sin_family != AF_INET) {
332 			error = EAFNOSUPPORT;
333 			goto done;
334 		}
335 #endif
336 		if (!prison_replace_wildcards(td, nam)) {
337 			error = EINVAL;
338 			goto done;
339 		}
340 		lport = sin->sin_port;
341 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
342 			/*
343 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
344 			 * allow complete duplication of binding if
345 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
346 			 * and a multicast address is bound on both
347 			 * new and duplicated sockets.
348 			 */
349 			if (so->so_options & SO_REUSEADDR)
350 				reuseport = SO_REUSEADDR | SO_REUSEPORT;
351 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
352 			sin->sin_port = 0;		/* yech... */
353 			bzero(&sin->sin_zero, sizeof sin->sin_zero);
354 			if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) {
355 				error = EADDRNOTAVAIL;
356 				goto done;
357 			}
358 		}
359 		if (lport != 0) {
360 			struct inpcb *t;
361 
362 			/* GROSS */
363 			if (ntohs(lport) < IPPORT_RESERVED &&
364 			    cred &&
365 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0)) {
366 				error = EACCES;
367 				goto done;
368 			}
369 			if (so->so_cred->cr_uid != 0 &&
370 			    !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
371 				t = in_pcblookup_local(pcbinfo,
372 						       sin->sin_addr,
373 						       lport,
374 						       INPLOOKUP_WILDCARD,
375 						       cred);
376 				if (t &&
377 				    (!in_nullhost(sin->sin_addr) ||
378 				     !in_nullhost(t->inp_laddr) ||
379 				     (t->inp_socket->so_options &
380 					 SO_REUSEPORT) == 0) &&
381 				    (so->so_cred->cr_uid !=
382 				     t->inp_socket->so_cred->cr_uid)) {
383 #ifdef INET6
384 					if (!in_nullhost(sin->sin_addr) ||
385 					    !in_nullhost(t->inp_laddr) ||
386 					    INP_SOCKAF(so) ==
387 					    INP_SOCKAF(t->inp_socket))
388 #endif
389 					{
390 						error = EADDRINUSE;
391 						goto done;
392 					}
393 				}
394 			}
395 			if (cred && !prison_replace_wildcards(td, nam)) {
396 				error = EADDRNOTAVAIL;
397 				goto done;
398 			}
399 			t = in_pcblookup_local(pcbinfo, sin->sin_addr, lport,
400 					       wild, cred);
401 			if (t && !(reuseport & t->inp_socket->so_options)) {
402 #ifdef INET6
403 				if (!in_nullhost(sin->sin_addr) ||
404 				    !in_nullhost(t->inp_laddr) ||
405 				    INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
406 #endif
407 				{
408 					error = EADDRINUSE;
409 					goto done;
410 				}
411 			}
412 		}
413 		inp->inp_laddr = sin->sin_addr;
414 	}
415 	if (lport == 0) {
416 		ushort first, last;
417 		int count;
418 
419 		jsin.sin_family = AF_INET;
420 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
421 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
422 			inp->inp_laddr.s_addr = INADDR_ANY;
423 			error = EINVAL;
424 			goto done;
425 		}
426 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
427 
428 		inp->inp_flags |= INP_ANONPORT;
429 
430 		if (inp->inp_flags & INP_HIGHPORT) {
431 			first = ipport_hifirstauto;	/* sysctl */
432 			last  = ipport_hilastauto;
433 			lastport = &pcbinfo->lasthi;
434 		} else if (inp->inp_flags & INP_LOWPORT) {
435 			if (cred &&
436 			    (error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
437 				inp->inp_laddr.s_addr = INADDR_ANY;
438 				goto done;
439 			}
440 			first = ipport_lowfirstauto;	/* 1023 */
441 			last  = ipport_lowlastauto;	/* 600 */
442 			lastport = &pcbinfo->lastlow;
443 		} else {
444 			first = ipport_firstauto;	/* sysctl */
445 			last  = ipport_lastauto;
446 			lastport = &pcbinfo->lastport;
447 		}
448 		/*
449 		 * Simple check to ensure all ports are not used up causing
450 		 * a deadlock here.
451 		 *
452 		 * We split the two cases (up and down) so that the direction
453 		 * is not being tested on each round of the loop.
454 		 */
455 		if (first > last) {
456 			/*
457 			 * counting down
458 			 */
459 			count = first - last;
460 
461 			do {
462 				if (count-- < 0) {	/* completely used? */
463 					inp->inp_laddr.s_addr = INADDR_ANY;
464 					error = EADDRNOTAVAIL;
465 					goto done;
466 				}
467 				--*lastport;
468 				if (*lastport > first || *lastport < last)
469 					*lastport = first;
470 				lport = htons(*lastport);
471 			} while (in_pcblookup_local(pcbinfo, inp->inp_laddr,
472 						    lport, wild, cred));
473 		} else {
474 			/*
475 			 * counting up
476 			 */
477 			count = last - first;
478 
479 			do {
480 				if (count-- < 0) {	/* completely used? */
481 					inp->inp_laddr.s_addr = INADDR_ANY;
482 					error = EADDRNOTAVAIL;
483 					goto done;
484 				}
485 				++*lastport;
486 				if (*lastport < first || *lastport > last)
487 					*lastport = first;
488 				lport = htons(*lastport);
489 			} while (in_pcblookup_local(pcbinfo, inp->inp_laddr,
490 						    lport, wild, cred));
491 		}
492 	}
493 	inp->inp_lport = lport;
494 
495 	jsin.sin_family = AF_INET;
496 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
497 	if (!prison_replace_wildcards(td, (struct sockaddr*)&jsin)) {
498 		inp->inp_laddr.s_addr = INADDR_ANY;
499 		inp->inp_lport = 0;
500 		error = EINVAL;
501 		goto done;
502 	}
503 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
504 
505 	if (in_pcbinsporthash(inp) != 0) {
506 		inp->inp_laddr.s_addr = INADDR_ANY;
507 		inp->inp_lport = 0;
508 		error = EAGAIN;
509 		goto done;
510 	}
511 	error = 0;
512 done:
513 	if (pcbinfo->porttoken)
514 		lwkt_reltoken(pcbinfo->porttoken);
515 	return error;
516 }
517 
518 static struct inpcb *
519 in_pcblookup_addrport(struct inpcbinfo *pcbinfo, struct in_addr laddr,
520     u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
521 {
522 	struct inpcb *inp;
523 	struct inpcbporthead *porthash;
524 	struct inpcbport *phd;
525 	struct inpcb *match = NULL;
526 
527 	/*
528 	 * If the porthashbase is shared across several cpus we need
529 	 * to lock.
