xref: /freebsd/sys/netinet/in_pcb.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
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/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/callout.h>
54 #include <sys/eventhandler.h>
55 #include <sys/domain.h>
56 #include <sys/protosw.h>
57 #include <sys/rmlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/refcount.h>
63 #include <sys/jail.h>
64 #include <sys/kernel.h>
65 #include <sys/sysctl.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70 
71 #include <vm/uma.h>
72 
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/if_types.h>
76 #include <net/if_llatbl.h>
77 #include <net/route.h>
78 #include <net/rss_config.h>
79 #include <net/vnet.h>
80 
81 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/ip_var.h>
85 #include <netinet/tcp_var.h>
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
88 #endif
89 #ifdef INET
90 #include <netinet/in_var.h>
91 #endif
92 #ifdef INET6
93 #include <netinet/ip6.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/in6_var.h>
96 #include <netinet6/ip6_var.h>
97 #endif /* INET6 */
98 
99 
100 #ifdef IPSEC
101 #include <netipsec/ipsec.h>
102 #include <netipsec/key.h>
103 #endif /* IPSEC */
104 
105 #include <security/mac/mac_framework.h>
106 
107 static struct callout	ipport_tick_callout;
108 
109 /*
110  * These configure the range of local port addresses assigned to
111  * "unspecified" outgoing connections/packets/whatever.
112  */
113 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
114 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
115 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
116 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
117 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
118 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
119 
120 /*
121  * Reserved ports accessible only to root. There are significant
122  * security considerations that must be accounted for when changing these,
123  * but the security benefits can be great. Please be careful.
124  */
125 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
126 VNET_DEFINE(int, ipport_reservedlow);
127 
128 /* Variables dealing with random ephemeral port allocation. */
129 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
130 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
131 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
132 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
133 VNET_DEFINE(int, ipport_tcpallocs);
134 static VNET_DEFINE(int, ipport_tcplastcount);
135 
136 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
137 
138 static void	in_pcbremlists(struct inpcb *inp);
139 #ifdef INET
140 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
141 			    struct in_addr faddr, u_int fport_arg,
142 			    struct in_addr laddr, u_int lport_arg,
143 			    int lookupflags, struct ifnet *ifp);
144 
145 #define RANGECHK(var, min, max) \
146 	if ((var) < (min)) { (var) = (min); } \
147 	else if ((var) > (max)) { (var) = (max); }
148 
149 static int
150 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
151 {
152 	int error;
153 
154 	error = sysctl_handle_int(oidp, arg1, arg2, req);
155 	if (error == 0) {
156 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
157 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
158 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
159 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
160 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
161 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
162 	}
163 	return (error);
164 }
165 
166 #undef RANGECHK
167 
168 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
169     "IP Ports");
170 
171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
172 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
173 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
175 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
176 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
177 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
178 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
179 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
180 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
181 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
182 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
184 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
185 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
186 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
187 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
188 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
189 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
190 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
191 	&VNET_NAME(ipport_reservedhigh), 0, "");
192 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
193 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
194 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
195 	CTLFLAG_VNET | CTLFLAG_RW,
196 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
198 	CTLFLAG_VNET | CTLFLAG_RW,
199 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
200 	"allocations before switching to a sequental one");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
202 	CTLFLAG_VNET | CTLFLAG_RW,
203 	&VNET_NAME(ipport_randomtime), 0,
204 	"Minimum time to keep sequental port "
205 	"allocation before switching to a random one");
206 #endif /* INET */
207 
208 /*
209  * in_pcb.c: manage the Protocol Control Blocks.
210  *
211  * NOTE: It is assumed that most of these functions will be called with
212  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
213  * functions often modify hash chains or addresses in pcbs.
214  */
215 
216 /*
217  * Initialize an inpcbinfo -- we should be able to reduce the number of
218  * arguments in time.
219  */
220 void
221 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
222     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
223     char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
224     uint32_t inpcbzone_flags, u_int hashfields)
225 {
226 
227 	INP_INFO_LOCK_INIT(pcbinfo, name);
228 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
229 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
230 #ifdef VIMAGE
231 	pcbinfo->ipi_vnet = curvnet;
232 #endif
233 	pcbinfo->ipi_listhead = listhead;
234 	LIST_INIT(pcbinfo->ipi_listhead);
235 	pcbinfo->ipi_count = 0;
236 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
237 	    &pcbinfo->ipi_hashmask);
238 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
239 	    &pcbinfo->ipi_porthashmask);
240 #ifdef PCBGROUP
241 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
242 #endif
243 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
244 	    NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
245 	    inpcbzone_flags);
246 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
247 	uma_zone_set_warning(pcbinfo->ipi_zone,
248 	    "kern.ipc.maxsockets limit reached");
249 }
250 
251 /*
252  * Destroy an inpcbinfo.
253  */
254 void
255 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
256 {
257 
258 	KASSERT(pcbinfo->ipi_count == 0,
259 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
260 
261 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
262 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
263 	    pcbinfo->ipi_porthashmask);
264 #ifdef PCBGROUP
265 	in_pcbgroup_destroy(pcbinfo);
266 #endif
267 	uma_zdestroy(pcbinfo->ipi_zone);
268 	INP_LIST_LOCK_DESTROY(pcbinfo);
269 	INP_HASH_LOCK_DESTROY(pcbinfo);
270 	INP_INFO_LOCK_DESTROY(pcbinfo);
271 }
272 
273 /*
274  * Allocate a PCB and associate it with the socket.
275  * On success return with the PCB locked.
276  */
277 int
278 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
279 {
280 	struct inpcb *inp;
281 	int error;
282 
283 #ifdef INVARIANTS
284 	if (pcbinfo == &V_tcbinfo) {
285 		INP_INFO_RLOCK_ASSERT(pcbinfo);
286 	} else {
287 		INP_INFO_WLOCK_ASSERT(pcbinfo);
288 	}
289 #endif
290 
291 	error = 0;
292 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
293 	if (inp == NULL)
294 		return (ENOBUFS);
295 	bzero(inp, inp_zero_size);
296 	inp->inp_pcbinfo = pcbinfo;
297 	inp->inp_socket = so;
298 	inp->inp_cred = crhold(so->so_cred);
299 	inp->inp_inc.inc_fibnum = so->so_fibnum;
300 #ifdef MAC
301 	error = mac_inpcb_init(inp, M_NOWAIT);
302 	if (error != 0)
303 		goto out;
304 	mac_inpcb_create(so, inp);
305 #endif
306 #ifdef IPSEC
307 	error = ipsec_init_policy(so, &inp->inp_sp);
308 	if (error != 0) {
309 #ifdef MAC
310 		mac_inpcb_destroy(inp);
311 #endif
312 		goto out;
313 	}
314 #endif /*IPSEC*/
315 #ifdef INET6
316 	if (INP_SOCKAF(so) == AF_INET6) {
317 		inp->inp_vflag |= INP_IPV6PROTO;
318 		if (V_ip6_v6only)
319 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
320 	}
321 #endif
322 	INP_WLOCK(inp);
323 	INP_LIST_WLOCK(pcbinfo);
324 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
325 	pcbinfo->ipi_count++;
326 	so->so_pcb = (caddr_t)inp;
327 #ifdef INET6
328 	if (V_ip6_auto_flowlabel)
329 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
330 #endif
331 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
332 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
333 	INP_LIST_WUNLOCK(pcbinfo);
334 #if defined(IPSEC) || defined(MAC)
335 out:
336 	if (error != 0) {
337 		crfree(inp->inp_cred);
338 		uma_zfree(pcbinfo->ipi_zone, inp);
339 	}
340 #endif
341 	return (error);
342 }
343 
344 #ifdef INET
345 int
346 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
347 {
348 	int anonport, error;
349 
350 	INP_WLOCK_ASSERT(inp);
351 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
352 
353 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
354 		return (EINVAL);
355 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
356 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
357 	    &inp->inp_lport, cred);
358 	if (error)
359 		return (error);
360 	if (in_pcbinshash(inp) != 0) {
361 		inp->inp_laddr.s_addr = INADDR_ANY;
362 		inp->inp_lport = 0;
363 		return (EAGAIN);
364 	}
365 	if (anonport)
366 		inp->inp_flags |= INP_ANONPORT;
367 	return (0);
368 }
369 #endif
370 
371 /*
372  * Select a local port (number) to use.
373  */
374 #if defined(INET) || defined(INET6)
375 int
376 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
377     struct ucred *cred, int lookupflags)
378 {
379 	struct inpcbinfo *pcbinfo;
380 	struct inpcb *tmpinp;
381 	unsigned short *lastport;
382 	int count, dorandom, error;
383 	u_short aux, first, last, lport;
384 #ifdef INET
385 	struct in_addr laddr;
386 #endif
387 
388 	pcbinfo = inp->inp_pcbinfo;
389 
390 	/*
391 	 * Because no actual state changes occur here, a global write lock on
392 	 * the pcbinfo isn't required.
393 	 */
394 	INP_LOCK_ASSERT(inp);
395 	INP_HASH_LOCK_ASSERT(pcbinfo);
396 
397 	if (inp->inp_flags & INP_HIGHPORT) {
398 		first = V_ipport_hifirstauto;	/* sysctl */
399 		last  = V_ipport_hilastauto;
400 		lastport = &pcbinfo->ipi_lasthi;
401 	} else if (inp->inp_flags & INP_LOWPORT) {
402 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
403 		if (error)
404 			return (error);
405 		first = V_ipport_lowfirstauto;	/* 1023 */
406 		last  = V_ipport_lowlastauto;	/* 600 */
407 		lastport = &pcbinfo->ipi_lastlow;
408 	} else {
409 		first = V_ipport_firstauto;	/* sysctl */
410 		last  = V_ipport_lastauto;
411 		lastport = &pcbinfo->ipi_lastport;
412 	}
413 	/*
414 	 * For UDP(-Lite), use random port allocation as long as the user
415 	 * allows it.  For TCP (and as of yet unknown) connections,
416 	 * use random port allocation only if the user allows it AND
417 	 * ipport_tick() allows it.
