xref: /freebsd/sys/netinet/in_pcb.c (revision a273973175e2a4cbbda3fd7b8cddd0bec0857082)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993, 1995
5  *	The Regents of the University of California.
6  * Copyright (c) 2007-2009 Robert N. M. Watson
7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
8  * All rights reserved.
9  *
10  * Portions of this software were developed by Robert N. M. Watson under
11  * contract to Juniper Networks, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_ddb.h"
44 #include "opt_ipsec.h"
45 #include "opt_inet.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
49 #include "opt_rss.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/sockio.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/refcount.h>
67 #include <sys/jail.h>
68 #include <sys/kernel.h>
69 #include <sys/sysctl.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif
74 
75 #include <vm/uma.h>
76 
77 #include <net/if.h>
78 #include <net/if_var.h>
79 #include <net/if_types.h>
80 #include <net/if_llatbl.h>
81 #include <net/route.h>
82 #include <net/rss_config.h>
83 #include <net/vnet.h>
84 
85 #if defined(INET) || defined(INET6)
86 #include <netinet/in.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/ip_var.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/udp.h>
91 #include <netinet/udp_var.h>
92 #endif
93 #ifdef INET
94 #include <netinet/in_var.h>
95 #endif
96 #ifdef INET6
97 #include <netinet/ip6.h>
98 #include <netinet6/in6_pcb.h>
99 #include <netinet6/in6_var.h>
100 #include <netinet6/ip6_var.h>
101 #endif /* INET6 */
102 
103 #include <netipsec/ipsec_support.h>
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  * Different protocols initialize their inpcbs differently - giving
218  * different name to the lock.  But they all are disposed the same.
219  */
220 static void
221 inpcb_fini(void *mem, int size)
222 {
223 	struct inpcb *inp = mem;
224 
225 	INP_LOCK_DESTROY(inp);
226 }
227 
228 /*
229  * Initialize an inpcbinfo -- we should be able to reduce the number of
230  * arguments in time.
231  */
232 void
233 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
234     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
235     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
236 {
237 
238 	INP_INFO_LOCK_INIT(pcbinfo, name);
239 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
240 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
241 #ifdef VIMAGE
242 	pcbinfo->ipi_vnet = curvnet;
243 #endif
244 	pcbinfo->ipi_listhead = listhead;
245 	LIST_INIT(pcbinfo->ipi_listhead);
246 	pcbinfo->ipi_count = 0;
247 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
248 	    &pcbinfo->ipi_hashmask);
249 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
250 	    &pcbinfo->ipi_porthashmask);
251 #ifdef PCBGROUP
252 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
253 #endif
254 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
255 	    NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
256 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
257 	uma_zone_set_warning(pcbinfo->ipi_zone,
258 	    "kern.ipc.maxsockets limit reached");
259 }
260 
261 /*
262  * Destroy an inpcbinfo.
263  */
264 void
265 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
266 {
267 
268 	KASSERT(pcbinfo->ipi_count == 0,
269 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
270 
271 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
272 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
273 	    pcbinfo->ipi_porthashmask);
274 #ifdef PCBGROUP
275 	in_pcbgroup_destroy(pcbinfo);
276 #endif
277 	uma_zdestroy(pcbinfo->ipi_zone);
278 	INP_LIST_LOCK_DESTROY(pcbinfo);
279 	INP_HASH_LOCK_DESTROY(pcbinfo);
280 	INP_INFO_LOCK_DESTROY(pcbinfo);
281 }
282 
283 /*
284  * Allocate a PCB and associate it with the socket.
285  * On success return with the PCB locked.
286  */
287 int
288 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
289 {
290 	struct inpcb *inp;
291 	int error;
292 
293 #ifdef INVARIANTS
294 	if (pcbinfo == &V_tcbinfo) {
295 		INP_INFO_RLOCK_ASSERT(pcbinfo);
296 	} else {
297 		INP_INFO_WLOCK_ASSERT(pcbinfo);
298 	}
299 #endif
300 
301 	error = 0;
302 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
303 	if (inp == NULL)
304 		return (ENOBUFS);
305 	bzero(&inp->inp_start_zero, inp_zero_size);
306 	inp->inp_pcbinfo = pcbinfo;
307 	inp->inp_socket = so;
308 	inp->inp_cred = crhold(so->so_cred);
309 	inp->inp_inc.inc_fibnum = so->so_fibnum;
310 #ifdef MAC
311 	error = mac_inpcb_init(inp, M_NOWAIT);
312 	if (error != 0)
313 		goto out;
314 	mac_inpcb_create(so, inp);
315 #endif
316 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
317 	error = ipsec_init_pcbpolicy(inp);
318 	if (error != 0) {
319 #ifdef MAC
320 		mac_inpcb_destroy(inp);
321 #endif
322 		goto out;
323 	}
324 #endif /*IPSEC*/
325 #ifdef INET6
326 	if (INP_SOCKAF(so) == AF_INET6) {
327 		inp->inp_vflag |= INP_IPV6PROTO;
328 		if (V_ip6_v6only)
329 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
330 	}
331 #endif
332 	INP_WLOCK(inp);
333 	INP_LIST_WLOCK(pcbinfo);
334 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
335 	pcbinfo->ipi_count++;
336 	so->so_pcb = (caddr_t)inp;
337 #ifdef INET6
338 	if (V_ip6_auto_flowlabel)
339 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
340 #endif
341 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
342 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
343 
344 	/*
345 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
346 	 * to be cleaned up.
347 	 */
348 	inp->inp_route.ro_flags = RT_LLE_CACHE;
349 	INP_LIST_WUNLOCK(pcbinfo);
350 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
351 out:
352 	if (error != 0) {
353 		crfree(inp->inp_cred);
354 		uma_zfree(pcbinfo->ipi_zone, inp);
355 	}
356 #endif
357 	return (error);
358 }
359 
360 #ifdef INET
361 int
362 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
363 {
364 	int anonport, error;
365 
366 	INP_WLOCK_ASSERT(inp);
367 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
368 
369 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
370 		return (EINVAL);
371 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
372 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
373 	    &inp->inp_lport, cred);
374 	if (error)
375 		return (error);
376 	if (in_pcbinshash(inp) != 0) {
377 		inp->inp_laddr.s_addr = INADDR_ANY;
378 		inp->inp_lport = 0;
379 		return (EAGAIN);
380 	}
381 	if (anonport)
382 		inp->inp_flags |= INP_ANONPORT;
383 	return (0);
384 }
385 #endif
386 
387 /*
388  * Select a local port (number) to use.
389  */
390 #if defined(INET) || defined(INET6)
391 int
392 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
393     struct ucred *cred, int lookupflags)
394 {
395 	struct inpcbinfo *pcbinfo;
396 	struct inpcb *tmpinp;
397 	unsigned short *lastport;
398 	int count, dorandom, error;
399 	u_short aux, first, last, lport;
400 #ifdef INET
401 	struct in_addr laddr;
402 #endif
403 
404 	pcbinfo = inp->inp_pcbinfo;
405 
406 	/*
407 	 * Because no actual state changes occur here, a global write lock on
408 	 * the pcbinfo isn't required.
409 	 */
410 	INP_LOCK_ASSERT(inp);
411 	INP_HASH_LOCK_ASSERT(pcbinfo);
412 
413 	if (inp->inp_flags & INP_HIGHPORT) {
414 		first = V_ipport_hifirstauto;	/* sysctl */
415 		last  = V_ipport_hilastauto;
416 		lastport = &pcbinfo->ipi_lasthi;
417 	} else if (inp->inp_flags & INP_LOWPORT) {
418 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
419 		if (error)
420 			return (error);
421 		first = V_ipport_lowfirstauto;	/* 1023 */
422 		last  = V_ipport_lowlastauto;	/* 600 */
423 		lastport = &pcbinfo->ipi_lastlow;
424 	} else {
425 		first = V_ipport_firstauto;	/* sysctl */
426 		last  = V_ipport_lastauto;
427 		lastport = &pcbinfo->ipi_lastport;
428 	}
429 	/*
430 	 * For UDP(-Lite), use random port allocation as long as the user
431 	 * allows it.  For TCP (and as of yet unknown) connections,
432 	 * use random port allocation only if the user allows it AND
433 	 * ipport_tick() allows it.
434 	 */
435 	if (V_ipport_randomized &&
436 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
437 		pcbinfo == &V_ulitecbinfo))
438 		dorandom = 1;
439 	else
440 		dorandom = 0;
441 	/*
442 	 * It makes no sense to do random port allocation if
443 	 * we have the only port available.
444 	 */
445 	if (first == last)
446 		dorandom = 0;
447 	/* Make sure to not include UDP(-Lite) packets in the count. */
448 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
449 		V_ipport_tcpallocs++;
450 	/*
451 	 * Instead of having two loops further down counting up or down
452 	 * make sure that first is always <= last and go with only one
453 	 * code path implementing all logic.
454 	 */
455 	if (first > last) {
456 		aux = first;
457 		first = last;
458 		last = aux;
459 	}
460 
461 #ifdef INET
462 	/* Make the compiler happy. */
463 	laddr.s_addr = 0;
464 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
465 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
466 		    __func__, inp));
467 		laddr = *laddrp;
468 	}
469 #endif
470 	tmpinp = NULL;	/* Make compiler happy. */
471 	lport = *lportp;
472 
473 	if (dorandom)
474 		*lastport = first + (arc4random() % (last - first));
475 
476 	count = last - first;
477 
478 	do {
479 		if (count-- < 0)	/* completely used? */
480 			return (EADDRNOTAVAIL);
481 		++*lastport;
482 		if (*lastport < first || *lastport > last)
483 			*lastport = first;
484 		lport = htons(*lastport);
485 
486 #ifdef INET6
487 		if ((inp->inp_vflag & INP_IPV6) != 0)
488 			tmpinp = in6_pcblookup_local(pcbinfo,
489 			    &inp->in6p_laddr, lport, lookupflags, cred);
490 #endif
491 #if defined(INET) && defined(INET6)
492 		else
493 #endif
494 #ifdef INET
495 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
496 			    lport, lookupflags, cred);
497 #endif
498 	} while (tmpinp != NULL);
499 
500 #ifdef INET
501 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
502 		laddrp->s_addr = laddr.s_addr;
503 #endif
504 	*lportp = lport;
505 
506 	return (0);
507 }
508 
509 /*
510  * Return cached socket options.
511  */
512 short
513 inp_so_options(const struct inpcb *inp)
514 {
515    short so_options;
516 
517    so_options = 0;
518 
519    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
520 	   so_options |= SO_REUSEPORT;
521    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
522 	   so_options |= SO_REUSEADDR;
523    return (so_options);
524 }
525 #endif /* INET || INET6 */
526 
527 /*
528  * Check if a new BINDMULTI socket is allowed to be created.
