xref: /dflybsd-src/sys/netinet/in_pcb.c (revision 3484c4aa8f394b9ac6147d925d119ffb4cd9edb6)
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_lgrpindex = -1;
289 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
290 	inp->inp_pcbinfo = pcbinfo;
291 	inp->inp_socket = so;
292 #ifdef IPSEC
293 	error = ipsec_init_policy(so, &inp->inp_sp);
294 	if (error != 0) {
295 		kfree(inp, M_PCB);
296 		return (error);
297 	}
298 #endif
299 #ifdef INET6
300 	if (INP_CHECK_SOCKAF(so, AF_INET6)) {
301 		if (ip6_auto_flowlabel)
302 			inp->inp_flags |= IN6P_AUTOFLOWLABEL;
303 		inp->inp_af = AF_INET6;
304 	} else
305 #endif
306 	inp->inp_af = AF_INET;
307 	soreference(so);
308 	so->so_pcb = inp;
309 
310 	in_pcbonlist(inp);
311 	return (0);
312 }
313 
314 /*
315  * Unlink a pcb with the intention of moving it to another cpu with a
316  * different pcbinfo.  While unlinked nothing should attempt to dereference
317  * inp_pcbinfo, NULL it out so we assert if it does.
318  */
319 void
320 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
321 {
322 	KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch"));
323 	KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
324 	    ("already linked"));
325 
326 	in_pcbofflist(inp);
327 	inp->inp_pcbinfo = NULL;
328 }
329 
330 void
331 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
332 {
333 	in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD);
334 }
335 
336 /*
337  * Relink a pcb into a new pcbinfo.
338  */
339 void
340 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags)
341 {
342 	KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo"));
343 	KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0,
344 	    ("already linked"));
345 
346 	inp->inp_pcbinfo = pcbinfo;
347 	in_pcbonlist(inp);
348 }
349 
350 void
351 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo)
352 {
353 	return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD);
354 }
355 
356 static int
357 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred)
358 {
359 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
360 	struct inpcbportinfo *portinfo;
361 	u_short first, last, lport, step;
362 	u_short *lastport;
363 	int count, error;
364 	int portinfo_first, portinfo_idx;
365 
366 	inp->inp_flags |= INP_ANONPORT;
367 
368 	step = pcbinfo->portinfo_mask + 1;
369 	portinfo_first = mycpuid & pcbinfo->portinfo_mask;
370 	portinfo_idx = portinfo_first;
371 loop:
372 	portinfo = &pcbinfo->portinfo[portinfo_idx];
373 
374 	if (inp->inp_flags & INP_HIGHPORT) {
375 		first = ipport_hifirstauto;	/* sysctl */
376 		last  = ipport_hilastauto;
377 		lastport = &portinfo->lasthi;
378 	} else if (inp->inp_flags & INP_LOWPORT) {
379 		if (cred &&
380 		    (error =
381 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
382 			inp->inp_laddr.s_addr = INADDR_ANY;
383 			return error;
384 		}
385 		first = ipport_lowfirstauto;	/* 1023 */
386 		last  = ipport_lowlastauto;	/* 600 */
387 		lastport = &portinfo->lastlow;
388 	} else {
389 		first = ipport_firstauto;	/* sysctl */
390 		last  = ipport_lastauto;
391 		lastport = &portinfo->lastport;
392 	}
393 
394 	/*
395 	 * This has to be atomic.  If the porthash is shared across multiple
396 	 * protocol threads (aka tcp) then the token must be held.
397 	 */
398 	GET_PORT_TOKEN(portinfo);
399 
400 	/*
401 	 * Simple check to ensure all ports are not used up causing
402 	 * a deadlock here.
403 	 *
404 	 * We split the two cases (up and down) so that the direction
405 	 * is not being tested on each round of the loop.
406 	 */
407 	if (first > last) {
408 		/*
409 		 * counting down
410 		 */
411 		in_pcbportrange(&first, &last, portinfo->offset, step);
412 		count = (first - last) / step;
413 
414 		do {
415 			if (count-- < 0) {	/* completely used? */
416 				error = EADDRNOTAVAIL;
417 				goto done;
418 			}
419 			*lastport -= step;
420 			if (*lastport > first || *lastport < last)
421 				*lastport = first;
422 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
423 			    portinfo->offset);
424 			lport = htons(*lastport);
425 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
426 		    wild, cred));
427 	} else {
428 		/*
429 		 * counting up
430 		 */
431 		in_pcbportrange(&last, &first, portinfo->offset, step);
432 		count = (last - first) / step;
433 
434 		do {
435 			if (count-- < 0) {	/* completely used? */
436 				error = EADDRNOTAVAIL;
437 				goto done;
438 			}
439 			*lastport += step;
440 			if (*lastport < first || *lastport > last)
441 				*lastport = first;
442 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
443 			    portinfo->offset);
444 			lport = htons(*lastport);
445 		} while (in_pcblookup_local(portinfo, inp->inp_laddr, lport,
446 		    wild, cred));
447 	}
448 	inp->inp_lport = lport;
449 	in_pcbinsporthash(portinfo, inp);
450 	error = 0;
451 done:
452 	REL_PORT_TOKEN(portinfo);
453 
454 	if (error) {
455 		/* Try next portinfo */
456 		portinfo_idx++;
457 		portinfo_idx &= pcbinfo->portinfo_mask;
458 		if (portinfo_idx != portinfo_first)
459 			goto loop;
460 		inp->inp_laddr.s_addr = INADDR_ANY;
461 	}
462 	return error;
463 }
464 
465 int
466 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
467 {
468 	struct socket *so = inp->inp_socket;
469 	struct sockaddr_in jsin;
470 	struct ucred *cred = NULL;
471 	int wild = 0;
472 
473 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
474 		return (EADDRNOTAVAIL);
475 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
476 		return (EINVAL);	/* already bound */
477 
478 	if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT)))
479 		wild = 1;    /* neither SO_REUSEADDR nor SO_REUSEPORT is set */
480 	if (td->td_proc)
481 		cred = td->td_proc->p_ucred;
482 
483 	if (nam != NULL) {
484 		struct sockaddr_in *sin = (struct sockaddr_in *)nam;
485 		struct inpcbinfo *pcbinfo;
486 		struct inpcbportinfo *portinfo;
487 		struct inpcb *t;
488 		u_short lport, lport_ho;
489 		int reuseport = (so->so_options & SO_REUSEPORT);
490 		int error;
491 
492 		if (nam->sa_len != sizeof *sin)
493 			return (EINVAL);
494 #ifdef notdef
495 		/*
496 		 * We should check the family, but old programs
497 		 * incorrectly fail to initialize it.
498 		 */
499 		if (sin->sin_family != AF_INET)
500 			return (EAFNOSUPPORT);
501 #endif
502 		if (!prison_replace_wildcards(td, nam))
503 			return (EINVAL);
504 
505 		lport = sin->sin_port;
506 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
507 			/*
508 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
509 			 * allow complete duplication of binding if
510 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
511 			 * and a multicast address is bound on both
512 			 * new and duplicated sockets.
513 			 */
514 			if (so->so_options & SO_REUSEADDR)
515 				reuseport = SO_REUSEADDR | SO_REUSEPORT;
516 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
517 			sin->sin_port = 0;		/* yech... */
518 			bzero(&sin->sin_zero, sizeof sin->sin_zero);
519 			if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL)
520 				return (EADDRNOTAVAIL);
521 		}
522 
523 		inp->inp_laddr = sin->sin_addr;
524 
525 		jsin.sin_family = AF_INET;
526 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
527 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
528 			inp->inp_laddr.s_addr = INADDR_ANY;
529 			return (EINVAL);
530 		}
531 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
532 
533 		if (lport == 0) {
534 			/* Auto-select local port */
535 			return in_pcbsetlport(inp, wild, cred);
536 		}
537 		lport_ho = ntohs(lport);
538 
539 		/* GROSS */
540 		if (lport_ho < IPPORT_RESERVED && cred &&
541 		    (error =
542 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
543 			inp->inp_laddr.s_addr = INADDR_ANY;
544 			return (error);
545 		}
546 
547 		/*
548 		 * Locate the proper portinfo based on lport
549 		 */
550 		pcbinfo = inp->inp_pcbinfo;
551 		portinfo =
552 		    &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
553 		KKASSERT((lport_ho & pcbinfo->portinfo_mask) ==
554 		    portinfo->offset);
555 
556 		/*
557 		 * This has to be atomic.  If the porthash is shared across
558 		 * multiple protocol threads (aka tcp) then the token must
559 		 * be held.
