xref: /freebsd/sys/netinet/in_pcb.c (revision e72055b7feba695a760d45f01f0f8268b1cb4a74)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2007-2009 Robert N. M. Watson
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * All rights reserved.
7  *
8  * Portions of this software were developed by Robert N. M. Watson under
9  * contract to Juniper Networks, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ddb.h"
42 #include "opt_ipsec.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_pcbgroup.h"
46 #include "opt_rss.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/callout.h>
53 #include <sys/domain.h>
54 #include <sys/protosw.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/refcount.h>
60 #include <sys/jail.h>
61 #include <sys/kernel.h>
62 #include <sys/sysctl.h>
63 
64 #ifdef DDB
65 #include <ddb/ddb.h>
66 #endif
67 
68 #include <vm/uma.h>
69 
70 #include <net/if.h>
71 #include <net/if_var.h>
72 #include <net/if_types.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75 
76 #if defined(INET) || defined(INET6)
77 #include <netinet/in.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_rss.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/tcp_var.h>
82 #include <netinet/udp.h>
83 #include <netinet/udp_var.h>
84 #endif
85 #ifdef INET
86 #include <netinet/in_var.h>
87 #endif
88 #ifdef INET6
89 #include <netinet/ip6.h>
90 #include <netinet6/in6_pcb.h>
91 #include <netinet6/in6_var.h>
92 #include <netinet6/ip6_var.h>
93 #endif /* INET6 */
94 
95 
96 #ifdef IPSEC
97 #include <netipsec/ipsec.h>
98 #include <netipsec/key.h>
99 #endif /* IPSEC */
100 
101 #include <security/mac/mac_framework.h>
102 
103 static struct callout	ipport_tick_callout;
104 
105 /*
106  * These configure the range of local port addresses assigned to
107  * "unspecified" outgoing connections/packets/whatever.
108  */
109 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
110 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
111 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
112 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
113 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
114 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
115 
116 /*
117  * Reserved ports accessible only to root. There are significant
118  * security considerations that must be accounted for when changing these,
119  * but the security benefits can be great. Please be careful.
120  */
121 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
122 VNET_DEFINE(int, ipport_reservedlow);
123 
124 /* Variables dealing with random ephemeral port allocation. */
125 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
126 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
127 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
128 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
129 VNET_DEFINE(int, ipport_tcpallocs);
130 static VNET_DEFINE(int, ipport_tcplastcount);
131 
132 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
133 
134 static void	in_pcbremlists(struct inpcb *inp);
135 #ifdef INET
136 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
137 			    struct in_addr faddr, u_int fport_arg,
138 			    struct in_addr laddr, u_int lport_arg,
139 			    int lookupflags, struct ifnet *ifp);
140 
141 #define RANGECHK(var, min, max) \
142 	if ((var) < (min)) { (var) = (min); } \
143 	else if ((var) > (max)) { (var) = (max); }
144 
145 static int
146 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
147 {
148 	int error;
149 
150 	error = sysctl_handle_int(oidp, arg1, arg2, req);
151 	if (error == 0) {
152 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
153 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
154 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
155 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
156 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
157 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
158 	}
159 	return (error);
160 }
161 
162 #undef RANGECHK
163 
164 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
165     "IP Ports");
166 
167 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
168 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowfirstauto), 0,
169 	&sysctl_net_ipport_check, "I", "");
170 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
171 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowlastauto), 0,
172 	&sysctl_net_ipport_check, "I", "");
173 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, first,
174 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_firstauto), 0,
175 	&sysctl_net_ipport_check, "I", "");
176 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, last,
177 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lastauto), 0,
178 	&sysctl_net_ipport_check, "I", "");
179 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
180 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hifirstauto), 0,
181 	&sysctl_net_ipport_check, "I", "");
182 SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
183 	CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hilastauto), 0,
184 	&sysctl_net_ipport_check, "I", "");
185 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
186 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, "");
187 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
188 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
189 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_RW,
190 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
191 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_RW,
192 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
193 	"allocations before switching to a sequental one");
194 SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_RW,
195 	&VNET_NAME(ipport_randomtime), 0,
196 	"Minimum time to keep sequental port "
197 	"allocation before switching to a random one");
198 #endif /* INET */
199 
200 /*
201  * in_pcb.c: manage the Protocol Control Blocks.
202  *
203  * NOTE: It is assumed that most of these functions will be called with
204  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
205  * functions often modify hash chains or addresses in pcbs.
206  */
207 
208 /*
209  * Initialize an inpcbinfo -- we should be able to reduce the number of
210  * arguments in time.
211  */
212 void
213 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
214     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
215     char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
216     uint32_t inpcbzone_flags, u_int hashfields)
217 {
218 
219 	INP_INFO_LOCK_INIT(pcbinfo, name);
220 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
221 #ifdef VIMAGE
222 	pcbinfo->ipi_vnet = curvnet;
223 #endif
224 	pcbinfo->ipi_listhead = listhead;
225 	LIST_INIT(pcbinfo->ipi_listhead);
226 	pcbinfo->ipi_count = 0;
227 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
228 	    &pcbinfo->ipi_hashmask);
229 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
230 	    &pcbinfo->ipi_porthashmask);
231 #ifdef PCBGROUP
232 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
233 #endif
234 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
235 	    NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
236 	    inpcbzone_flags);
237 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
238 	uma_zone_set_warning(pcbinfo->ipi_zone,
239 	    "kern.ipc.maxsockets limit reached");
240 }
241 
242 /*
243  * Destroy an inpcbinfo.
244  */
245 void
246 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
247 {
248 
249 	KASSERT(pcbinfo->ipi_count == 0,
250 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
251 
252 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
253 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
254 	    pcbinfo->ipi_porthashmask);
255 #ifdef PCBGROUP
256 	in_pcbgroup_destroy(pcbinfo);
257 #endif
258 	uma_zdestroy(pcbinfo->ipi_zone);
259 	INP_HASH_LOCK_DESTROY(pcbinfo);
260 	INP_INFO_LOCK_DESTROY(pcbinfo);
261 }
262 
263 /*
264  * Allocate a PCB and associate it with the socket.
265  * On success return with the PCB locked.
266  */
267 int
268 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
269 {
270 	struct inpcb *inp;
271 	int error;
272 
273 	INP_INFO_WLOCK_ASSERT(pcbinfo);
274 	error = 0;
275 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
276 	if (inp == NULL)
277 		return (ENOBUFS);
278 	bzero(inp, inp_zero_size);
279 	inp->inp_pcbinfo = pcbinfo;
280 	inp->inp_socket = so;
281 	inp->inp_cred = crhold(so->so_cred);
282 	inp->inp_inc.inc_fibnum = so->so_fibnum;
283 #ifdef MAC
284 	error = mac_inpcb_init(inp, M_NOWAIT);
285 	if (error != 0)
286 		goto out;
287 	mac_inpcb_create(so, inp);
288 #endif
289 #ifdef IPSEC
290 	error = ipsec_init_policy(so, &inp->inp_sp);
291 	if (error != 0) {
292 #ifdef MAC
293 		mac_inpcb_destroy(inp);
294 #endif
295 		goto out;
296 	}
297 #endif /*IPSEC*/
298 #ifdef INET6
299 	if (INP_SOCKAF(so) == AF_INET6) {
300 		inp->inp_vflag |= INP_IPV6PROTO;
301 		if (V_ip6_v6only)
302 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
303 	}
304 #endif
305 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
306 	pcbinfo->ipi_count++;
307 	so->so_pcb = (caddr_t)inp;
308 #ifdef INET6
309 	if (V_ip6_auto_flowlabel)
310 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
311 #endif
312 	INP_WLOCK(inp);
313 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
314 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
315 #if defined(IPSEC) || defined(MAC)
316 out:
317 	if (error != 0) {
318 		crfree(inp->inp_cred);
319 		uma_zfree(pcbinfo->ipi_zone, inp);
320 	}
321 #endif
322 	return (error);
323 }
324 
325 #ifdef INET
326 int
327 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
328 {
329 	int anonport, error;
330 
331 	INP_WLOCK_ASSERT(inp);
332 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
333 
334 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
335 		return (EINVAL);
336 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
337 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
338 	    &inp->inp_lport, cred);
339 	if (error)
340 		return (error);
341 	if (in_pcbinshash(inp) != 0) {
342 		inp->inp_laddr.s_addr = INADDR_ANY;
343 		inp->inp_lport = 0;
344 		return (EAGAIN);
345 	}
346 	if (anonport)
347 		inp->inp_flags |= INP_ANONPORT;
348 	return (0);
349 }
350 #endif
351 
352 /*
353  * Select a local port (number) to use.
354  */
355 #if defined(INET) || defined(INET6)
356 int
357 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
358     struct ucred *cred, int lookupflags)
359 {
360 	struct inpcbinfo *pcbinfo;
361 	struct inpcb *tmpinp;
362 	unsigned short *lastport;
363 	int count, dorandom, error;
364 	u_short aux, first, last, lport;
365 #ifdef INET
366 	struct in_addr laddr;
367 #endif
368 
369 	pcbinfo = inp->inp_pcbinfo;
370 
371 	/*
372 	 * Because no actual state changes occur here, a global write lock on
373 	 * the pcbinfo isn't required.