530 	 */
531 	if (pcbinfo->porttoken)
532 		lwkt_gettoken(pcbinfo->porttoken);
533 
534 	/*
535 	 * Best fit PCB lookup.
536 	 *
537 	 * First see if this local port is in use by looking on the
538 	 * port hash list.
539 	 */
540 	porthash = &pcbinfo->porthashbase[
541 			INP_PCBPORTHASH(lport, pcbinfo->porthashmask)];
542 	LIST_FOREACH(phd, porthash, phd_hash) {
543 		if (phd->phd_port == lport)
544 			break;
545 	}
546 	if (phd != NULL) {
547 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
548 #ifdef INET6
549 			if ((inp->inp_vflag & INP_IPV4) == 0)
550 				continue;
551 #endif
552 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
553 			    inp->inp_laddr.s_addr != laddr.s_addr)
554 				continue;
555 
556 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
557 			    inp->inp_faddr.s_addr != faddr.s_addr)
558 				continue;
559 
560 			if (inp->inp_fport != 0 && inp->inp_fport != fport)
561 				continue;
562 
563 			if (cred == NULL ||
564 			    cred->cr_prison ==
565 			    inp->inp_socket->so_cred->cr_prison) {
566 				match = inp;
567 				break;
568 			}
569 		}
570 	}
571 	if (pcbinfo->porttoken)
572 		lwkt_reltoken(pcbinfo->porttoken);
573 	return (match);
574 }
575 
576 int
577 in_pcbconn_bind(struct inpcb *inp, const struct sockaddr *nam,
578     struct thread *td)
579 {
580 	struct proc *p = td->td_proc;
581 	unsigned short *lastport;
582 	const struct sockaddr_in *sin = (const struct sockaddr_in *)nam;
583 	struct sockaddr_in jsin;
584 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
585 	struct ucred *cred = NULL;
586 	u_short lport = 0;
587 	ushort first, last;
588 	int count, error, dup = 0;
589 
590 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
591 		return (EADDRNOTAVAIL);
592 
593 	KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
594 	if (inp->inp_lport != 0)
595 		return (EINVAL);	/* already bound */
596 
597 	KKASSERT(p);
598 	cred = p->p_ucred;
599 
600 	/*
601 	 * This has to be atomic.  If the porthash is shared across multiple
602 	 * protocol threads (aka tcp) then the token will be non-NULL.
603 	 */
604 	if (pcbinfo->porttoken)
605 		lwkt_gettoken(pcbinfo->porttoken);
606 
607 	jsin.sin_family = AF_INET;
608 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
609 	if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
610 		inp->inp_laddr.s_addr = INADDR_ANY;
611 		error = EINVAL;
612 		goto done;
613 	}
614 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
615 
616 	inp->inp_flags |= INP_ANONPORT;
617 
618 	if (inp->inp_flags & INP_HIGHPORT) {
619 		first = ipport_hifirstauto;	/* sysctl */
620 		last  = ipport_hilastauto;
621 		lastport = &pcbinfo->lasthi;
622 	} else if (inp->inp_flags & INP_LOWPORT) {
623 		if (cred &&
624 		    (error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
625 			inp->inp_laddr.s_addr = INADDR_ANY;
626 			goto done;
627 		}
628 		first = ipport_lowfirstauto;	/* 1023 */
629 		last  = ipport_lowlastauto;	/* 600 */
630 		lastport = &pcbinfo->lastlow;
631 	} else {
632 		first = ipport_firstauto;	/* sysctl */
633 		last  = ipport_lastauto;
634 		lastport = &pcbinfo->lastport;
635 	}
636 
637 again:
638 	/*
639 	 * Simple check to ensure all ports are not used up causing
640 	 * a deadlock here.
641 	 *
642 	 * We split the two cases (up and down) so that the direction
643 	 * is not being tested on each round of the loop.
