xref: /freebsd/sys/netinet/in_pcb.c (revision 95d45410b5100e07f6f98450bcd841a8945d4726)
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 #if defined(INET) || defined(INET6)
353 int
354 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
355     struct ucred *cred, int lookupflags)
356 {
357 	struct inpcbinfo *pcbinfo;
358 	struct inpcb *tmpinp;
359 	unsigned short *lastport;
360 	int count, dorandom, error;
361 	u_short aux, first, last, lport;
362 #ifdef INET
363 	struct in_addr laddr;
364 #endif
365 
366 	pcbinfo = inp->inp_pcbinfo;
367 
368 	/*
369 	 * Because no actual state changes occur here, a global write lock on
370 	 * the pcbinfo isn't required.
371 	 */
372 	INP_LOCK_ASSERT(inp);
373 	INP_HASH_LOCK_ASSERT(pcbinfo);
374 
375 	if (inp->inp_flags & INP_HIGHPORT) {
376 		first = V_ipport_hifirstauto;	/* sysctl */
377 		last  = V_ipport_hilastauto;
378 		lastport = &pcbinfo->ipi_lasthi;
379 	} else if (inp->inp_flags & INP_LOWPORT) {
380 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
381 		if (error)
382 			return (error);
383 		first = V_ipport_lowfirstauto;	/* 1023 */
384 		last  = V_ipport_lowlastauto;	/* 600 */
385 		lastport = &pcbinfo->ipi_lastlow;
386 	} else {
387 		first = V_ipport_firstauto;	/* sysctl */
388 		last  = V_ipport_lastauto;
389 		lastport = &pcbinfo->ipi_lastport;
390 	}
391 	/*
392 	 * For UDP(-Lite), use random port allocation as long as the user
393 	 * allows it.  For TCP (and as of yet unknown) connections,
394 	 * use random port allocation only if the user allows it AND
395 	 * ipport_tick() allows it.
396 	 */
397 	if (V_ipport_randomized &&
398 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
399 		pcbinfo == &V_ulitecbinfo))
400 		dorandom = 1;
401 	else
402 		dorandom = 0;
403 	/*
404 	 * It makes no sense to do random port allocation if
405 	 * we have the only port available.
406 	 */
407 	if (first == last)
408 		dorandom = 0;
409 	/* Make sure to not include UDP(-Lite) packets in the count. */
410 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
411 		V_ipport_tcpallocs++;
412 	/*
413 	 * Instead of having two loops further down counting up or down
414 	 * make sure that first is always <= last and go with only one
415 	 * code path implementing all logic.
416 	 */
417 	if (first > last) {
418 		aux = first;
419 		first = last;
420 		last = aux;
421 	}
422 
423 #ifdef INET
424 	/* Make the compiler happy. */
425 	laddr.s_addr = 0;
426 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
427 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
428 		    __func__, inp));
429 		laddr = *laddrp;
430 	}
431 #endif
432 	tmpinp = NULL;	/* Make compiler happy. */
433 	lport = *lportp;
434 
435 	if (dorandom)
436 		*lastport = first + (arc4random() % (last - first));
437 
438 	count = last - first;
439 
440 	do {
441 		if (count-- < 0)	/* completely used? */
442 			return (EADDRNOTAVAIL);
443 		++*lastport;
444 		if (*lastport < first || *lastport > last)
445 			*lastport = first;
446 		lport = htons(*lastport);
447 
448 #ifdef INET6
449 		if ((inp->inp_vflag & INP_IPV6) != 0)
450 			tmpinp = in6_pcblookup_local(pcbinfo,
451 			    &inp->in6p_laddr, lport, lookupflags, cred);
452 #endif
453 #if defined(INET) && defined(INET6)
454 		else
455 #endif
456 #ifdef INET
457 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
458 			    lport, lookupflags, cred);
459 #endif
460 	} while (tmpinp != NULL);
461 
462 #ifdef INET
463 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
464 		laddrp->s_addr = laddr.s_addr;
465 #endif
466 	*lportp = lport;
467 
468 	return (0);
469 }
470 
471 /*
472  * Return cached socket options.
473  */
474 short
475 inp_so_options(const struct inpcb *inp)
476 {
477    short so_options;
478 
479    so_options = 0;
480 
481    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
482 	   so_options |= SO_REUSEPORT;
483    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
484 	   so_options |= SO_REUSEADDR;
485    return (so_options);
486 }
487 #endif /* INET || INET6 */
488 
489 /*
490  * Check if a new BINDMULTI socket is allowed to be created.
491  *
492  * ni points to the new inp.
493  * oi points to the exisitng inp.
494  *
495  * This checks whether the existing inp also has BINDMULTI and
496  * whether the credentials match.
497  */
498 int
499 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
500 {
501 	/* Check permissions match */
502 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
503 	    (ni->inp_cred->cr_uid !=
504 	    oi->inp_cred->cr_uid))
505 		return (0);
506 
507 	/* Check the existing inp has BINDMULTI set */
508 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
509 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
510 		return (0);
511 
512 	/*
513 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
514 	 * it is and it matches the checks.
515 	 */
516 	return (1);
517 }
518 
519 #ifdef INET
520 /*
521  * Set up a bind operation on a PCB, performing port allocation
522  * as required, but do not actually modify the PCB. Callers can
523  * either complete the bind by setting inp_laddr/inp_lport and
524  * calling in_pcbinshash(), or they can just use the resulting
525  * port and address to authorise the sending of a once-off packet.
526  *
527  * On error, the values of *laddrp and *lportp are not changed.
528  */
529 int
530 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
531     u_short *lportp, struct ucred *cred)
532 {
533 	struct socket *so = inp->inp_socket;
534 	struct sockaddr_in *sin;
535 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
536 	struct in_addr laddr;
537 	u_short lport = 0;
538 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
539 	int error;
540 
541 	/*
542 	 * No state changes, so read locks are sufficient here.
543 	 */
544 	INP_LOCK_ASSERT(inp);
545 	INP_HASH_LOCK_ASSERT(pcbinfo);
546 
547 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
548 		return (EADDRNOTAVAIL);
549 	laddr.s_addr = *laddrp;
550 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
551 		return (EINVAL);
552 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
553 		lookupflags = INPLOOKUP_WILDCARD;
554 	if (nam == NULL) {
555 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
556 			return (error);
557 	} else {
558 		sin = (struct sockaddr_in *)nam;
559 		if (nam->sa_len != sizeof (*sin))
560 			return (EINVAL);
561 #ifdef notdef
562 		/*
563 		 * We should check the family, but old programs
564 		 * incorrectly fail to initialize it.
565 		 */
566 		if (sin->sin_family != AF_INET)
567 			return (EAFNOSUPPORT);
568 #endif
569 		error = prison_local_ip4(cred, &sin->sin_addr);
570 		if (error)
571 			return (error);
572 		if (sin->sin_port != *lportp) {
573 			/* Don't allow the port to change. */
574 			if (*lportp != 0)
575 				return (EINVAL);
576 			lport = sin->sin_port;
577 		}
578 		/* NB: lport is left as 0 if the port isn't being changed. */
579 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
580 			/*
581 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
582 			 * allow complete duplication of binding if
583 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
584 			 * and a multicast address is bound on both
585 			 * new and duplicated sockets.
586 			 */
587 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
588 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
589 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
590 			sin->sin_port = 0;		/* yech... */
591 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
592 			/*
593 			 * Is the address a local IP address?
594 			 * If INP_BINDANY is set, then the socket may be bound
595 			 * to any endpoint address, local or not.
596 			 */
597 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
598 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
599 				return (EADDRNOTAVAIL);
600 		}
601 		laddr = sin->sin_addr;
602 		if (lport) {
603 			struct inpcb *t;
604 			struct tcptw *tw;
605 
606 			/* GROSS */
607 			if (ntohs(lport) <= V_ipport_reservedhigh &&
608 			    ntohs(lport) >= V_ipport_reservedlow &&
609 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
610 			    0))
611 				return (EACCES);
612 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
613 			    priv_check_cred(inp->inp_cred,
614 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
615 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
616 				    lport, INPLOOKUP_WILDCARD, cred);
617 	/*
618 	 * XXX
619 	 * This entire block sorely needs a rewrite.
620 	 */
621 				if (t &&
622 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
623 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
624 				    (so->so_type != SOCK_STREAM ||
625 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
626 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
627 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
628 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
629 				    (inp->inp_cred->cr_uid !=
630 				     t->inp_cred->cr_uid))
631 					return (EADDRINUSE);
632 
633 				/*
634 				 * If the socket is a BINDMULTI socket, then
635 				 * the credentials need to match and the
636 				 * original socket also has to have been bound
637 				 * with BINDMULTI.