418 	 */
419 	if (V_ipport_randomized &&
420 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
421 		pcbinfo == &V_ulitecbinfo))
422 		dorandom = 1;
423 	else
424 		dorandom = 0;
425 	/*
426 	 * It makes no sense to do random port allocation if
427 	 * we have the only port available.
428 	 */
429 	if (first == last)
430 		dorandom = 0;
431 	/* Make sure to not include UDP(-Lite) packets in the count. */
432 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
433 		V_ipport_tcpallocs++;
434 	/*
435 	 * Instead of having two loops further down counting up or down
436 	 * make sure that first is always <= last and go with only one
437 	 * code path implementing all logic.
438 	 */
439 	if (first > last) {
440 		aux = first;
441 		first = last;
442 		last = aux;
443 	}
444 
445 #ifdef INET
446 	/* Make the compiler happy. */
447 	laddr.s_addr = 0;
448 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
449 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
450 		    __func__, inp));
451 		laddr = *laddrp;
452 	}
453 #endif
454 	tmpinp = NULL;	/* Make compiler happy. */
455 	lport = *lportp;
456 
457 	if (dorandom)
458 		*lastport = first + (arc4random() % (last - first));
459 
460 	count = last - first;
461 
462 	do {
463 		if (count-- < 0)	/* completely used? */
464 			return (EADDRNOTAVAIL);
465 		++*lastport;
466 		if (*lastport < first || *lastport > last)
467 			*lastport = first;
468 		lport = htons(*lastport);
469 
470 #ifdef INET6
471 		if ((inp->inp_vflag & INP_IPV6) != 0)
472 			tmpinp = in6_pcblookup_local(pcbinfo,
473 			    &inp->in6p_laddr, lport, lookupflags, cred);
474 #endif
475 #if defined(INET) && defined(INET6)
476 		else
477 #endif
478 #ifdef INET
479 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
480 			    lport, lookupflags, cred);
481 #endif
482 	} while (tmpinp != NULL);
483 
484 #ifdef INET
485 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
486 		laddrp->s_addr = laddr.s_addr;
487 #endif
488 	*lportp = lport;
489 
490 	return (0);
491 }
492 
493 /*
494  * Return cached socket options.
495  */
496 short
497 inp_so_options(const struct inpcb *inp)
498 {
499    short so_options;
500 
501    so_options = 0;
502 
503    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
504 	   so_options |= SO_REUSEPORT;
505    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
506 	   so_options |= SO_REUSEADDR;
507    return (so_options);
508 }
509 #endif /* INET || INET6 */
510 
511 /*
512  * Check if a new BINDMULTI socket is allowed to be created.
513  *
514  * ni points to the new inp.
515  * oi points to the exisitng inp.
516  *
517  * This checks whether the existing inp also has BINDMULTI and
518  * whether the credentials match.
519  */
520 int
521 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
522 {
523 	/* Check permissions match */
524 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
525 	    (ni->inp_cred->cr_uid !=
526 	    oi->inp_cred->cr_uid))
527 		return (0);
528 
529 	/* Check the existing inp has BINDMULTI set */
530 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
531 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
532 		return (0);
533 
534 	/*
535 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
536 	 * it is and it matches the checks.
537 	 */
538 	return (1);
539 }
540 
541 #ifdef INET
542 /*
543  * Set up a bind operation on a PCB, performing port allocation
544  * as required, but do not actually modify the PCB. Callers can
545  * either complete the bind by setting inp_laddr/inp_lport and
546  * calling in_pcbinshash(), or they can just use the resulting
547  * port and address to authorise the sending of a once-off packet.
548  *
549  * On error, the values of *laddrp and *lportp are not changed.
550  */
551 int
552 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
553     u_short *lportp, struct ucred *cred)
554 {
555 	struct socket *so = inp->inp_socket;
556 	struct sockaddr_in *sin;
557 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
558 	struct in_addr laddr;
559 	u_short lport = 0;
560 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
561 	int error;
562 
563 	/*
564 	 * No state changes, so read locks are sufficient here.
565 	 */
566 	INP_LOCK_ASSERT(inp);
567 	INP_HASH_LOCK_ASSERT(pcbinfo);
568 
569 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
570 		return (EADDRNOTAVAIL);
571 	laddr.s_addr = *laddrp;
572 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
573 		return (EINVAL);
574 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
575 		lookupflags = INPLOOKUP_WILDCARD;
576 	if (nam == NULL) {
577 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
578 			return (error);
579 	} else {
580 		sin = (struct sockaddr_in *)nam;
581 		if (nam->sa_len != sizeof (*sin))
582 			return (EINVAL);
583 #ifdef notdef
584 		/*
585 		 * We should check the family, but old programs
586 		 * incorrectly fail to initialize it.
587 		 */
588 		if (sin->sin_family != AF_INET)
589 			return (EAFNOSUPPORT);
590 #endif
591 		error = prison_local_ip4(cred, &sin->sin_addr);
592 		if (error)
593 			return (error);
594 		if (sin->sin_port != *lportp) {
595 			/* Don't allow the port to change. */
596 			if (*lportp != 0)
597 				return (EINVAL);
598 			lport = sin->sin_port;
599 		}
600 		/* NB: lport is left as 0 if the port isn't being changed. */
601 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
602 			/*
603 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
604 			 * allow complete duplication of binding if
605 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
606 			 * and a multicast address is bound on both
607 			 * new and duplicated sockets.
608 			 */
609 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
610 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
611 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
612 			sin->sin_port = 0;		/* yech... */
613 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
614 			/*
615 			 * Is the address a local IP address?
616 			 * If INP_BINDANY is set, then the socket may be bound
617 			 * to any endpoint address, local or not.
618 			 */
619 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
620 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
621 				return (EADDRNOTAVAIL);
622 		}
623 		laddr = sin->sin_addr;
624 		if (lport) {
625 			struct inpcb *t;
626 			struct tcptw *tw;
627 
628 			/* GROSS */
629 			if (ntohs(lport) <= V_ipport_reservedhigh &&
630 			    ntohs(lport) >= V_ipport_reservedlow &&
631 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
632 			    0))
633 				return (EACCES);
634 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
635 			    priv_check_cred(inp->inp_cred,
636 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
637 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
638 				    lport, INPLOOKUP_WILDCARD, cred);
639 	/*
640 	 * XXX
641 	 * This entire block sorely needs a rewrite.
642 	 */
643 				if (t &&
644 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
645 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
646 				    (so->so_type != SOCK_STREAM ||
647 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
648 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
649 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
650 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
651 				    (inp->inp_cred->cr_uid !=
652 				     t->inp_cred->cr_uid))
653 					return (EADDRINUSE);
654 
655 				/*
656 				 * If the socket is a BINDMULTI socket, then
657 				 * the credentials need to match and the
658 				 * original socket also has to have been bound
659 				 * with BINDMULTI.
660 				 */
661 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
662 					return (EADDRINUSE);
663 			}
664 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
665 			    lport, lookupflags, cred);
666 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
667 				/*
668 				 * XXXRW: If an incpb has had its timewait
669 				 * state recycled, we treat the address as
670 				 * being in use (for now).  This is better
671 				 * than a panic, but not desirable.
672 				 */
673 				tw = intotw(t);
674 				if (tw == NULL ||
675 				    (reuseport & tw->tw_so_options) == 0)
676 					return (EADDRINUSE);
677 			} else if (t &&
678 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
679 			    (reuseport & inp_so_options(t)) == 0) {
680 #ifdef INET6
681 				if (ntohl(sin->sin_addr.s_addr) !=
682 				    INADDR_ANY ||
683 				    ntohl(t->inp_laddr.s_addr) !=
684 				    INADDR_ANY ||
685 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
686 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
687 #endif
688 				return (EADDRINUSE);
689 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
690 					return (EADDRINUSE);
691 			}
692 		}
693 	}
694 	if (*lportp != 0)
695 		lport = *lportp;
696 	if (lport == 0) {
697 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
698 		if (error != 0)
699 			return (error);
700 
701 	}
702 	*laddrp = laddr.s_addr;
703 	*lportp = lport;
704 	return (0);
705 }
706 
707 /*
708  * Connect from a socket to a specified address.
709  * Both address and port must be specified in argument sin.
710  * If don't have a local address for this socket yet,
711  * then pick one.
712  */
713 int
714 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
715     struct ucred *cred, struct mbuf *m)
716 {
717 	u_short lport, fport;
718 	in_addr_t laddr, faddr;
719 	int anonport, error;
720 
721 	INP_WLOCK_ASSERT(inp);
722 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
723 
724 	lport = inp->inp_lport;
725 	laddr = inp->inp_laddr.s_addr;
726 	anonport = (lport == 0);
727 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
728 	    NULL, cred);
729 	if (error)
730 		return (error);
731 
732 	/* Do the initial binding of the local address if required. */
733 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
734 		inp->inp_lport = lport;
735 		inp->inp_laddr.s_addr = laddr;
736 		if (in_pcbinshash(inp) != 0) {
737 			inp->inp_laddr.s_addr = INADDR_ANY;
738 			inp->inp_lport = 0;
739 			return (EAGAIN);
740 		}
741 	}
742 
743 	/* Commit the remaining changes. */
744 	inp->inp_lport = lport;
745 	inp->inp_laddr.s_addr = laddr;
746 	inp->inp_faddr.s_addr = faddr;
747 	inp->inp_fport = fport;
748 	in_pcbrehash_mbuf(inp, m);
749 
750 	if (anonport)
751 		inp->inp_flags |= INP_ANONPORT;
752 	return (0);
753 }
754 
755 int
756 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
757 {
758 
759 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
760 }
761 
762 /*
763  * Do proper source address selection on an unbound socket in case
764  * of connect. Take jails into account as well.