644 	 */
645 	if (first > last) {
646 		/*
647 		 * counting down
648 		 */
649 		count = first - last;
650 
651 		do {
652 			if (count-- < 0) {	/* completely used? */
653 				inp->inp_laddr.s_addr = INADDR_ANY;
654 				error = EADDRNOTAVAIL;
655 				goto done;
656 			}
657 			--*lastport;
658 			if (*lastport > first || *lastport < last)
659 				*lastport = first;
660 			lport = htons(*lastport);
661 		} while (in_pcblookup_addrport(pcbinfo, inp->inp_laddr, lport,
662 				sin->sin_addr, sin->sin_port, cred));
663 	} else {
664 		/*
665 		 * counting up
666 		 */
667 		count = last - first;
668 
669 		do {
670 			if (count-- < 0) {	/* completely used? */
671 				inp->inp_laddr.s_addr = INADDR_ANY;
672 				error = EADDRNOTAVAIL;
673 				goto done;
674 			}
675 			++*lastport;
676 			if (*lastport < first || *lastport > last)
677 				*lastport = first;
678 			lport = htons(*lastport);
679 		} while (in_pcblookup_addrport(pcbinfo, inp->inp_laddr, lport,
680 				sin->sin_addr, sin->sin_port, cred));
681 	}
682 
683 	/* This could happen on loopback interface */
684 	if (sin->sin_port == lport &&
685 	    sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
686 		if (dup) {
687 			/*
688 			 * Duplicate again; give up
689 			 */
690 			inp->inp_laddr.s_addr = INADDR_ANY;
691 			error = EADDRNOTAVAIL;
692 			goto done;
693 		}
694 		dup = 1;
695 		goto again;
696 	}
697 	inp->inp_lport = lport;
698 
699 	jsin.sin_family = AF_INET;
700 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
701 	if (!prison_replace_wildcards(td, (struct sockaddr*)&jsin)) {
702 		inp->inp_laddr.s_addr = INADDR_ANY;
703 		inp->inp_lport = 0;
704 		error = EINVAL;
705 		goto done;
706 	}
707 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
708 
709 	if (in_pcbinsporthash(inp) != 0) {
710 		inp->inp_laddr.s_addr = INADDR_ANY;
711 		inp->inp_lport = 0;
712 		error = EAGAIN;
713 		goto done;
714 	}
715 	error = 0;
716 done:
717 	if (pcbinfo->porttoken)
718 		lwkt_reltoken(pcbinfo->porttoken);
719 	return error;
720 }
721 
722 /*
723  *   Transform old in_pcbconnect() into an inner subroutine for new
724  *   in_pcbconnect(): Do some validity-checking on the remote
725  *   address (in mbuf 'nam') and then determine local host address
726  *   (i.e., which interface) to use to access that remote host.
727  *
728  *   This preserves definition of in_pcbconnect(), while supporting a
729  *   slightly different version for T/TCP.  (This is more than
730  *   a bit of a kludge, but cleaning up the internal interfaces would
731  *   have forced minor changes in every protocol).
732  */
733 int
734 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
735 	struct sockaddr_in **plocal_sin, struct thread *td)
736 {
737 	struct in_ifaddr *ia;
738 	struct ucred *cred = NULL;
739 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
740 	struct sockaddr *jsin;
741 	int jailed = 0, alloc_route = 0;
742 
743 	if (nam->sa_len != sizeof *sin)
744 		return (EINVAL);
745 	if (sin->sin_family != AF_INET)
746 		return (EAFNOSUPPORT);
747 	if (sin->sin_port == 0)
748 		return (EADDRNOTAVAIL);
749 	if (td && td->td_proc && td->td_proc->p_ucred)
750 		cred = td->td_proc->p_ucred;
751 	if (cred && cred->cr_prison)
752 		jailed = 1;
753 	if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
754 		ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
755 		/*
756 		 * If the destination address is INADDR_ANY,
757 		 * use the primary local address.
758 		 * If the supplied address is INADDR_BROADCAST,
759 		 * and the primary interface supports broadcast,
760 		 * choose the broadcast address for that interface.
761 		 */
762 		if (sin->sin_addr.s_addr == INADDR_ANY)
763 			sin->sin_addr = IA_SIN(ia)->sin_addr;
764 		else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
765 		    (ia->ia_ifp->if_flags & IFF_BROADCAST))
766 			sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
767 	}
768 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
769 		struct route *ro;
770 
771 		ia = NULL;
772 		/*
773 		 * If route is known or can be allocated now,
774 		 * our src addr is taken from the i/f, else punt.
775 		 * Note that we should check the address family of the cached
776 		 * destination, in case of sharing the cache with IPv6.
777 		 */
778 		ro = &inp->inp_route;
779 		if (ro->ro_rt &&
780 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
781 		     ro->ro_dst.sa_family != AF_INET ||
782 		     satosin(&ro->ro_dst)->sin_addr.s_addr !=
783 				      sin->sin_addr.s_addr ||
784 		     inp->inp_socket->so_options & SO_DONTROUTE)) {
785 			RTFREE(ro->ro_rt);
786 			ro->ro_rt = NULL;
787 		}
788 		if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
789 		    (ro->ro_rt == NULL ||
790 		    ro->ro_rt->rt_ifp == NULL)) {
791 			/* No route yet, so try to acquire one */
792 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
793 			ro->ro_dst.sa_family = AF_INET;
794 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
795 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
796 				sin->sin_addr;
797 			rtalloc(ro);
798 			alloc_route = 1;
799 		}
800 		/*
801 		 * If we found a route, use the address
802 		 * corresponding to the outgoing interface
803 		 * unless it is the loopback (in case a route
804 		 * to our address on another net goes to loopback).
805 		 */
806 		if (ro->ro_rt && !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
807 			if (jailed) {
808 				if (jailed_ip(cred->cr_prison,
809 				    ro->ro_rt->rt_ifa->ifa_addr)) {
810 					ia = ifatoia(ro->ro_rt->rt_ifa);
811 				}
812 			} else {
813 				ia = ifatoia(ro->ro_rt->rt_ifa);
814 			}
815 		}
816 		if (ia == NULL) {
817 			u_short fport = sin->sin_port;
818 
819 			sin->sin_port = 0;
820 			ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
821 			if (ia && jailed && !jailed_ip(cred->cr_prison,
822 			    sintosa(&ia->ia_addr)))
823 				ia = NULL;
824 			if (ia == NULL)
825 				ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
826 			if (ia && jailed && !jailed_ip(cred->cr_prison,
827 			    sintosa(&ia->ia_addr)))
828 				ia = NULL;
829 			sin->sin_port = fport;
830 			if (ia == NULL &&
831 			    !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
832 				ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
833 			if (ia && jailed && !jailed_ip(cred->cr_prison,
834 			    sintosa(&ia->ia_addr)))
835 				ia = NULL;
836 
837 			if (!jailed && ia == NULL)
838 				goto fail;
839 		}
840 		/*
841 		 * If the destination address is multicast and an outgoing
842 		 * interface has been set as a multicast option, use the
843 		 * address of that interface as our source address.