638 				 */
639 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
640 					return (EADDRINUSE);
641 			}
642 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
643 			    lport, lookupflags, cred);
644 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
645 				/*
646 				 * XXXRW: If an incpb has had its timewait
647 				 * state recycled, we treat the address as
648 				 * being in use (for now).  This is better
649 				 * than a panic, but not desirable.
650 				 */
651 				tw = intotw(t);
652 				if (tw == NULL ||
653 				    (reuseport & tw->tw_so_options) == 0)
654 					return (EADDRINUSE);
655 			} else if (t &&
656 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
657 			    (reuseport & inp_so_options(t)) == 0) {
658 #ifdef INET6
659 				if (ntohl(sin->sin_addr.s_addr) !=
660 				    INADDR_ANY ||
661 				    ntohl(t->inp_laddr.s_addr) !=
662 				    INADDR_ANY ||
663 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
664 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
665 #endif
666 				return (EADDRINUSE);
667 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
668 					return (EADDRINUSE);
669 			}
670 		}
671 	}
672 	if (*lportp != 0)
673 		lport = *lportp;
674 	if (lport == 0) {
675 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
676 		if (error != 0)
677 			return (error);
678 
679 	}
680 	*laddrp = laddr.s_addr;
681 	*lportp = lport;
682 	return (0);
683 }
684 
685 /*
686  * Connect from a socket to a specified address.
687  * Both address and port must be specified in argument sin.
688  * If don't have a local address for this socket yet,
689  * then pick one.
690  */
691 int
692 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
693     struct ucred *cred, struct mbuf *m)
694 {
695 	u_short lport, fport;
696 	in_addr_t laddr, faddr;
697 	int anonport, error;
698 
699 	INP_WLOCK_ASSERT(inp);
700 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
701 
702 	lport = inp->inp_lport;
703 	laddr = inp->inp_laddr.s_addr;
704 	anonport = (lport == 0);
705 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
706 	    NULL, cred);
707 	if (error)
708 		return (error);
709 
710 	/* Do the initial binding of the local address if required. */
711 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
712 		inp->inp_lport = lport;
713 		inp->inp_laddr.s_addr = laddr;
714 		if (in_pcbinshash(inp) != 0) {
715 			inp->inp_laddr.s_addr = INADDR_ANY;
716 			inp->inp_lport = 0;
717 			return (EAGAIN);
718 		}
719 	}
720 
721 	/* Commit the remaining changes. */
722 	inp->inp_lport = lport;
723 	inp->inp_laddr.s_addr = laddr;
724 	inp->inp_faddr.s_addr = faddr;
725 	inp->inp_fport = fport;
726 	in_pcbrehash_mbuf(inp, m);
727 
728 	if (anonport)
729 		inp->inp_flags |= INP_ANONPORT;
730 	return (0);
731 }
732 
733 int
734 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
735 {
736 
737 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
738 }
739 
740 /*
741  * Do proper source address selection on an unbound socket in case
742  * of connect. Take jails into account as well.
743  */
744 int
745 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
746     struct ucred *cred)
747 {
748 	struct ifaddr *ifa;
749 	struct sockaddr *sa;
750 	struct sockaddr_in *sin;
751 	struct route sro;
752 	int error;
753 
754 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
755 
756 	/*
757 	 * Bypass source address selection and use the primary jail IP
758 	 * if requested.
759 	 */
760 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
761 		return (0);
762 
763 	error = 0;
764 	bzero(&sro, sizeof(sro));
765 
766 	sin = (struct sockaddr_in *)&sro.ro_dst;
767 	sin->sin_family = AF_INET;
768 	sin->sin_len = sizeof(struct sockaddr_in);
769 	sin->sin_addr.s_addr = faddr->s_addr;
770 
771 	/*
772 	 * If route is known our src addr is taken from the i/f,
773 	 * else punt.
774 	 *
775 	 * Find out route to destination.
776 	 */
777 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
778 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
779 
780 	/*
781 	 * If we found a route, use the address corresponding to
782 	 * the outgoing interface.
783 	 *
784 	 * Otherwise assume faddr is reachable on a directly connected
785 	 * network and try to find a corresponding interface to take
786 	 * the source address from.
787 	 */
788 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
789 		struct in_ifaddr *ia;
790 		struct ifnet *ifp;
791 
792 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin));
793 		if (ia == NULL)
794 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0));
795 		if (ia == NULL) {
796 			error = ENETUNREACH;
797 			goto done;
798 		}
799 
800 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
801 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
802 			ifa_free(&ia->ia_ifa);
803 			goto done;
804 		}
805 
806 		ifp = ia->ia_ifp;
807 		ifa_free(&ia->ia_ifa);
808 		ia = NULL;
809 		IF_ADDR_RLOCK(ifp);
810 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
811 
812 			sa = ifa->ifa_addr;
813 			if (sa->sa_family != AF_INET)
814 				continue;
815 			sin = (struct sockaddr_in *)sa;
816 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
817 				ia = (struct in_ifaddr *)ifa;
818 				break;
819 			}
820 		}
821 		if (ia != NULL) {
822 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
823 			IF_ADDR_RUNLOCK(ifp);
824 			goto done;
825 		}
826 		IF_ADDR_RUNLOCK(ifp);
827 
828 		/* 3. As a last resort return the 'default' jail address. */
829 		error = prison_get_ip4(cred, laddr);
830 		goto done;
831 	}
832 
833 	/*
834 	 * If the outgoing interface on the route found is not
835 	 * a loopback interface, use the address from that interface.
836 	 * In case of jails do those three steps:
837 	 * 1. check if the interface address belongs to the jail. If so use it.
838 	 * 2. check if we have any address on the outgoing interface
839 	 *    belonging to this jail. If so use it.
840 	 * 3. as a last resort return the 'default' jail address.
841 	 */
842 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
843 		struct in_ifaddr *ia;
844 		struct ifnet *ifp;
845 
846 		/* If not jailed, use the default returned. */
847 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
848 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
849 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
850 			goto done;
851 		}
852 
853 		/* Jailed. */
854 		/* 1. Check if the iface address belongs to the jail. */
855 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
856 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
857 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
858 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
859 			goto done;
860 		}
861 
862 		/*
863 		 * 2. Check if we have any address on the outgoing interface
864 		 *    belonging to this jail.
865 		 */
866 		ia = NULL;
867 		ifp = sro.ro_rt->rt_ifp;
868 		IF_ADDR_RLOCK(ifp);
869 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
870 			sa = ifa->ifa_addr;
871 			if (sa->sa_family != AF_INET)
872 				continue;
873 			sin = (struct sockaddr_in *)sa;
874 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
875 				ia = (struct in_ifaddr *)ifa;
876 				break;
877 			}
878 		}
879 		if (ia != NULL) {
880 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
881 			IF_ADDR_RUNLOCK(ifp);
882 			goto done;
883 		}
884 		IF_ADDR_RUNLOCK(ifp);
885 
886 		/* 3. As a last resort return the 'default' jail address. */
887 		error = prison_get_ip4(cred, laddr);
888 		goto done;
889 	}
890 
891 	/*
892 	 * The outgoing interface is marked with 'loopback net', so a route
893 	 * to ourselves is here.
894 	 * Try to find the interface of the destination address and then
895 	 * take the address from there. That interface is not necessarily
896 	 * a loopback interface.
897 	 * In case of jails, check that it is an address of the jail
898 	 * and if we cannot find, fall back to the 'default' jail address.