529  *
530  * ni points to the new inp.
531  * oi points to the exisitng inp.
532  *
533  * This checks whether the existing inp also has BINDMULTI and
534  * whether the credentials match.
535  */
536 int
537 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
538 {
539 	/* Check permissions match */
540 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
541 	    (ni->inp_cred->cr_uid !=
542 	    oi->inp_cred->cr_uid))
543 		return (0);
544 
545 	/* Check the existing inp has BINDMULTI set */
546 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
547 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
548 		return (0);
549 
550 	/*
551 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
552 	 * it is and it matches the checks.
553 	 */
554 	return (1);
555 }
556 
557 #ifdef INET
558 /*
559  * Set up a bind operation on a PCB, performing port allocation
560  * as required, but do not actually modify the PCB. Callers can
561  * either complete the bind by setting inp_laddr/inp_lport and
562  * calling in_pcbinshash(), or they can just use the resulting
563  * port and address to authorise the sending of a once-off packet.
564  *
565  * On error, the values of *laddrp and *lportp are not changed.
566  */
567 int
568 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
569     u_short *lportp, struct ucred *cred)
570 {
571 	struct socket *so = inp->inp_socket;
572 	struct sockaddr_in *sin;
573 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
574 	struct in_addr laddr;
575 	u_short lport = 0;
576 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
577 	int error;
578 
579 	/*
580 	 * No state changes, so read locks are sufficient here.
581 	 */
582 	INP_LOCK_ASSERT(inp);
583 	INP_HASH_LOCK_ASSERT(pcbinfo);
584 
585 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
586 		return (EADDRNOTAVAIL);
587 	laddr.s_addr = *laddrp;
588 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
589 		return (EINVAL);
590 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
591 		lookupflags = INPLOOKUP_WILDCARD;
592 	if (nam == NULL) {
593 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
594 			return (error);
595 	} else {
596 		sin = (struct sockaddr_in *)nam;
597 		if (nam->sa_len != sizeof (*sin))
598 			return (EINVAL);
599 #ifdef notdef
600 		/*
601 		 * We should check the family, but old programs
602 		 * incorrectly fail to initialize it.
603 		 */
604 		if (sin->sin_family != AF_INET)
605 			return (EAFNOSUPPORT);
606 #endif
607 		error = prison_local_ip4(cred, &sin->sin_addr);
608 		if (error)
609 			return (error);
610 		if (sin->sin_port != *lportp) {
611 			/* Don't allow the port to change. */
612 			if (*lportp != 0)
613 				return (EINVAL);
614 			lport = sin->sin_port;
615 		}
616 		/* NB: lport is left as 0 if the port isn't being changed. */
617 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
618 			/*
619 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
620 			 * allow complete duplication of binding if
621 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
622 			 * and a multicast address is bound on both
623 			 * new and duplicated sockets.
624 			 */
625 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
626 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
627 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
628 			sin->sin_port = 0;		/* yech... */
629 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
630 			/*
631 			 * Is the address a local IP address?
632 			 * If INP_BINDANY is set, then the socket may be bound
633 			 * to any endpoint address, local or not.
634 			 */
635 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
636 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
637 				return (EADDRNOTAVAIL);
638 		}
639 		laddr = sin->sin_addr;
640 		if (lport) {
641 			struct inpcb *t;
642 			struct tcptw *tw;
643 
644 			/* GROSS */
645 			if (ntohs(lport) <= V_ipport_reservedhigh &&
646 			    ntohs(lport) >= V_ipport_reservedlow &&
647 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
648 			    0))
649 				return (EACCES);
650 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
651 			    priv_check_cred(inp->inp_cred,
652 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
653 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
654 				    lport, INPLOOKUP_WILDCARD, cred);
655 	/*
656 	 * XXX
657 	 * This entire block sorely needs a rewrite.
658 	 */
659 				if (t &&
660 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
661 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
662 				    (so->so_type != SOCK_STREAM ||
663 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
664 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
665 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
666 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
667 				    (inp->inp_cred->cr_uid !=
668 				     t->inp_cred->cr_uid))
669 					return (EADDRINUSE);
670 
671 				/*
672 				 * If the socket is a BINDMULTI socket, then
673 				 * the credentials need to match and the
674 				 * original socket also has to have been bound
675 				 * with BINDMULTI.
676 				 */
677 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
678 					return (EADDRINUSE);
679 			}
680 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
681 			    lport, lookupflags, cred);
682 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
683 				/*
684 				 * XXXRW: If an incpb has had its timewait
685 				 * state recycled, we treat the address as
686 				 * being in use (for now).  This is better
687 				 * than a panic, but not desirable.
688 				 */
689 				tw = intotw(t);
690 				if (tw == NULL ||
691 				    (reuseport & tw->tw_so_options) == 0)
692 					return (EADDRINUSE);
693 			} else if (t &&
694 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
695 			    (reuseport & inp_so_options(t)) == 0) {
696 #ifdef INET6
697 				if (ntohl(sin->sin_addr.s_addr) !=
698 				    INADDR_ANY ||
699 				    ntohl(t->inp_laddr.s_addr) !=
700 				    INADDR_ANY ||
701 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
702 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
703 #endif
704 				return (EADDRINUSE);
705 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
706 					return (EADDRINUSE);
707 			}
708 		}
709 	}
710 	if (*lportp != 0)
711 		lport = *lportp;
712 	if (lport == 0) {
713 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
714 		if (error != 0)
715 			return (error);
716 
717 	}
718 	*laddrp = laddr.s_addr;
719 	*lportp = lport;
720 	return (0);
721 }
722 
723 /*
724  * Connect from a socket to a specified address.
725  * Both address and port must be specified in argument sin.
726  * If don't have a local address for this socket yet,
727  * then pick one.
728  */
729 int
730 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
731     struct ucred *cred, struct mbuf *m)
732 {
733 	u_short lport, fport;
734 	in_addr_t laddr, faddr;
735 	int anonport, error;
736 
737 	INP_WLOCK_ASSERT(inp);
738 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
739 
740 	lport = inp->inp_lport;
741 	laddr = inp->inp_laddr.s_addr;
742 	anonport = (lport == 0);
743 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
744 	    NULL, cred);
745 	if (error)
746 		return (error);
747 
748 	/* Do the initial binding of the local address if required. */
749 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
750 		inp->inp_lport = lport;
751 		inp->inp_laddr.s_addr = laddr;
752 		if (in_pcbinshash(inp) != 0) {
753 			inp->inp_laddr.s_addr = INADDR_ANY;
754 			inp->inp_lport = 0;
755 			return (EAGAIN);
756 		}
757 	}
758 
759 	/* Commit the remaining changes. */
760 	inp->inp_lport = lport;
761 	inp->inp_laddr.s_addr = laddr;
762 	inp->inp_faddr.s_addr = faddr;
763 	inp->inp_fport = fport;
764 	in_pcbrehash_mbuf(inp, m);
765 
766 	if (anonport)
767 		inp->inp_flags |= INP_ANONPORT;
768 	return (0);
769 }
770 
771 int
772 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
773 {
774 
775 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
776 }
777 
778 /*
779  * Do proper source address selection on an unbound socket in case
780  * of connect. Take jails into account as well.
781  */
782 int
783 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
784     struct ucred *cred)
785 {
786 	struct ifaddr *ifa;
787 	struct sockaddr *sa;
788 	struct sockaddr_in *sin;
789 	struct route sro;
790 	int error;
791 
792 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
793 
794 	/*
795 	 * Bypass source address selection and use the primary jail IP
796 	 * if requested.
797 	 */
798 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
799 		return (0);
800 
801 	error = 0;
802 	bzero(&sro, sizeof(sro));
803 
804 	sin = (struct sockaddr_in *)&sro.ro_dst;
805 	sin->sin_family = AF_INET;
806 	sin->sin_len = sizeof(struct sockaddr_in);
807 	sin->sin_addr.s_addr = faddr->s_addr;
808 
809 	/*
810 	 * If route is known our src addr is taken from the i/f,
811 	 * else punt.
812 	 *
813 	 * Find out route to destination.
814 	 */
815 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
816 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
817 
818 	/*
819 	 * If we found a route, use the address corresponding to
820 	 * the outgoing interface.
821 	 *
822 	 * Otherwise assume faddr is reachable on a directly connected
823 	 * network and try to find a corresponding interface to take
824 	 * the source address from.
825 	 */
826 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
827 		struct in_ifaddr *ia;
828 		struct ifnet *ifp;
829 
830 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
831 					inp->inp_socket->so_fibnum));
832 		if (ia == NULL)
833 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
834 						inp->inp_socket->so_fibnum));
835 		if (ia == NULL) {
836 			error = ENETUNREACH;
837 			goto done;
838 		}
839 
840 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
841 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
842 			ifa_free(&ia->ia_ifa);
843 			goto done;
844 		}
845 
846 		ifp = ia->ia_ifp;
847 		ifa_free(&ia->ia_ifa);
848 		ia = NULL;
849 		IF_ADDR_RLOCK(ifp);
850 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
851 
852 			sa = ifa->ifa_addr;
853 			if (sa->sa_family != AF_INET)
854 				continue;
855 			sin = (struct sockaddr_in *)sa;
856 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
857 				ia = (struct in_ifaddr *)ifa;
858 				break;
859 			}
860 		}
861 		if (ia != NULL) {
862 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
863 			IF_ADDR_RUNLOCK(ifp);
864 			goto done;
865 		}
866 		IF_ADDR_RUNLOCK(ifp);
867 
868 		/* 3. As a last resort return the 'default' jail address. */
869 		error = prison_get_ip4(cred, laddr);
870 		goto done;
871 	}
872 
873 	/*
874 	 * If the outgoing interface on the route found is not
875 	 * a loopback interface, use the address from that interface.
876 	 * In case of jails do those three steps:
877 	 * 1. check if the interface address belongs to the jail. If so use it.
878 	 * 2. check if we have any address on the outgoing interface
879 	 *    belonging to this jail. If so use it.
880 	 * 3. as a last resort return the 'default' jail address.
881 	 */
882 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
883 		struct in_ifaddr *ia;
884 		struct ifnet *ifp;
885 
886 		/* If not jailed, use the default returned. */
887 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
888 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
889 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
890 			goto done;
891 		}
892 
893 		/* Jailed. */
894 		/* 1. Check if the iface address belongs to the jail. */
895 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
896 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
897 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
898 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
899 			goto done;
900 		}
901 
902 		/*
903 		 * 2. Check if we have any address on the outgoing interface
904 		 *    belonging to this jail.