374 	 */
375 	INP_LOCK_ASSERT(inp);
376 	INP_HASH_LOCK_ASSERT(pcbinfo);
377 
378 	if (inp->inp_flags & INP_HIGHPORT) {
379 		first = V_ipport_hifirstauto;	/* sysctl */
380 		last  = V_ipport_hilastauto;
381 		lastport = &pcbinfo->ipi_lasthi;
382 	} else if (inp->inp_flags & INP_LOWPORT) {
383 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
384 		if (error)
385 			return (error);
386 		first = V_ipport_lowfirstauto;	/* 1023 */
387 		last  = V_ipport_lowlastauto;	/* 600 */
388 		lastport = &pcbinfo->ipi_lastlow;
389 	} else {
390 		first = V_ipport_firstauto;	/* sysctl */
391 		last  = V_ipport_lastauto;
392 		lastport = &pcbinfo->ipi_lastport;
393 	}
394 	/*
395 	 * For UDP(-Lite), use random port allocation as long as the user
396 	 * allows it.  For TCP (and as of yet unknown) connections,
397 	 * use random port allocation only if the user allows it AND
398 	 * ipport_tick() allows it.
399 	 */
400 	if (V_ipport_randomized &&
401 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
402 		pcbinfo == &V_ulitecbinfo))
403 		dorandom = 1;
404 	else
405 		dorandom = 0;
406 	/*
407 	 * It makes no sense to do random port allocation if
408 	 * we have the only port available.
409 	 */
410 	if (first == last)
411 		dorandom = 0;
412 	/* Make sure to not include UDP(-Lite) packets in the count. */
413 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
414 		V_ipport_tcpallocs++;
415 	/*
416 	 * Instead of having two loops further down counting up or down
417 	 * make sure that first is always <= last and go with only one
418 	 * code path implementing all logic.
419 	 */
420 	if (first > last) {
421 		aux = first;
422 		first = last;
423 		last = aux;
424 	}
425 
426 #ifdef INET
427 	/* Make the compiler happy. */
428 	laddr.s_addr = 0;
429 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
430 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
431 		    __func__, inp));
432 		laddr = *laddrp;
433 	}
434 #endif
435 	tmpinp = NULL;	/* Make compiler happy. */
436 	lport = *lportp;
437 
438 	if (dorandom)
439 		*lastport = first + (arc4random() % (last - first));
440 
441 	count = last - first;
442 
443 	do {
444 		if (count-- < 0)	/* completely used? */
445 			return (EADDRNOTAVAIL);
446 		++*lastport;
447 		if (*lastport < first || *lastport > last)
448 			*lastport = first;
449 		lport = htons(*lastport);
450 
451 #ifdef INET6
452 		if ((inp->inp_vflag & INP_IPV6) != 0)
453 			tmpinp = in6_pcblookup_local(pcbinfo,
454 			    &inp->in6p_laddr, lport, lookupflags, cred);
455 #endif
456 #if defined(INET) && defined(INET6)
457 		else
458 #endif
459 #ifdef INET
460 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
461 			    lport, lookupflags, cred);
462 #endif
463 	} while (tmpinp != NULL);
464 
465 #ifdef INET
466 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
467 		laddrp->s_addr = laddr.s_addr;
468 #endif
469 	*lportp = lport;
470 
471 	return (0);
472 }
473 
474 /*
475  * Return cached socket options.
476  */
477 short
478 inp_so_options(const struct inpcb *inp)
479 {
480    short so_options;
481 
482    so_options = 0;
483 
484    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
485 	   so_options |= SO_REUSEPORT;
486    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
487 	   so_options |= SO_REUSEADDR;
488    return (so_options);
489 }
490 #endif /* INET || INET6 */
491 
492 /*
493  * Check if a new BINDMULTI socket is allowed to be created.
494  *
495  * ni points to the new inp.
496  * oi points to the exisitng inp.
497  *
498  * This checks whether the existing inp also has BINDMULTI and
499  * whether the credentials match.
500  */
501 int
502 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
503 {
504 	/* Check permissions match */
505 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
506 	    (ni->inp_cred->cr_uid !=
507 	    oi->inp_cred->cr_uid))
508 		return (0);
509 
510 	/* Check the existing inp has BINDMULTI set */
511 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
512 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
513 		return (0);
514 
515 	/*
516 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
517 	 * it is and it matches the checks.
518 	 */
519 	return (1);
520 }
521 
522 #ifdef INET
523 /*
524  * Set up a bind operation on a PCB, performing port allocation
525  * as required, but do not actually modify the PCB. Callers can
526  * either complete the bind by setting inp_laddr/inp_lport and
527  * calling in_pcbinshash(), or they can just use the resulting
528  * port and address to authorise the sending of a once-off packet.
529  *
530  * On error, the values of *laddrp and *lportp are not changed.
531  */
532 int
533 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
534     u_short *lportp, struct ucred *cred)
535 {
536 	struct socket *so = inp->inp_socket;
537 	struct sockaddr_in *sin;
538 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
539 	struct in_addr laddr;
540 	u_short lport = 0;
541 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
542 	int error;
543 
544 	/*
545 	 * No state changes, so read locks are sufficient here.
546 	 */
547 	INP_LOCK_ASSERT(inp);
548 	INP_HASH_LOCK_ASSERT(pcbinfo);
549 
550 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
551 		return (EADDRNOTAVAIL);
552 	laddr.s_addr = *laddrp;
553 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
554 		return (EINVAL);
555 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
556 		lookupflags = INPLOOKUP_WILDCARD;
557 	if (nam == NULL) {
558 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
559 			return (error);
560 	} else {
561 		sin = (struct sockaddr_in *)nam;
562 		if (nam->sa_len != sizeof (*sin))
563 			return (EINVAL);
564 #ifdef notdef
565 		/*
566 		 * We should check the family, but old programs
567 		 * incorrectly fail to initialize it.
568 		 */
569 		if (sin->sin_family != AF_INET)
570 			return (EAFNOSUPPORT);
571 #endif
572 		error = prison_local_ip4(cred, &sin->sin_addr);
573 		if (error)
574 			return (error);
575 		if (sin->sin_port != *lportp) {
576 			/* Don't allow the port to change. */
577 			if (*lportp != 0)
578 				return (EINVAL);
579 			lport = sin->sin_port;
580 		}
581 		/* NB: lport is left as 0 if the port isn't being changed. */
582 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
583 			/*
584 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
585 			 * allow complete duplication of binding if
586 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
587 			 * and a multicast address is bound on both
588 			 * new and duplicated sockets.
589 			 */
590 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
591 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
592 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
593 			sin->sin_port = 0;		/* yech... */
594 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
595 			/*
596 			 * Is the address a local IP address?
597 			 * If INP_BINDANY is set, then the socket may be bound
598 			 * to any endpoint address, local or not.
599 			 */
600 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
601 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
602 				return (EADDRNOTAVAIL);
603 		}
604 		laddr = sin->sin_addr;
605 		if (lport) {
606 			struct inpcb *t;
607 			struct tcptw *tw;
608 
609 			/* GROSS */
610 			if (ntohs(lport) <= V_ipport_reservedhigh &&
611 			    ntohs(lport) >= V_ipport_reservedlow &&
612 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
613 			    0))
614 				return (EACCES);
615 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
616 			    priv_check_cred(inp->inp_cred,
617 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
618 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
619 				    lport, INPLOOKUP_WILDCARD, cred);
620 	/*
621 	 * XXX
622 	 * This entire block sorely needs a rewrite.
623 	 */
624 				if (t &&
625 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
626 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
627 				    (so->so_type != SOCK_STREAM ||
628 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
629 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
630 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
631 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
632 				    (inp->inp_cred->cr_uid !=
633 				     t->inp_cred->cr_uid))
634 					return (EADDRINUSE);
635 
636 				/*
637 				 * If the socket is a BINDMULTI socket, then
638 				 * the credentials need to match and the
639 				 * original socket also has to have been bound
640 				 * with BINDMULTI.
641 				 */
642 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
643 					return (EADDRINUSE);
644 			}
645 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
646 			    lport, lookupflags, cred);
647 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
648 				/*
649 				 * XXXRW: If an incpb has had its timewait
650 				 * state recycled, we treat the address as
651 				 * being in use (for now).  This is better
652 				 * than a panic, but not desirable.
653 				 */
654 				tw = intotw(t);
655 				if (tw == NULL ||
656 				    (reuseport & tw->tw_so_options) == 0)
657 					return (EADDRINUSE);
658 			} else if (t &&
659 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
660 			    (reuseport & inp_so_options(t)) == 0) {
661 #ifdef INET6
662 				if (ntohl(sin->sin_addr.s_addr) !=
663 				    INADDR_ANY ||
664 				    ntohl(t->inp_laddr.s_addr) !=
665 				    INADDR_ANY ||
666 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
667 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
668 #endif
669 				return (EADDRINUSE);
670 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
671 					return (EADDRINUSE);
672 			}
673 		}
674 	}
675 	if (*lportp != 0)
676 		lport = *lportp;
677 	if (lport == 0) {
678 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
679 		if (error != 0)
680 			return (error);
681 
682 	}
683 	*laddrp = laddr.s_addr;
684 	*lportp = lport;
685 	return (0);
686 }
687 
688 /*
689  * Connect from a socket to a specified address.
690  * Both address and port must be specified in argument sin.
691  * If don't have a local address for this socket yet,
692  * then pick one.
