xref: /freebsd/sys/netinet/in_pcb.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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/rss_config.h>
75 #include <net/vnet.h>
76 
77 #if defined(INET) || defined(INET6)
78 #include <netinet/in.h>
79 #include <netinet/in_pcb.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_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
168 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
169 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
171 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
172 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
174 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
175 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
177 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
178 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
180 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
181 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
183 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
184 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
185 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
186 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
187 	&VNET_NAME(ipport_reservedhigh), 0, "");
188 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
189 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
190 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
191 	CTLFLAG_VNET | CTLFLAG_RW,
192 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
194 	CTLFLAG_VNET | CTLFLAG_RW,
195 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
196 	"allocations before switching to a sequental one");
197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
198 	CTLFLAG_VNET | CTLFLAG_RW,
199 	&VNET_NAME(ipport_randomtime), 0,
200 	"Minimum time to keep sequental port "
201 	"allocation before switching to a random one");
202 #endif /* INET */
203 
204 /*
205  * in_pcb.c: manage the Protocol Control Blocks.
206  *
207  * NOTE: It is assumed that most of these functions will be called with
208  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
209  * functions often modify hash chains or addresses in pcbs.
210  */
211 
212 /*
213  * Initialize an inpcbinfo -- we should be able to reduce the number of
214  * arguments in time.
215  */
216 void
217 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
218     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
219     char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
220     uint32_t inpcbzone_flags, u_int hashfields)
221 {
222 
223 	INP_INFO_LOCK_INIT(pcbinfo, name);
224 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
225 #ifdef VIMAGE
226 	pcbinfo->ipi_vnet = curvnet;
227 #endif
228 	pcbinfo->ipi_listhead = listhead;
229 	LIST_INIT(pcbinfo->ipi_listhead);
230 	pcbinfo->ipi_count = 0;
231 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
232 	    &pcbinfo->ipi_hashmask);
233 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
234 	    &pcbinfo->ipi_porthashmask);
235 #ifdef PCBGROUP
236 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
237 #endif
238 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
239 	    NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
240 	    inpcbzone_flags);
241 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
242 	uma_zone_set_warning(pcbinfo->ipi_zone,
243 	    "kern.ipc.maxsockets limit reached");
244 }
245 
246 /*
247  * Destroy an inpcbinfo.
248  */
249 void
250 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
251 {
252 
253 	KASSERT(pcbinfo->ipi_count == 0,
254 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
255 
256 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
257 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
258 	    pcbinfo->ipi_porthashmask);
259 #ifdef PCBGROUP
260 	in_pcbgroup_destroy(pcbinfo);
261 #endif
262 	uma_zdestroy(pcbinfo->ipi_zone);
263 	INP_HASH_LOCK_DESTROY(pcbinfo);
264 	INP_INFO_LOCK_DESTROY(pcbinfo);
265 }
266 
267 /*
268  * Allocate a PCB and associate it with the socket.
269  * On success return with the PCB locked.
270  */
271 int
272 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
273 {
274 	struct inpcb *inp;
275 	int error;
276 
277 	INP_INFO_WLOCK_ASSERT(pcbinfo);
278 	error = 0;
279 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
280 	if (inp == NULL)
281 		return (ENOBUFS);
282 	bzero(inp, inp_zero_size);
283 	inp->inp_pcbinfo = pcbinfo;
284 	inp->inp_socket = so;
285 	inp->inp_cred = crhold(so->so_cred);
286 	inp->inp_inc.inc_fibnum = so->so_fibnum;
287 #ifdef MAC
288 	error = mac_inpcb_init(inp, M_NOWAIT);
289 	if (error != 0)
290 		goto out;
291 	mac_inpcb_create(so, inp);
292 #endif
293 #ifdef IPSEC
294 	error = ipsec_init_policy(so, &inp->inp_sp);
295 	if (error != 0) {
296 #ifdef MAC
297 		mac_inpcb_destroy(inp);
298 #endif
299 		goto out;
300 	}
301 #endif /*IPSEC*/
302 #ifdef INET6
303 	if (INP_SOCKAF(so) == AF_INET6) {
304 		inp->inp_vflag |= INP_IPV6PROTO;
305 		if (V_ip6_v6only)
306 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
307 	}
308 #endif
309 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
310 	pcbinfo->ipi_count++;
311 	so->so_pcb = (caddr_t)inp;
312 #ifdef INET6
313 	if (V_ip6_auto_flowlabel)
314 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
315 #endif
316 	INP_WLOCK(inp);
317 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
318 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
319 #if defined(IPSEC) || defined(MAC)
320 out:
321 	if (error != 0) {
322 		crfree(inp->inp_cred);
323 		uma_zfree(pcbinfo->ipi_zone, inp);
324 	}
325 #endif
326 	return (error);
327 }
328 
329 #ifdef INET
330 int
331 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
332 {
333 	int anonport, error;
334 
335 	INP_WLOCK_ASSERT(inp);
336 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
337 
338 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
339 		return (EINVAL);
340 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
341 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
342 	    &inp->inp_lport, cred);
343 	if (error)
344 		return (error);
345 	if (in_pcbinshash(inp) != 0) {
346 		inp->inp_laddr.s_addr = INADDR_ANY;
347 		inp->inp_lport = 0;
348 		return (EAGAIN);
349 	}
350 	if (anonport)
351 		inp->inp_flags |= INP_ANONPORT;
352 	return (0);
353 }
354 #endif
355 
356 /*
357  * Select a local port (number) to use.
358  */
359 #if defined(INET) || defined(INET6)
360 int
361 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
362     struct ucred *cred, int lookupflags)
363 {
364 	struct inpcbinfo *pcbinfo;
365 	struct inpcb *tmpinp;
366 	unsigned short *lastport;
367 	int count, dorandom, error;
368 	u_short aux, first, last, lport;
369 #ifdef INET
370 	struct in_addr laddr;
371 #endif
372 
373 	pcbinfo = inp->inp_pcbinfo;
374 
375 	/*
376 	 * Because no actual state changes occur here, a global write lock on
377 	 * the pcbinfo isn't required.
378 	 */
379 	INP_LOCK_ASSERT(inp);
380 	INP_HASH_LOCK_ASSERT(pcbinfo);
381 
382 	if (inp->inp_flags & INP_HIGHPORT) {
383 		first = V_ipport_hifirstauto;	/* sysctl */
384 		last  = V_ipport_hilastauto;
385 		lastport = &pcbinfo->ipi_lasthi;
386 	} else if (inp->inp_flags & INP_LOWPORT) {
387 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
388 		if (error)
389 			return (error);
390 		first = V_ipport_lowfirstauto;	/* 1023 */
391 		last  = V_ipport_lowlastauto;	/* 600 */
392 		lastport = &pcbinfo->ipi_lastlow;
393 	} else {
394 		first = V_ipport_firstauto;	/* sysctl */
395 		last  = V_ipport_lastauto;
396 		lastport = &pcbinfo->ipi_lastport;
397 	}
398 	/*
399 	 * For UDP(-Lite), use random port allocation as long as the user
400 	 * allows it.  For TCP (and as of yet unknown) connections,
401 	 * use random port allocation only if the user allows it AND
402 	 * ipport_tick() allows it.
403 	 */
404 	if (V_ipport_randomized &&
405 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
406 		pcbinfo == &V_ulitecbinfo))
407 		dorandom = 1;
408 	else
409 		dorandom = 0;
410 	/*
411 	 * It makes no sense to do random port allocation if
412 	 * we have the only port available.
413 	 */
414 	if (first == last)
415 		dorandom = 0;
416 	/* Make sure to not include UDP(-Lite) packets in the count. */
417 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
418 		V_ipport_tcpallocs++;
419 	/*
420 	 * Instead of having two loops further down counting up or down
421 	 * make sure that first is always <= last and go with only one
422 	 * code path implementing all logic.
423 	 */
424 	if (first > last) {
425 		aux = first;
426 		first = last;
427 		last = aux;
428 	}
429 
430 #ifdef INET
431 	/* Make the compiler happy. */
432 	laddr.s_addr = 0;
433 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
434 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
435 		    __func__, inp));
436 		laddr = *laddrp;
437 	}
438 #endif
439 	tmpinp = NULL;	/* Make compiler happy. */
440 	lport = *lportp;
441 
442 	if (dorandom)
443 		*lastport = first + (arc4random() % (last - first));
444 
445 	count = last - first;
446 
447 	do {
448 		if (count-- < 0)	/* completely used? */
449 			return (EADDRNOTAVAIL);
450 		++*lastport;
451 		if (*lastport < first || *lastport > last)
452 			*lastport = first;
453 		lport = htons(*lastport);
454 
455 #ifdef INET6
456 		if ((inp->inp_vflag & INP_IPV6) != 0)
457 			tmpinp = in6_pcblookup_local(pcbinfo,
458 			    &inp->in6p_laddr, lport, lookupflags, cred);
459 #endif
460 #if defined(INET) && defined(INET6)
461 		else
462 #endif
463 #ifdef INET
464 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
465 			    lport, lookupflags, cred);
466 #endif
467 	} while (tmpinp != NULL);
468 
469 #ifdef INET
470 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
471 		laddrp->s_addr = laddr.s_addr;
472 #endif
473 	*lportp = lport;
474 
475 	return (0);
476 }
477 
478 /*
479  * Return cached socket options.