844 		 */
845 		if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
846 		    inp->inp_moptions != NULL) {
847 			struct ip_moptions *imo;
848 			struct ifnet *ifp;
849 
850 			imo = inp->inp_moptions;
851 			if (imo->imo_multicast_ifp != NULL) {
852 				struct in_ifaddr_container *iac;
853 
854 				ifp = imo->imo_multicast_ifp;
855 				ia = NULL;
856 				TAILQ_FOREACH(iac,
857 				&in_ifaddrheads[mycpuid], ia_link) {
858 					if (iac->ia->ia_ifp == ifp) {
859 						ia = iac->ia;
860 						break;
861 					}
862 				}
863 				if (ia == NULL)
864 					goto fail;
865 			}
866 		}
867 		/*
868 		 * Don't do pcblookup call here; return interface in plocal_sin
869 		 * and exit to caller, that will do the lookup.
870 		 */
871 		if (ia == NULL && jailed) {
872 			if ((jsin = prison_get_nonlocal(cred->cr_prison, AF_INET, NULL)) != NULL ||
873 			    (jsin = prison_get_local(cred->cr_prison, AF_INET, NULL)) != NULL) {
874 				*plocal_sin = satosin(jsin);
875 			} else {
876 				/* IPv6 only Jail */
877 				goto fail;
878 			}
879 		} else {
880 			*plocal_sin = &ia->ia_addr;
881 		}
882 	}
883 	return (0);
884 fail:
885 	if (alloc_route) {
886 		struct route *ro = &inp->inp_route;
887 
888 		if (ro->ro_rt != NULL)
889 			RTFREE(ro->ro_rt);
890 		bzero(ro, sizeof(*ro));
891 	}
892 	return (EADDRNOTAVAIL);
893 }
894 
895 /*
896  * Outer subroutine:
897  * Connect from a socket to a specified address.
898  * Both address and port must be specified in argument sin.
899  * If don't have a local address for this socket yet,
900  * then pick one.
901  */
902 int
903 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
904 {
905 	struct sockaddr_in *if_sin;
906 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
907 	int error;
908 
909 	/* Call inner routine to assign local interface address. */
910 	if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
911 		return (error);
912 
913 	if (in_pcblookup_hash(inp->inp_cpcbinfo, sin->sin_addr, sin->sin_port,
914 			      inp->inp_laddr.s_addr ?
915 				inp->inp_laddr : if_sin->sin_addr,
916 			      inp->inp_lport, FALSE, NULL) != NULL) {
917 		return (EADDRINUSE);
918 	}
919 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
920 		if (inp->inp_lport == 0) {
921 			error = in_pcbbind(inp, NULL, td);
922 			if (error)
923 				return (error);
924 		}
925 		inp->inp_laddr = if_sin->sin_addr;
926 	}
927 	inp->inp_faddr = sin->sin_addr;
928 	inp->inp_fport = sin->sin_port;
929 	in_pcbinsconnhash(inp);
930 	return (0);
931 }
932 
933 void
934 in_pcbdisconnect(struct inpcb *inp)
935 {
936 
937 	inp->inp_faddr.s_addr = INADDR_ANY;
938 	inp->inp_fport = 0;
939 	in_pcbremconnhash(inp);
940 	if (inp->inp_socket->so_state & SS_NOFDREF)
941 		in_pcbdetach(inp);
942 }
943 
944 void
945 in_pcbdetach(struct inpcb *inp)
946 {
947 	struct socket *so = inp->inp_socket;
948 	struct inpcbinfo *ipi = inp->inp_pcbinfo;
949 
950 #ifdef IPSEC
951 	ipsec4_delete_pcbpolicy(inp);
952 #endif /*IPSEC*/
953 	inp->inp_gencnt = ++ipi->ipi_gencnt;
954 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
955 	in_pcbremlists(inp);
956 	so->so_pcb = NULL;
957 	sofree(so);			/* remove pcb ref */
958 	if (inp->inp_options)
959 		m_free(inp->inp_options);
960 	if (inp->inp_route.ro_rt)
961 		rtfree(inp->inp_route.ro_rt);
962 	ip_freemoptions(inp->inp_moptions);
963 	inp->inp_vflag = 0;
964 	kfree(inp, M_PCB);
965 }
966 
967 /*
968  * The calling convention of in_setsockaddr() and in_setpeeraddr() was
969  * modified to match the pru_sockaddr() and pru_peeraddr() entry points
970  * in struct pr_usrreqs, so that protocols can just reference then directly
971  * without the need for a wrapper function.  The socket must have a valid
972  * (i.e., non-nil) PCB, but it should be impossible to get an invalid one
973  * except through a kernel programming error, so it is acceptable to panic
974  * (or in this case trap) if the PCB is invalid.  (Actually, we don't trap
975  * because there actually /is/ a programming error somewhere... XXX)
976  */
977 int
978 in_setsockaddr(struct socket *so, struct sockaddr **nam)
979 {
980 	struct inpcb *inp;
981 	struct sockaddr_in *sin;
982 
983 	/*
984 	 * Do the malloc first in case it blocks.
985 	 */
986 	MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
987 		M_WAITOK | M_ZERO);
988 	sin->sin_family = AF_INET;
989 	sin->sin_len = sizeof *sin;
990 
991 	crit_enter();
992 	inp = so->so_pcb;
993 	if (!inp) {
994 		crit_exit();
995 		kfree(sin, M_SONAME);
996 		return (ECONNRESET);
997 	}
998 	sin->sin_port = inp->inp_lport;
999 	sin->sin_addr = inp->inp_laddr;
1000 	crit_exit();
1001 
1002 	*nam = (struct sockaddr *)sin;
1003 	return (0);
1004 }
1005 
1006 void
1007 in_setsockaddr_dispatch(netmsg_t msg)
1008 {
1009 	int error;
1010 
1011 	error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1012 	lwkt_replymsg(&msg->lmsg, error);
1013 }
1014 
1015 int
1016 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1017 {
1018 	struct inpcb *inp;
1019 	struct sockaddr_in *sin;
1020 
1021 	/*
1022 	 * Do the malloc first in case it blocks.