905 		 */
906 		ia = NULL;
907 		ifp = sro.ro_rt->rt_ifp;
908 		IF_ADDR_RLOCK(ifp);
909 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
910 			sa = ifa->ifa_addr;
911 			if (sa->sa_family != AF_INET)
912 				continue;
913 			sin = (struct sockaddr_in *)sa;
914 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
915 				ia = (struct in_ifaddr *)ifa;
916 				break;
917 			}
918 		}
919 		if (ia != NULL) {
920 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
921 			IF_ADDR_RUNLOCK(ifp);
922 			goto done;
923 		}
924 		IF_ADDR_RUNLOCK(ifp);
925 
926 		/* 3. As a last resort return the 'default' jail address. */
927 		error = prison_get_ip4(cred, laddr);
928 		goto done;
929 	}
930 
931 	/*
932 	 * The outgoing interface is marked with 'loopback net', so a route
933 	 * to ourselves is here.
934 	 * Try to find the interface of the destination address and then
935 	 * take the address from there. That interface is not necessarily
936 	 * a loopback interface.
937 	 * In case of jails, check that it is an address of the jail
938 	 * and if we cannot find, fall back to the 'default' jail address.
939 	 */
940 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
941 		struct sockaddr_in sain;
942 		struct in_ifaddr *ia;
943 
944 		bzero(&sain, sizeof(struct sockaddr_in));
945 		sain.sin_family = AF_INET;
946 		sain.sin_len = sizeof(struct sockaddr_in);
947 		sain.sin_addr.s_addr = faddr->s_addr;
948 
949 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
950 					inp->inp_socket->so_fibnum));
951 		if (ia == NULL)
952 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
953 						inp->inp_socket->so_fibnum));
954 		if (ia == NULL)
955 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
956 
957 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
958 			if (ia == NULL) {
959 				error = ENETUNREACH;
960 				goto done;
961 			}
962 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
963 			ifa_free(&ia->ia_ifa);
964 			goto done;
965 		}
966 
967 		/* Jailed. */
968 		if (ia != NULL) {
969 			struct ifnet *ifp;
970 
971 			ifp = ia->ia_ifp;
972 			ifa_free(&ia->ia_ifa);
973 			ia = NULL;
974 			IF_ADDR_RLOCK(ifp);
975 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
976 
977 				sa = ifa->ifa_addr;
978 				if (sa->sa_family != AF_INET)
979 					continue;
980 				sin = (struct sockaddr_in *)sa;
981 				if (prison_check_ip4(cred,
982 				    &sin->sin_addr) == 0) {
983 					ia = (struct in_ifaddr *)ifa;
984 					break;
985 				}
986 			}
987 			if (ia != NULL) {
988 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
989 				IF_ADDR_RUNLOCK(ifp);
990 				goto done;
991 			}
992 			IF_ADDR_RUNLOCK(ifp);
993 		}
994 
995 		/* 3. As a last resort return the 'default' jail address. */
996 		error = prison_get_ip4(cred, laddr);
997 		goto done;
998 	}
999 
1000 done:
1001 	if (sro.ro_rt != NULL)
1002 		RTFREE(sro.ro_rt);
1003 	return (error);
1004 }
1005 
1006 /*
1007  * Set up for a connect from a socket to the specified address.
1008  * On entry, *laddrp and *lportp should contain the current local
1009  * address and port for the PCB; these are updated to the values
1010  * that should be placed in inp_laddr and inp_lport to complete
1011  * the connect.
1012  *
1013  * On success, *faddrp and *fportp will be set to the remote address
1014  * and port. These are not updated in the error case.
1015  *
1016  * If the operation fails because the connection already exists,
1017  * *oinpp will be set to the PCB of that connection so that the
1018  * caller can decide to override it. In all other cases, *oinpp
1019  * is set to NULL.
1020  */
1021 int
1022 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1023     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1024     struct inpcb **oinpp, struct ucred *cred)
1025 {
1026 	struct rm_priotracker in_ifa_tracker;
1027 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1028 	struct in_ifaddr *ia;
1029 	struct inpcb *oinp;
1030 	struct in_addr laddr, faddr;
1031 	u_short lport, fport;
1032 	int error;
1033 
1034 	/*
1035 	 * Because a global state change doesn't actually occur here, a read
1036 	 * lock is sufficient.
1037 	 */
1038 	INP_LOCK_ASSERT(inp);
1039 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1040 
1041 	if (oinpp != NULL)
1042 		*oinpp = NULL;
1043 	if (nam->sa_len != sizeof (*sin))
1044 		return (EINVAL);
1045 	if (sin->sin_family != AF_INET)
1046 		return (EAFNOSUPPORT);
1047 	if (sin->sin_port == 0)
1048 		return (EADDRNOTAVAIL);
1049 	laddr.s_addr = *laddrp;
1050 	lport = *lportp;
1051 	faddr = sin->sin_addr;
1052 	fport = sin->sin_port;
1053 
1054 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1055 		/*
1056 		 * If the destination address is INADDR_ANY,
1057 		 * use the primary local address.
1058 		 * If the supplied address is INADDR_BROADCAST,
1059 		 * and the primary interface supports broadcast,
1060 		 * choose the broadcast address for that interface.
1061 		 */
1062 		if (faddr.s_addr == INADDR_ANY) {
1063 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1064 			faddr =
1065 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1066 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1067 			if (cred != NULL &&
1068 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1069 				return (error);
1070 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1071 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1072 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1073 			    IFF_BROADCAST)
1074 				faddr = satosin(&TAILQ_FIRST(
1075 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1076 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1077 		}
1078 	}
1079 	if (laddr.s_addr == INADDR_ANY) {
1080 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1081 		/*
1082 		 * If the destination address is multicast and an outgoing
1083 		 * interface has been set as a multicast option, prefer the
1084 		 * address of that interface as our source address.
1085 		 */
1086 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1087 		    inp->inp_moptions != NULL) {
1088 			struct ip_moptions *imo;
1089 			struct ifnet *ifp;
1090 
1091 			imo = inp->inp_moptions;
1092 			if (imo->imo_multicast_ifp != NULL) {
1093 				ifp = imo->imo_multicast_ifp;
1094 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1095 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1096 					if ((ia->ia_ifp == ifp) &&
1097 					    (cred == NULL ||
1098 					    prison_check_ip4(cred,
1099 					    &ia->ia_addr.sin_addr) == 0))
1100 						break;
1101 				}
1102 				if (ia == NULL)
1103 					error = EADDRNOTAVAIL;
1104 				else {
1105 					laddr = ia->ia_addr.sin_addr;
1106 					error = 0;
1107 				}
1108 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1109 			}
1110 		}
1111 		if (error)
1112 			return (error);
1113 	}
1114 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1115 	    laddr, lport, 0, NULL);
1116 	if (oinp != NULL) {
1117 		if (oinpp != NULL)
1118 			*oinpp = oinp;
1119 		return (EADDRINUSE);
1120 	}
1121 	if (lport == 0) {
1122 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1123 		    cred);
1124 		if (error)
1125 			return (error);
1126 	}
1127 	*laddrp = laddr.s_addr;
1128 	*lportp = lport;
1129 	*faddrp = faddr.s_addr;
1130 	*fportp = fport;
1131 	return (0);
1132 }
1133 
1134 void
1135 in_pcbdisconnect(struct inpcb *inp)
1136 {
1137 
1138 	INP_WLOCK_ASSERT(inp);
1139 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1140 
1141 	inp->inp_faddr.s_addr = INADDR_ANY;
1142 	inp->inp_fport = 0;
1143 	in_pcbrehash(inp);
1144 }
1145 #endif /* INET */
1146 
1147 /*
1148  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1149  * For most protocols, this will be invoked immediately prior to calling
1150  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1151  * socket, in which case in_pcbfree() is deferred.
1152  */
1153 void
1154 in_pcbdetach(struct inpcb *inp)
1155 {
1156 
1157 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1158 
1159 #ifdef RATELIMIT
1160 	if (inp->inp_snd_tag != NULL)
1161 		in_pcbdetach_txrtlmt(inp);
1162 #endif
1163 	inp->inp_socket->so_pcb = NULL;
1164 	inp->inp_socket = NULL;
1165 }
1166 
1167 /*
1168  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1169  * stability of an inpcb pointer despite the inpcb lock being released.  This
1170  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1171  * but where the inpcb lock may already held, or when acquiring a reference
1172  * via a pcbgroup.
1173  *
1174  * in_pcbref() should be used only to provide brief memory stability, and
1175  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1176  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1177  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1178  * lock and rele are the *only* safe operations that may be performed on the
1179  * inpcb.
1180  *
1181  * While the inpcb will not be freed, releasing the inpcb lock means that the
1182  * connection's state may change, so the caller should be careful to
1183  * revalidate any cached state on reacquiring the lock.  Drop the reference
1184  * using in_pcbrele().
1185  */
1186 void
1187 in_pcbref(struct inpcb *inp)
1188 {
1189 
1190 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1191 
1192 	refcount_acquire(&inp->inp_refcount);
1193 }
1194 
1195 /*
1196  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1197  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1198  * return a flag indicating whether or not the inpcb remains valid.  If it is
1199  * valid, we return with the inpcb lock held.
1200  *
1201  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1202  * reference on an inpcb.  Historically more work was done here (actually, in
1203  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1204  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1205  * about memory stability (and continued use of the write lock).
1206  */
1207 int
1208 in_pcbrele_rlocked(struct inpcb *inp)
1209 {
1210 	struct inpcbinfo *pcbinfo;
1211 
1212 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1213 
1214 	INP_RLOCK_ASSERT(inp);
1215 
1216 	if (refcount_release(&inp->inp_refcount) == 0) {
1217 		/*
1218 		 * If the inpcb has been freed, let the caller know, even if
1219 		 * this isn't the last reference.
1220 		 */
1221 		if (inp->inp_flags2 & INP_FREED) {
1222 			INP_RUNLOCK(inp);
1223 			return (1);
1224 		}
1225 		return (0);
1226 	}
1227 
1228 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1229 
1230 	INP_RUNLOCK(inp);
1231 	pcbinfo = inp->inp_pcbinfo;
1232 	uma_zfree(pcbinfo->ipi_zone, inp);
1233 	return (1);
1234 }
1235 
1236 int
1237 in_pcbrele_wlocked(struct inpcb *inp)
1238 {
1239 	struct inpcbinfo *pcbinfo;
1240 
1241 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1242 
1243 	INP_WLOCK_ASSERT(inp);
1244 
1245 	if (refcount_release(&inp->inp_refcount) == 0) {
1246 		/*
1247 		 * If the inpcb has been freed, let the caller know, even if
1248 		 * this isn't the last reference.