899 	 */
900 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
901 		struct sockaddr_in sain;
902 		struct in_ifaddr *ia;
903 
904 		bzero(&sain, sizeof(struct sockaddr_in));
905 		sain.sin_family = AF_INET;
906 		sain.sin_len = sizeof(struct sockaddr_in);
907 		sain.sin_addr.s_addr = faddr->s_addr;
908 
909 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain)));
910 		if (ia == NULL)
911 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0));
912 		if (ia == NULL)
913 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
914 
915 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
916 			if (ia == NULL) {
917 				error = ENETUNREACH;
918 				goto done;
919 			}
920 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
921 			ifa_free(&ia->ia_ifa);
922 			goto done;
923 		}
924 
925 		/* Jailed. */
926 		if (ia != NULL) {
927 			struct ifnet *ifp;
928 
929 			ifp = ia->ia_ifp;
930 			ifa_free(&ia->ia_ifa);
931 			ia = NULL;
932 			IF_ADDR_RLOCK(ifp);
933 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
934 
935 				sa = ifa->ifa_addr;
936 				if (sa->sa_family != AF_INET)
937 					continue;
938 				sin = (struct sockaddr_in *)sa;
939 				if (prison_check_ip4(cred,
940 				    &sin->sin_addr) == 0) {
941 					ia = (struct in_ifaddr *)ifa;
942 					break;
943 				}
944 			}
945 			if (ia != NULL) {
946 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
947 				IF_ADDR_RUNLOCK(ifp);
948 				goto done;
949 			}
950 			IF_ADDR_RUNLOCK(ifp);
951 		}
952 
953 		/* 3. As a last resort return the 'default' jail address. */
954 		error = prison_get_ip4(cred, laddr);
955 		goto done;
956 	}
957 
958 done:
959 	if (sro.ro_rt != NULL)
960 		RTFREE(sro.ro_rt);
961 	return (error);
962 }
963 
964 /*
965  * Set up for a connect from a socket to the specified address.
966  * On entry, *laddrp and *lportp should contain the current local
967  * address and port for the PCB; these are updated to the values
968  * that should be placed in inp_laddr and inp_lport to complete
969  * the connect.
970  *
971  * On success, *faddrp and *fportp will be set to the remote address
972  * and port. These are not updated in the error case.
973  *
974  * If the operation fails because the connection already exists,
975  * *oinpp will be set to the PCB of that connection so that the
976  * caller can decide to override it. In all other cases, *oinpp
977  * is set to NULL.
978  */
979 int
980 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
981     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
982     struct inpcb **oinpp, struct ucred *cred)
983 {
984 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
985 	struct in_ifaddr *ia;
986 	struct inpcb *oinp;
987 	struct in_addr laddr, faddr;
988 	u_short lport, fport;
989 	int error;
990 
991 	/*
992 	 * Because a global state change doesn't actually occur here, a read
993 	 * lock is sufficient.
994 	 */
995 	INP_LOCK_ASSERT(inp);
996 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
997 
998 	if (oinpp != NULL)
999 		*oinpp = NULL;
1000 	if (nam->sa_len != sizeof (*sin))
1001 		return (EINVAL);
1002 	if (sin->sin_family != AF_INET)
1003 		return (EAFNOSUPPORT);
1004 	if (sin->sin_port == 0)
1005 		return (EADDRNOTAVAIL);
1006 	laddr.s_addr = *laddrp;
1007 	lport = *lportp;
1008 	faddr = sin->sin_addr;
1009 	fport = sin->sin_port;
1010 
1011 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1012 		/*
1013 		 * If the destination address is INADDR_ANY,
1014 		 * use the primary local address.
1015 		 * If the supplied address is INADDR_BROADCAST,
1016 		 * and the primary interface supports broadcast,
1017 		 * choose the broadcast address for that interface.
1018 		 */
1019 		if (faddr.s_addr == INADDR_ANY) {
1020 			IN_IFADDR_RLOCK();
1021 			faddr =
1022 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1023 			IN_IFADDR_RUNLOCK();
1024 			if (cred != NULL &&
1025 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1026 				return (error);
1027 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1028 			IN_IFADDR_RLOCK();
1029 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1030 			    IFF_BROADCAST)
1031 				faddr = satosin(&TAILQ_FIRST(
1032 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1033 			IN_IFADDR_RUNLOCK();
1034 		}
1035 	}
1036 	if (laddr.s_addr == INADDR_ANY) {
1037 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1038 		/*
1039 		 * If the destination address is multicast and an outgoing
1040 		 * interface has been set as a multicast option, prefer the
1041 		 * address of that interface as our source address.
1042 		 */
1043 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1044 		    inp->inp_moptions != NULL) {
1045 			struct ip_moptions *imo;
1046 			struct ifnet *ifp;
1047 
1048 			imo = inp->inp_moptions;
1049 			if (imo->imo_multicast_ifp != NULL) {
1050 				ifp = imo->imo_multicast_ifp;
1051 				IN_IFADDR_RLOCK();
1052 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1053 					if ((ia->ia_ifp == ifp) &&
1054 					    (cred == NULL ||
1055 					    prison_check_ip4(cred,
1056 					    &ia->ia_addr.sin_addr) == 0))
1057 						break;
1058 				}
1059 				if (ia == NULL)
1060 					error = EADDRNOTAVAIL;
1061 				else {
1062 					laddr = ia->ia_addr.sin_addr;
1063 					error = 0;
1064 				}
1065 				IN_IFADDR_RUNLOCK();
1066 			}
1067 		}
1068 		if (error)
1069 			return (error);
1070 	}
1071 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1072 	    laddr, lport, 0, NULL);
1073 	if (oinp != NULL) {
1074 		if (oinpp != NULL)
1075 			*oinpp = oinp;
1076 		return (EADDRINUSE);
1077 	}
1078 	if (lport == 0) {
1079 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1080 		    cred);
1081 		if (error)
1082 			return (error);
1083 	}
1084 	*laddrp = laddr.s_addr;
1085 	*lportp = lport;
1086 	*faddrp = faddr.s_addr;
1087 	*fportp = fport;
1088 	return (0);
1089 }
1090 
1091 void
1092 in_pcbdisconnect(struct inpcb *inp)
1093 {
1094 
1095 	INP_WLOCK_ASSERT(inp);
1096 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1097 
1098 	inp->inp_faddr.s_addr = INADDR_ANY;
1099 	inp->inp_fport = 0;
1100 	in_pcbrehash(inp);
1101 }
1102 #endif /* INET */
1103 
1104 /*
1105  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1106  * For most protocols, this will be invoked immediately prior to calling
1107  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1108  * socket, in which case in_pcbfree() is deferred.
1109  */
1110 void
1111 in_pcbdetach(struct inpcb *inp)
1112 {
1113 
1114 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1115 
1116 	inp->inp_socket->so_pcb = NULL;
1117 	inp->inp_socket = NULL;
1118 }
1119 
1120 /*
1121  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1122  * stability of an inpcb pointer despite the inpcb lock being released.  This
1123  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1124  * but where the inpcb lock may already held, or when acquiring a reference
1125  * via a pcbgroup.
1126  *
1127  * in_pcbref() should be used only to provide brief memory stability, and
1128  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1129  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1130  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1131  * lock and rele are the *only* safe operations that may be performed on the
1132  * inpcb.
1133  *
1134  * While the inpcb will not be freed, releasing the inpcb lock means that the
1135  * connection's state may change, so the caller should be careful to
1136  * revalidate any cached state on reacquiring the lock.  Drop the reference
1137  * using in_pcbrele().
1138  */
1139 void
1140 in_pcbref(struct inpcb *inp)
1141 {
1142 
1143 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1144 
1145 	refcount_acquire(&inp->inp_refcount);
1146 }
1147 
1148 /*
1149  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1150  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1151  * return a flag indicating whether or not the inpcb remains valid.  If it is
1152  * valid, we return with the inpcb lock held.
1153  *
1154  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1155  * reference on an inpcb.  Historically more work was done here (actually, in
1156  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1157  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1158  * about memory stability (and continued use of the write lock).
1159  */
1160 int
1161 in_pcbrele_rlocked(struct inpcb *inp)
1162 {
1163 	struct inpcbinfo *pcbinfo;
1164 
1165 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1166 
1167 	INP_RLOCK_ASSERT(inp);
1168 
1169 	if (refcount_release(&inp->inp_refcount) == 0) {
1170 		/*
1171 		 * If the inpcb has been freed, let the caller know, even if
1172 		 * this isn't the last reference.
1173 		 */
1174 		if (inp->inp_flags2 & INP_FREED) {
1175 			INP_RUNLOCK(inp);
1176 			return (1);
1177 		}
1178 		return (0);
1179 	}
1180 
1181 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1182 
1183 	INP_RUNLOCK(inp);
1184 	pcbinfo = inp->inp_pcbinfo;
1185 	uma_zfree(pcbinfo->ipi_zone, inp);
1186 	return (1);
1187 }
1188 
1189 int
1190 in_pcbrele_wlocked(struct inpcb *inp)
1191 {
1192 	struct inpcbinfo *pcbinfo;
1193 
1194 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1195 
1196 	INP_WLOCK_ASSERT(inp);
1197 
1198 	if (refcount_release(&inp->inp_refcount) == 0)
1199 		return (0);
1200 
1201 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1202 
1203 	INP_WUNLOCK(inp);
1204 	pcbinfo = inp->inp_pcbinfo;
1205 	uma_zfree(pcbinfo->ipi_zone, inp);
1206 	return (1);
1207 }
1208 
1209 /*
1210  * Temporary wrapper.