693  */
694 int
695 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
696     struct ucred *cred, struct mbuf *m)
697 {
698 	u_short lport, fport;
699 	in_addr_t laddr, faddr;
700 	int anonport, error;
701 
702 	INP_WLOCK_ASSERT(inp);
703 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
704 
705 	lport = inp->inp_lport;
706 	laddr = inp->inp_laddr.s_addr;
707 	anonport = (lport == 0);
708 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
709 	    NULL, cred);
710 	if (error)
711 		return (error);
712 
713 	/* Do the initial binding of the local address if required. */
714 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
715 		inp->inp_lport = lport;
716 		inp->inp_laddr.s_addr = laddr;
717 		if (in_pcbinshash(inp) != 0) {
718 			inp->inp_laddr.s_addr = INADDR_ANY;
719 			inp->inp_lport = 0;
720 			return (EAGAIN);
721 		}
722 	}
723 
724 	/* Commit the remaining changes. */
725 	inp->inp_lport = lport;
726 	inp->inp_laddr.s_addr = laddr;
727 	inp->inp_faddr.s_addr = faddr;
728 	inp->inp_fport = fport;
729 	in_pcbrehash_mbuf(inp, m);
730 
731 	if (anonport)
732 		inp->inp_flags |= INP_ANONPORT;
733 	return (0);
734 }
735 
736 int
737 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
738 {
739 
740 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
741 }
742 
743 /*
744  * Do proper source address selection on an unbound socket in case
745  * of connect. Take jails into account as well.
746  */
747 int
748 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
749     struct ucred *cred)
750 {
751 	struct ifaddr *ifa;
752 	struct sockaddr *sa;
753 	struct sockaddr_in *sin;
754 	struct route sro;
755 	int error;
756 
757 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
758 
759 	/*
760 	 * Bypass source address selection and use the primary jail IP
761 	 * if requested.
762 	 */
763 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
764 		return (0);
765 
766 	error = 0;
767 	bzero(&sro, sizeof(sro));
768 
769 	sin = (struct sockaddr_in *)&sro.ro_dst;
770 	sin->sin_family = AF_INET;
771 	sin->sin_len = sizeof(struct sockaddr_in);
772 	sin->sin_addr.s_addr = faddr->s_addr;
773 
774 	/*
775 	 * If route is known our src addr is taken from the i/f,
776 	 * else punt.
777 	 *
778 	 * Find out route to destination.
779 	 */
780 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
781 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
782 
783 	/*
784 	 * If we found a route, use the address corresponding to
785 	 * the outgoing interface.
786 	 *
787 	 * Otherwise assume faddr is reachable on a directly connected
788 	 * network and try to find a corresponding interface to take
789 	 * the source address from.
790 	 */
791 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
792 		struct in_ifaddr *ia;
793 		struct ifnet *ifp;
794 
795 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
796 					RT_ALL_FIBS));
797 		if (ia == NULL)
798 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
799 						RT_ALL_FIBS));
800 		if (ia == NULL) {
801 			error = ENETUNREACH;
802 			goto done;
803 		}
804 
805 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
806 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
807 			ifa_free(&ia->ia_ifa);
808 			goto done;
809 		}
810 
811 		ifp = ia->ia_ifp;
812 		ifa_free(&ia->ia_ifa);
813 		ia = NULL;
814 		IF_ADDR_RLOCK(ifp);
815 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
816 
817 			sa = ifa->ifa_addr;
818 			if (sa->sa_family != AF_INET)
819 				continue;
820 			sin = (struct sockaddr_in *)sa;
821 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
822 				ia = (struct in_ifaddr *)ifa;
823 				break;
824 			}
825 		}
826 		if (ia != NULL) {
827 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
828 			IF_ADDR_RUNLOCK(ifp);
829 			goto done;
830 		}
831 		IF_ADDR_RUNLOCK(ifp);
832 
833 		/* 3. As a last resort return the 'default' jail address. */
834 		error = prison_get_ip4(cred, laddr);
835 		goto done;
836 	}
837 
838 	/*
839 	 * If the outgoing interface on the route found is not
840 	 * a loopback interface, use the address from that interface.
841 	 * In case of jails do those three steps:
842 	 * 1. check if the interface address belongs to the jail. If so use it.
843 	 * 2. check if we have any address on the outgoing interface
844 	 *    belonging to this jail. If so use it.
845 	 * 3. as a last resort return the 'default' jail address.
846 	 */
847 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
848 		struct in_ifaddr *ia;
849 		struct ifnet *ifp;
850 
851 		/* If not jailed, use the default returned. */
852 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
853 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
854 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
855 			goto done;
856 		}
857 
858 		/* Jailed. */
859 		/* 1. Check if the iface address belongs to the jail. */
860 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
861 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
862 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
863 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
864 			goto done;
865 		}
866 
867 		/*
868 		 * 2. Check if we have any address on the outgoing interface
869 		 *    belonging to this jail.
870 		 */
871 		ia = NULL;
872 		ifp = sro.ro_rt->rt_ifp;
873 		IF_ADDR_RLOCK(ifp);
874 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
875 			sa = ifa->ifa_addr;
876 			if (sa->sa_family != AF_INET)
877 				continue;
878 			sin = (struct sockaddr_in *)sa;
879 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
880 				ia = (struct in_ifaddr *)ifa;
881 				break;
882 			}
883 		}
884 		if (ia != NULL) {
885 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
886 			IF_ADDR_RUNLOCK(ifp);
887 			goto done;
888 		}
889 		IF_ADDR_RUNLOCK(ifp);
890 
891 		/* 3. As a last resort return the 'default' jail address. */
892 		error = prison_get_ip4(cred, laddr);
893 		goto done;
894 	}
895 
896 	/*
897 	 * The outgoing interface is marked with 'loopback net', so a route
898 	 * to ourselves is here.
899 	 * Try to find the interface of the destination address and then
900 	 * take the address from there. That interface is not necessarily
901 	 * a loopback interface.
902 	 * In case of jails, check that it is an address of the jail
903 	 * and if we cannot find, fall back to the 'default' jail address.
904 	 */
905 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
906 		struct sockaddr_in sain;
907 		struct in_ifaddr *ia;
908 
909 		bzero(&sain, sizeof(struct sockaddr_in));
910 		sain.sin_family = AF_INET;
911 		sain.sin_len = sizeof(struct sockaddr_in);
912 		sain.sin_addr.s_addr = faddr->s_addr;
913 
914 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), RT_ALL_FIBS));
915 		if (ia == NULL)
916 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
917 						RT_ALL_FIBS));
918 		if (ia == NULL)
919 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
920 
921 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
922 			if (ia == NULL) {
923 				error = ENETUNREACH;
924 				goto done;
925 			}
926 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
927 			ifa_free(&ia->ia_ifa);
928 			goto done;
929 		}
930 
931 		/* Jailed. */
932 		if (ia != NULL) {
933 			struct ifnet *ifp;
934 
935 			ifp = ia->ia_ifp;
936 			ifa_free(&ia->ia_ifa);
937 			ia = NULL;
938 			IF_ADDR_RLOCK(ifp);
939 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
940 
941 				sa = ifa->ifa_addr;
942 				if (sa->sa_family != AF_INET)
943 					continue;
944 				sin = (struct sockaddr_in *)sa;
945 				if (prison_check_ip4(cred,
946 				    &sin->sin_addr) == 0) {
947 					ia = (struct in_ifaddr *)ifa;
948 					break;
949 				}
950 			}
951 			if (ia != NULL) {
952 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
953 				IF_ADDR_RUNLOCK(ifp);
954 				goto done;
955 			}
956 			IF_ADDR_RUNLOCK(ifp);
957 		}
958 
959 		/* 3. As a last resort return the 'default' jail address. */
960 		error = prison_get_ip4(cred, laddr);
961 		goto done;
962 	}
963 
964 done:
965 	if (sro.ro_rt != NULL)
966 		RTFREE(sro.ro_rt);
967 	return (error);
968 }
969 
970 /*
971  * Set up for a connect from a socket to the specified address.
972  * On entry, *laddrp and *lportp should contain the current local
973  * address and port for the PCB; these are updated to the values
974  * that should be placed in inp_laddr and inp_lport to complete
975  * the connect.
976  *
977  * On success, *faddrp and *fportp will be set to the remote address
978  * and port. These are not updated in the error case.
979  *
980  * If the operation fails because the connection already exists,
981  * *oinpp will be set to the PCB of that connection so that the
982  * caller can decide to override it. In all other cases, *oinpp
983  * is set to NULL.
984  */
985 int
986 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
987     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
988     struct inpcb **oinpp, struct ucred *cred)
989 {
990 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
991 	struct in_ifaddr *ia;
992 	struct inpcb *oinp;
993 	struct in_addr laddr, faddr;
994 	u_short lport, fport;
995 	int error;
996 
997 	/*
998 	 * Because a global state change doesn't actually occur here, a read
999 	 * lock is sufficient.
1000 	 */
1001 	INP_LOCK_ASSERT(inp);
1002 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1003 
1004 	if (oinpp != NULL)
1005 		*oinpp = NULL;
1006 	if (nam->sa_len != sizeof (*sin))
1007 		return (EINVAL);
1008 	if (sin->sin_family != AF_INET)
1009 		return (EAFNOSUPPORT);
1010 	if (sin->sin_port == 0)
1011 		return (EADDRNOTAVAIL);
1012 	laddr.s_addr = *laddrp;
1013 	lport = *lportp;
1014 	faddr = sin->sin_addr;
1015 	fport = sin->sin_port;
1016 
1017 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1018 		/*
1019 		 * If the destination address is INADDR_ANY,
1020 		 * use the primary local address.
1021 		 * If the supplied address is INADDR_BROADCAST,
1022 		 * and the primary interface supports broadcast,
1023 		 * choose the broadcast address for that interface.