765  */
766 int
767 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
768     struct ucred *cred)
769 {
770 	struct ifaddr *ifa;
771 	struct sockaddr *sa;
772 	struct sockaddr_in *sin;
773 	struct route sro;
774 	int error;
775 
776 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
777 
778 	/*
779 	 * Bypass source address selection and use the primary jail IP
780 	 * if requested.
781 	 */
782 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
783 		return (0);
784 
785 	error = 0;
786 	bzero(&sro, sizeof(sro));
787 
788 	sin = (struct sockaddr_in *)&sro.ro_dst;
789 	sin->sin_family = AF_INET;
790 	sin->sin_len = sizeof(struct sockaddr_in);
791 	sin->sin_addr.s_addr = faddr->s_addr;
792 
793 	/*
794 	 * If route is known our src addr is taken from the i/f,
795 	 * else punt.
796 	 *
797 	 * Find out route to destination.
798 	 */
799 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
800 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
801 
802 	/*
803 	 * If we found a route, use the address corresponding to
804 	 * the outgoing interface.
805 	 *
806 	 * Otherwise assume faddr is reachable on a directly connected
807 	 * network and try to find a corresponding interface to take
808 	 * the source address from.
809 	 */
810 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
811 		struct in_ifaddr *ia;
812 		struct ifnet *ifp;
813 
814 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
815 					inp->inp_socket->so_fibnum));
816 		if (ia == NULL)
817 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
818 						inp->inp_socket->so_fibnum));
819 		if (ia == NULL) {
820 			error = ENETUNREACH;
821 			goto done;
822 		}
823 
824 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
825 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
826 			ifa_free(&ia->ia_ifa);
827 			goto done;
828 		}
829 
830 		ifp = ia->ia_ifp;
831 		ifa_free(&ia->ia_ifa);
832 		ia = NULL;
833 		IF_ADDR_RLOCK(ifp);
834 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
835 
836 			sa = ifa->ifa_addr;
837 			if (sa->sa_family != AF_INET)
838 				continue;
839 			sin = (struct sockaddr_in *)sa;
840 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
841 				ia = (struct in_ifaddr *)ifa;
842 				break;
843 			}
844 		}
845 		if (ia != NULL) {
846 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
847 			IF_ADDR_RUNLOCK(ifp);
848 			goto done;
849 		}
850 		IF_ADDR_RUNLOCK(ifp);
851 
852 		/* 3. As a last resort return the 'default' jail address. */
853 		error = prison_get_ip4(cred, laddr);
854 		goto done;
855 	}
856 
857 	/*
858 	 * If the outgoing interface on the route found is not
859 	 * a loopback interface, use the address from that interface.
860 	 * In case of jails do those three steps:
861 	 * 1. check if the interface address belongs to the jail. If so use it.
862 	 * 2. check if we have any address on the outgoing interface
863 	 *    belonging to this jail. If so use it.
864 	 * 3. as a last resort return the 'default' jail address.
865 	 */
866 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
867 		struct in_ifaddr *ia;
868 		struct ifnet *ifp;
869 
870 		/* If not jailed, use the default returned. */
871 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
872 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
873 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
874 			goto done;
875 		}
876 
877 		/* Jailed. */
878 		/* 1. Check if the iface address belongs to the jail. */
879 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
880 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
881 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
882 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
883 			goto done;
884 		}
885 
886 		/*
887 		 * 2. Check if we have any address on the outgoing interface
888 		 *    belonging to this jail.
889 		 */
890 		ia = NULL;
891 		ifp = sro.ro_rt->rt_ifp;
892 		IF_ADDR_RLOCK(ifp);
893 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
894 			sa = ifa->ifa_addr;
895 			if (sa->sa_family != AF_INET)
896 				continue;
897 			sin = (struct sockaddr_in *)sa;
898 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
899 				ia = (struct in_ifaddr *)ifa;
900 				break;
901 			}
902 		}
903 		if (ia != NULL) {
904 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
905 			IF_ADDR_RUNLOCK(ifp);
906 			goto done;
907 		}
908 		IF_ADDR_RUNLOCK(ifp);
909 
910 		/* 3. As a last resort return the 'default' jail address. */
911 		error = prison_get_ip4(cred, laddr);
912 		goto done;
913 	}
914 
915 	/*
916 	 * The outgoing interface is marked with 'loopback net', so a route
917 	 * to ourselves is here.
918 	 * Try to find the interface of the destination address and then
919 	 * take the address from there. That interface is not necessarily
920 	 * a loopback interface.
921 	 * In case of jails, check that it is an address of the jail
922 	 * and if we cannot find, fall back to the 'default' jail address.
923 	 */
924 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
925 		struct sockaddr_in sain;
926 		struct in_ifaddr *ia;
927 
928 		bzero(&sain, sizeof(struct sockaddr_in));
929 		sain.sin_family = AF_INET;
930 		sain.sin_len = sizeof(struct sockaddr_in);
931 		sain.sin_addr.s_addr = faddr->s_addr;
932 
933 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
934 					inp->inp_socket->so_fibnum));
935 		if (ia == NULL)
936 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
937 						inp->inp_socket->so_fibnum));
938 		if (ia == NULL)
939 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
940 
941 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
942 			if (ia == NULL) {
943 				error = ENETUNREACH;
944 				goto done;
945 			}
946 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
947 			ifa_free(&ia->ia_ifa);
948 			goto done;
949 		}
950 
951 		/* Jailed. */
952 		if (ia != NULL) {
953 			struct ifnet *ifp;
954 
955 			ifp = ia->ia_ifp;
956 			ifa_free(&ia->ia_ifa);
957 			ia = NULL;
958 			IF_ADDR_RLOCK(ifp);
959 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
960 
961 				sa = ifa->ifa_addr;
962 				if (sa->sa_family != AF_INET)
963 					continue;
964 				sin = (struct sockaddr_in *)sa;
965 				if (prison_check_ip4(cred,
966 				    &sin->sin_addr) == 0) {
967 					ia = (struct in_ifaddr *)ifa;
968 					break;
969 				}
970 			}
971 			if (ia != NULL) {
972 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
973 				IF_ADDR_RUNLOCK(ifp);
974 				goto done;
975 			}
976 			IF_ADDR_RUNLOCK(ifp);
977 		}
978 
979 		/* 3. As a last resort return the 'default' jail address. */
980 		error = prison_get_ip4(cred, laddr);
981 		goto done;
982 	}
983 
984 done:
985 	if (sro.ro_rt != NULL)
986 		RTFREE(sro.ro_rt);
987 	return (error);
988 }
989 
990 /*
991  * Set up for a connect from a socket to the specified address.
992  * On entry, *laddrp and *lportp should contain the current local
993  * address and port for the PCB; these are updated to the values
994  * that should be placed in inp_laddr and inp_lport to complete
995  * the connect.
996  *
997  * On success, *faddrp and *fportp will be set to the remote address
998  * and port. These are not updated in the error case.
999  *
1000  * If the operation fails because the connection already exists,
1001  * *oinpp will be set to the PCB of that connection so that the
1002  * caller can decide to override it. In all other cases, *oinpp
1003  * is set to NULL.
1004  */
1005 int
1006 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1007     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1008     struct inpcb **oinpp, struct ucred *cred)
1009 {
1010 	struct rm_priotracker in_ifa_tracker;
1011 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1012 	struct in_ifaddr *ia;
1013 	struct inpcb *oinp;
1014 	struct in_addr laddr, faddr;
1015 	u_short lport, fport;
1016 	int error;
1017 
1018 	/*
1019 	 * Because a global state change doesn't actually occur here, a read
1020 	 * lock is sufficient.
1021 	 */
1022 	INP_LOCK_ASSERT(inp);
1023 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1024 
1025 	if (oinpp != NULL)
1026 		*oinpp = NULL;
1027 	if (nam->sa_len != sizeof (*sin))
1028 		return (EINVAL);
1029 	if (sin->sin_family != AF_INET)
1030 		return (EAFNOSUPPORT);
1031 	if (sin->sin_port == 0)
1032 		return (EADDRNOTAVAIL);
1033 	laddr.s_addr = *laddrp;
1034 	lport = *lportp;
1035 	faddr = sin->sin_addr;
1036 	fport = sin->sin_port;
1037 
1038 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1039 		/*
1040 		 * If the destination address is INADDR_ANY,
1041 		 * use the primary local address.
1042 		 * If the supplied address is INADDR_BROADCAST,
1043 		 * and the primary interface supports broadcast,
1044 		 * choose the broadcast address for that interface.
1045 		 */
1046 		if (faddr.s_addr == INADDR_ANY) {
1047 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1048 			faddr =
1049 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1050 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1051 			if (cred != NULL &&
1052 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1053 				return (error);
1054 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1055 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1056 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1057 			    IFF_BROADCAST)
1058 				faddr = satosin(&TAILQ_FIRST(
1059 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1060 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1061 		}
1062 	}
1063 	if (laddr.s_addr == INADDR_ANY) {
1064 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1065 		/*
1066 		 * If the destination address is multicast and an outgoing
1067 		 * interface has been set as a multicast option, prefer the
1068 		 * address of that interface as our source address.