1211  */
1212 int
1213 in_pcbrele(struct inpcb *inp)
1214 {
1215 
1216 	return (in_pcbrele_wlocked(inp));
1217 }
1218 
1219 /*
1220  * Unconditionally schedule an inpcb to be freed by decrementing its
1221  * reference count, which should occur only after the inpcb has been detached
1222  * from its socket.  If another thread holds a temporary reference (acquired
1223  * using in_pcbref()) then the free is deferred until that reference is
1224  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1225  * work, including removal from global lists, is done in this context, where
1226  * the pcbinfo lock is held.
1227  */
1228 void
1229 in_pcbfree(struct inpcb *inp)
1230 {
1231 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1232 
1233 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1234 
1235 	INP_INFO_WLOCK_ASSERT(pcbinfo);
1236 	INP_WLOCK_ASSERT(inp);
1237 
1238 	/* XXXRW: Do as much as possible here. */
1239 #ifdef IPSEC
1240 	if (inp->inp_sp != NULL)
1241 		ipsec_delete_pcbpolicy(inp);
1242 #endif
1243 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1244 	in_pcbremlists(inp);
1245 #ifdef INET6
1246 	if (inp->inp_vflag & INP_IPV6PROTO) {
1247 		ip6_freepcbopts(inp->in6p_outputopts);
1248 		if (inp->in6p_moptions != NULL)
1249 			ip6_freemoptions(inp->in6p_moptions);
1250 	}
1251 #endif
1252 	if (inp->inp_options)
1253 		(void)m_free(inp->inp_options);
1254 #ifdef INET
1255 	if (inp->inp_moptions != NULL)
1256 		inp_freemoptions(inp->inp_moptions);
1257 #endif
1258 	inp->inp_vflag = 0;
1259 	inp->inp_flags2 |= INP_FREED;
1260 	crfree(inp->inp_cred);
1261 #ifdef MAC
1262 	mac_inpcb_destroy(inp);
1263 #endif
1264 	if (!in_pcbrele_wlocked(inp))
1265 		INP_WUNLOCK(inp);
1266 }
1267 
1268 /*
1269  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1270  * port reservation, and preventing it from being returned by inpcb lookups.
1271  *
1272  * It is used by TCP to mark an inpcb as unused and avoid future packet
1273  * delivery or event notification when a socket remains open but TCP has
1274  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1275  * or a RST on the wire, and allows the port binding to be reused while still
1276  * maintaining the invariant that so_pcb always points to a valid inpcb until
1277  * in_pcbdetach().
1278  *
1279  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1280  * in_pcbnotifyall() and in_pcbpurgeif0()?
1281  */
1282 void
1283 in_pcbdrop(struct inpcb *inp)
1284 {
1285 
1286 	INP_WLOCK_ASSERT(inp);
1287 
1288 	/*
1289 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1290 	 * the hash lock...?
1291 	 */
1292 	inp->inp_flags |= INP_DROPPED;
1293 	if (inp->inp_flags & INP_INHASHLIST) {
1294 		struct inpcbport *phd = inp->inp_phd;
1295 
1296 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1297 		LIST_REMOVE(inp, inp_hash);
1298 		LIST_REMOVE(inp, inp_portlist);
1299 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1300 			LIST_REMOVE(phd, phd_hash);
1301 			free(phd, M_PCB);
1302 		}
1303 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1304 		inp->inp_flags &= ~INP_INHASHLIST;
1305 #ifdef PCBGROUP
1306 		in_pcbgroup_remove(inp);
1307 #endif
1308 	}
1309 }
1310 
1311 #ifdef INET
1312 /*
1313  * Common routines to return the socket addresses associated with inpcbs.
1314  */
1315 struct sockaddr *
1316 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1317 {
1318 	struct sockaddr_in *sin;
1319 
1320 	sin = malloc(sizeof *sin, M_SONAME,
1321 		M_WAITOK | M_ZERO);
1322 	sin->sin_family = AF_INET;
1323 	sin->sin_len = sizeof(*sin);
1324 	sin->sin_addr = *addr_p;
1325 	sin->sin_port = port;
1326 
1327 	return (struct sockaddr *)sin;
1328 }
1329 
1330 int
1331 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1332 {
1333 	struct inpcb *inp;
1334 	struct in_addr addr;
1335 	in_port_t port;
1336 
1337 	inp = sotoinpcb(so);
1338 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1339 
1340 	INP_RLOCK(inp);
1341 	port = inp->inp_lport;
1342 	addr = inp->inp_laddr;
1343 	INP_RUNLOCK(inp);
1344 
1345 	*nam = in_sockaddr(port, &addr);
1346 	return 0;
1347 }
1348 
1349 int
1350 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1351 {
1352 	struct inpcb *inp;
1353 	struct in_addr addr;
1354 	in_port_t port;
1355 
1356 	inp = sotoinpcb(so);
1357 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1358 
1359 	INP_RLOCK(inp);
1360 	port = inp->inp_fport;
1361 	addr = inp->inp_faddr;
1362 	INP_RUNLOCK(inp);
1363 
1364 	*nam = in_sockaddr(port, &addr);
1365 	return 0;
1366 }
1367 
1368 void
1369 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1370     struct inpcb *(*notify)(struct inpcb *, int))
1371 {
1372 	struct inpcb *inp, *inp_temp;
1373 
1374 	INP_INFO_WLOCK(pcbinfo);
1375 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1376 		INP_WLOCK(inp);
1377 #ifdef INET6
1378 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1379 			INP_WUNLOCK(inp);
1380 			continue;
1381 		}
1382 #endif
1383 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1384 		    inp->inp_socket == NULL) {
1385 			INP_WUNLOCK(inp);
1386 			continue;
1387 		}
1388 		if ((*notify)(inp, errno))
1389 			INP_WUNLOCK(inp);
1390 	}
1391 	INP_INFO_WUNLOCK(pcbinfo);
1392 }
1393 
1394 void
1395 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1396 {
1397 	struct inpcb *inp;
1398 	struct ip_moptions *imo;
1399 	int i, gap;
1400 
1401 	INP_INFO_RLOCK(pcbinfo);
1402 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1403 		INP_WLOCK(inp);
1404 		imo = inp->inp_moptions;
1405 		if ((inp->inp_vflag & INP_IPV4) &&
1406 		    imo != NULL) {
1407 			/*
1408 			 * Unselect the outgoing interface if it is being
1409 			 * detached.
1410 			 */
1411 			if (imo->imo_multicast_ifp == ifp)
1412 				imo->imo_multicast_ifp = NULL;
1413 
1414 			/*
1415 			 * Drop multicast group membership if we joined
1416 			 * through the interface being detached.
1417 			 */
1418 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1419 			    i++) {
1420 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1421 					in_delmulti(imo->imo_membership[i]);
1422 					gap++;
1423 				} else if (gap != 0)
1424 					imo->imo_membership[i - gap] =
1425 					    imo->imo_membership[i];
1426 			}
1427 			imo->imo_num_memberships -= gap;
1428 		}
1429 		INP_WUNLOCK(inp);
1430 	}
1431 	INP_INFO_RUNLOCK(pcbinfo);
1432 }
1433 
1434 /*
1435  * Lookup a PCB based on the local address and port.  Caller must hold the
1436  * hash lock.  No inpcb locks or references are acquired.
1437  */
1438 #define INP_LOOKUP_MAPPED_PCB_COST	3
1439 struct inpcb *
1440 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1441     u_short lport, int lookupflags, struct ucred *cred)
1442 {
1443 	struct inpcb *inp;
1444 #ifdef INET6
1445 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1446 #else
1447 	int matchwild = 3;
1448 #endif
1449 	int wildcard;
1450 
1451 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1452 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1453 
1454 	INP_HASH_LOCK_ASSERT(pcbinfo);
1455 
1456 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1457 		struct inpcbhead *head;
1458 		/*
1459 		 * Look for an unconnected (wildcard foreign addr) PCB that
1460 		 * matches the local address and port we're looking for.
1461 		 */
1462 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1463 		    0, pcbinfo->ipi_hashmask)];
1464 		LIST_FOREACH(inp, head, inp_hash) {
1465 #ifdef INET6
1466 			/* XXX inp locking */
1467 			if ((inp->inp_vflag & INP_IPV4) == 0)
1468 				continue;
1469 #endif
1470 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1471 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1472 			    inp->inp_lport == lport) {
1473 				/*
1474 				 * Found?
