xref: /freebsd/sys/netinet/in_pcb.c (revision 895f86f15fbf6540071feb9328c3c50ed1f027b8)
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/domain.h>
55 #include <sys/protosw.h>
56 #include <sys/rmlock.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/refcount.h>
62 #include <sys/jail.h>
63 #include <sys/kernel.h>
64 #include <sys/sysctl.h>
65 
66 #ifdef DDB
67 #include <ddb/ddb.h>
68 #endif
69 
70 #include <vm/uma.h>
71 
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/if_types.h>
75 #include <net/route.h>
76 #include <net/rss_config.h>
77 #include <net/vnet.h>
78 
79 #if defined(INET) || defined(INET6)
80 #include <netinet/in.h>
81 #include <netinet/in_pcb.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/tcp_var.h>
84 #include <netinet/udp.h>
85 #include <netinet/udp_var.h>
86 #endif
87 #ifdef INET
88 #include <netinet/in_var.h>
89 #endif
90 #ifdef INET6
91 #include <netinet/ip6.h>
92 #include <netinet6/in6_pcb.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet6/ip6_var.h>
95 #endif /* INET6 */
96 
97 
98 #ifdef IPSEC
99 #include <netipsec/ipsec.h>
100 #include <netipsec/key.h>
101 #endif /* IPSEC */
102 
103 #include <security/mac/mac_framework.h>
104 
105 static struct callout	ipport_tick_callout;
106 
107 /*
108  * These configure the range of local port addresses assigned to
109  * "unspecified" outgoing connections/packets/whatever.
110  */
111 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
112 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
113 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
114 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
115 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
116 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
117 
118 /*
119  * Reserved ports accessible only to root. There are significant
120  * security considerations that must be accounted for when changing these,
121  * but the security benefits can be great. Please be careful.
122  */
123 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
124 VNET_DEFINE(int, ipport_reservedlow);
125 
126 /* Variables dealing with random ephemeral port allocation. */
127 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
128 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
129 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
130 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
131 VNET_DEFINE(int, ipport_tcpallocs);
132 static VNET_DEFINE(int, ipport_tcplastcount);
133 
134 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
135 
136 static void	in_pcbremlists(struct inpcb *inp);
137 #ifdef INET
138 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
139 			    struct in_addr faddr, u_int fport_arg,
140 			    struct in_addr laddr, u_int lport_arg,
141 			    int lookupflags, struct ifnet *ifp);
142 
143 #define RANGECHK(var, min, max) \
144 	if ((var) < (min)) { (var) = (min); } \
145 	else if ((var) > (max)) { (var) = (max); }
146 
147 static int
148 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
149 {
150 	int error;
151 
152 	error = sysctl_handle_int(oidp, arg1, arg2, req);
153 	if (error == 0) {
154 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
155 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
156 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
157 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
158 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
159 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
160 	}
161 	return (error);
162 }
163 
164 #undef RANGECHK
165 
166 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
167     "IP Ports");
168 
169 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
170 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
171 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
173 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
174 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
176 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
177 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
179 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
182 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
185 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
188 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
189 	&VNET_NAME(ipport_reservedhigh), 0, "");
190 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
191 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
192 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
193 	CTLFLAG_VNET | CTLFLAG_RW,
194 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
195 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
196 	CTLFLAG_VNET | CTLFLAG_RW,
197 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
198 	"allocations before switching to a sequental one");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
200 	CTLFLAG_VNET | CTLFLAG_RW,
201 	&VNET_NAME(ipport_randomtime), 0,
202 	"Minimum time to keep sequental port "
203 	"allocation before switching to a random one");
204 #endif /* INET */
205 
206 /*
207  * in_pcb.c: manage the Protocol Control Blocks.
208  *
209  * NOTE: It is assumed that most of these functions will be called with
210  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
211  * functions often modify hash chains or addresses in pcbs.
212  */
213 
214 /*
215  * Initialize an inpcbinfo -- we should be able to reduce the number of
216  * arguments in time.
217  */
218 void
219 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
220     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
221     char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
222     uint32_t inpcbzone_flags, u_int hashfields)
223 {
224 
225 	INP_INFO_LOCK_INIT(pcbinfo, name);
226 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
227 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
228 #ifdef VIMAGE
229 	pcbinfo->ipi_vnet = curvnet;
230 #endif
231 	pcbinfo->ipi_listhead = listhead;
232 	LIST_INIT(pcbinfo->ipi_listhead);
233 	pcbinfo->ipi_count = 0;
234 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
235 	    &pcbinfo->ipi_hashmask);
236 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
237 	    &pcbinfo->ipi_porthashmask);
238 #ifdef PCBGROUP
239 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
240 #endif
241 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
242 	    NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
243 	    inpcbzone_flags);
244 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
245 	uma_zone_set_warning(pcbinfo->ipi_zone,
246 	    "kern.ipc.maxsockets limit reached");
247 }
248 
249 /*
250  * Destroy an inpcbinfo.
251  */
252 void
253 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
254 {
255 
256 	KASSERT(pcbinfo->ipi_count == 0,
257 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
258 
259 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
260 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
261 	    pcbinfo->ipi_porthashmask);
262 #ifdef PCBGROUP
263 	in_pcbgroup_destroy(pcbinfo);
264 #endif
265 	uma_zdestroy(pcbinfo->ipi_zone);
266 	INP_LIST_LOCK_DESTROY(pcbinfo);
267 	INP_HASH_LOCK_DESTROY(pcbinfo);
268 	INP_INFO_LOCK_DESTROY(pcbinfo);
269 }
270 
271 /*
272  * Allocate a PCB and associate it with the socket.
273  * On success return with the PCB locked.
274  */
275 int
276 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
277 {
278 	struct inpcb *inp;
279 	int error;
280 
281 #ifdef INVARIANTS
282 	if (pcbinfo == &V_tcbinfo) {
283 		INP_INFO_RLOCK_ASSERT(pcbinfo);
284 	} else {
285 		INP_INFO_WLOCK_ASSERT(pcbinfo);
286 	}
287 #endif
288 
289 	error = 0;
290 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
291 	if (inp == NULL)
292 		return (ENOBUFS);
293 	bzero(inp, inp_zero_size);
294 	inp->inp_pcbinfo = pcbinfo;
295 	inp->inp_socket = so;
296 	inp->inp_cred = crhold(so->so_cred);
297 	inp->inp_inc.inc_fibnum = so->so_fibnum;
298 #ifdef MAC
299 	error = mac_inpcb_init(inp, M_NOWAIT);
300 	if (error != 0)
301 		goto out;
302 	mac_inpcb_create(so, inp);
303 #endif
304 #ifdef IPSEC
305 	error = ipsec_init_policy(so, &inp->inp_sp);
306 	if (error != 0) {
307 #ifdef MAC
308 		mac_inpcb_destroy(inp);
309 #endif
310 		goto out;
311 	}
312 #endif /*IPSEC*/
313 #ifdef INET6
314 	if (INP_SOCKAF(so) == AF_INET6) {
315 		inp->inp_vflag |= INP_IPV6PROTO;
316 		if (V_ip6_v6only)
317 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
318 	}
319 #endif
320 	INP_WLOCK(inp);
321 	INP_LIST_WLOCK(pcbinfo);
322 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
323 	pcbinfo->ipi_count++;
324 	so->so_pcb = (caddr_t)inp;
325 #ifdef INET6
326 	if (V_ip6_auto_flowlabel)
327 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
328 #endif
329 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
330 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
331 	INP_LIST_WUNLOCK(pcbinfo);
332 #if defined(IPSEC) || defined(MAC)
333 out:
334 	if (error != 0) {
335 		crfree(inp->inp_cred);
336 		uma_zfree(pcbinfo->ipi_zone, inp);
337 	}
338 #endif
339 	return (error);
340 }
341 
342 #ifdef INET
343 int
344 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
345 {
346 	int anonport, error;
347 
348 	INP_WLOCK_ASSERT(inp);
349 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
350 
351 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
352 		return (EINVAL);
353 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
354 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
355 	    &inp->inp_lport, cred);
356 	if (error)
357 		return (error);
358 	if (in_pcbinshash(inp) != 0) {
359 		inp->inp_laddr.s_addr = INADDR_ANY;
360 		inp->inp_lport = 0;
361 		return (EAGAIN);
362 	}
363 	if (anonport)
364 		inp->inp_flags |= INP_ANONPORT;
365 	return (0);
366 }
367 #endif
368 
369 /*
370  * Select a local port (number) to use.
371  */
372 #if defined(INET) || defined(INET6)
373 int
374 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
375     struct ucred *cred, int lookupflags)
376 {
377 	struct inpcbinfo *pcbinfo;
378 	struct inpcb *tmpinp;
379 	unsigned short *lastport;
380 	int count, dorandom, error;
381 	u_short aux, first, last, lport;
382 #ifdef INET
383 	struct in_addr laddr;
384 #endif
385 
386 	pcbinfo = inp->inp_pcbinfo;
387 
388 	/*
389 	 * Because no actual state changes occur here, a global write lock on
390 	 * the pcbinfo isn't required.