1069 		 */
1070 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1071 		    inp->inp_moptions != NULL) {
1072 			struct ip_moptions *imo;
1073 			struct ifnet *ifp;
1074 
1075 			imo = inp->inp_moptions;
1076 			if (imo->imo_multicast_ifp != NULL) {
1077 				ifp = imo->imo_multicast_ifp;
1078 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1079 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1080 					if ((ia->ia_ifp == ifp) &&
1081 					    (cred == NULL ||
1082 					    prison_check_ip4(cred,
1083 					    &ia->ia_addr.sin_addr) == 0))
1084 						break;
1085 				}
1086 				if (ia == NULL)
1087 					error = EADDRNOTAVAIL;
1088 				else {
1089 					laddr = ia->ia_addr.sin_addr;
1090 					error = 0;
1091 				}
1092 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1093 			}
1094 		}
1095 		if (error)
1096 			return (error);
1097 	}
1098 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1099 	    laddr, lport, 0, NULL);
1100 	if (oinp != NULL) {
1101 		if (oinpp != NULL)
1102 			*oinpp = oinp;
1103 		return (EADDRINUSE);
1104 	}
1105 	if (lport == 0) {
1106 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1107 		    cred);
1108 		if (error)
1109 			return (error);
1110 	}
1111 	*laddrp = laddr.s_addr;
1112 	*lportp = lport;
1113 	*faddrp = faddr.s_addr;
1114 	*fportp = fport;
1115 	return (0);
1116 }
1117 
1118 void
1119 in_pcbdisconnect(struct inpcb *inp)
1120 {
1121 
1122 	INP_WLOCK_ASSERT(inp);
1123 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1124 
1125 	inp->inp_faddr.s_addr = INADDR_ANY;
1126 	inp->inp_fport = 0;
1127 	in_pcbrehash(inp);
1128 }
1129 #endif /* INET */
1130 
1131 /*
1132  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1133  * For most protocols, this will be invoked immediately prior to calling
1134  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1135  * socket, in which case in_pcbfree() is deferred.
1136  */
1137 void
1138 in_pcbdetach(struct inpcb *inp)
1139 {
1140 
1141 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1142 
1143 	inp->inp_socket->so_pcb = NULL;
1144 	inp->inp_socket = NULL;
1145 }
1146 
1147 /*
1148  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1149  * stability of an inpcb pointer despite the inpcb lock being released.  This
1150  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1151  * but where the inpcb lock may already held, or when acquiring a reference
1152  * via a pcbgroup.
1153  *
1154  * in_pcbref() should be used only to provide brief memory stability, and
1155  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1156  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1157  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1158  * lock and rele are the *only* safe operations that may be performed on the
1159  * inpcb.
1160  *
1161  * While the inpcb will not be freed, releasing the inpcb lock means that the
1162  * connection's state may change, so the caller should be careful to
1163  * revalidate any cached state on reacquiring the lock.  Drop the reference
1164  * using in_pcbrele().
1165  */
1166 void
1167 in_pcbref(struct inpcb *inp)
1168 {
1169 
1170 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1171 
1172 	refcount_acquire(&inp->inp_refcount);
1173 }
1174 
1175 /*
1176  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1177  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1178  * return a flag indicating whether or not the inpcb remains valid.  If it is
1179  * valid, we return with the inpcb lock held.
1180  *
1181  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1182  * reference on an inpcb.  Historically more work was done here (actually, in
1183  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1184  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1185  * about memory stability (and continued use of the write lock).
1186  */
1187 int
1188 in_pcbrele_rlocked(struct inpcb *inp)
1189 {
1190 	struct inpcbinfo *pcbinfo;
1191 
1192 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1193 
1194 	INP_RLOCK_ASSERT(inp);
1195 
1196 	if (refcount_release(&inp->inp_refcount) == 0) {
1197 		/*
1198 		 * If the inpcb has been freed, let the caller know, even if
1199 		 * this isn't the last reference.
1200 		 */
1201 		if (inp->inp_flags2 & INP_FREED) {
1202 			INP_RUNLOCK(inp);
1203 			return (1);
1204 		}
1205 		return (0);
1206 	}
1207 
1208 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1209 
1210 	INP_RUNLOCK(inp);
1211 	pcbinfo = inp->inp_pcbinfo;
1212 	uma_zfree(pcbinfo->ipi_zone, inp);
1213 	return (1);
1214 }
1215 
1216 int
1217 in_pcbrele_wlocked(struct inpcb *inp)
1218 {
1219 	struct inpcbinfo *pcbinfo;
1220 
1221 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1222 
1223 	INP_WLOCK_ASSERT(inp);
1224 
1225 	if (refcount_release(&inp->inp_refcount) == 0) {
1226 		/*
1227 		 * If the inpcb has been freed, let the caller know, even if
1228 		 * this isn't the last reference.
1229 		 */
1230 		if (inp->inp_flags2 & INP_FREED) {
1231 			INP_WUNLOCK(inp);
1232 			return (1);
1233 		}
1234 		return (0);
1235 	}
1236 
1237 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1238 
1239 	INP_WUNLOCK(inp);
1240 	pcbinfo = inp->inp_pcbinfo;
1241 	uma_zfree(pcbinfo->ipi_zone, inp);
1242 	return (1);
1243 }
1244 
1245 /*
1246  * Temporary wrapper.
1247  */
1248 int
1249 in_pcbrele(struct inpcb *inp)
1250 {
1251 
1252 	return (in_pcbrele_wlocked(inp));
1253 }
1254 
1255 /*
1256  * Unconditionally schedule an inpcb to be freed by decrementing its
1257  * reference count, which should occur only after the inpcb has been detached
1258  * from its socket.  If another thread holds a temporary reference (acquired
1259  * using in_pcbref()) then the free is deferred until that reference is
1260  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1261  * work, including removal from global lists, is done in this context, where
1262  * the pcbinfo lock is held.
1263  */
1264 void
1265 in_pcbfree(struct inpcb *inp)
1266 {
1267 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1268 
1269 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1270 
1271 #ifdef INVARIANTS
1272 	if (pcbinfo == &V_tcbinfo) {
1273 		INP_INFO_LOCK_ASSERT(pcbinfo);
1274 	} else {
1275 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1276 	}
1277 #endif
1278 	INP_WLOCK_ASSERT(inp);
1279 
1280 	/* XXXRW: Do as much as possible here. */
1281 #ifdef IPSEC
1282 	if (inp->inp_sp != NULL)
1283 		ipsec_delete_pcbpolicy(inp);
1284 #endif
1285 	INP_LIST_WLOCK(pcbinfo);
1286 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1287 	in_pcbremlists(inp);
1288 	INP_LIST_WUNLOCK(pcbinfo);
1289 #ifdef INET6
1290 	if (inp->inp_vflag & INP_IPV6PROTO) {
1291 		ip6_freepcbopts(inp->in6p_outputopts);
1292 		if (inp->in6p_moptions != NULL)
1293 			ip6_freemoptions(inp->in6p_moptions);
1294 	}
1295 #endif
1296 	if (inp->inp_options)
1297 		(void)m_free(inp->inp_options);
1298 #ifdef INET
1299 	if (inp->inp_moptions != NULL)
1300 		inp_freemoptions(inp->inp_moptions);
1301 #endif
1302 	if (inp->inp_route.ro_rt) {
1303 		RTFREE(inp->inp_route.ro_rt);
1304 		inp->inp_route.ro_rt = (struct rtentry *)NULL;
1305 	}
1306 	if (inp->inp_route.ro_lle)
1307 		LLE_FREE(inp->inp_route.ro_lle);	/* zeros ro_lle */
1308 
1309 	inp->inp_vflag = 0;
1310 	inp->inp_flags2 |= INP_FREED;
1311 	crfree(inp->inp_cred);
1312 #ifdef MAC
1313 	mac_inpcb_destroy(inp);
1314 #endif
1315 	if (!in_pcbrele_wlocked(inp))
1316 		INP_WUNLOCK(inp);
1317 }
1318 
1319 /*
1320  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1321  * port reservation, and preventing it from being returned by inpcb lookups.
1322  *
1323  * It is used by TCP to mark an inpcb as unused and avoid future packet
1324  * delivery or event notification when a socket remains open but TCP has
1325  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1326  * or a RST on the wire, and allows the port binding to be reused while still
1327  * maintaining the invariant that so_pcb always points to a valid inpcb until
1328  * in_pcbdetach().
1329  *
1330  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1331  * in_pcbnotifyall() and in_pcbpurgeif0()?
1332  */
1333 void
1334 in_pcbdrop(struct inpcb *inp)
1335 {
1336 
1337 	INP_WLOCK_ASSERT(inp);
1338 
1339 	/*
1340 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1341 	 * the hash lock...?
1342 	 */
1343 	inp->inp_flags |= INP_DROPPED;
1344 	if (inp->inp_flags & INP_INHASHLIST) {
1345 		struct inpcbport *phd = inp->inp_phd;
1346 
1347 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1348 		LIST_REMOVE(inp, inp_hash);
1349 		LIST_REMOVE(inp, inp_portlist);
1350 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1351 			LIST_REMOVE(phd, phd_hash);
1352 			free(phd, M_PCB);
1353 		}
1354 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1355 		inp->inp_flags &= ~INP_INHASHLIST;
1356 #ifdef PCBGROUP
1357 		in_pcbgroup_remove(inp);
1358 #endif
1359 	}
1360 }
1361 
1362 #ifdef INET
1363 /*
1364  * Common routines to return the socket addresses associated with inpcbs.