480  */
481 short
482 inp_so_options(const struct inpcb *inp)
483 {
484    short so_options;
485 
486    so_options = 0;
487 
488    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
489 	   so_options |= SO_REUSEPORT;
490    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
491 	   so_options |= SO_REUSEADDR;
492    return (so_options);
493 }
494 #endif /* INET || INET6 */
495 
496 /*
497  * Check if a new BINDMULTI socket is allowed to be created.
498  *
499  * ni points to the new inp.
500  * oi points to the exisitng inp.
501  *
502  * This checks whether the existing inp also has BINDMULTI and
503  * whether the credentials match.
504  */
505 int
506 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
507 {
508 	/* Check permissions match */
509 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
510 	    (ni->inp_cred->cr_uid !=
511 	    oi->inp_cred->cr_uid))
512 		return (0);
513 
514 	/* Check the existing inp has BINDMULTI set */
515 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
516 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
517 		return (0);
518 
519 	/*
520 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
521 	 * it is and it matches the checks.
522 	 */
523 	return (1);
524 }
525 
526 #ifdef INET
527 /*
528  * Set up a bind operation on a PCB, performing port allocation
529  * as required, but do not actually modify the PCB. Callers can
530  * either complete the bind by setting inp_laddr/inp_lport and
531  * calling in_pcbinshash(), or they can just use the resulting
532  * port and address to authorise the sending of a once-off packet.
533  *
534  * On error, the values of *laddrp and *lportp are not changed.
535  */
536 int
537 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
538     u_short *lportp, struct ucred *cred)
539 {
540 	struct socket *so = inp->inp_socket;
541 	struct sockaddr_in *sin;
542 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
543 	struct in_addr laddr;
544 	u_short lport = 0;
545 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
546 	int error;
547 
548 	/*
549 	 * No state changes, so read locks are sufficient here.
550 	 */
551 	INP_LOCK_ASSERT(inp);
552 	INP_HASH_LOCK_ASSERT(pcbinfo);
553 
554 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
555 		return (EADDRNOTAVAIL);
556 	laddr.s_addr = *laddrp;
557 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
558 		return (EINVAL);
559 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
560 		lookupflags = INPLOOKUP_WILDCARD;
561 	if (nam == NULL) {
562 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
563 			return (error);
564 	} else {
565 		sin = (struct sockaddr_in *)nam;
566 		if (nam->sa_len != sizeof (*sin))
567 			return (EINVAL);
568 #ifdef notdef
569 		/*
570 		 * We should check the family, but old programs
571 		 * incorrectly fail to initialize it.
572 		 */
573 		if (sin->sin_family != AF_INET)
574 			return (EAFNOSUPPORT);
575 #endif
576 		error = prison_local_ip4(cred, &sin->sin_addr);
577 		if (error)
578 			return (error);
579 		if (sin->sin_port != *lportp) {
580 			/* Don't allow the port to change. */
581 			if (*lportp != 0)
582 				return (EINVAL);
583 			lport = sin->sin_port;
584 		}
585 		/* NB: lport is left as 0 if the port isn't being changed. */
586 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
587 			/*
588 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
589 			 * allow complete duplication of binding if
590 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
591 			 * and a multicast address is bound on both
592 			 * new and duplicated sockets.
593 			 */
594 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
595 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
596 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
597 			sin->sin_port = 0;		/* yech... */
598 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
599 			/*
600 			 * Is the address a local IP address?
601 			 * If INP_BINDANY is set, then the socket may be bound
602 			 * to any endpoint address, local or not.
603 			 */
604 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
605 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
606 				return (EADDRNOTAVAIL);
607 		}
608 		laddr = sin->sin_addr;
609 		if (lport) {
610 			struct inpcb *t;
611 			struct tcptw *tw;
612 
613 			/* GROSS */
614 			if (ntohs(lport) <= V_ipport_reservedhigh &&
615 			    ntohs(lport) >= V_ipport_reservedlow &&
616 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
617 			    0))
618 				return (EACCES);
619 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
620 			    priv_check_cred(inp->inp_cred,
621 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
622 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
623 				    lport, INPLOOKUP_WILDCARD, cred);
624 	/*
625 	 * XXX
626 	 * This entire block sorely needs a rewrite.
627 	 */
628 				if (t &&
629 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
630 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
631 				    (so->so_type != SOCK_STREAM ||
632 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
633 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
634 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
635 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
636 				    (inp->inp_cred->cr_uid !=
637 				     t->inp_cred->cr_uid))
638 					return (EADDRINUSE);
639 
640 				/*
641 				 * If the socket is a BINDMULTI socket, then
642 				 * the credentials need to match and the
643 				 * original socket also has to have been bound
644 				 * with BINDMULTI.
645 				 */
646 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
647 					return (EADDRINUSE);
648 			}
649 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
650 			    lport, lookupflags, cred);
651 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
652 				/*
653 				 * XXXRW: If an incpb has had its timewait
654 				 * state recycled, we treat the address as
655 				 * being in use (for now).  This is better
656 				 * than a panic, but not desirable.
657 				 */
658 				tw = intotw(t);
659 				if (tw == NULL ||
660 				    (reuseport & tw->tw_so_options) == 0)
661 					return (EADDRINUSE);
662 			} else if (t &&
663 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
664 			    (reuseport & inp_so_options(t)) == 0) {
665 #ifdef INET6
666 				if (ntohl(sin->sin_addr.s_addr) !=
667 				    INADDR_ANY ||
668 				    ntohl(t->inp_laddr.s_addr) !=
669 				    INADDR_ANY ||
670 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
671 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
672 #endif
673 				return (EADDRINUSE);
674 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
675 					return (EADDRINUSE);
676 			}
677 		}
678 	}
679 	if (*lportp != 0)
680 		lport = *lportp;
681 	if (lport == 0) {
682 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
683 		if (error != 0)
684 			return (error);
685 
686 	}
687 	*laddrp = laddr.s_addr;
688 	*lportp = lport;
689 	return (0);
690 }
691 
692 /*
693  * Connect from a socket to a specified address.
694  * Both address and port must be specified in argument sin.
695  * If don't have a local address for this socket yet,
696  * then pick one.
697  */
698 int
699 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
700     struct ucred *cred, struct mbuf *m)
701 {
702 	u_short lport, fport;
703 	in_addr_t laddr, faddr;
704 	int anonport, error;
705 
706 	INP_WLOCK_ASSERT(inp);
707 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
708 
709 	lport = inp->inp_lport;
710 	laddr = inp->inp_laddr.s_addr;
711 	anonport = (lport == 0);
712 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
713 	    NULL, cred);
714 	if (error)
715 		return (error);
716 
717 	/* Do the initial binding of the local address if required. */
718 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
719 		inp->inp_lport = lport;
720 		inp->inp_laddr.s_addr = laddr;
721 		if (in_pcbinshash(inp) != 0) {
722 			inp->inp_laddr.s_addr = INADDR_ANY;
723 			inp->inp_lport = 0;
724 			return (EAGAIN);
725 		}
726 	}
727 
728 	/* Commit the remaining changes. */
729 	inp->inp_lport = lport;
730 	inp->inp_laddr.s_addr = laddr;
731 	inp->inp_faddr.s_addr = faddr;
732 	inp->inp_fport = fport;
733 	in_pcbrehash_mbuf(inp, m);
734 
735 	if (anonport)
736 		inp->inp_flags |= INP_ANONPORT;
737 	return (0);
738 }
739 
740 int
741 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
742 {
743 
744 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
745 }
746 
747 /*
748  * Do proper source address selection on an unbound socket in case
749  * of connect. Take jails into account as well.
750  */
751 int
752 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
753     struct ucred *cred)
754 {
755 	struct ifaddr *ifa;
756 	struct sockaddr *sa;
757 	struct sockaddr_in *sin;
758 	struct route sro;
759 	int error;
760 
761 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
762 
763 	/*
764 	 * Bypass source address selection and use the primary jail IP
765 	 * if requested.
766 	 */
767 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
768 		return (0);
769 
770 	error = 0;
771 	bzero(&sro, sizeof(sro));
772 
773 	sin = (struct sockaddr_in *)&sro.ro_dst;
774 	sin->sin_family = AF_INET;
775 	sin->sin_len = sizeof(struct sockaddr_in);
776 	sin->sin_addr.s_addr = faddr->s_addr;
777 
778 	/*
779 	 * If route is known our src addr is taken from the i/f,
780 	 * else punt.
