xref: /dflybsd-src/sys/netinet/in_pcb.c (revision 2d1142198014c216eccccff4280c7b9ca04d1346)
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 #include <net/netisr2.h>
93 
94 #include <netinet/in.h>
95 #include <netinet/in_pcb.h>
96 #include <netinet/in_var.h>
97 #include <netinet/ip_var.h>
98 #ifdef INET6
99 #include <netinet/ip6.h>
100 #include <netinet6/ip6_var.h>
101 #endif /* INET6 */
102 
103 #ifdef IPSEC
104 #include <netinet6/ipsec.h>
105 #include <netproto/key/key.h>
106 #include <netproto/ipsec/esp_var.h>
107 #endif
108 
109 #ifdef FAST_IPSEC
110 #if defined(IPSEC) || defined(IPSEC_ESP)
111 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!"
112 #endif
113 
114 #include <netproto/ipsec/ipsec.h>
115 #include <netproto/ipsec/key.h>
116 #define	IPSEC
117 #endif /* FAST_IPSEC */
118 
119 #define INP_LOCALGROUP_SIZMIN	8
120 #define INP_LOCALGROUP_SIZMAX	256
121 
122 struct in_addr zeroin_addr;
123 
124 /*
125  * These configure the range of local port addresses assigned to
126  * "unspecified" outgoing connections/packets/whatever.
127  */
128 int ipport_lowfirstauto = IPPORT_RESERVED - 1;	/* 1023 */
129 int ipport_lowlastauto = IPPORT_RESERVEDSTART;	/* 600 */
130 
131 int ipport_firstauto = IPPORT_RESERVED;		/* 1024 */
132 int ipport_lastauto = IPPORT_USERRESERVED;	/* 5000 */
133 
134 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO;	/* 49152 */
135 int ipport_hilastauto = IPPORT_HILASTAUTO;	/* 65535 */
136 
137 #define RANGECHK(var, min, max) \
138 	if ((var) < (min)) { (var) = (min); } \
139 	else if ((var) > (max)) { (var) = (max); }
140 
141 int udpencap_enable = 1;	/* enabled by default */
142 int udpencap_port = 4500;	/* triggers decapsulation */
143 
144 /*
145  * Per-netisr inpcb markers.
146  * NOTE: they should only be used in netisrs.
147  */
148 static struct inpcb		*in_pcbmarkers;
149 static struct inpcontainer	*in_pcbcontainer_markers;
150 
151 static int
152 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
153 {
154 	int error;
155 
156 	error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
157 	if (!error) {
158 		RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
159 		RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
160 
161 		RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX);
162 		RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX);
163 
164 		RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX);
165 		RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX);
166 	}
167 	return (error);
168 }
169 
170 #undef RANGECHK
171 
172 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports");
173 
174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW,
175 	   &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW,
177 	   &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", "");
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW,
179 	   &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", "");
180 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW,
181 	   &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW,
183 	   &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW,
185 	   &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", "");
186 
187 /*
188  * in_pcb.c: manage the Protocol Control Blocks.
189  *
190  * NOTE: It is assumed that most of these functions will be called from
191  * a critical section.  XXX - There are, unfortunately, a few exceptions
192  * to this rule that should be fixed.
193  *
194  * NOTE: The caller should initialize the cpu field to the cpu running the
195  * protocol stack associated with this inpcbinfo.
196  */
197 
198 void
199 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared)
200 {
201 	KASSERT(cpu >= 0 && cpu < ncpus, ("invalid cpu%d", cpu));
202 	pcbinfo->cpu = cpu;
203 
204 	LIST_INIT(&pcbinfo->pcblisthead);
205 	pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB,
206 				    M_WAITOK | M_ZERO);
207 
208 	if (shared) {
209 		pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token),
210 		    M_PCB, M_WAITOK);
211 		lwkt_token_init(pcbinfo->infotoken, "infotoken");
212 	} else {
213 		pcbinfo->infotoken = NULL;
214 	}
215 }
216 
217 struct baddynamicports baddynamicports;
218 
219 /*
220  * Check if the specified port is invalid for dynamic allocation.
221  */
222 int
223 in_baddynamic(u_int16_t port, u_int16_t proto)
224 {
225 	switch (proto) {
226 	case IPPROTO_TCP:
227 		return (DP_ISSET(baddynamicports.tcp, port));
228 	case IPPROTO_UDP:
229 #ifdef IPSEC
230 		/* Cannot preset this as it is a sysctl */
231 		if (port == udpencap_port)
232 			return (1);
233 #endif
234 		return (DP_ISSET(baddynamicports.udp, port));
235 	default:
236 		return (0);
237 	}
238 }
239 
240 void
241 in_pcbonlist(struct inpcb *inp)
242 {
243 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
244 
245 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
246 	    ("not in the correct netisr"));
247 	KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist"));
248 	inp->inp_flags |= INP_ONLIST;
249 
250 	GET_PCBINFO_TOKEN(pcbinfo);
251 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list);
252 	pcbinfo->ipi_count++;
253 	REL_PCBINFO_TOKEN(pcbinfo);
254 }
255 
256 void
257 in_pcbofflist(struct inpcb *inp)
258 {
259 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
260 
261 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
262 	    ("not in the correct netisr"));
263 	KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist"));
264 	inp->inp_flags &= ~INP_ONLIST;
265 
266 	GET_PCBINFO_TOKEN(pcbinfo);
267 	LIST_REMOVE(inp, inp_list);
268 	KASSERT(pcbinfo->ipi_count > 0,
269 	    ("invalid inpcb count %d", pcbinfo->ipi_count));
270 	pcbinfo->ipi_count--;
271 	REL_PCBINFO_TOKEN(pcbinfo);
272 }
273 
274 /*
275  * Allocate a PCB and associate it with the socket.
276  */
277 int
278 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
279 {
280 	struct inpcb *inp;
281 #ifdef IPSEC
282 	int error;
283 #endif
284 
285 	inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK);
286 	if (inp == NULL)
287 		return (ENOMEM);
288 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
289 	inp->inp_pcbinfo = pcbinfo;
290 	inp->inp_socket = so;
291 #ifdef IPSEC
292 	error = ipsec_init_policy(so, &inp->inp_sp);
293 	if (error != 0) {
294 		kfree(inp, M_PCB);
295 		return (error);
296 	}
297 #endif
298 #ifdef INET6
299 	if (INP_SOCKAF(so) == AF_INET6 && ip6_v6only)
300 		inp->inp_flags |= IN6P_IPV6_V6ONLY;
301 	if (ip6_auto_flowlabel)
302 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
303 #endif
304 	soreference(so);
305 	so->so_pcb = inp;
306 
307 	in_pcbonlist(inp);
308 	return (0);
309 }
310 
311 /*
312  * Unlink a pcb with the intention of moving it to another cpu with a
313  * different pcbinfo.  While unlinked nothing should attempt to dereference
314  * inp_pcbinfo, NULL it out so we assert if it does.
315  */
316 void
317 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
318 {
319 	KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
320 	KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
321 	    ("already linked"));
322 
323 	in_pcbofflist(inp);
324 	inp->inp_pcbinfo = NULL;
325 }
326 
327 /*
328  * Relink a pcb into a new pcbinfo.
329  */
330 void
331 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
332 {
333 	KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
334 	KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0,
335 	    ("already linked"));
336 
337 	inp->inp_pcbinfo = pcbinfo;
338 	in_pcbonlist(inp);
339 }
340 
341 static int
342 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
343 {
344 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
345 	struct inpcbportinfo *portinfo;
346 	u_short first, last, lport, step;
347 	u_short *lastport;
348 	int count, error;
349 	int portinfo_first, portinfo_idx;
350 
351 	inp->inp_flags |= INP_ANONPORT;
352 
353 	step = pcbinfo->portinfo_mask + 1;
354 	portinfo_first = mycpuid & pcbinfo->portinfo_mask;
355 	portinfo_idx = portinfo_first;
356 loop:
357 	portinfo = &pcbinfo->portinfo[portinfo_idx];
358 
359 	if (inp->inp_flags & INP_HIGHPORT) {
360 		first = ipport_hifirstauto;	/* sysctl */
361 		last  = ipport_hilastauto;
362 		lastport = &portinfo->lasthi;
363 	} else if (inp->inp_flags & INP_LOWPORT) {
364 		if (cred &&
365 		    (error =
366 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
367 			inp->inp_laddr.s_addr = INADDR_ANY;
368 			return error;
369 		}
370 		first = ipport_lowfirstauto;	/* 1023 */
371 		last  = ipport_lowlastauto;	/* 600 */
372 		lastport = &portinfo->lastlow;
373 	} else {
374 		first = ipport_firstauto;	/* sysctl */
375 		last  = ipport_lastauto;
376 		lastport = &portinfo->lastport;
377 	}
378 
379 	/*
380 	 * This has to be atomic.  If the porthash is shared across multiple
381 	 * protocol threads (aka tcp) then the token must be held.
382 	 */
383 	GET_PORT_TOKEN(portinfo);
384 
385 	/*
386 	 * Simple check to ensure all ports are not used up causing
387 	 * a deadlock here.