1365  */
1366 struct sockaddr *
1367 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1368 {
1369 	struct sockaddr_in *sin;
1370 
1371 	sin = malloc(sizeof *sin, M_SONAME,
1372 		M_WAITOK | M_ZERO);
1373 	sin->sin_family = AF_INET;
1374 	sin->sin_len = sizeof(*sin);
1375 	sin->sin_addr = *addr_p;
1376 	sin->sin_port = port;
1377 
1378 	return (struct sockaddr *)sin;
1379 }
1380 
1381 int
1382 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1383 {
1384 	struct inpcb *inp;
1385 	struct in_addr addr;
1386 	in_port_t port;
1387 
1388 	inp = sotoinpcb(so);
1389 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1390 
1391 	INP_RLOCK(inp);
1392 	port = inp->inp_lport;
1393 	addr = inp->inp_laddr;
1394 	INP_RUNLOCK(inp);
1395 
1396 	*nam = in_sockaddr(port, &addr);
1397 	return 0;
1398 }
1399 
1400 int
1401 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1402 {
1403 	struct inpcb *inp;
1404 	struct in_addr addr;
1405 	in_port_t port;
1406 
1407 	inp = sotoinpcb(so);
1408 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1409 
1410 	INP_RLOCK(inp);
1411 	port = inp->inp_fport;
1412 	addr = inp->inp_faddr;
1413 	INP_RUNLOCK(inp);
1414 
1415 	*nam = in_sockaddr(port, &addr);
1416 	return 0;
1417 }
1418 
1419 void
1420 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1421     struct inpcb *(*notify)(struct inpcb *, int))
1422 {
1423 	struct inpcb *inp, *inp_temp;
1424 
1425 	INP_INFO_WLOCK(pcbinfo);
1426 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1427 		INP_WLOCK(inp);
1428 #ifdef INET6
1429 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1430 			INP_WUNLOCK(inp);
1431 			continue;
1432 		}
1433 #endif
1434 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1435 		    inp->inp_socket == NULL) {
1436 			INP_WUNLOCK(inp);
1437 			continue;
1438 		}
1439 		if ((*notify)(inp, errno))
1440 			INP_WUNLOCK(inp);
1441 	}
1442 	INP_INFO_WUNLOCK(pcbinfo);
1443 }
1444 
1445 void
1446 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1447 {
1448 	struct inpcb *inp;
1449 	struct ip_moptions *imo;
1450 	int i, gap;
1451 
1452 	INP_INFO_WLOCK(pcbinfo);
1453 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1454 		INP_WLOCK(inp);
1455 		imo = inp->inp_moptions;
1456 		if ((inp->inp_vflag & INP_IPV4) &&
1457 		    imo != NULL) {
1458 			/*
1459 			 * Unselect the outgoing interface if it is being
1460 			 * detached.
1461 			 */
1462 			if (imo->imo_multicast_ifp == ifp)
1463 				imo->imo_multicast_ifp = NULL;
1464 
1465 			/*
1466 			 * Drop multicast group membership if we joined
1467 			 * through the interface being detached.
1468 			 */
1469 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1470 			    i++) {
1471 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1472 					in_delmulti(imo->imo_membership[i]);
1473 					gap++;
1474 				} else if (gap != 0)
1475 					imo->imo_membership[i - gap] =
1476 					    imo->imo_membership[i];
1477 			}
1478 			imo->imo_num_memberships -= gap;
1479 		}
1480 		INP_WUNLOCK(inp);
1481 	}
1482 	INP_INFO_WUNLOCK(pcbinfo);
1483 }
1484 
1485 /*
1486  * Lookup a PCB based on the local address and port.  Caller must hold the
1487  * hash lock.  No inpcb locks or references are acquired.
1488  */
1489 #define INP_LOOKUP_MAPPED_PCB_COST	3
1490 struct inpcb *
1491 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1492     u_short lport, int lookupflags, struct ucred *cred)
1493 {
1494 	struct inpcb *inp;
1495 #ifdef INET6
1496 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1497 #else
1498 	int matchwild = 3;
1499 #endif
1500 	int wildcard;
1501 
1502 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1503 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1504 
1505 	INP_HASH_LOCK_ASSERT(pcbinfo);
1506 
1507 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1508 		struct inpcbhead *head;
1509 		/*
1510 		 * Look for an unconnected (wildcard foreign addr) PCB that
1511 		 * matches the local address and port we're looking for.
1512 		 */
1513 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1514 		    0, pcbinfo->ipi_hashmask)];
1515 		LIST_FOREACH(inp, head, inp_hash) {
1516 #ifdef INET6
1517 			/* XXX inp locking */
1518 			if ((inp->inp_vflag & INP_IPV4) == 0)
1519 				continue;
1520 #endif
1521 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1522 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1523 			    inp->inp_lport == lport) {
1524 				/*
1525 				 * Found?
1526 				 */
1527 				if (cred == NULL ||
1528 				    prison_equal_ip4(cred->cr_prison,
1529 					inp->inp_cred->cr_prison))
1530 					return (inp);
1531 			}
1532 		}
1533 		/*
1534 		 * Not found.
1535 		 */
1536 		return (NULL);
1537 	} else {
1538 		struct inpcbporthead *porthash;
1539 		struct inpcbport *phd;
1540 		struct inpcb *match = NULL;
1541 		/*
1542 		 * Best fit PCB lookup.
1543 		 *
1544 		 * First see if this local port is in use by looking on the
1545 		 * port hash list.
1546 		 */
1547 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1548 		    pcbinfo->ipi_porthashmask)];
1549 		LIST_FOREACH(phd, porthash, phd_hash) {
1550 			if (phd->phd_port == lport)
1551 				break;
1552 		}
1553 		if (phd != NULL) {
1554 			/*
1555 			 * Port is in use by one or more PCBs. Look for best
1556 			 * fit.
1557 			 */
1558 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1559 				wildcard = 0;
1560 				if (cred != NULL &&
1561 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1562 					cred->cr_prison))
1563 					continue;
1564 #ifdef INET6
1565 				/* XXX inp locking */
1566 				if ((inp->inp_vflag & INP_IPV4) == 0)
1567 					continue;
1568 				/*
1569 				 * We never select the PCB that has
1570 				 * INP_IPV6 flag and is bound to :: if
1571 				 * we have another PCB which is bound
1572 				 * to 0.0.0.0.  If a PCB has the
1573 				 * INP_IPV6 flag, then we set its cost
1574 				 * higher than IPv4 only PCBs.
1575 				 *
1576 				 * Note that the case only happens
1577 				 * when a socket is bound to ::, under
1578 				 * the condition that the use of the
1579 				 * mapped address is allowed.
1580 				 */
1581 				if ((inp->inp_vflag & INP_IPV6) != 0)
1582 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1583 #endif
1584 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1585 					wildcard++;
1586 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1587 					if (laddr.s_addr == INADDR_ANY)
1588 						wildcard++;
1589 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1590 						continue;
1591 				} else {
1592 					if (laddr.s_addr != INADDR_ANY)
1593 						wildcard++;
1594 				}
1595 				if (wildcard < matchwild) {
1596 					match = inp;
1597 					matchwild = wildcard;
1598 					if (matchwild == 0)
1599 						break;
1600 				}
1601 			}
1602 		}
1603 		return (match);
1604 	}
1605 }
1606 #undef INP_LOOKUP_MAPPED_PCB_COST
1607 
1608 #ifdef PCBGROUP
1609 /*
1610  * Lookup PCB in hash list, using pcbgroup tables.
1611  */
1612 static struct inpcb *
1613 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1614     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1615     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1616 {
1617 	struct inpcbhead *head;
1618 	struct inpcb *inp, *tmpinp;
1619 	u_short fport = fport_arg, lport = lport_arg;
1620 
1621 	/*
1622 	 * First look for an exact match.
1623 	 */
1624 	tmpinp = NULL;
1625 	INP_GROUP_LOCK(pcbgroup);
1626 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1627 	    pcbgroup->ipg_hashmask)];
1628 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1629 #ifdef INET6
1630 		/* XXX inp locking */
1631 		if ((inp->inp_vflag & INP_IPV4) == 0)
1632 			continue;
1633 #endif
1634 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1635 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1636 		    inp->inp_fport == fport &&
1637 		    inp->inp_lport == lport) {
1638 			/*
1639 			 * XXX We should be able to directly return
1640 			 * the inp here, without any checks.
1641 			 * Well unless both bound with SO_REUSEPORT?
1642 			 */
1643 			if (prison_flag(inp->inp_cred, PR_IP4))
1644 				goto found;
1645 			if (tmpinp == NULL)
1646 				tmpinp = inp;
1647 		}
1648 	}
1649 	if (tmpinp != NULL) {
1650 		inp = tmpinp;
1651 		goto found;
1652 	}
1653 
1654 #ifdef	RSS
1655 	/*
1656 	 * For incoming connections, we may wish to do a wildcard
1657 	 * match for an RSS-local socket.
1658 	 */
1659 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1660 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1661 #ifdef INET6
1662 		struct inpcb *local_wild_mapped = NULL;
1663 #endif
1664 		struct inpcb *jail_wild = NULL;
1665 		struct inpcbhead *head;
1666 		int injail;
1667 
1668 		/*
1669 		 * Order of socket selection - we always prefer jails.
1670 		 *      1. jailed, non-wild.
1671 		 *      2. jailed, wild.
1672 		 *      3. non-jailed, non-wild.
1673 		 *      4. non-jailed, wild.