1475 				 */
1476 				if (cred == NULL ||
1477 				    prison_equal_ip4(cred->cr_prison,
1478 					inp->inp_cred->cr_prison))
1479 					return (inp);
1480 			}
1481 		}
1482 		/*
1483 		 * Not found.
1484 		 */
1485 		return (NULL);
1486 	} else {
1487 		struct inpcbporthead *porthash;
1488 		struct inpcbport *phd;
1489 		struct inpcb *match = NULL;
1490 		/*
1491 		 * Best fit PCB lookup.
1492 		 *
1493 		 * First see if this local port is in use by looking on the
1494 		 * port hash list.
1495 		 */
1496 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1497 		    pcbinfo->ipi_porthashmask)];
1498 		LIST_FOREACH(phd, porthash, phd_hash) {
1499 			if (phd->phd_port == lport)
1500 				break;
1501 		}
1502 		if (phd != NULL) {
1503 			/*
1504 			 * Port is in use by one or more PCBs. Look for best
1505 			 * fit.
1506 			 */
1507 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1508 				wildcard = 0;
1509 				if (cred != NULL &&
1510 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1511 					cred->cr_prison))
1512 					continue;
1513 #ifdef INET6
1514 				/* XXX inp locking */
1515 				if ((inp->inp_vflag & INP_IPV4) == 0)
1516 					continue;
1517 				/*
1518 				 * We never select the PCB that has
1519 				 * INP_IPV6 flag and is bound to :: if
1520 				 * we have another PCB which is bound
1521 				 * to 0.0.0.0.  If a PCB has the
1522 				 * INP_IPV6 flag, then we set its cost
1523 				 * higher than IPv4 only PCBs.
1524 				 *
1525 				 * Note that the case only happens
1526 				 * when a socket is bound to ::, under
1527 				 * the condition that the use of the
1528 				 * mapped address is allowed.
1529 				 */
1530 				if ((inp->inp_vflag & INP_IPV6) != 0)
1531 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1532 #endif
1533 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1534 					wildcard++;
1535 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1536 					if (laddr.s_addr == INADDR_ANY)
1537 						wildcard++;
1538 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1539 						continue;
1540 				} else {
1541 					if (laddr.s_addr != INADDR_ANY)
1542 						wildcard++;
1543 				}
1544 				if (wildcard < matchwild) {
1545 					match = inp;
1546 					matchwild = wildcard;
1547 					if (matchwild == 0)
1548 						break;
1549 				}
1550 			}
1551 		}
1552 		return (match);
1553 	}
1554 }
1555 #undef INP_LOOKUP_MAPPED_PCB_COST
1556 
1557 #ifdef PCBGROUP
1558 /*
1559  * Lookup PCB in hash list, using pcbgroup tables.
1560  */
1561 static struct inpcb *
1562 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1563     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1564     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1565 {
1566 	struct inpcbhead *head;
1567 	struct inpcb *inp, *tmpinp;
1568 	u_short fport = fport_arg, lport = lport_arg;
1569 
1570 	/*
1571 	 * First look for an exact match.
1572 	 */
1573 	tmpinp = NULL;
1574 	INP_GROUP_LOCK(pcbgroup);
1575 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1576 	    pcbgroup->ipg_hashmask)];
1577 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1578 #ifdef INET6
1579 		/* XXX inp locking */
1580 		if ((inp->inp_vflag & INP_IPV4) == 0)
1581 			continue;
1582 #endif
1583 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1584 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1585 		    inp->inp_fport == fport &&
1586 		    inp->inp_lport == lport) {
1587 			/*
1588 			 * XXX We should be able to directly return
1589 			 * the inp here, without any checks.
1590 			 * Well unless both bound with SO_REUSEPORT?
1591 			 */
1592 			if (prison_flag(inp->inp_cred, PR_IP4))
1593 				goto found;
1594 			if (tmpinp == NULL)
1595 				tmpinp = inp;
1596 		}
1597 	}
1598 	if (tmpinp != NULL) {
1599 		inp = tmpinp;
1600 		goto found;
1601 	}
1602 
1603 #ifdef	RSS
1604 	/*
1605 	 * For incoming connections, we may wish to do a wildcard
1606 	 * match for an RSS-local socket.
1607 	 */
1608 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1609 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1610 #ifdef INET6
1611 		struct inpcb *local_wild_mapped = NULL;
1612 #endif
1613 		struct inpcb *jail_wild = NULL;
1614 		struct inpcbhead *head;
1615 		int injail;
1616 
1617 		/*
1618 		 * Order of socket selection - we always prefer jails.
1619 		 *      1. jailed, non-wild.
1620 		 *      2. jailed, wild.
1621 		 *      3. non-jailed, non-wild.
1622 		 *      4. non-jailed, wild.
1623 		 */
1624 
1625 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1626 		    lport, 0, pcbgroup->ipg_hashmask)];
1627 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1628 #ifdef INET6
1629 			/* XXX inp locking */
1630 			if ((inp->inp_vflag & INP_IPV4) == 0)
1631 				continue;
1632 #endif
1633 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1634 			    inp->inp_lport != lport)
1635 				continue;
1636 
1637 			/* XXX inp locking */
1638 			if (ifp && ifp->if_type == IFT_FAITH &&
1639 			    (inp->inp_flags & INP_FAITH) == 0)
1640 				continue;
1641 
1642 			injail = prison_flag(inp->inp_cred, PR_IP4);
1643 			if (injail) {
1644 				if (prison_check_ip4(inp->inp_cred,
1645 				    &laddr) != 0)
1646 					continue;
1647 			} else {
1648 				if (local_exact != NULL)
1649 					continue;
1650 			}
1651 
1652 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1653 				if (injail)
1654 					goto found;
1655 				else
1656 					local_exact = inp;
1657 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1658 #ifdef INET6
1659 				/* XXX inp locking, NULL check */
1660 				if (inp->inp_vflag & INP_IPV6PROTO)
1661 					local_wild_mapped = inp;
1662 				else
1663 #endif
1664 					if (injail)
1665 						jail_wild = inp;
1666 					else
1667 						local_wild = inp;
1668 			}
1669 		} /* LIST_FOREACH */
1670 
1671 		inp = jail_wild;
1672 		if (inp == NULL)
1673 			inp = local_exact;
1674 		if (inp == NULL)
1675 			inp = local_wild;
1676 #ifdef INET6
1677 		if (inp == NULL)
1678 			inp = local_wild_mapped;
1679 #endif
1680 		if (inp != NULL)
1681 			goto found;
1682 	}
1683 #endif
1684 
1685 	/*
1686 	 * Then look for a wildcard match, if requested.
1687 	 */
1688 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1689 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1690 #ifdef INET6
1691 		struct inpcb *local_wild_mapped = NULL;
1692 #endif
1693 		struct inpcb *jail_wild = NULL;
1694 		struct inpcbhead *head;
1695 		int injail;
1696 
1697 		/*
1698 		 * Order of socket selection - we always prefer jails.
1699 		 *      1. jailed, non-wild.
1700 		 *      2. jailed, wild.
1701 		 *      3. non-jailed, non-wild.
1702 		 *      4. non-jailed, wild.