1023 	 */
1024 	MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
1025 		M_WAITOK | M_ZERO);
1026 	sin->sin_family = AF_INET;
1027 	sin->sin_len = sizeof *sin;
1028 
1029 	crit_enter();
1030 	inp = so->so_pcb;
1031 	if (!inp) {
1032 		crit_exit();
1033 		kfree(sin, M_SONAME);
1034 		return (ECONNRESET);
1035 	}
1036 	sin->sin_port = inp->inp_fport;
1037 	sin->sin_addr = inp->inp_faddr;
1038 	crit_exit();
1039 
1040 	*nam = (struct sockaddr *)sin;
1041 	return (0);
1042 }
1043 
1044 void
1045 in_setpeeraddr_dispatch(netmsg_t msg)
1046 {
1047 	int error;
1048 
1049 	error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1050 	lwkt_replymsg(&msg->lmsg, error);
1051 }
1052 
1053 void
1054 in_pcbnotifyall(struct inpcbhead *head, struct in_addr faddr, int err,
1055 		void (*notify)(struct inpcb *, int))
1056 {
1057 	struct inpcb *inp, *ninp;
1058 
1059 	/*
1060 	 * note: if INP_PLACEMARKER is set we must ignore the rest of
1061 	 * the structure and skip it.
1062 	 */
1063 	crit_enter();
1064 	LIST_FOREACH_MUTABLE(inp, head, inp_list, ninp) {
1065 		if (inp->inp_flags & INP_PLACEMARKER)
1066 			continue;
1067 #ifdef INET6
1068 		if (!(inp->inp_vflag & INP_IPV4))
1069 			continue;
1070 #endif
1071 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1072 		    inp->inp_socket == NULL)
1073 			continue;
1074 		(*notify)(inp, err);		/* can remove inp from list! */
1075 	}
1076 	crit_exit();
1077 }
1078 
1079 void
1080 in_pcbpurgeif0(struct inpcb *head, struct ifnet *ifp)
1081 {
1082 	struct inpcb *inp;
1083 	struct ip_moptions *imo;
1084 	int i, gap;
1085 
1086 	for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) {
1087 		if (inp->inp_flags & INP_PLACEMARKER)
1088 			continue;
1089 		imo = inp->inp_moptions;
1090 		if ((inp->inp_vflag & INP_IPV4) && imo != NULL) {
1091 			/*
1092 			 * Unselect the outgoing interface if it is being
1093 			 * detached.
1094 			 */
1095 			if (imo->imo_multicast_ifp == ifp)
1096 				imo->imo_multicast_ifp = NULL;
1097 
1098 			/*
1099 			 * Drop multicast group membership if we joined
1100 			 * through the interface being detached.
1101 			 */
1102 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1103 			    i++) {
1104 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1105 					in_delmulti(imo->imo_membership[i]);
1106 					gap++;
1107 				} else if (gap != 0)
1108 					imo->imo_membership[i - gap] =
1109 					    imo->imo_membership[i];
1110 			}
1111 			imo->imo_num_memberships -= gap;
1112 		}
1113 	}
1114 }
1115 
1116 /*
1117  * Check for alternatives when higher level complains
1118  * about service problems.  For now, invalidate cached
1119  * routing information.  If the route was created dynamically
1120  * (by a redirect), time to try a default gateway again.
1121  */
1122 void
1123 in_losing(struct inpcb *inp)
1124 {
1125 	struct rtentry *rt;
1126 	struct rt_addrinfo rtinfo;
1127 
1128 	if ((rt = inp->inp_route.ro_rt)) {
1129 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
1130 		rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1131 		rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1132 		rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1133 		rtinfo.rti_flags = rt->rt_flags;
1134 		rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1135 		if (rt->rt_flags & RTF_DYNAMIC)
1136 			rtrequest1_global(RTM_DELETE, &rtinfo, NULL, NULL);
1137 		inp->inp_route.ro_rt = NULL;
1138 		rtfree(rt);
1139 		/*
1140 		 * A new route can be allocated
1141 		 * the next time output is attempted.
1142 		 */
1143 	}
1144 }
1145 
1146 /*
1147  * After a routing change, flush old routing
1148  * and allocate a (hopefully) better one.
1149  */
1150 void
1151 in_rtchange(struct inpcb *inp, int err)
1152 {
1153 	if (inp->inp_route.ro_rt) {
1154 		rtfree(inp->inp_route.ro_rt);
1155 		inp->inp_route.ro_rt = NULL;
1156 		/*
1157 		 * A new route can be allocated the next time
1158 		 * output is attempted.
1159 		 */
1160 	}
1161 }
1162 
1163 /*
1164  * Lookup a PCB based on the local address and port.
1165  */
1166 struct inpcb *
1167 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1168 		   u_int lport_arg, int wild_okay, struct ucred *cred)
1169 {
1170 	struct inpcb *inp;
1171 	int matchwild = 3, wildcard;
1172 	u_short lport = lport_arg;
1173 	struct inpcbporthead *porthash;
1174 	struct inpcbport *phd;
1175 	struct inpcb *match = NULL;
1176 
1177 	/*
1178 	 * If the porthashbase is shared across several cpus we need
1179 	 * to lock.
1180 	 */
1181 	if (pcbinfo->porttoken)
1182 		lwkt_gettoken(pcbinfo->porttoken);
1183 
1184 	/*
1185 	 * Best fit PCB lookup.