781 	 *
782 	 * Find out route to destination.
783 	 */
784 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
785 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
786 
787 	/*
788 	 * If we found a route, use the address corresponding to
789 	 * the outgoing interface.
790 	 *
791 	 * Otherwise assume faddr is reachable on a directly connected
792 	 * network and try to find a corresponding interface to take
793 	 * the source address from.
794 	 */
795 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
796 		struct in_ifaddr *ia;
797 		struct ifnet *ifp;
798 
799 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
800 					inp->inp_socket->so_fibnum));
801 		if (ia == NULL)
802 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
803 						inp->inp_socket->so_fibnum));
804 		if (ia == NULL) {
805 			error = ENETUNREACH;
806 			goto done;
807 		}
808 
809 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
810 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
811 			ifa_free(&ia->ia_ifa);
812 			goto done;
813 		}
814 
815 		ifp = ia->ia_ifp;
816 		ifa_free(&ia->ia_ifa);
817 		ia = NULL;
818 		IF_ADDR_RLOCK(ifp);
819 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
820 
821 			sa = ifa->ifa_addr;
822 			if (sa->sa_family != AF_INET)
823 				continue;
824 			sin = (struct sockaddr_in *)sa;
825 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
826 				ia = (struct in_ifaddr *)ifa;
827 				break;
828 			}
829 		}
830 		if (ia != NULL) {
831 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
832 			IF_ADDR_RUNLOCK(ifp);
833 			goto done;
834 		}
835 		IF_ADDR_RUNLOCK(ifp);
836 
837 		/* 3. As a last resort return the 'default' jail address. */
838 		error = prison_get_ip4(cred, laddr);
839 		goto done;
840 	}
841 
842 	/*
843 	 * If the outgoing interface on the route found is not
844 	 * a loopback interface, use the address from that interface.
845 	 * In case of jails do those three steps:
846 	 * 1. check if the interface address belongs to the jail. If so use it.
847 	 * 2. check if we have any address on the outgoing interface
848 	 *    belonging to this jail. If so use it.
849 	 * 3. as a last resort return the 'default' jail address.
850 	 */
851 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
852 		struct in_ifaddr *ia;
853 		struct ifnet *ifp;
854 
855 		/* If not jailed, use the default returned. */
856 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
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 		/* Jailed. */
863 		/* 1. Check if the iface address belongs to the jail. */
864 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
865 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
866 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
867 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
868 			goto done;
869 		}
870 
871 		/*
872 		 * 2. Check if we have any address on the outgoing interface
873 		 *    belonging to this jail.
874 		 */
875 		ia = NULL;
876 		ifp = sro.ro_rt->rt_ifp;
877 		IF_ADDR_RLOCK(ifp);
878 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
879 			sa = ifa->ifa_addr;
880 			if (sa->sa_family != AF_INET)
881 				continue;
882 			sin = (struct sockaddr_in *)sa;
883 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
884 				ia = (struct in_ifaddr *)ifa;
885 				break;
886 			}
887 		}
888 		if (ia != NULL) {
889 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
890 			IF_ADDR_RUNLOCK(ifp);
891 			goto done;
892 		}
893 		IF_ADDR_RUNLOCK(ifp);
894 
895 		/* 3. As a last resort return the 'default' jail address. */
896 		error = prison_get_ip4(cred, laddr);
897 		goto done;
898 	}
899 
900 	/*
901 	 * The outgoing interface is marked with 'loopback net', so a route
902 	 * to ourselves is here.
903 	 * Try to find the interface of the destination address and then
904 	 * take the address from there. That interface is not necessarily
905 	 * a loopback interface.
906 	 * In case of jails, check that it is an address of the jail
907 	 * and if we cannot find, fall back to the 'default' jail address.
908 	 */
909 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
910 		struct sockaddr_in sain;
911 		struct in_ifaddr *ia;
912 
913 		bzero(&sain, sizeof(struct sockaddr_in));
914 		sain.sin_family = AF_INET;
915 		sain.sin_len = sizeof(struct sockaddr_in);
916 		sain.sin_addr.s_addr = faddr->s_addr;
917 
918 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
919 					inp->inp_socket->so_fibnum));
920 		if (ia == NULL)
921 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
922 						inp->inp_socket->so_fibnum));
923 		if (ia == NULL)
924 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
925 
926 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
927 			if (ia == NULL) {
928 				error = ENETUNREACH;
929 				goto done;
930 			}
931 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
932 			ifa_free(&ia->ia_ifa);
933 			goto done;
934 		}
935 
936 		/* Jailed. */
937 		if (ia != NULL) {
938 			struct ifnet *ifp;
939 
940 			ifp = ia->ia_ifp;
941 			ifa_free(&ia->ia_ifa);
942 			ia = NULL;
943 			IF_ADDR_RLOCK(ifp);
944 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
945 
946 				sa = ifa->ifa_addr;
947 				if (sa->sa_family != AF_INET)
948 					continue;
949 				sin = (struct sockaddr_in *)sa;
950 				if (prison_check_ip4(cred,
951 				    &sin->sin_addr) == 0) {
952 					ia = (struct in_ifaddr *)ifa;
953 					break;
954 				}
955 			}
956 			if (ia != NULL) {
957 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
958 				IF_ADDR_RUNLOCK(ifp);
959 				goto done;
960 			}
961 			IF_ADDR_RUNLOCK(ifp);
962 		}
963 
964 		/* 3. As a last resort return the 'default' jail address. */
965 		error = prison_get_ip4(cred, laddr);
966 		goto done;
967 	}
968 
969 done:
970 	if (sro.ro_rt != NULL)
971 		RTFREE(sro.ro_rt);
972 	return (error);
973 }
974 
975 /*
976  * Set up for a connect from a socket to the specified address.
977  * On entry, *laddrp and *lportp should contain the current local
978  * address and port for the PCB; these are updated to the values
979  * that should be placed in inp_laddr and inp_lport to complete
980  * the connect.
981  *
982  * On success, *faddrp and *fportp will be set to the remote address
983  * and port. These are not updated in the error case.
984  *
985  * If the operation fails because the connection already exists,
986  * *oinpp will be set to the PCB of that connection so that the
987  * caller can decide to override it. In all other cases, *oinpp
988  * is set to NULL.
989  */
990 int
991 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
992     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
993     struct inpcb **oinpp, struct ucred *cred)
994 {
995 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
996 	struct in_ifaddr *ia;
997 	struct inpcb *oinp;
998 	struct in_addr laddr, faddr;
999 	u_short lport, fport;
1000 	int error;
1001 
1002 	/*
1003 	 * Because a global state change doesn't actually occur here, a read
1004 	 * lock is sufficient.
1005 	 */
1006 	INP_LOCK_ASSERT(inp);
1007 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1008 
1009 	if (oinpp != NULL)
1010 		*oinpp = NULL;
1011 	if (nam->sa_len != sizeof (*sin))
1012 		return (EINVAL);
1013 	if (sin->sin_family != AF_INET)
1014 		return (EAFNOSUPPORT);
1015 	if (sin->sin_port == 0)
1016 		return (EADDRNOTAVAIL);
1017 	laddr.s_addr = *laddrp;
1018 	lport = *lportp;
1019 	faddr = sin->sin_addr;
1020 	fport = sin->sin_port;
1021 
1022 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1023 		/*
1024 		 * If the destination address is INADDR_ANY,
1025 		 * use the primary local address.
1026 		 * If the supplied address is INADDR_BROADCAST,
1027 		 * and the primary interface supports broadcast,
1028 		 * choose the broadcast address for that interface.
1029 		 */
1030 		if (faddr.s_addr == INADDR_ANY) {
1031 			IN_IFADDR_RLOCK();
1032 			faddr =
1033 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1034 			IN_IFADDR_RUNLOCK();
1035 			if (cred != NULL &&
1036 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1037 				return (error);
1038 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1039 			IN_IFADDR_RLOCK();
1040 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1041 			    IFF_BROADCAST)
1042 				faddr = satosin(&TAILQ_FIRST(
1043 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1044 			IN_IFADDR_RUNLOCK();
1045 		}
1046 	}
1047 	if (laddr.s_addr == INADDR_ANY) {
1048 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1049 		/*
1050 		 * If the destination address is multicast and an outgoing
1051 		 * interface has been set as a multicast option, prefer the
1052 		 * address of that interface as our source address.