391 	 */
392 	INP_LOCK_ASSERT(inp);
393 	INP_HASH_LOCK_ASSERT(pcbinfo);
394 
395 	if (inp->inp_flags & INP_HIGHPORT) {
396 		first = V_ipport_hifirstauto;	/* sysctl */
397 		last  = V_ipport_hilastauto;
398 		lastport = &pcbinfo->ipi_lasthi;
399 	} else if (inp->inp_flags & INP_LOWPORT) {
400 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
401 		if (error)
402 			return (error);
403 		first = V_ipport_lowfirstauto;	/* 1023 */
404 		last  = V_ipport_lowlastauto;	/* 600 */
405 		lastport = &pcbinfo->ipi_lastlow;
406 	} else {
407 		first = V_ipport_firstauto;	/* sysctl */
408 		last  = V_ipport_lastauto;
409 		lastport = &pcbinfo->ipi_lastport;
410 	}
411 	/*
412 	 * For UDP(-Lite), use random port allocation as long as the user
413 	 * allows it.  For TCP (and as of yet unknown) connections,
414 	 * use random port allocation only if the user allows it AND
415 	 * ipport_tick() allows it.
416 	 */
417 	if (V_ipport_randomized &&
418 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
419 		pcbinfo == &V_ulitecbinfo))
420 		dorandom = 1;
421 	else
422 		dorandom = 0;
423 	/*
424 	 * It makes no sense to do random port allocation if
425 	 * we have the only port available.
426 	 */
427 	if (first == last)
428 		dorandom = 0;
429 	/* Make sure to not include UDP(-Lite) packets in the count. */
430 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
431 		V_ipport_tcpallocs++;
432 	/*
433 	 * Instead of having two loops further down counting up or down
434 	 * make sure that first is always <= last and go with only one
435 	 * code path implementing all logic.
436 	 */
437 	if (first > last) {
438 		aux = first;
439 		first = last;
440 		last = aux;
441 	}
442 
443 #ifdef INET
444 	/* Make the compiler happy. */
445 	laddr.s_addr = 0;
446 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
447 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
448 		    __func__, inp));
449 		laddr = *laddrp;
450 	}
451 #endif
452 	tmpinp = NULL;	/* Make compiler happy. */
453 	lport = *lportp;
454 
455 	if (dorandom)
456 		*lastport = first + (arc4random() % (last - first));
457 
458 	count = last - first;
459 
460 	do {
461 		if (count-- < 0)	/* completely used? */
462 			return (EADDRNOTAVAIL);
463 		++*lastport;
464 		if (*lastport < first || *lastport > last)
465 			*lastport = first;
466 		lport = htons(*lastport);
467 
468 #ifdef INET6
469 		if ((inp->inp_vflag & INP_IPV6) != 0)
470 			tmpinp = in6_pcblookup_local(pcbinfo,
471 			    &inp->in6p_laddr, lport, lookupflags, cred);
472 #endif
473 #if defined(INET) && defined(INET6)
474 		else
475 #endif
476 #ifdef INET
477 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
478 			    lport, lookupflags, cred);
479 #endif
480 	} while (tmpinp != NULL);
481 
482 #ifdef INET
483 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
484 		laddrp->s_addr = laddr.s_addr;
485 #endif
486 	*lportp = lport;
487 
488 	return (0);
489 }
490 
491 /*
492  * Return cached socket options.
493  */
494 short
495 inp_so_options(const struct inpcb *inp)
496 {
497    short so_options;
498 
499    so_options = 0;
500 
501    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
502 	   so_options |= SO_REUSEPORT;
503    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
504 	   so_options |= SO_REUSEADDR;
505    return (so_options);
506 }
507 #endif /* INET || INET6 */
508 
509 /*
510  * Check if a new BINDMULTI socket is allowed to be created.
511  *
512  * ni points to the new inp.
513  * oi points to the exisitng inp.
514  *
515  * This checks whether the existing inp also has BINDMULTI and
516  * whether the credentials match.
517  */
518 int
519 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
520 {
521 	/* Check permissions match */
522 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
523 	    (ni->inp_cred->cr_uid !=
524 	    oi->inp_cred->cr_uid))
525 		return (0);
526 
527 	/* Check the existing inp has BINDMULTI set */
528 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
529 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
530 		return (0);
531 
532 	/*
533 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
534 	 * it is and it matches the checks.
535 	 */
536 	return (1);
537 }
538 
539 #ifdef INET
540 /*
541  * Set up a bind operation on a PCB, performing port allocation
542  * as required, but do not actually modify the PCB. Callers can
543  * either complete the bind by setting inp_laddr/inp_lport and
544  * calling in_pcbinshash(), or they can just use the resulting
545  * port and address to authorise the sending of a once-off packet.
546  *
547  * On error, the values of *laddrp and *lportp are not changed.
548  */
549 int
550 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
551     u_short *lportp, struct ucred *cred)
552 {
553 	struct socket *so = inp->inp_socket;
554 	struct sockaddr_in *sin;
555 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
556 	struct in_addr laddr;
557 	u_short lport = 0;
558 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
559 	int error;
560 
561 	/*
562 	 * No state changes, so read locks are sufficient here.
563 	 */
564 	INP_LOCK_ASSERT(inp);
565 	INP_HASH_LOCK_ASSERT(pcbinfo);
566 
567 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
568 		return (EADDRNOTAVAIL);
569 	laddr.s_addr = *laddrp;
570 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
571 		return (EINVAL);
572 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
573 		lookupflags = INPLOOKUP_WILDCARD;
574 	if (nam == NULL) {
575 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
576 			return (error);
577 	} else {
578 		sin = (struct sockaddr_in *)nam;
579 		if (nam->sa_len != sizeof (*sin))
580 			return (EINVAL);
581 #ifdef notdef
582 		/*
583 		 * We should check the family, but old programs
584 		 * incorrectly fail to initialize it.
585 		 */
586 		if (sin->sin_family != AF_INET)
587 			return (EAFNOSUPPORT);
588 #endif
589 		error = prison_local_ip4(cred, &sin->sin_addr);
590 		if (error)
591 			return (error);
592 		if (sin->sin_port != *lportp) {
593 			/* Don't allow the port to change. */
594 			if (*lportp != 0)
595 				return (EINVAL);
596 			lport = sin->sin_port;
597 		}
598 		/* NB: lport is left as 0 if the port isn't being changed. */
599 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
600 			/*
601 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
602 			 * allow complete duplication of binding if
603 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
604 			 * and a multicast address is bound on both
605 			 * new and duplicated sockets.
606 			 */
607 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
608 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
609 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
610 			sin->sin_port = 0;		/* yech... */
611 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
612 			/*
613 			 * Is the address a local IP address?
614 			 * If INP_BINDANY is set, then the socket may be bound
615 			 * to any endpoint address, local or not.
616 			 */
617 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
618 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
619 				return (EADDRNOTAVAIL);
620 		}
621 		laddr = sin->sin_addr;
622 		if (lport) {
623 			struct inpcb *t;
624 			struct tcptw *tw;
625 
626 			/* GROSS */
627 			if (ntohs(lport) <= V_ipport_reservedhigh &&
628 			    ntohs(lport) >= V_ipport_reservedlow &&
629 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
630 			    0))
631 				return (EACCES);
632 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
633 			    priv_check_cred(inp->inp_cred,
634 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
635 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
636 				    lport, INPLOOKUP_WILDCARD, cred);
637 	/*
638 	 * XXX
639 	 * This entire block sorely needs a rewrite.
640 	 */
641 				if (t &&
642 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
643 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
644 				    (so->so_type != SOCK_STREAM ||
645 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
646 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
647 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
648 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
649 				    (inp->inp_cred->cr_uid !=
650 				     t->inp_cred->cr_uid))
651 					return (EADDRINUSE);
652 
653 				/*
654 				 * If the socket is a BINDMULTI socket, then
655 				 * the credentials need to match and the
656 				 * original socket also has to have been bound
657 				 * with BINDMULTI.
658 				 */
659 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
660 					return (EADDRINUSE);
661 			}
662 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
663 			    lport, lookupflags, cred);
664 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
665 				/*
666 				 * XXXRW: If an incpb has had its timewait
667 				 * state recycled, we treat the address as
668 				 * being in use (for now).  This is better
669 				 * than a panic, but not desirable.
670 				 */
671 				tw = intotw(t);
672 				if (tw == NULL ||
673 				    (reuseport & tw->tw_so_options) == 0)
674 					return (EADDRINUSE);
675 			} else if (t &&
676 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
677 			    (reuseport & inp_so_options(t)) == 0) {
678 #ifdef INET6
679 				if (ntohl(sin->sin_addr.s_addr) !=
680 				    INADDR_ANY ||
681 				    ntohl(t->inp_laddr.s_addr) !=
682 				    INADDR_ANY ||
683 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
684 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
685 #endif
686 				return (EADDRINUSE);
687 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
688 					return (EADDRINUSE);
689 			}
690 		}
691 	}
692 	if (*lportp != 0)
693 		lport = *lportp;
694 	if (lport == 0) {
695 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
696 		if (error != 0)
697 			return (error);
698 
699 	}
700 	*laddrp = laddr.s_addr;
701 	*lportp = lport;
702 	return (0);
703 }
704 
705 /*
706  * Connect from a socket to a specified address.
707  * Both address and port must be specified in argument sin.
708  * If don't have a local address for this socket yet,
709  * then pick one.