560 		 */
561 		GET_PORT_TOKEN(portinfo);
562 
563 		if (so->so_cred->cr_uid != 0 &&
564 		    !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
565 			t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
566 			    INPLOOKUP_WILDCARD, cred);
567 			if (t &&
568 			    (so->so_cred->cr_uid !=
569 			     t->inp_socket->so_cred->cr_uid)) {
570 				inp->inp_laddr.s_addr = INADDR_ANY;
571 				error = EADDRINUSE;
572 				goto done;
573 			}
574 		}
575 		if (cred && !prison_replace_wildcards(td, nam)) {
576 			inp->inp_laddr.s_addr = INADDR_ANY;
577 			error = EADDRNOTAVAIL;
578 			goto done;
579 		}
580 		t = in_pcblookup_local(portinfo, sin->sin_addr, lport,
581 		    wild, cred);
582 		if (t && !(reuseport & t->inp_socket->so_options)) {
583 			inp->inp_laddr.s_addr = INADDR_ANY;
584 			error = EADDRINUSE;
585 			goto done;
586 		}
587 		inp->inp_lport = lport;
588 		in_pcbinsporthash(portinfo, inp);
589 		error = 0;
590 done:
591 		REL_PORT_TOKEN(portinfo);
592 		return (error);
593 	} else {
594 		jsin.sin_family = AF_INET;
595 		jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
596 		if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
597 			inp->inp_laddr.s_addr = INADDR_ANY;
598 			return (EINVAL);
599 		}
600 		inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
601 
602 		return in_pcbsetlport(inp, wild, cred);
603 	}
604 }
605 
606 static struct inpcb *
607 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr,
608     u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred)
609 {
610 	struct inpcb *inp;
611 	struct inpcbporthead *porthash;
612 	struct inpcbport *phd;
613 	struct inpcb *match = NULL;
614 
615 	/*
616 	 * If the porthashbase is shared across several cpus, it must
617 	 * have been locked.
618 	 */
619 	ASSERT_PORT_TOKEN_HELD(portinfo);
620 
621 	/*
622 	 * Best fit PCB lookup.
623 	 *
624 	 * First see if this local port is in use by looking on the
625 	 * port hash list.
626 	 */
627 	porthash = &portinfo->porthashbase[
628 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
629 	LIST_FOREACH(phd, porthash, phd_hash) {
630 		if (phd->phd_port == lport)
631 			break;
632 	}
633 	if (phd != NULL) {
634 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
635 #ifdef INET6
636 			if (!INP_ISIPV4(inp))
637 				continue;
638 #endif
639 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
640 			    inp->inp_laddr.s_addr != laddr.s_addr)
641 				continue;
642 
643 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
644 			    inp->inp_faddr.s_addr != faddr.s_addr)
645 				continue;
646 
647 			if (inp->inp_fport != 0 && inp->inp_fport != fport)
648 				continue;
649 
650 			if (cred == NULL ||
651 			    cred->cr_prison ==
652 			    inp->inp_socket->so_cred->cr_prison) {
653 				match = inp;
654 				break;
655 			}
656 		}
657 	}
658 	return (match);
659 }
660 
661 int
662 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote,
663     struct thread *td)
664 {
665 	struct proc *p = td->td_proc;
666 	unsigned short *lastport;
667 	const struct sockaddr_in *sin = (const struct sockaddr_in *)remote;
668 	struct sockaddr_in jsin;
669 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
670 	struct inpcbportinfo *portinfo;
671 	struct ucred *cred = NULL;
672 	u_short first, last, lport, step;
673 	int count, error, dup;
674 	int portinfo_first, portinfo_idx;
675 
676 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */
677 		return (EADDRNOTAVAIL);
678 
679 	KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY);
680 	if (inp->inp_lport != 0)
681 		return (EINVAL);	/* already bound */
682 
683 	KKASSERT(p);
684 	cred = p->p_ucred;
685 
686 	jsin.sin_family = AF_INET;
687 	jsin.sin_addr.s_addr = inp->inp_laddr.s_addr;
688 	if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) {
689 		inp->inp_laddr.s_addr = INADDR_ANY;
690 		return (EINVAL);
691 	}
692 	inp->inp_laddr.s_addr = jsin.sin_addr.s_addr;
693 
694 	inp->inp_flags |= INP_ANONPORT;
695 
696 	step = pcbinfo->portinfo_mask + 1;
697 	portinfo_first = mycpuid & pcbinfo->portinfo_mask;
698 	portinfo_idx = portinfo_first;
699 loop:
700 	portinfo = &pcbinfo->portinfo[portinfo_idx];
701 	dup = 0;
702 
703 	if (inp->inp_flags & INP_HIGHPORT) {
704 		first = ipport_hifirstauto;	/* sysctl */
705 		last  = ipport_hilastauto;
706 		lastport = &portinfo->lasthi;
707 	} else if (inp->inp_flags & INP_LOWPORT) {
708 		if (cred &&
709 		    (error =
710 		     priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) {
711 			inp->inp_laddr.s_addr = INADDR_ANY;
712 			return (error);
713 		}
714 		first = ipport_lowfirstauto;	/* 1023 */
715 		last  = ipport_lowlastauto;	/* 600 */
716 		lastport = &portinfo->lastlow;
717 	} else {
718 		first = ipport_firstauto;	/* sysctl */
719 		last  = ipport_lastauto;
720 		lastport = &portinfo->lastport;
721 	}
722 
723 	/*
724 	 * This has to be atomic.  If the porthash is shared across multiple
725 	 * protocol threads (aka tcp) then the token must be held.
726 	 */
727 	GET_PORT_TOKEN(portinfo);
728 
729 again:
730 	/*
731 	 * Simple check to ensure all ports are not used up causing
732 	 * a deadlock here.
733 	 *
734 	 * We split the two cases (up and down) so that the direction
735 	 * is not being tested on each round of the loop.
736 	 */
737 	if (first > last) {
738 		/*
739 		 * counting down
740 		 */
741 		in_pcbportrange(&first, &last, portinfo->offset, step);
742 		count = (first - last) / step;
743 
744 		do {
745 			if (count-- < 0) {	/* completely used? */
746 				error = EADDRNOTAVAIL;
747 				goto done;
748 			}
749 			*lastport -= step;
750 			if (*lastport > first || *lastport < last)
751 				*lastport = first;
752 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
753 			    portinfo->offset);
754 			lport = htons(*lastport);
755 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
756 		    lport, sin->sin_addr, sin->sin_port, cred));
757 	} else {
758 		/*
759 		 * counting up
760 		 */
761 		in_pcbportrange(&last, &first, portinfo->offset, step);
762 		count = (last - first) / step;
763 
764 		do {
765 			if (count-- < 0) {	/* completely used? */
766 				error = EADDRNOTAVAIL;
767 				goto done;
768 			}
769 			*lastport += step;
770 			if (*lastport < first || *lastport > last)
771 				*lastport = first;
772 			KKASSERT((*lastport & pcbinfo->portinfo_mask) ==
773 			    portinfo->offset);
774 			lport = htons(*lastport);
775 		} while (in_pcblookup_localremote(portinfo, inp->inp_laddr,
776 		    lport, sin->sin_addr, sin->sin_port, cred));
777 	}
778 
779 	/* This could happen on loopback interface */
780 	if (sin->sin_port == lport &&
781 	    sin->sin_addr.s_addr == inp->inp_laddr.s_addr) {
782 		if (dup) {
783 			/*
784 			 * Duplicate again; give up
785 			 */
786 			error = EADDRNOTAVAIL;
787 			goto done;
788 		}
789 		dup = 1;
790 		goto again;
791 	}
792 	inp->inp_lport = lport;
793 	in_pcbinsporthash(portinfo, inp);
794 	error = 0;
795 done:
796 	REL_PORT_TOKEN(portinfo);
797 
798 	if (error) {
799 		/* Try next portinfo */
800 		portinfo_idx++;
801 		portinfo_idx &= pcbinfo->portinfo_mask;
802 		if (portinfo_idx != portinfo_first)
803 			goto loop;
804 		inp->inp_laddr.s_addr = INADDR_ANY;
805 	}
806 	return error;
807 }
808 
809 /*
810  *   Transform old in_pcbconnect() into an inner subroutine for new
811  *   in_pcbconnect(): Do some validity-checking on the remote
812  *   address (in mbuf 'nam') and then determine local host address
813  *   (i.e., which interface) to use to access that remote host.