1053 		 */
1054 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1055 		    inp->inp_moptions != NULL) {
1056 			struct ip_moptions *imo;
1057 			struct ifnet *ifp;
1058 
1059 			imo = inp->inp_moptions;
1060 			if (imo->imo_multicast_ifp != NULL) {
1061 				ifp = imo->imo_multicast_ifp;
1062 				IN_IFADDR_RLOCK();
1063 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1064 					if ((ia->ia_ifp == ifp) &&
1065 					    (cred == NULL ||
1066 					    prison_check_ip4(cred,
1067 					    &ia->ia_addr.sin_addr) == 0))
1068 						break;
1069 				}
1070 				if (ia == NULL)
1071 					error = EADDRNOTAVAIL;
1072 				else {
1073 					laddr = ia->ia_addr.sin_addr;
1074 					error = 0;
1075 				}
1076 				IN_IFADDR_RUNLOCK();
1077 			}
1078 		}
1079 		if (error)
1080 			return (error);
1081 	}
1082 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1083 	    laddr, lport, 0, NULL);
1084 	if (oinp != NULL) {
1085 		if (oinpp != NULL)
1086 			*oinpp = oinp;
1087 		return (EADDRINUSE);
1088 	}
1089 	if (lport == 0) {
1090 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1091 		    cred);
1092 		if (error)
1093 			return (error);
1094 	}
1095 	*laddrp = laddr.s_addr;
1096 	*lportp = lport;
1097 	*faddrp = faddr.s_addr;
1098 	*fportp = fport;
1099 	return (0);
1100 }
1101 
1102 void
1103 in_pcbdisconnect(struct inpcb *inp)
1104 {
1105 
1106 	INP_WLOCK_ASSERT(inp);
1107 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1108 
1109 	inp->inp_faddr.s_addr = INADDR_ANY;
1110 	inp->inp_fport = 0;
1111 	in_pcbrehash(inp);
1112 }
1113 #endif /* INET */
1114 
1115 /*
1116  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1117  * For most protocols, this will be invoked immediately prior to calling
1118  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1119  * socket, in which case in_pcbfree() is deferred.
1120  */
1121 void
1122 in_pcbdetach(struct inpcb *inp)
1123 {
1124 
1125 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1126 
1127 	inp->inp_socket->so_pcb = NULL;
1128 	inp->inp_socket = NULL;
1129 }
1130 
1131 /*
1132  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1133  * stability of an inpcb pointer despite the inpcb lock being released.  This
1134  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1135  * but where the inpcb lock may already held, or when acquiring a reference
1136  * via a pcbgroup.
1137  *
1138  * in_pcbref() should be used only to provide brief memory stability, and
1139  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1140  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1141  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1142  * lock and rele are the *only* safe operations that may be performed on the
1143  * inpcb.
1144  *
1145  * While the inpcb will not be freed, releasing the inpcb lock means that the
1146  * connection's state may change, so the caller should be careful to
1147  * revalidate any cached state on reacquiring the lock.  Drop the reference
1148  * using in_pcbrele().
1149  */
1150 void
1151 in_pcbref(struct inpcb *inp)
1152 {
1153 
1154 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1155 
1156 	refcount_acquire(&inp->inp_refcount);
1157 }
1158 
1159 /*
1160  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1161  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1162  * return a flag indicating whether or not the inpcb remains valid.  If it is
1163  * valid, we return with the inpcb lock held.
1164  *
1165  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1166  * reference on an inpcb.  Historically more work was done here (actually, in
1167  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1168  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1169  * about memory stability (and continued use of the write lock).
1170  */
1171 int
1172 in_pcbrele_rlocked(struct inpcb *inp)
1173 {
1174 	struct inpcbinfo *pcbinfo;
1175 
1176 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1177 
1178 	INP_RLOCK_ASSERT(inp);
1179 
1180 	if (refcount_release(&inp->inp_refcount) == 0) {
1181 		/*
1182 		 * If the inpcb has been freed, let the caller know, even if
1183 		 * this isn't the last reference.
1184 		 */
1185 		if (inp->inp_flags2 & INP_FREED) {
1186 			INP_RUNLOCK(inp);
1187 			return (1);
1188 		}
1189 		return (0);
1190 	}
1191 
1192 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1193 
1194 	INP_RUNLOCK(inp);
1195 	pcbinfo = inp->inp_pcbinfo;
1196 	uma_zfree(pcbinfo->ipi_zone, inp);
1197 	return (1);
1198 }
1199 
1200 int
1201 in_pcbrele_wlocked(struct inpcb *inp)
1202 {
1203 	struct inpcbinfo *pcbinfo;
1204 
1205 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1206 
1207 	INP_WLOCK_ASSERT(inp);
1208 
1209 	if (refcount_release(&inp->inp_refcount) == 0)
1210 		return (0);
1211 
1212 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1213 
1214 	INP_WUNLOCK(inp);
1215 	pcbinfo = inp->inp_pcbinfo;
1216 	uma_zfree(pcbinfo->ipi_zone, inp);
1217 	return (1);
1218 }
1219 
1220 /*
1221  * Temporary wrapper.
1222  */
1223 int
1224 in_pcbrele(struct inpcb *inp)
1225 {
1226 
1227 	return (in_pcbrele_wlocked(inp));
1228 }
1229 
1230 /*
1231  * Unconditionally schedule an inpcb to be freed by decrementing its
1232  * reference count, which should occur only after the inpcb has been detached
1233  * from its socket.  If another thread holds a temporary reference (acquired
1234  * using in_pcbref()) then the free is deferred until that reference is
1235  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1236  * work, including removal from global lists, is done in this context, where
1237  * the pcbinfo lock is held.
1238  */
1239 void
1240 in_pcbfree(struct inpcb *inp)
1241 {
1242 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1243 
1244 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1245 
1246 	INP_INFO_WLOCK_ASSERT(pcbinfo);
1247 	INP_WLOCK_ASSERT(inp);
1248 
1249 	/* XXXRW: Do as much as possible here. */
1250 #ifdef IPSEC
1251 	if (inp->inp_sp != NULL)
1252 		ipsec_delete_pcbpolicy(inp);
1253 #endif
1254 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1255 	in_pcbremlists(inp);
1256 #ifdef INET6
1257 	if (inp->inp_vflag & INP_IPV6PROTO) {
1258 		ip6_freepcbopts(inp->in6p_outputopts);
1259 		if (inp->in6p_moptions != NULL)
1260 			ip6_freemoptions(inp->in6p_moptions);
1261 	}
1262 #endif
1263 	if (inp->inp_options)
1264 		(void)m_free(inp->inp_options);
1265 #ifdef INET
1266 	if (inp->inp_moptions != NULL)
1267 		inp_freemoptions(inp->inp_moptions);
1268 #endif
1269 	inp->inp_vflag = 0;
1270 	inp->inp_flags2 |= INP_FREED;
1271 	crfree(inp->inp_cred);
1272 #ifdef MAC
1273 	mac_inpcb_destroy(inp);
1274 #endif
1275 	if (!in_pcbrele_wlocked(inp))
1276 		INP_WUNLOCK(inp);
1277 }
1278 
1279 /*
1280  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1281  * port reservation, and preventing it from being returned by inpcb lookups.
1282  *
1283  * It is used by TCP to mark an inpcb as unused and avoid future packet
1284  * delivery or event notification when a socket remains open but TCP has
1285  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1286  * or a RST on the wire, and allows the port binding to be reused while still
1287  * maintaining the invariant that so_pcb always points to a valid inpcb until
1288  * in_pcbdetach().
1289  *
1290  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1291  * in_pcbnotifyall() and in_pcbpurgeif0()?
1292  */
1293 void
1294 in_pcbdrop(struct inpcb *inp)
1295 {
1296 
1297 	INP_WLOCK_ASSERT(inp);
1298 
1299 	/*
1300 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1301 	 * the hash lock...?
1302 	 */
1303 	inp->inp_flags |= INP_DROPPED;
1304 	if (inp->inp_flags & INP_INHASHLIST) {
1305 		struct inpcbport *phd = inp->inp_phd;
1306 
1307 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1308 		LIST_REMOVE(inp, inp_hash);
1309 		LIST_REMOVE(inp, inp_portlist);
1310 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1311 			LIST_REMOVE(phd, phd_hash);
1312 			free(phd, M_PCB);
1313 		}
1314 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1315 		inp->inp_flags &= ~INP_INHASHLIST;
1316 #ifdef PCBGROUP
1317 		in_pcbgroup_remove(inp);
1318 #endif
1319 	}
1320 }
1321 
1322 #ifdef INET
1323 /*
1324  * Common routines to return the socket addresses associated with inpcbs.
