xref: /freebsd/sys/netinet/in_pcb.c (revision 43faedc1339a9624c7acedb7f3e5624e64da5b99)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2007-2009 Robert N. M. Watson
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * All rights reserved.
7  *
8  * Portions of this software were developed by Robert N. M. Watson under
9  * contract to Juniper Networks, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ddb.h"
42 #include "opt_ipsec.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_pcbgroup.h"
46 #include "opt_rss.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/callout.h>
54 #include <sys/eventhandler.h>
55 #include <sys/domain.h>
56 #include <sys/protosw.h>
57 #include <sys/rmlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/refcount.h>
63 #include <sys/jail.h>
64 #include <sys/kernel.h>
65 #include <sys/sysctl.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #endif
70 
71 #include <vm/uma.h>
72 
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/if_types.h>
76 #include <net/route.h>
77 #include <net/rss_config.h>
78 #include <net/vnet.h>
79 
80 #if defined(INET) || defined(INET6)
81 #include <netinet/in.h>
82 #include <netinet/in_pcb.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/tcp_var.h>
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
87 #endif
88 #ifdef INET
89 #include <netinet/in_var.h>
90 #endif
91 #ifdef INET6
92 #include <netinet/ip6.h>
93 #include <netinet6/in6_pcb.h>
94 #include <netinet6/in6_var.h>
95 #include <netinet6/ip6_var.h>
96 #endif /* INET6 */
97 
98 
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/key.h>
102 #endif /* IPSEC */
103 
104 #include <security/mac/mac_framework.h>
105 
106 static struct callout	ipport_tick_callout;
107 
108 /*
109  * These configure the range of local port addresses assigned to
110  * "unspecified" outgoing connections/packets/whatever.
111  */
112 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
113 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
114 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
115 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
116 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
117 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
118 
119 /*
120  * Reserved ports accessible only to root. There are significant
121  * security considerations that must be accounted for when changing these,
122  * but the security benefits can be great. Please be careful.
123  */
124 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
125 VNET_DEFINE(int, ipport_reservedlow);
126 
127 /* Variables dealing with random ephemeral port allocation. */
128 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
129 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
130 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
131 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
132 VNET_DEFINE(int, ipport_tcpallocs);
133 static VNET_DEFINE(int, ipport_tcplastcount);
134 
135 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
136 
137 static void	in_pcbremlists(struct inpcb *inp);
138 #ifdef INET
139 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
140 			    struct in_addr faddr, u_int fport_arg,
141 			    struct in_addr laddr, u_int lport_arg,
142 			    int lookupflags, struct ifnet *ifp);
143 
144 #define RANGECHK(var, min, max) \
145 	if ((var) < (min)) { (var) = (min); } \
146 	else if ((var) > (max)) { (var) = (max); }
147 
148 static int
149 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
150 {
151 	int error;
152 
153 	error = sysctl_handle_int(oidp, arg1, arg2, req);
154 	if (error == 0) {
155 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
156 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
157 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
158 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
159 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
160 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
161 	}
162 	return (error);
163 }
164 
165 #undef RANGECHK
166 
167 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
168     "IP Ports");
169 
170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
171 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
172 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
174 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
175 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
177 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
178 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
180 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
181 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
183 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
184 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
186 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
187 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
188 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
189 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
190 	&VNET_NAME(ipport_reservedhigh), 0, "");
191 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
192 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
194 	CTLFLAG_VNET | CTLFLAG_RW,
195 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
197 	CTLFLAG_VNET | CTLFLAG_RW,
198 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
199 	"allocations before switching to a sequental one");
200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
201 	CTLFLAG_VNET | CTLFLAG_RW,
202 	&VNET_NAME(ipport_randomtime), 0,
203 	"Minimum time to keep sequental port "
204 	"allocation before switching to a random one");
205 #endif /* INET */
206 
207 /*
208  * in_pcb.c: manage the Protocol Control Blocks.
209  *
210  * NOTE: It is assumed that most of these functions will be called with
211  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
212  * functions often modify hash chains or addresses in pcbs.
213  */
214 
215 /*
216  * Initialize an inpcbinfo -- we should be able to reduce the number of
217  * arguments in time.