710  */
711 int
712 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
713     struct ucred *cred, struct mbuf *m)
714 {
715 	u_short lport, fport;
716 	in_addr_t laddr, faddr;
717 	int anonport, error;
718 
719 	INP_WLOCK_ASSERT(inp);
720 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
721 
722 	lport = inp->inp_lport;
723 	laddr = inp->inp_laddr.s_addr;
724 	anonport = (lport == 0);
725 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
726 	    NULL, cred);
727 	if (error)
728 		return (error);
729 
730 	/* Do the initial binding of the local address if required. */
731 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
732 		inp->inp_lport = lport;
733 		inp->inp_laddr.s_addr = laddr;
734 		if (in_pcbinshash(inp) != 0) {
735 			inp->inp_laddr.s_addr = INADDR_ANY;
736 			inp->inp_lport = 0;
737 			return (EAGAIN);
738 		}
739 	}
740 
741 	/* Commit the remaining changes. */
742 	inp->inp_lport = lport;
743 	inp->inp_laddr.s_addr = laddr;
744 	inp->inp_faddr.s_addr = faddr;
745 	inp->inp_fport = fport;
746 	in_pcbrehash_mbuf(inp, m);
747 
748 	if (anonport)
749 		inp->inp_flags |= INP_ANONPORT;
750 	return (0);
751 }
752 
753 int
754 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
755 {
756 
757 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
758 }
759 
760 /*
761  * Do proper source address selection on an unbound socket in case
762  * of connect. Take jails into account as well.
763  */
764 int
765 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
766     struct ucred *cred)
767 {
768 	struct ifaddr *ifa;
769 	struct sockaddr *sa;
770 	struct sockaddr_in *sin;
771 	struct route sro;
772 	int error;
773 
774 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
775 
776 	/*
777 	 * Bypass source address selection and use the primary jail IP
778 	 * if requested.
779 	 */
780 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
781 		return (0);
782 
783 	error = 0;
784 	bzero(&sro, sizeof(sro));
785 
786 	sin = (struct sockaddr_in *)&sro.ro_dst;
787 	sin->sin_family = AF_INET;
788 	sin->sin_len = sizeof(struct sockaddr_in);
789 	sin->sin_addr.s_addr = faddr->s_addr;
790 
791 	/*
792 	 * If route is known our src addr is taken from the i/f,
793 	 * else punt.
794 	 *
795 	 * Find out route to destination.
796 	 */
797 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
798 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
799 
800 	/*
801 	 * If we found a route, use the address corresponding to
802 	 * the outgoing interface.
803 	 *
804 	 * Otherwise assume faddr is reachable on a directly connected
805 	 * network and try to find a corresponding interface to take
806 	 * the source address from.
807 	 */
808 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
809 		struct in_ifaddr *ia;
810 		struct ifnet *ifp;
811 
812 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
813 					inp->inp_socket->so_fibnum));
814 		if (ia == NULL)
815 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
816 						inp->inp_socket->so_fibnum));
817 		if (ia == NULL) {
818 			error = ENETUNREACH;
819 			goto done;
820 		}
821 
822 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
823 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
824 			ifa_free(&ia->ia_ifa);
825 			goto done;
826 		}
827 
828 		ifp = ia->ia_ifp;
829 		ifa_free(&ia->ia_ifa);
830 		ia = NULL;
831 		IF_ADDR_RLOCK(ifp);
832 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
833 
834 			sa = ifa->ifa_addr;
835 			if (sa->sa_family != AF_INET)
836 				continue;
837 			sin = (struct sockaddr_in *)sa;
838 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
839 				ia = (struct in_ifaddr *)ifa;
840 				break;
841 			}
842 		}
843 		if (ia != NULL) {
844 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
845 			IF_ADDR_RUNLOCK(ifp);
846 			goto done;
847 		}
848 		IF_ADDR_RUNLOCK(ifp);
849 
850 		/* 3. As a last resort return the 'default' jail address. */
851 		error = prison_get_ip4(cred, laddr);
852 		goto done;
853 	}
854 
855 	/*
856 	 * If the outgoing interface on the route found is not
857 	 * a loopback interface, use the address from that interface.
858 	 * In case of jails do those three steps:
859 	 * 1. check if the interface address belongs to the jail. If so use it.
860 	 * 2. check if we have any address on the outgoing interface
861 	 *    belonging to this jail. If so use it.
862 	 * 3. as a last resort return the 'default' jail address.
863 	 */
864 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
865 		struct in_ifaddr *ia;
866 		struct ifnet *ifp;
867 
868 		/* If not jailed, use the default returned. */
869 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
870 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
871 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
872 			goto done;
873 		}
874 
875 		/* Jailed. */
876 		/* 1. Check if the iface address belongs to the jail. */
877 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
878 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
879 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
880 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
881 			goto done;
882 		}
883 
884 		/*
885 		 * 2. Check if we have any address on the outgoing interface
886 		 *    belonging to this jail.
887 		 */
888 		ia = NULL;
889 		ifp = sro.ro_rt->rt_ifp;
890 		IF_ADDR_RLOCK(ifp);
891 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
892 			sa = ifa->ifa_addr;
893 			if (sa->sa_family != AF_INET)
894 				continue;
895 			sin = (struct sockaddr_in *)sa;
896 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
897 				ia = (struct in_ifaddr *)ifa;
898 				break;
899 			}
900 		}
901 		if (ia != NULL) {
902 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
903 			IF_ADDR_RUNLOCK(ifp);
904 			goto done;
905 		}
906 		IF_ADDR_RUNLOCK(ifp);
907 
908 		/* 3. As a last resort return the 'default' jail address. */
909 		error = prison_get_ip4(cred, laddr);
910 		goto done;
911 	}
912 
913 	/*
914 	 * The outgoing interface is marked with 'loopback net', so a route
915 	 * to ourselves is here.
916 	 * Try to find the interface of the destination address and then
917 	 * take the address from there. That interface is not necessarily
918 	 * a loopback interface.
919 	 * In case of jails, check that it is an address of the jail
920 	 * and if we cannot find, fall back to the 'default' jail address.
921 	 */
922 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
923 		struct sockaddr_in sain;
924 		struct in_ifaddr *ia;
925 
926 		bzero(&sain, sizeof(struct sockaddr_in));
927 		sain.sin_family = AF_INET;
928 		sain.sin_len = sizeof(struct sockaddr_in);
929 		sain.sin_addr.s_addr = faddr->s_addr;
930 
931 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
932 					inp->inp_socket->so_fibnum));
933 		if (ia == NULL)
934 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
935 						inp->inp_socket->so_fibnum));
936 		if (ia == NULL)
937 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
938 
939 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
940 			if (ia == NULL) {
941 				error = ENETUNREACH;
942 				goto done;
943 			}
944 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
945 			ifa_free(&ia->ia_ifa);
946 			goto done;
947 		}
948 
949 		/* Jailed. */
950 		if (ia != NULL) {
951 			struct ifnet *ifp;
952 
953 			ifp = ia->ia_ifp;
954 			ifa_free(&ia->ia_ifa);
955 			ia = NULL;
956 			IF_ADDR_RLOCK(ifp);
957 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
958 
959 				sa = ifa->ifa_addr;
960 				if (sa->sa_family != AF_INET)
961 					continue;
962 				sin = (struct sockaddr_in *)sa;
963 				if (prison_check_ip4(cred,
964 				    &sin->sin_addr) == 0) {
965 					ia = (struct in_ifaddr *)ifa;
966 					break;
967 				}
968 			}
969 			if (ia != NULL) {
970 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
971 				IF_ADDR_RUNLOCK(ifp);
972 				goto done;
973 			}
974 			IF_ADDR_RUNLOCK(ifp);
975 		}
976 
977 		/* 3. As a last resort return the 'default' jail address. */
978 		error = prison_get_ip4(cred, laddr);
979 		goto done;
980 	}
981 
982 done:
983 	if (sro.ro_rt != NULL)
984 		RTFREE(sro.ro_rt);
985 	return (error);
986 }
987 
988 /*
989  * Set up for a connect from a socket to the specified address.
990  * On entry, *laddrp and *lportp should contain the current local
991  * address and port for the PCB; these are updated to the values
992  * that should be placed in inp_laddr and inp_lport to complete
993  * the connect.
994  *
995  * On success, *faddrp and *fportp will be set to the remote address
996  * and port. These are not updated in the error case.
997  *
998  * If the operation fails because the connection already exists,
999  * *oinpp will be set to the PCB of that connection so that the
1000  * caller can decide to override it. In all other cases, *oinpp
1001  * is set to NULL.
1002  */
1003 int
1004 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1005     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1006     struct inpcb **oinpp, struct ucred *cred)
1007 {
1008 	struct rm_priotracker in_ifa_tracker;
1009 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1010 	struct in_ifaddr *ia;
1011 	struct inpcb *oinp;
1012 	struct in_addr laddr, faddr;
1013 	u_short lport, fport;
1014 	int error;
1015 
1016 	/*
1017 	 * Because a global state change doesn't actually occur here, a read
1018 	 * lock is sufficient.
1019 	 */
1020 	INP_LOCK_ASSERT(inp);
1021 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1022 
1023 	if (oinpp != NULL)
1024 		*oinpp = NULL;
1025 	if (nam->sa_len != sizeof (*sin))
1026 		return (EINVAL);
1027 	if (sin->sin_family != AF_INET)
1028 		return (EAFNOSUPPORT);
1029 	if (sin->sin_port == 0)
1030 		return (EADDRNOTAVAIL);
1031 	laddr.s_addr = *laddrp;
1032 	lport = *lportp;
1033 	faddr = sin->sin_addr;
1034 	fport = sin->sin_port;
1035 
1036 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1037 		/*
1038 		 * If the destination address is INADDR_ANY,
1039 		 * use the primary local address.
1040 		 * If the supplied address is INADDR_BROADCAST,
1041 		 * and the primary interface supports broadcast,
1042 		 * choose the broadcast address for that interface.