814  *
815  *   This preserves definition of in_pcbconnect(), while supporting a
816  *   slightly different version for T/TCP.  (This is more than
817  *   a bit of a kludge, but cleaning up the internal interfaces would
818  *   have forced minor changes in every protocol).
819  */
820 int
821 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam,
822     struct sockaddr_in **plocal_sin, struct thread *td, int find)
823 {
824 	struct in_ifaddr *ia;
825 	struct ucred *cred = NULL;
826 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
827 	struct sockaddr *jsin;
828 	int jailed = 0, alloc_route = 0;
829 
830 	if (nam->sa_len != sizeof *sin)
831 		return (EINVAL);
832 	if (sin->sin_family != AF_INET)
833 		return (EAFNOSUPPORT);
834 	if (sin->sin_port == 0)
835 		return (EADDRNOTAVAIL);
836 	if (td && td->td_proc && td->td_proc->p_ucred)
837 		cred = td->td_proc->p_ucred;
838 	if (cred && cred->cr_prison)
839 		jailed = 1;
840 	if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) {
841 		ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
842 		/*
843 		 * If the destination address is INADDR_ANY,
844 		 * use the primary local address.
845 		 * If the supplied address is INADDR_BROADCAST,
846 		 * and the primary interface supports broadcast,
847 		 * choose the broadcast address for that interface.
848 		 */
849 		if (sin->sin_addr.s_addr == INADDR_ANY)
850 			sin->sin_addr = IA_SIN(ia)->sin_addr;
851 		else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST &&
852 		    (ia->ia_ifp->if_flags & IFF_BROADCAST))
853 			sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr;
854 	}
855 	if (find) {
856 		struct route *ro;
857 
858 		ia = NULL;
859 		/*
860 		 * If route is known or can be allocated now,
861 		 * our src addr is taken from the i/f, else punt.
862 		 * Note that we should check the address family of the cached
863 		 * destination, in case of sharing the cache with IPv6.
864 		 */
865 		ro = &inp->inp_route;
866 		if (ro->ro_rt &&
867 		    (!(ro->ro_rt->rt_flags & RTF_UP) ||
868 		     ro->ro_dst.sa_family != AF_INET ||
869 		     satosin(&ro->ro_dst)->sin_addr.s_addr !=
870 				      sin->sin_addr.s_addr ||
871 		     inp->inp_socket->so_options & SO_DONTROUTE)) {
872 			RTFREE(ro->ro_rt);
873 			ro->ro_rt = NULL;
874 		}
875 		if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
876 		    (ro->ro_rt == NULL ||
877 		    ro->ro_rt->rt_ifp == NULL)) {
878 			/* No route yet, so try to acquire one */
879 			bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
880 			ro->ro_dst.sa_family = AF_INET;
881 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
882 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
883 				sin->sin_addr;
884 			rtalloc(ro);
885 			alloc_route = 1;
886 		}
887 		/*
888 		 * If we found a route, use the address
889 		 * corresponding to the outgoing interface
890 		 * unless it is the loopback (in case a route
891 		 * to our address on another net goes to loopback).
892 		 */
893 		if (ro->ro_rt &&
894 		    !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) {
895 			if (jailed) {
896 				if (jailed_ip(cred->cr_prison,
897 				    ro->ro_rt->rt_ifa->ifa_addr)) {
898 					ia = ifatoia(ro->ro_rt->rt_ifa);
899 				}
900 			} else {
901 				ia = ifatoia(ro->ro_rt->rt_ifa);
902 			}
903 		}
904 		if (ia == NULL) {
905 			u_short fport = sin->sin_port;
906 
907 			sin->sin_port = 0;
908 			ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
909 			if (ia && jailed && !jailed_ip(cred->cr_prison,
910 			    sintosa(&ia->ia_addr)))
911 				ia = NULL;
912 			if (ia == NULL)
913 				ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
914 			if (ia && jailed && !jailed_ip(cred->cr_prison,
915 			    sintosa(&ia->ia_addr)))
916 				ia = NULL;
917 			sin->sin_port = fport;
918 			if (ia == NULL &&
919 			    !TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
920 				ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia;
921 			if (ia && jailed && !jailed_ip(cred->cr_prison,
922 			    sintosa(&ia->ia_addr)))
923 				ia = NULL;
924 
925 			if (!jailed && ia == NULL)
926 				goto fail;
927 		}
928 		/*
929 		 * If the destination address is multicast and an outgoing
930 		 * interface has been set as a multicast option, use the
931 		 * address of that interface as our source address.
932 		 */
933 		if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
934 		    inp->inp_moptions != NULL) {
935 			struct ip_moptions *imo;
936 			struct ifnet *ifp;
937 
938 			imo = inp->inp_moptions;
939 			if (imo->imo_multicast_ifp != NULL) {
940 				struct in_ifaddr_container *iac;
941 
942 				ifp = imo->imo_multicast_ifp;
943 				ia = NULL;
944 				TAILQ_FOREACH(iac,
945 				&in_ifaddrheads[mycpuid], ia_link) {
946 					if (iac->ia->ia_ifp == ifp) {
947 						ia = iac->ia;
948 						break;
949 					}
950 				}
951 				if (ia == NULL)
952 					goto fail;
953 			}
954 		}
955 		/*
956 		 * Don't do pcblookup call here; return interface in plocal_sin
957 		 * and exit to caller, that will do the lookup.
958 		 */
959 		if (ia == NULL && jailed) {
960 			if ((jsin = prison_get_nonlocal(
961 				cred->cr_prison, AF_INET, NULL)) != NULL ||
962 			    (jsin = prison_get_local(
963 				cred->cr_prison, AF_INET, NULL)) != NULL) {
964 				*plocal_sin = satosin(jsin);
965 			} else {
966 				/* IPv6 only Jail */
967 				goto fail;
968 			}
969 		} else {
970 			*plocal_sin = &ia->ia_addr;
971 		}
972 	}
973 	return (0);
974 fail:
975 	if (alloc_route)
976 		in_pcbresetroute(inp);
977 	return (EADDRNOTAVAIL);
978 }
979 
980 int
981 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam,
982     struct sockaddr_in **plocal_sin, struct thread *td)
983 {
984 	return in_pcbladdr_find(inp, nam, plocal_sin, td,
985 	    (inp->inp_laddr.s_addr == INADDR_ANY));
986 }
987 
988 /*
989  * Outer subroutine:
990  * Connect from a socket to a specified address.
991  * Both address and port must be specified in argument sin.
992  * If don't have a local address for this socket yet,
993  * then pick one.
994  */
995 int
996 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td)
997 {
998 	struct sockaddr_in *if_sin;
999 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1000 	int error;
1001 
1002 	/* Call inner routine to assign local interface address. */
1003 	if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0)
1004 		return (error);
1005 
1006 	if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port,
1007 			      inp->inp_laddr.s_addr ?
1008 				inp->inp_laddr : if_sin->sin_addr,
1009 			      inp->inp_lport, FALSE, NULL) != NULL) {
1010 		return (EADDRINUSE);
1011 	}
1012 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
1013 		if (inp->inp_lport == 0) {
1014 			error = in_pcbbind(inp, NULL, td);
1015 			if (error)
1016 				return (error);
1017 		}
1018 		inp->inp_laddr = if_sin->sin_addr;
1019 	}
1020 	inp->inp_faddr = sin->sin_addr;
1021 	inp->inp_fport = sin->sin_port;
1022 	in_pcbinsconnhash(inp);
1023 	return (0);
1024 }
1025 
1026 void
1027 in_pcbdisconnect(struct inpcb *inp)
1028 {
1029 
1030 	in_pcbremconnhash(inp);
1031 	inp->inp_faddr.s_addr = INADDR_ANY;
1032 	inp->inp_fport = 0;
1033 }
1034 
1035 void
1036 in_pcbdetach(struct inpcb *inp)
1037 {
1038 	struct socket *so = inp->inp_socket;
1039 	struct inpcbinfo *ipi = inp->inp_pcbinfo;
1040 
1041 #ifdef IPSEC
1042 	ipsec4_delete_pcbpolicy(inp);
1043 #endif /*IPSEC*/
1044 	inp->inp_gencnt = ++ipi->ipi_gencnt;
1045 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
1046 	in_pcbremlists(inp);
1047 	so->so_pcb = NULL;
1048 	sofree(so);			/* remove pcb ref */
1049 	if (inp->inp_options)
1050 		m_free(inp->inp_options);
1051 	if (inp->inp_route.ro_rt)
1052 		rtfree(inp->inp_route.ro_rt);
1053 	ip_freemoptions(inp->inp_moptions);
1054 	kfree(inp, M_PCB);
1055 }
1056 
1057 /*
1058  * The socket may have an invalid PCB, i.e. NULL.  For example, a TCP
1059  * socket received RST.