1024 		 */
1025 		if (faddr.s_addr == INADDR_ANY) {
1026 			IN_IFADDR_RLOCK();
1027 			faddr =
1028 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1029 			IN_IFADDR_RUNLOCK();
1030 			if (cred != NULL &&
1031 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1032 				return (error);
1033 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1034 			IN_IFADDR_RLOCK();
1035 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1036 			    IFF_BROADCAST)
1037 				faddr = satosin(&TAILQ_FIRST(
1038 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1039 			IN_IFADDR_RUNLOCK();
1040 		}
1041 	}
1042 	if (laddr.s_addr == INADDR_ANY) {
1043 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1044 		/*
1045 		 * If the destination address is multicast and an outgoing
1046 		 * interface has been set as a multicast option, prefer the
1047 		 * address of that interface as our source address.
1048 		 */
1049 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1050 		    inp->inp_moptions != NULL) {
1051 			struct ip_moptions *imo;
1052 			struct ifnet *ifp;
1053 
1054 			imo = inp->inp_moptions;
1055 			if (imo->imo_multicast_ifp != NULL) {
1056 				ifp = imo->imo_multicast_ifp;
1057 				IN_IFADDR_RLOCK();
1058 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1059 					if ((ia->ia_ifp == ifp) &&
1060 					    (cred == NULL ||
1061 					    prison_check_ip4(cred,
1062 					    &ia->ia_addr.sin_addr) == 0))
1063 						break;
1064 				}
1065 				if (ia == NULL)
1066 					error = EADDRNOTAVAIL;
1067 				else {
1068 					laddr = ia->ia_addr.sin_addr;
1069 					error = 0;
1070 				}
1071 				IN_IFADDR_RUNLOCK();
1072 			}
1073 		}
1074 		if (error)
1075 			return (error);
1076 	}
1077 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1078 	    laddr, lport, 0, NULL);
1079 	if (oinp != NULL) {
1080 		if (oinpp != NULL)
1081 			*oinpp = oinp;
1082 		return (EADDRINUSE);
1083 	}
1084 	if (lport == 0) {
1085 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1086 		    cred);
1087 		if (error)
1088 			return (error);
1089 	}
1090 	*laddrp = laddr.s_addr;
1091 	*lportp = lport;
1092 	*faddrp = faddr.s_addr;
1093 	*fportp = fport;
1094 	return (0);
1095 }
1096 
1097 void
1098 in_pcbdisconnect(struct inpcb *inp)
1099 {
1100 
1101 	INP_WLOCK_ASSERT(inp);
1102 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1103 
1104 	inp->inp_faddr.s_addr = INADDR_ANY;
1105 	inp->inp_fport = 0;
1106 	in_pcbrehash(inp);
1107 }
1108 #endif /* INET */
1109 
1110 /*
1111  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1112  * For most protocols, this will be invoked immediately prior to calling
1113  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1114  * socket, in which case in_pcbfree() is deferred.
1115  */
1116 void
1117 in_pcbdetach(struct inpcb *inp)
1118 {
1119 
1120 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1121 
1122 	inp->inp_socket->so_pcb = NULL;
1123 	inp->inp_socket = NULL;
1124 }
1125 
1126 /*
1127  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1128  * stability of an inpcb pointer despite the inpcb lock being released.  This
1129  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1130  * but where the inpcb lock may already held, or when acquiring a reference
1131  * via a pcbgroup.
1132  *
1133  * in_pcbref() should be used only to provide brief memory stability, and
1134  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1135  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1136  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1137  * lock and rele are the *only* safe operations that may be performed on the
1138  * inpcb.
1139  *
1140  * While the inpcb will not be freed, releasing the inpcb lock means that the
1141  * connection's state may change, so the caller should be careful to
1142  * revalidate any cached state on reacquiring the lock.  Drop the reference
1143  * using in_pcbrele().
1144  */
1145 void
1146 in_pcbref(struct inpcb *inp)
1147 {
1148 
1149 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1150 
1151 	refcount_acquire(&inp->inp_refcount);
1152 }
1153 
1154 /*
1155  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1156  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1157  * return a flag indicating whether or not the inpcb remains valid.  If it is
1158  * valid, we return with the inpcb lock held.
1159  *
1160  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1161  * reference on an inpcb.  Historically more work was done here (actually, in
1162  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1163  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1164  * about memory stability (and continued use of the write lock).
1165  */
1166 int
1167 in_pcbrele_rlocked(struct inpcb *inp)
1168 {
1169 	struct inpcbinfo *pcbinfo;
1170 
1171 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1172 
1173 	INP_RLOCK_ASSERT(inp);
1174 
1175 	if (refcount_release(&inp->inp_refcount) == 0) {
1176 		/*
1177 		 * If the inpcb has been freed, let the caller know, even if
1178 		 * this isn't the last reference.
1179 		 */
1180 		if (inp->inp_flags2 & INP_FREED) {
1181 			INP_RUNLOCK(inp);
1182 			return (1);
1183 		}
1184 		return (0);
1185 	}
1186 
1187 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1188 
1189 	INP_RUNLOCK(inp);
1190 	pcbinfo = inp->inp_pcbinfo;
1191 	uma_zfree(pcbinfo->ipi_zone, inp);
1192 	return (1);
1193 }
1194 
1195 int
1196 in_pcbrele_wlocked(struct inpcb *inp)
1197 {
1198 	struct inpcbinfo *pcbinfo;
1199 
1200 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1201 
1202 	INP_WLOCK_ASSERT(inp);
1203 
1204 	if (refcount_release(&inp->inp_refcount) == 0)
1205 		return (0);
1206 
1207 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1208 
1209 	INP_WUNLOCK(inp);
1210 	pcbinfo = inp->inp_pcbinfo;
1211 	uma_zfree(pcbinfo->ipi_zone, inp);
1212 	return (1);
1213 }
1214 
1215 /*
1216  * Temporary wrapper.
1217  */
1218 int
1219 in_pcbrele(struct inpcb *inp)
1220 {
1221 
1222 	return (in_pcbrele_wlocked(inp));
1223 }
1224 
1225 /*
1226  * Unconditionally schedule an inpcb to be freed by decrementing its
1227  * reference count, which should occur only after the inpcb has been detached
1228  * from its socket.  If another thread holds a temporary reference (acquired
1229  * using in_pcbref()) then the free is deferred until that reference is
1230  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1231  * work, including removal from global lists, is done in this context, where
1232  * the pcbinfo lock is held.
1233  */
1234 void
1235 in_pcbfree(struct inpcb *inp)
1236 {
1237 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1238 
1239 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1240 
1241 	INP_INFO_WLOCK_ASSERT(pcbinfo);
1242 	INP_WLOCK_ASSERT(inp);
1243 
1244 	/* XXXRW: Do as much as possible here. */
1245 #ifdef IPSEC
1246 	if (inp->inp_sp != NULL)
1247 		ipsec_delete_pcbpolicy(inp);
1248 #endif
1249 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1250 	in_pcbremlists(inp);
1251 #ifdef INET6
1252 	if (inp->inp_vflag & INP_IPV6PROTO) {
1253 		ip6_freepcbopts(inp->in6p_outputopts);
1254 		if (inp->in6p_moptions != NULL)
1255 			ip6_freemoptions(inp->in6p_moptions);
1256 	}
1257 #endif
1258 	if (inp->inp_options)
1259 		(void)m_free(inp->inp_options);
1260 #ifdef INET
1261 	if (inp->inp_moptions != NULL)
1262 		inp_freemoptions(inp->inp_moptions);
1263 #endif
1264 	inp->inp_vflag = 0;
1265 	inp->inp_flags2 |= INP_FREED;
1266 	crfree(inp->inp_cred);
1267 #ifdef MAC
1268 	mac_inpcb_destroy(inp);
1269 #endif
1270 	if (!in_pcbrele_wlocked(inp))
1271 		INP_WUNLOCK(inp);
1272 }
1273 
1274 /*
1275  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1276  * port reservation, and preventing it from being returned by inpcb lookups.
1277  *
1278  * It is used by TCP to mark an inpcb as unused and avoid future packet
1279  * delivery or event notification when a socket remains open but TCP has
1280  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1281  * or a RST on the wire, and allows the port binding to be reused while still
1282  * maintaining the invariant that so_pcb always points to a valid inpcb until
1283  * in_pcbdetach().
1284  *
1285  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1286  * in_pcbnotifyall() and in_pcbpurgeif0()?
1287  */
1288 void
1289 in_pcbdrop(struct inpcb *inp)
1290 {
1291 
1292 	INP_WLOCK_ASSERT(inp);
1293 
1294 	/*
1295 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1296 	 * the hash lock...?
1297 	 */
1298 	inp->inp_flags |= INP_DROPPED;
1299 	if (inp->inp_flags & INP_INHASHLIST) {
1300 		struct inpcbport *phd = inp->inp_phd;
1301 
1302 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1303 		LIST_REMOVE(inp, inp_hash);
1304 		LIST_REMOVE(inp, inp_portlist);
1305 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1306 			LIST_REMOVE(phd, phd_hash);
1307 			free(phd, M_PCB);
1308 		}
1309 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1310 		inp->inp_flags &= ~INP_INHASHLIST;
1311 #ifdef PCBGROUP
1312 		in_pcbgroup_remove(inp);
1313 #endif
1314 	}
1315 }
1316 
1317 #ifdef INET
1318 /*
1319  * Common routines to return the socket addresses associated with inpcbs.