388 	 *
389 	 * We split the two cases (up and down) so that the direction
390 	 * is not being tested on each round of the loop.
391 	 */
392 	if (first > last) {
393 		/*
394 		 * counting down
395 		 */
396 		in_pcbportrange(&first, &last, portinfo->offset, step);
397 		count = (first - last) / step;
398 
399 		do {
400 			if (count-- < 0) {	/* completely used? */
401 				error = EADDRNOTAVAIL;
402 				goto done;
403 			}
404 			*lastport -= step;
405 			if (*lastport > first || *lastport < last)
406 				*lastport = first;
407 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
408 			    portinfo->offset);
409 			lport = htons(*lastport);
410 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
411 		    wild, cred));
412 	} else {
413 		/*
414 		 * counting up
415 		 */
416 		in_pcbportrange(&last, &first, portinfo->offset, step);
417 		count = (last - first) / step;
418 
419 		do {
420 			if (count-- < 0) {	/* completely used? */
421 				error = EADDRNOTAVAIL;
422 				goto done;
423 			}
424 			*lastport += step;
425 			if (*lastport < first || *lastport > last)
426 				*lastport = first;
427 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
428 			    portinfo->offset);
429 			lport = htons(*lastport);
430 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
431 		    wild, cred));
432 	}
433 	inp->inp_lport = lport;
434 	in_pcbinsporthash(portinfo, inp);
435 	error = 0;
436 done:
437 	REL_PORT_TOKEN(portinfo);
438 
439 	if (error) {
440 		/* Try next portinfo */
441 		portinfo_idx++;
442 		portinfo_idx &= pcbinfo->portinfo_mask;
443 		if (portinfo_idx != portinfo_first)
444 			goto loop;
445 		inp->inp_laddr.s_addr = INADDR_ANY;
446 	}
447 	return error;
448 }
449 
450 int
451 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
452 {
453 	struct socket *so = inp->inp_socket;
454 	struct sockaddr_in jsin;
455 	struct ucred *cred = NULL;
456 	int wild = 0;
457 
458 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
459 		return (EADDRNOTAVAIL);
460 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
461 		return (EINVAL);	/* already bound */
462 
463 	if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
464 		wild = 1;    /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
465 	if (td->td_proc)
466 		cred = td->td_proc->p_ucred;
467 
468 	if (nam != NULL) {
469 		struct sockaddr_in *sin = (struct sockaddr_in *)nam;
470 		struct inpcbinfo *pcbinfo;
471 		struct inpcbportinfo *portinfo;
472 		struct inpcb *t;
473 		u_short lport, lport_ho;
474 		int reuseport = (so->so_options & SO_REUSEPORT);
475 		int error;
476 
477 		if (nam->sa_len != sizeof *sin)
478 			return (EINVAL);
479 #ifdef notdef
480 		/*
481 		 * We should check the family, but old programs
482 		 * incorrectly fail to initialize it.
483 		 */
484 		if (sin->sin_family != AF_INET)
485 			return (EAFNOSUPPORT);
486 #endif
487 		if (!prison_replace_wildcards(td, nam))
488 			return (EINVAL);
489 
490 		lport = sin->sin_port;
491 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
492 			/*
493 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
494 			 * allow complete duplication of binding if
495 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
496 			 * and a multicast address is bound on both
497 			 * new and duplicated sockets.
498 			 */
499 			if (so->so_options & SO_REUSEADDR)
500 				reuseport = SO_REUSEADDR | SO_REUSEPORT;
501 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
502 			sin->sin_port = 0;		/* yech... */
503 			bzero(&sin->sin_zero, sizeof sin->sin_zero);
504 			if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
505 				return (EADDRNOTAVAIL);
506 		}
507 
508 		inp->inp_laddr = sin->sin_addr;
509 
510 		jsin.sin_family = AF_INET;
511 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
512 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
513 			inp->inp_laddr.s_addr = INADDR_ANY;
514 			return (EINVAL);
515 		}
516 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
517 
518 		if (lport == 0) {
519 			/* Auto-select local port */
520 			return in_pcbsetlport(inp, wild, cred);
521 		}
522 		lport_ho = ntohs(lport);
523 
524 		/* GROSS */
525 		if (lport_ho < IPPORT_RESERVED && cred &&
526 		    (error =
527 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
528 			inp->inp_laddr.s_addr = INADDR_ANY;
529 			return (error);
530 		}
531 
532 		/*
533 		 * Locate the proper portinfo based on lport
534 		 */
535 		pcbinfo = inp->inp_pcbinfo;
536 		portinfo =
537 		    &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
538 		KKASSERT((lport_ho & pcbinfo->portinfo_mask) ==
539 		    portinfo->offset);
540 
541 		/*
542 		 * This has to be atomic.  If the porthash is shared across
543 		 * multiple protocol threads (aka tcp) then the token must
544 		 * be held.
545 		 */
546 		GET_PORT_TOKEN(portinfo);
547 
548 		if (so->so_cred->cr_uid != 0 &&
549 		    !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
550 			t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
551 			    INPLOOKUP_WILDCARD, cred);
552 			if (t &&
553 			    (!in_nullhost(sin->sin_addr) ||
554 			     !in_nullhost(t->inp_laddr) ||
555 			     (t->inp_socket->so_options & SO_REUSEPORT) == 0) &&
556 			    (so->so_cred->cr_uid !=
557 			     t->inp_socket->so_cred->cr_uid)) {
558 #ifdef INET6
559 				if (!in_nullhost(sin->sin_addr) ||
560 				    !in_nullhost(t->inp_laddr) ||
561 				    INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
562 #endif
563 				{
564 					inp->inp_laddr.s_addr = INADDR_ANY;
565 					error = EADDRINUSE;
566 					goto done;
567 				}
568 			}
569 		}
570 		if (cred && !prison_replace_wildcards(td, nam)) {
571 			inp->inp_laddr.s_addr = INADDR_ANY;
572 			error = EADDRNOTAVAIL;
573 			goto done;
574 		}
575 		t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
576 		    wild, cred);
577 		if (t && !(reuseport & t->inp_socket->so_options)) {
578 #ifdef INET6
579 			if (!in_nullhost(sin->sin_addr) ||
580 			    !in_nullhost(t->inp_laddr) ||
581 			    INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket))
582 #endif
583 			{
584 				inp->inp_laddr.s_addr = INADDR_ANY;
585 				error = EADDRINUSE;
586 				goto done;
587 			}
588 		}
589 		inp->inp_lport = lport;
590 		in_pcbinsporthash(portinfo, inp);
591 		error = 0;
592 done:
593 		REL_PORT_TOKEN(portinfo);
594 		return (error);
595 	} else {
596 		jsin.sin_family = AF_INET;
597 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
598 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
599 			inp->inp_laddr.s_addr = INADDR_ANY;
600 			return (EINVAL);
601 		}
602 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
603 
604 		return in_pcbsetlport(inp, wild, cred);
605 	}
606 }
607 
608 static struct inpcb *
609 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr,
610     u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
611 {
612 	struct inpcb *inp;
613 	struct inpcbporthead *porthash;
614 	struct inpcbport *phd;
615 	struct inpcb *match = NULL;
616 
617 	/*
618 	 * If the porthashbase is shared across several cpus, it must
619 	 * have been locked.
620 	 */
621 	ASSERT_PORT_TOKEN_HELD(portinfo);
622 
623 	/*
624 	 * Best fit PCB lookup.
625 	 *
626 	 * First see if this local port is in use by looking on the
627 	 * port hash list.
628 	 */
629 	porthash = &portinfo->porthashbase[
630 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
631 	LIST_FOREACH(phd, porthash, phd_hash) {
632 		if (phd->phd_port == lport)
633 			break;
634 	}
635 	if (phd != NULL) {
636 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
637 #ifdef INET6
638 			if ((inp->inp_vflag & INP_IPV4) == 0)
639 				continue;
640 #endif
641 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
642 			    inp->inp_laddr.s_addr != laddr.s_addr)
643 				continue;
644 
645 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
646 			    inp->inp_faddr.s_addr != faddr.s_addr)
647 				continue;
648 
649 			if (inp->inp_fport != 0 && inp->inp_fport != fport)
650 				continue;
651 
652 			if (cred == NULL ||
653 			    cred->cr_prison ==
654 			    inp->inp_socket->so_cred->cr_prison) {
655 				match = inp;
656 				break;
657 			}
658 		}
659 	}
660 	return (match);
661 }
662 
663 int
664 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
665     struct thread *td)
666 {
667 	struct proc *p = td->td_proc;
668 	unsigned short *lastport;
669 	const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
670 	struct sockaddr_in jsin;
671 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
672 	struct inpcbportinfo *portinfo;
673 	struct ucred *cred = NULL;
674 	u_short first, last, lport, step;
675 	int count, error, dup;
676 	int portinfo_first, portinfo_idx;
677 
678 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
679 		return (EADDRNOTAVAIL);
680 
681 	KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
682 	if (inp->inp_lport != 0)
683 		return (EINVAL);	/* already bound */
684 
685 	KKASSERT(p);
686 	cred = p->p_ucred;
687 
688 	jsin.sin_family = AF_INET;
689 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
690 	if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
691 		inp->inp_laddr.s_addr = INADDR_ANY;
692 		return (EINVAL);
693 	}
694 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
695 
696 	inp->inp_flags |= INP_ANONPORT;
697 
698 	step = pcbinfo->portinfo_mask + 1;
699 	portinfo_first = mycpuid & pcbinfo->portinfo_mask;
700 	portinfo_idx = portinfo_first;
701 loop:
702 	portinfo = &pcbinfo->portinfo[portinfo_idx];
703 	dup = 0;
704 
705 	if (inp->inp_flags & INP_HIGHPORT) {
706 		first = ipport_hifirstauto;	/* sysctl */
707 		last  = ipport_hilastauto;
708 		lastport = &portinfo->lasthi;
709 	} else if (inp->inp_flags & INP_LOWPORT) {
710 		if (cred &&
711 		    (error =
712 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
713 			inp->inp_laddr.s_addr = INADDR_ANY;
714 			return (error);
715 		}
716 		first = ipport_lowfirstauto;	/* 1023 */
717 		last  = ipport_lowlastauto;	/* 600 */
718 		lastport = &portinfo->lastlow;
719 	} else {
720 		first = ipport_firstauto;	/* sysctl */
721 		last  = ipport_lastauto;
722 		lastport = &portinfo->lastport;
723 	}
724 
725 	/*
726 	 * This has to be atomic.  If the porthash is shared across multiple
727 	 * protocol threads (aka tcp) then the token must be held.