1060  */
1061 static int
1062 in_setsockaddr(struct socket *so, struct sockaddr **nam)
1063 {
1064 	struct inpcb *inp;
1065 	struct sockaddr_in *sin;
1066 
1067 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1068 	inp = so->so_pcb;
1069 	if (!inp)
1070 		return (ECONNRESET);
1071 
1072 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1073 	sin->sin_family = AF_INET;
1074 	sin->sin_len = sizeof *sin;
1075 	sin->sin_port = inp->inp_lport;
1076 	sin->sin_addr = inp->inp_laddr;
1077 
1078 	*nam = (struct sockaddr *)sin;
1079 	return (0);
1080 }
1081 
1082 void
1083 in_setsockaddr_dispatch(netmsg_t msg)
1084 {
1085 	int error;
1086 
1087 	error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1088 	lwkt_replymsg(&msg->lmsg, error);
1089 }
1090 
1091 /*
1092  * The socket may have an invalid PCB, i.e. NULL.  For example, a TCP
1093  * socket received RST.
1094  */
1095 int
1096 in_setpeeraddr(struct socket *so, struct sockaddr **nam)
1097 {
1098 	struct inpcb *inp;
1099 	struct sockaddr_in *sin;
1100 
1101 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
1102 	inp = so->so_pcb;
1103 	if (!inp)
1104 		return (ECONNRESET);
1105 
1106 	sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO);
1107 	sin->sin_family = AF_INET;
1108 	sin->sin_len = sizeof *sin;
1109 	sin->sin_port = inp->inp_fport;
1110 	sin->sin_addr = inp->inp_faddr;
1111 
1112 	*nam = (struct sockaddr *)sin;
1113 	return (0);
1114 }
1115 
1116 void
1117 in_setpeeraddr_dispatch(netmsg_t msg)
1118 {
1119 	int error;
1120 
1121 	error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam);
1122 	lwkt_replymsg(&msg->lmsg, error);
1123 }
1124 
1125 void
1126 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err,
1127     inp_notify_t notify)
1128 {
1129 	struct inpcb *inp, *marker;
1130 
1131 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1132 	    ("not in the correct netisr"));
1133 	marker = &in_pcbmarkers[mycpuid];
1134 
1135 	/*
1136 	 * NOTE:
1137 	 * - If INP_PLACEMARKER is set we must ignore the rest of the
1138 	 *   structure and skip it.
1139 	 * - It is safe to nuke inpcbs here, since we are in their own
1140 	 *   netisr.
1141 	 */
1142 	GET_PCBINFO_TOKEN(pcbinfo);
1143 
1144 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1145 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1146 		LIST_REMOVE(marker, inp_list);
1147 		LIST_INSERT_AFTER(inp, marker, inp_list);
1148 
1149 		if (inp->inp_flags & INP_PLACEMARKER)
1150 			continue;
1151 #ifdef INET6
1152 		if (!INP_ISIPV4(inp))
1153 			continue;
1154 #endif
1155 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1156 		    inp->inp_socket == NULL)
1157 			continue;
1158 		(*notify)(inp, err);		/* can remove inp from list! */
1159 	}
1160 	LIST_REMOVE(marker, inp_list);
1161 
1162 	REL_PCBINFO_TOKEN(pcbinfo);
1163 }
1164 
1165 void
1166 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1167 {
1168 	struct inpcb *inp, *marker;
1169 
1170 	/*
1171 	 * We only need to make sure that we are in netisr0, where all
1172 	 * multicast operation happen.  We could check inpcbinfo which
1173 	 * does not belong to netisr0 by holding the inpcbinfo's token.
1174 	 * In this case, the pcbinfo must be able to be shared, i.e.
1175 	 * pcbinfo->infotoken is not NULL.
1176 	 */
1177 	ASSERT_IN_NETISR(0);
1178 	KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL,
1179 	    ("pcbinfo could not be shared"));
1180 
1181 	/*
1182 	 * Get a marker for the current netisr (netisr0).
1183 	 *
1184 	 * It is possible that the multicast address deletion blocks,
1185 	 * which could cause temporary token releasing.  So we use
1186 	 * inpcb marker here to get a coherent view of the inpcb list.
1187 	 *
1188 	 * While, on the other hand, moptions are only added and deleted
1189 	 * in netisr0, so we would not see staled moption or miss moption
1190 	 * even if the token was released due to the blocking multicast
1191 	 * address deletion.
1192 	 */
1193 	marker = &in_pcbmarkers[mycpuid];
1194 
1195 	GET_PCBINFO_TOKEN(pcbinfo);
1196 
1197 	LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
1198 	while ((inp = LIST_NEXT(marker, inp_list)) != NULL) {
1199 		struct ip_moptions *imo;
1200 
1201 		LIST_REMOVE(marker, inp_list);
1202 		LIST_INSERT_AFTER(inp, marker, inp_list);
1203 
1204 		if (inp->inp_flags & INP_PLACEMARKER)
1205 			continue;
1206 		imo = inp->inp_moptions;
1207 		if (INP_ISIPV4(inp) && imo != NULL) {
1208 			int i, gap;
1209 
1210 			/*
1211 			 * Unselect the outgoing interface if it is being
1212 			 * detached.
1213 			 */
1214 			if (imo->imo_multicast_ifp == ifp)
1215 				imo->imo_multicast_ifp = NULL;
1216 
1217 			/*
1218 			 * Drop multicast group membership if we joined
1219 			 * through the interface being detached.
1220 			 */
1221 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1222 			    i++) {
1223 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1224 					/*
1225 					 * NOTE:
1226 					 * This could block and the pcbinfo
1227 					 * token could be passively released.
1228 					 */
1229 					in_delmulti(imo->imo_membership[i]);
1230 					gap++;
1231 				} else if (gap != 0)
1232 					imo->imo_membership[i - gap] =
1233 					    imo->imo_membership[i];
1234 			}
1235 			imo->imo_num_memberships -= gap;
1236 		}
1237 	}
1238 	LIST_REMOVE(marker, inp_list);
1239 
1240 	REL_PCBINFO_TOKEN(pcbinfo);
1241 }
1242 
1243 /*
1244  * Check for alternatives when higher level complains
1245  * about service problems.  For now, invalidate cached
1246  * routing information.  If the route was created dynamically
1247  * (by a redirect), time to try a default gateway again.
1248  */
1249 void
1250 in_losing(struct inpcb *inp)
1251 {
1252 	struct rtentry *rt;
1253 	struct rt_addrinfo rtinfo;
1254 
1255 	if ((rt = inp->inp_route.ro_rt)) {
1256 		bzero(&rtinfo, sizeof(struct rt_addrinfo));
1257 		rtinfo.rti_info[RTAX_DST] = rt_key(rt);
1258 		rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
1259 		rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt);
1260 		rtinfo.rti_flags = rt->rt_flags;
1261 		rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0);
1262 		if (rt->rt_flags & RTF_DYNAMIC) {
1263 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1264 			    rt_mask(rt), rt->rt_flags, NULL);
1265 		}
1266 		inp->inp_route.ro_rt = NULL;
1267 		rtfree(rt);
1268 		/*
1269 		 * A new route can be allocated
1270 		 * the next time output is attempted.
1271 		 */
1272 	}
1273 }
1274 
1275 /*
1276  * After a routing change, flush old routing
1277  * and allocate a (hopefully) better one.
1278  */
1279 void
1280 in_rtchange(struct inpcb *inp, int err)
1281 {
1282 	if (inp->inp_route.ro_rt) {
1283 		rtfree(inp->inp_route.ro_rt);
1284 		inp->inp_route.ro_rt = NULL;
1285 		/*
1286 		 * A new route can be allocated the next time
1287 		 * output is attempted.
1288 		 */
1289 	}
1290 }
1291 
1292 /*
1293  * Lookup a PCB based on the local address and port.
1294  */
1295 struct inpcb *
1296 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr,
1297 		   u_int lport_arg, int wild_okay, struct ucred *cred)
1298 {
1299 	struct inpcb *inp;
1300 	int matchwild = 3, wildcard;
1301 	u_short lport = lport_arg;
1302 	struct inpcbporthead *porthash;
1303 	struct inpcbport *phd;
1304 	struct inpcb *match = NULL;
1305 
1306 	/*
1307 	 * If the porthashbase is shared across several cpus, it must
1308 	 * have been locked.