1249 		 */
1250 		if (inp->inp_flags2 & INP_FREED) {
1251 			INP_WUNLOCK(inp);
1252 			return (1);
1253 		}
1254 		return (0);
1255 	}
1256 
1257 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1258 
1259 	INP_WUNLOCK(inp);
1260 	pcbinfo = inp->inp_pcbinfo;
1261 	uma_zfree(pcbinfo->ipi_zone, inp);
1262 	return (1);
1263 }
1264 
1265 /*
1266  * Temporary wrapper.
1267  */
1268 int
1269 in_pcbrele(struct inpcb *inp)
1270 {
1271 
1272 	return (in_pcbrele_wlocked(inp));
1273 }
1274 
1275 /*
1276  * Unconditionally schedule an inpcb to be freed by decrementing its
1277  * reference count, which should occur only after the inpcb has been detached
1278  * from its socket.  If another thread holds a temporary reference (acquired
1279  * using in_pcbref()) then the free is deferred until that reference is
1280  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1281  * work, including removal from global lists, is done in this context, where
1282  * the pcbinfo lock is held.
1283  */
1284 void
1285 in_pcbfree(struct inpcb *inp)
1286 {
1287 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1288 
1289 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1290 
1291 #ifdef INVARIANTS
1292 	if (pcbinfo == &V_tcbinfo) {
1293 		INP_INFO_LOCK_ASSERT(pcbinfo);
1294 	} else {
1295 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1296 	}
1297 #endif
1298 	INP_WLOCK_ASSERT(inp);
1299 
1300 	/* XXXRW: Do as much as possible here. */
1301 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1302 	if (inp->inp_sp != NULL)
1303 		ipsec_delete_pcbpolicy(inp);
1304 #endif
1305 	INP_LIST_WLOCK(pcbinfo);
1306 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1307 	in_pcbremlists(inp);
1308 	INP_LIST_WUNLOCK(pcbinfo);
1309 #ifdef INET6
1310 	if (inp->inp_vflag & INP_IPV6PROTO) {
1311 		ip6_freepcbopts(inp->in6p_outputopts);
1312 		if (inp->in6p_moptions != NULL)
1313 			ip6_freemoptions(inp->in6p_moptions);
1314 	}
1315 #endif
1316 	if (inp->inp_options)
1317 		(void)m_free(inp->inp_options);
1318 #ifdef INET
1319 	if (inp->inp_moptions != NULL)
1320 		inp_freemoptions(inp->inp_moptions);
1321 #endif
1322 	RO_INVALIDATE_CACHE(&inp->inp_route);
1323 
1324 	inp->inp_vflag = 0;
1325 	inp->inp_flags2 |= INP_FREED;
1326 	crfree(inp->inp_cred);
1327 #ifdef MAC
1328 	mac_inpcb_destroy(inp);
1329 #endif
1330 	if (!in_pcbrele_wlocked(inp))
1331 		INP_WUNLOCK(inp);
1332 }
1333 
1334 /*
1335  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1336  * port reservation, and preventing it from being returned by inpcb lookups.
1337  *
1338  * It is used by TCP to mark an inpcb as unused and avoid future packet
1339  * delivery or event notification when a socket remains open but TCP has
1340  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1341  * or a RST on the wire, and allows the port binding to be reused while still
1342  * maintaining the invariant that so_pcb always points to a valid inpcb until
1343  * in_pcbdetach().
1344  *
1345  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1346  * in_pcbnotifyall() and in_pcbpurgeif0()?
1347  */
1348 void
1349 in_pcbdrop(struct inpcb *inp)
1350 {
1351 
1352 	INP_WLOCK_ASSERT(inp);
1353 
1354 	/*
1355 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1356 	 * the hash lock...?
1357 	 */
1358 	inp->inp_flags |= INP_DROPPED;
1359 	if (inp->inp_flags & INP_INHASHLIST) {
1360 		struct inpcbport *phd = inp->inp_phd;
1361 
1362 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1363 		LIST_REMOVE(inp, inp_hash);
1364 		LIST_REMOVE(inp, inp_portlist);
1365 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1366 			LIST_REMOVE(phd, phd_hash);
1367 			free(phd, M_PCB);
1368 		}
1369 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1370 		inp->inp_flags &= ~INP_INHASHLIST;
1371 #ifdef PCBGROUP
1372 		in_pcbgroup_remove(inp);
1373 #endif
1374 	}
1375 }
1376 
1377 #ifdef INET
1378 /*
1379  * Common routines to return the socket addresses associated with inpcbs.
1380  */
1381 struct sockaddr *
1382 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1383 {
1384 	struct sockaddr_in *sin;
1385 
1386 	sin = malloc(sizeof *sin, M_SONAME,
1387 		M_WAITOK | M_ZERO);
1388 	sin->sin_family = AF_INET;
1389 	sin->sin_len = sizeof(*sin);
1390 	sin->sin_addr = *addr_p;
1391 	sin->sin_port = port;
1392 
1393 	return (struct sockaddr *)sin;
1394 }
1395 
1396 int
1397 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1398 {
1399 	struct inpcb *inp;
1400 	struct in_addr addr;
1401 	in_port_t port;
1402 
1403 	inp = sotoinpcb(so);
1404 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1405 
1406 	INP_RLOCK(inp);
1407 	port = inp->inp_lport;
1408 	addr = inp->inp_laddr;
1409 	INP_RUNLOCK(inp);
1410 
1411 	*nam = in_sockaddr(port, &addr);
1412 	return 0;
1413 }
1414 
1415 int
1416 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1417 {
1418 	struct inpcb *inp;
1419 	struct in_addr addr;
1420 	in_port_t port;
1421 
1422 	inp = sotoinpcb(so);
1423 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1424 
1425 	INP_RLOCK(inp);
1426 	port = inp->inp_fport;
1427 	addr = inp->inp_faddr;
1428 	INP_RUNLOCK(inp);
1429 
1430 	*nam = in_sockaddr(port, &addr);
1431 	return 0;
1432 }
1433 
1434 void
1435 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1436     struct inpcb *(*notify)(struct inpcb *, int))
1437 {
1438 	struct inpcb *inp, *inp_temp;
1439 
1440 	INP_INFO_WLOCK(pcbinfo);
1441 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1442 		INP_WLOCK(inp);
1443 #ifdef INET6
1444 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1445 			INP_WUNLOCK(inp);
1446 			continue;
1447 		}
1448 #endif
1449 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1450 		    inp->inp_socket == NULL) {
1451 			INP_WUNLOCK(inp);
1452 			continue;
1453 		}
1454 		if ((*notify)(inp, errno))
1455 			INP_WUNLOCK(inp);
1456 	}
1457 	INP_INFO_WUNLOCK(pcbinfo);
1458 }
1459 
1460 void
1461 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1462 {
1463 	struct inpcb *inp;
1464 	struct ip_moptions *imo;
1465 	int i, gap;
1466 
1467 	INP_INFO_WLOCK(pcbinfo);
1468 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1469 		INP_WLOCK(inp);
1470 		imo = inp->inp_moptions;
1471 		if ((inp->inp_vflag & INP_IPV4) &&
1472 		    imo != NULL) {
1473 			/*
1474 			 * Unselect the outgoing interface if it is being
1475 			 * detached.
1476 			 */
1477 			if (imo->imo_multicast_ifp == ifp)
1478 				imo->imo_multicast_ifp = NULL;
1479 
1480 			/*
1481 			 * Drop multicast group membership if we joined
1482 			 * through the interface being detached.
1483 			 */
1484 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1485 			    i++) {
1486 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1487 					in_delmulti(imo->imo_membership[i]);
1488 					gap++;
1489 				} else if (gap != 0)
1490 					imo->imo_membership[i - gap] =
1491 					    imo->imo_membership[i];
1492 			}
1493 			imo->imo_num_memberships -= gap;
1494 		}
1495 		INP_WUNLOCK(inp);
1496 	}
1497 	INP_INFO_WUNLOCK(pcbinfo);
1498 }
1499 
1500 /*
1501  * Lookup a PCB based on the local address and port.  Caller must hold the
1502  * hash lock.  No inpcb locks or references are acquired.
1503  */
1504 #define INP_LOOKUP_MAPPED_PCB_COST	3
1505 struct inpcb *
1506 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1507     u_short lport, int lookupflags, struct ucred *cred)
1508 {
1509 	struct inpcb *inp;
1510 #ifdef INET6
1511 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1512 #else
1513 	int matchwild = 3;
1514 #endif
1515 	int wildcard;
1516 
1517 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1518 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1519 
1520 	INP_HASH_LOCK_ASSERT(pcbinfo);
1521 
1522 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1523 		struct inpcbhead *head;
1524 		/*
1525 		 * Look for an unconnected (wildcard foreign addr) PCB that
1526 		 * matches the local address and port we're looking for.
1527 		 */
1528 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1529 		    0, pcbinfo->ipi_hashmask)];
1530 		LIST_FOREACH(inp, head, inp_hash) {
1531 #ifdef INET6
1532 			/* XXX inp locking */
1533 			if ((inp->inp_vflag & INP_IPV4) == 0)
1534 				continue;
1535 #endif
1536 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1537 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1538 			    inp->inp_lport == lport) {
1539 				/*
1540 				 * Found?
1541 				 */
1542 				if (cred == NULL ||
1543 				    prison_equal_ip4(cred->cr_prison,
1544 					inp->inp_cred->cr_prison))
1545 					return (inp);
1546 			}
1547 		}
1548 		/*
1549 		 * Not found.
1550 		 */
1551 		return (NULL);
1552 	} else {
1553 		struct inpcbporthead *porthash;
1554 		struct inpcbport *phd;
1555 		struct inpcb *match = NULL;
1556 		/*
1557 		 * Best fit PCB lookup.
1558 		 *
1559 		 * First see if this local port is in use by looking on the
1560 		 * port hash list.
1561 		 */
1562 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1563 		    pcbinfo->ipi_porthashmask)];
1564 		LIST_FOREACH(phd, porthash, phd_hash) {
1565 			if (phd->phd_port == lport)
1566 				break;
1567 		}
1568 		if (phd != NULL) {
1569 			/*
1570 			 * Port is in use by one or more PCBs. Look for best
1571 			 * fit.