728 	 */
729 	GET_PORT_TOKEN(portinfo);
730 
731 again:
732 	/*
733 	 * Simple check to ensure all ports are not used up causing
734 	 * a deadlock here.
735 	 *
736 	 * We split the two cases (up and down) so that the direction
737 	 * is not being tested on each round of the loop.
738 	 */
739 	if (first > last) {
740 		/*
741 		 * counting down
742 		 */
743 		in_pcbportrange(&first, &last, portinfo->offset, step);
744 		count = (first - last) / step;
745 
746 		do {
747 			if (count-- < 0) {	/* completely used? */
748 				error = EADDRNOTAVAIL;
749 				goto done;
750 			}
751 			*lastport -= step;
752 			if (*lastport > first || *lastport < last)
753 				*lastport = first;
754 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
755 			    portinfo->offset);
756 			lport = htons(*lastport);
757 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
758 		    lport, sin->sin_addr, sin->sin_port, cred));
759 	} else {
760 		/*
761 		 * counting up
762 		 */
763 		in_pcbportrange(&last, &first, portinfo->offset, step);
764 		count = (last - first) / step;
765 
766 		do {
767 			if (count-- < 0) {	/* completely used? */
768 				error = EADDRNOTAVAIL;
769 				goto done;
770 			}
771 			*lastport += step;
772 			if (*lastport < first || *lastport > last)
773 				*lastport = first;
774 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
775 			    portinfo->offset);
776 			lport = htons(*lastport);
777 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
778 		    lport, sin->sin_addr, sin->sin_port, cred));
779 	}
780 
781 	/* This could happen on loopback interface */
782 	if (sin->sin_port == lport &&
783 	    sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
784 		if (dup) {
785 			/*
786 			 * Duplicate again; give up
787 			 */
788 			error = EADDRNOTAVAIL;
789 			goto done;
790 		}
791 		dup = 1;
792 		goto again;
793 	}
794 	inp->inp_lport = lport;
795 	in_pcbinsporthash(portinfo, inp);
796 	error = 0;
797 done:
798 	REL_PORT_TOKEN(portinfo);
799 
800 	if (error) {
801 		/* Try next portinfo */
802 		portinfo_idx++;
803 		portinfo_idx &= pcbinfo->portinfo_mask;
804 		if (portinfo_idx != portinfo_first)
805 			goto loop;
806 		inp->inp_laddr.s_addr = INADDR_ANY;
807 	}
808 	return error;
809 }
810 
811 /*
812  *   Transform old in_pcbconnect() into an inner subroutine for new
813  *   in_pcbconnect(): Do some validity-checking on the remote
814  *   address (in mbuf 'nam') and then determine local host address
815  *   (i.e., which interface) to use to access that remote host.
816  *
817  *   This preserves definition of in_pcbconnect(), while supporting a
818  *   slightly different version for T/TCP.  (This is more than
819  *   a bit of a kludge, but cleaning up the internal interfaces would
820  *   have forced minor changes in every protocol).
821  */
822 int
823 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
824     struct sockaddr_in **plocal_sin, struct thread *td, int find)
825 {
826 	struct in_ifaddr *ia;
827 	struct ucred *cred = NULL;
828 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
829 	struct sockaddr *jsin;
830 	int jailed = 0, alloc_route = 0;
831 
832 	if (nam->sa_len != sizeof *sin)
833 		return (EINVAL);
834 	if (sin->sin_family != AF_INET)
835 		return (EAFNOSUPPORT);
836 	if (sin->sin_port == 0)
837 		return (EADDRNOTAVAIL);
838 	if (td && td->td_proc && td->td_proc->p_ucred)
839 		cred = td->td_proc->p_ucred;
840 	if (cred && cred->cr_prison)
841 		jailed = 1;
842 	if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
843 		ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
844 		/*
845 		 * If the destination address is INADDR_ANY,
846 		 * use the primary local address.
847 		 * If the supplied address is INADDR_BROADCAST,
848 		 * and the primary interface supports broadcast,
849 		 * choose the broadcast address for that interface.
850 		 */
851 		if (sin->sin_addr.s_addr == INADDR_ANY)
852 			sin->sin_addr = IA_SIN(ia)->sin_addr;
853 		else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
854 		    (ia->ia_ifp->if_flags & IFF_BROADCAST))
855 			sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
856 	}
857 	if (find) {
858 		struct route *ro;
859 
860 		ia = NULL;
861 		/*
862 		 * If route is known or can be allocated now,
863 		 * our src addr is taken from the i/f, else punt.
864 		 * Note that we should check the address family of the cached
865 		 * destination, in case of sharing the cache with IPv6.
866 		 */
867 		ro = &inp->inp_route;
868 		if (ro->ro_rt &&
869 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
870 		     ro->ro_dst.sa_family != AF_INET ||
871 		     satosin(&ro->ro_dst)->sin_addr.s_addr !=
872 				      sin->sin_addr.s_addr ||
873 		     inp->inp_socket->so_options & SO_DONTROUTE)) {
874 			RTFREE(ro->ro_rt);
875 			ro->ro_rt = NULL;
876 		}
877 		if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
878 		    (ro->ro_rt == NULL ||
879 		    ro->ro_rt->rt_ifp == NULL)) {
880 			/* No route yet, so try to acquire one */
881 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
882 			ro->ro_dst.sa_family = AF_INET;
883 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
884 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
885 				sin->sin_addr;
886 			rtalloc(ro);
887 			alloc_route = 1;
888 		}
889 		/*
890 		 * If we found a route, use the address
891 		 * corresponding to the outgoing interface
892 		 * unless it is the loopback (in case a route
893 		 * to our address on another net goes to loopback).
894 		 */
895 		if (ro->ro_rt &&
896 		    !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
897 			if (jailed) {
898 				if (jailed_ip(cred->cr_prison,
899 				    ro->ro_rt->rt_ifa->ifa_addr)) {
900 					ia = ifatoia(ro->ro_rt->rt_ifa);
901 				}
902 			} else {
903 				ia = ifatoia(ro->ro_rt->rt_ifa);
904 			}
905 		}
906 		if (ia == NULL) {
907 			u_short fport = sin->sin_port;
908 
909 			sin->sin_port = 0;
910 			ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
911 			if (ia && jailed && !jailed_ip(cred->cr_prison,
912 			    sintosa(&ia->ia_addr)))
913 				ia = NULL;
914 			if (ia == NULL)
915 				ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
916 			if (ia && jailed && !jailed_ip(cred->cr_prison,
917 			    sintosa(&ia->ia_addr)))
918 				ia = NULL;
919 			sin->sin_port = fport;
920 			if (ia == NULL &&
921 			    !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
922 				ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
923 			if (ia && jailed && !jailed_ip(cred->cr_prison,
924 			    sintosa(&ia->ia_addr)))
925 				ia = NULL;
926 
927 			if (!jailed && ia == NULL)
928 				goto fail;
929 		}
930 		/*
931 		 * If the destination address is multicast and an outgoing
932 		 * interface has been set as a multicast option, use the
933 		 * address of that interface as our source address.
934 		 */
935 		if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
936 		    inp->inp_moptions != NULL) {
937 			struct ip_moptions *imo;
938 			struct ifnet *ifp;
939 
940 			imo = inp->inp_moptions;
941 			if (imo->imo_multicast_ifp != NULL) {
942 				struct in_ifaddr_container *iac;
943 
944 				ifp = imo->imo_multicast_ifp;
945 				ia = NULL;
946 				TAILQ_FOREACH(iac,
947 				&in_ifaddrheads[mycpuid], ia_link) {
948 					if (iac->ia->ia_ifp == ifp) {
949 						ia = iac->ia;
950 						break;
951 					}
952 				}
953 				if (ia == NULL)
954 					goto fail;
955 			}
956 		}
957 		/*
958 		 * Don't do pcblookup call here; return interface in plocal_sin
959 		 * and exit to caller, that will do the lookup.