1309 	 */
1310 	ASSERT_PORT_TOKEN_HELD(portinfo);
1311 
1312 	/*
1313 	 * Best fit PCB lookup.
1314 	 *
1315 	 * First see if this local port is in use by looking on the
1316 	 * port hash list.
1317 	 */
1318 	porthash = &portinfo->porthashbase[
1319 			INP_PCBPORTHASH(lport, portinfo->porthashmask)];
1320 	LIST_FOREACH(phd, porthash, phd_hash) {
1321 		if (phd->phd_port == lport)
1322 			break;
1323 	}
1324 	if (phd != NULL) {
1325 		/*
1326 		 * Port is in use by one or more PCBs. Look for best
1327 		 * fit.
1328 		 */
1329 		LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1330 			wildcard = 0;
1331 #ifdef INET6
1332 			if (!INP_ISIPV4(inp))
1333 				continue;
1334 #endif
1335 			if (inp->inp_faddr.s_addr != INADDR_ANY)
1336 				wildcard++;
1337 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
1338 				if (laddr.s_addr == INADDR_ANY)
1339 					wildcard++;
1340 				else if (inp->inp_laddr.s_addr != laddr.s_addr)
1341 					continue;
1342 			} else {
1343 				if (laddr.s_addr != INADDR_ANY)
1344 					wildcard++;
1345 			}
1346 			if (wildcard && !wild_okay)
1347 				continue;
1348 			if (wildcard < matchwild &&
1349 			    (cred == NULL ||
1350 			     cred->cr_prison ==
1351 					inp->inp_socket->so_cred->cr_prison)) {
1352 				match = inp;
1353 				matchwild = wildcard;
1354 				if (matchwild == 0) {
1355 					break;
1356 				}
1357 			}
1358 		}
1359 	}
1360 	return (match);
1361 }
1362 
1363 struct inpcb *
1364 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo,
1365     const struct inpcb *inp)
1366 {
1367 	const struct inp_localgrphead *hdr;
1368 	const struct inp_localgroup *grp;
1369 	int i;
1370 
1371 	if (pcbinfo->localgrphashbase == NULL)
1372 		return NULL;
1373 
1374 	GET_PCBINFO_TOKEN(pcbinfo);
1375 
1376 	hdr = &pcbinfo->localgrphashbase[
1377 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1378 
1379 	LIST_FOREACH(grp, hdr, il_list) {
1380 		if (grp->il_af == inp->inp_af &&
1381 		    grp->il_lport == inp->inp_lport &&
1382 		    memcmp(&grp->il_dependladdr,
1383 			&inp->inp_inc.inc_ie.ie_dependladdr,
1384 			sizeof(grp->il_dependladdr)) == 0) {
1385 			break;
1386 		}
1387 	}
1388 	if (grp == NULL || grp->il_inpcnt == 1) {
1389 		REL_PCBINFO_TOKEN(pcbinfo);
1390 		return NULL;
1391 	}
1392 
1393 	KASSERT(grp->il_inpcnt >= 2,
1394 	    ("invalid localgroup inp count %d", grp->il_inpcnt));
1395 	for (i = 0; i < grp->il_inpcnt; ++i) {
1396 		if (grp->il_inp[i] == inp) {
1397 			int last = grp->il_inpcnt - 1;
1398 
1399 			if (i == last)
1400 				last = grp->il_inpcnt - 2;
1401 			REL_PCBINFO_TOKEN(pcbinfo);
1402 			return grp->il_inp[last];
1403 		}
1404 	}
1405 	REL_PCBINFO_TOKEN(pcbinfo);
1406 	return NULL;
1407 }
1408 
1409 static struct inpcb *
1410 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo,
1411     struct in_addr laddr, uint16_t lport, uint32_t pkt_hash)
1412 {
1413 	struct inpcb *local_wild = NULL;
1414 	const struct inp_localgrphead *hdr;
1415 	const struct inp_localgroup *grp;
1416 
1417 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1418 
1419 	hdr = &pcbinfo->localgrphashbase[
1420 	    INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)];
1421 
1422 	/*
1423 	 * Order of socket selection:
1424 	 * 1. non-wild.
1425 	 * 2. wild.
1426 	 *
1427 	 * NOTE: Local group does not contain jailed sockets
1428 	 */
1429 	LIST_FOREACH(grp, hdr, il_list) {
1430 #ifdef INET6
1431 		if (grp->il_af != AF_INET)
1432 			continue;
1433 #endif
1434 		if (grp->il_lport == lport) {
1435 			int idx;
1436 
1437 			/*
1438 			 * Modulo-N is used here, which greatly reduces
1439 			 * completion queue token contention, thus more
1440 			 * cpu time is saved.
1441 			 */
1442 			idx = pkt_hash % grp->il_inpcnt;
1443 			if (grp->il_laddr.s_addr == laddr.s_addr)
1444 				return grp->il_inp[idx];
1445 			else if (grp->il_laddr.s_addr == INADDR_ANY)
1446 				local_wild = grp->il_inp[idx];
1447 		}
1448 	}
1449 	if (local_wild != NULL)
1450 		return local_wild;
1451 	return NULL;
1452 }
1453 
1454 /*
1455  * Lookup PCB in hash list.
1456  */
1457 struct inpcb *
1458 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1459     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1460     boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m)
1461 {
1462 	struct inpcbhead *head;
1463 	struct inpcb *inp, *jinp=NULL;
1464 	u_short fport = fport_arg, lport = lport_arg;
1465 
1466 	/*
1467 	 * First look for an exact match.
1468 	 */
1469 	head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport,
1470 	    laddr.s_addr, lport, pcbinfo->hashmask)];
1471 	LIST_FOREACH(inp, head, inp_hash) {
1472 #ifdef INET6
1473 		if (!INP_ISIPV4(inp))
1474 			continue;
1475 #endif
1476 		if (in_hosteq(inp->inp_faddr, faddr) &&
1477 		    in_hosteq(inp->inp_laddr, laddr) &&
1478 		    inp->inp_fport == fport && inp->inp_lport == lport) {
1479 			/* found */
1480 			if (inp->inp_socket == NULL ||
1481 			    inp->inp_socket->so_cred->cr_prison == NULL) {
1482 				return (inp);
1483 			} else {
1484 				if  (jinp == NULL)
1485 					jinp = inp;
1486 			}
1487 		}
1488 	}
1489 	if (jinp != NULL)
1490 		return (jinp);
1491 
1492 	if (wildcard) {
1493 		struct inpcb *local_wild = NULL;
1494 		struct inpcb *jinp_wild = NULL;
1495 		struct inpcontainer *ic;
1496 		struct inpcontainerhead *chead;
1497 		struct sockaddr_in jsin;
1498 		struct ucred *cred;
1499 
1500 		GET_PCBINFO_TOKEN(pcbinfo);
1501 
1502 		/*
1503 		 * Check local group first
1504 		 */
1505 		if (pcbinfo->localgrphashbase != NULL &&
1506 		    m != NULL && (m->m_flags & M_HASH) &&
1507 		    !(ifp && ifp->if_type == IFT_FAITH)) {
1508 			inp = inp_localgroup_lookup(pcbinfo,
1509 			    laddr, lport, m->m_pkthdr.hash);
1510 			if (inp != NULL) {
1511 				REL_PCBINFO_TOKEN(pcbinfo);
1512 				return inp;
1513 			}
1514 		}
1515 
1516 		/*
1517 		 * Order of socket selection:
1518 		 * 1. non-jailed, non-wild.
1519 		 * 2. non-jailed, wild.
1520 		 * 3. jailed, non-wild.
1521 		 * 4. jailed, wild.