1325  */
1326 struct sockaddr *
1327 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1328 {
1329 	struct sockaddr_in *sin;
1330 
1331 	sin = malloc(sizeof *sin, M_SONAME,
1332 		M_WAITOK | M_ZERO);
1333 	sin->sin_family = AF_INET;
1334 	sin->sin_len = sizeof(*sin);
1335 	sin->sin_addr = *addr_p;
1336 	sin->sin_port = port;
1337 
1338 	return (struct sockaddr *)sin;
1339 }
1340 
1341 int
1342 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1343 {
1344 	struct inpcb *inp;
1345 	struct in_addr addr;
1346 	in_port_t port;
1347 
1348 	inp = sotoinpcb(so);
1349 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1350 
1351 	INP_RLOCK(inp);
1352 	port = inp->inp_lport;
1353 	addr = inp->inp_laddr;
1354 	INP_RUNLOCK(inp);
1355 
1356 	*nam = in_sockaddr(port, &addr);
1357 	return 0;
1358 }
1359 
1360 int
1361 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1362 {
1363 	struct inpcb *inp;
1364 	struct in_addr addr;
1365 	in_port_t port;
1366 
1367 	inp = sotoinpcb(so);
1368 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1369 
1370 	INP_RLOCK(inp);
1371 	port = inp->inp_fport;
1372 	addr = inp->inp_faddr;
1373 	INP_RUNLOCK(inp);
1374 
1375 	*nam = in_sockaddr(port, &addr);
1376 	return 0;
1377 }
1378 
1379 void
1380 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1381     struct inpcb *(*notify)(struct inpcb *, int))
1382 {
1383 	struct inpcb *inp, *inp_temp;
1384 
1385 	INP_INFO_WLOCK(pcbinfo);
1386 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1387 		INP_WLOCK(inp);
1388 #ifdef INET6
1389 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1390 			INP_WUNLOCK(inp);
1391 			continue;
1392 		}
1393 #endif
1394 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1395 		    inp->inp_socket == NULL) {
1396 			INP_WUNLOCK(inp);
1397 			continue;
1398 		}
1399 		if ((*notify)(inp, errno))
1400 			INP_WUNLOCK(inp);
1401 	}
1402 	INP_INFO_WUNLOCK(pcbinfo);
1403 }
1404 
1405 void
1406 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1407 {
1408 	struct inpcb *inp;
1409 	struct ip_moptions *imo;
1410 	int i, gap;
1411 
1412 	INP_INFO_RLOCK(pcbinfo);
1413 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1414 		INP_WLOCK(inp);
1415 		imo = inp->inp_moptions;
1416 		if ((inp->inp_vflag & INP_IPV4) &&
1417 		    imo != NULL) {
1418 			/*
1419 			 * Unselect the outgoing interface if it is being
1420 			 * detached.
1421 			 */
1422 			if (imo->imo_multicast_ifp == ifp)
1423 				imo->imo_multicast_ifp = NULL;
1424 
1425 			/*
1426 			 * Drop multicast group membership if we joined
1427 			 * through the interface being detached.
1428 			 */
1429 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1430 			    i++) {
1431 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1432 					in_delmulti(imo->imo_membership[i]);
1433 					gap++;
1434 				} else if (gap != 0)
1435 					imo->imo_membership[i - gap] =
1436 					    imo->imo_membership[i];
1437 			}
1438 			imo->imo_num_memberships -= gap;
1439 		}
1440 		INP_WUNLOCK(inp);
1441 	}
1442 	INP_INFO_RUNLOCK(pcbinfo);
1443 }
1444 
1445 /*
1446  * Lookup a PCB based on the local address and port.  Caller must hold the
1447  * hash lock.  No inpcb locks or references are acquired.
1448  */
1449 #define INP_LOOKUP_MAPPED_PCB_COST	3
1450 struct inpcb *
1451 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1452     u_short lport, int lookupflags, struct ucred *cred)
1453 {
1454 	struct inpcb *inp;
1455 #ifdef INET6
1456 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1457 #else
1458 	int matchwild = 3;
1459 #endif
1460 	int wildcard;
1461 
1462 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1463 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1464 
1465 	INP_HASH_LOCK_ASSERT(pcbinfo);
1466 
1467 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1468 		struct inpcbhead *head;
1469 		/*
1470 		 * Look for an unconnected (wildcard foreign addr) PCB that
1471 		 * matches the local address and port we're looking for.
1472 		 */
1473 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1474 		    0, pcbinfo->ipi_hashmask)];
1475 		LIST_FOREACH(inp, head, inp_hash) {
1476 #ifdef INET6
1477 			/* XXX inp locking */
1478 			if ((inp->inp_vflag & INP_IPV4) == 0)
1479 				continue;
1480 #endif
1481 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1482 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1483 			    inp->inp_lport == lport) {
1484 				/*
1485 				 * Found?
1486 				 */
1487 				if (cred == NULL ||
1488 				    prison_equal_ip4(cred->cr_prison,
1489 					inp->inp_cred->cr_prison))
1490 					return (inp);
1491 			}
1492 		}
1493 		/*
1494 		 * Not found.
1495 		 */
1496 		return (NULL);
1497 	} else {
1498 		struct inpcbporthead *porthash;
1499 		struct inpcbport *phd;
1500 		struct inpcb *match = NULL;
1501 		/*
1502 		 * Best fit PCB lookup.
1503 		 *
1504 		 * First see if this local port is in use by looking on the
1505 		 * port hash list.
1506 		 */
1507 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1508 		    pcbinfo->ipi_porthashmask)];
1509 		LIST_FOREACH(phd, porthash, phd_hash) {
1510 			if (phd->phd_port == lport)
1511 				break;
1512 		}
1513 		if (phd != NULL) {
1514 			/*
1515 			 * Port is in use by one or more PCBs. Look for best
1516 			 * fit.
1517 			 */
1518 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1519 				wildcard = 0;
1520 				if (cred != NULL &&
1521 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1522 					cred->cr_prison))
1523 					continue;
1524 #ifdef INET6
1525 				/* XXX inp locking */
1526 				if ((inp->inp_vflag & INP_IPV4) == 0)
1527 					continue;
1528 				/*
1529 				 * We never select the PCB that has
1530 				 * INP_IPV6 flag and is bound to :: if
1531 				 * we have another PCB which is bound
1532 				 * to 0.0.0.0.  If a PCB has the
1533 				 * INP_IPV6 flag, then we set its cost
1534 				 * higher than IPv4 only PCBs.
1535 				 *
1536 				 * Note that the case only happens
1537 				 * when a socket is bound to ::, under
1538 				 * the condition that the use of the
1539 				 * mapped address is allowed.
1540 				 */
1541 				if ((inp->inp_vflag & INP_IPV6) != 0)
1542 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1543 #endif
1544 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1545 					wildcard++;
1546 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1547 					if (laddr.s_addr == INADDR_ANY)
1548 						wildcard++;
1549 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1550 						continue;
1551 				} else {
1552 					if (laddr.s_addr != INADDR_ANY)
1553 						wildcard++;
1554 				}
1555 				if (wildcard < matchwild) {
1556 					match = inp;
1557 					matchwild = wildcard;
1558 					if (matchwild == 0)
1559 						break;
1560 				}
1561 			}
1562 		}
1563 		return (match);
1564 	}
1565 }
1566 #undef INP_LOOKUP_MAPPED_PCB_COST
1567 
1568 #ifdef PCBGROUP
1569 /*
1570  * Lookup PCB in hash list, using pcbgroup tables.
1571  */
1572 static struct inpcb *
1573 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1574     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1575     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1576 {
1577 	struct inpcbhead *head;
1578 	struct inpcb *inp, *tmpinp;
1579 	u_short fport = fport_arg, lport = lport_arg;
1580 
1581 	/*
1582 	 * First look for an exact match.
1583 	 */
1584 	tmpinp = NULL;
1585 	INP_GROUP_LOCK(pcbgroup);
1586 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1587 	    pcbgroup->ipg_hashmask)];
1588 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1589 #ifdef INET6
1590 		/* XXX inp locking */
1591 		if ((inp->inp_vflag & INP_IPV4) == 0)
1592 			continue;
1593 #endif
1594 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1595 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1596 		    inp->inp_fport == fport &&
1597 		    inp->inp_lport == lport) {
1598 			/*
1599 			 * XXX We should be able to directly return
1600 			 * the inp here, without any checks.
1601 			 * Well unless both bound with SO_REUSEPORT?
1602 			 */
1603 			if (prison_flag(inp->inp_cred, PR_IP4))
1604 				goto found;
1605 			if (tmpinp == NULL)
1606 				tmpinp = inp;
1607 		}
1608 	}
1609 	if (tmpinp != NULL) {
1610 		inp = tmpinp;
1611 		goto found;
1612 	}
1613 
1614 #ifdef	RSS
1615 	/*
1616 	 * For incoming connections, we may wish to do a wildcard
1617 	 * match for an RSS-local socket.
1618 	 */
1619 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1620 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1621 #ifdef INET6
1622 		struct inpcb *local_wild_mapped = NULL;
1623 #endif
1624 		struct inpcb *jail_wild = NULL;
1625 		struct inpcbhead *head;
1626 		int injail;
1627 
1628 		/*
1629 		 * Order of socket selection - we always prefer jails.