1186 	 *
1187 	 * First see if this local port is in use by looking on the
1188 	 * port hash list.
1189 	 */
1190 	porthash = &pcbinfo->porthashbase[
1191 			INP_PCBPORTHASH(lport, pcbinfo->porthashmask)];
1192 	LIST_FOREACH(phd, porthash, phd_hash) {
1193 		if (phd->phd_port == lport)
1194 			break;
1195 	}
1196 	if (phd != NULL) {
1197 		/*
1198 		 * Port is in use by one or more PCBs. Look for best
1199 		 * fit.
1200 		 */
1201 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1202 			wildcard = 0;
1203 #ifdef INET6
1204 			if ((inp->inp_vflag & INP_IPV4) == 0)
1205 				continue;
1206 #endif
1207 			if (inp->inp_faddr.s_addr != INADDR_ANY)
1208 				wildcard++;
1209 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
1210 				if (laddr.s_addr == INADDR_ANY)
1211 					wildcard++;
1212 				else if (inp->inp_laddr.s_addr != laddr.s_addr)
1213 					continue;
1214 			} else {
1215 				if (laddr.s_addr != INADDR_ANY)
1216 					wildcard++;
1217 			}
1218 			if (wildcard && !wild_okay)
1219 				continue;
1220 			if (wildcard < matchwild &&
1221 			    (cred == NULL ||
1222 			     cred->cr_prison ==
1223 					inp->inp_socket->so_cred->cr_prison)) {
1224 				match = inp;
1225 				matchwild = wildcard;
1226 				if (matchwild == 0) {
1227 					break;
1228 				}
1229 			}
1230 		}
1231 	}
1232 	if (pcbinfo->porttoken)
1233 		lwkt_reltoken(pcbinfo->porttoken);
1234 	return (match);
1235 }
1236 
1237 /*
1238  * Lookup PCB in hash list.
1239  */
1240 struct inpcb *
1241 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1242 		  u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1243 		  boolean_t wildcard, struct ifnet *ifp)
1244 {
1245 	struct inpcbhead *head;
1246 	struct inpcb *inp, *jinp=NULL;
1247 	u_short fport = fport_arg, lport = lport_arg;
1248 
1249 	/*
1250 	 * First look for an exact match.
1251 	 */
1252 	head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1253 	    laddr.s_addr, lport, pcbinfo->hashmask)];
1254 	LIST_FOREACH(inp, head, inp_hash) {
1255 #ifdef INET6
1256 		if (!(inp->inp_vflag & INP_IPV4))
1257 			continue;
1258 #endif
1259 		if (in_hosteq(inp->inp_faddr, faddr) &&
1260 		    in_hosteq(inp->inp_laddr, laddr) &&
1261 		    inp->inp_fport == fport && inp->inp_lport == lport) {
1262 			/* found */
1263 			if (inp->inp_socket == NULL ||
1264 			    inp->inp_socket->so_cred->cr_prison == NULL) {
1265 				return (inp);
1266 			} else {
1267 				if  (jinp == NULL)
1268 					jinp = inp;
1269 			}
1270 		}
1271 	}
1272 	if (jinp != NULL)
1273 		return (jinp);
1274 	if (wildcard) {
1275 		struct inpcb *local_wild = NULL;
1276 		struct inpcb *jinp_wild = NULL;
1277 #ifdef INET6
1278 		struct inpcb *local_wild_mapped = NULL;
1279 #endif
1280 		struct inpcontainer *ic;
1281 		struct inpcontainerhead *chead;
1282 		struct sockaddr_in jsin;
1283 		struct ucred *cred;
1284 
1285 		/*
1286 		 * Order of socket selection:
1287 		 * 1. non-jailed, non-wild.
1288 		 * 2. non-jailed, wild.
1289 		 * 3. jailed, non-wild.
1290 		 * 4. jailed, wild.
1291 		 */
1292 		jsin.sin_family = AF_INET;
1293 		chead = &pcbinfo->wildcardhashbase[
1294 		    INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1295 		LIST_FOREACH(ic, chead, ic_list) {
1296 			inp = ic->ic_inp;
1297 			jsin.sin_addr.s_addr = laddr.s_addr;
1298 #ifdef INET6
1299 			if (!(inp->inp_vflag & INP_IPV4))
1300 				continue;
1301 #endif
1302 			if (inp->inp_socket != NULL)
1303 				cred = inp->inp_socket->so_cred;
1304 			else
1305 				cred = NULL;
1306 			if (cred != NULL && jailed(cred)) {
1307 				if (jinp != NULL)
1308 					continue;
1309 				else
1310 					if (!jailed_ip(cred->cr_prison,
1311 					    (struct sockaddr *)&jsin))
1312 						continue;
1313 			}
1314 			if (inp->inp_lport == lport) {
1315 				if (ifp && ifp->if_type == IFT_FAITH &&
1316 				    !(inp->inp_flags & INP_FAITH))
1317 					continue;
1318 				if (inp->inp_laddr.s_addr == laddr.s_addr) {
1319 					if (cred != NULL && jailed(cred))
1320 						jinp = inp;
1321 					else
1322 						return (inp);
1323 				}
1324 				if (inp->inp_laddr.s_addr == INADDR_ANY) {
1325 #ifdef INET6
1326 					if (INP_CHECK_SOCKAF(inp->inp_socket,
1327 							     AF_INET6))
1328 						local_wild_mapped = inp;
1329 					else
1330 #endif
1331 						if (cred != NULL &&
1332 						    jailed(cred))
1333 							jinp_wild = inp;
1334 						else
1335 							local_wild = inp;
1336 				}
1337 			}
1338 		}
1339 		if (local_wild != NULL)
1340 			return (local_wild);
1341 #ifdef INET6
1342 		if (local_wild_mapped != NULL)
1343 			return (local_wild_mapped);
1344 #endif
1345 		if (jinp != NULL)
1346 			return (jinp);
1347 		return (jinp_wild);
1348 	}
1349 
1350 	/*
1351 	 * Not found.