1320  */
1321 struct sockaddr *
1322 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1323 {
1324 	struct sockaddr_in *sin;
1325 
1326 	sin = malloc(sizeof *sin, M_SONAME,
1327 		M_WAITOK | M_ZERO);
1328 	sin->sin_family = AF_INET;
1329 	sin->sin_len = sizeof(*sin);
1330 	sin->sin_addr = *addr_p;
1331 	sin->sin_port = port;
1332 
1333 	return (struct sockaddr *)sin;
1334 }
1335 
1336 int
1337 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1338 {
1339 	struct inpcb *inp;
1340 	struct in_addr addr;
1341 	in_port_t port;
1342 
1343 	inp = sotoinpcb(so);
1344 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1345 
1346 	INP_RLOCK(inp);
1347 	port = inp->inp_lport;
1348 	addr = inp->inp_laddr;
1349 	INP_RUNLOCK(inp);
1350 
1351 	*nam = in_sockaddr(port, &addr);
1352 	return 0;
1353 }
1354 
1355 int
1356 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1357 {
1358 	struct inpcb *inp;
1359 	struct in_addr addr;
1360 	in_port_t port;
1361 
1362 	inp = sotoinpcb(so);
1363 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1364 
1365 	INP_RLOCK(inp);
1366 	port = inp->inp_fport;
1367 	addr = inp->inp_faddr;
1368 	INP_RUNLOCK(inp);
1369 
1370 	*nam = in_sockaddr(port, &addr);
1371 	return 0;
1372 }
1373 
1374 void
1375 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1376     struct inpcb *(*notify)(struct inpcb *, int))
1377 {
1378 	struct inpcb *inp, *inp_temp;
1379 
1380 	INP_INFO_WLOCK(pcbinfo);
1381 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1382 		INP_WLOCK(inp);
1383 #ifdef INET6
1384 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1385 			INP_WUNLOCK(inp);
1386 			continue;
1387 		}
1388 #endif
1389 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1390 		    inp->inp_socket == NULL) {
1391 			INP_WUNLOCK(inp);
1392 			continue;
1393 		}
1394 		if ((*notify)(inp, errno))
1395 			INP_WUNLOCK(inp);
1396 	}
1397 	INP_INFO_WUNLOCK(pcbinfo);
1398 }
1399 
1400 void
1401 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1402 {
1403 	struct inpcb *inp;
1404 	struct ip_moptions *imo;
1405 	int i, gap;
1406 
1407 	INP_INFO_RLOCK(pcbinfo);
1408 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1409 		INP_WLOCK(inp);
1410 		imo = inp->inp_moptions;
1411 		if ((inp->inp_vflag & INP_IPV4) &&
1412 		    imo != NULL) {
1413 			/*
1414 			 * Unselect the outgoing interface if it is being
1415 			 * detached.
1416 			 */
1417 			if (imo->imo_multicast_ifp == ifp)
1418 				imo->imo_multicast_ifp = NULL;
1419 
1420 			/*
1421 			 * Drop multicast group membership if we joined
1422 			 * through the interface being detached.
1423 			 */
1424 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1425 			    i++) {
1426 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1427 					in_delmulti(imo->imo_membership[i]);
1428 					gap++;
1429 				} else if (gap != 0)
1430 					imo->imo_membership[i - gap] =
1431 					    imo->imo_membership[i];
1432 			}
1433 			imo->imo_num_memberships -= gap;
1434 		}
1435 		INP_WUNLOCK(inp);
1436 	}
1437 	INP_INFO_RUNLOCK(pcbinfo);
1438 }
1439 
1440 /*
1441  * Lookup a PCB based on the local address and port.  Caller must hold the
1442  * hash lock.  No inpcb locks or references are acquired.
1443  */
1444 #define INP_LOOKUP_MAPPED_PCB_COST	3
1445 struct inpcb *
1446 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1447     u_short lport, int lookupflags, struct ucred *cred)
1448 {
1449 	struct inpcb *inp;
1450 #ifdef INET6
1451 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1452 #else
1453 	int matchwild = 3;
1454 #endif
1455 	int wildcard;
1456 
1457 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1458 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1459 
1460 	INP_HASH_LOCK_ASSERT(pcbinfo);
1461 
1462 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1463 		struct inpcbhead *head;
1464 		/*
1465 		 * Look for an unconnected (wildcard foreign addr) PCB that
1466 		 * matches the local address and port we're looking for.
1467 		 */
1468 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1469 		    0, pcbinfo->ipi_hashmask)];
1470 		LIST_FOREACH(inp, head, inp_hash) {
1471 #ifdef INET6
1472 			/* XXX inp locking */
1473 			if ((inp->inp_vflag & INP_IPV4) == 0)
1474 				continue;
1475 #endif
1476 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1477 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1478 			    inp->inp_lport == lport) {
1479 				/*
1480 				 * Found?
1481 				 */
1482 				if (cred == NULL ||
1483 				    prison_equal_ip4(cred->cr_prison,
1484 					inp->inp_cred->cr_prison))
1485 					return (inp);
1486 			}
1487 		}
1488 		/*
1489 		 * Not found.
1490 		 */
1491 		return (NULL);
1492 	} else {
1493 		struct inpcbporthead *porthash;
1494 		struct inpcbport *phd;
1495 		struct inpcb *match = NULL;
1496 		/*
1497 		 * Best fit PCB lookup.
1498 		 *
1499 		 * First see if this local port is in use by looking on the
1500 		 * port hash list.
1501 		 */
1502 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1503 		    pcbinfo->ipi_porthashmask)];
1504 		LIST_FOREACH(phd, porthash, phd_hash) {
1505 			if (phd->phd_port == lport)
1506 				break;
1507 		}
1508 		if (phd != NULL) {
1509 			/*
1510 			 * Port is in use by one or more PCBs. Look for best
1511 			 * fit.
1512 			 */
1513 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1514 				wildcard = 0;
1515 				if (cred != NULL &&
1516 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1517 					cred->cr_prison))
1518 					continue;
1519 #ifdef INET6
1520 				/* XXX inp locking */
1521 				if ((inp->inp_vflag & INP_IPV4) == 0)
1522 					continue;
1523 				/*
1524 				 * We never select the PCB that has
1525 				 * INP_IPV6 flag and is bound to :: if
1526 				 * we have another PCB which is bound
1527 				 * to 0.0.0.0.  If a PCB has the
1528 				 * INP_IPV6 flag, then we set its cost
1529 				 * higher than IPv4 only PCBs.
1530 				 *
1531 				 * Note that the case only happens
1532 				 * when a socket is bound to ::, under
1533 				 * the condition that the use of the
1534 				 * mapped address is allowed.
1535 				 */
1536 				if ((inp->inp_vflag & INP_IPV6) != 0)
1537 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1538 #endif
1539 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1540 					wildcard++;
1541 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1542 					if (laddr.s_addr == INADDR_ANY)
1543 						wildcard++;
1544 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1545 						continue;
1546 				} else {
1547 					if (laddr.s_addr != INADDR_ANY)
1548 						wildcard++;
1549 				}
1550 				if (wildcard < matchwild) {
1551 					match = inp;
1552 					matchwild = wildcard;
1553 					if (matchwild == 0)
1554 						break;
1555 				}
1556 			}
1557 		}
1558 		return (match);
1559 	}
1560 }
1561 #undef INP_LOOKUP_MAPPED_PCB_COST
1562 
1563 #ifdef PCBGROUP
1564 /*
1565  * Lookup PCB in hash list, using pcbgroup tables.
1566  */
1567 static struct inpcb *
1568 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1569     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1570     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1571 {
1572 	struct inpcbhead *head;
1573 	struct inpcb *inp, *tmpinp;
1574 	u_short fport = fport_arg, lport = lport_arg;
1575 
1576 	/*
1577 	 * First look for an exact match.
1578 	 */
1579 	tmpinp = NULL;
1580 	INP_GROUP_LOCK(pcbgroup);
1581 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1582 	    pcbgroup->ipg_hashmask)];
1583 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1584 #ifdef INET6
1585 		/* XXX inp locking */
1586 		if ((inp->inp_vflag & INP_IPV4) == 0)
1587 			continue;
1588 #endif
1589 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1590 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1591 		    inp->inp_fport == fport &&
1592 		    inp->inp_lport == lport) {
1593 			/*
1594 			 * XXX We should be able to directly return
1595 			 * the inp here, without any checks.
1596 			 * Well unless both bound with SO_REUSEPORT?
1597 			 */
1598 			if (prison_flag(inp->inp_cred, PR_IP4))
1599 				goto found;
1600 			if (tmpinp == NULL)
1601 				tmpinp = inp;
1602 		}
1603 	}
1604 	if (tmpinp != NULL) {
1605 		inp = tmpinp;
1606 		goto found;
1607 	}
1608 
1609 #ifdef	RSS
1610 	/*
1611 	 * For incoming connections, we may wish to do a wildcard
1612 	 * match for an RSS-local socket.
1613 	 */
1614 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1615 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1616 #ifdef INET6
1617 		struct inpcb *local_wild_mapped = NULL;
1618 #endif
1619 		struct inpcb *jail_wild = NULL;
1620 		struct inpcbhead *head;
1621 		int injail;
1622 
1623 		/*
1624 		 * Order of socket selection - we always prefer jails.
1625 		 *      1. jailed, non-wild.
1626 		 *      2. jailed, wild.
1627 		 *      3. non-jailed, non-wild.
1628 		 *      4. non-jailed, wild.