1703 		 */
1704 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1705 		    0, pcbinfo->ipi_wildmask)];
1706 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1707 #ifdef INET6
1708 			/* XXX inp locking */
1709 			if ((inp->inp_vflag & INP_IPV4) == 0)
1710 				continue;
1711 #endif
1712 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1713 			    inp->inp_lport != lport)
1714 				continue;
1715 
1716 			/* XXX inp locking */
1717 			if (ifp && ifp->if_type == IFT_FAITH &&
1718 			    (inp->inp_flags & INP_FAITH) == 0)
1719 				continue;
1720 
1721 			injail = prison_flag(inp->inp_cred, PR_IP4);
1722 			if (injail) {
1723 				if (prison_check_ip4(inp->inp_cred,
1724 				    &laddr) != 0)
1725 					continue;
1726 			} else {
1727 				if (local_exact != NULL)
1728 					continue;
1729 			}
1730 
1731 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1732 				if (injail)
1733 					goto found;
1734 				else
1735 					local_exact = inp;
1736 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1737 #ifdef INET6
1738 				/* XXX inp locking, NULL check */
1739 				if (inp->inp_vflag & INP_IPV6PROTO)
1740 					local_wild_mapped = inp;
1741 				else
1742 #endif
1743 					if (injail)
1744 						jail_wild = inp;
1745 					else
1746 						local_wild = inp;
1747 			}
1748 		} /* LIST_FOREACH */
1749 		inp = jail_wild;
1750 		if (inp == NULL)
1751 			inp = local_exact;
1752 		if (inp == NULL)
1753 			inp = local_wild;
1754 #ifdef INET6
1755 		if (inp == NULL)
1756 			inp = local_wild_mapped;
1757 #endif
1758 		if (inp != NULL)
1759 			goto found;
1760 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1761 	INP_GROUP_UNLOCK(pcbgroup);
1762 	return (NULL);
1763 
1764 found:
1765 	in_pcbref(inp);
1766 	INP_GROUP_UNLOCK(pcbgroup);
1767 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1768 		INP_WLOCK(inp);
1769 		if (in_pcbrele_wlocked(inp))
1770 			return (NULL);
1771 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1772 		INP_RLOCK(inp);
1773 		if (in_pcbrele_rlocked(inp))
1774 			return (NULL);
1775 	} else
1776 		panic("%s: locking bug", __func__);
1777 	return (inp);
1778 }
1779 #endif /* PCBGROUP */
1780 
1781 /*
1782  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1783  * that the caller has locked the hash list, and will not perform any further
1784  * locking or reference operations on either the hash list or the connection.
1785  */
1786 static struct inpcb *
1787 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1788     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1789     struct ifnet *ifp)
1790 {
1791 	struct inpcbhead *head;
1792 	struct inpcb *inp, *tmpinp;
1793 	u_short fport = fport_arg, lport = lport_arg;
1794 
1795 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1796 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1797 
1798 	INP_HASH_LOCK_ASSERT(pcbinfo);
1799 
1800 	/*
1801 	 * First look for an exact match.
1802 	 */
1803 	tmpinp = NULL;
1804 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1805 	    pcbinfo->ipi_hashmask)];
1806 	LIST_FOREACH(inp, head, inp_hash) {
1807 #ifdef INET6
1808 		/* XXX inp locking */
1809 		if ((inp->inp_vflag & INP_IPV4) == 0)
1810 			continue;
1811 #endif
1812 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1813 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1814 		    inp->inp_fport == fport &&
1815 		    inp->inp_lport == lport) {
1816 			/*
1817 			 * XXX We should be able to directly return
1818 			 * the inp here, without any checks.
1819 			 * Well unless both bound with SO_REUSEPORT?
1820 			 */
1821 			if (prison_flag(inp->inp_cred, PR_IP4))
1822 				return (inp);
1823 			if (tmpinp == NULL)
1824 				tmpinp = inp;
1825 		}
1826 	}
1827 	if (tmpinp != NULL)
1828 		return (tmpinp);
1829 
1830 	/*
1831 	 * Then look for a wildcard match, if requested.
1832 	 */
1833 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1834 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1835 #ifdef INET6
1836 		struct inpcb *local_wild_mapped = NULL;
1837 #endif
1838 		struct inpcb *jail_wild = NULL;
1839 		int injail;
1840 
1841 		/*
1842 		 * Order of socket selection - we always prefer jails.
1843 		 *      1. jailed, non-wild.
1844 		 *      2. jailed, wild.
1845 		 *      3. non-jailed, non-wild.
1846 		 *      4. non-jailed, wild.
1847 		 */
1848 
1849 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1850 		    0, pcbinfo->ipi_hashmask)];
1851 		LIST_FOREACH(inp, head, inp_hash) {
1852 #ifdef INET6
1853 			/* XXX inp locking */
1854 			if ((inp->inp_vflag & INP_IPV4) == 0)
1855 				continue;
1856 #endif
1857 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1858 			    inp->inp_lport != lport)
1859 				continue;
1860 
1861 			/* XXX inp locking */
1862 			if (ifp && ifp->if_type == IFT_FAITH &&
1863 			    (inp->inp_flags & INP_FAITH) == 0)
1864 				continue;
1865 
1866 			injail = prison_flag(inp->inp_cred, PR_IP4);
1867 			if (injail) {
1868 				if (prison_check_ip4(inp->inp_cred,
1869 				    &laddr) != 0)
1870 					continue;
1871 			} else {
1872 				if (local_exact != NULL)
1873 					continue;
1874 			}
1875 
1876 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1877 				if (injail)
1878 					return (inp);
1879 				else
1880 					local_exact = inp;
1881 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1882 #ifdef INET6
1883 				/* XXX inp locking, NULL check */
1884 				if (inp->inp_vflag & INP_IPV6PROTO)
1885 					local_wild_mapped = inp;
1886 				else
1887 #endif
1888 					if (injail)
1889 						jail_wild = inp;
1890 					else
1891 						local_wild = inp;
1892 			}
1893 		} /* LIST_FOREACH */
1894 		if (jail_wild != NULL)
1895 			return (jail_wild);
1896 		if (local_exact != NULL)
1897 			return (local_exact);
1898 		if (local_wild != NULL)
1899 			return (local_wild);
1900 #ifdef INET6
1901 		if (local_wild_mapped != NULL)
1902 			return (local_wild_mapped);
1903 #endif
1904 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1905 
1906 	return (NULL);
1907 }
1908 
1909 /*
1910  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1911  * hash list lock, and will return the inpcb locked (i.e., requires
1912  * INPLOOKUP_LOCKPCB).
1913  */
1914 static struct inpcb *
1915 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1916     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1917     struct ifnet *ifp)
1918 {
1919 	struct inpcb *inp;
1920 
1921 	INP_HASH_RLOCK(pcbinfo);
1922 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1923 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1924 	if (inp != NULL) {
1925 		in_pcbref(inp);
1926 		INP_HASH_RUNLOCK(pcbinfo);
1927 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1928 			INP_WLOCK(inp);
1929 			if (in_pcbrele_wlocked(inp))
1930 				return (NULL);
1931 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1932 			INP_RLOCK(inp);
1933 			if (in_pcbrele_rlocked(inp))
1934 				return (NULL);
1935 		} else
1936 			panic("%s: locking bug", __func__);
1937 	} else
1938 		INP_HASH_RUNLOCK(pcbinfo);
1939 	return (inp);
1940 }
1941 
1942 /*
1943  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1944  * from which a pre-calculated hash value may be extracted.
1945  *
1946  * Possibly more of this logic should be in in_pcbgroup.c.
1947  */
1948 struct inpcb *
1949 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1950     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1951 {
1952 #if defined(PCBGROUP) && !defined(RSS)
1953 	struct inpcbgroup *pcbgroup;
1954 #endif
1955 
1956 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1957 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1958 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1959 	    ("%s: LOCKPCB not set", __func__));
1960 
1961 	/*
1962 	 * When not using RSS, use connection groups in preference to the
1963 	 * reservation table when looking up 4-tuples.  When using RSS, just
1964 	 * use the reservation table, due to the cost of the Toeplitz hash
1965 	 * in software.
1966 	 *
1967 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
1968 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
1969 	 * in software.
1970 	 */
1971 #if defined(PCBGROUP) && !defined(RSS)
1972 	if (in_pcbgroup_enabled(pcbinfo)) {
1973 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1974 		    fport);
1975 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1976 		    laddr, lport, lookupflags, ifp));
1977 	}
1978 #endif
1979 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1980 	    lookupflags, ifp));
1981 }
1982 
1983 struct inpcb *
1984 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1985     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1986     struct ifnet *ifp, struct mbuf *m)
1987 {
1988 #ifdef PCBGROUP
1989 	struct inpcbgroup *pcbgroup;
1990 #endif
1991 
1992 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1993 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1994 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1995 	    ("%s: LOCKPCB not set", __func__));
1996 
1997 #ifdef PCBGROUP
1998 	/*
1999 	 * If we can use a hardware-generated hash to look up the connection
2000 	 * group, use that connection group to find the inpcb.  Otherwise
2001 	 * fall back on a software hash -- or the reservation table if we're
2002 	 * using RSS.
2003 	 *
2004 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2005 	 */
2006 	if (in_pcbgroup_enabled(pcbinfo) &&
2007 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2008 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2009 		    m->m_pkthdr.flowid);
2010 		if (pcbgroup != NULL)
2011 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2012 			    fport, laddr, lport, lookupflags, ifp));
2013 #ifndef RSS
2014 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2015 		    fport);
2016 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2017 		    laddr, lport, lookupflags, ifp));
2018 #endif
2019 	}
2020 #endif
2021 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2022 	    lookupflags, ifp));
2023 }
2024 #endif /* INET */
2025 
2026 /*
2027  * Insert PCB onto various hash lists.