1352 	 */
1353 	return (NULL);
1354 }
1355 
1356 /*
1357  * Insert PCB into connection hash table.
1358  */
1359 void
1360 in_pcbinsconnhash(struct inpcb *inp)
1361 {
1362 	struct inpcbinfo *pcbinfo = inp->inp_cpcbinfo;
1363 	struct inpcbhead *bucket;
1364 	u_int32_t hashkey_faddr, hashkey_laddr;
1365 
1366 #ifdef INET6
1367 	if (inp->inp_vflag & INP_IPV6) {
1368 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1369 		hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1370 	} else {
1371 #endif
1372 		hashkey_faddr = inp->inp_faddr.s_addr;
1373 		hashkey_laddr = inp->inp_laddr.s_addr;
1374 #ifdef INET6
1375 	}
1376 #endif
1377 
1378 	KASSERT(!(inp->inp_flags & INP_WILDCARD),
1379 		("already on wildcardhash\n"));
1380 	KASSERT(!(inp->inp_flags & INP_CONNECTED),
1381 		("already on connhash\n"));
1382 	inp->inp_flags |= INP_CONNECTED;
1383 
1384 	/*
1385 	 * Insert into the connection hash table.
1386 	 */
1387 	bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1388 	    inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1389 	LIST_INSERT_HEAD(bucket, inp, inp_hash);
1390 }
1391 
1392 /*
1393  * Remove PCB from connection hash table.
1394  */
1395 void
1396 in_pcbremconnhash(struct inpcb *inp)
1397 {
1398 	KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1399 	LIST_REMOVE(inp, inp_hash);
1400 	inp->inp_flags &= ~INP_CONNECTED;
1401 }
1402 
1403 /*
1404  * Insert PCB into port hash table.
1405  */
1406 int
1407 in_pcbinsporthash(struct inpcb *inp)
1408 {
1409 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1410 	struct inpcbporthead *pcbporthash;
1411 	struct inpcbport *phd;
1412 
1413 	/*
1414 	 * If the porthashbase is shared across several cpus we need
1415 	 * to lock.
1416 	 */
1417 	if (pcbinfo->porttoken)
1418 		lwkt_gettoken(pcbinfo->porttoken);
1419 
1420 	/*
1421 	 * Insert into the port hash table.
1422 	 */
1423 	pcbporthash = &pcbinfo->porthashbase[
1424 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->porthashmask)];
1425 
1426 	/* Go through port list and look for a head for this lport. */
1427 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
1428 		if (phd->phd_port == inp->inp_lport)
1429 			break;
1430 	}
1431 
1432 	/* If none exists, malloc one and tack it on. */
1433 	if (phd == NULL) {
1434 		KKASSERT(pcbinfo->portsave != NULL);
1435 		phd = pcbinfo->portsave;
1436 		pcbinfo->portsave = NULL;
1437 		phd->phd_port = inp->inp_lport;
1438 		LIST_INIT(&phd->phd_pcblist);
1439 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1440 	}
1441 
1442 	inp->inp_phd = phd;
1443 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1444 
1445 	if (pcbinfo->porttoken)
1446 		lwkt_reltoken(pcbinfo->porttoken);
1447 	if (pcbinfo->portsave == NULL) {
1448 		pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1449 					    M_PCB, M_INTWAIT | M_ZERO);
1450 	}
1451 	return (0);
1452 }
1453 
1454 void
1455 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1456 {
1457 	struct inpcontainer *ic;
1458 	struct inpcontainerhead *bucket;
1459 
1460 	bucket = &pcbinfo->wildcardhashbase[
1461 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1462 
1463 	ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
1464 	ic->ic_inp = inp;
1465 	LIST_INSERT_HEAD(bucket, ic, ic_list);
1466 }
1467 
1468 /*
1469  * Insert PCB into wildcard hash table.
1470  */
1471 void
1472 in_pcbinswildcardhash(struct inpcb *inp)
1473 {
1474 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1475 
1476 	KASSERT(!(inp->inp_flags & INP_CONNECTED),
1477 		("already on connhash\n"));
1478 	KASSERT(!(inp->inp_flags & INP_WILDCARD),
1479 		("already on wildcardhash\n"));
1480 	inp->inp_flags |= INP_WILDCARD;
1481 
1482 	in_pcbinswildcardhash_oncpu(inp, pcbinfo);
1483 }
1484 
1485 void
1486 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1487 {
1488 	struct inpcontainer *ic;
1489 	struct inpcontainerhead *head;
1490 
1491 	/* find bucket */
1492 	head = &pcbinfo->wildcardhashbase[
1493 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1494 
1495 	LIST_FOREACH(ic, head, ic_list) {
1496 		if (ic->ic_inp == inp)
1497 			goto found;
1498 	}
1499 	return;			/* not found! */
1500 
1501 found:
1502 	LIST_REMOVE(ic, ic_list);	/* remove container from bucket chain */
1503 	kfree(ic, M_TEMP);		/* deallocate container */
1504 }
1505 
1506 /*
1507  * Remove PCB from wildcard hash table.
1508  */
1509 void
1510 in_pcbremwildcardhash(struct inpcb *inp)
1511 {
1512 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1513 
1514 	KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
1515 	in_pcbremwildcardhash_oncpu(inp, pcbinfo);
1516 	inp->inp_flags &= ~INP_WILDCARD;
1517 }
1518 
1519 /*
1520  * Remove PCB from various lists.