1522 		 */
1523 		jsin.sin_family = AF_INET;
1524 		chead = &pcbinfo->wildcardhashbase[
1525 		    INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)];
1526 		LIST_FOREACH(ic, chead, ic_list) {
1527 			inp = ic->ic_inp;
1528 			if (inp->inp_flags & INP_PLACEMARKER)
1529 				continue;
1530 
1531 			jsin.sin_addr.s_addr = laddr.s_addr;
1532 #ifdef INET6
1533 			if (!INP_ISIPV4(inp))
1534 				continue;
1535 #endif
1536 			if (inp->inp_socket != NULL)
1537 				cred = inp->inp_socket->so_cred;
1538 			else
1539 				cred = NULL;
1540 			if (cred != NULL && jailed(cred)) {
1541 				if (jinp != NULL)
1542 					continue;
1543 				else
1544 					if (!jailed_ip(cred->cr_prison,
1545 					    (struct sockaddr *)&jsin))
1546 						continue;
1547 			}
1548 			if (inp->inp_lport == lport) {
1549 				if (ifp && ifp->if_type == IFT_FAITH &&
1550 				    !(inp->inp_flags & INP_FAITH))
1551 					continue;
1552 				if (inp->inp_laddr.s_addr == laddr.s_addr) {
1553 					if (cred != NULL && jailed(cred)) {
1554 						jinp = inp;
1555 					} else {
1556 						REL_PCBINFO_TOKEN(pcbinfo);
1557 						return (inp);
1558 					}
1559 				}
1560 				if (inp->inp_laddr.s_addr == INADDR_ANY) {
1561 					if (cred != NULL && jailed(cred))
1562 						jinp_wild = inp;
1563 					else
1564 						local_wild = inp;
1565 				}
1566 			}
1567 		}
1568 
1569 		REL_PCBINFO_TOKEN(pcbinfo);
1570 
1571 		if (local_wild != NULL)
1572 			return (local_wild);
1573 		if (jinp != NULL)
1574 			return (jinp);
1575 		return (jinp_wild);
1576 	}
1577 
1578 	/*
1579 	 * Not found.
1580 	 */
1581 	return (NULL);
1582 }
1583 
1584 struct inpcb *
1585 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1586     u_int fport_arg, struct in_addr laddr, u_int lport_arg,
1587     boolean_t wildcard, struct ifnet *ifp)
1588 {
1589 	return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg,
1590 	    laddr, lport_arg, wildcard, ifp, NULL);
1591 }
1592 
1593 /*
1594  * Insert PCB into connection hash table.
1595  */
1596 void
1597 in_pcbinsconnhash(struct inpcb *inp)
1598 {
1599 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1600 	struct inpcbhead *bucket;
1601 	u_int32_t hashkey_faddr, hashkey_laddr;
1602 
1603 #ifdef INET6
1604 	if (INP_ISIPV6(inp)) {
1605 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */;
1606 		hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */;
1607 	} else {
1608 #endif
1609 		hashkey_faddr = inp->inp_faddr.s_addr;
1610 		hashkey_laddr = inp->inp_laddr.s_addr;
1611 #ifdef INET6
1612 	}
1613 #endif
1614 
1615 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1616 	    ("not in the correct netisr"));
1617 	ASSERT_INP_NOTINHASH(inp);
1618 	inp->inp_flags |= INP_CONNECTED;
1619 
1620 	/*
1621 	 * Insert into the connection hash table.
1622 	 */
1623 	bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr,
1624 	    inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)];
1625 	LIST_INSERT_HEAD(bucket, inp, inp_hash);
1626 }
1627 
1628 /*
1629  * Remove PCB from connection hash table.
1630  */
1631 void
1632 in_pcbremconnhash(struct inpcb *inp)
1633 {
1634 	struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo;
1635 
1636 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1637 	    ("not in the correct netisr"));
1638 	KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected"));
1639 
1640 	LIST_REMOVE(inp, inp_hash);
1641 	inp->inp_flags &= ~INP_CONNECTED;
1642 }
1643 
1644 /*
1645  * Insert PCB into port hash table.
1646  */
1647 void
1648 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp)
1649 {
1650 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1651 	struct inpcbporthead *pcbporthash;
1652 	struct inpcbport *phd;
1653 
1654 	/*
1655 	 * If the porthashbase is shared across several cpus, it must
1656 	 * have been locked.
1657 	 */
1658 	ASSERT_PORT_TOKEN_HELD(portinfo);
1659 
1660 	/*
1661 	 * Insert into the port hash table.
1662 	 */
1663 	pcbporthash = &portinfo->porthashbase[
1664 	    INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)];
1665 
1666 	/* Go through port list and look for a head for this lport. */
1667 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
1668 		if (phd->phd_port == inp->inp_lport)
1669 			break;
1670 	}
1671 
1672 	/* If none exists, use saved one and tack it on. */
1673 	if (phd == NULL) {
1674 		KKASSERT(pcbinfo->portsave != NULL);
1675 		phd = pcbinfo->portsave;
1676 		pcbinfo->portsave = NULL;
1677 		phd->phd_port = inp->inp_lport;
1678 		LIST_INIT(&phd->phd_pcblist);
1679 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
1680 	}
1681 
1682 	inp->inp_portinfo = portinfo;
1683 	inp->inp_phd = phd;
1684 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
1685 
1686 	/*
1687 	 * Malloc one inpcbport for later use.  It is safe to use
1688 	 * "wait" malloc here (port token would be released, if
1689 	 * malloc ever blocked), since all changes to the porthash
1690 	 * are done.
1691 	 */
1692 	if (pcbinfo->portsave == NULL) {
1693 		pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave),
1694 					    M_PCB, M_INTWAIT | M_ZERO);
1695 	}
1696 }
1697 
1698 void
1699 in_pcbinsporthash_lport(struct inpcb *inp)
1700 {
1701 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1702 	struct inpcbportinfo *portinfo;
1703 	u_short lport_ho;
1704 
1705 	/* Locate the proper portinfo based on lport */
1706 	lport_ho = ntohs(inp->inp_lport);
1707 	portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask];
1708 	KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset);
1709 
1710 	GET_PORT_TOKEN(portinfo);
1711 	in_pcbinsporthash(portinfo, inp);
1712 	REL_PORT_TOKEN(portinfo);
1713 }
1714 
1715 static struct inp_localgroup *
1716 inp_localgroup_alloc(u_char af, uint16_t port,
1717     const union in_dependaddr *addr, int size)
1718 {
1719 	struct inp_localgroup *grp;
1720 
1721 	grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]),
1722 	    M_TEMP, M_INTWAIT | M_ZERO);
1723 	grp->il_af = af;
1724 	grp->il_lport = port;
1725 	grp->il_dependladdr = *addr;
1726 	grp->il_inpsiz = size;
1727 
1728 	return grp;
1729 }
1730 
1731 static void
1732 inp_localgroup_free(struct inp_localgroup *grp)
1733 {
1734 	kfree(grp, M_TEMP);
1735 }
1736 
1737 static void
1738 inp_localgroup_destroy(struct inp_localgroup *grp)
1739 {
1740 	LIST_REMOVE(grp, il_list);
1741 	inp_localgroup_free(grp);
1742 }
1743 
1744 static void
1745 inp_localgroup_copy(struct inp_localgroup *grp,
1746     const struct inp_localgroup *old_grp)
1747 {
1748 	int i;
1749 
1750 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
1751 	    ("invalid new local group size %d and old local group count %d",
1752 	     grp->il_inpsiz, old_grp->il_inpcnt));
1753 	for (i = 0; i < old_grp->il_inpcnt; ++i)
1754 		grp->il_inp[i] = old_grp->il_inp[i];
1755 	grp->il_inpcnt = old_grp->il_inpcnt;
1756 }
1757 
1758 static void
1759 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1760 {
1761 	struct inp_localgrphead *hdr;
1762 	struct inp_localgroup *grp, *grp_alloc = NULL;
1763 	struct ucred *cred;
1764 	int i, idx;
1765 
1766 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1767 
1768 	if (pcbinfo->localgrphashbase == NULL)
1769 		return;
1770 
1771 	/*
1772 	 * XXX don't allow jailed socket to join local group
1773 	 */
1774 	if (inp->inp_socket != NULL)
1775 		cred = inp->inp_socket->so_cred;
1776 	else
1777 		cred = NULL;
1778 	if (cred != NULL && jailed(cred))
1779 		return;
1780 
1781 	hdr = &pcbinfo->localgrphashbase[
1782 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1783 
1784 again:
1785 	LIST_FOREACH(grp, hdr, il_list) {
1786 		if (grp->il_af == inp->inp_af &&
1787 		    grp->il_lport == inp->inp_lport &&
1788 		    memcmp(&grp->il_dependladdr,
1789 		        &inp->inp_inc.inc_ie.ie_dependladdr,
1790 		        sizeof(grp->il_dependladdr)) == 0) {
1791 			break;
1792 		}
1793 	}
1794 	if (grp == NULL) {
1795 		/*
1796 		 * Create a new local group
1797 		 */
1798 		if (grp_alloc == NULL) {
1799 			grp_alloc = inp_localgroup_alloc(inp->inp_af,
1800 			    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
1801 			    INP_LOCALGROUP_SIZMIN);
1802 			/*
1803 			 * Local group allocation could block and the
1804 			 * local group w/ the same property might have
1805 			 * been added by others when we were blocked;
1806 			 * check again.