218  */
219 void
220 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
221     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
222     char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini,
223     uint32_t inpcbzone_flags, u_int hashfields)
224 {
225 
226 	INP_INFO_LOCK_INIT(pcbinfo, name);
227 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
228 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
229 #ifdef VIMAGE
230 	pcbinfo->ipi_vnet = curvnet;
231 #endif
232 	pcbinfo->ipi_listhead = listhead;
233 	LIST_INIT(pcbinfo->ipi_listhead);
234 	pcbinfo->ipi_count = 0;
235 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
236 	    &pcbinfo->ipi_hashmask);
237 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
238 	    &pcbinfo->ipi_porthashmask);
239 #ifdef PCBGROUP
240 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
241 #endif
242 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
243 	    NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR,
244 	    inpcbzone_flags);
245 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
246 	uma_zone_set_warning(pcbinfo->ipi_zone,
247 	    "kern.ipc.maxsockets limit reached");
248 }
249 
250 /*
251  * Destroy an inpcbinfo.
252  */
253 void
254 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
255 {
256 
257 	KASSERT(pcbinfo->ipi_count == 0,
258 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
259 
260 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
261 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
262 	    pcbinfo->ipi_porthashmask);
263 #ifdef PCBGROUP
264 	in_pcbgroup_destroy(pcbinfo);
265 #endif
266 	uma_zdestroy(pcbinfo->ipi_zone);
267 	INP_LIST_LOCK_DESTROY(pcbinfo);
268 	INP_HASH_LOCK_DESTROY(pcbinfo);
269 	INP_INFO_LOCK_DESTROY(pcbinfo);
270 }
271 
272 /*
273  * Allocate a PCB and associate it with the socket.
274  * On success return with the PCB locked.
275  */
276 int
277 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
278 {
279 	struct inpcb *inp;
280 	int error;
281 
282 #ifdef INVARIANTS
283 	if (pcbinfo == &V_tcbinfo) {
284 		INP_INFO_RLOCK_ASSERT(pcbinfo);
285 	} else {
286 		INP_INFO_WLOCK_ASSERT(pcbinfo);
287 	}
288 #endif
289 
290 	error = 0;
291 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
292 	if (inp == NULL)
293 		return (ENOBUFS);
294 	bzero(inp, inp_zero_size);
295 	inp->inp_pcbinfo = pcbinfo;
296 	inp->inp_socket = so;
297 	inp->inp_cred = crhold(so->so_cred);
298 	inp->inp_inc.inc_fibnum = so->so_fibnum;
299 #ifdef MAC
300 	error = mac_inpcb_init(inp, M_NOWAIT);
301 	if (error != 0)
302 		goto out;
303 	mac_inpcb_create(so, inp);
304 #endif
305 #ifdef IPSEC
306 	error = ipsec_init_policy(so, &inp->inp_sp);
307 	if (error != 0) {
308 #ifdef MAC
309 		mac_inpcb_destroy(inp);
310 #endif
311 		goto out;
312 	}
313 #endif /*IPSEC*/
314 #ifdef INET6
315 	if (INP_SOCKAF(so) == AF_INET6) {
316 		inp->inp_vflag |= INP_IPV6PROTO;
317 		if (V_ip6_v6only)
318 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
319 	}
320 #endif
321 	INP_WLOCK(inp);
322 	INP_LIST_WLOCK(pcbinfo);
323 	LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
324 	pcbinfo->ipi_count++;
325 	so->so_pcb = (caddr_t)inp;
326 #ifdef INET6
327 	if (V_ip6_auto_flowlabel)
328 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
329 #endif
330 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
331 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
332 	INP_LIST_WUNLOCK(pcbinfo);
333 #if defined(IPSEC) || defined(MAC)
334 out:
335 	if (error != 0) {
336 		crfree(inp->inp_cred);
337 		uma_zfree(pcbinfo->ipi_zone, inp);
338 	}
339 #endif
340 	return (error);
341 }
342 
343 #ifdef INET
344 int
345 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
346 {
347 	int anonport, error;
348 
349 	INP_WLOCK_ASSERT(inp);
350 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
351 
352 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
353 		return (EINVAL);
354 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
355 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
356 	    &inp->inp_lport, cred);
357 	if (error)
358 		return (error);
359 	if (in_pcbinshash(inp) != 0) {
360 		inp->inp_laddr.s_addr = INADDR_ANY;
361 		inp->inp_lport = 0;
362 		return (EAGAIN);
363 	}
364 	if (anonport)
365 		inp->inp_flags |= INP_ANONPORT;
366 	return (0);
367 }
368 #endif
369 
370 /*
371  * Select a local port (number) to use.