2028  */
2029 static int
2030 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2031 {
2032 	struct inpcbhead *pcbhash;
2033 	struct inpcbporthead *pcbporthash;
2034 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2035 	struct inpcbport *phd;
2036 	u_int32_t hashkey_faddr;
2037 
2038 	INP_WLOCK_ASSERT(inp);
2039 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2040 
2041 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2042 	    ("in_pcbinshash: INP_INHASHLIST"));
2043 
2044 #ifdef INET6
2045 	if (inp->inp_vflag & INP_IPV6)
2046 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
2047 	else
2048 #endif
2049 	hashkey_faddr = inp->inp_faddr.s_addr;
2050 
2051 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2052 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2053 
2054 	pcbporthash = &pcbinfo->ipi_porthashbase[
2055 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2056 
2057 	/*
2058 	 * Go through port list and look for a head for this lport.
2059 	 */
2060 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2061 		if (phd->phd_port == inp->inp_lport)
2062 			break;
2063 	}
2064 	/*
2065 	 * If none exists, malloc one and tack it on.
2066 	 */
2067 	if (phd == NULL) {
2068 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2069 		if (phd == NULL) {
2070 			return (ENOBUFS); /* XXX */
2071 		}
2072 		phd->phd_port = inp->inp_lport;
2073 		LIST_INIT(&phd->phd_pcblist);
2074 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2075 	}
2076 	inp->inp_phd = phd;
2077 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2078 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2079 	inp->inp_flags |= INP_INHASHLIST;
2080 #ifdef PCBGROUP
2081 	if (do_pcbgroup_update)
2082 		in_pcbgroup_update(inp);
2083 #endif
2084 	return (0);
2085 }
2086 
2087 /*
2088  * For now, there are two public interfaces to insert an inpcb into the hash
2089  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2090  * is used only in the TCP syncache, where in_pcbinshash is called before the
2091  * full 4-tuple is set for the inpcb, and we don't want to install in the
2092  * pcbgroup until later.
2093  *
2094  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2095  * connection groups, and partially initialised inpcbs should not be exposed
2096  * to either reservation hash tables or pcbgroups.
2097  */
2098 int
2099 in_pcbinshash(struct inpcb *inp)
2100 {
2101 
2102 	return (in_pcbinshash_internal(inp, 1));
2103 }
2104 
2105 int
2106 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2107 {
2108 
2109 	return (in_pcbinshash_internal(inp, 0));
2110 }
2111 
2112 /*
2113  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2114  * changed. NOTE: This does not handle the case of the lport changing (the
2115  * hashed port list would have to be updated as well), so the lport must
2116  * not change after in_pcbinshash() has been called.
2117  */
2118 void
2119 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2120 {
2121 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2122 	struct inpcbhead *head;
2123 	u_int32_t hashkey_faddr;
2124 
2125 	INP_WLOCK_ASSERT(inp);
2126 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2127 
2128 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2129 	    ("in_pcbrehash: !INP_INHASHLIST"));
2130 
2131 #ifdef INET6
2132 	if (inp->inp_vflag & INP_IPV6)
2133 		hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */;
2134 	else
2135 #endif
2136 	hashkey_faddr = inp->inp_faddr.s_addr;
2137 
2138 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2139 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2140 
2141 	LIST_REMOVE(inp, inp_hash);
2142 	LIST_INSERT_HEAD(head, inp, inp_hash);
2143 
2144 #ifdef PCBGROUP
2145 	if (m != NULL)
2146 		in_pcbgroup_update_mbuf(inp, m);
2147 	else
2148 		in_pcbgroup_update(inp);
2149 #endif
2150 }
2151 
2152 void
2153 in_pcbrehash(struct inpcb *inp)
2154 {
2155 
2156 	in_pcbrehash_mbuf(inp, NULL);
2157 }
2158 
2159 /*
2160  * Remove PCB from various lists.
2161  */
2162 static void
2163 in_pcbremlists(struct inpcb *inp)
2164 {
2165 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2166 
2167 	INP_INFO_WLOCK_ASSERT(pcbinfo);
2168 	INP_WLOCK_ASSERT(inp);
2169 
2170 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2171 	if (inp->inp_flags & INP_INHASHLIST) {
2172 		struct inpcbport *phd = inp->inp_phd;
2173 
2174 		INP_HASH_WLOCK(pcbinfo);
2175 		LIST_REMOVE(inp, inp_hash);
2176 		LIST_REMOVE(inp, inp_portlist);
2177 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2178 			LIST_REMOVE(phd, phd_hash);
2179 			free(phd, M_PCB);
2180 		}
2181 		INP_HASH_WUNLOCK(pcbinfo);
2182 		inp->inp_flags &= ~INP_INHASHLIST;
2183 	}
2184 	LIST_REMOVE(inp, inp_list);
2185 	pcbinfo->ipi_count--;
2186 #ifdef PCBGROUP
2187 	in_pcbgroup_remove(inp);
2188 #endif
2189 }
2190 
2191 /*
2192  * A set label operation has occurred at the socket layer, propagate the
2193  * label change into the in_pcb for the socket.
2194  */
2195 void
2196 in_pcbsosetlabel(struct socket *so)
2197 {
2198 #ifdef MAC
2199 	struct inpcb *inp;
2200 
2201 	inp = sotoinpcb(so);
2202 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2203 
2204 	INP_WLOCK(inp);
2205 	SOCK_LOCK(so);
2206 	mac_inpcb_sosetlabel(so, inp);
2207 	SOCK_UNLOCK(so);
2208 	INP_WUNLOCK(inp);
2209 #endif
2210 }
2211 
2212 /*
2213  * ipport_tick runs once per second, determining if random port allocation
2214  * should be continued.  If more than ipport_randomcps ports have been
2215  * allocated in the last second, then we return to sequential port
2216  * allocation. We return to random allocation only once we drop below
2217  * ipport_randomcps for at least ipport_randomtime seconds.
2218  */
2219 static void
2220 ipport_tick(void *xtp)
2221 {
2222 	VNET_ITERATOR_DECL(vnet_iter);
2223 
2224 	VNET_LIST_RLOCK_NOSLEEP();
2225 	VNET_FOREACH(vnet_iter) {
2226 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2227 		if (V_ipport_tcpallocs <=
2228 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2229 			if (V_ipport_stoprandom > 0)
2230 				V_ipport_stoprandom--;
2231 		} else
2232 			V_ipport_stoprandom = V_ipport_randomtime;
2233 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2234 		CURVNET_RESTORE();
2235 	}
2236 	VNET_LIST_RUNLOCK_NOSLEEP();
2237 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2238 }
2239 
2240 static void
2241 ip_fini(void *xtp)
2242 {
2243 
2244 	callout_stop(&ipport_tick_callout);
2245 }
2246 
2247 /*
2248  * The ipport_callout should start running at about the time we attach the
2249  * inet or inet6 domains.