1043 		 */
1044 		if (faddr.s_addr == INADDR_ANY) {
1045 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1046 			faddr =
1047 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1048 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1049 			if (cred != NULL &&
1050 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1051 				return (error);
1052 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1053 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1054 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1055 			    IFF_BROADCAST)
1056 				faddr = satosin(&TAILQ_FIRST(
1057 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1058 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1059 		}
1060 	}
1061 	if (laddr.s_addr == INADDR_ANY) {
1062 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1063 		/*
1064 		 * If the destination address is multicast and an outgoing
1065 		 * interface has been set as a multicast option, prefer the
1066 		 * address of that interface as our source address.
1067 		 */
1068 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1069 		    inp->inp_moptions != NULL) {
1070 			struct ip_moptions *imo;
1071 			struct ifnet *ifp;
1072 
1073 			imo = inp->inp_moptions;
1074 			if (imo->imo_multicast_ifp != NULL) {
1075 				ifp = imo->imo_multicast_ifp;
1076 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1077 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1078 					if ((ia->ia_ifp == ifp) &&
1079 					    (cred == NULL ||
1080 					    prison_check_ip4(cred,
1081 					    &ia->ia_addr.sin_addr) == 0))
1082 						break;
1083 				}
1084 				if (ia == NULL)
1085 					error = EADDRNOTAVAIL;
1086 				else {
1087 					laddr = ia->ia_addr.sin_addr;
1088 					error = 0;
1089 				}
1090 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1091 			}
1092 		}
1093 		if (error)
1094 			return (error);
1095 	}
1096 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1097 	    laddr, lport, 0, NULL);
1098 	if (oinp != NULL) {
1099 		if (oinpp != NULL)
1100 			*oinpp = oinp;
1101 		return (EADDRINUSE);
1102 	}
1103 	if (lport == 0) {
1104 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1105 		    cred);
1106 		if (error)
1107 			return (error);
1108 	}
1109 	*laddrp = laddr.s_addr;
1110 	*lportp = lport;
1111 	*faddrp = faddr.s_addr;
1112 	*fportp = fport;
1113 	return (0);
1114 }
1115 
1116 void
1117 in_pcbdisconnect(struct inpcb *inp)
1118 {
1119 
1120 	INP_WLOCK_ASSERT(inp);
1121 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1122 
1123 	inp->inp_faddr.s_addr = INADDR_ANY;
1124 	inp->inp_fport = 0;
1125 	in_pcbrehash(inp);
1126 }
1127 #endif /* INET */
1128 
1129 /*
1130  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1131  * For most protocols, this will be invoked immediately prior to calling
1132  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1133  * socket, in which case in_pcbfree() is deferred.
1134  */
1135 void
1136 in_pcbdetach(struct inpcb *inp)
1137 {
1138 
1139 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1140 
1141 	inp->inp_socket->so_pcb = NULL;
1142 	inp->inp_socket = NULL;
1143 }
1144 
1145 /*
1146  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1147  * stability of an inpcb pointer despite the inpcb lock being released.  This
1148  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1149  * but where the inpcb lock may already held, or when acquiring a reference
1150  * via a pcbgroup.
1151  *
1152  * in_pcbref() should be used only to provide brief memory stability, and
1153  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1154  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1155  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1156  * lock and rele are the *only* safe operations that may be performed on the
1157  * inpcb.
1158  *
1159  * While the inpcb will not be freed, releasing the inpcb lock means that the
1160  * connection's state may change, so the caller should be careful to
1161  * revalidate any cached state on reacquiring the lock.  Drop the reference
1162  * using in_pcbrele().
1163  */
1164 void
1165 in_pcbref(struct inpcb *inp)
1166 {
1167 
1168 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1169 
1170 	refcount_acquire(&inp->inp_refcount);
1171 }
1172 
1173 /*
1174  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1175  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1176  * return a flag indicating whether or not the inpcb remains valid.  If it is
1177  * valid, we return with the inpcb lock held.
1178  *
1179  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1180  * reference on an inpcb.  Historically more work was done here (actually, in
1181  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1182  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1183  * about memory stability (and continued use of the write lock).
1184  */
1185 int
1186 in_pcbrele_rlocked(struct inpcb *inp)
1187 {
1188 	struct inpcbinfo *pcbinfo;
1189 
1190 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1191 
1192 	INP_RLOCK_ASSERT(inp);
1193 
1194 	if (refcount_release(&inp->inp_refcount) == 0) {
1195 		/*
1196 		 * If the inpcb has been freed, let the caller know, even if
1197 		 * this isn't the last reference.
1198 		 */
1199 		if (inp->inp_flags2 & INP_FREED) {
1200 			INP_RUNLOCK(inp);
1201 			return (1);
1202 		}
1203 		return (0);
1204 	}
1205 
1206 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1207 
1208 	INP_RUNLOCK(inp);
1209 	pcbinfo = inp->inp_pcbinfo;
1210 	uma_zfree(pcbinfo->ipi_zone, inp);
1211 	return (1);
1212 }
1213 
1214 int
1215 in_pcbrele_wlocked(struct inpcb *inp)
1216 {
1217 	struct inpcbinfo *pcbinfo;
1218 
1219 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1220 
1221 	INP_WLOCK_ASSERT(inp);
1222 
1223 	if (refcount_release(&inp->inp_refcount) == 0) {
1224 		/*
1225 		 * If the inpcb has been freed, let the caller know, even if
1226 		 * this isn't the last reference.
1227 		 */
1228 		if (inp->inp_flags2 & INP_FREED) {
1229 			INP_WUNLOCK(inp);
1230 			return (1);
1231 		}
1232 		return (0);
1233 	}
1234 
1235 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1236 
1237 	INP_WUNLOCK(inp);
1238 	pcbinfo = inp->inp_pcbinfo;
1239 	uma_zfree(pcbinfo->ipi_zone, inp);
1240 	return (1);
1241 }
1242 
1243 /*
1244  * Temporary wrapper.
1245  */
1246 int
1247 in_pcbrele(struct inpcb *inp)
1248 {
1249 
1250 	return (in_pcbrele_wlocked(inp));
1251 }
1252 
1253 /*
1254  * Unconditionally schedule an inpcb to be freed by decrementing its
1255  * reference count, which should occur only after the inpcb has been detached
1256  * from its socket.  If another thread holds a temporary reference (acquired
1257  * using in_pcbref()) then the free is deferred until that reference is
1258  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1259  * work, including removal from global lists, is done in this context, where
1260  * the pcbinfo lock is held.
1261  */
1262 void
1263 in_pcbfree(struct inpcb *inp)
1264 {
1265 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1266 
1267 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1268 
1269 #ifdef INVARIANTS
1270 	if (pcbinfo == &V_tcbinfo) {
1271 		INP_INFO_LOCK_ASSERT(pcbinfo);
1272 	} else {
1273 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1274 	}
1275 #endif
1276 	INP_WLOCK_ASSERT(inp);
1277 
1278 	/* XXXRW: Do as much as possible here. */
1279 #ifdef IPSEC
1280 	if (inp->inp_sp != NULL)
1281 		ipsec_delete_pcbpolicy(inp);
1282 #endif
1283 	INP_LIST_WLOCK(pcbinfo);
1284 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1285 	in_pcbremlists(inp);
1286 	INP_LIST_WUNLOCK(pcbinfo);
1287 #ifdef INET6
1288 	if (inp->inp_vflag & INP_IPV6PROTO) {
1289 		ip6_freepcbopts(inp->in6p_outputopts);
1290 		if (inp->in6p_moptions != NULL)
1291 			ip6_freemoptions(inp->in6p_moptions);
1292 	}
1293 #endif
1294 	if (inp->inp_options)
1295 		(void)m_free(inp->inp_options);
1296 #ifdef INET
1297 	if (inp->inp_moptions != NULL)
1298 		inp_freemoptions(inp->inp_moptions);
1299 #endif
1300 	inp->inp_vflag = 0;
1301 	inp->inp_flags2 |= INP_FREED;
1302 	crfree(inp->inp_cred);
1303 #ifdef MAC
1304 	mac_inpcb_destroy(inp);
1305 #endif
1306 	if (!in_pcbrele_wlocked(inp))
1307 		INP_WUNLOCK(inp);
1308 }
1309 
1310 /*
1311  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1312  * port reservation, and preventing it from being returned by inpcb lookups.
1313  *
1314  * It is used by TCP to mark an inpcb as unused and avoid future packet
1315  * delivery or event notification when a socket remains open but TCP has
1316  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1317  * or a RST on the wire, and allows the port binding to be reused while still
1318  * maintaining the invariant that so_pcb always points to a valid inpcb until
1319  * in_pcbdetach().
1320  *
1321  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1322  * in_pcbnotifyall() and in_pcbpurgeif0()?
1323  */
1324 void
1325 in_pcbdrop(struct inpcb *inp)
1326 {
1327 
1328 	INP_WLOCK_ASSERT(inp);
1329 
1330 	/*
1331 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1332 	 * the hash lock...?
1333 	 */
1334 	inp->inp_flags |= INP_DROPPED;
1335 	if (inp->inp_flags & INP_INHASHLIST) {
1336 		struct inpcbport *phd = inp->inp_phd;
1337 
1338 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1339 		LIST_REMOVE(inp, inp_hash);
1340 		LIST_REMOVE(inp, inp_portlist);
1341 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1342 			LIST_REMOVE(phd, phd_hash);
1343 			free(phd, M_PCB);
1344 		}
1345 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1346 		inp->inp_flags &= ~INP_INHASHLIST;
1347 #ifdef PCBGROUP
1348 		in_pcbgroup_remove(inp);
1349 #endif
1350 	}
1351 }
1352 
1353 #ifdef INET
1354 /*
1355  * Common routines to return the socket addresses associated with inpcbs.