1572 			 */
1573 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1574 				wildcard = 0;
1575 				if (cred != NULL &&
1576 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1577 					cred->cr_prison))
1578 					continue;
1579 #ifdef INET6
1580 				/* XXX inp locking */
1581 				if ((inp->inp_vflag & INP_IPV4) == 0)
1582 					continue;
1583 				/*
1584 				 * We never select the PCB that has
1585 				 * INP_IPV6 flag and is bound to :: if
1586 				 * we have another PCB which is bound
1587 				 * to 0.0.0.0.  If a PCB has the
1588 				 * INP_IPV6 flag, then we set its cost
1589 				 * higher than IPv4 only PCBs.
1590 				 *
1591 				 * Note that the case only happens
1592 				 * when a socket is bound to ::, under
1593 				 * the condition that the use of the
1594 				 * mapped address is allowed.
1595 				 */
1596 				if ((inp->inp_vflag & INP_IPV6) != 0)
1597 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1598 #endif
1599 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1600 					wildcard++;
1601 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1602 					if (laddr.s_addr == INADDR_ANY)
1603 						wildcard++;
1604 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1605 						continue;
1606 				} else {
1607 					if (laddr.s_addr != INADDR_ANY)
1608 						wildcard++;
1609 				}
1610 				if (wildcard < matchwild) {
1611 					match = inp;
1612 					matchwild = wildcard;
1613 					if (matchwild == 0)
1614 						break;
1615 				}
1616 			}
1617 		}
1618 		return (match);
1619 	}
1620 }
1621 #undef INP_LOOKUP_MAPPED_PCB_COST
1622 
1623 #ifdef PCBGROUP
1624 /*
1625  * Lookup PCB in hash list, using pcbgroup tables.
1626  */
1627 static struct inpcb *
1628 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1629     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1630     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1631 {
1632 	struct inpcbhead *head;
1633 	struct inpcb *inp, *tmpinp;
1634 	u_short fport = fport_arg, lport = lport_arg;
1635 
1636 	/*
1637 	 * First look for an exact match.
1638 	 */
1639 	tmpinp = NULL;
1640 	INP_GROUP_LOCK(pcbgroup);
1641 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1642 	    pcbgroup->ipg_hashmask)];
1643 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1644 #ifdef INET6
1645 		/* XXX inp locking */
1646 		if ((inp->inp_vflag & INP_IPV4) == 0)
1647 			continue;
1648 #endif
1649 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1650 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1651 		    inp->inp_fport == fport &&
1652 		    inp->inp_lport == lport) {
1653 			/*
1654 			 * XXX We should be able to directly return
1655 			 * the inp here, without any checks.
1656 			 * Well unless both bound with SO_REUSEPORT?
1657 			 */
1658 			if (prison_flag(inp->inp_cred, PR_IP4))
1659 				goto found;
1660 			if (tmpinp == NULL)
1661 				tmpinp = inp;
1662 		}
1663 	}
1664 	if (tmpinp != NULL) {
1665 		inp = tmpinp;
1666 		goto found;
1667 	}
1668 
1669 #ifdef	RSS
1670 	/*
1671 	 * For incoming connections, we may wish to do a wildcard
1672 	 * match for an RSS-local socket.
1673 	 */
1674 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1675 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1676 #ifdef INET6
1677 		struct inpcb *local_wild_mapped = NULL;
1678 #endif
1679 		struct inpcb *jail_wild = NULL;
1680 		struct inpcbhead *head;
1681 		int injail;
1682 
1683 		/*
1684 		 * Order of socket selection - we always prefer jails.
1685 		 *      1. jailed, non-wild.
1686 		 *      2. jailed, wild.
1687 		 *      3. non-jailed, non-wild.
1688 		 *      4. non-jailed, wild.
1689 		 */
1690 
1691 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1692 		    lport, 0, pcbgroup->ipg_hashmask)];
1693 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1694 #ifdef INET6
1695 			/* XXX inp locking */
1696 			if ((inp->inp_vflag & INP_IPV4) == 0)
1697 				continue;
1698 #endif
1699 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1700 			    inp->inp_lport != lport)
1701 				continue;
1702 
1703 			injail = prison_flag(inp->inp_cred, PR_IP4);
1704 			if (injail) {
1705 				if (prison_check_ip4(inp->inp_cred,
1706 				    &laddr) != 0)
1707 					continue;
1708 			} else {
1709 				if (local_exact != NULL)
1710 					continue;
1711 			}
1712 
1713 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1714 				if (injail)
1715 					goto found;
1716 				else
1717 					local_exact = inp;
1718 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1719 #ifdef INET6
1720 				/* XXX inp locking, NULL check */
1721 				if (inp->inp_vflag & INP_IPV6PROTO)
1722 					local_wild_mapped = inp;
1723 				else
1724 #endif
1725 					if (injail)
1726 						jail_wild = inp;
1727 					else
1728 						local_wild = inp;
1729 			}
1730 		} /* LIST_FOREACH */
1731 
1732 		inp = jail_wild;
1733 		if (inp == NULL)
1734 			inp = local_exact;
1735 		if (inp == NULL)
1736 			inp = local_wild;
1737 #ifdef INET6
1738 		if (inp == NULL)
1739 			inp = local_wild_mapped;
1740 #endif
1741 		if (inp != NULL)
1742 			goto found;
1743 	}
1744 #endif
1745 
1746 	/*
1747 	 * Then look for a wildcard match, if requested.
1748 	 */
1749 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1750 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1751 #ifdef INET6
1752 		struct inpcb *local_wild_mapped = NULL;
1753 #endif
1754 		struct inpcb *jail_wild = NULL;
1755 		struct inpcbhead *head;
1756 		int injail;
1757 
1758 		/*
1759 		 * Order of socket selection - we always prefer jails.
1760 		 *      1. jailed, non-wild.
1761 		 *      2. jailed, wild.
1762 		 *      3. non-jailed, non-wild.
1763 		 *      4. non-jailed, wild.
1764 		 */
1765 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1766 		    0, pcbinfo->ipi_wildmask)];
1767 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1768 #ifdef INET6
1769 			/* XXX inp locking */
1770 			if ((inp->inp_vflag & INP_IPV4) == 0)
1771 				continue;
1772 #endif
1773 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1774 			    inp->inp_lport != lport)
1775 				continue;
1776 
1777 			injail = prison_flag(inp->inp_cred, PR_IP4);
1778 			if (injail) {
1779 				if (prison_check_ip4(inp->inp_cred,
1780 				    &laddr) != 0)
1781 					continue;
1782 			} else {
1783 				if (local_exact != NULL)
1784 					continue;
1785 			}
1786 
1787 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1788 				if (injail)
1789 					goto found;
1790 				else
1791 					local_exact = inp;
1792 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1793 #ifdef INET6
1794 				/* XXX inp locking, NULL check */
1795 				if (inp->inp_vflag & INP_IPV6PROTO)
1796 					local_wild_mapped = inp;
1797 				else
1798 #endif
1799 					if (injail)
1800 						jail_wild = inp;
1801 					else
1802 						local_wild = inp;
1803 			}
1804 		} /* LIST_FOREACH */
1805 		inp = jail_wild;
1806 		if (inp == NULL)
1807 			inp = local_exact;
1808 		if (inp == NULL)
1809 			inp = local_wild;
1810 #ifdef INET6
1811 		if (inp == NULL)
1812 			inp = local_wild_mapped;
1813 #endif
1814 		if (inp != NULL)
1815 			goto found;
1816 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1817 	INP_GROUP_UNLOCK(pcbgroup);
1818 	return (NULL);
1819 
1820 found:
1821 	in_pcbref(inp);
1822 	INP_GROUP_UNLOCK(pcbgroup);
1823 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1824 		INP_WLOCK(inp);
1825 		if (in_pcbrele_wlocked(inp))
1826 			return (NULL);
1827 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1828 		INP_RLOCK(inp);
1829 		if (in_pcbrele_rlocked(inp))
1830 			return (NULL);
1831 	} else
1832 		panic("%s: locking bug", __func__);
1833 	return (inp);
1834 }
1835 #endif /* PCBGROUP */
1836 
1837 /*
1838  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1839  * that the caller has locked the hash list, and will not perform any further
1840  * locking or reference operations on either the hash list or the connection.
1841  */
1842 static struct inpcb *
1843 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1844     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1845     struct ifnet *ifp)
1846 {
1847 	struct inpcbhead *head;
1848 	struct inpcb *inp, *tmpinp;
1849 	u_short fport = fport_arg, lport = lport_arg;
1850 
1851 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1852 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1853 
1854 	INP_HASH_LOCK_ASSERT(pcbinfo);
1855 
1856 	/*
1857 	 * First look for an exact match.
1858 	 */
1859 	tmpinp = NULL;
1860 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1861 	    pcbinfo->ipi_hashmask)];
1862 	LIST_FOREACH(inp, head, inp_hash) {
1863 #ifdef INET6
1864 		/* XXX inp locking */
1865 		if ((inp->inp_vflag & INP_IPV4) == 0)
1866 			continue;
1867 #endif
1868 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1869 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1870 		    inp->inp_fport == fport &&
1871 		    inp->inp_lport == lport) {
1872 			/*
1873 			 * XXX We should be able to directly return
1874 			 * the inp here, without any checks.
1875 			 * Well unless both bound with SO_REUSEPORT?
1876 			 */
1877 			if (prison_flag(inp->inp_cred, PR_IP4))
1878 				return (inp);
1879 			if (tmpinp == NULL)
1880 				tmpinp = inp;
1881 		}
1882 	}
1883 	if (tmpinp != NULL)
1884 		return (tmpinp);
1885 
1886 	/*
1887 	 * Then look for a wildcard match, if requested.
1888 	 */
1889 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1890 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1891 #ifdef INET6
1892 		struct inpcb *local_wild_mapped = NULL;
1893 #endif
1894 		struct inpcb *jail_wild = NULL;
1895 		int injail;
1896 
1897 		/*
1898 		 * Order of socket selection - we always prefer jails.
1899 		 *      1. jailed, non-wild.
1900 		 *      2. jailed, wild.
1901 		 *      3. non-jailed, non-wild.
1902 		 *      4. non-jailed, wild.