372  */
373 #if defined(INET) || defined(INET6)
374 int
375 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
376     struct ucred *cred, int lookupflags)
377 {
378 	struct inpcbinfo *pcbinfo;
379 	struct inpcb *tmpinp;
380 	unsigned short *lastport;
381 	int count, dorandom, error;
382 	u_short aux, first, last, lport;
383 #ifdef INET
384 	struct in_addr laddr;
385 #endif
386 
387 	pcbinfo = inp->inp_pcbinfo;
388 
389 	/*
390 	 * Because no actual state changes occur here, a global write lock on
391 	 * the pcbinfo isn't required.
392 	 */
393 	INP_LOCK_ASSERT(inp);
394 	INP_HASH_LOCK_ASSERT(pcbinfo);
395 
396 	if (inp->inp_flags & INP_HIGHPORT) {
397 		first = V_ipport_hifirstauto;	/* sysctl */
398 		last  = V_ipport_hilastauto;
399 		lastport = &pcbinfo->ipi_lasthi;
400 	} else if (inp->inp_flags & INP_LOWPORT) {
401 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
402 		if (error)
403 			return (error);
404 		first = V_ipport_lowfirstauto;	/* 1023 */
405 		last  = V_ipport_lowlastauto;	/* 600 */
406 		lastport = &pcbinfo->ipi_lastlow;
407 	} else {
408 		first = V_ipport_firstauto;	/* sysctl */
409 		last  = V_ipport_lastauto;
410 		lastport = &pcbinfo->ipi_lastport;
411 	}
412 	/*
413 	 * For UDP(-Lite), use random port allocation as long as the user
414 	 * allows it.  For TCP (and as of yet unknown) connections,
415 	 * use random port allocation only if the user allows it AND
416 	 * ipport_tick() allows it.
417 	 */
418 	if (V_ipport_randomized &&
419 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
420 		pcbinfo == &V_ulitecbinfo))
421 		dorandom = 1;
422 	else
423 		dorandom = 0;
424 	/*
425 	 * It makes no sense to do random port allocation if
426 	 * we have the only port available.
427 	 */
428 	if (first == last)
429 		dorandom = 0;
430 	/* Make sure to not include UDP(-Lite) packets in the count. */
431 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
432 		V_ipport_tcpallocs++;
433 	/*
434 	 * Instead of having two loops further down counting up or down
435 	 * make sure that first is always <= last and go with only one
436 	 * code path implementing all logic.
437 	 */
438 	if (first > last) {
439 		aux = first;
440 		first = last;
441 		last = aux;
442 	}
443 
444 #ifdef INET
445 	/* Make the compiler happy. */
446 	laddr.s_addr = 0;
447 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
448 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
449 		    __func__, inp));
450 		laddr = *laddrp;
451 	}
452 #endif
453 	tmpinp = NULL;	/* Make compiler happy. */
454 	lport = *lportp;
455 
456 	if (dorandom)
457 		*lastport = first + (arc4random() % (last - first));
458 
459 	count = last - first;
460 
461 	do {
462 		if (count-- < 0)	/* completely used? */
463 			return (EADDRNOTAVAIL);
464 		++*lastport;
465 		if (*lastport < first || *lastport > last)
466 			*lastport = first;
467 		lport = htons(*lastport);
468 
469 #ifdef INET6
470 		if ((inp->inp_vflag & INP_IPV6) != 0)
471 			tmpinp = in6_pcblookup_local(pcbinfo,
472 			    &inp->in6p_laddr, lport, lookupflags, cred);
473 #endif
474 #if defined(INET) && defined(INET6)
475 		else
476 #endif
477 #ifdef INET
478 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
479 			    lport, lookupflags, cred);
480 #endif
481 	} while (tmpinp != NULL);
482 
483 #ifdef INET
484 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
485 		laddrp->s_addr = laddr.s_addr;
486 #endif
487 	*lportp = lport;
488 
489 	return (0);
490 }
491 
492 /*
493  * Return cached socket options.