1356  */
1357 struct sockaddr *
1358 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1359 {
1360 	struct sockaddr_in *sin;
1361 
1362 	sin = malloc(sizeof *sin, M_SONAME,
1363 		M_WAITOK | M_ZERO);
1364 	sin->sin_family = AF_INET;
1365 	sin->sin_len = sizeof(*sin);
1366 	sin->sin_addr = *addr_p;
1367 	sin->sin_port = port;
1368 
1369 	return (struct sockaddr *)sin;
1370 }
1371 
1372 int
1373 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1374 {
1375 	struct inpcb *inp;
1376 	struct in_addr addr;
1377 	in_port_t port;
1378 
1379 	inp = sotoinpcb(so);
1380 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1381 
1382 	INP_RLOCK(inp);
1383 	port = inp->inp_lport;
1384 	addr = inp->inp_laddr;
1385 	INP_RUNLOCK(inp);
1386 
1387 	*nam = in_sockaddr(port, &addr);
1388 	return 0;
1389 }
1390 
1391 int
1392 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1393 {
1394 	struct inpcb *inp;
1395 	struct in_addr addr;
1396 	in_port_t port;
1397 
1398 	inp = sotoinpcb(so);
1399 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1400 
1401 	INP_RLOCK(inp);
1402 	port = inp->inp_fport;
1403 	addr = inp->inp_faddr;
1404 	INP_RUNLOCK(inp);
1405 
1406 	*nam = in_sockaddr(port, &addr);
1407 	return 0;
1408 }
1409 
1410 void
1411 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1412     struct inpcb *(*notify)(struct inpcb *, int))
1413 {
1414 	struct inpcb *inp, *inp_temp;
1415 
1416 	INP_INFO_WLOCK(pcbinfo);
1417 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1418 		INP_WLOCK(inp);
1419 #ifdef INET6
1420 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1421 			INP_WUNLOCK(inp);
1422 			continue;
1423 		}
1424 #endif
1425 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1426 		    inp->inp_socket == NULL) {
1427 			INP_WUNLOCK(inp);
1428 			continue;
1429 		}
1430 		if ((*notify)(inp, errno))
1431 			INP_WUNLOCK(inp);
1432 	}
1433 	INP_INFO_WUNLOCK(pcbinfo);
1434 }
1435 
1436 void
1437 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1438 {
1439 	struct inpcb *inp;
1440 	struct ip_moptions *imo;
1441 	int i, gap;
1442 
1443 	INP_INFO_WLOCK(pcbinfo);
1444 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1445 		INP_WLOCK(inp);
1446 		imo = inp->inp_moptions;
1447 		if ((inp->inp_vflag & INP_IPV4) &&
1448 		    imo != NULL) {
1449 			/*
1450 			 * Unselect the outgoing interface if it is being
1451 			 * detached.
1452 			 */
1453 			if (imo->imo_multicast_ifp == ifp)
1454 				imo->imo_multicast_ifp = NULL;
1455 
1456 			/*
1457 			 * Drop multicast group membership if we joined
1458 			 * through the interface being detached.
1459 			 */
1460 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1461 			    i++) {
1462 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1463 					in_delmulti(imo->imo_membership[i]);
1464 					gap++;
1465 				} else if (gap != 0)
1466 					imo->imo_membership[i - gap] =
1467 					    imo->imo_membership[i];
1468 			}
1469 			imo->imo_num_memberships -= gap;
1470 		}
1471 		INP_WUNLOCK(inp);
1472 	}
1473 	INP_INFO_WUNLOCK(pcbinfo);
1474 }
1475 
1476 /*
1477  * Lookup a PCB based on the local address and port.  Caller must hold the
1478  * hash lock.  No inpcb locks or references are acquired.
1479  */
1480 #define INP_LOOKUP_MAPPED_PCB_COST	3
1481 struct inpcb *
1482 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1483     u_short lport, int lookupflags, struct ucred *cred)
1484 {
1485 	struct inpcb *inp;
1486 #ifdef INET6
1487 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1488 #else
1489 	int matchwild = 3;
1490 #endif
1491 	int wildcard;
1492 
1493 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1494 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1495 
1496 	INP_HASH_LOCK_ASSERT(pcbinfo);
1497 
1498 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1499 		struct inpcbhead *head;
1500 		/*
1501 		 * Look for an unconnected (wildcard foreign addr) PCB that
1502 		 * matches the local address and port we're looking for.
1503 		 */
1504 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1505 		    0, pcbinfo->ipi_hashmask)];
1506 		LIST_FOREACH(inp, head, inp_hash) {
1507 #ifdef INET6
1508 			/* XXX inp locking */
1509 			if ((inp->inp_vflag & INP_IPV4) == 0)
1510 				continue;
1511 #endif
1512 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1513 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1514 			    inp->inp_lport == lport) {
1515 				/*
1516 				 * Found?
1517 				 */
1518 				if (cred == NULL ||
1519 				    prison_equal_ip4(cred->cr_prison,
1520 					inp->inp_cred->cr_prison))
1521 					return (inp);
1522 			}
1523 		}
1524 		/*
1525 		 * Not found.
1526 		 */
1527 		return (NULL);
1528 	} else {
1529 		struct inpcbporthead *porthash;
1530 		struct inpcbport *phd;
1531 		struct inpcb *match = NULL;
1532 		/*
1533 		 * Best fit PCB lookup.
1534 		 *
1535 		 * First see if this local port is in use by looking on the
1536 		 * port hash list.
1537 		 */
1538 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1539 		    pcbinfo->ipi_porthashmask)];
1540 		LIST_FOREACH(phd, porthash, phd_hash) {
1541 			if (phd->phd_port == lport)
1542 				break;
1543 		}
1544 		if (phd != NULL) {
1545 			/*
1546 			 * Port is in use by one or more PCBs. Look for best
1547 			 * fit.
1548 			 */
1549 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1550 				wildcard = 0;
1551 				if (cred != NULL &&
1552 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1553 					cred->cr_prison))
1554 					continue;
1555 #ifdef INET6
1556 				/* XXX inp locking */
1557 				if ((inp->inp_vflag & INP_IPV4) == 0)
1558 					continue;
1559 				/*
1560 				 * We never select the PCB that has
1561 				 * INP_IPV6 flag and is bound to :: if
1562 				 * we have another PCB which is bound
1563 				 * to 0.0.0.0.  If a PCB has the
1564 				 * INP_IPV6 flag, then we set its cost
1565 				 * higher than IPv4 only PCBs.
1566 				 *
1567 				 * Note that the case only happens
1568 				 * when a socket is bound to ::, under
1569 				 * the condition that the use of the
1570 				 * mapped address is allowed.
1571 				 */
1572 				if ((inp->inp_vflag & INP_IPV6) != 0)
1573 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1574 #endif
1575 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1576 					wildcard++;
1577 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1578 					if (laddr.s_addr == INADDR_ANY)
1579 						wildcard++;
1580 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1581 						continue;
1582 				} else {
1583 					if (laddr.s_addr != INADDR_ANY)
1584 						wildcard++;
1585 				}
1586 				if (wildcard < matchwild) {
1587 					match = inp;
1588 					matchwild = wildcard;
1589 					if (matchwild == 0)
1590 						break;
1591 				}
1592 			}
1593 		}
1594 		return (match);
1595 	}
1596 }
1597 #undef INP_LOOKUP_MAPPED_PCB_COST
1598 
1599 #ifdef PCBGROUP
1600 /*
1601  * Lookup PCB in hash list, using pcbgroup tables.
1602  */
1603 static struct inpcb *
1604 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1605     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1606     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1607 {
1608 	struct inpcbhead *head;
1609 	struct inpcb *inp, *tmpinp;
1610 	u_short fport = fport_arg, lport = lport_arg;
1611 
1612 	/*
1613 	 * First look for an exact match.
1614 	 */
1615 	tmpinp = NULL;
1616 	INP_GROUP_LOCK(pcbgroup);
1617 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1618 	    pcbgroup->ipg_hashmask)];
1619 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1620 #ifdef INET6
1621 		/* XXX inp locking */
1622 		if ((inp->inp_vflag & INP_IPV4) == 0)
1623 			continue;
1624 #endif
1625 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1626 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1627 		    inp->inp_fport == fport &&
1628 		    inp->inp_lport == lport) {
1629 			/*
1630 			 * XXX We should be able to directly return
1631 			 * the inp here, without any checks.
1632 			 * Well unless both bound with SO_REUSEPORT?
1633 			 */
1634 			if (prison_flag(inp->inp_cred, PR_IP4))
1635 				goto found;
1636 			if (tmpinp == NULL)
1637 				tmpinp = inp;
1638 		}
1639 	}
1640 	if (tmpinp != NULL) {
1641 		inp = tmpinp;
1642 		goto found;
1643 	}
1644 
1645 #ifdef	RSS
1646 	/*
1647 	 * For incoming connections, we may wish to do a wildcard
1648 	 * match for an RSS-local socket.
1649 	 */
1650 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1651 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1652 #ifdef INET6
1653 		struct inpcb *local_wild_mapped = NULL;
1654 #endif
1655 		struct inpcb *jail_wild = NULL;
1656 		struct inpcbhead *head;
1657 		int injail;
1658 
1659 		/*
1660 		 * Order of socket selection - we always prefer jails.