1630 		 *      1. jailed, non-wild.
1631 		 *      2. jailed, wild.
1632 		 *      3. non-jailed, non-wild.
1633 		 *      4. non-jailed, wild.
1634 		 */
1635 
1636 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1637 		    lport, 0, pcbgroup->ipg_hashmask)];
1638 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1639 #ifdef INET6
1640 			/* XXX inp locking */
1641 			if ((inp->inp_vflag & INP_IPV4) == 0)
1642 				continue;
1643 #endif
1644 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1645 			    inp->inp_lport != lport)
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 			injail = prison_flag(inp->inp_cred, PR_IP4);
1723 			if (injail) {
1724 				if (prison_check_ip4(inp->inp_cred,
1725 				    &laddr) != 0)
1726 					continue;
1727 			} else {
1728 				if (local_exact != NULL)
1729 					continue;
1730 			}
1731 
1732 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1733 				if (injail)
1734 					goto found;
1735 				else
1736 					local_exact = inp;
1737 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1738 #ifdef INET6
1739 				/* XXX inp locking, NULL check */
1740 				if (inp->inp_vflag & INP_IPV6PROTO)
1741 					local_wild_mapped = inp;
1742 				else
1743 #endif
1744 					if (injail)
1745 						jail_wild = inp;
1746 					else
1747 						local_wild = inp;
1748 			}
1749 		} /* LIST_FOREACH */
1750 		inp = jail_wild;
1751 		if (inp == NULL)
1752 			inp = local_exact;
1753 		if (inp == NULL)
1754 			inp = local_wild;
1755 #ifdef INET6
1756 		if (inp == NULL)
1757 			inp = local_wild_mapped;
1758 #endif
1759 		if (inp != NULL)
1760 			goto found;
1761 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1762 	INP_GROUP_UNLOCK(pcbgroup);
1763 	return (NULL);
1764 
1765 found:
1766 	in_pcbref(inp);
1767 	INP_GROUP_UNLOCK(pcbgroup);
1768 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1769 		INP_WLOCK(inp);
1770 		if (in_pcbrele_wlocked(inp))
1771 			return (NULL);
1772 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1773 		INP_RLOCK(inp);
1774 		if (in_pcbrele_rlocked(inp))
1775 			return (NULL);
1776 	} else
1777 		panic("%s: locking bug", __func__);
1778 	return (inp);
1779 }
1780 #endif /* PCBGROUP */
1781 
1782 /*
1783  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1784  * that the caller has locked the hash list, and will not perform any further
1785  * locking or reference operations on either the hash list or the connection.
1786  */
1787 static struct inpcb *
1788 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1789     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1790     struct ifnet *ifp)
1791 {
1792 	struct inpcbhead *head;
1793 	struct inpcb *inp, *tmpinp;
1794 	u_short fport = fport_arg, lport = lport_arg;
1795 
1796 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1797 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1798 
1799 	INP_HASH_LOCK_ASSERT(pcbinfo);
1800 
1801 	/*
1802 	 * First look for an exact match.
1803 	 */
1804 	tmpinp = NULL;
1805 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1806 	    pcbinfo->ipi_hashmask)];
1807 	LIST_FOREACH(inp, head, inp_hash) {
1808 #ifdef INET6
1809 		/* XXX inp locking */
1810 		if ((inp->inp_vflag & INP_IPV4) == 0)
1811 			continue;
1812 #endif
1813 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1814 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1815 		    inp->inp_fport == fport &&
1816 		    inp->inp_lport == lport) {
1817 			/*
1818 			 * XXX We should be able to directly return
1819 			 * the inp here, without any checks.
1820 			 * Well unless both bound with SO_REUSEPORT?
1821 			 */
1822 			if (prison_flag(inp->inp_cred, PR_IP4))
1823 				return (inp);
1824 			if (tmpinp == NULL)
1825 				tmpinp = inp;
1826 		}
1827 	}
1828 	if (tmpinp != NULL)
1829 		return (tmpinp);
1830 
1831 	/*
1832 	 * Then look for a wildcard match, if requested.
1833 	 */
1834 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1835 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1836 #ifdef INET6
1837 		struct inpcb *local_wild_mapped = NULL;
1838 #endif
1839 		struct inpcb *jail_wild = NULL;
1840 		int injail;
1841 
1842 		/*
1843 		 * Order of socket selection - we always prefer jails.
1844 		 *      1. jailed, non-wild.
1845 		 *      2. jailed, wild.
1846 		 *      3. non-jailed, non-wild.
1847 		 *      4. non-jailed, wild.
1848 		 */
1849 
1850 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1851 		    0, pcbinfo->ipi_hashmask)];
1852 		LIST_FOREACH(inp, head, inp_hash) {
1853 #ifdef INET6
1854 			/* XXX inp locking */
1855 			if ((inp->inp_vflag & INP_IPV4) == 0)
1856 				continue;
1857 #endif
1858 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1859 			    inp->inp_lport != lport)
1860 				continue;
1861 
1862 			injail = prison_flag(inp->inp_cred, PR_IP4);
1863 			if (injail) {
1864 				if (prison_check_ip4(inp->inp_cred,
1865 				    &laddr) != 0)
1866 					continue;
1867 			} else {
1868 				if (local_exact != NULL)
1869 					continue;
1870 			}
1871 
1872 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1873 				if (injail)
1874 					return (inp);
1875 				else
1876 					local_exact = inp;
1877 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1878 #ifdef INET6
1879 				/* XXX inp locking, NULL check */
1880 				if (inp->inp_vflag & INP_IPV6PROTO)
1881 					local_wild_mapped = inp;
1882 				else
1883 #endif
1884 					if (injail)
1885 						jail_wild = inp;
1886 					else
1887 						local_wild = inp;
1888 			}
1889 		} /* LIST_FOREACH */
1890 		if (jail_wild != NULL)
1891 			return (jail_wild);
1892 		if (local_exact != NULL)
1893 			return (local_exact);
1894 		if (local_wild != NULL)
1895 			return (local_wild);
1896 #ifdef INET6
1897 		if (local_wild_mapped != NULL)
1898 			return (local_wild_mapped);
1899 #endif
1900 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1901 
1902 	return (NULL);
1903 }
1904 
1905 /*
1906  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1907  * hash list lock, and will return the inpcb locked (i.e., requires
1908  * INPLOOKUP_LOCKPCB).
1909  */
1910 static struct inpcb *
1911 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1912     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1913     struct ifnet *ifp)
1914 {
1915 	struct inpcb *inp;
1916 
1917 	INP_HASH_RLOCK(pcbinfo);
1918 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1919 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1920 	if (inp != NULL) {
1921 		in_pcbref(inp);
1922 		INP_HASH_RUNLOCK(pcbinfo);
1923 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1924 			INP_WLOCK(inp);
1925 			if (in_pcbrele_wlocked(inp))
1926 				return (NULL);
1927 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1928 			INP_RLOCK(inp);
1929 			if (in_pcbrele_rlocked(inp))
1930 				return (NULL);
1931 		} else
1932 			panic("%s: locking bug", __func__);
1933 	} else
1934 		INP_HASH_RUNLOCK(pcbinfo);
1935 	return (inp);
1936 }
1937 
1938 /*
1939  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1940  * from which a pre-calculated hash value may be extracted.
1941  *
1942  * Possibly more of this logic should be in in_pcbgroup.c.
1943  */
1944 struct inpcb *
1945 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1946     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1947 {
1948 #if defined(PCBGROUP) && !defined(RSS)
1949 	struct inpcbgroup *pcbgroup;
1950 #endif
1951 
1952 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1953 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1954 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1955 	    ("%s: LOCKPCB not set", __func__));
1956 
1957 	/*
1958 	 * When not using RSS, use connection groups in preference to the
1959 	 * reservation table when looking up 4-tuples.  When using RSS, just
1960 	 * use the reservation table, due to the cost of the Toeplitz hash
1961 	 * in software.
1962 	 *
1963 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
1964 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
1965 	 * in software.
1966 	 */
1967 #if defined(PCBGROUP) && !defined(RSS)
1968 	if (in_pcbgroup_enabled(pcbinfo)) {
1969 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
1970 		    fport);
1971 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
1972 		    laddr, lport, lookupflags, ifp));
1973 	}
1974 #endif
1975 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
1976 	    lookupflags, ifp));
1977 }
1978 
1979 struct inpcb *
1980 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1981     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1982     struct ifnet *ifp, struct mbuf *m)
1983 {
1984 #ifdef PCBGROUP
1985 	struct inpcbgroup *pcbgroup;
1986 #endif
1987 
1988 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1989 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1990 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1991 	    ("%s: LOCKPCB not set", __func__));
1992 
1993 #ifdef PCBGROUP
1994 	/*
1995 	 * If we can use a hardware-generated hash to look up the connection
1996 	 * group, use that connection group to find the inpcb.  Otherwise
1997 	 * fall back on a software hash -- or the reservation table if we're
1998 	 * using RSS.