1903 		 */
1904 
1905 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1906 		    0, pcbinfo->ipi_hashmask)];
1907 		LIST_FOREACH(inp, head, inp_hash) {
1908 #ifdef INET6
1909 			/* XXX inp locking */
1910 			if ((inp->inp_vflag & INP_IPV4) == 0)
1911 				continue;
1912 #endif
1913 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1914 			    inp->inp_lport != lport)
1915 				continue;
1916 
1917 			injail = prison_flag(inp->inp_cred, PR_IP4);
1918 			if (injail) {
1919 				if (prison_check_ip4(inp->inp_cred,
1920 				    &laddr) != 0)
1921 					continue;
1922 			} else {
1923 				if (local_exact != NULL)
1924 					continue;
1925 			}
1926 
1927 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1928 				if (injail)
1929 					return (inp);
1930 				else
1931 					local_exact = inp;
1932 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1933 #ifdef INET6
1934 				/* XXX inp locking, NULL check */
1935 				if (inp->inp_vflag & INP_IPV6PROTO)
1936 					local_wild_mapped = inp;
1937 				else
1938 #endif
1939 					if (injail)
1940 						jail_wild = inp;
1941 					else
1942 						local_wild = inp;
1943 			}
1944 		} /* LIST_FOREACH */
1945 		if (jail_wild != NULL)
1946 			return (jail_wild);
1947 		if (local_exact != NULL)
1948 			return (local_exact);
1949 		if (local_wild != NULL)
1950 			return (local_wild);
1951 #ifdef INET6
1952 		if (local_wild_mapped != NULL)
1953 			return (local_wild_mapped);
1954 #endif
1955 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1956 
1957 	return (NULL);
1958 }
1959 
1960 /*
1961  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1962  * hash list lock, and will return the inpcb locked (i.e., requires
1963  * INPLOOKUP_LOCKPCB).
1964  */
1965 static struct inpcb *
1966 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1967     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1968     struct ifnet *ifp)
1969 {
1970 	struct inpcb *inp;
1971 
1972 	INP_HASH_RLOCK(pcbinfo);
1973 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1974 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1975 	if (inp != NULL) {
1976 		in_pcbref(inp);
1977 		INP_HASH_RUNLOCK(pcbinfo);
1978 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1979 			INP_WLOCK(inp);
1980 			if (in_pcbrele_wlocked(inp))
1981 				return (NULL);
1982 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1983 			INP_RLOCK(inp);
1984 			if (in_pcbrele_rlocked(inp))
1985 				return (NULL);
1986 		} else
1987 			panic("%s: locking bug", __func__);
1988 	} else
1989 		INP_HASH_RUNLOCK(pcbinfo);
1990 	return (inp);
1991 }
1992 
1993 /*
1994  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1995  * from which a pre-calculated hash value may be extracted.
1996  *
1997  * Possibly more of this logic should be in in_pcbgroup.c.
1998  */
1999 struct inpcb *
2000 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2001     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2002 {
2003 #if defined(PCBGROUP) && !defined(RSS)
2004 	struct inpcbgroup *pcbgroup;
2005 #endif
2006 
2007 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2008 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2009 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2010 	    ("%s: LOCKPCB not set", __func__));
2011 
2012 	/*
2013 	 * When not using RSS, use connection groups in preference to the
2014 	 * reservation table when looking up 4-tuples.  When using RSS, just
2015 	 * use the reservation table, due to the cost of the Toeplitz hash
2016 	 * in software.
2017 	 *
2018 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2019 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2020 	 * in software.
2021 	 */
2022 #if defined(PCBGROUP) && !defined(RSS)
2023 	if (in_pcbgroup_enabled(pcbinfo)) {
2024 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2025 		    fport);
2026 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2027 		    laddr, lport, lookupflags, ifp));
2028 	}
2029 #endif
2030 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2031 	    lookupflags, ifp));
2032 }
2033 
2034 struct inpcb *
2035 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2036     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2037     struct ifnet *ifp, struct mbuf *m)
2038 {
2039 #ifdef PCBGROUP
2040 	struct inpcbgroup *pcbgroup;
2041 #endif
2042 
2043 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2044 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2045 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2046 	    ("%s: LOCKPCB not set", __func__));
2047 
2048 #ifdef PCBGROUP
2049 	/*
2050 	 * If we can use a hardware-generated hash to look up the connection
2051 	 * group, use that connection group to find the inpcb.  Otherwise
2052 	 * fall back on a software hash -- or the reservation table if we're
2053 	 * using RSS.
2054 	 *
2055 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2056 	 */
2057 	if (in_pcbgroup_enabled(pcbinfo) &&
2058 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2059 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2060 		    m->m_pkthdr.flowid);
2061 		if (pcbgroup != NULL)
2062 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2063 			    fport, laddr, lport, lookupflags, ifp));
2064 #ifndef RSS
2065 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2066 		    fport);
2067 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2068 		    laddr, lport, lookupflags, ifp));
2069 #endif
2070 	}
2071 #endif
2072 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2073 	    lookupflags, ifp));
2074 }
2075 #endif /* INET */
2076 
2077 /*
2078  * Insert PCB onto various hash lists.
2079  */
2080 static int
2081 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2082 {
2083 	struct inpcbhead *pcbhash;
2084 	struct inpcbporthead *pcbporthash;
2085 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2086 	struct inpcbport *phd;
2087 	u_int32_t hashkey_faddr;
2088 
2089 	INP_WLOCK_ASSERT(inp);
2090 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2091 
2092 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2093 	    ("in_pcbinshash: INP_INHASHLIST"));
2094 
2095 #ifdef INET6
2096 	if (inp->inp_vflag & INP_IPV6)
2097 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2098 	else
2099 #endif
2100 	hashkey_faddr = inp->inp_faddr.s_addr;
2101 
2102 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2103 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2104 
2105 	pcbporthash = &pcbinfo->ipi_porthashbase[
2106 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2107 
2108 	/*
2109 	 * Go through port list and look for a head for this lport.
2110 	 */
2111 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2112 		if (phd->phd_port == inp->inp_lport)
2113 			break;
2114 	}
2115 	/*
2116 	 * If none exists, malloc one and tack it on.
2117 	 */
2118 	if (phd == NULL) {
2119 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2120 		if (phd == NULL) {
2121 			return (ENOBUFS); /* XXX */
2122 		}
2123 		phd->phd_port = inp->inp_lport;
2124 		LIST_INIT(&phd->phd_pcblist);
2125 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2126 	}
2127 	inp->inp_phd = phd;
2128 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2129 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2130 	inp->inp_flags |= INP_INHASHLIST;
2131 #ifdef PCBGROUP
2132 	if (do_pcbgroup_update)
2133 		in_pcbgroup_update(inp);
2134 #endif
2135 	return (0);
2136 }
2137 
2138 /*
2139  * For now, there are two public interfaces to insert an inpcb into the hash
2140  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2141  * is used only in the TCP syncache, where in_pcbinshash is called before the
2142  * full 4-tuple is set for the inpcb, and we don't want to install in the
2143  * pcbgroup until later.
2144  *
2145  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2146  * connection groups, and partially initialised inpcbs should not be exposed
2147  * to either reservation hash tables or pcbgroups.
2148  */
2149 int
2150 in_pcbinshash(struct inpcb *inp)
2151 {
2152 
2153 	return (in_pcbinshash_internal(inp, 1));
2154 }
2155 
2156 int
2157 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2158 {
2159 
2160 	return (in_pcbinshash_internal(inp, 0));
2161 }
2162 
2163 /*
2164  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2165  * changed. NOTE: This does not handle the case of the lport changing (the
2166  * hashed port list would have to be updated as well), so the lport must
2167  * not change after in_pcbinshash() has been called.
2168  */
2169 void
2170 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2171 {
2172 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2173 	struct inpcbhead *head;
2174 	u_int32_t hashkey_faddr;
2175 
2176 	INP_WLOCK_ASSERT(inp);
2177 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2178 
2179 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2180 	    ("in_pcbrehash: !INP_INHASHLIST"));
2181 
2182 #ifdef INET6
2183 	if (inp->inp_vflag & INP_IPV6)
2184 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2185 	else
2186 #endif
2187 	hashkey_faddr = inp->inp_faddr.s_addr;
2188 
2189 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2190 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2191 
2192 	LIST_REMOVE(inp, inp_hash);
2193 	LIST_INSERT_HEAD(head, inp, inp_hash);
2194 
2195 #ifdef PCBGROUP
2196 	if (m != NULL)
2197 		in_pcbgroup_update_mbuf(inp, m);
2198 	else
2199 		in_pcbgroup_update(inp);
2200 #endif
2201 }
2202 
2203 void
2204 in_pcbrehash(struct inpcb *inp)
2205 {
2206 
2207 	in_pcbrehash_mbuf(inp, NULL);
2208 }
2209 
2210 /*
2211  * Remove PCB from various lists.
2212  */
2213 static void
2214 in_pcbremlists(struct inpcb *inp)
2215 {
2216 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2217 
2218 #ifdef INVARIANTS
2219 	if (pcbinfo == &V_tcbinfo) {
2220 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2221 	} else {
2222 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2223 	}
2224 #endif
2225 
2226 	INP_WLOCK_ASSERT(inp);
2227 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2228 
2229 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2230 	if (inp->inp_flags & INP_INHASHLIST) {
2231 		struct inpcbport *phd = inp->inp_phd;
2232 
2233 		INP_HASH_WLOCK(pcbinfo);
2234 		LIST_REMOVE(inp, inp_hash);
2235 		LIST_REMOVE(inp, inp_portlist);
2236 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2237 			LIST_REMOVE(phd, phd_hash);
2238 			free(phd, M_PCB);
2239 		}
2240 		INP_HASH_WUNLOCK(pcbinfo);
2241 		inp->inp_flags &= ~INP_INHASHLIST;
2242 	}
2243 	LIST_REMOVE(inp, inp_list);
2244 	pcbinfo->ipi_count--;
2245 #ifdef PCBGROUP
2246 	in_pcbgroup_remove(inp);
2247 #endif
2248 }
2249 
2250 /*
2251  * Check for alternatives when higher level complains
2252  * about service problems.  For now, invalidate cached
2253  * routing information.  If the route was created dynamically
2254  * (by a redirect), time to try a default gateway again.
2255  */
2256 void
2257 in_losing(struct inpcb *inp)
2258 {
2259 
2260 	RO_INVALIDATE_CACHE(&inp->inp_route);
2261 	return;
2262 }
2263 
2264 /*
2265  * A set label operation has occurred at the socket layer, propagate the
2266  * label change into the in_pcb for the socket.
2267  */
2268 void
2269 in_pcbsosetlabel(struct socket *so)
2270 {
2271 #ifdef MAC
2272 	struct inpcb *inp;
2273 
2274 	inp = sotoinpcb(so);
2275 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2276 
2277 	INP_WLOCK(inp);
2278 	SOCK_LOCK(so);
2279 	mac_inpcb_sosetlabel(so, inp);
2280 	SOCK_UNLOCK(so);
2281 	INP_WUNLOCK(inp);
2282 #endif
2283 }
2284 
2285 /*
2286  * ipport_tick runs once per second, determining if random port allocation
2287  * should be continued.  If more than ipport_randomcps ports have been
2288  * allocated in the last second, then we return to sequential port
2289  * allocation. We return to random allocation only once we drop below
2290  * ipport_randomcps for at least ipport_randomtime seconds.