960 		 */
961 		if (ia == NULL && jailed) {
962 			if ((jsin = prison_get_nonlocal(
963 				cred->cr_prison, AF_INET, NULL)) != NULL ||
964 			    (jsin = prison_get_local(
965 				cred->cr_prison, AF_INET, NULL)) != NULL) {
966 				*plocal_sin = satosin(jsin);
967 			} else {
968 				/* IPv6 only Jail */
969 				goto fail;
970 			}
971 		} else {
972 			*plocal_sin = &ia->ia_addr;
973 		}
974 	}
975 	return (0);
976 fail:
977 	if (alloc_route) {
978 		struct route *ro = &inp->inp_route;
979 
980 		if (ro->ro_rt != NULL)
981 			RTFREE(ro->ro_rt);
982 		bzero(ro, sizeof(*ro));
983 	}
984 	return (EADDRNOTAVAIL);
985 }
986 
987 int
988 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
989     struct sockaddr_in **plocal_sin, struct thread *td)
990 {
991 	return in_pcbladdr_find(inp, nam, plocal_sin, td,
992 	    (inp->inp_laddr.s_addr == INADDR_ANY));
993 }
994 
995 /*
996  * Outer subroutine:
997  * Connect from a socket to a specified address.
998  * Both address and port must be specified in argument sin.
999  * If don't have a local address for this socket yet,
1000  * then pick one.
1001  */
1002 int
1003 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
1004 {
1005 	struct sockaddr_in *if_sin;
1006 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1007 	int error;
1008 
1009 	/* Call inner routine to assign local interface address. */
1010 	if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1011 		return (error);
1012 
1013 	if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1014 			      inp->inp_laddr.s_addr ?
1015 				inp->inp_laddr : if_sin->sin_addr,
1016 			      inp->inp_lport, FALSE, NULL) != NULL) {
1017 		return (EADDRINUSE);
1018 	}
1019 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
1020 		if (inp->inp_lport == 0) {
1021 			error = in_pcbbind(inp, NULL, td);
1022 			if (error)
1023 				return (error);
1024 		}
1025 		inp->inp_laddr = if_sin->sin_addr;
1026 	}
1027 	inp->inp_faddr = sin->sin_addr;
1028 	inp->inp_fport = sin->sin_port;
1029 	in_pcbinsconnhash(inp);
1030 	return (0);
1031 }
1032 
1033 void
1034 in_pcbdisconnect(struct inpcb *inp)
1035 {
1036 
1037 	in_pcbremconnhash(inp);
1038 	inp->inp_faddr.s_addr = INADDR_ANY;
1039 	inp->inp_fport = 0;
1040 }
1041 
1042 void
1043 in_pcbdetach(struct inpcb *inp)
1044 {
1045 	struct socket *so = inp->inp_socket;
1046 	struct inpcbinfo *ipi = inp->inp_pcbinfo;
1047 
1048 #ifdef IPSEC
1049 	ipsec4_delete_pcbpolicy(inp);
1050 #endif /*IPSEC*/
1051 	inp->inp_gencnt = ++ipi->ipi_gencnt;
1052 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1053 	in_pcbremlists(inp);
1054 	so->so_pcb = NULL;
1055 	sofree(so);			/* remove pcb ref */
1056 	if (inp->inp_options)
1057 		m_free(inp->inp_options);
1058 	if (inp->inp_route.ro_rt)
1059 		rtfree(inp->inp_route.ro_rt);
1060 	ip_freemoptions(inp->inp_moptions);
1061 	inp->inp_vflag = 0;
1062 	kfree(inp, M_PCB);
1063 }
1064 
1065 /*
1066  * The calling convention of in_setsockaddr() and in_setpeeraddr() was
1067  * modified to match the pru_sockaddr() and pru_peeraddr() entry points
1068  * in struct pr_usrreqs, so that protocols can just reference then directly
1069  * without the need for a wrapper function.  The socket must have a valid
1070  * (i.e., non-nil) PCB, but it should be impossible to get an invalid one
1071  * except through a kernel programming error, so it is acceptable to panic
1072  * (or in this case trap) if the PCB is invalid.  (Actually, we don't trap
1073  * because there actually /is/ a programming error somewhere... XXX)
1074  */
1075 int
1076 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1077 {
1078 	struct inpcb *inp;
1079 	struct sockaddr_in *sin;
1080 
1081 	/*
1082 	 * Do the malloc first in case it blocks.
1083 	 */
1084 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1085 	sin->sin_family = AF_INET;
1086 	sin->sin_len = sizeof *sin;
1087 
1088 	crit_enter();
1089 	inp = so->so_pcb;
1090 	if (!inp) {
1091 		crit_exit();
1092 		kfree(sin, M_SONAME);
1093 		return (ECONNRESET);
1094 	}
1095 	sin->sin_port = inp->inp_lport;
1096 	sin->sin_addr = inp->inp_laddr;
1097 	crit_exit();
1098 
1099 	*nam = (struct sockaddr *)sin;
1100 	return (0);
1101 }
1102 
1103 void
1104 in_setsockaddr_dispatch(netmsg_t msg)
1105 {
1106 	int error;
1107 
1108 	error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1109 	lwkt_replymsg(&msg->lmsg, error);
1110 }
1111 
1112 int
1113 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1114 {
1115 	struct inpcb *inp;
1116 	struct sockaddr_in *sin;
1117 
1118 	/*
1119 	 * Do the malloc first in case it blocks.
1120 	 */
1121 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1122 	sin->sin_family = AF_INET;
1123 	sin->sin_len = sizeof *sin;
1124 
1125 	crit_enter();
1126 	inp = so->so_pcb;
1127 	if (!inp) {
1128 		crit_exit();
1129 		kfree(sin, M_SONAME);
1130 		return (ECONNRESET);
1131 	}
1132 	sin->sin_port = inp->inp_fport;
1133 	sin->sin_addr = inp->inp_faddr;
1134 	crit_exit();
1135 
1136 	*nam = (struct sockaddr *)sin;
1137 	return (0);
1138 }
1139 
1140 void
1141 in_setpeeraddr_dispatch(netmsg_t msg)
1142 {
1143 	int error;
1144 
1145 	error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1146 	lwkt_replymsg(&msg->lmsg, error);
1147 }
1148 
1149 void
1150 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1151 		void (*notify)(struct inpcb *, int))
1152 {
1153 	struct inpcb *inp, *marker;
1154 
1155 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1156 	    ("not in the correct netisr"));
1157 	marker = &in_pcbmarkers[mycpuid];
1158 
1159 	/*
1160 	 * NOTE:
1161 	 * - If INP_PLACEMARKER is set we must ignore the rest of the
1162 	 *   structure and skip it.
1163 	 * - It is safe to nuke inpcbs here, since we are in their own
1164 	 *   netisr.
1165 	 */
1166 	GET_PCBINFO_TOKEN(pcbinfo);
1167 
1168 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1169 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1170 		LIST_REMOVE(marker, inp_list);
1171 		LIST_INSERT_AFTER(inp, marker, inp_list);
1172 
1173 		if (inp->inp_flags & INP_PLACEMARKER)
1174 			continue;
1175 #ifdef INET6
1176 		if (!(inp->inp_vflag & INP_IPV4))
1177 			continue;
1178 #endif
1179 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1180 		    inp->inp_socket == NULL)
1181 			continue;
1182 		(*notify)(inp, err);		/* can remove inp from list! */
1183 	}
1184 	LIST_REMOVE(marker, inp_list);
1185 
1186 	REL_PCBINFO_TOKEN(pcbinfo);
1187 }
1188 
1189 void
1190 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1191 {
1192 	struct inpcb *inp, *marker;
1193 
1194 	/*
1195 	 * We only need to make sure that we are in netisr0, where all
1196 	 * multicast operation happen.  We could check inpcbinfo which
1197 	 * does not belong to netisr0 by holding the inpcbinfo's token.
1198 	 * In this case, the pcbinfo must be able to be shared, i.e.
1199 	 * pcbinfo->infotoken is not NULL.
1200 	 */
1201 	KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1202 	    ("not in netisr0"));
1203 	KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1204 	    ("pcbinfo could not be shared"));
1205 
1206 	/*
1207 	 * Get a marker for the current netisr (netisr0).
1208 	 *
1209 	 * It is possible that the multicast address deletion blocks,
1210 	 * which could cause temporary token releasing.  So we use
1211 	 * inpcb marker here to get a coherent view of the inpcb list.
1212 	 *
1213 	 * While, on the other hand, moptions are only added and deleted
1214 	 * in netisr0, so we would not see staled moption or miss moption
1215 	 * even if the token was released due to the blocking multicast
1216 	 * address deletion.