1807 			 */
1808 			goto again;
1809 		} else {
1810 			/* Local group has been allocated; link it */
1811 			grp = grp_alloc;
1812 			grp_alloc = NULL;
1813 			LIST_INSERT_HEAD(hdr, grp, il_list);
1814 		}
1815 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
1816 		if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) {
1817 			static int limit_logged = 0;
1818 
1819 			if (!limit_logged) {
1820 				limit_logged = 1;
1821 				kprintf("local group port %d, "
1822 				    "limit reached\n", ntohs(grp->il_lport));
1823 			}
1824 			if (grp_alloc != NULL) {
1825 				/*
1826 				 * This would happen if the local group
1827 				 * w/ the same property was expanded when
1828 				 * our local group allocation blocked.
1829 				 */
1830 				inp_localgroup_free(grp_alloc);
1831 			}
1832 			return;
1833 		}
1834 
1835 		/*
1836 		 * Expand this local group
1837 		 */
1838 		if (grp_alloc == NULL ||
1839 		    grp->il_inpcnt >= grp_alloc->il_inpsiz) {
1840 			if (grp_alloc != NULL)
1841 				inp_localgroup_free(grp_alloc);
1842 			grp_alloc = inp_localgroup_alloc(grp->il_af,
1843 			    grp->il_lport, &grp->il_dependladdr,
1844 			    grp->il_inpsiz * 2);
1845 			/*
1846 			 * Local group allocation could block and the
1847 			 * local group w/ the same property might have
1848 			 * been expanded by others when we were blocked;
1849 			 * check again.
1850 			 */
1851 			goto again;
1852 		}
1853 
1854 		/*
1855 		 * Save the old local group, link the new one, and then
1856 		 * destroy the old local group
1857 		 */
1858 		inp_localgroup_copy(grp_alloc, grp);
1859 		LIST_INSERT_HEAD(hdr, grp_alloc, il_list);
1860 		inp_localgroup_destroy(grp);
1861 
1862 		grp = grp_alloc;
1863 		grp_alloc = NULL;
1864 	} else {
1865 		/*
1866 		 * Found the local group
1867 		 */
1868 		if (grp_alloc != NULL) {
1869 			/*
1870 			 * This would happen if the local group w/ the
1871 			 * same property was added or expanded when our
1872 			 * local group allocation blocked.
1873 			 */
1874 			inp_localgroup_free(grp_alloc);
1875 			grp_alloc = NULL;
1876 		}
1877 	}
1878 
1879 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
1880 	    ("invalid local group size %d and count %d",
1881 	     grp->il_inpsiz, grp->il_inpcnt));
1882 
1883 	/*
1884 	 * Keep the local group sorted by the inpcb local group index
1885 	 * in ascending order.
1886 	 *
1887 	 * This eases the multi-process userland application which uses
1888 	 * SO_REUSEPORT sockets and binds process to the owner cpu of
1889 	 * the SO_REUSEPORT socket:
1890 	 * If we didn't sort the local group by the inpcb local group
1891 	 * index and one of the process owning an inpcb in this local
1892 	 * group restarted, e.g. crashed and restarted by watchdog,
1893 	 * other processes owning a inpcb in this local group would have
1894 	 * to detect that event, refetch its socket's owner cpu, and
1895 	 * re-bind.
1896 	 */
1897 	idx = grp->il_inpcnt;
1898 	for (i = 0; i < idx; ++i) {
1899 		struct inpcb *oinp = grp->il_inp[i];
1900 
1901 		if (oinp->inp_lgrpindex > i) {
1902 			if (inp->inp_lgrpindex < 0) {
1903 				inp->inp_lgrpindex = i;
1904 			} else if (inp->inp_lgrpindex != i) {
1905 				if (bootverbose) {
1906 					kprintf("inp %p: grpidx %d, "
1907 					    "assigned to %d, cpu%d\n",
1908 					    inp, inp->inp_lgrpindex, i,
1909 					    mycpuid);
1910 				}
1911 			}
1912 			grp->il_inp[i] = inp;
1913 
1914 			/* Pull down inpcbs */
1915 			for (; i < grp->il_inpcnt; ++i) {
1916 				struct inpcb *oinp1 = grp->il_inp[i + 1];
1917 
1918 				grp->il_inp[i + 1] = oinp;
1919 				oinp = oinp1;
1920 			}
1921 			grp->il_inpcnt++;
1922 			return;
1923 		}
1924 	}
1925 
1926 	if (inp->inp_lgrpindex < 0) {
1927 		inp->inp_lgrpindex = idx;
1928 	} else if (inp->inp_lgrpindex != idx) {
1929 		if (bootverbose) {
1930 			kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n",
1931 			    inp, inp->inp_lgrpindex, idx, mycpuid);
1932 		}
1933 	}
1934 	grp->il_inp[idx] = inp;
1935 	grp->il_inpcnt++;
1936 }
1937 
1938 void
1939 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1940 {
1941 	struct inpcontainer *ic;
1942 	struct inpcontainerhead *bucket;
1943 
1944 	GET_PCBINFO_TOKEN(pcbinfo);
1945 
1946 	in_pcbinslocalgrphash_oncpu(inp, pcbinfo);
1947 
1948 	bucket = &pcbinfo->wildcardhashbase[
1949 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
1950 
1951 	ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT);
1952 	ic->ic_inp = inp;
1953 	LIST_INSERT_HEAD(bucket, ic, ic_list);
1954 
1955 	REL_PCBINFO_TOKEN(pcbinfo);
1956 }
1957 
1958 /*
1959  * Insert PCB into wildcard hash table.
1960  */
1961 void
1962 in_pcbinswildcardhash(struct inpcb *inp)
1963 {
1964 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1965 
1966 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
1967 	    ("not in correct netisr"));
1968 	ASSERT_INP_NOTINHASH(inp);
1969 	inp->inp_flags |= INP_WILDCARD;
1970 
1971 	in_pcbinswildcardhash_oncpu(inp, pcbinfo);
1972 }
1973 
1974 static void
1975 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
1976 {
1977 	struct inp_localgrphead *hdr;
1978 	struct inp_localgroup *grp;
1979 
1980 	ASSERT_PCBINFO_TOKEN_HELD(pcbinfo);
1981 
1982 	if (pcbinfo->localgrphashbase == NULL)
1983 		return;
1984 
1985 	hdr = &pcbinfo->localgrphashbase[
1986 	    INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)];
1987 
1988 	LIST_FOREACH(grp, hdr, il_list) {
1989 		int i;
1990 
1991 		for (i = 0; i < grp->il_inpcnt; ++i) {
1992 			if (grp->il_inp[i] != inp)
1993 				continue;
1994 
1995 			if (grp->il_inpcnt == 1) {
1996 				/* Destroy this local group */
1997 				inp_localgroup_destroy(grp);
1998 			} else {
1999 				/* Pull up inpcbs */
2000 				for (; i + 1 < grp->il_inpcnt; ++i)
2001 					grp->il_inp[i] = grp->il_inp[i + 1];
2002 				grp->il_inpcnt--;
2003 			}
2004 			return;
2005 		}
2006 	}
2007 }
2008 
2009 void
2010 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo)
2011 {
2012 	struct inpcontainer *ic;
2013 	struct inpcontainerhead *head;
2014 
2015 	GET_PCBINFO_TOKEN(pcbinfo);
2016 
2017 	in_pcbremlocalgrphash_oncpu(inp, pcbinfo);
2018 
2019 	/* find bucket */
2020 	head = &pcbinfo->wildcardhashbase[
2021 	    INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)];
2022 
2023 	LIST_FOREACH(ic, head, ic_list) {
2024 		if (ic->ic_inp == inp)
2025 			goto found;
2026 	}
2027 	REL_PCBINFO_TOKEN(pcbinfo);
2028 	return;			/* not found! */
2029 
2030 found:
2031 	LIST_REMOVE(ic, ic_list);	/* remove container from bucket chain */
2032 	REL_PCBINFO_TOKEN(pcbinfo);
2033 	kfree(ic, M_TEMP);		/* deallocate container */
2034 }
2035 
2036 /*
2037  * Remove PCB from wildcard hash table.
2038  */
2039 void
2040 in_pcbremwildcardhash(struct inpcb *inp)
2041 {
2042 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2043 
2044 	KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu),
2045 	    ("not in correct netisr"));
2046 	KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard"));
2047 
2048 	in_pcbremwildcardhash_oncpu(inp, pcbinfo);
2049 	inp->inp_lgrpindex = -1;
2050 	inp->inp_flags &= ~INP_WILDCARD;
2051 }
2052 
2053 /*
2054  * Remove PCB from various lists.