1661 		 *      1. jailed, non-wild.
1662 		 *      2. jailed, wild.
1663 		 *      3. non-jailed, non-wild.
1664 		 *      4. non-jailed, wild.
1665 		 */
1666 
1667 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1668 		    lport, 0, pcbgroup->ipg_hashmask)];
1669 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1670 #ifdef INET6
1671 			/* XXX inp locking */
1672 			if ((inp->inp_vflag & INP_IPV4) == 0)
1673 				continue;
1674 #endif
1675 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1676 			    inp->inp_lport != lport)
1677 				continue;
1678 
1679 			injail = prison_flag(inp->inp_cred, PR_IP4);
1680 			if (injail) {
1681 				if (prison_check_ip4(inp->inp_cred,
1682 				    &laddr) != 0)
1683 					continue;
1684 			} else {
1685 				if (local_exact != NULL)
1686 					continue;
1687 			}
1688 
1689 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1690 				if (injail)
1691 					goto found;
1692 				else
1693 					local_exact = inp;
1694 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1695 #ifdef INET6
1696 				/* XXX inp locking, NULL check */
1697 				if (inp->inp_vflag & INP_IPV6PROTO)
1698 					local_wild_mapped = inp;
1699 				else
1700 #endif
1701 					if (injail)
1702 						jail_wild = inp;
1703 					else
1704 						local_wild = inp;
1705 			}
1706 		} /* LIST_FOREACH */
1707 
1708 		inp = jail_wild;
1709 		if (inp == NULL)
1710 			inp = local_exact;
1711 		if (inp == NULL)
1712 			inp = local_wild;
1713 #ifdef INET6
1714 		if (inp == NULL)
1715 			inp = local_wild_mapped;
1716 #endif
1717 		if (inp != NULL)
1718 			goto found;
1719 	}
1720 #endif
1721 
1722 	/*
1723 	 * Then look for a wildcard match, if requested.
1724 	 */
1725 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1726 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1727 #ifdef INET6
1728 		struct inpcb *local_wild_mapped = NULL;
1729 #endif
1730 		struct inpcb *jail_wild = NULL;
1731 		struct inpcbhead *head;
1732 		int injail;
1733 
1734 		/*
1735 		 * Order of socket selection - we always prefer jails.
1736 		 *      1. jailed, non-wild.
1737 		 *      2. jailed, wild.
1738 		 *      3. non-jailed, non-wild.
1739 		 *      4. non-jailed, wild.
1740 		 */
1741 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1742 		    0, pcbinfo->ipi_wildmask)];
1743 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1744 #ifdef INET6
1745 			/* XXX inp locking */
1746 			if ((inp->inp_vflag & INP_IPV4) == 0)
1747 				continue;
1748 #endif
1749 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1750 			    inp->inp_lport != lport)
1751 				continue;
1752 
1753 			injail = prison_flag(inp->inp_cred, PR_IP4);
1754 			if (injail) {
1755 				if (prison_check_ip4(inp->inp_cred,
1756 				    &laddr) != 0)
1757 					continue;
1758 			} else {
1759 				if (local_exact != NULL)
1760 					continue;
1761 			}
1762 
1763 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1764 				if (injail)
1765 					goto found;
1766 				else
1767 					local_exact = inp;
1768 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1769 #ifdef INET6
1770 				/* XXX inp locking, NULL check */
1771 				if (inp->inp_vflag & INP_IPV6PROTO)
1772 					local_wild_mapped = inp;
1773 				else
1774 #endif
1775 					if (injail)
1776 						jail_wild = inp;
1777 					else
1778 						local_wild = inp;
1779 			}
1780 		} /* LIST_FOREACH */
1781 		inp = jail_wild;
1782 		if (inp == NULL)
1783 			inp = local_exact;
1784 		if (inp == NULL)
1785 			inp = local_wild;
1786 #ifdef INET6
1787 		if (inp == NULL)
1788 			inp = local_wild_mapped;
1789 #endif
1790 		if (inp != NULL)
1791 			goto found;
1792 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1793 	INP_GROUP_UNLOCK(pcbgroup);
1794 	return (NULL);
1795 
1796 found:
1797 	in_pcbref(inp);
1798 	INP_GROUP_UNLOCK(pcbgroup);
1799 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1800 		INP_WLOCK(inp);
1801 		if (in_pcbrele_wlocked(inp))
1802 			return (NULL);
1803 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1804 		INP_RLOCK(inp);
1805 		if (in_pcbrele_rlocked(inp))
1806 			return (NULL);
1807 	} else
1808 		panic("%s: locking bug", __func__);
1809 	return (inp);
1810 }
1811 #endif /* PCBGROUP */
1812 
1813 /*
1814  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1815  * that the caller has locked the hash list, and will not perform any further
1816  * locking or reference operations on either the hash list or the connection.
1817  */
1818 static struct inpcb *
1819 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1820     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1821     struct ifnet *ifp)
1822 {
1823 	struct inpcbhead *head;
1824 	struct inpcb *inp, *tmpinp;
1825 	u_short fport = fport_arg, lport = lport_arg;
1826 
1827 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1828 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1829 
1830 	INP_HASH_LOCK_ASSERT(pcbinfo);
1831 
1832 	/*
1833 	 * First look for an exact match.
1834 	 */
1835 	tmpinp = NULL;
1836 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1837 	    pcbinfo->ipi_hashmask)];
1838 	LIST_FOREACH(inp, head, inp_hash) {
1839 #ifdef INET6
1840 		/* XXX inp locking */
1841 		if ((inp->inp_vflag & INP_IPV4) == 0)
1842 			continue;
1843 #endif
1844 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1845 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1846 		    inp->inp_fport == fport &&
1847 		    inp->inp_lport == lport) {
1848 			/*
1849 			 * XXX We should be able to directly return
1850 			 * the inp here, without any checks.
1851 			 * Well unless both bound with SO_REUSEPORT?
1852 			 */
1853 			if (prison_flag(inp->inp_cred, PR_IP4))
1854 				return (inp);
1855 			if (tmpinp == NULL)
1856 				tmpinp = inp;
1857 		}
1858 	}
1859 	if (tmpinp != NULL)
1860 		return (tmpinp);
1861 
1862 	/*
1863 	 * Then look for a wildcard match, if requested.
1864 	 */
1865 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1866 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1867 #ifdef INET6
1868 		struct inpcb *local_wild_mapped = NULL;
1869 #endif
1870 		struct inpcb *jail_wild = NULL;
1871 		int injail;
1872 
1873 		/*
1874 		 * Order of socket selection - we always prefer jails.
1875 		 *      1. jailed, non-wild.
1876 		 *      2. jailed, wild.
1877 		 *      3. non-jailed, non-wild.
1878 		 *      4. non-jailed, wild.
1879 		 */
1880 
1881 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1882 		    0, pcbinfo->ipi_hashmask)];
1883 		LIST_FOREACH(inp, head, inp_hash) {
1884 #ifdef INET6
1885 			/* XXX inp locking */
1886 			if ((inp->inp_vflag & INP_IPV4) == 0)
1887 				continue;
1888 #endif
1889 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1890 			    inp->inp_lport != lport)
1891 				continue;
1892 
1893 			injail = prison_flag(inp->inp_cred, PR_IP4);
1894 			if (injail) {
1895 				if (prison_check_ip4(inp->inp_cred,
1896 				    &laddr) != 0)
1897 					continue;
1898 			} else {
1899 				if (local_exact != NULL)
1900 					continue;
1901 			}
1902 
1903 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1904 				if (injail)
1905 					return (inp);
1906 				else
1907 					local_exact = inp;
1908 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1909 #ifdef INET6
1910 				/* XXX inp locking, NULL check */
1911 				if (inp->inp_vflag & INP_IPV6PROTO)
1912 					local_wild_mapped = inp;
1913 				else
1914 #endif
1915 					if (injail)
1916 						jail_wild = inp;
1917 					else
1918 						local_wild = inp;
1919 			}
1920 		} /* LIST_FOREACH */
1921 		if (jail_wild != NULL)
1922 			return (jail_wild);
1923 		if (local_exact != NULL)
1924 			return (local_exact);
1925 		if (local_wild != NULL)
1926 			return (local_wild);
1927 #ifdef INET6
1928 		if (local_wild_mapped != NULL)
1929 			return (local_wild_mapped);
1930 #endif
1931 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1932 
1933 	return (NULL);
1934 }
1935 
1936 /*
1937  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1938  * hash list lock, and will return the inpcb locked (i.e., requires
1939  * INPLOOKUP_LOCKPCB).
1940  */
1941 static struct inpcb *
1942 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1943     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1944     struct ifnet *ifp)
1945 {
1946 	struct inpcb *inp;
1947 
1948 	INP_HASH_RLOCK(pcbinfo);
1949 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1950 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1951 	if (inp != NULL) {
1952 		in_pcbref(inp);
1953 		INP_HASH_RUNLOCK(pcbinfo);
1954 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1955 			INP_WLOCK(inp);
1956 			if (in_pcbrele_wlocked(inp))
1957 				return (NULL);
1958 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1959 			INP_RLOCK(inp);
1960 			if (in_pcbrele_rlocked(inp))
1961 				return (NULL);
1962 		} else
1963 			panic("%s: locking bug", __func__);
1964 	} else
1965 		INP_HASH_RUNLOCK(pcbinfo);
1966 	return (inp);
1967 }
1968 
1969 /*
1970  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1971  * from which a pre-calculated hash value may be extracted.
1972  *
1973  * Possibly more of this logic should be in in_pcbgroup.c.