1217 	 */
1218 	marker = &in_pcbmarkers[mycpuid];
1219 
1220 	GET_PCBINFO_TOKEN(pcbinfo);
1221 
1222 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1223 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1224 		struct ip_moptions *imo;
1225 
1226 		LIST_REMOVE(marker, inp_list);
1227 		LIST_INSERT_AFTER(inp, marker, inp_list);
1228 
1229 		if (inp->inp_flags & INP_PLACEMARKER)
1230 			continue;
1231 		imo = inp->inp_moptions;
1232 		if ((inp->inp_vflag & INP_IPV4) && imo != NULL) {
1233 			int i, gap;
1234 
1235 			/*
1236 			 * Unselect the outgoing interface if it is being
1237 			 * detached.
1238 			 */
1239 			if (imo->imo_multicast_ifp == ifp)
1240 				imo->imo_multicast_ifp = NULL;
1241 
1242 			/*
1243 			 * Drop multicast group membership if we joined
1244 			 * through the interface being detached.
1245 			 */
1246 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1247 			    i++) {
1248 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1249 					/*
1250 					 * NOTE:
1251 					 * This could block and the pcbinfo
1252 					 * token could be passively released.
1253 					 */
1254 					in_delmulti(imo->imo_membership[i]);
1255 					gap++;
1256 				} else if (gap != 0)
1257 					imo->imo_membership[i - gap] =
1258 					    imo->imo_membership[i];
1259 			}
1260 			imo->imo_num_memberships -= gap;
1261 		}
1262 	}
1263 	LIST_REMOVE(marker, inp_list);
1264 
1265 	REL_PCBINFO_TOKEN(pcbinfo);
1266 }
1267 
1268 /*
1269  * Check for alternatives when higher level complains
1270  * about service problems.  For now, invalidate cached
1271  * routing information.  If the route was created dynamically
1272  * (by a redirect), time to try a default gateway again.
1273  */
1274 void
1275 in_losing(struct inpcb *inp)
1276 {
1277 	struct rtentry *rt;
1278 	struct rt_addrinfo rtinfo;
1279 
1280 	if ((rt = inp->inp_route.ro_rt)) {
1281 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
1282 		rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1283 		rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1284 		rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1285 		rtinfo.rti_flags = rt->rt_flags;
1286 		rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1287 		if (rt->rt_flags & RTF_DYNAMIC) {
1288 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1289 			    rt_mask(rt), rt->rt_flags, NULL);
1290 		}
1291 		inp->inp_route.ro_rt = NULL;
1292 		rtfree(rt);
1293 		/*
1294 		 * A new route can be allocated
1295 		 * the next time output is attempted.
1296 		 */
1297 	}
1298 }
1299 
1300 /*
1301  * After a routing change, flush old routing
1302  * and allocate a (hopefully) better one.
1303  */
1304 void
1305 in_rtchange(struct inpcb *inp, int err)
1306 {
1307 	if (inp->inp_route.ro_rt) {
1308 		rtfree(inp->inp_route.ro_rt);
1309 		inp->inp_route.ro_rt = NULL;
1310 		/*
1311 		 * A new route can be allocated the next time
1312 		 * output is attempted.
1313 		 */
1314 	}
1315 }
1316 
1317 /*
1318  * Lookup a PCB based on the local address and port.
1319  */
1320 struct inpcb *
1321 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr,
1322 		   u_int lport_arg, int wild_okay, struct ucred *cred)
1323 {
1324 	struct inpcb *inp;
1325 	int matchwild = 3, wildcard;
1326 	u_short lport = lport_arg;
1327 	struct inpcbporthead *porthash;
1328 	struct inpcbport *phd;
1329 	struct inpcb *match = NULL;
1330 
1331 	/*
1332 	 * If the porthashbase is shared across several cpus, it must
1333 	 * have been locked.
1334 	 */
1335 	ASSERT_PORT_TOKEN_HELD(portinfo);
1336 
1337 	/*
1338 	 * Best fit PCB lookup.
1339 	 *
1340 	 * First see if this local port is in use by looking on the
1341 	 * port hash list.
1342 	 */
1343 	porthash = &portinfo->porthashbase[
1344 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
1345 	LIST_FOREACH(phd, porthash, phd_hash) {
1346 		if (phd->phd_port == lport)
1347 			break;
1348 	}
1349 	if (phd != NULL) {
1350 		/*
1351 		 * Port is in use by one or more PCBs. Look for best
1352 		 * fit.
1353 		 */
1354 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1355 			wildcard = 0;
1356 #ifdef INET6
1357 			if ((inp->inp_vflag & INP_IPV4) == 0)
1358 				continue;
1359 #endif
1360 			if (inp->inp_faddr.s_addr != INADDR_ANY)
1361 				wildcard++;
1362 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
1363 				if (laddr.s_addr == INADDR_ANY)
1364 					wildcard++;
1365 				else if (inp->inp_laddr.s_addr != laddr.s_addr)
1366 					continue;
1367 			} else {
1368 				if (laddr.s_addr != INADDR_ANY)
1369 					wildcard++;
1370 			}
1371 			if (wildcard && !wild_okay)
1372 				continue;
1373 			if (wildcard < matchwild &&
1374 			    (cred == NULL ||
1375 			     cred->cr_prison ==
1376 					inp->inp_socket->so_cred->cr_prison)) {
1377 				match = inp;
1378 				matchwild = wildcard;
1379 				if (matchwild == 0) {
1380 					break;
1381 				}
1382 			}
1383 		}
1384 	}
1385 	return (match);
1386 }
1387 
1388 struct inpcb *
1389 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1390     const struct inpcb *inp)
1391 {
1392 	const struct inp_localgrphead *hdr;
1393 	const struct inp_localgroup *grp;
1394 	int i;
1395 
1396 	if (pcbinfo->localgrphashbase == NULL)
1397 		return NULL;
1398 
1399 	GET_PCBINFO_TOKEN(pcbinfo);
1400 
1401 	hdr = &pcbinfo->localgrphashbase[
1402 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1403 
1404 	LIST_FOREACH(grp, hdr, il_list) {
1405 		if (grp->il_vflag == inp->inp_vflag &&
1406 		    grp->il_lport == inp->inp_lport &&
1407 		    memcmp(&grp->il_dependladdr,
1408 			&inp->inp_inc.inc_ie.ie_dependladdr,
1409 			sizeof(grp->il_dependladdr)) == 0) {
1410 			break;
1411 		}
1412 	}
1413 	if (grp == NULL || grp->il_inpcnt == 1) {
1414 		REL_PCBINFO_TOKEN(pcbinfo);
1415 		return NULL;
1416 	}
1417 
1418 	KASSERT(grp->il_inpcnt >= 2,
1419 	    ("invalid localgroup inp count %d", grp->il_inpcnt));
1420 	for (i = 0; i < grp->il_inpcnt; ++i) {
1421 		if (grp->il_inp[i] == inp) {
1422 			int last = grp->il_inpcnt - 1;
1423 
1424 			if (i == last)
1425 				last = grp->il_inpcnt - 2;
1426 			REL_PCBINFO_TOKEN(pcbinfo);
1427 			return grp->il_inp[last];
1428 		}
1429 	}
1430 	REL_PCBINFO_TOKEN(pcbinfo);
1431 	return NULL;
1432 }
1433 
1434 static struct inpcb *
1435 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1436     struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1437 {
1438 	struct inpcb *local_wild = NULL;
1439 	const struct inp_localgrphead *hdr;
1440 	const struct inp_localgroup *grp;
1441 
1442 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1443 
1444 	hdr = &pcbinfo->localgrphashbase[
1445 	    INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1446 #ifdef INP_LOCALGROUP_HASHTHR
1447 	pkt_hash >>= ncpus2_shift;
1448 #endif
1449 
1450 	/*
1451 	 * Order of socket selection:
1452 	 * 1. non-wild.
1453 	 * 2. wild.
1454 	 *
1455 	 * NOTE:
1456 	 * - Local group does not contain jailed sockets
1457 	 * - Local group does not contain IPv4 mapped INET6 wild sockets
1458 	 */
1459 	LIST_FOREACH(grp, hdr, il_list) {
1460 #ifdef INET6
1461 		if (!(grp->il_vflag & INP_IPV4))
1462 			continue;
1463 #endif
1464 		if (grp->il_lport == lport) {
1465 			int idx;
1466 
1467 #ifdef INP_LOCALGROUP_HASHTHR
1468 			idx = pkt_hash / grp->il_factor;
1469 			KASSERT(idx < grp->il_inpcnt && idx >= 0,
1470 			    ("invalid hash %04x, cnt %d or fact %d",
1471 			     pkt_hash, grp->il_inpcnt, grp->il_factor));
1472 #else
1473 			/*
1474 			 * Modulo-N is used here, which greatly reduces
1475 			 * completion queue token contention, thus more
1476 			 * cpu time is saved.
1477 			 */
1478 			idx = pkt_hash % grp->il_inpcnt;
1479 #endif
1480 
1481 			if (grp->il_laddr.s_addr == laddr.s_addr)
1482 				return grp->il_inp[idx];
1483 			else if (grp->il_laddr.s_addr == INADDR_ANY)
1484 				local_wild = grp->il_inp[idx];
1485 		}
1486 	}
1487 	if (local_wild != NULL)
1488 		return local_wild;
1489 	return NULL;
1490 }
1491 
1492 /*
1493  * Lookup PCB in hash list.