1999 	 *
2000 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2001 	 */
2002 	if (in_pcbgroup_enabled(pcbinfo) &&
2003 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2004 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2005 		    m->m_pkthdr.flowid);
2006 		if (pcbgroup != NULL)
2007 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2008 			    fport, laddr, lport, lookupflags, ifp));
2009 #ifndef RSS
2010 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2011 		    fport);
2012 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2013 		    laddr, lport, lookupflags, ifp));
2014 #endif
2015 	}
2016 #endif
2017 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2018 	    lookupflags, ifp));
2019 }
2020 #endif /* INET */
2021 
2022 /*
2023  * Insert PCB onto various hash lists.
2024  */
2025 static int
2026 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2027 {
2028 	struct inpcbhead *pcbhash;
2029 	struct inpcbporthead *pcbporthash;
2030 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2031 	struct inpcbport *phd;
2032 	u_int32_t hashkey_faddr;
2033 
2034 	INP_WLOCK_ASSERT(inp);
2035 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2036 
2037 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2038 	    ("in_pcbinshash: INP_INHASHLIST"));
2039 
2040 #ifdef INET6
2041 	if (inp->inp_vflag & INP_IPV6)
2042 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2043 	else
2044 #endif
2045 	hashkey_faddr = inp->inp_faddr.s_addr;
2046 
2047 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2048 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2049 
2050 	pcbporthash = &pcbinfo->ipi_porthashbase[
2051 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2052 
2053 	/*
2054 	 * Go through port list and look for a head for this lport.
2055 	 */
2056 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2057 		if (phd->phd_port == inp->inp_lport)
2058 			break;
2059 	}
2060 	/*
2061 	 * If none exists, malloc one and tack it on.
2062 	 */
2063 	if (phd == NULL) {
2064 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2065 		if (phd == NULL) {
2066 			return (ENOBUFS); /* XXX */
2067 		}
2068 		phd->phd_port = inp->inp_lport;
2069 		LIST_INIT(&phd->phd_pcblist);
2070 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2071 	}
2072 	inp->inp_phd = phd;
2073 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2074 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2075 	inp->inp_flags |= INP_INHASHLIST;
2076 #ifdef PCBGROUP
2077 	if (do_pcbgroup_update)
2078 		in_pcbgroup_update(inp);
2079 #endif
2080 	return (0);
2081 }
2082 
2083 /*
2084  * For now, there are two public interfaces to insert an inpcb into the hash
2085  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2086  * is used only in the TCP syncache, where in_pcbinshash is called before the
2087  * full 4-tuple is set for the inpcb, and we don't want to install in the
2088  * pcbgroup until later.
2089  *
2090  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2091  * connection groups, and partially initialised inpcbs should not be exposed
2092  * to either reservation hash tables or pcbgroups.
2093  */
2094 int
2095 in_pcbinshash(struct inpcb *inp)
2096 {
2097 
2098 	return (in_pcbinshash_internal(inp, 1));
2099 }
2100 
2101 int
2102 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2103 {
2104 
2105 	return (in_pcbinshash_internal(inp, 0));
2106 }
2107 
2108 /*
2109  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2110  * changed. NOTE: This does not handle the case of the lport changing (the
2111  * hashed port list would have to be updated as well), so the lport must
2112  * not change after in_pcbinshash() has been called.
2113  */
2114 void
2115 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2116 {
2117 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2118 	struct inpcbhead *head;
2119 	u_int32_t hashkey_faddr;
2120 
2121 	INP_WLOCK_ASSERT(inp);
2122 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2123 
2124 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2125 	    ("in_pcbrehash: !INP_INHASHLIST"));
2126 
2127 #ifdef INET6
2128 	if (inp->inp_vflag & INP_IPV6)
2129 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2130 	else
2131 #endif
2132 	hashkey_faddr = inp->inp_faddr.s_addr;
2133 
2134 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2135 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2136 
2137 	LIST_REMOVE(inp, inp_hash);
2138 	LIST_INSERT_HEAD(head, inp, inp_hash);
2139 
2140 #ifdef PCBGROUP
2141 	if (m != NULL)
2142 		in_pcbgroup_update_mbuf(inp, m);
2143 	else
2144 		in_pcbgroup_update(inp);
2145 #endif
2146 }
2147 
2148 void
2149 in_pcbrehash(struct inpcb *inp)
2150 {
2151 
2152 	in_pcbrehash_mbuf(inp, NULL);
2153 }
2154 
2155 /*
2156  * Remove PCB from various lists.
2157  */
2158 static void
2159 in_pcbremlists(struct inpcb *inp)
2160 {
2161 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2162 
2163 	INP_INFO_WLOCK_ASSERT(pcbinfo);
2164 	INP_WLOCK_ASSERT(inp);
2165 
2166 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2167 	if (inp->inp_flags & INP_INHASHLIST) {
2168 		struct inpcbport *phd = inp->inp_phd;
2169 
2170 		INP_HASH_WLOCK(pcbinfo);
2171 		LIST_REMOVE(inp, inp_hash);
2172 		LIST_REMOVE(inp, inp_portlist);
2173 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2174 			LIST_REMOVE(phd, phd_hash);
2175 			free(phd, M_PCB);
2176 		}
2177 		INP_HASH_WUNLOCK(pcbinfo);
2178 		inp->inp_flags &= ~INP_INHASHLIST;
2179 	}
2180 	LIST_REMOVE(inp, inp_list);
2181 	pcbinfo->ipi_count--;
2182 #ifdef PCBGROUP
2183 	in_pcbgroup_remove(inp);
2184 #endif
2185 }
2186 
2187 /*
2188  * A set label operation has occurred at the socket layer, propagate the
2189  * label change into the in_pcb for the socket.
2190  */
2191 void
2192 in_pcbsosetlabel(struct socket *so)
2193 {
2194 #ifdef MAC
2195 	struct inpcb *inp;
2196 
2197 	inp = sotoinpcb(so);
2198 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2199 
2200 	INP_WLOCK(inp);
2201 	SOCK_LOCK(so);
2202 	mac_inpcb_sosetlabel(so, inp);
2203 	SOCK_UNLOCK(so);
2204 	INP_WUNLOCK(inp);
2205 #endif
2206 }
2207 
2208 /*
2209  * ipport_tick runs once per second, determining if random port allocation
2210  * should be continued.  If more than ipport_randomcps ports have been
2211  * allocated in the last second, then we return to sequential port
2212  * allocation. We return to random allocation only once we drop below
2213  * ipport_randomcps for at least ipport_randomtime seconds.
2214  */
2215 static void
2216 ipport_tick(void *xtp)
2217 {
2218 	VNET_ITERATOR_DECL(vnet_iter);
2219 
2220 	VNET_LIST_RLOCK_NOSLEEP();
2221 	VNET_FOREACH(vnet_iter) {
2222 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2223 		if (V_ipport_tcpallocs <=
2224 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2225 			if (V_ipport_stoprandom > 0)
2226 				V_ipport_stoprandom--;
2227 		} else
2228 			V_ipport_stoprandom = V_ipport_randomtime;
2229 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2230 		CURVNET_RESTORE();
2231 	}
2232 	VNET_LIST_RUNLOCK_NOSLEEP();
2233 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2234 }
2235 
2236 static void
2237 ip_fini(void *xtp)
2238 {
2239 
2240 	callout_stop(&ipport_tick_callout);
2241 }
2242 
2243 /*
2244  * The ipport_callout should start running at about the time we attach the
2245  * inet or inet6 domains.