1974  */
1975 struct inpcb *
1976 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1977     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1978 {
1979 #if defined(PCBGROUP) && !defined(RSS)
1980 	struct inpcbgroup *pcbgroup;
1981 #endif
1982 
1983 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1984 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1985 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1986 	    ("%s: LOCKPCB not set", __func__));
1987 
1988 	/*
1989 	 * When not using RSS, use connection groups in preference to the
1990 	 * reservation table when looking up 4-tuples.  When using RSS, just
1991 	 * use the reservation table, due to the cost of the Toeplitz hash
1992 	 * in software.
1993 	 *
1994 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
1995 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
1996 	 * in software.
1997 	 */
1998 #if defined(PCBGROUP) && !defined(RSS)
1999 	if (in_pcbgroup_enabled(pcbinfo)) {
2000 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2001 		    fport);
2002 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2003 		    laddr, lport, lookupflags, ifp));
2004 	}
2005 #endif
2006 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2007 	    lookupflags, ifp));
2008 }
2009 
2010 struct inpcb *
2011 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2012     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2013     struct ifnet *ifp, struct mbuf *m)
2014 {
2015 #ifdef PCBGROUP
2016 	struct inpcbgroup *pcbgroup;
2017 #endif
2018 
2019 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2020 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2021 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2022 	    ("%s: LOCKPCB not set", __func__));
2023 
2024 #ifdef PCBGROUP
2025 	/*
2026 	 * If we can use a hardware-generated hash to look up the connection
2027 	 * group, use that connection group to find the inpcb.  Otherwise
2028 	 * fall back on a software hash -- or the reservation table if we're
2029 	 * using RSS.
2030 	 *
2031 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2032 	 */
2033 	if (in_pcbgroup_enabled(pcbinfo) &&
2034 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2035 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2036 		    m->m_pkthdr.flowid);
2037 		if (pcbgroup != NULL)
2038 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2039 			    fport, laddr, lport, lookupflags, ifp));
2040 #ifndef RSS
2041 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2042 		    fport);
2043 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2044 		    laddr, lport, lookupflags, ifp));
2045 #endif
2046 	}
2047 #endif
2048 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2049 	    lookupflags, ifp));
2050 }
2051 #endif /* INET */
2052 
2053 /*
2054  * Insert PCB onto various hash lists.
2055  */
2056 static int
2057 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2058 {
2059 	struct inpcbhead *pcbhash;
2060 	struct inpcbporthead *pcbporthash;
2061 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2062 	struct inpcbport *phd;
2063 	u_int32_t hashkey_faddr;
2064 
2065 	INP_WLOCK_ASSERT(inp);
2066 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2067 
2068 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2069 	    ("in_pcbinshash: INP_INHASHLIST"));
2070 
2071 #ifdef INET6
2072 	if (inp->inp_vflag & INP_IPV6)
2073 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2074 	else
2075 #endif
2076 	hashkey_faddr = inp->inp_faddr.s_addr;
2077 
2078 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2079 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2080 
2081 	pcbporthash = &pcbinfo->ipi_porthashbase[
2082 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2083 
2084 	/*
2085 	 * Go through port list and look for a head for this lport.
2086 	 */
2087 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2088 		if (phd->phd_port == inp->inp_lport)
2089 			break;
2090 	}
2091 	/*
2092 	 * If none exists, malloc one and tack it on.
2093 	 */
2094 	if (phd == NULL) {
2095 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2096 		if (phd == NULL) {
2097 			return (ENOBUFS); /* XXX */
2098 		}
2099 		phd->phd_port = inp->inp_lport;
2100 		LIST_INIT(&phd->phd_pcblist);
2101 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2102 	}
2103 	inp->inp_phd = phd;
2104 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2105 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2106 	inp->inp_flags |= INP_INHASHLIST;
2107 #ifdef PCBGROUP
2108 	if (do_pcbgroup_update)
2109 		in_pcbgroup_update(inp);
2110 #endif
2111 	return (0);
2112 }
2113 
2114 /*
2115  * For now, there are two public interfaces to insert an inpcb into the hash
2116  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2117  * is used only in the TCP syncache, where in_pcbinshash is called before the
2118  * full 4-tuple is set for the inpcb, and we don't want to install in the
2119  * pcbgroup until later.
2120  *
2121  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2122  * connection groups, and partially initialised inpcbs should not be exposed
2123  * to either reservation hash tables or pcbgroups.
2124  */
2125 int
2126 in_pcbinshash(struct inpcb *inp)
2127 {
2128 
2129 	return (in_pcbinshash_internal(inp, 1));
2130 }
2131 
2132 int
2133 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2134 {
2135 
2136 	return (in_pcbinshash_internal(inp, 0));
2137 }
2138 
2139 /*
2140  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2141  * changed. NOTE: This does not handle the case of the lport changing (the
2142  * hashed port list would have to be updated as well), so the lport must
2143  * not change after in_pcbinshash() has been called.
2144  */
2145 void
2146 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2147 {
2148 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2149 	struct inpcbhead *head;
2150 	u_int32_t hashkey_faddr;
2151 
2152 	INP_WLOCK_ASSERT(inp);
2153 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2154 
2155 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2156 	    ("in_pcbrehash: !INP_INHASHLIST"));
2157 
2158 #ifdef INET6
2159 	if (inp->inp_vflag & INP_IPV6)
2160 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2161 	else
2162 #endif
2163 	hashkey_faddr = inp->inp_faddr.s_addr;
2164 
2165 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2166 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2167 
2168 	LIST_REMOVE(inp, inp_hash);
2169 	LIST_INSERT_HEAD(head, inp, inp_hash);
2170 
2171 #ifdef PCBGROUP
2172 	if (m != NULL)
2173 		in_pcbgroup_update_mbuf(inp, m);
2174 	else
2175 		in_pcbgroup_update(inp);
2176 #endif
2177 }
2178 
2179 void
2180 in_pcbrehash(struct inpcb *inp)
2181 {
2182 
2183 	in_pcbrehash_mbuf(inp, NULL);
2184 }
2185 
2186 /*
2187  * Remove PCB from various lists.
2188  */
2189 static void
2190 in_pcbremlists(struct inpcb *inp)
2191 {
2192 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2193 
2194 #ifdef INVARIANTS
2195 	if (pcbinfo == &V_tcbinfo) {
2196 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2197 	} else {
2198 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2199 	}
2200 #endif
2201 
2202 	INP_WLOCK_ASSERT(inp);
2203 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2204 
2205 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2206 	if (inp->inp_flags & INP_INHASHLIST) {
2207 		struct inpcbport *phd = inp->inp_phd;
2208 
2209 		INP_HASH_WLOCK(pcbinfo);
2210 		LIST_REMOVE(inp, inp_hash);
2211 		LIST_REMOVE(inp, inp_portlist);
2212 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2213 			LIST_REMOVE(phd, phd_hash);
2214 			free(phd, M_PCB);
2215 		}
2216 		INP_HASH_WUNLOCK(pcbinfo);
2217 		inp->inp_flags &= ~INP_INHASHLIST;
2218 	}
2219 	LIST_REMOVE(inp, inp_list);
2220 	pcbinfo->ipi_count--;
2221 #ifdef PCBGROUP
2222 	in_pcbgroup_remove(inp);
2223 #endif
2224 }
2225 
2226 /*
2227  * A set label operation has occurred at the socket layer, propagate the
2228  * label change into the in_pcb for the socket.
2229  */
2230 void
2231 in_pcbsosetlabel(struct socket *so)
2232 {
2233 #ifdef MAC
2234 	struct inpcb *inp;
2235 
2236 	inp = sotoinpcb(so);
2237 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2238 
2239 	INP_WLOCK(inp);
2240 	SOCK_LOCK(so);
2241 	mac_inpcb_sosetlabel(so, inp);
2242 	SOCK_UNLOCK(so);
2243 	INP_WUNLOCK(inp);
2244 #endif
2245 }
2246 
2247 /*
2248  * ipport_tick runs once per second, determining if random port allocation
2249  * should be continued.  If more than ipport_randomcps ports have been
2250  * allocated in the last second, then we return to sequential port
2251  * allocation. We return to random allocation only once we drop below
2252  * ipport_randomcps for at least ipport_randomtime seconds.
2253  */
2254 static void
2255 ipport_tick(void *xtp)
2256 {
2257 	VNET_ITERATOR_DECL(vnet_iter);
2258 
2259 	VNET_LIST_RLOCK_NOSLEEP();
2260 	VNET_FOREACH(vnet_iter) {
2261 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2262 		if (V_ipport_tcpallocs <=
2263 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2264 			if (V_ipport_stoprandom > 0)
2265 				V_ipport_stoprandom--;
2266 		} else
2267 			V_ipport_stoprandom = V_ipport_randomtime;
2268 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2269 		CURVNET_RESTORE();
2270 	}
2271 	VNET_LIST_RUNLOCK_NOSLEEP();
2272 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2273 }
2274 
2275 static void
2276 ip_fini(void *xtp)
2277 {
2278 
2279 	callout_stop(&ipport_tick_callout);
2280 }
2281 
2282 /*
2283  * The ipport_callout should start running at about the time we attach the
2284  * inet or inet6 domains.