1494  */
1495 struct inpcb *
1496 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1497     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1498     boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1499 {
1500 	struct inpcbhead *head;
1501 	struct inpcb *inp, *jinp=NULL;
1502 	u_short fport = fport_arg, lport = lport_arg;
1503 
1504 	/*
1505 	 * First look for an exact match.
1506 	 */
1507 	head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1508 	    laddr.s_addr, lport, pcbinfo->hashmask)];
1509 	LIST_FOREACH(inp, head, inp_hash) {
1510 #ifdef INET6
1511 		if (!(inp->inp_vflag & INP_IPV4))
1512 			continue;
1513 #endif
1514 		if (in_hosteq(inp->inp_faddr, faddr) &&
1515 		    in_hosteq(inp->inp_laddr, laddr) &&
1516 		    inp->inp_fport == fport && inp->inp_lport == lport) {
1517 			/* found */
1518 			if (inp->inp_socket == NULL ||
1519 			    inp->inp_socket->so_cred->cr_prison == NULL) {
1520 				return (inp);
1521 			} else {
1522 				if  (jinp == NULL)
1523 					jinp = inp;
1524 			}
1525 		}
1526 	}
1527 	if (jinp != NULL)
1528 		return (jinp);
1529 
1530 	if (wildcard) {
1531 		struct inpcb *local_wild = NULL;
1532 		struct inpcb *jinp_wild = NULL;
1533 #ifdef INET6
1534 		struct inpcb *local_wild_mapped = NULL;
1535 #endif
1536 		struct inpcontainer *ic;
1537 		struct inpcontainerhead *chead;
1538 		struct sockaddr_in jsin;
1539 		struct ucred *cred;
1540 
1541 		GET_PCBINFO_TOKEN(pcbinfo);
1542 
1543 		/*
1544 		 * Check local group first
1545 		 */
1546 		if (pcbinfo->localgrphashbase != NULL &&
1547 		    m != NULL && (m->m_flags & M_HASH) &&
1548 		    !(ifp && ifp->if_type == IFT_FAITH)) {
1549 			inp = inp_localgroup_lookup(pcbinfo,
1550 			    laddr, lport, m->m_pkthdr.hash);
1551 			if (inp != NULL) {
1552 				REL_PCBINFO_TOKEN(pcbinfo);
1553 				return inp;
1554 			}
1555 		}
1556 
1557 		/*
1558 		 * Order of socket selection:
1559 		 * 1. non-jailed, non-wild.
1560 		 * 2. non-jailed, wild.
1561 		 * 3. jailed, non-wild.
1562 		 * 4. jailed, wild.
1563 		 */
1564 		jsin.sin_family = AF_INET;
1565 		chead = &pcbinfo->wildcardhashbase[
1566 		    INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1567 		LIST_FOREACH(ic, chead, ic_list) {
1568 			inp = ic->ic_inp;
1569 			if (inp->inp_flags & INP_PLACEMARKER)
1570 				continue;
1571 
1572 			jsin.sin_addr.s_addr = laddr.s_addr;
1573 #ifdef INET6
1574 			if (!(inp->inp_vflag & INP_IPV4))
1575 				continue;
1576 #endif
1577 			if (inp->inp_socket != NULL)
1578 				cred = inp->inp_socket->so_cred;
1579 			else
1580 				cred = NULL;
1581 			if (cred != NULL && jailed(cred)) {
1582 				if (jinp != NULL)
1583 					continue;
1584 				else
1585 					if (!jailed_ip(cred->cr_prison,
1586 					    (struct sockaddr *)&jsin))
1587 						continue;
1588 			}
1589 			if (inp->inp_lport == lport) {
1590 				if (ifp && ifp->if_type == IFT_FAITH &&
1591 				    !(inp->inp_flags & INP_FAITH))
1592 					continue;
1593 				if (inp->inp_laddr.s_addr == laddr.s_addr) {
1594 					if (cred != NULL && jailed(cred)) {
1595 						jinp = inp;
1596 					} else {
1597 						REL_PCBINFO_TOKEN(pcbinfo);
1598 						return (inp);
1599 					}
1600 				}
1601 				if (inp->inp_laddr.s_addr == INADDR_ANY) {
1602 #ifdef INET6
1603 					if (INP_CHECK_SOCKAF(inp->inp_socket,
1604 							     AF_INET6))
1605 						local_wild_mapped = inp;
1606 					else
1607 #endif
1608 						if (cred != NULL &&
1609 						    jailed(cred))
1610 							jinp_wild = inp;
1611 						else
1612 							local_wild = inp;
1613 				}
1614 			}
1615 		}
1616 
1617 		REL_PCBINFO_TOKEN(pcbinfo);
1618 
1619 		if (local_wild != NULL)
1620 			return (local_wild);
1621 #ifdef INET6
1622 		if (local_wild_mapped != NULL)
1623 			return (local_wild_mapped);
1624 #endif
1625 		if (jinp != NULL)
1626 			return (jinp);
1627 		return (jinp_wild);
1628 	}
1629 
1630 	/*
1631 	 * Not found.
1632 	 */
1633 	return (NULL);
1634 }
1635 
1636 struct inpcb *
1637 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1638     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1639     boolean_t wildcard, struct ifnet *ifp)
1640 {
1641 	return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1642 	    laddr, lport_arg, wildcard, ifp, NULL);
1643 }
1644 
1645 /*
1646  * Insert PCB into connection hash table.
1647  */
1648 void
1649 in_pcbinsconnhash(struct inpcb *inp)
1650 {
1651 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1652 	struct inpcbhead *bucket;
1653 	u_int32_t hashkey_faddr, hashkey_laddr;
1654 
1655 #ifdef INET6
1656 	if (inp->inp_vflag & INP_IPV6) {
1657 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1658 		hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1659 	} else {
1660 #endif
1661 		hashkey_faddr = inp->inp_faddr.s_addr;
1662 		hashkey_laddr = inp->inp_laddr.s_addr;
1663 #ifdef INET6
1664 	}
1665 #endif
1666 
1667 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1668 	    ("not in the correct netisr"));
1669 	KASSERT(!(inp->inp_flags & INP_WILDCARD), ("already on wildcardhash"));
1670 	KASSERT(!(inp->inp_flags & INP_CONNECTED), ("already on connhash"));
1671 	inp->inp_flags |= INP_CONNECTED;
1672 
1673 	/*
1674 	 * Insert into the connection hash table.
1675 	 */
1676 	bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1677 	    inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1678 	LIST_INSERT_HEAD(bucket, inp, inp_hash);
1679 }
1680 
1681 /*
1682  * Remove PCB from connection hash table.
1683  */
1684 void
1685 in_pcbremconnhash(struct inpcb *inp)
1686 {
1687 	struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1688 
1689 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1690 	    ("not in the correct netisr"));
1691 	KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1692 
1693 	LIST_REMOVE(inp, inp_hash);
1694 	inp->inp_flags &= ~INP_CONNECTED;
1695 }
1696 
1697 /*
1698  * Insert PCB into port hash table.
1699  */
1700 void
1701 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp)
1702 {
1703 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1704 	struct inpcbporthead *pcbporthash;
1705 	struct inpcbport *phd;
1706 
1707 	/*
1708 	 * If the porthashbase is shared across several cpus, it must
1709 	 * have been locked.
1710 	 */
1711 	ASSERT_PORT_TOKEN_HELD(portinfo);
1712 
1713 	/*
1714 	 * Insert into the port hash table.
1715 	 */
1716 	pcbporthash = &portinfo->porthashbase[
1717 	    INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)];
1718 
1719 	/* Go through port list and look for a head for this lport. */
1720 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
1721 		if (phd->phd_port == inp->inp_lport)
1722 			break;
1723 	}
1724 
1725 	/* If none exists, use saved one and tack it on. */
1726 	if (phd == NULL) {
1727 		KKASSERT(pcbinfo->portsave != NULL);
1728 		phd = pcbinfo->portsave;
1729 		pcbinfo->portsave = NULL;
1730 		phd->phd_port = inp->inp_lport;
1731 		LIST_INIT(&phd->phd_pcblist);
1732 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1733 	}
1734 
1735 	inp->inp_portinfo = portinfo;
1736 	inp->inp_phd = phd;
1737 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1738 
1739 	/*
1740 	 * Malloc one inpcbport for later use.  It is safe to use
1741 	 * "wait" malloc here (port token would be released, if
1742 	 * malloc ever blocked), since all changes to the porthash
1743 	 * are done.