494  */
495 short
496 inp_so_options(const struct inpcb *inp)
497 {
498    short so_options;
499 
500    so_options = 0;
501 
502    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
503 	   so_options |= SO_REUSEPORT;
504    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
505 	   so_options |= SO_REUSEADDR;
506    return (so_options);
507 }
508 #endif /* INET || INET6 */
509 
510 /*
511  * Check if a new BINDMULTI socket is allowed to be created.
512  *
513  * ni points to the new inp.
514  * oi points to the exisitng inp.
515  *
516  * This checks whether the existing inp also has BINDMULTI and
517  * whether the credentials match.
518  */
519 int
520 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
521 {
522 	/* Check permissions match */
523 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
524 	    (ni->inp_cred->cr_uid !=
525 	    oi->inp_cred->cr_uid))
526 		return (0);
527 
528 	/* Check the existing inp has BINDMULTI set */
529 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
530 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
531 		return (0);
532 
533 	/*
534 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
535 	 * it is and it matches the checks.
536 	 */
537 	return (1);
538 }
539 
540 #ifdef INET
541 /*
542  * Set up a bind operation on a PCB, performing port allocation
543  * as required, but do not actually modify the PCB. Callers can
544  * either complete the bind by setting inp_laddr/inp_lport and
545  * calling in_pcbinshash(), or they can just use the resulting
546  * port and address to authorise the sending of a once-off packet.
547  *
548  * On error, the values of *laddrp and *lportp are not changed.
549  */
550 int
551 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
552     u_short *lportp, struct ucred *cred)
553 {
554 	struct socket *so = inp->inp_socket;
555 	struct sockaddr_in *sin;
556 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
557 	struct in_addr laddr;
558 	u_short lport = 0;
559 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
560 	int error;
561 
562 	/*
563 	 * No state changes, so read locks are sufficient here.
564 	 */
565 	INP_LOCK_ASSERT(inp);
566 	INP_HASH_LOCK_ASSERT(pcbinfo);
567 
568 	if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
569 		return (EADDRNOTAVAIL);
570 	laddr.s_addr = *laddrp;
571 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
572 		return (EINVAL);
573 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
574 		lookupflags = INPLOOKUP_WILDCARD;
575 	if (nam == NULL) {
576 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
577 			return (error);
578 	} else {
579 		sin = (struct sockaddr_in *)nam;
580 		if (nam->sa_len != sizeof (*sin))
581 			return (EINVAL);
582 #ifdef notdef
583 		/*
584 		 * We should check the family, but old programs
585 		 * incorrectly fail to initialize it.
586 		 */
587 		if (sin->sin_family != AF_INET)
588 			return (EAFNOSUPPORT);
589 #endif
590 		error = prison_local_ip4(cred, &sin->sin_addr);
591 		if (error)
592 			return (error);
593 		if (sin->sin_port != *lportp) {
594 			/* Don't allow the port to change. */
595 			if (*lportp != 0)
596 				return (EINVAL);
597 			lport = sin->sin_port;
598 		}
599 		/* NB: lport is left as 0 if the port isn't being changed. */
600 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
601 			/*
602 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
603 			 * allow complete duplication of binding if
604 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
605 			 * and a multicast address is bound on both
606 			 * new and duplicated sockets.
607 			 */
608 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
609 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
610 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
611 			sin->sin_port = 0;		/* yech... */
612 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
613 			/*
614 			 * Is the address a local IP address?
615 			 * If INP_BINDANY is set, then the socket may be bound
616 			 * to any endpoint address, local or not.
617 			 */
618 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
619 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
620 				return (EADDRNOTAVAIL);
621 		}
622 		laddr = sin->sin_addr;
623 		if (lport) {
624 			struct inpcb *t;
625 			struct tcptw *tw;
626 
627 			/* GROSS */
628 			if (ntohs(lport) <= V_ipport_reservedhigh &&
629 			    ntohs(lport) >= V_ipport_reservedlow &&
630 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
631 			    0))
632 				return (EACCES);
633 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
634 			    priv_check_cred(inp->inp_cred,
635 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
636 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
637 				    lport, INPLOOKUP_WILDCARD, cred);
638 	/*
639 	 * XXX
640 	 * This entire block sorely needs a rewrite.
641 	 */
642 				if (t &&
643 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
644 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
645 				    (so->so_type != SOCK_STREAM ||
646 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
647 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
648 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
649 				     (t->inp_flags2 & INP_REUSEPORT) == 0) &&
650 				    (inp->inp_cred->cr_uid !=
651 				     t->inp_cred->cr_uid))
652 					return (EADDRINUSE);
653 
654 				/*
655 				 * If the socket is a BINDMULTI socket, then
656 				 * the credentials need to match and the
657 				 * original socket also has to have been bound
658 				 * with BINDMULTI.