1629 		 */
1630 
1631 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1632 		    lport, 0, pcbgroup->ipg_hashmask)];
1633 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1634 #ifdef INET6
1635 			/* XXX inp locking */
1636 			if ((inp->inp_vflag & INP_IPV4) == 0)
1637 				continue;
1638 #endif
1639 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1640 			    inp->inp_lport != lport)
1641 				continue;
1642 
1643 			/* XXX inp locking */
1644 			if (ifp && ifp->if_type == IFT_FAITH &&
1645 			    (inp->inp_flags & INP_FAITH) == 0)
1646 				continue;
1647 
1648 			injail = prison_flag(inp->inp_cred, PR_IP4);
1649 			if (injail) {
1650 				if (prison_check_ip4(inp->inp_cred,
1651 				    &laddr) != 0)
1652 					continue;
1653 			} else {
1654 				if (local_exact != NULL)
1655 					continue;
1656 			}
1657 
1658 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1659 				if (injail)
1660 					goto found;
1661 				else
1662 					local_exact = inp;
1663 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1664 #ifdef INET6
1665 				/* XXX inp locking, NULL check */
1666 				if (inp->inp_vflag & INP_IPV6PROTO)
1667 					local_wild_mapped = inp;
1668 				else
1669 #endif
1670 					if (injail)
1671 						jail_wild = inp;
1672 					else
1673 						local_wild = inp;
1674 			}
1675 		} /* LIST_FOREACH */
1676 
1677 		inp = jail_wild;
1678 		if (inp == NULL)
1679 			inp = local_exact;
1680 		if (inp == NULL)
1681 			inp = local_wild;
1682 #ifdef INET6
1683 		if (inp == NULL)
1684 			inp = local_wild_mapped;
1685 #endif
1686 		if (inp != NULL)
1687 			goto found;
1688 	}
1689 #endif
1690 
1691 	/*
1692 	 * Then look for a wildcard match, if requested.
1693 	 */
1694 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1695 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1696 #ifdef INET6
1697 		struct inpcb *local_wild_mapped = NULL;
1698 #endif
1699 		struct inpcb *jail_wild = NULL;
1700 		struct inpcbhead *head;
1701 		int injail;
1702 
1703 		/*
1704 		 * Order of socket selection - we always prefer jails.
1705 		 *      1. jailed, non-wild.
1706 		 *      2. jailed, wild.
1707 		 *      3. non-jailed, non-wild.
1708 		 *      4. non-jailed, wild.
1709 		 */
1710 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1711 		    0, pcbinfo->ipi_wildmask)];
1712 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1713 #ifdef INET6
1714 			/* XXX inp locking */
1715 			if ((inp->inp_vflag & INP_IPV4) == 0)
1716 				continue;
1717 #endif
1718 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1719 			    inp->inp_lport != lport)
1720 				continue;
1721 
1722 			/* XXX inp locking */
1723 			if (ifp && ifp->if_type == IFT_FAITH &&
1724 			    (inp->inp_flags & INP_FAITH) == 0)
1725 				continue;
1726 
1727 			injail = prison_flag(inp->inp_cred, PR_IP4);
1728 			if (injail) {
1729 				if (prison_check_ip4(inp->inp_cred,
1730 				    &laddr) != 0)
1731 					continue;
1732 			} else {
1733 				if (local_exact != NULL)
1734 					continue;
1735 			}
1736 
1737 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1738 				if (injail)
1739 					goto found;
1740 				else
1741 					local_exact = inp;
1742 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1743 #ifdef INET6
1744 				/* XXX inp locking, NULL check */
1745 				if (inp->inp_vflag & INP_IPV6PROTO)
1746 					local_wild_mapped = inp;
1747 				else
1748 #endif
1749 					if (injail)
1750 						jail_wild = inp;
1751 					else
1752 						local_wild = inp;
1753 			}
1754 		} /* LIST_FOREACH */
1755 		inp = jail_wild;
1756 		if (inp == NULL)
1757 			inp = local_exact;
1758 		if (inp == NULL)
1759 			inp = local_wild;
1760 #ifdef INET6
1761 		if (inp == NULL)
1762 			inp = local_wild_mapped;
1763 #endif
1764 		if (inp != NULL)
1765 			goto found;
1766 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1767 	INP_GROUP_UNLOCK(pcbgroup);
1768 	return (NULL);
1769 
1770 found:
1771 	in_pcbref(inp);
1772 	INP_GROUP_UNLOCK(pcbgroup);
1773 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1774 		INP_WLOCK(inp);
1775 		if (in_pcbrele_wlocked(inp))
1776 			return (NULL);
1777 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1778 		INP_RLOCK(inp);
1779 		if (in_pcbrele_rlocked(inp))
1780 			return (NULL);
1781 	} else
1782 		panic("%s: locking bug", __func__);
1783 	return (inp);
1784 }
1785 #endif /* PCBGROUP */
1786 
1787 /*
1788  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1789  * that the caller has locked the hash list, and will not perform any further
1790  * locking or reference operations on either the hash list or the connection.
1791  */
1792 static struct inpcb *
1793 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1794     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1795     struct ifnet *ifp)
1796 {
1797 	struct inpcbhead *head;
1798 	struct inpcb *inp, *tmpinp;
1799 	u_short fport = fport_arg, lport = lport_arg;
1800 
1801 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1802 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1803 
1804 	INP_HASH_LOCK_ASSERT(pcbinfo);
1805 
1806 	/*
1807 	 * First look for an exact match.
1808 	 */
1809 	tmpinp = NULL;
1810 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1811 	    pcbinfo->ipi_hashmask)];
1812 	LIST_FOREACH(inp, head, inp_hash) {
1813 #ifdef INET6
1814 		/* XXX inp locking */
1815 		if ((inp->inp_vflag & INP_IPV4) == 0)
1816 			continue;
1817 #endif
1818 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1819 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1820 		    inp->inp_fport == fport &&
1821 		    inp->inp_lport == lport) {
1822 			/*
1823 			 * XXX We should be able to directly return
1824 			 * the inp here, without any checks.
1825 			 * Well unless both bound with SO_REUSEPORT?
1826 			 */
1827 			if (prison_flag(inp->inp_cred, PR_IP4))
1828 				return (inp);
1829 			if (tmpinp == NULL)
1830 				tmpinp = inp;
1831 		}
1832 	}
1833 	if (tmpinp != NULL)
1834 		return (tmpinp);
1835 
1836 	/*
1837 	 * Then look for a wildcard match, if requested.
1838 	 */
1839 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1840 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1841 #ifdef INET6
1842 		struct inpcb *local_wild_mapped = NULL;
1843 #endif
1844 		struct inpcb *jail_wild = NULL;
1845 		int injail;
1846 
1847 		/*
1848 		 * Order of socket selection - we always prefer jails.
1849 		 *      1. jailed, non-wild.
1850 		 *      2. jailed, wild.
1851 		 *      3. non-jailed, non-wild.
1852 		 *      4. non-jailed, wild.
1853 		 */
1854 
1855 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1856 		    0, pcbinfo->ipi_hashmask)];
1857 		LIST_FOREACH(inp, head, inp_hash) {
1858 #ifdef INET6
1859 			/* XXX inp locking */
1860 			if ((inp->inp_vflag & INP_IPV4) == 0)
1861 				continue;
1862 #endif
1863 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1864 			    inp->inp_lport != lport)
1865 				continue;
1866 
1867 			/* XXX inp locking */
1868 			if (ifp && ifp->if_type == IFT_FAITH &&
1869 			    (inp->inp_flags & INP_FAITH) == 0)
1870 				continue;
1871 
1872 			injail = prison_flag(inp->inp_cred, PR_IP4);
1873 			if (injail) {
1874 				if (prison_check_ip4(inp->inp_cred,
1875 				    &laddr) != 0)
1876 					continue;
1877 			} else {
1878 				if (local_exact != NULL)
1879 					continue;
1880 			}
1881 
1882 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1883 				if (injail)
1884 					return (inp);
1885 				else
1886 					local_exact = inp;
1887 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1888 #ifdef INET6
1889 				/* XXX inp locking, NULL check */
1890 				if (inp->inp_vflag & INP_IPV6PROTO)
1891 					local_wild_mapped = inp;
1892 				else
1893 #endif
1894 					if (injail)
1895 						jail_wild = inp;
1896 					else
1897 						local_wild = inp;
1898 			}
1899 		} /* LIST_FOREACH */
1900 		if (jail_wild != NULL)
1901 			return (jail_wild);
1902 		if (local_exact != NULL)
1903 			return (local_exact);
1904 		if (local_wild != NULL)
1905 			return (local_wild);
1906 #ifdef INET6
1907 		if (local_wild_mapped != NULL)
1908 			return (local_wild_mapped);
1909 #endif
1910 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1911 
1912 	return (NULL);
1913 }
1914 
1915 /*
1916  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1917  * hash list lock, and will return the inpcb locked (i.e., requires
1918  * INPLOOKUP_LOCKPCB).
1919  */
1920 static struct inpcb *
1921 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1922     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1923     struct ifnet *ifp)
1924 {
1925 	struct inpcb *inp;
1926 
1927 	INP_HASH_RLOCK(pcbinfo);
1928 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1929 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1930 	if (inp != NULL) {
1931 		in_pcbref(inp);
1932 		INP_HASH_RUNLOCK(pcbinfo);
1933 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1934 			INP_WLOCK(inp);
1935 			if (in_pcbrele_wlocked(inp))
1936 				return (NULL);
1937 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1938 			INP_RLOCK(inp);
1939 			if (in_pcbrele_rlocked(inp))
1940 				return (NULL);
1941 		} else
1942 			panic("%s: locking bug", __func__);
1943 	} else
1944 		INP_HASH_RUNLOCK(pcbinfo);
1945 	return (inp);
1946 }
1947 
1948 /*
1949  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1950  * from which a pre-calculated hash value may be extracted.