2285  */
2286 static void
2287 ipport_tick_init(const void *unused __unused)
2288 {
2289 
2290 	/* Start ipport_tick. */
2291 	callout_init(&ipport_tick_callout, 1);
2292 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2293 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2294 		SHUTDOWN_PRI_DEFAULT);
2295 }
2296 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2297     ipport_tick_init, NULL);
2298 
2299 void
2300 inp_wlock(struct inpcb *inp)
2301 {
2302 
2303 	INP_WLOCK(inp);
2304 }
2305 
2306 void
2307 inp_wunlock(struct inpcb *inp)
2308 {
2309 
2310 	INP_WUNLOCK(inp);
2311 }
2312 
2313 void
2314 inp_rlock(struct inpcb *inp)
2315 {
2316 
2317 	INP_RLOCK(inp);
2318 }
2319 
2320 void
2321 inp_runlock(struct inpcb *inp)
2322 {
2323 
2324 	INP_RUNLOCK(inp);
2325 }
2326 
2327 #ifdef INVARIANTS
2328 void
2329 inp_lock_assert(struct inpcb *inp)
2330 {
2331 
2332 	INP_WLOCK_ASSERT(inp);
2333 }
2334 
2335 void
2336 inp_unlock_assert(struct inpcb *inp)
2337 {
2338 
2339 	INP_UNLOCK_ASSERT(inp);
2340 }
2341 #endif
2342 
2343 void
2344 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2345 {
2346 	struct inpcb *inp;
2347 
2348 	INP_INFO_WLOCK(&V_tcbinfo);
2349 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2350 		INP_WLOCK(inp);
2351 		func(inp, arg);
2352 		INP_WUNLOCK(inp);
2353 	}
2354 	INP_INFO_WUNLOCK(&V_tcbinfo);
2355 }
2356 
2357 struct socket *
2358 inp_inpcbtosocket(struct inpcb *inp)
2359 {
2360 
2361 	INP_WLOCK_ASSERT(inp);
2362 	return (inp->inp_socket);
2363 }
2364 
2365 struct tcpcb *
2366 inp_inpcbtotcpcb(struct inpcb *inp)
2367 {
2368 
2369 	INP_WLOCK_ASSERT(inp);
2370 	return ((struct tcpcb *)inp->inp_ppcb);
2371 }
2372 
2373 int
2374 inp_ip_tos_get(const struct inpcb *inp)
2375 {
2376 
2377 	return (inp->inp_ip_tos);
2378 }
2379 
2380 void
2381 inp_ip_tos_set(struct inpcb *inp, int val)
2382 {
2383 
2384 	inp->inp_ip_tos = val;
2385 }
2386 
2387 void
2388 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2389     uint32_t *faddr, uint16_t *fp)
2390 {
2391 
2392 	INP_LOCK_ASSERT(inp);
2393 	*laddr = inp->inp_laddr.s_addr;
2394 	*faddr = inp->inp_faddr.s_addr;
2395 	*lp = inp->inp_lport;
2396 	*fp = inp->inp_fport;
2397 }
2398 
2399 struct inpcb *
2400 so_sotoinpcb(struct socket *so)
2401 {
2402 
2403 	return (sotoinpcb(so));
2404 }
2405 
2406 struct tcpcb *
2407 so_sototcpcb(struct socket *so)
2408 {
2409 
2410 	return (sototcpcb(so));
2411 }
2412 
2413 #ifdef DDB
2414 static void
2415 db_print_indent(int indent)
2416 {
2417 	int i;
2418 
2419 	for (i = 0; i < indent; i++)
2420 		db_printf(" ");
2421 }
2422 
2423 static void
2424 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2425 {
2426 	char faddr_str[48], laddr_str[48];
2427 
2428 	db_print_indent(indent);
2429 	db_printf("%s at %p\n", name, inc);
2430 
2431 	indent += 2;
2432 
2433 #ifdef INET6
2434 	if (inc->inc_flags & INC_ISIPV6) {
2435 		/* IPv6. */
2436 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2437 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2438 	} else
2439 #endif
2440 	{
2441 		/* IPv4. */
2442 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2443 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2444 	}
2445 	db_print_indent(indent);
2446 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2447 	    ntohs(inc->inc_lport));
2448 	db_print_indent(indent);
2449 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2450 	    ntohs(inc->inc_fport));
2451 }
2452 
2453 static void
2454 db_print_inpflags(int inp_flags)
2455 {
2456 	int comma;
2457 
2458 	comma = 0;
2459 	if (inp_flags & INP_RECVOPTS) {
2460 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2461 		comma = 1;
2462 	}
2463 	if (inp_flags & INP_RECVRETOPTS) {
2464 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2465 		comma = 1;
2466 	}
2467 	if (inp_flags & INP_RECVDSTADDR) {
2468 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2469 		comma = 1;
2470 	}
2471 	if (inp_flags & INP_HDRINCL) {
2472 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2473 		comma = 1;
2474 	}
2475 	if (inp_flags & INP_HIGHPORT) {
2476 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2477 		comma = 1;
2478 	}
2479 	if (inp_flags & INP_LOWPORT) {
2480 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2481 		comma = 1;
2482 	}
2483 	if (inp_flags & INP_ANONPORT) {
2484 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2485 		comma = 1;
2486 	}
2487 	if (inp_flags & INP_RECVIF) {
2488 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2489 		comma = 1;
2490 	}
2491 	if (inp_flags & INP_MTUDISC) {
2492 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2493 		comma = 1;
2494 	}
2495 	if (inp_flags & INP_RECVTTL) {
2496 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2497 		comma = 1;
2498 	}
2499 	if (inp_flags & INP_DONTFRAG) {
2500 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2501 		comma = 1;
2502 	}
2503 	if (inp_flags & INP_RECVTOS) {
2504 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2505 		comma = 1;
2506 	}
2507 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2508 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2509 		comma = 1;
2510 	}
2511 	if (inp_flags & IN6P_PKTINFO) {
2512 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2513 		comma = 1;
2514 	}
2515 	if (inp_flags & IN6P_HOPLIMIT) {
2516 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2517 		comma = 1;
2518 	}
2519 	if (inp_flags & IN6P_HOPOPTS) {
2520 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2521 		comma = 1;
2522 	}
2523 	if (inp_flags & IN6P_DSTOPTS) {
2524 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2525 		comma = 1;
2526 	}
2527 	if (inp_flags & IN6P_RTHDR) {
2528 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2529 		comma = 1;
2530 	}
2531 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2532 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2533 		comma = 1;
2534 	}
2535 	if (inp_flags & IN6P_TCLASS) {
2536 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2537 		comma = 1;
2538 	}
2539 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2540 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2541 		comma = 1;
2542 	}
2543 	if (inp_flags & INP_TIMEWAIT) {
2544 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2545 		comma  = 1;
2546 	}
2547 	if (inp_flags & INP_ONESBCAST) {
2548 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2549 		comma  = 1;
2550 	}
2551 	if (inp_flags & INP_DROPPED) {
2552 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2553 		comma  = 1;
2554 	}
2555 	if (inp_flags & INP_SOCKREF) {
2556 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2557 		comma  = 1;
2558 	}
2559 	if (inp_flags & IN6P_RFC2292) {
2560 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2561 		comma = 1;
2562 	}
2563 	if (inp_flags & IN6P_MTU) {
2564 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2565 		comma = 1;
2566 	}
2567 }
2568 
2569 static void
2570 db_print_inpvflag(u_char inp_vflag)
2571 {
2572 	int comma;
2573 
2574 	comma = 0;
2575 	if (inp_vflag & INP_IPV4) {
2576 		db_printf("%sINP_IPV4", comma ? ", " : "");
2577 		comma  = 1;
2578 	}
2579 	if (inp_vflag & INP_IPV6) {
2580 		db_printf("%sINP_IPV6", comma ? ", " : "");
2581 		comma  = 1;
2582 	}
2583 	if (inp_vflag & INP_IPV6PROTO) {
2584 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2585 		comma  = 1;
2586 	}
2587 }
2588 
2589 static void
2590 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2591 {
2592 
2593 	db_print_indent(indent);
2594 	db_printf("%s at %p\n", name, inp);
2595 
2596 	indent += 2;
2597 
2598 	db_print_indent(indent);
2599 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2600 
2601 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2602 
2603 	db_print_indent(indent);
2604 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2605 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2606 
2607 	db_print_indent(indent);
2608 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2609 	   inp->inp_label, inp->inp_flags);
2610 	db_print_inpflags(inp->inp_flags);
2611 	db_printf(")\n");
2612 
2613 	db_print_indent(indent);
2614 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2615 	    inp->inp_vflag);
2616 	db_print_inpvflag(inp->inp_vflag);
2617 	db_printf(")\n");
2618 
2619 	db_print_indent(indent);
2620 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2621 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2622 
2623 	db_print_indent(indent);
2624 #ifdef INET6
2625 	if (inp->inp_vflag & INP_IPV6) {
2626 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2627 		    "in6p_moptions: %p\n", inp->in6p_options,
2628 		    inp->in6p_outputopts, inp->in6p_moptions);
2629 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2630 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2631 		    inp->in6p_hops);
2632 	} else
2633 #endif
2634 	{
2635 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2636 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2637 		    inp->inp_options, inp->inp_moptions);
2638 	}
2639 
2640 	db_print_indent(indent);
2641 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2642 	    (uintmax_t)inp->inp_gencnt);
2643 }
2644 
2645 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2646 {
2647 	struct inpcb *inp;
2648 
2649 	if (!have_addr) {
2650 		db_printf("usage: show inpcb <addr>\n");
2651 		return;
2652 	}
2653 	inp = (struct inpcb *)addr;
2654 
2655 	db_print_inpcb(inp, "inpcb", 0);
2656 }
2657 #endif /* DDB */
2658