659 				 */
660 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
661 					return (EADDRINUSE);
662 			}
663 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
664 			    lport, lookupflags, cred);
665 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
666 				/*
667 				 * XXXRW: If an incpb has had its timewait
668 				 * state recycled, we treat the address as
669 				 * being in use (for now).  This is better
670 				 * than a panic, but not desirable.
671 				 */
672 				tw = intotw(t);
673 				if (tw == NULL ||
674 				    (reuseport & tw->tw_so_options) == 0)
675 					return (EADDRINUSE);
676 			} else if (t &&
677 			    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
678 			    (reuseport & inp_so_options(t)) == 0) {
679 #ifdef INET6
680 				if (ntohl(sin->sin_addr.s_addr) !=
681 				    INADDR_ANY ||
682 				    ntohl(t->inp_laddr.s_addr) !=
683 				    INADDR_ANY ||
684 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
685 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
686 #endif
687 				return (EADDRINUSE);
688 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
689 					return (EADDRINUSE);
690 			}
691 		}
692 	}
693 	if (*lportp != 0)
694 		lport = *lportp;
695 	if (lport == 0) {
696 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
697 		if (error != 0)
698 			return (error);
699 
700 	}
701 	*laddrp = laddr.s_addr;
702 	*lportp = lport;
703 	return (0);
704 }
705 
706 /*
707  * Connect from a socket to a specified address.
708  * Both address and port must be specified in argument sin.
709  * If don't have a local address for this socket yet,
710  * then pick one.
711  */
712 int
713 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
714     struct ucred *cred, struct mbuf *m)
715 {
716 	u_short lport, fport;
717 	in_addr_t laddr, faddr;
718 	int anonport, error;
719 
720 	INP_WLOCK_ASSERT(inp);
721 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
722 
723 	lport = inp->inp_lport;
724 	laddr = inp->inp_laddr.s_addr;
725 	anonport = (lport == 0);
726 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
727 	    NULL, cred);
728 	if (error)
729 		return (error);
730 
731 	/* Do the initial binding of the local address if required. */
732 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
733 		inp->inp_lport = lport;
734 		inp->inp_laddr.s_addr = laddr;
735 		if (in_pcbinshash(inp) != 0) {
736 			inp->inp_laddr.s_addr = INADDR_ANY;
737 			inp->inp_lport = 0;
738 			return (EAGAIN);
739 		}
740 	}
741 
742 	/* Commit the remaining changes. */
743 	inp->inp_lport = lport;
744 	inp->inp_laddr.s_addr = laddr;
745 	inp->inp_faddr.s_addr = faddr;
746 	inp->inp_fport = fport;
747 	in_pcbrehash_mbuf(inp, m);
748 
749 	if (anonport)
750 		inp->inp_flags |= INP_ANONPORT;
751 	return (0);
752 }
753 
754 int
755 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
756 {
757 
758 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
759 }
760 
761 /*
762  * Do proper source address selection on an unbound socket in case
763  * of connect. Take jails into account as well.
764  */
765 int
766 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
767     struct ucred *cred)
768 {
769 	struct ifaddr *ifa;
770 	struct sockaddr *sa;
771 	struct sockaddr_in *sin;
772 	struct route sro;
773 	int error;
774 
775 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
776 
777 	/*
778 	 * Bypass source address selection and use the primary jail IP
779 	 * if requested.
780 	 */
781 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
782 		return (0);
783 
784 	error = 0;
785 	bzero(&sro, sizeof(sro));
786 
787 	sin = (struct sockaddr_in *)&sro.ro_dst;
788 	sin->sin_family = AF_INET;
789 	sin->sin_len = sizeof(struct sockaddr_in);
790 	sin->sin_addr.s_addr = faddr->s_addr;
791 
792 	/*
793 	 * If route is known our src addr is taken from the i/f,
794 	 * else punt.
795 	 *
796 	 * Find out route to destination.
797 	 */
798 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
799 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
800 
801 	/*
802 	 * If we found a route, use the address corresponding to
803 	 * the outgoing interface.