1674 		 */
1675 
1676 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1677 		    lport, 0, pcbgroup->ipg_hashmask)];
1678 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1679 #ifdef INET6
1680 			/* XXX inp locking */
1681 			if ((inp->inp_vflag & INP_IPV4) == 0)
1682 				continue;
1683 #endif
1684 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1685 			    inp->inp_lport != lport)
1686 				continue;
1687 
1688 			injail = prison_flag(inp->inp_cred, PR_IP4);
1689 			if (injail) {
1690 				if (prison_check_ip4(inp->inp_cred,
1691 				    &laddr) != 0)
1692 					continue;
1693 			} else {
1694 				if (local_exact != NULL)
1695 					continue;
1696 			}
1697 
1698 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1699 				if (injail)
1700 					goto found;
1701 				else
1702 					local_exact = inp;
1703 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1704 #ifdef INET6
1705 				/* XXX inp locking, NULL check */
1706 				if (inp->inp_vflag & INP_IPV6PROTO)
1707 					local_wild_mapped = inp;
1708 				else
1709 #endif
1710 					if (injail)
1711 						jail_wild = inp;
1712 					else
1713 						local_wild = inp;
1714 			}
1715 		} /* LIST_FOREACH */
1716 
1717 		inp = jail_wild;
1718 		if (inp == NULL)
1719 			inp = local_exact;
1720 		if (inp == NULL)
1721 			inp = local_wild;
1722 #ifdef INET6
1723 		if (inp == NULL)
1724 			inp = local_wild_mapped;
1725 #endif
1726 		if (inp != NULL)
1727 			goto found;
1728 	}
1729 #endif
1730 
1731 	/*
1732 	 * Then look for a wildcard match, if requested.
1733 	 */
1734 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1735 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1736 #ifdef INET6
1737 		struct inpcb *local_wild_mapped = NULL;
1738 #endif
1739 		struct inpcb *jail_wild = NULL;
1740 		struct inpcbhead *head;
1741 		int injail;
1742 
1743 		/*
1744 		 * Order of socket selection - we always prefer jails.
1745 		 *      1. jailed, non-wild.
1746 		 *      2. jailed, wild.
1747 		 *      3. non-jailed, non-wild.
1748 		 *      4. non-jailed, wild.
1749 		 */
1750 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1751 		    0, pcbinfo->ipi_wildmask)];
1752 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1753 #ifdef INET6
1754 			/* XXX inp locking */
1755 			if ((inp->inp_vflag & INP_IPV4) == 0)
1756 				continue;
1757 #endif
1758 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1759 			    inp->inp_lport != lport)
1760 				continue;
1761 
1762 			injail = prison_flag(inp->inp_cred, PR_IP4);
1763 			if (injail) {
1764 				if (prison_check_ip4(inp->inp_cred,
1765 				    &laddr) != 0)
1766 					continue;
1767 			} else {
1768 				if (local_exact != NULL)
1769 					continue;
1770 			}
1771 
1772 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1773 				if (injail)
1774 					goto found;
1775 				else
1776 					local_exact = inp;
1777 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1778 #ifdef INET6
1779 				/* XXX inp locking, NULL check */
1780 				if (inp->inp_vflag & INP_IPV6PROTO)
1781 					local_wild_mapped = inp;
1782 				else
1783 #endif
1784 					if (injail)
1785 						jail_wild = inp;
1786 					else
1787 						local_wild = inp;
1788 			}
1789 		} /* LIST_FOREACH */
1790 		inp = jail_wild;
1791 		if (inp == NULL)
1792 			inp = local_exact;
1793 		if (inp == NULL)
1794 			inp = local_wild;
1795 #ifdef INET6
1796 		if (inp == NULL)
1797 			inp = local_wild_mapped;
1798 #endif
1799 		if (inp != NULL)
1800 			goto found;
1801 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1802 	INP_GROUP_UNLOCK(pcbgroup);
1803 	return (NULL);
1804 
1805 found:
1806 	in_pcbref(inp);
1807 	INP_GROUP_UNLOCK(pcbgroup);
1808 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1809 		INP_WLOCK(inp);
1810 		if (in_pcbrele_wlocked(inp))
1811 			return (NULL);
1812 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1813 		INP_RLOCK(inp);
1814 		if (in_pcbrele_rlocked(inp))
1815 			return (NULL);
1816 	} else
1817 		panic("%s: locking bug", __func__);
1818 	return (inp);
1819 }
1820 #endif /* PCBGROUP */
1821 
1822 /*
1823  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1824  * that the caller has locked the hash list, and will not perform any further
1825  * locking or reference operations on either the hash list or the connection.
1826  */
1827 static struct inpcb *
1828 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1829     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1830     struct ifnet *ifp)
1831 {
1832 	struct inpcbhead *head;
1833 	struct inpcb *inp, *tmpinp;
1834 	u_short fport = fport_arg, lport = lport_arg;
1835 
1836 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1837 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1838 
1839 	INP_HASH_LOCK_ASSERT(pcbinfo);
1840 
1841 	/*
1842 	 * First look for an exact match.
1843 	 */
1844 	tmpinp = NULL;
1845 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1846 	    pcbinfo->ipi_hashmask)];
1847 	LIST_FOREACH(inp, head, inp_hash) {
1848 #ifdef INET6
1849 		/* XXX inp locking */
1850 		if ((inp->inp_vflag & INP_IPV4) == 0)
1851 			continue;
1852 #endif
1853 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1854 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1855 		    inp->inp_fport == fport &&
1856 		    inp->inp_lport == lport) {
1857 			/*
1858 			 * XXX We should be able to directly return
1859 			 * the inp here, without any checks.
1860 			 * Well unless both bound with SO_REUSEPORT?
1861 			 */
1862 			if (prison_flag(inp->inp_cred, PR_IP4))
1863 				return (inp);
1864 			if (tmpinp == NULL)
1865 				tmpinp = inp;
1866 		}
1867 	}
1868 	if (tmpinp != NULL)
1869 		return (tmpinp);
1870 
1871 	/*
1872 	 * Then look for a wildcard match, if requested.
1873 	 */
1874 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1875 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1876 #ifdef INET6
1877 		struct inpcb *local_wild_mapped = NULL;
1878 #endif
1879 		struct inpcb *jail_wild = NULL;
1880 		int injail;
1881 
1882 		/*
1883 		 * Order of socket selection - we always prefer jails.
1884 		 *      1. jailed, non-wild.
1885 		 *      2. jailed, wild.
1886 		 *      3. non-jailed, non-wild.
1887 		 *      4. non-jailed, wild.
1888 		 */
1889 
1890 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1891 		    0, pcbinfo->ipi_hashmask)];
1892 		LIST_FOREACH(inp, head, inp_hash) {
1893 #ifdef INET6
1894 			/* XXX inp locking */
1895 			if ((inp->inp_vflag & INP_IPV4) == 0)
1896 				continue;
1897 #endif
1898 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1899 			    inp->inp_lport != lport)
1900 				continue;
1901 
1902 			injail = prison_flag(inp->inp_cred, PR_IP4);
1903 			if (injail) {
1904 				if (prison_check_ip4(inp->inp_cred,
1905 				    &laddr) != 0)
1906 					continue;
1907 			} else {
1908 				if (local_exact != NULL)
1909 					continue;
1910 			}
1911 
1912 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1913 				if (injail)
1914 					return (inp);
1915 				else
1916 					local_exact = inp;
1917 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1918 #ifdef INET6
1919 				/* XXX inp locking, NULL check */
1920 				if (inp->inp_vflag & INP_IPV6PROTO)
1921 					local_wild_mapped = inp;
1922 				else
1923 #endif
1924 					if (injail)
1925 						jail_wild = inp;
1926 					else
1927 						local_wild = inp;
1928 			}
1929 		} /* LIST_FOREACH */
1930 		if (jail_wild != NULL)
1931 			return (jail_wild);
1932 		if (local_exact != NULL)
1933 			return (local_exact);
1934 		if (local_wild != NULL)
1935 			return (local_wild);
1936 #ifdef INET6
1937 		if (local_wild_mapped != NULL)
1938 			return (local_wild_mapped);
1939 #endif
1940 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1941 
1942 	return (NULL);
1943 }
1944 
1945 /*
1946  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1947  * hash list lock, and will return the inpcb locked (i.e., requires
1948  * INPLOOKUP_LOCKPCB).
1949  */
1950 static struct inpcb *
1951 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1952     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1953     struct ifnet *ifp)
1954 {
1955 	struct inpcb *inp;
1956 
1957 	INP_HASH_RLOCK(pcbinfo);
1958 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1959 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1960 	if (inp != NULL) {
1961 		in_pcbref(inp);
1962 		INP_HASH_RUNLOCK(pcbinfo);
1963 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1964 			INP_WLOCK(inp);
1965 			if (in_pcbrele_wlocked(inp))
1966 				return (NULL);
1967 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1968 			INP_RLOCK(inp);
1969 			if (in_pcbrele_rlocked(inp))
1970 				return (NULL);
1971 		} else
1972 			panic("%s: locking bug", __func__);
1973 	} else
1974 		INP_HASH_RUNLOCK(pcbinfo);
1975 	return (inp);
1976 }
1977 
1978 /*
1979  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1980  * from which a pre-calculated hash value may be extracted.
1981  *
1982  * Possibly more of this logic should be in in_pcbgroup.c.
1983  */
1984 struct inpcb *
1985 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1986     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1987 {
1988 #if defined(PCBGROUP) && !defined(RSS)
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 	/*
1998 	 * When not using RSS, use connection groups in preference to the
1999 	 * reservation table when looking up 4-tuples.  When using RSS, just
2000 	 * use the reservation table, due to the cost of the Toeplitz hash
2001 	 * in software.