2246  */
2247 static void
2248 ipport_tick_init(const void *unused __unused)
2249 {
2250 
2251 	/* Start ipport_tick. */
2252 	callout_init(&ipport_tick_callout, 1);
2253 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2254 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2255 		SHUTDOWN_PRI_DEFAULT);
2256 }
2257 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2258     ipport_tick_init, NULL);
2259 
2260 void
2261 inp_wlock(struct inpcb *inp)
2262 {
2263 
2264 	INP_WLOCK(inp);
2265 }
2266 
2267 void
2268 inp_wunlock(struct inpcb *inp)
2269 {
2270 
2271 	INP_WUNLOCK(inp);
2272 }
2273 
2274 void
2275 inp_rlock(struct inpcb *inp)
2276 {
2277 
2278 	INP_RLOCK(inp);
2279 }
2280 
2281 void
2282 inp_runlock(struct inpcb *inp)
2283 {
2284 
2285 	INP_RUNLOCK(inp);
2286 }
2287 
2288 #ifdef INVARIANTS
2289 void
2290 inp_lock_assert(struct inpcb *inp)
2291 {
2292 
2293 	INP_WLOCK_ASSERT(inp);
2294 }
2295 
2296 void
2297 inp_unlock_assert(struct inpcb *inp)
2298 {
2299 
2300 	INP_UNLOCK_ASSERT(inp);
2301 }
2302 #endif
2303 
2304 void
2305 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2306 {
2307 	struct inpcb *inp;
2308 
2309 	INP_INFO_RLOCK(&V_tcbinfo);
2310 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2311 		INP_WLOCK(inp);
2312 		func(inp, arg);
2313 		INP_WUNLOCK(inp);
2314 	}
2315 	INP_INFO_RUNLOCK(&V_tcbinfo);
2316 }
2317 
2318 struct socket *
2319 inp_inpcbtosocket(struct inpcb *inp)
2320 {
2321 
2322 	INP_WLOCK_ASSERT(inp);
2323 	return (inp->inp_socket);
2324 }
2325 
2326 struct tcpcb *
2327 inp_inpcbtotcpcb(struct inpcb *inp)
2328 {
2329 
2330 	INP_WLOCK_ASSERT(inp);
2331 	return ((struct tcpcb *)inp->inp_ppcb);
2332 }
2333 
2334 int
2335 inp_ip_tos_get(const struct inpcb *inp)
2336 {
2337 
2338 	return (inp->inp_ip_tos);
2339 }
2340 
2341 void
2342 inp_ip_tos_set(struct inpcb *inp, int val)
2343 {
2344 
2345 	inp->inp_ip_tos = val;
2346 }
2347 
2348 void
2349 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2350     uint32_t *faddr, uint16_t *fp)
2351 {
2352 
2353 	INP_LOCK_ASSERT(inp);
2354 	*laddr = inp->inp_laddr.s_addr;
2355 	*faddr = inp->inp_faddr.s_addr;
2356 	*lp = inp->inp_lport;
2357 	*fp = inp->inp_fport;
2358 }
2359 
2360 struct inpcb *
2361 so_sotoinpcb(struct socket *so)
2362 {
2363 
2364 	return (sotoinpcb(so));
2365 }
2366 
2367 struct tcpcb *
2368 so_sototcpcb(struct socket *so)
2369 {
2370 
2371 	return (sototcpcb(so));
2372 }
2373 
2374 #ifdef DDB
2375 static void
2376 db_print_indent(int indent)
2377 {
2378 	int i;
2379 
2380 	for (i = 0; i < indent; i++)
2381 		db_printf(" ");
2382 }
2383 
2384 static void
2385 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2386 {
2387 	char faddr_str[48], laddr_str[48];
2388 
2389 	db_print_indent(indent);
2390 	db_printf("%s at %p\n", name, inc);
2391 
2392 	indent += 2;
2393 
2394 #ifdef INET6
2395 	if (inc->inc_flags & INC_ISIPV6) {
2396 		/* IPv6. */
2397 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2398 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2399 	} else
2400 #endif
2401 	{
2402 		/* IPv4. */
2403 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2404 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2405 	}
2406 	db_print_indent(indent);
2407 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2408 	    ntohs(inc->inc_lport));
2409 	db_print_indent(indent);
2410 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2411 	    ntohs(inc->inc_fport));
2412 }
2413 
2414 static void
2415 db_print_inpflags(int inp_flags)
2416 {
2417 	int comma;
2418 
2419 	comma = 0;
2420 	if (inp_flags & INP_RECVOPTS) {
2421 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2422 		comma = 1;
2423 	}
2424 	if (inp_flags & INP_RECVRETOPTS) {
2425 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2426 		comma = 1;
2427 	}
2428 	if (inp_flags & INP_RECVDSTADDR) {
2429 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2430 		comma = 1;
2431 	}
2432 	if (inp_flags & INP_HDRINCL) {
2433 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2434 		comma = 1;
2435 	}
2436 	if (inp_flags & INP_HIGHPORT) {
2437 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2438 		comma = 1;
2439 	}
2440 	if (inp_flags & INP_LOWPORT) {
2441 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2442 		comma = 1;
2443 	}
2444 	if (inp_flags & INP_ANONPORT) {
2445 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2446 		comma = 1;
2447 	}
2448 	if (inp_flags & INP_RECVIF) {
2449 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2450 		comma = 1;
2451 	}
2452 	if (inp_flags & INP_MTUDISC) {
2453 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2454 		comma = 1;
2455 	}
2456 	if (inp_flags & INP_RECVTTL) {
2457 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2458 		comma = 1;
2459 	}
2460 	if (inp_flags & INP_DONTFRAG) {
2461 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2462 		comma = 1;
2463 	}
2464 	if (inp_flags & INP_RECVTOS) {
2465 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2466 		comma = 1;
2467 	}
2468 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2469 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2470 		comma = 1;
2471 	}
2472 	if (inp_flags & IN6P_PKTINFO) {
2473 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2474 		comma = 1;
2475 	}
2476 	if (inp_flags & IN6P_HOPLIMIT) {
2477 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2478 		comma = 1;
2479 	}
2480 	if (inp_flags & IN6P_HOPOPTS) {
2481 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2482 		comma = 1;
2483 	}
2484 	if (inp_flags & IN6P_DSTOPTS) {
2485 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2486 		comma = 1;
2487 	}
2488 	if (inp_flags & IN6P_RTHDR) {
2489 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2490 		comma = 1;
2491 	}
2492 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2493 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2494 		comma = 1;
2495 	}
2496 	if (inp_flags & IN6P_TCLASS) {
2497 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2498 		comma = 1;
2499 	}
2500 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2501 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2502 		comma = 1;
2503 	}
2504 	if (inp_flags & INP_TIMEWAIT) {
2505 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2506 		comma  = 1;
2507 	}
2508 	if (inp_flags & INP_ONESBCAST) {
2509 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2510 		comma  = 1;
2511 	}
2512 	if (inp_flags & INP_DROPPED) {
2513 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2514 		comma  = 1;
2515 	}
2516 	if (inp_flags & INP_SOCKREF) {
2517 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2518 		comma  = 1;
2519 	}
2520 	if (inp_flags & IN6P_RFC2292) {
2521 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2522 		comma = 1;
2523 	}
2524 	if (inp_flags & IN6P_MTU) {
2525 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2526 		comma = 1;
2527 	}
2528 }
2529 
2530 static void
2531 db_print_inpvflag(u_char inp_vflag)
2532 {
2533 	int comma;
2534 
2535 	comma = 0;
2536 	if (inp_vflag & INP_IPV4) {
2537 		db_printf("%sINP_IPV4", comma ? ", " : "");
2538 		comma  = 1;
2539 	}
2540 	if (inp_vflag & INP_IPV6) {
2541 		db_printf("%sINP_IPV6", comma ? ", " : "");
2542 		comma  = 1;
2543 	}
2544 	if (inp_vflag & INP_IPV6PROTO) {
2545 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2546 		comma  = 1;
2547 	}
2548 }
2549 
2550 static void
2551 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2552 {
2553 
2554 	db_print_indent(indent);
2555 	db_printf("%s at %p\n", name, inp);
2556 
2557 	indent += 2;
2558 
2559 	db_print_indent(indent);
2560 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2561 
2562 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2563 
2564 	db_print_indent(indent);
2565 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2566 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2567 
2568 	db_print_indent(indent);
2569 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2570 	   inp->inp_label, inp->inp_flags);
2571 	db_print_inpflags(inp->inp_flags);
2572 	db_printf(")\n");
2573 
2574 	db_print_indent(indent);
2575 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2576 	    inp->inp_vflag);
2577 	db_print_inpvflag(inp->inp_vflag);
2578 	db_printf(")\n");
2579 
2580 	db_print_indent(indent);
2581 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2582 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2583 
2584 	db_print_indent(indent);
2585 #ifdef INET6
2586 	if (inp->inp_vflag & INP_IPV6) {
2587 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2588 		    "in6p_moptions: %p\n", inp->in6p_options,
2589 		    inp->in6p_outputopts, inp->in6p_moptions);
2590 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2591 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2592 		    inp->in6p_hops);
2593 	} else
2594 #endif
2595 	{
2596 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2597 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2598 		    inp->inp_options, inp->inp_moptions);
2599 	}
2600 
2601 	db_print_indent(indent);
2602 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2603 	    (uintmax_t)inp->inp_gencnt);
2604 }
2605 
2606 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2607 {
2608 	struct inpcb *inp;
2609 
2610 	if (!have_addr) {
2611 		db_printf("usage: show inpcb <addr>\n");
2612 		return;
2613 	}
2614 	inp = (struct inpcb *)addr;
2615 
2616 	db_print_inpcb(inp, "inpcb", 0);
2617 }
2618 #endif /* DDB */
2619