804 	 *
805 	 * Otherwise assume faddr is reachable on a directly connected
806 	 * network and try to find a corresponding interface to take
807 	 * the source address from.
808 	 */
809 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
810 		struct in_ifaddr *ia;
811 		struct ifnet *ifp;
812 
813 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
814 					inp->inp_socket->so_fibnum));
815 		if (ia == NULL)
816 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
817 						inp->inp_socket->so_fibnum));
818 		if (ia == NULL) {
819 			error = ENETUNREACH;
820 			goto done;
821 		}
822 
823 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
824 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
825 			ifa_free(&ia->ia_ifa);
826 			goto done;
827 		}
828 
829 		ifp = ia->ia_ifp;
830 		ifa_free(&ia->ia_ifa);
831 		ia = NULL;
832 		IF_ADDR_RLOCK(ifp);
833 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
834 
835 			sa = ifa->ifa_addr;
836 			if (sa->sa_family != AF_INET)
837 				continue;
838 			sin = (struct sockaddr_in *)sa;
839 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
840 				ia = (struct in_ifaddr *)ifa;
841 				break;
842 			}
843 		}
844 		if (ia != NULL) {
845 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
846 			IF_ADDR_RUNLOCK(ifp);
847 			goto done;
848 		}
849 		IF_ADDR_RUNLOCK(ifp);
850 
851 		/* 3. As a last resort return the 'default' jail address. */
852 		error = prison_get_ip4(cred, laddr);
853 		goto done;
854 	}
855 
856 	/*
857 	 * If the outgoing interface on the route found is not
858 	 * a loopback interface, use the address from that interface.
859 	 * In case of jails do those three steps:
860 	 * 1. check if the interface address belongs to the jail. If so use it.
861 	 * 2. check if we have any address on the outgoing interface
862 	 *    belonging to this jail. If so use it.
863 	 * 3. as a last resort return the 'default' jail address.
864 	 */
865 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
866 		struct in_ifaddr *ia;
867 		struct ifnet *ifp;
868 
869 		/* If not jailed, use the default returned. */
870 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
871 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
872 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
873 			goto done;
874 		}
875 
876 		/* Jailed. */
877 		/* 1. Check if the iface address belongs to the jail. */
878 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
879 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
880 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
881 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
882 			goto done;
883 		}
884 
885 		/*
886 		 * 2. Check if we have any address on the outgoing interface
887 		 *    belonging to this jail.
888 		 */
889 		ia = NULL;
890 		ifp = sro.ro_rt->rt_ifp;
891 		IF_ADDR_RLOCK(ifp);
892 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
893 			sa = ifa->ifa_addr;
894 			if (sa->sa_family != AF_INET)
895 				continue;
896 			sin = (struct sockaddr_in *)sa;
897 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
898 				ia = (struct in_ifaddr *)ifa;
899 				break;
900 			}
901 		}
902 		if (ia != NULL) {
903 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
904 			IF_ADDR_RUNLOCK(ifp);
905 			goto done;
906 		}
907 		IF_ADDR_RUNLOCK(ifp);
908 
909 		/* 3. As a last resort return the 'default' jail address. */
910 		error = prison_get_ip4(cred, laddr);
911 		goto done;
912 	}
913 
914 	/*
915 	 * The outgoing interface is marked with 'loopback net', so a route
916 	 * to ourselves is here.
917 	 * Try to find the interface of the destination address and then
918 	 * take the address from there. That interface is not necessarily
919 	 * a loopback interface.
920 	 * In case of jails, check that it is an address of the jail
921 	 * and if we cannot find, fall back to the 'default' jail address.