1951  *
1952  * Possibly more of this logic should be in in_pcbgroup.c.
1953  */
1954 struct inpcb *
1955 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1956     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1957 {
1958 #if defined(PCBGROUP) && !defined(RSS)
1959 	struct inpcbgroup *pcbgroup;
1960 #endif
1961 
1962 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1963 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1964 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1965 	    ("%s: LOCKPCB not set", __func__));
1966 
1967 	/*
1968 	 * When not using RSS, use connection groups in preference to the
1969 	 * reservation table when looking up 4-tuples.  When using RSS, just
1970 	 * use the reservation table, due to the cost of the Toeplitz hash
1971 	 * in software.
1972 	 *
1973 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
1974 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
1975 	 * in software.
1976 	 */
1977 #if defined(PCBGROUP) && !defined(RSS)
1978 	if (in_pcbgroup_enabled(pcbinfo)) {
1979 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1980 		    fport);
1981 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1982 		    laddr, lport, lookupflags, ifp));
1983 	}
1984 #endif
1985 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1986 	    lookupflags, ifp));
1987 }
1988 
1989 struct inpcb *
1990 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1991     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1992     struct ifnet *ifp, struct mbuf *m)
1993 {
1994 #ifdef PCBGROUP
1995 	struct inpcbgroup *pcbgroup;
1996 #endif
1997 
1998 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1999 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2000 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2001 	    ("%s: LOCKPCB not set", __func__));
2002 
2003 #ifdef PCBGROUP
2004 	/*
2005 	 * If we can use a hardware-generated hash to look up the connection
2006 	 * group, use that connection group to find the inpcb.  Otherwise
2007 	 * fall back on a software hash -- or the reservation table if we're
2008 	 * using RSS.
2009 	 *
2010 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2011 	 */
2012 	if (in_pcbgroup_enabled(pcbinfo) &&
2013 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2014 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2015 		    m->m_pkthdr.flowid);
2016 		if (pcbgroup != NULL)
2017 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2018 			    fport, laddr, lport, lookupflags, ifp));
2019 #ifndef RSS
2020 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2021 		    fport);
2022 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2023 		    laddr, lport, lookupflags, ifp));
2024 #endif
2025 	}
2026 #endif
2027 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2028 	    lookupflags, ifp));
2029 }
2030 #endif /* INET */
2031 
2032 /*
2033  * Insert PCB onto various hash lists.
2034  */
2035 static int
2036 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2037 {
2038 	struct inpcbhead *pcbhash;
2039 	struct inpcbporthead *pcbporthash;
2040 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2041 	struct inpcbport *phd;
2042 	u_int32_t hashkey_faddr;
2043 
2044 	INP_WLOCK_ASSERT(inp);
2045 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2046 
2047 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2048 	    ("in_pcbinshash: INP_INHASHLIST"));
2049 
2050 #ifdef INET6
2051 	if (inp->inp_vflag & INP_IPV6)
2052 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2053 	else
2054 #endif
2055 	hashkey_faddr = inp->inp_faddr.s_addr;
2056 
2057 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2058 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2059 
2060 	pcbporthash = &pcbinfo->ipi_porthashbase[
2061 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2062 
2063 	/*
2064 	 * Go through port list and look for a head for this lport.
2065 	 */
2066 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2067 		if (phd->phd_port == inp->inp_lport)
2068 			break;
2069 	}
2070 	/*
2071 	 * If none exists, malloc one and tack it on.
2072 	 */
2073 	if (phd == NULL) {
2074 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2075 		if (phd == NULL) {
2076 			return (ENOBUFS); /* XXX */
2077 		}
2078 		phd->phd_port = inp->inp_lport;
2079 		LIST_INIT(&phd->phd_pcblist);
2080 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2081 	}
2082 	inp->inp_phd = phd;
2083 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2084 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2085 	inp->inp_flags |= INP_INHASHLIST;
2086 #ifdef PCBGROUP
2087 	if (do_pcbgroup_update)
2088 		in_pcbgroup_update(inp);
2089 #endif
2090 	return (0);
2091 }
2092 
2093 /*
2094  * For now, there are two public interfaces to insert an inpcb into the hash
2095  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2096  * is used only in the TCP syncache, where in_pcbinshash is called before the
2097  * full 4-tuple is set for the inpcb, and we don't want to install in the
2098  * pcbgroup until later.
2099  *
2100  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2101  * connection groups, and partially initialised inpcbs should not be exposed
2102  * to either reservation hash tables or pcbgroups.
2103  */
2104 int
2105 in_pcbinshash(struct inpcb *inp)
2106 {
2107 
2108 	return (in_pcbinshash_internal(inp, 1));
2109 }
2110 
2111 int
2112 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2113 {
2114 
2115 	return (in_pcbinshash_internal(inp, 0));
2116 }
2117 
2118 /*
2119  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2120  * changed. NOTE: This does not handle the case of the lport changing (the
2121  * hashed port list would have to be updated as well), so the lport must
2122  * not change after in_pcbinshash() has been called.
2123  */
2124 void
2125 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2126 {
2127 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2128 	struct inpcbhead *head;
2129 	u_int32_t hashkey_faddr;
2130 
2131 	INP_WLOCK_ASSERT(inp);
2132 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2133 
2134 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2135 	    ("in_pcbrehash: !INP_INHASHLIST"));
2136 
2137 #ifdef INET6
2138 	if (inp->inp_vflag & INP_IPV6)
2139 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2140 	else
2141 #endif
2142 	hashkey_faddr = inp->inp_faddr.s_addr;
2143 
2144 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2145 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2146 
2147 	LIST_REMOVE(inp, inp_hash);
2148 	LIST_INSERT_HEAD(head, inp, inp_hash);
2149 
2150 #ifdef PCBGROUP
2151 	if (m != NULL)
2152 		in_pcbgroup_update_mbuf(inp, m);
2153 	else
2154 		in_pcbgroup_update(inp);
2155 #endif
2156 }
2157 
2158 void
2159 in_pcbrehash(struct inpcb *inp)
2160 {
2161 
2162 	in_pcbrehash_mbuf(inp, NULL);
2163 }
2164 
2165 /*
2166  * Remove PCB from various lists.
2167  */
2168 static void
2169 in_pcbremlists(struct inpcb *inp)
2170 {
2171 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2172 
2173 	INP_INFO_WLOCK_ASSERT(pcbinfo);
2174 	INP_WLOCK_ASSERT(inp);
2175 
2176 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2177 	if (inp->inp_flags & INP_INHASHLIST) {
2178 		struct inpcbport *phd = inp->inp_phd;
2179 
2180 		INP_HASH_WLOCK(pcbinfo);
2181 		LIST_REMOVE(inp, inp_hash);
2182 		LIST_REMOVE(inp, inp_portlist);
2183 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2184 			LIST_REMOVE(phd, phd_hash);
2185 			free(phd, M_PCB);
2186 		}
2187 		INP_HASH_WUNLOCK(pcbinfo);
2188 		inp->inp_flags &= ~INP_INHASHLIST;
2189 	}
2190 	LIST_REMOVE(inp, inp_list);
2191 	pcbinfo->ipi_count--;
2192 #ifdef PCBGROUP
2193 	in_pcbgroup_remove(inp);
2194 #endif
2195 }
2196 
2197 /*
2198  * A set label operation has occurred at the socket layer, propagate the
2199  * label change into the in_pcb for the socket.
2200  */
2201 void
2202 in_pcbsosetlabel(struct socket *so)
2203 {
2204 #ifdef MAC
2205 	struct inpcb *inp;
2206 
2207 	inp = sotoinpcb(so);
2208 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2209 
2210 	INP_WLOCK(inp);
2211 	SOCK_LOCK(so);
2212 	mac_inpcb_sosetlabel(so, inp);
2213 	SOCK_UNLOCK(so);
2214 	INP_WUNLOCK(inp);
2215 #endif
2216 }
2217 
2218 /*
2219  * ipport_tick runs once per second, determining if random port allocation
2220  * should be continued.  If more than ipport_randomcps ports have been
2221  * allocated in the last second, then we return to sequential port
2222  * allocation. We return to random allocation only once we drop below
2223  * ipport_randomcps for at least ipport_randomtime seconds.
2224  */
2225 static void
2226 ipport_tick(void *xtp)
2227 {
2228 	VNET_ITERATOR_DECL(vnet_iter);
2229 
2230 	VNET_LIST_RLOCK_NOSLEEP();
2231 	VNET_FOREACH(vnet_iter) {
2232 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2233 		if (V_ipport_tcpallocs <=
2234 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2235 			if (V_ipport_stoprandom > 0)
2236 				V_ipport_stoprandom--;
2237 		} else
2238 			V_ipport_stoprandom = V_ipport_randomtime;
2239 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2240 		CURVNET_RESTORE();
2241 	}
2242 	VNET_LIST_RUNLOCK_NOSLEEP();
2243 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2244 }
2245 
2246 static void
2247 ip_fini(void *xtp)
2248 {
2249 
2250 	callout_stop(&ipport_tick_callout);
2251 }
2252 
2253 /*
2254  * The ipport_callout should start running at about the time we attach the
2255  * inet or inet6 domains.