2055  */
2056 void
2057 in_pcbremlists(struct inpcb *inp)
2058 {
2059 	if (inp->inp_lport) {
2060 		struct inpcbportinfo *portinfo;
2061 		struct inpcbport *phd;
2062 
2063 		/*
2064 		 * NOTE:
2065 		 * inp->inp_portinfo is _not_ necessary same as
2066 		 * inp->inp_pcbinfo->portinfo.
2067 		 */
2068 		portinfo = inp->inp_portinfo;
2069 		GET_PORT_TOKEN(portinfo);
2070 
2071 		phd = inp->inp_phd;
2072 		LIST_REMOVE(inp, inp_portlist);
2073 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2074 			LIST_REMOVE(phd, phd_hash);
2075 			kfree(phd, M_PCB);
2076 		}
2077 
2078 		REL_PORT_TOKEN(portinfo);
2079 	}
2080 	if (inp->inp_flags & INP_WILDCARD) {
2081 		in_pcbremwildcardhash(inp);
2082 	} else if (inp->inp_flags & INP_CONNECTED) {
2083 		in_pcbremconnhash(inp);
2084 	}
2085 
2086 	if (inp->inp_flags & INP_ONLIST)
2087 		in_pcbofflist(inp);
2088 }
2089 
2090 int
2091 prison_xinpcb(struct thread *td, struct inpcb *inp)
2092 {
2093 	struct ucred *cr;
2094 
2095 	if (td->td_proc == NULL)
2096 		return (0);
2097 	cr = td->td_proc->p_ucred;
2098 	if (cr->cr_prison == NULL)
2099 		return (0);
2100 	if (inp->inp_socket && inp->inp_socket->so_cred &&
2101 	    inp->inp_socket->so_cred->cr_prison &&
2102 	    cr->cr_prison == inp->inp_socket->so_cred->cr_prison)
2103 		return (0);
2104 	return (1);
2105 }
2106 
2107 int
2108 in_pcblist_global(SYSCTL_HANDLER_ARGS)
2109 {
2110 	struct inpcbinfo *pcbinfo_arr = arg1;
2111 	int pcbinfo_arrlen = arg2;
2112 	struct inpcb *marker;
2113 	int cpu, origcpu;
2114 	int error, n;
2115 
2116 	KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1,
2117 	    ("invalid pcbinfo count %d", pcbinfo_arrlen));
2118 
2119 	/*
2120 	 * The process of preparing the TCB list is too time-consuming and
2121 	 * resource-intensive to repeat twice on every request.
2122 	 */
2123 	n = 0;
2124 	if (req->oldptr == NULL) {
2125 		for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu)
2126 			n += pcbinfo_arr[cpu].ipi_count;
2127 		req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb);
2128 		return 0;
2129 	}
2130 
2131 	if (req->newptr != NULL)
2132 		return EPERM;
2133 
2134 	marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
2135 	marker->inp_flags |= INP_PLACEMARKER;
2136 
2137 	/*
2138 	 * OK, now we're committed to doing something.  Re-fetch ipi_count
2139 	 * after obtaining the generation count.
2140 	 */
2141 	error = 0;
2142 	origcpu = mycpuid;
2143 	for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) {
2144 		struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu];
2145 		struct inpcb *inp;
2146 		struct xinpcb xi;
2147 		int i;
2148 
2149 		lwkt_migratecpu(cpu);
2150 
2151 		GET_PCBINFO_TOKEN(pcbinfo);
2152 
2153 		n = pcbinfo->ipi_count;
2154 
2155 		LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list);
2156 		i = 0;
2157 		while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
2158 			LIST_REMOVE(marker, inp_list);
2159 			LIST_INSERT_AFTER(inp, marker, inp_list);
2160 
2161 			if (inp->inp_flags & INP_PLACEMARKER)
2162 				continue;
2163 			if (prison_xinpcb(req->td, inp))
2164 				continue;
2165 
2166 			bzero(&xi, sizeof xi);
2167 			xi.xi_len = sizeof xi;
2168 			bcopy(inp, &xi.xi_inp, sizeof *inp);
2169 			if (inp->inp_socket)
2170 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
2171 			if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0)
2172 				break;
2173 			++i;
2174 		}
2175 		LIST_REMOVE(marker, inp_list);
2176 
2177 		REL_PCBINFO_TOKEN(pcbinfo);
2178 
2179 		if (error == 0 && i < n) {
2180 			bzero(&xi, sizeof xi);
2181 			xi.xi_len = sizeof xi;
2182 			while (i < n) {
2183 				error = SYSCTL_OUT(req, &xi, sizeof xi);
2184 				if (error)
2185 					break;
2186 				++i;
2187 			}
2188 		}
2189 	}
2190 
2191 	lwkt_migratecpu(origcpu);
2192 	kfree(marker, M_TEMP);
2193 	return error;
2194 }
2195 
2196 int
2197 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS)
2198 {
2199 	return in_pcblist_global(oidp, arg1, ncpus2, req);
2200 }
2201 
2202 void
2203 in_savefaddr(struct socket *so, const struct sockaddr *faddr)
2204 {
2205 	struct sockaddr_in *sin;
2206 
2207 	KASSERT(faddr->sa_family == AF_INET,
2208 	    ("not AF_INET faddr %d", faddr->sa_family));
2209 
2210 	sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO);
2211 	sin->sin_family = AF_INET;
2212 	sin->sin_len = sizeof(*sin);
2213 	sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port;
2214 	sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr;
2215 
2216 	so->so_faddr = (struct sockaddr *)sin;
2217 }
2218 
2219 void
2220 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize,
2221     boolean_t shared, u_short offset)
2222 {
2223 	memset(portinfo, 0, sizeof(*portinfo));
2224 
2225 	portinfo->offset = offset;
2226 	portinfo->lastport = offset;
2227 	portinfo->lastlow = offset;
2228 	portinfo->lasthi = offset;
2229 
2230 	portinfo->porthashbase = hashinit(hashsize, M_PCB,
2231 	    &portinfo->porthashmask);
2232 
2233 	if (shared) {
2234 		portinfo->porttoken = kmalloc(sizeof(struct lwkt_token),
2235 		    M_PCB, M_WAITOK);
2236 		lwkt_token_init(portinfo->porttoken, "porttoken");
2237 	}
2238 }
2239 
2240 void
2241 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step)
2242 {
2243 	int hi, lo;
2244 
2245 	if (step == 1)
2246 		return;
2247 
2248 	hi = *hi0;
2249 	lo = *lo0;
2250 
2251 	hi = rounddown2(hi, step);
2252 	hi += ofs;
2253 	if (hi > (int)*hi0)
2254 		hi -= step;
2255 
2256 	lo = roundup2(lo, step);
2257 	lo -= (step - ofs);
2258 	if (lo < (int)*lo0)
2259 		lo += step;
2260 
2261 	*hi0 = hi;
2262 	*lo0 = lo;
2263 }
2264 
2265 void
2266 in_pcbglobalinit(void)
2267 {
2268 	int cpu;
2269 
2270 	in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB,
2271 	    M_WAITOK | M_ZERO);
2272 	in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer),
2273 	    M_PCB, M_WAITOK | M_ZERO);
2274 
2275 	for (cpu = 0; cpu < ncpus; ++cpu) {
2276 		struct inpcontainer *ic = &in_pcbcontainer_markers[cpu];
2277 		struct inpcb *marker = &in_pcbmarkers[cpu];
2278 
2279 		marker->inp_flags |= INP_PLACEMARKER;
2280 		ic->ic_inp = marker;
2281 	}
2282 }
2283 
2284 struct inpcb *
2285 in_pcbmarker(int cpuid)
2286 {
2287 	KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2288 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2289 
2290 	return &in_pcbmarkers[cpuid];
2291 }
2292 
2293 struct inpcontainer *
2294 in_pcbcontainer_marker(int cpuid)
2295 {
2296 	KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid));
2297 	KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
2298 
2299 	return &in_pcbcontainer_markers[cpuid];
2300 }
2301 
2302 void
2303 in_pcbresetroute(struct inpcb *inp)
2304 {
2305 	struct route *ro = &inp->inp_route;
2306 
2307 	if (ro->ro_rt != NULL)
2308 		RTFREE(ro->ro_rt);
2309 	bzero(ro, sizeof(*ro));
2310 }
2311