2291  */
2292 static void
2293 ipport_tick(void *xtp)
2294 {
2295 	VNET_ITERATOR_DECL(vnet_iter);
2296 
2297 	VNET_LIST_RLOCK_NOSLEEP();
2298 	VNET_FOREACH(vnet_iter) {
2299 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2300 		if (V_ipport_tcpallocs <=
2301 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2302 			if (V_ipport_stoprandom > 0)
2303 				V_ipport_stoprandom--;
2304 		} else
2305 			V_ipport_stoprandom = V_ipport_randomtime;
2306 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2307 		CURVNET_RESTORE();
2308 	}
2309 	VNET_LIST_RUNLOCK_NOSLEEP();
2310 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2311 }
2312 
2313 static void
2314 ip_fini(void *xtp)
2315 {
2316 
2317 	callout_stop(&ipport_tick_callout);
2318 }
2319 
2320 /*
2321  * The ipport_callout should start running at about the time we attach the
2322  * inet or inet6 domains.
2323  */
2324 static void
2325 ipport_tick_init(const void *unused __unused)
2326 {
2327 
2328 	/* Start ipport_tick. */
2329 	callout_init(&ipport_tick_callout, 1);
2330 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2331 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2332 		SHUTDOWN_PRI_DEFAULT);
2333 }
2334 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2335     ipport_tick_init, NULL);
2336 
2337 void
2338 inp_wlock(struct inpcb *inp)
2339 {
2340 
2341 	INP_WLOCK(inp);
2342 }
2343 
2344 void
2345 inp_wunlock(struct inpcb *inp)
2346 {
2347 
2348 	INP_WUNLOCK(inp);
2349 }
2350 
2351 void
2352 inp_rlock(struct inpcb *inp)
2353 {
2354 
2355 	INP_RLOCK(inp);
2356 }
2357 
2358 void
2359 inp_runlock(struct inpcb *inp)
2360 {
2361 
2362 	INP_RUNLOCK(inp);
2363 }
2364 
2365 #ifdef INVARIANT_SUPPORT
2366 void
2367 inp_lock_assert(struct inpcb *inp)
2368 {
2369 
2370 	INP_WLOCK_ASSERT(inp);
2371 }
2372 
2373 void
2374 inp_unlock_assert(struct inpcb *inp)
2375 {
2376 
2377 	INP_UNLOCK_ASSERT(inp);
2378 }
2379 #endif
2380 
2381 void
2382 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2383 {
2384 	struct inpcb *inp;
2385 
2386 	INP_INFO_WLOCK(&V_tcbinfo);
2387 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2388 		INP_WLOCK(inp);
2389 		func(inp, arg);
2390 		INP_WUNLOCK(inp);
2391 	}
2392 	INP_INFO_WUNLOCK(&V_tcbinfo);
2393 }
2394 
2395 struct socket *
2396 inp_inpcbtosocket(struct inpcb *inp)
2397 {
2398 
2399 	INP_WLOCK_ASSERT(inp);
2400 	return (inp->inp_socket);
2401 }
2402 
2403 struct tcpcb *
2404 inp_inpcbtotcpcb(struct inpcb *inp)
2405 {
2406 
2407 	INP_WLOCK_ASSERT(inp);
2408 	return ((struct tcpcb *)inp->inp_ppcb);
2409 }
2410 
2411 int
2412 inp_ip_tos_get(const struct inpcb *inp)
2413 {
2414 
2415 	return (inp->inp_ip_tos);
2416 }
2417 
2418 void
2419 inp_ip_tos_set(struct inpcb *inp, int val)
2420 {
2421 
2422 	inp->inp_ip_tos = val;
2423 }
2424 
2425 void
2426 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2427     uint32_t *faddr, uint16_t *fp)
2428 {
2429 
2430 	INP_LOCK_ASSERT(inp);
2431 	*laddr = inp->inp_laddr.s_addr;
2432 	*faddr = inp->inp_faddr.s_addr;
2433 	*lp = inp->inp_lport;
2434 	*fp = inp->inp_fport;
2435 }
2436 
2437 struct inpcb *
2438 so_sotoinpcb(struct socket *so)
2439 {
2440 
2441 	return (sotoinpcb(so));
2442 }
2443 
2444 struct tcpcb *
2445 so_sototcpcb(struct socket *so)
2446 {
2447 
2448 	return (sototcpcb(so));
2449 }
2450 
2451 /*
2452  * Create an external-format (``xinpcb'') structure using the information in
2453  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2454  * reduce the spew of irrelevant information over this interface, to isolate
2455  * user code from changes in the kernel structure, and potentially to provide
2456  * information-hiding if we decide that some of this information should be
2457  * hidden from users.
2458  */
2459 void
2460 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2461 {
2462 
2463 	xi->xi_len = sizeof(struct xinpcb);
2464 	if (inp->inp_socket)
2465 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2466 	else
2467 		bzero(&xi->xi_socket, sizeof(struct xsocket));
2468 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2469 	xi->inp_gencnt = inp->inp_gencnt;
2470 	xi->inp_ppcb = inp->inp_ppcb;
2471 	xi->inp_flow = inp->inp_flow;
2472 	xi->inp_flowid = inp->inp_flowid;
2473 	xi->inp_flowtype = inp->inp_flowtype;
2474 	xi->inp_flags = inp->inp_flags;
2475 	xi->inp_flags2 = inp->inp_flags2;
2476 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2477 	xi->in6p_cksum = inp->in6p_cksum;
2478 	xi->in6p_hops = inp->in6p_hops;
2479 	xi->inp_ip_tos = inp->inp_ip_tos;
2480 	xi->inp_vflag = inp->inp_vflag;
2481 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2482 	xi->inp_ip_p = inp->inp_ip_p;
2483 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2484 }
2485 
2486 #ifdef DDB
2487 static void
2488 db_print_indent(int indent)
2489 {
2490 	int i;
2491 
2492 	for (i = 0; i < indent; i++)
2493 		db_printf(" ");
2494 }
2495 
2496 static void
2497 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2498 {
2499 	char faddr_str[48], laddr_str[48];
2500 
2501 	db_print_indent(indent);
2502 	db_printf("%s at %p\n", name, inc);
2503 
2504 	indent += 2;
2505 
2506 #ifdef INET6
2507 	if (inc->inc_flags & INC_ISIPV6) {
2508 		/* IPv6. */
2509 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2510 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2511 	} else
2512 #endif
2513 	{
2514 		/* IPv4. */
2515 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2516 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2517 	}
2518 	db_print_indent(indent);
2519 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2520 	    ntohs(inc->inc_lport));
2521 	db_print_indent(indent);
2522 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2523 	    ntohs(inc->inc_fport));
2524 }
2525 
2526 static void
2527 db_print_inpflags(int inp_flags)
2528 {
2529 	int comma;
2530 
2531 	comma = 0;
2532 	if (inp_flags & INP_RECVOPTS) {
2533 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2534 		comma = 1;
2535 	}
2536 	if (inp_flags & INP_RECVRETOPTS) {
2537 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2538 		comma = 1;
2539 	}
2540 	if (inp_flags & INP_RECVDSTADDR) {
2541 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2542 		comma = 1;
2543 	}
2544 	if (inp_flags & INP_ORIGDSTADDR) {
2545 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2546 		comma = 1;
2547 	}
2548 	if (inp_flags & INP_HDRINCL) {
2549 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2550 		comma = 1;
2551 	}
2552 	if (inp_flags & INP_HIGHPORT) {
2553 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2554 		comma = 1;
2555 	}
2556 	if (inp_flags & INP_LOWPORT) {
2557 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2558 		comma = 1;
2559 	}
2560 	if (inp_flags & INP_ANONPORT) {
2561 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2562 		comma = 1;
2563 	}
2564 	if (inp_flags & INP_RECVIF) {
2565 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2566 		comma = 1;
2567 	}
2568 	if (inp_flags & INP_MTUDISC) {
2569 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2570 		comma = 1;
2571 	}
2572 	if (inp_flags & INP_RECVTTL) {
2573 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2574 		comma = 1;
2575 	}
2576 	if (inp_flags & INP_DONTFRAG) {
2577 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2578 		comma = 1;
2579 	}
2580 	if (inp_flags & INP_RECVTOS) {
2581 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2582 		comma = 1;
2583 	}
2584 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2585 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2586 		comma = 1;
2587 	}
2588 	if (inp_flags & IN6P_PKTINFO) {
2589 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2590 		comma = 1;
2591 	}
2592 	if (inp_flags & IN6P_HOPLIMIT) {
2593 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2594 		comma = 1;
2595 	}
2596 	if (inp_flags & IN6P_HOPOPTS) {
2597 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2598 		comma = 1;
2599 	}
2600 	if (inp_flags & IN6P_DSTOPTS) {
2601 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2602 		comma = 1;
2603 	}
2604 	if (inp_flags & IN6P_RTHDR) {
2605 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2606 		comma = 1;
2607 	}
2608 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2609 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2610 		comma = 1;
2611 	}
2612 	if (inp_flags & IN6P_TCLASS) {
2613 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2614 		comma = 1;
2615 	}
2616 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2617 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2618 		comma = 1;
2619 	}
2620 	if (inp_flags & INP_TIMEWAIT) {
2621 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2622 		comma  = 1;
2623 	}
2624 	if (inp_flags & INP_ONESBCAST) {
2625 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2626 		comma  = 1;
2627 	}
2628 	if (inp_flags & INP_DROPPED) {
2629 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2630 		comma  = 1;
2631 	}
2632 	if (inp_flags & INP_SOCKREF) {
2633 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2634 		comma  = 1;
2635 	}
2636 	if (inp_flags & IN6P_RFC2292) {
2637 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2638 		comma = 1;
2639 	}
2640 	if (inp_flags & IN6P_MTU) {
2641 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2642 		comma = 1;
2643 	}
2644 }
2645 
2646 static void
2647 db_print_inpvflag(u_char inp_vflag)
2648 {
2649 	int comma;
2650 
2651 	comma = 0;
2652 	if (inp_vflag & INP_IPV4) {
2653 		db_printf("%sINP_IPV4", comma ? ", " : "");
2654 		comma  = 1;
2655 	}
2656 	if (inp_vflag & INP_IPV6) {
2657 		db_printf("%sINP_IPV6", comma ? ", " : "");
2658 		comma  = 1;
2659 	}
2660 	if (inp_vflag & INP_IPV6PROTO) {
2661 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2662 		comma  = 1;
2663 	}
2664 }
2665 
2666 static void
2667 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2668 {
2669 
2670 	db_print_indent(indent);
2671 	db_printf("%s at %p\n", name, inp);
2672 
2673 	indent += 2;
2674 
2675 	db_print_indent(indent);
2676 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2677 
2678 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2679 
2680 	db_print_indent(indent);
2681 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2682 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2683 
2684 	db_print_indent(indent);
2685 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2686 	   inp->inp_label, inp->inp_flags);
2687 	db_print_inpflags(inp->inp_flags);
2688 	db_printf(")\n");
2689 
2690 	db_print_indent(indent);
2691 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2692 	    inp->inp_vflag);
2693 	db_print_inpvflag(inp->inp_vflag);
2694 	db_printf(")\n");
2695 
2696 	db_print_indent(indent);
2697 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2698 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2699 
2700 	db_print_indent(indent);
2701 #ifdef INET6
2702 	if (inp->inp_vflag & INP_IPV6) {
2703 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2704 		    "in6p_moptions: %p\n", inp->in6p_options,
2705 		    inp->in6p_outputopts, inp->in6p_moptions);
2706 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2707 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2708 		    inp->in6p_hops);
2709 	} else
2710 #endif
2711 	{
2712 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2713 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2714 		    inp->inp_options, inp->inp_moptions);
2715 	}
2716 
2717 	db_print_indent(indent);
2718 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2719 	    (uintmax_t)inp->inp_gencnt);
2720 }
2721 
2722 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2723 {
2724 	struct inpcb *inp;
2725 
2726 	if (!have_addr) {
2727 		db_printf("usage: show inpcb <addr>\n");
2728 		return;
2729 	}
2730 	inp = (struct inpcb *)addr;
2731 
2732 	db_print_inpcb(inp, "inpcb", 0);
2733 }
2734 #endif /* DDB */
2735 
2736 #ifdef RATELIMIT
2737 /*
2738  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2739  * if any.