1744 	 */
1745 	if (pcbinfo->portsave == NULL) {
1746 		pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1747 					    M_PCB, M_INTWAIT | M_ZERO);
1748 	}
1749 }
1750 
1751 void
1752 in_pcbinsporthash_lport(struct inpcb *inp)
1753 {
1754 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1755 	struct inpcbportinfo *portinfo;
1756 	u_short lport_ho;
1757 
1758 	/* Locate the proper portinfo based on lport */
1759 	lport_ho = ntohs(inp->inp_lport);
1760 	portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
1761 	KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset);
1762 
1763 	GET_PORT_TOKEN(portinfo);
1764 	in_pcbinsporthash(portinfo, inp);
1765 	REL_PORT_TOKEN(portinfo);
1766 }
1767 
1768 static struct inp_localgroup *
1769 inp_localgroup_alloc(u_char vflag,
1770     uint16_t port, const union in_dependaddr *addr, int size)
1771 {
1772 	struct inp_localgroup *grp;
1773 
1774 	grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1775 	    M_TEMP, M_INTWAIT | M_ZERO);
1776 	grp->il_vflag = vflag;
1777 	grp->il_lport = port;
1778 	grp->il_dependladdr = *addr;
1779 	grp->il_inpsiz = size;
1780 
1781 	return grp;
1782 }
1783 
1784 static void
1785 inp_localgroup_free(struct inp_localgroup *grp)
1786 {
1787 	kfree(grp, M_TEMP);
1788 }
1789 
1790 static void
1791 inp_localgroup_destroy(struct inp_localgroup *grp)
1792 {
1793 	LIST_REMOVE(grp, il_list);
1794 	inp_localgroup_free(grp);
1795 }
1796 
1797 static void
1798 inp_localgroup_copy(struct inp_localgroup *grp,
1799     const struct inp_localgroup *old_grp)
1800 {
1801 	int i;
1802 
1803 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1804 	    ("invalid new local group size %d and old local group count %d",
1805 	     grp->il_inpsiz, old_grp->il_inpcnt));
1806 	for (i = 0; i < old_grp->il_inpcnt; ++i)
1807 		grp->il_inp[i] = old_grp->il_inp[i];
1808 	grp->il_inpcnt = old_grp->il_inpcnt;
1809 	grp->il_factor = old_grp->il_factor;
1810 }
1811 
1812 static void
1813 inp_localgroup_factor(struct inp_localgroup *grp)
1814 {
1815 	grp->il_factor =
1816 	    ((uint32_t)(0xffff >> ncpus2_shift) / grp->il_inpcnt) + 1;
1817 	KASSERT(grp->il_factor != 0, ("invalid local group factor, "
1818 	    "ncpus2_shift %d, inpcnt %d", ncpus2_shift, grp->il_inpcnt));
1819 }
1820 
1821 static void
1822 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1823 {
1824 	struct inp_localgrphead *hdr;
1825 	struct inp_localgroup *grp, *grp_alloc = NULL;
1826 	struct ucred *cred;
1827 
1828 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1829 
1830 	if (pcbinfo->localgrphashbase == NULL)
1831 		return;
1832 
1833 	/*
1834 	 * XXX don't allow jailed socket to join local group
1835 	 */
1836 	if (inp->inp_socket != NULL)
1837 		cred = inp->inp_socket->so_cred;
1838 	else
1839 		cred = NULL;
1840 	if (cred != NULL && jailed(cred))
1841 		return;
1842 
1843 #ifdef INET6
1844 	/*
1845 	 * XXX don't allow IPv4 mapped INET6 wild socket
1846 	 */
1847 	if ((inp->inp_vflag & INP_IPV4) &&
1848 	    inp->inp_laddr.s_addr == INADDR_ANY &&
1849 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6))
1850 		return;
1851 #endif
1852 
1853 	hdr = &pcbinfo->localgrphashbase[
1854 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1855 
1856 again:
1857 	LIST_FOREACH(grp, hdr, il_list) {
1858 		if (grp->il_vflag == inp->inp_vflag &&
1859 		    grp->il_lport == inp->inp_lport &&
1860 		    memcmp(&grp->il_dependladdr,
1861 		        &inp->inp_inc.inc_ie.ie_dependladdr,
1862 		        sizeof(grp->il_dependladdr)) == 0) {
1863 			break;
1864 		}
1865 	}
1866 	if (grp == NULL) {
1867 		/*
1868 		 * Create a new local group
1869 		 */
1870 		if (grp_alloc == NULL) {
1871 			grp_alloc = inp_localgroup_alloc(inp->inp_vflag,
1872 			    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1873 			    INP_LOCALGROUP_SIZMIN);
1874 			/*
1875 			 * Local group allocation could block and the
1876 			 * local group w/ the same property might have
1877 			 * been added by others when we were blocked;
1878 			 * check again.
1879 			 */
1880 			goto again;
1881 		} else {
1882 			/* Local group has been allocated; link it */
1883 			grp = grp_alloc;
1884 			grp_alloc = NULL;
1885 			LIST_INSERT_HEAD(hdr, grp, il_list);
1886 		}
1887 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
1888 		if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1889 			static int limit_logged = 0;
1890 
1891 			if (!limit_logged) {
1892 				limit_logged = 1;
1893 				kprintf("local group port %d, "
1894 				    "limit reached\n", ntohs(grp->il_lport));
1895 			}
1896 			if (grp_alloc != NULL) {
1897 				/*
1898 				 * This would happen if the local group
1899 				 * w/ the same property was expanded when
1900 				 * our local group allocation blocked.
1901 				 */
1902 				inp_localgroup_free(grp_alloc);
1903 			}
1904 			return;
1905 		}
1906 
1907 		/*
1908 		 * Expand this local group
1909 		 */
1910 		if (grp_alloc == NULL ||
1911 		    grp->il_inpcnt >= grp_alloc->il_inpsiz) {
1912 			if (grp_alloc != NULL)
1913 				inp_localgroup_free(grp_alloc);
1914 			grp_alloc = inp_localgroup_alloc(grp->il_vflag,
1915 			    grp->il_lport, &grp->il_dependladdr,
1916 			    grp->il_inpsiz * 2);
1917 			/*
1918 			 * Local group allocation could block and the
1919 			 * local group w/ the same property might have
1920 			 * been expanded by others when we were blocked;
1921 			 * check again.
1922 			 */
1923 			goto again;
1924 		}
1925 
1926 		/*
1927 		 * Save the old local group, link the new one, and then
1928 		 * destroy the old local group
1929 		 */
1930 		inp_localgroup_copy(grp_alloc, grp);
1931 		LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
1932 		inp_localgroup_destroy(grp);
1933 
1934 		grp = grp_alloc;
1935 		grp_alloc = NULL;
1936 	} else {
1937 		/*
1938 		 * Found the local group
1939 		 */
1940 		if (grp_alloc != NULL) {
1941 			/*
1942 			 * This would happen if the local group w/ the
1943 			 * same property was added or expanded when our
1944 			 * local group allocation blocked.
1945 			 */
1946 			inp_localgroup_free(grp_alloc);
1947 			grp_alloc = NULL;
1948 		}
1949 	}
1950 
1951 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
1952 	    ("invalid local group size %d and count %d",
1953 	     grp->il_inpsiz, grp->il_inpcnt));
1954 	grp->il_inp[grp->il_inpcnt] = inp;
1955 	grp->il_inpcnt++;
1956 	inp_localgroup_factor(grp);
1957 }
1958 
1959 void
1960 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1961 {
1962 	struct inpcontainer *ic;
1963 	struct inpcontainerhead *bucket;
1964 
1965 	GET_PCBINFO_TOKEN(pcbinfo);
1966 
1967 	in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
1968 
1969 	bucket = &pcbinfo->wildcardhashbase[
1970 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1971 
1972 	ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
1973 	ic->ic_inp = inp;
1974 	LIST_INSERT_HEAD(bucket, ic, ic_list);
1975 
1976 	REL_PCBINFO_TOKEN(pcbinfo);
1977 }
1978 
1979 /*
1980  * Insert PCB into wildcard hash table.
1981  */
1982 void
1983 in_pcbinswildcardhash(struct inpcb *inp)
1984 {
1985 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1986 
1987 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1988 	    ("not in correct netisr"));
1989 	KASSERT(!(inp->inp_flags & INP_CONNECTED),
1990 		("already on connhash"));
1991 	KASSERT(!(inp->inp_flags & INP_WILDCARD),
1992 		("already on wildcardhash"));
1993 	inp->inp_flags |= INP_WILDCARD;
1994 
1995 	in_pcbinswildcardhash_oncpu(inp, pcbinfo);
1996 }
1997 
1998 static void
1999 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2000 {
2001 	struct inp_localgrphead *hdr;
2002 	struct inp_localgroup *grp;
2003 
2004 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
2005 
2006 	if (pcbinfo->localgrphashbase == NULL)
2007 		return;
2008 
2009 	hdr = &pcbinfo->localgrphashbase[
2010 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
2011 
2012 	LIST_FOREACH(grp, hdr, il_list) {
2013 		int i;
2014 
2015 		for (i = 0; i < grp->il_inpcnt; ++i) {
2016 			if (grp->il_inp[i] != inp)
2017 				continue;
2018 
2019 			if (grp->il_inpcnt == 1) {
2020 				/* Destroy this local group */
2021 				inp_localgroup_destroy(grp);
2022 			} else {
2023 				/* Pull up inpcbs */
2024 				for (; i + 1 < grp->il_inpcnt; ++i)
2025 					grp->il_inp[i] = grp->il_inp[i + 1];
2026 				grp->il_inpcnt--;
2027 				inp_localgroup_factor(grp);
2028 			}
2029 			return;
2030 		}
2031 	}
2032 }
2033 
2034 void
2035 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2036 {
2037 	struct inpcontainer *ic;
2038 	struct inpcontainerhead *head;
2039 
2040 	GET_PCBINFO_TOKEN(pcbinfo);
2041 
2042 	in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2043 
2044 	/* find bucket */
2045 	head = &pcbinfo->wildcardhashbase[
2046 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2047 
2048 	LIST_FOREACH(ic, head, ic_list) {
2049 		if (ic->ic_inp == inp)
2050 			goto found;
2051 	}
2052 	REL_PCBINFO_TOKEN(pcbinfo);
2053 	return;			/* not found! */
2054 
2055 found:
2056 	LIST_REMOVE(ic, ic_list);	/* remove container from bucket chain */
2057 	REL_PCBINFO_TOKEN(pcbinfo);
2058 	kfree(ic, M_TEMP);		/* deallocate container */
2059 }
2060 
2061 /*
2062  * Remove PCB from wildcard hash table.