2256  */
2257 static void
2258 ipport_tick_init(const void *unused __unused)
2259 {
2260 
2261 	/* Start ipport_tick. */
2262 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
2263 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2264 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2265 		SHUTDOWN_PRI_DEFAULT);
2266 }
2267 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2268     ipport_tick_init, NULL);
2269 
2270 void
2271 inp_wlock(struct inpcb *inp)
2272 {
2273 
2274 	INP_WLOCK(inp);
2275 }
2276 
2277 void
2278 inp_wunlock(struct inpcb *inp)
2279 {
2280 
2281 	INP_WUNLOCK(inp);
2282 }
2283 
2284 void
2285 inp_rlock(struct inpcb *inp)
2286 {
2287 
2288 	INP_RLOCK(inp);
2289 }
2290 
2291 void
2292 inp_runlock(struct inpcb *inp)
2293 {
2294 
2295 	INP_RUNLOCK(inp);
2296 }
2297 
2298 #ifdef INVARIANTS
2299 void
2300 inp_lock_assert(struct inpcb *inp)
2301 {
2302 
2303 	INP_WLOCK_ASSERT(inp);
2304 }
2305 
2306 void
2307 inp_unlock_assert(struct inpcb *inp)
2308 {
2309 
2310 	INP_UNLOCK_ASSERT(inp);
2311 }
2312 #endif
2313 
2314 void
2315 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2316 {
2317 	struct inpcb *inp;
2318 
2319 	INP_INFO_RLOCK(&V_tcbinfo);
2320 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2321 		INP_WLOCK(inp);
2322 		func(inp, arg);
2323 		INP_WUNLOCK(inp);
2324 	}
2325 	INP_INFO_RUNLOCK(&V_tcbinfo);
2326 }
2327 
2328 struct socket *
2329 inp_inpcbtosocket(struct inpcb *inp)
2330 {
2331 
2332 	INP_WLOCK_ASSERT(inp);
2333 	return (inp->inp_socket);
2334 }
2335 
2336 struct tcpcb *
2337 inp_inpcbtotcpcb(struct inpcb *inp)
2338 {
2339 
2340 	INP_WLOCK_ASSERT(inp);
2341 	return ((struct tcpcb *)inp->inp_ppcb);
2342 }
2343 
2344 int
2345 inp_ip_tos_get(const struct inpcb *inp)
2346 {
2347 
2348 	return (inp->inp_ip_tos);
2349 }
2350 
2351 void
2352 inp_ip_tos_set(struct inpcb *inp, int val)
2353 {
2354 
2355 	inp->inp_ip_tos = val;
2356 }
2357 
2358 void
2359 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2360     uint32_t *faddr, uint16_t *fp)
2361 {
2362 
2363 	INP_LOCK_ASSERT(inp);
2364 	*laddr = inp->inp_laddr.s_addr;
2365 	*faddr = inp->inp_faddr.s_addr;
2366 	*lp = inp->inp_lport;
2367 	*fp = inp->inp_fport;
2368 }
2369 
2370 struct inpcb *
2371 so_sotoinpcb(struct socket *so)
2372 {
2373 
2374 	return (sotoinpcb(so));
2375 }
2376 
2377 struct tcpcb *
2378 so_sototcpcb(struct socket *so)
2379 {
2380 
2381 	return (sototcpcb(so));
2382 }
2383 
2384 #ifdef DDB
2385 static void
2386 db_print_indent(int indent)
2387 {
2388 	int i;
2389 
2390 	for (i = 0; i < indent; i++)
2391 		db_printf(" ");
2392 }
2393 
2394 static void
2395 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2396 {
2397 	char faddr_str[48], laddr_str[48];
2398 
2399 	db_print_indent(indent);
2400 	db_printf("%s at %p\n", name, inc);
2401 
2402 	indent += 2;
2403 
2404 #ifdef INET6
2405 	if (inc->inc_flags & INC_ISIPV6) {
2406 		/* IPv6. */
2407 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2408 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2409 	} else
2410 #endif
2411 	{
2412 		/* IPv4. */
2413 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2414 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2415 	}
2416 	db_print_indent(indent);
2417 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2418 	    ntohs(inc->inc_lport));
2419 	db_print_indent(indent);
2420 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2421 	    ntohs(inc->inc_fport));
2422 }
2423 
2424 static void
2425 db_print_inpflags(int inp_flags)
2426 {
2427 	int comma;
2428 
2429 	comma = 0;
2430 	if (inp_flags & INP_RECVOPTS) {
2431 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2432 		comma = 1;
2433 	}
2434 	if (inp_flags & INP_RECVRETOPTS) {
2435 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2436 		comma = 1;
2437 	}
2438 	if (inp_flags & INP_RECVDSTADDR) {
2439 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2440 		comma = 1;
2441 	}
2442 	if (inp_flags & INP_HDRINCL) {
2443 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2444 		comma = 1;
2445 	}
2446 	if (inp_flags & INP_HIGHPORT) {
2447 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2448 		comma = 1;
2449 	}
2450 	if (inp_flags & INP_LOWPORT) {
2451 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2452 		comma = 1;
2453 	}
2454 	if (inp_flags & INP_ANONPORT) {
2455 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2456 		comma = 1;
2457 	}
2458 	if (inp_flags & INP_RECVIF) {
2459 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2460 		comma = 1;
2461 	}
2462 	if (inp_flags & INP_MTUDISC) {
2463 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2464 		comma = 1;
2465 	}
2466 	if (inp_flags & INP_FAITH) {
2467 		db_printf("%sINP_FAITH", comma ? ", " : "");
2468 		comma = 1;
2469 	}
2470 	if (inp_flags & INP_RECVTTL) {
2471 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2472 		comma = 1;
2473 	}
2474 	if (inp_flags & INP_DONTFRAG) {
2475 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2476 		comma = 1;
2477 	}
2478 	if (inp_flags & INP_RECVTOS) {
2479 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2480 		comma = 1;
2481 	}
2482 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2483 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2484 		comma = 1;
2485 	}
2486 	if (inp_flags & IN6P_PKTINFO) {
2487 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2488 		comma = 1;
2489 	}
2490 	if (inp_flags & IN6P_HOPLIMIT) {
2491 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2492 		comma = 1;
2493 	}
2494 	if (inp_flags & IN6P_HOPOPTS) {
2495 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2496 		comma = 1;
2497 	}
2498 	if (inp_flags & IN6P_DSTOPTS) {
2499 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2500 		comma = 1;
2501 	}
2502 	if (inp_flags & IN6P_RTHDR) {
2503 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2504 		comma = 1;
2505 	}
2506 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2507 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2508 		comma = 1;
2509 	}
2510 	if (inp_flags & IN6P_TCLASS) {
2511 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2512 		comma = 1;
2513 	}
2514 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2515 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2516 		comma = 1;
2517 	}
2518 	if (inp_flags & INP_TIMEWAIT) {
2519 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2520 		comma  = 1;
2521 	}
2522 	if (inp_flags & INP_ONESBCAST) {
2523 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2524 		comma  = 1;
2525 	}
2526 	if (inp_flags & INP_DROPPED) {
2527 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2528 		comma  = 1;
2529 	}
2530 	if (inp_flags & INP_SOCKREF) {
2531 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2532 		comma  = 1;
2533 	}
2534 	if (inp_flags & IN6P_RFC2292) {
2535 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2536 		comma = 1;
2537 	}
2538 	if (inp_flags & IN6P_MTU) {
2539 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2540 		comma = 1;
2541 	}
2542 }
2543 
2544 static void
2545 db_print_inpvflag(u_char inp_vflag)
2546 {
2547 	int comma;
2548 
2549 	comma = 0;
2550 	if (inp_vflag & INP_IPV4) {
2551 		db_printf("%sINP_IPV4", comma ? ", " : "");
2552 		comma  = 1;
2553 	}
2554 	if (inp_vflag & INP_IPV6) {
2555 		db_printf("%sINP_IPV6", comma ? ", " : "");
2556 		comma  = 1;
2557 	}
2558 	if (inp_vflag & INP_IPV6PROTO) {
2559 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2560 		comma  = 1;
2561 	}
2562 }
2563 
2564 static void
2565 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2566 {
2567 
2568 	db_print_indent(indent);
2569 	db_printf("%s at %p\n", name, inp);
2570 
2571 	indent += 2;
2572 
2573 	db_print_indent(indent);
2574 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2575 
2576 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2577 
2578 	db_print_indent(indent);
2579 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2580 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2581 
2582 	db_print_indent(indent);
2583 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2584 	   inp->inp_label, inp->inp_flags);
2585 	db_print_inpflags(inp->inp_flags);
2586 	db_printf(")\n");
2587 
2588 	db_print_indent(indent);
2589 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2590 	    inp->inp_vflag);
2591 	db_print_inpvflag(inp->inp_vflag);
2592 	db_printf(")\n");
2593 
2594 	db_print_indent(indent);
2595 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2596 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2597 
2598 	db_print_indent(indent);
2599 #ifdef INET6
2600 	if (inp->inp_vflag & INP_IPV6) {
2601 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2602 		    "in6p_moptions: %p\n", inp->in6p_options,
2603 		    inp->in6p_outputopts, inp->in6p_moptions);
2604 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2605 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2606 		    inp->in6p_hops);
2607 	} else
2608 #endif
2609 	{
2610 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2611 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2612 		    inp->inp_options, inp->inp_moptions);
2613 	}
2614 
2615 	db_print_indent(indent);
2616 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2617 	    (uintmax_t)inp->inp_gencnt);
2618 }
2619 
2620 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2621 {
2622 	struct inpcb *inp;
2623 
2624 	if (!have_addr) {
2625 		db_printf("usage: show inpcb <addr>\n");
2626 		return;
2627 	}
2628 	inp = (struct inpcb *)addr;
2629 
2630 	db_print_inpcb(inp, "inpcb", 0);
2631 }
2632 #endif /* DDB */
2633