2002 	 *
2003 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2004 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2005 	 * in software.
2006 	 */
2007 #if defined(PCBGROUP) && !defined(RSS)
2008 	if (in_pcbgroup_enabled(pcbinfo)) {
2009 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2010 		    fport);
2011 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2012 		    laddr, lport, lookupflags, ifp));
2013 	}
2014 #endif
2015 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2016 	    lookupflags, ifp));
2017 }
2018 
2019 struct inpcb *
2020 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2021     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2022     struct ifnet *ifp, struct mbuf *m)
2023 {
2024 #ifdef PCBGROUP
2025 	struct inpcbgroup *pcbgroup;
2026 #endif
2027 
2028 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2029 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2030 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2031 	    ("%s: LOCKPCB not set", __func__));
2032 
2033 #ifdef PCBGROUP
2034 	/*
2035 	 * If we can use a hardware-generated hash to look up the connection
2036 	 * group, use that connection group to find the inpcb.  Otherwise
2037 	 * fall back on a software hash -- or the reservation table if we're
2038 	 * using RSS.
2039 	 *
2040 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2041 	 */
2042 	if (in_pcbgroup_enabled(pcbinfo) &&
2043 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2044 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2045 		    m->m_pkthdr.flowid);
2046 		if (pcbgroup != NULL)
2047 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2048 			    fport, laddr, lport, lookupflags, ifp));
2049 #ifndef RSS
2050 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2051 		    fport);
2052 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2053 		    laddr, lport, lookupflags, ifp));
2054 #endif
2055 	}
2056 #endif
2057 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2058 	    lookupflags, ifp));
2059 }
2060 #endif /* INET */
2061 
2062 /*
2063  * Insert PCB onto various hash lists.
2064  */
2065 static int
2066 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2067 {
2068 	struct inpcbhead *pcbhash;
2069 	struct inpcbporthead *pcbporthash;
2070 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2071 	struct inpcbport *phd;
2072 	u_int32_t hashkey_faddr;
2073 
2074 	INP_WLOCK_ASSERT(inp);
2075 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2076 
2077 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2078 	    ("in_pcbinshash: INP_INHASHLIST"));
2079 
2080 #ifdef INET6
2081 	if (inp->inp_vflag & INP_IPV6)
2082 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2083 	else
2084 #endif
2085 	hashkey_faddr = inp->inp_faddr.s_addr;
2086 
2087 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2088 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2089 
2090 	pcbporthash = &pcbinfo->ipi_porthashbase[
2091 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2092 
2093 	/*
2094 	 * Go through port list and look for a head for this lport.
2095 	 */
2096 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2097 		if (phd->phd_port == inp->inp_lport)
2098 			break;
2099 	}
2100 	/*
2101 	 * If none exists, malloc one and tack it on.
2102 	 */
2103 	if (phd == NULL) {
2104 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2105 		if (phd == NULL) {
2106 			return (ENOBUFS); /* XXX */
2107 		}
2108 		phd->phd_port = inp->inp_lport;
2109 		LIST_INIT(&phd->phd_pcblist);
2110 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2111 	}
2112 	inp->inp_phd = phd;
2113 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2114 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2115 	inp->inp_flags |= INP_INHASHLIST;
2116 #ifdef PCBGROUP
2117 	if (do_pcbgroup_update)
2118 		in_pcbgroup_update(inp);
2119 #endif
2120 	return (0);
2121 }
2122 
2123 /*
2124  * For now, there are two public interfaces to insert an inpcb into the hash
2125  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2126  * is used only in the TCP syncache, where in_pcbinshash is called before the
2127  * full 4-tuple is set for the inpcb, and we don't want to install in the
2128  * pcbgroup until later.
2129  *
2130  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2131  * connection groups, and partially initialised inpcbs should not be exposed
2132  * to either reservation hash tables or pcbgroups.
2133  */
2134 int
2135 in_pcbinshash(struct inpcb *inp)
2136 {
2137 
2138 	return (in_pcbinshash_internal(inp, 1));
2139 }
2140 
2141 int
2142 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2143 {
2144 
2145 	return (in_pcbinshash_internal(inp, 0));
2146 }
2147 
2148 /*
2149  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2150  * changed. NOTE: This does not handle the case of the lport changing (the
2151  * hashed port list would have to be updated as well), so the lport must
2152  * not change after in_pcbinshash() has been called.
2153  */
2154 void
2155 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2156 {
2157 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2158 	struct inpcbhead *head;
2159 	u_int32_t hashkey_faddr;
2160 
2161 	INP_WLOCK_ASSERT(inp);
2162 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2163 
2164 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2165 	    ("in_pcbrehash: !INP_INHASHLIST"));
2166 
2167 #ifdef INET6
2168 	if (inp->inp_vflag & INP_IPV6)
2169 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2170 	else
2171 #endif
2172 	hashkey_faddr = inp->inp_faddr.s_addr;
2173 
2174 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2175 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2176 
2177 	LIST_REMOVE(inp, inp_hash);
2178 	LIST_INSERT_HEAD(head, inp, inp_hash);
2179 
2180 #ifdef PCBGROUP
2181 	if (m != NULL)
2182 		in_pcbgroup_update_mbuf(inp, m);
2183 	else
2184 		in_pcbgroup_update(inp);
2185 #endif
2186 }
2187 
2188 void
2189 in_pcbrehash(struct inpcb *inp)
2190 {
2191 
2192 	in_pcbrehash_mbuf(inp, NULL);
2193 }
2194 
2195 /*
2196  * Remove PCB from various lists.
2197  */
2198 static void
2199 in_pcbremlists(struct inpcb *inp)
2200 {
2201 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2202 
2203 #ifdef INVARIANTS
2204 	if (pcbinfo == &V_tcbinfo) {
2205 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2206 	} else {
2207 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2208 	}
2209 #endif
2210 
2211 	INP_WLOCK_ASSERT(inp);
2212 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2213 
2214 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2215 	if (inp->inp_flags & INP_INHASHLIST) {
2216 		struct inpcbport *phd = inp->inp_phd;
2217 
2218 		INP_HASH_WLOCK(pcbinfo);
2219 		LIST_REMOVE(inp, inp_hash);
2220 		LIST_REMOVE(inp, inp_portlist);
2221 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2222 			LIST_REMOVE(phd, phd_hash);
2223 			free(phd, M_PCB);
2224 		}
2225 		INP_HASH_WUNLOCK(pcbinfo);
2226 		inp->inp_flags &= ~INP_INHASHLIST;
2227 	}
2228 	LIST_REMOVE(inp, inp_list);
2229 	pcbinfo->ipi_count--;
2230 #ifdef PCBGROUP
2231 	in_pcbgroup_remove(inp);
2232 #endif
2233 }
2234 
2235 /*
2236  * Check for alternatives when higher level complains
2237  * about service problems.  For now, invalidate cached
2238  * routing information.  If the route was created dynamically
2239  * (by a redirect), time to try a default gateway again.
2240  */
2241 void
2242 in_losing(struct inpcb *inp)
2243 {
2244 
2245 	if (inp->inp_route.ro_rt) {
2246 		RTFREE(inp->inp_route.ro_rt);
2247 		inp->inp_route.ro_rt = (struct rtentry *)NULL;
2248 	}
2249 	if (inp->inp_route.ro_lle)
2250 		LLE_FREE(inp->inp_route.ro_lle);	/* zeros ro_lle */
2251 	return;
2252 }
2253 
2254 /*
2255  * A set label operation has occurred at the socket layer, propagate the
2256  * label change into the in_pcb for the socket.
2257  */
2258 void
2259 in_pcbsosetlabel(struct socket *so)
2260 {
2261 #ifdef MAC
2262 	struct inpcb *inp;
2263 
2264 	inp = sotoinpcb(so);
2265 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2266 
2267 	INP_WLOCK(inp);
2268 	SOCK_LOCK(so);
2269 	mac_inpcb_sosetlabel(so, inp);
2270 	SOCK_UNLOCK(so);
2271 	INP_WUNLOCK(inp);
2272 #endif
2273 }
2274 
2275 /*
2276  * ipport_tick runs once per second, determining if random port allocation
2277  * should be continued.  If more than ipport_randomcps ports have been
2278  * allocated in the last second, then we return to sequential port
2279  * allocation. We return to random allocation only once we drop below
2280  * ipport_randomcps for at least ipport_randomtime seconds.
2281  */
2282 static void
2283 ipport_tick(void *xtp)
2284 {
2285 	VNET_ITERATOR_DECL(vnet_iter);
2286 
2287 	VNET_LIST_RLOCK_NOSLEEP();
2288 	VNET_FOREACH(vnet_iter) {
2289 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2290 		if (V_ipport_tcpallocs <=
2291 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2292 			if (V_ipport_stoprandom > 0)
2293 				V_ipport_stoprandom--;
2294 		} else
2295 			V_ipport_stoprandom = V_ipport_randomtime;
2296 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2297 		CURVNET_RESTORE();
2298 	}
2299 	VNET_LIST_RUNLOCK_NOSLEEP();
2300 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2301 }
2302 
2303 static void
2304 ip_fini(void *xtp)
2305 {
2306 
2307 	callout_stop(&ipport_tick_callout);
2308 }
2309 
2310 /*
2311  * The ipport_callout should start running at about the time we attach the
2312  * inet or inet6 domains.