2063  */
2064 void
2065 in_pcbremwildcardhash(struct inpcb *inp)
2066 {
2067 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2068 
2069 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2070 	    ("not in correct netisr"));
2071 	KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2072 
2073 	in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2074 	inp->inp_flags &= ~INP_WILDCARD;
2075 }
2076 
2077 /*
2078  * Remove PCB from various lists.
2079  */
2080 void
2081 in_pcbremlists(struct inpcb *inp)
2082 {
2083 	if (inp->inp_lport) {
2084 		struct inpcbportinfo *portinfo;
2085 		struct inpcbport *phd;
2086 
2087 		/*
2088 		 * NOTE:
2089 		 * inp->inp_portinfo is _not_ necessary same as
2090 		 * inp->inp_pcbinfo->portinfo.
2091 		 */
2092 		portinfo = inp->inp_portinfo;
2093 		GET_PORT_TOKEN(portinfo);
2094 
2095 		phd = inp->inp_phd;
2096 		LIST_REMOVE(inp, inp_portlist);
2097 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2098 			LIST_REMOVE(phd, phd_hash);
2099 			kfree(phd, M_PCB);
2100 		}
2101 
2102 		REL_PORT_TOKEN(portinfo);
2103 	}
2104 	if (inp->inp_flags & INP_WILDCARD) {
2105 		in_pcbremwildcardhash(inp);
2106 	} else if (inp->inp_flags & INP_CONNECTED) {
2107 		in_pcbremconnhash(inp);
2108 	}
2109 
2110 	if (inp->inp_flags & INP_ONLIST)
2111 		in_pcbofflist(inp);
2112 }
2113 
2114 int
2115 prison_xinpcb(struct thread *td, struct inpcb *inp)
2116 {
2117 	struct ucred *cr;
2118 
2119 	if (td->td_proc == NULL)
2120 		return (0);
2121 	cr = td->td_proc->p_ucred;
2122 	if (cr->cr_prison == NULL)
2123 		return (0);
2124 	if (inp->inp_socket && inp->inp_socket->so_cred &&
2125 	    inp->inp_socket->so_cred->cr_prison &&
2126 	    cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2127 		return (0);
2128 	return (1);
2129 }
2130 
2131 int
2132 in_pcblist_global(SYSCTL_HANDLER_ARGS)
2133 {
2134 	struct inpcbinfo *pcbinfo_arr = arg1;
2135 	int pcbinfo_arrlen = arg2;
2136 	struct inpcb *marker;
2137 	int cpu, origcpu;
2138 	int error, n;
2139 
2140 	KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1,
2141 	    ("invalid pcbinfo count %d", pcbinfo_arrlen));
2142 
2143 	/*
2144 	 * The process of preparing the TCB list is too time-consuming and
2145 	 * resource-intensive to repeat twice on every request.
2146 	 */
2147 	n = 0;
2148 	if (req->oldptr == NULL) {
2149 		for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2150 			n += pcbinfo_arr[cpu].ipi_count;
2151 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2152 		return 0;
2153 	}
2154 
2155 	if (req->newptr != NULL)
2156 		return EPERM;
2157 
2158 	marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2159 	marker->inp_flags |= INP_PLACEMARKER;
2160 
2161 	/*
2162 	 * OK, now we're committed to doing something.  Re-fetch ipi_count
2163 	 * after obtaining the generation count.
2164 	 */
2165 	error = 0;
2166 	origcpu = mycpuid;
2167 	for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2168 		struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2169 		struct inpcb *inp;
2170 		struct xinpcb xi;
2171 		int i;
2172 
2173 		lwkt_migratecpu(cpu);
2174 
2175 		GET_PCBINFO_TOKEN(pcbinfo);
2176 
2177 		n = pcbinfo->ipi_count;
2178 
2179 		LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2180 		i = 0;
2181 		while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2182 			LIST_REMOVE(marker, inp_list);
2183 			LIST_INSERT_AFTER(inp, marker, inp_list);
2184 
2185 			if (inp->inp_flags & INP_PLACEMARKER)
2186 				continue;
2187 			if (prison_xinpcb(req->td, inp))
2188 				continue;
2189 
2190 			bzero(&xi, sizeof xi);
2191 			xi.xi_len = sizeof xi;
2192 			bcopy(inp, &xi.xi_inp, sizeof *inp);
2193 			if (inp->inp_socket)
2194 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
2195 			if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2196 				break;
2197 			++i;
2198 		}
2199 		LIST_REMOVE(marker, inp_list);
2200 
2201 		REL_PCBINFO_TOKEN(pcbinfo);
2202 
2203 		if (error == 0 && i < n) {
2204 			bzero(&xi, sizeof xi);
2205 			xi.xi_len = sizeof xi;
2206 			while (i < n) {
2207 				error = SYSCTL_OUT(req, &xi, sizeof xi);
2208 				if (error)
2209 					break;
2210 				++i;
2211 			}
2212 		}
2213 	}
2214 
2215 	lwkt_migratecpu(origcpu);
2216 	kfree(marker, M_TEMP);
2217 	return error;
2218 }
2219 
2220 int
2221 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS)
2222 {
2223 	return in_pcblist_global(oidp, arg1, ncpus2, req);
2224 }
2225 
2226 void
2227 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2228 {
2229 	struct sockaddr_in *sin;
2230 
2231 	KASSERT(faddr->sa_family == AF_INET,
2232 	    ("not AF_INET faddr %d", faddr->sa_family));
2233 
2234 	sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2235 	sin->sin_family = AF_INET;
2236 	sin->sin_len = sizeof(*sin);
2237 	sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2238 	sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2239 
2240 	so->so_faddr = (struct sockaddr *)sin;
2241 }
2242 
2243 void
2244 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2245     boolean_t shared, u_short offset)
2246 {
2247 	memset(portinfo, 0, sizeof(*portinfo));
2248 
2249 	portinfo->offset = offset;
2250 	portinfo->lastport = offset;
2251 	portinfo->lastlow = offset;
2252 	portinfo->lasthi = offset;
2253 
2254 	portinfo->porthashbase = hashinit(hashsize, M_PCB,
2255 	    &portinfo->porthashmask);
2256 
2257 	if (shared) {
2258 		portinfo->porttoken = kmalloc(sizeof(struct lwkt_token),
2259 		    M_PCB, M_WAITOK);
2260 		lwkt_token_init(portinfo->porttoken, "porttoken");
2261 	}
2262 }
2263 
2264 void
2265 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2266 {
2267 	int hi, lo;
2268 
2269 	if (step == 1)
2270 		return;
2271 
2272 	hi = *hi0;
2273 	lo = *lo0;
2274 
2275 	hi = rounddown2(hi, step);
2276 	hi += ofs;
2277 	if (hi > (int)*hi0)
2278 		hi -= step;
2279 
2280 	lo = roundup2(lo, step);
2281 	lo -= (step - ofs);
2282 	if (lo < (int)*lo0)
2283 		lo += step;
2284 
2285 	*hi0 = hi;
2286 	*lo0 = lo;
2287 }
2288 
2289 void
2290 in_pcbglobalinit(void)
2291 {
2292 	int cpu;
2293 
2294 	in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB,
2295 	    M_WAITOK | M_ZERO);
2296 	in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer),
2297 	    M_PCB, M_WAITOK | M_ZERO);
2298 
2299 	for (cpu = 0; cpu < ncpus; ++cpu) {
2300 		struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2301 		struct inpcb *marker = &in_pcbmarkers[cpu];
2302 
2303 		marker->inp_flags |= INP_PLACEMARKER;
2304 		ic->ic_inp = marker;
2305 	}
2306 }
2307 
2308 struct inpcb *
2309 in_pcbmarker(int cpuid)
2310 {
2311 	KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2312 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2313 
2314 	return &in_pcbmarkers[cpuid];
2315 }
2316 
2317 struct inpcontainer *
2318 in_pcbcontainer_marker(int cpuid)
2319 {
2320 	KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2321 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2322 
2323 	return &in_pcbcontainer_markers[cpuid];
2324 }
2325