2740  */
2741 int
2742 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2743 {
2744 	union if_snd_tag_modify_params params = {
2745 		.rate_limit.max_rate = max_pacing_rate,
2746 	};
2747 	struct m_snd_tag *mst;
2748 	struct ifnet *ifp;
2749 	int error;
2750 
2751 	mst = inp->inp_snd_tag;
2752 	if (mst == NULL)
2753 		return (EINVAL);
2754 
2755 	ifp = mst->ifp;
2756 	if (ifp == NULL)
2757 		return (EINVAL);
2758 
2759 	if (ifp->if_snd_tag_modify == NULL) {
2760 		error = EOPNOTSUPP;
2761 	} else {
2762 		error = ifp->if_snd_tag_modify(mst, &params);
2763 	}
2764 	return (error);
2765 }
2766 
2767 /*
2768  * Query existing TX rate limit based on the existing
2769  * "inp->inp_snd_tag", if any.
2770  */
2771 int
2772 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
2773 {
2774 	union if_snd_tag_query_params params = { };
2775 	struct m_snd_tag *mst;
2776 	struct ifnet *ifp;
2777 	int error;
2778 
2779 	mst = inp->inp_snd_tag;
2780 	if (mst == NULL)
2781 		return (EINVAL);
2782 
2783 	ifp = mst->ifp;
2784 	if (ifp == NULL)
2785 		return (EINVAL);
2786 
2787 	if (ifp->if_snd_tag_query == NULL) {
2788 		error = EOPNOTSUPP;
2789 	} else {
2790 		error = ifp->if_snd_tag_query(mst, &params);
2791 		if (error == 0 &&  p_max_pacing_rate != NULL)
2792 			*p_max_pacing_rate = params.rate_limit.max_rate;
2793 	}
2794 	return (error);
2795 }
2796 
2797 /*
2798  * Query existing TX queue level based on the existing
2799  * "inp->inp_snd_tag", if any.
2800  */
2801 int
2802 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
2803 {
2804 	union if_snd_tag_query_params params = { };
2805 	struct m_snd_tag *mst;
2806 	struct ifnet *ifp;
2807 	int error;
2808 
2809 	mst = inp->inp_snd_tag;
2810 	if (mst == NULL)
2811 		return (EINVAL);
2812 
2813 	ifp = mst->ifp;
2814 	if (ifp == NULL)
2815 		return (EINVAL);
2816 
2817 	if (ifp->if_snd_tag_query == NULL)
2818 		return (EOPNOTSUPP);
2819 
2820 	error = ifp->if_snd_tag_query(mst, &params);
2821 	if (error == 0 &&  p_txqueue_level != NULL)
2822 		*p_txqueue_level = params.rate_limit.queue_level;
2823 	return (error);
2824 }
2825 
2826 /*
2827  * Allocate a new TX rate limit send tag from the network interface
2828  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
2829  */
2830 int
2831 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
2832     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
2833 {
2834 	union if_snd_tag_alloc_params params = {
2835 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
2836 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
2837 		.rate_limit.hdr.flowid = flowid,
2838 		.rate_limit.hdr.flowtype = flowtype,
2839 		.rate_limit.max_rate = max_pacing_rate,
2840 	};
2841 	int error;
2842 
2843 	INP_WLOCK_ASSERT(inp);
2844 
2845 	if (inp->inp_snd_tag != NULL)
2846 		return (EINVAL);
2847 
2848 	if (ifp->if_snd_tag_alloc == NULL) {
2849 		error = EOPNOTSUPP;
2850 	} else {
2851 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
2852 
2853 		/*
2854 		 * At success increment the refcount on
2855 		 * the send tag's network interface:
2856 		 */
2857 		if (error == 0)
2858 			if_ref(inp->inp_snd_tag->ifp);
2859 	}
2860 	return (error);
2861 }
2862 
2863 /*
2864  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
2865  * if any:
2866  */
2867 void
2868 in_pcbdetach_txrtlmt(struct inpcb *inp)
2869 {
2870 	struct m_snd_tag *mst;
2871 	struct ifnet *ifp;
2872 
2873 	INP_WLOCK_ASSERT(inp);
2874 
2875 	mst = inp->inp_snd_tag;
2876 	inp->inp_snd_tag = NULL;
2877 
2878 	if (mst == NULL)
2879 		return;
2880 
2881 	ifp = mst->ifp;
2882 	if (ifp == NULL)
2883 		return;
2884 
2885 	/*
2886 	 * If the device was detached while we still had reference(s)
2887 	 * on the ifp, we assume if_snd_tag_free() was replaced with
2888 	 * stubs.
2889 	 */
2890 	ifp->if_snd_tag_free(mst);
2891 
2892 	/* release reference count on network interface */
2893 	if_rele(ifp);
2894 }
2895 
2896 /*
2897  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
2898  * is set in the fast path and will attach/detach/modify the TX rate
2899  * limit send tag based on the socket's so_max_pacing_rate value.
2900  */
2901 void
2902 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
2903 {
2904 	struct socket *socket;
2905 	uint32_t max_pacing_rate;
2906 	bool did_upgrade;
2907 	int error;
2908 
2909 	if (inp == NULL)
2910 		return;
2911 
2912 	socket = inp->inp_socket;
2913 	if (socket == NULL)
2914 		return;
2915 
2916 	if (!INP_WLOCKED(inp)) {
2917 		/*
2918 		 * NOTE: If the write locking fails, we need to bail
2919 		 * out and use the non-ratelimited ring for the
2920 		 * transmit until there is a new chance to get the
2921 		 * write lock.
2922 		 */
2923 		if (!INP_TRY_UPGRADE(inp))
2924 			return;
2925 		did_upgrade = 1;
2926 	} else {
2927 		did_upgrade = 0;
2928 	}
2929 
2930 	/*
2931 	 * NOTE: The so_max_pacing_rate value is read unlocked,
2932 	 * because atomic updates are not required since the variable
2933 	 * is checked at every mbuf we send. It is assumed that the
2934 	 * variable read itself will be atomic.
2935 	 */
2936 	max_pacing_rate = socket->so_max_pacing_rate;
2937 
2938 	/*
2939 	 * NOTE: When attaching to a network interface a reference is
2940 	 * made to ensure the network interface doesn't go away until
2941 	 * all ratelimit connections are gone. The network interface
2942 	 * pointers compared below represent valid network interfaces,
2943 	 * except when comparing towards NULL.
2944 	 */
2945 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
2946 		error = 0;
2947 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
2948 		if (inp->inp_snd_tag != NULL)
2949 			in_pcbdetach_txrtlmt(inp);
2950 		error = 0;
2951 	} else if (inp->inp_snd_tag == NULL) {
2952 		/*
2953 		 * In order to utilize packet pacing with RSS, we need
2954 		 * to wait until there is a valid RSS hash before we
2955 		 * can proceed:
2956 		 */
2957 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
2958 			error = EAGAIN;
2959 		} else {
2960 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
2961 			    mb->m_pkthdr.flowid, max_pacing_rate);
2962 		}
2963 	} else {
2964 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
2965 	}
2966 	if (error == 0 || error == EOPNOTSUPP)
2967 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
2968 	if (did_upgrade)
2969 		INP_DOWNGRADE(inp);
2970 }
2971 
2972 /*
2973  * Track route changes for TX rate limiting.
2974  */
2975 void
2976 in_pcboutput_eagain(struct inpcb *inp)
2977 {
2978 	struct socket *socket;
2979 	bool did_upgrade;
2980 
2981 	if (inp == NULL)
2982 		return;
2983 
2984 	socket = inp->inp_socket;
2985 	if (socket == NULL)
2986 		return;
2987 
2988 	if (inp->inp_snd_tag == NULL)
2989 		return;
2990 
2991 	if (!INP_WLOCKED(inp)) {
2992 		/*
2993 		 * NOTE: If the write locking fails, we need to bail
2994 		 * out and use the non-ratelimited ring for the
2995 		 * transmit until there is a new chance to get the
2996 		 * write lock.
2997 		 */
2998 		if (!INP_TRY_UPGRADE(inp))
2999 			return;
3000 		did_upgrade = 1;
3001 	} else {
3002 		did_upgrade = 0;
3003 	}
3004 
3005 	/* detach rate limiting */
3006 	in_pcbdetach_txrtlmt(inp);
3007 
3008 	/* make sure new mbuf send tag allocation is made */
3009 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3010 
3011 	if (did_upgrade)
3012 		INP_DOWNGRADE(inp);
3013 }
3014 #endif /* RATELIMIT */
3015