2313  */
2314 static void
2315 ipport_tick_init(const void *unused __unused)
2316 {
2317 
2318 	/* Start ipport_tick. */
2319 	callout_init(&ipport_tick_callout, 1);
2320 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2321 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2322 		SHUTDOWN_PRI_DEFAULT);
2323 }
2324 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2325     ipport_tick_init, NULL);
2326 
2327 void
2328 inp_wlock(struct inpcb *inp)
2329 {
2330 
2331 	INP_WLOCK(inp);
2332 }
2333 
2334 void
2335 inp_wunlock(struct inpcb *inp)
2336 {
2337 
2338 	INP_WUNLOCK(inp);
2339 }
2340 
2341 void
2342 inp_rlock(struct inpcb *inp)
2343 {
2344 
2345 	INP_RLOCK(inp);
2346 }
2347 
2348 void
2349 inp_runlock(struct inpcb *inp)
2350 {
2351 
2352 	INP_RUNLOCK(inp);
2353 }
2354 
2355 #ifdef INVARIANTS
2356 void
2357 inp_lock_assert(struct inpcb *inp)
2358 {
2359 
2360 	INP_WLOCK_ASSERT(inp);
2361 }
2362 
2363 void
2364 inp_unlock_assert(struct inpcb *inp)
2365 {
2366 
2367 	INP_UNLOCK_ASSERT(inp);
2368 }
2369 #endif
2370 
2371 void
2372 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2373 {
2374 	struct inpcb *inp;
2375 
2376 	INP_INFO_WLOCK(&V_tcbinfo);
2377 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2378 		INP_WLOCK(inp);
2379 		func(inp, arg);
2380 		INP_WUNLOCK(inp);
2381 	}
2382 	INP_INFO_WUNLOCK(&V_tcbinfo);
2383 }
2384 
2385 struct socket *
2386 inp_inpcbtosocket(struct inpcb *inp)
2387 {
2388 
2389 	INP_WLOCK_ASSERT(inp);
2390 	return (inp->inp_socket);
2391 }
2392 
2393 struct tcpcb *
2394 inp_inpcbtotcpcb(struct inpcb *inp)
2395 {
2396 
2397 	INP_WLOCK_ASSERT(inp);
2398 	return ((struct tcpcb *)inp->inp_ppcb);
2399 }
2400 
2401 int
2402 inp_ip_tos_get(const struct inpcb *inp)
2403 {
2404 
2405 	return (inp->inp_ip_tos);
2406 }
2407 
2408 void
2409 inp_ip_tos_set(struct inpcb *inp, int val)
2410 {
2411 
2412 	inp->inp_ip_tos = val;
2413 }
2414 
2415 void
2416 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2417     uint32_t *faddr, uint16_t *fp)
2418 {
2419 
2420 	INP_LOCK_ASSERT(inp);
2421 	*laddr = inp->inp_laddr.s_addr;
2422 	*faddr = inp->inp_faddr.s_addr;
2423 	*lp = inp->inp_lport;
2424 	*fp = inp->inp_fport;
2425 }
2426 
2427 struct inpcb *
2428 so_sotoinpcb(struct socket *so)
2429 {
2430 
2431 	return (sotoinpcb(so));
2432 }
2433 
2434 struct tcpcb *
2435 so_sototcpcb(struct socket *so)
2436 {
2437 
2438 	return (sototcpcb(so));
2439 }
2440 
2441 #ifdef DDB
2442 static void
2443 db_print_indent(int indent)
2444 {
2445 	int i;
2446 
2447 	for (i = 0; i < indent; i++)
2448 		db_printf(" ");
2449 }
2450 
2451 static void
2452 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2453 {
2454 	char faddr_str[48], laddr_str[48];
2455 
2456 	db_print_indent(indent);
2457 	db_printf("%s at %p\n", name, inc);
2458 
2459 	indent += 2;
2460 
2461 #ifdef INET6
2462 	if (inc->inc_flags & INC_ISIPV6) {
2463 		/* IPv6. */
2464 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2465 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2466 	} else
2467 #endif
2468 	{
2469 		/* IPv4. */
2470 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2471 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2472 	}
2473 	db_print_indent(indent);
2474 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2475 	    ntohs(inc->inc_lport));
2476 	db_print_indent(indent);
2477 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2478 	    ntohs(inc->inc_fport));
2479 }
2480 
2481 static void
2482 db_print_inpflags(int inp_flags)
2483 {
2484 	int comma;
2485 
2486 	comma = 0;
2487 	if (inp_flags & INP_RECVOPTS) {
2488 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2489 		comma = 1;
2490 	}
2491 	if (inp_flags & INP_RECVRETOPTS) {
2492 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2493 		comma = 1;
2494 	}
2495 	if (inp_flags & INP_RECVDSTADDR) {
2496 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2497 		comma = 1;
2498 	}
2499 	if (inp_flags & INP_HDRINCL) {
2500 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2501 		comma = 1;
2502 	}
2503 	if (inp_flags & INP_HIGHPORT) {
2504 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2505 		comma = 1;
2506 	}
2507 	if (inp_flags & INP_LOWPORT) {
2508 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2509 		comma = 1;
2510 	}
2511 	if (inp_flags & INP_ANONPORT) {
2512 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2513 		comma = 1;
2514 	}
2515 	if (inp_flags & INP_RECVIF) {
2516 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2517 		comma = 1;
2518 	}
2519 	if (inp_flags & INP_MTUDISC) {
2520 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2521 		comma = 1;
2522 	}
2523 	if (inp_flags & INP_RECVTTL) {
2524 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2525 		comma = 1;
2526 	}
2527 	if (inp_flags & INP_DONTFRAG) {
2528 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2529 		comma = 1;
2530 	}
2531 	if (inp_flags & INP_RECVTOS) {
2532 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2533 		comma = 1;
2534 	}
2535 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2536 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2537 		comma = 1;
2538 	}
2539 	if (inp_flags & IN6P_PKTINFO) {
2540 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2541 		comma = 1;
2542 	}
2543 	if (inp_flags & IN6P_HOPLIMIT) {
2544 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2545 		comma = 1;
2546 	}
2547 	if (inp_flags & IN6P_HOPOPTS) {
2548 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2549 		comma = 1;
2550 	}
2551 	if (inp_flags & IN6P_DSTOPTS) {
2552 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2553 		comma = 1;
2554 	}
2555 	if (inp_flags & IN6P_RTHDR) {
2556 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2557 		comma = 1;
2558 	}
2559 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2560 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2561 		comma = 1;
2562 	}
2563 	if (inp_flags & IN6P_TCLASS) {
2564 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2565 		comma = 1;
2566 	}
2567 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2568 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2569 		comma = 1;
2570 	}
2571 	if (inp_flags & INP_TIMEWAIT) {
2572 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2573 		comma  = 1;
2574 	}
2575 	if (inp_flags & INP_ONESBCAST) {
2576 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2577 		comma  = 1;
2578 	}
2579 	if (inp_flags & INP_DROPPED) {
2580 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2581 		comma  = 1;
2582 	}
2583 	if (inp_flags & INP_SOCKREF) {
2584 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2585 		comma  = 1;
2586 	}
2587 	if (inp_flags & IN6P_RFC2292) {
2588 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2589 		comma = 1;
2590 	}
2591 	if (inp_flags & IN6P_MTU) {
2592 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2593 		comma = 1;
2594 	}
2595 }
2596 
2597 static void
2598 db_print_inpvflag(u_char inp_vflag)
2599 {
2600 	int comma;
2601 
2602 	comma = 0;
2603 	if (inp_vflag & INP_IPV4) {
2604 		db_printf("%sINP_IPV4", comma ? ", " : "");
2605 		comma  = 1;
2606 	}
2607 	if (inp_vflag & INP_IPV6) {
2608 		db_printf("%sINP_IPV6", comma ? ", " : "");
2609 		comma  = 1;
2610 	}
2611 	if (inp_vflag & INP_IPV6PROTO) {
2612 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2613 		comma  = 1;
2614 	}
2615 }
2616 
2617 static void
2618 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2619 {
2620 
2621 	db_print_indent(indent);
2622 	db_printf("%s at %p\n", name, inp);
2623 
2624 	indent += 2;
2625 
2626 	db_print_indent(indent);
2627 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2628 
2629 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2630 
2631 	db_print_indent(indent);
2632 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2633 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2634 
2635 	db_print_indent(indent);
2636 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2637 	   inp->inp_label, inp->inp_flags);
2638 	db_print_inpflags(inp->inp_flags);
2639 	db_printf(")\n");
2640 
2641 	db_print_indent(indent);
2642 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2643 	    inp->inp_vflag);
2644 	db_print_inpvflag(inp->inp_vflag);
2645 	db_printf(")\n");
2646 
2647 	db_print_indent(indent);
2648 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2649 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2650 
2651 	db_print_indent(indent);
2652 #ifdef INET6
2653 	if (inp->inp_vflag & INP_IPV6) {
2654 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2655 		    "in6p_moptions: %p\n", inp->in6p_options,
2656 		    inp->in6p_outputopts, inp->in6p_moptions);
2657 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2658 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2659 		    inp->in6p_hops);
2660 	} else
2661 #endif
2662 	{
2663 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2664 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2665 		    inp->inp_options, inp->inp_moptions);
2666 	}
2667 
2668 	db_print_indent(indent);
2669 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2670 	    (uintmax_t)inp->inp_gencnt);
2671 }
2672 
2673 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2674 {
2675 	struct inpcb *inp;
2676 
2677 	if (!have_addr) {
2678 		db_printf("usage: show inpcb <addr>\n");
2679 		return;
2680 	}
2681 	inp = (struct inpcb *)addr;
2682 
2683 	db_print_inpcb(inp, "inpcb", 0);
2684 }
2685 #endif /* DDB */
2686