1521  */
1522 void
1523 in_pcbremlists(struct inpcb *inp)
1524 {
1525 	struct inpcbinfo *pcbinfo;
1526 
1527 	if (inp->inp_lport) {
1528 		struct inpcbport *phd;
1529 
1530 		pcbinfo = inp->inp_pcbinfo;
1531 		if (pcbinfo->porttoken)
1532 			lwkt_gettoken(pcbinfo->porttoken);
1533 
1534 		phd = inp->inp_phd;
1535 		LIST_REMOVE(inp, inp_portlist);
1536 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1537 			LIST_REMOVE(phd, phd_hash);
1538 			kfree(phd, M_PCB);
1539 		}
1540 		if (pcbinfo->porttoken)
1541 			lwkt_reltoken(pcbinfo->porttoken);
1542 	}
1543 	if (inp->inp_flags & INP_WILDCARD) {
1544 		in_pcbremwildcardhash(inp);
1545 	} else if (inp->inp_flags & INP_CONNECTED) {
1546 		in_pcbremconnhash(inp);
1547 	}
1548 	LIST_REMOVE(inp, inp_list);
1549 	inp->inp_pcbinfo->ipi_count--;
1550 }
1551 
1552 int
1553 prison_xinpcb(struct thread *td, struct inpcb *inp)
1554 {
1555 	struct ucred *cr;
1556 
1557 	if (td->td_proc == NULL)
1558 		return (0);
1559 	cr = td->td_proc->p_ucred;
1560 	if (cr->cr_prison == NULL)
1561 		return (0);
1562 	if (inp->inp_socket && inp->inp_socket->so_cred &&
1563 	    inp->inp_socket->so_cred->cr_prison &&
1564 	    cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
1565 		return (0);
1566 	return (1);
1567 }
1568 
1569 int
1570 in_pcblist_global(SYSCTL_HANDLER_ARGS)
1571 {
1572 	struct inpcbinfo *pcbinfo = arg1;
1573 	struct inpcb *inp, *marker;
1574 	struct xinpcb xi;
1575 	int error, i, n;
1576 
1577 	/*
1578 	 * The process of preparing the TCB list is too time-consuming and
1579 	 * resource-intensive to repeat twice on every request.
1580 	 */
1581 	if (req->oldptr == NULL) {
1582 		n = pcbinfo->ipi_count;
1583 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
1584 		return 0;
1585 	}
1586 
1587 	if (req->newptr != NULL)
1588 		return EPERM;
1589 
1590 	/*
1591 	 * OK, now we're committed to doing something.  Re-fetch ipi_count
1592 	 * after obtaining the generation count.
1593 	 */
1594 	n = pcbinfo->ipi_count;
1595 
1596 	marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
1597 	marker->inp_flags |= INP_PLACEMARKER;
1598 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1599 
1600 	i = 0;
1601 	error = 0;
1602 
1603 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
1604 		LIST_REMOVE(marker, inp_list);
1605 		LIST_INSERT_AFTER(inp, marker, inp_list);
1606 
1607 		if (inp->inp_flags & INP_PLACEMARKER)
1608 			continue;
1609 		if (prison_xinpcb(req->td, inp))
1610 			continue;
1611 		bzero(&xi, sizeof xi);
1612 		xi.xi_len = sizeof xi;
1613 		bcopy(inp, &xi.xi_inp, sizeof *inp);
1614 		if (inp->inp_socket)
1615 			sotoxsocket(inp->inp_socket, &xi.xi_socket);
1616 		if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
1617 			break;
1618 		++i;
1619 	}
1620 	LIST_REMOVE(marker, inp_list);
1621 	if (error == 0 && i < n) {
1622 		bzero(&xi, sizeof xi);
1623 		xi.xi_len = sizeof xi;
1624 		while (i < n) {
1625 			error = SYSCTL_OUT(req, &xi, sizeof xi);
1626 			++i;
1627 		}
1628 	}
1629 	kfree(marker, M_TEMP);
1630 	return(error);
1631 }
1632 
1633 int
1634 in_pcblist_global_nomarker(SYSCTL_HANDLER_ARGS, struct xinpcb **xi0, int *nxi0)
1635 {
1636 	struct inpcbinfo *pcbinfo = arg1;
1637 	struct inpcb *inp;
1638 	struct xinpcb *xi;
1639 	int nxi;
1640 
1641 	*nxi0 = 0;
1642 	*xi0 = NULL;
1643 
1644 	/*
1645 	 * The process of preparing the PCB list is too time-consuming and
1646 	 * resource-intensive to repeat twice on every request.
1647 	 */
1648 	if (req->oldptr == NULL) {
1649 		int n = pcbinfo->ipi_count;
1650 
1651 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
1652 		return 0;
1653 	}
1654 
1655 	if (req->newptr != NULL)
1656 		return EPERM;
1657 
1658 	if (pcbinfo->ipi_count == 0)
1659 		return 0;
1660 
1661 	nxi = 0;
1662 	xi = kmalloc(pcbinfo->ipi_count * sizeof(*xi), M_TEMP,
1663 		     M_WAITOK | M_ZERO | M_NULLOK);
1664 	if (xi == NULL)
1665 		return ENOMEM;
1666 
1667 	LIST_FOREACH(inp, &pcbinfo->pcblisthead, inp_list) {
1668 		struct xinpcb *xi_ptr = &xi[nxi];
1669 
1670 		if (prison_xinpcb(req->td, inp))
1671 			continue;
1672 
1673 		xi_ptr->xi_len = sizeof(*xi_ptr);
1674 		bcopy(inp, &xi_ptr->xi_inp, sizeof(*inp));
1675 		if (inp->inp_socket)
1676 			sotoxsocket(inp->inp_socket, &xi_ptr->xi_socket);
1677 		++nxi;
1678 	}
1679 
1680 	if (nxi == 0) {
1681 		kfree(xi, M_TEMP);
1682 		return 0;
1683 	}
1684 
1685 	*nxi0 = nxi;
1686 	*xi0 = xi;
1687 
1688 	return 0;
1689 }
1690