2250  */
2251 static void
2252 ipport_tick_init(const void *unused __unused)
2253 {
2254 
2255 	/* Start ipport_tick. */
2256 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
2257 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2258 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2259 		SHUTDOWN_PRI_DEFAULT);
2260 }
2261 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2262     ipport_tick_init, NULL);
2263 
2264 void
2265 inp_wlock(struct inpcb *inp)
2266 {
2267 
2268 	INP_WLOCK(inp);
2269 }
2270 
2271 void
2272 inp_wunlock(struct inpcb *inp)
2273 {
2274 
2275 	INP_WUNLOCK(inp);
2276 }
2277 
2278 void
2279 inp_rlock(struct inpcb *inp)
2280 {
2281 
2282 	INP_RLOCK(inp);
2283 }
2284 
2285 void
2286 inp_runlock(struct inpcb *inp)
2287 {
2288 
2289 	INP_RUNLOCK(inp);
2290 }
2291 
2292 #ifdef INVARIANTS
2293 void
2294 inp_lock_assert(struct inpcb *inp)
2295 {
2296 
2297 	INP_WLOCK_ASSERT(inp);
2298 }
2299 
2300 void
2301 inp_unlock_assert(struct inpcb *inp)
2302 {
2303 
2304 	INP_UNLOCK_ASSERT(inp);
2305 }
2306 #endif
2307 
2308 void
2309 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2310 {
2311 	struct inpcb *inp;
2312 
2313 	INP_INFO_RLOCK(&V_tcbinfo);
2314 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2315 		INP_WLOCK(inp);
2316 		func(inp, arg);
2317 		INP_WUNLOCK(inp);
2318 	}
2319 	INP_INFO_RUNLOCK(&V_tcbinfo);
2320 }
2321 
2322 struct socket *
2323 inp_inpcbtosocket(struct inpcb *inp)
2324 {
2325 
2326 	INP_WLOCK_ASSERT(inp);
2327 	return (inp->inp_socket);
2328 }
2329 
2330 struct tcpcb *
2331 inp_inpcbtotcpcb(struct inpcb *inp)
2332 {
2333 
2334 	INP_WLOCK_ASSERT(inp);
2335 	return ((struct tcpcb *)inp->inp_ppcb);
2336 }
2337 
2338 int
2339 inp_ip_tos_get(const struct inpcb *inp)
2340 {
2341 
2342 	return (inp->inp_ip_tos);
2343 }
2344 
2345 void
2346 inp_ip_tos_set(struct inpcb *inp, int val)
2347 {
2348 
2349 	inp->inp_ip_tos = val;
2350 }
2351 
2352 void
2353 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2354     uint32_t *faddr, uint16_t *fp)
2355 {
2356 
2357 	INP_LOCK_ASSERT(inp);
2358 	*laddr = inp->inp_laddr.s_addr;
2359 	*faddr = inp->inp_faddr.s_addr;
2360 	*lp = inp->inp_lport;
2361 	*fp = inp->inp_fport;
2362 }
2363 
2364 struct inpcb *
2365 so_sotoinpcb(struct socket *so)
2366 {
2367 
2368 	return (sotoinpcb(so));
2369 }
2370 
2371 struct tcpcb *
2372 so_sototcpcb(struct socket *so)
2373 {
2374 
2375 	return (sototcpcb(so));
2376 }
2377 
2378 #ifdef DDB
2379 static void
2380 db_print_indent(int indent)
2381 {
2382 	int i;
2383 
2384 	for (i = 0; i < indent; i++)
2385 		db_printf(" ");
2386 }
2387 
2388 static void
2389 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2390 {
2391 	char faddr_str[48], laddr_str[48];
2392 
2393 	db_print_indent(indent);
2394 	db_printf("%s at %p\n", name, inc);
2395 
2396 	indent += 2;
2397 
2398 #ifdef INET6
2399 	if (inc->inc_flags & INC_ISIPV6) {
2400 		/* IPv6. */
2401 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2402 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2403 	} else
2404 #endif
2405 	{
2406 		/* IPv4. */
2407 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2408 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2409 	}
2410 	db_print_indent(indent);
2411 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2412 	    ntohs(inc->inc_lport));
2413 	db_print_indent(indent);
2414 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2415 	    ntohs(inc->inc_fport));
2416 }
2417 
2418 static void
2419 db_print_inpflags(int inp_flags)
2420 {
2421 	int comma;
2422 
2423 	comma = 0;
2424 	if (inp_flags & INP_RECVOPTS) {
2425 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2426 		comma = 1;
2427 	}
2428 	if (inp_flags & INP_RECVRETOPTS) {
2429 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2430 		comma = 1;
2431 	}
2432 	if (inp_flags & INP_RECVDSTADDR) {
2433 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2434 		comma = 1;
2435 	}
2436 	if (inp_flags & INP_HDRINCL) {
2437 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2438 		comma = 1;
2439 	}
2440 	if (inp_flags & INP_HIGHPORT) {
2441 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2442 		comma = 1;
2443 	}
2444 	if (inp_flags & INP_LOWPORT) {
2445 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2446 		comma = 1;
2447 	}
2448 	if (inp_flags & INP_ANONPORT) {
2449 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2450 		comma = 1;
2451 	}
2452 	if (inp_flags & INP_RECVIF) {
2453 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2454 		comma = 1;
2455 	}
2456 	if (inp_flags & INP_MTUDISC) {
2457 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2458 		comma = 1;
2459 	}
2460 	if (inp_flags & INP_FAITH) {
2461 		db_printf("%sINP_FAITH", comma ? ", " : "");
2462 		comma = 1;
2463 	}
2464 	if (inp_flags & INP_RECVTTL) {
2465 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2466 		comma = 1;
2467 	}
2468 	if (inp_flags & INP_DONTFRAG) {
2469 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2470 		comma = 1;
2471 	}
2472 	if (inp_flags & INP_RECVTOS) {
2473 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2474 		comma = 1;
2475 	}
2476 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2477 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2478 		comma = 1;
2479 	}
2480 	if (inp_flags & IN6P_PKTINFO) {
2481 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2482 		comma = 1;
2483 	}
2484 	if (inp_flags & IN6P_HOPLIMIT) {
2485 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2486 		comma = 1;
2487 	}
2488 	if (inp_flags & IN6P_HOPOPTS) {
2489 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2490 		comma = 1;
2491 	}
2492 	if (inp_flags & IN6P_DSTOPTS) {
2493 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2494 		comma = 1;
2495 	}
2496 	if (inp_flags & IN6P_RTHDR) {
2497 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2498 		comma = 1;
2499 	}
2500 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2501 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2502 		comma = 1;
2503 	}
2504 	if (inp_flags & IN6P_TCLASS) {
2505 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2506 		comma = 1;
2507 	}
2508 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2509 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2510 		comma = 1;
2511 	}
2512 	if (inp_flags & INP_TIMEWAIT) {
2513 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2514 		comma  = 1;
2515 	}
2516 	if (inp_flags & INP_ONESBCAST) {
2517 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2518 		comma  = 1;
2519 	}
2520 	if (inp_flags & INP_DROPPED) {
2521 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2522 		comma  = 1;
2523 	}
2524 	if (inp_flags & INP_SOCKREF) {
2525 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2526 		comma  = 1;
2527 	}
2528 	if (inp_flags & IN6P_RFC2292) {
2529 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2530 		comma = 1;
2531 	}
2532 	if (inp_flags & IN6P_MTU) {
2533 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2534 		comma = 1;
2535 	}
2536 }
2537 
2538 static void
2539 db_print_inpvflag(u_char inp_vflag)
2540 {
2541 	int comma;
2542 
2543 	comma = 0;
2544 	if (inp_vflag & INP_IPV4) {
2545 		db_printf("%sINP_IPV4", comma ? ", " : "");
2546 		comma  = 1;
2547 	}
2548 	if (inp_vflag & INP_IPV6) {
2549 		db_printf("%sINP_IPV6", comma ? ", " : "");
2550 		comma  = 1;
2551 	}
2552 	if (inp_vflag & INP_IPV6PROTO) {
2553 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2554 		comma  = 1;
2555 	}
2556 }
2557 
2558 static void
2559 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2560 {
2561 
2562 	db_print_indent(indent);
2563 	db_printf("%s at %p\n", name, inp);
2564 
2565 	indent += 2;
2566 
2567 	db_print_indent(indent);
2568 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2569 
2570 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2571 
2572 	db_print_indent(indent);
2573 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2574 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2575 
2576 	db_print_indent(indent);
2577 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2578 	   inp->inp_label, inp->inp_flags);
2579 	db_print_inpflags(inp->inp_flags);
2580 	db_printf(")\n");
2581 
2582 	db_print_indent(indent);
2583 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2584 	    inp->inp_vflag);
2585 	db_print_inpvflag(inp->inp_vflag);
2586 	db_printf(")\n");
2587 
2588 	db_print_indent(indent);
2589 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2590 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2591 
2592 	db_print_indent(indent);
2593 #ifdef INET6
2594 	if (inp->inp_vflag & INP_IPV6) {
2595 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2596 		    "in6p_moptions: %p\n", inp->in6p_options,
2597 		    inp->in6p_outputopts, inp->in6p_moptions);
2598 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2599 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2600 		    inp->in6p_hops);
2601 	} else
2602 #endif
2603 	{
2604 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2605 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2606 		    inp->inp_options, inp->inp_moptions);
2607 	}
2608 
2609 	db_print_indent(indent);
2610 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2611 	    (uintmax_t)inp->inp_gencnt);
2612 }
2613 
2614 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2615 {
2616 	struct inpcb *inp;
2617 
2618 	if (!have_addr) {
2619 		db_printf("usage: show inpcb <addr>\n");
2620 		return;
2621 	}
2622 	inp = (struct inpcb *)addr;
2623 
2624 	db_print_inpcb(inp, "inpcb", 0);
2625 }
2626 #endif /* DDB */
2627