922 	 */
923 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
924 		struct sockaddr_in sain;
925 		struct in_ifaddr *ia;
926 
927 		bzero(&sain, sizeof(struct sockaddr_in));
928 		sain.sin_family = AF_INET;
929 		sain.sin_len = sizeof(struct sockaddr_in);
930 		sain.sin_addr.s_addr = faddr->s_addr;
931 
932 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
933 					inp->inp_socket->so_fibnum));
934 		if (ia == NULL)
935 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
936 						inp->inp_socket->so_fibnum));
937 		if (ia == NULL)
938 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
939 
940 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
941 			if (ia == NULL) {
942 				error = ENETUNREACH;
943 				goto done;
944 			}
945 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
946 			ifa_free(&ia->ia_ifa);
947 			goto done;
948 		}
949 
950 		/* Jailed. */
951 		if (ia != NULL) {
952 			struct ifnet *ifp;
953 
954 			ifp = ia->ia_ifp;
955 			ifa_free(&ia->ia_ifa);
956 			ia = NULL;
957 			IF_ADDR_RLOCK(ifp);
958 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
959 
960 				sa = ifa->ifa_addr;
961 				if (sa->sa_family != AF_INET)
962 					continue;
963 				sin = (struct sockaddr_in *)sa;
964 				if (prison_check_ip4(cred,
965 				    &sin->sin_addr) == 0) {
966 					ia = (struct in_ifaddr *)ifa;
967 					break;
968 				}
969 			}
970 			if (ia != NULL) {
971 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
972 				IF_ADDR_RUNLOCK(ifp);
973 				goto done;
974 			}
975 			IF_ADDR_RUNLOCK(ifp);
976 		}
977 
978 		/* 3. As a last resort return the 'default' jail address. */
979 		error = prison_get_ip4(cred, laddr);
980 		goto done;
981 	}
982 
983 done:
984 	if (sro.ro_rt != NULL)
985 		RTFREE(sro.ro_rt);
986 	return (error);
987 }
988 
989 /*
990  * Set up for a connect from a socket to the specified address.
991  * On entry, *laddrp and *lportp should contain the current local
992  * address and port for the PCB; these are updated to the values
993  * that should be placed in inp_laddr and inp_lport to complete
994  * the connect.
995  *
996  * On success, *faddrp and *fportp will be set to the remote address
997  * and port. These are not updated in the error case.
998  *
999  * If the operation fails because the connection already exists,
1000  * *oinpp will be set to the PCB of that connection so that the
1001  * caller can decide to override it. In all other cases, *oinpp
1002  * is set to NULL.
1003  */
1004 int
1005 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1006     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1007     struct inpcb **oinpp, struct ucred *cred)
1008 {
1009 	struct rm_priotracker in_ifa_tracker;
1010 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1011 	struct in_ifaddr *ia;
1012 	struct inpcb *oinp;
1013 	struct in_addr laddr, faddr;
1014 	u_short lport, fport;
1015 	int error;
1016 
1017 	/*
1018 	 * Because a global state change doesn't actually occur here, a read
1019 	 * lock is sufficient.
1020 	 */
1021 	INP_LOCK_ASSERT(inp);
1022 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1023 
1024 	if (oinpp != NULL)
1025 		*oinpp = NULL;
1026 	if (nam->sa_len != sizeof (*sin))
1027 		return (EINVAL);
1028 	if (sin->sin_family != AF_INET)
1029 		return (EAFNOSUPPORT);
1030 	if (sin->sin_port == 0)
1031 		return (EADDRNOTAVAIL);
1032 	laddr.s_addr = *laddrp;
1033 	lport = *lportp;
1034 	faddr = sin->sin_addr;
1035 	fport = sin->sin_port;
1036 
1037 	if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1038 		/*
1039 		 * If the destination address is INADDR_ANY,
1040 		 * use the primary local address.
1041 		 * If the supplied address is INADDR_BROADCAST,
1042 		 * and the primary interface supports broadcast,
1043 		 * choose the broadcast address for that interface.
1044 		 */
1045 		if (faddr.s_addr == INADDR_ANY) {
1046 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1047 			faddr =
1048 			    IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1049 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1050 			if (cred != NULL &&
1051 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1052 				return (error);
1053 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1054 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1055 			if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1056 			    IFF_BROADCAST)
1057 				faddr = satosin(&TAILQ_FIRST(
1058 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1059 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1060 		}
1061 	}
1062 	if (laddr.s_addr == INADDR_ANY) {
1063 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1064 		/*
1065 		 * If the destination address is multicast and an outgoing
1066 		 * interface has been set as a multicast option, prefer the
1067 		 * address of that interface as our source address.
1068 		 */
1069 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1070 		    inp->inp_moptions != NULL) {
1071 			struct ip_moptions *imo;
1072 			struct ifnet *ifp;
1073 
1074 			imo = inp->inp_moptions;
1075 			if (imo->imo_multicast_ifp != NULL) {
1076 				ifp = imo->imo_multicast_ifp;
1077 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1078 				TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1079 					if ((ia->ia_ifp == ifp) &&
1080 					    (cred == NULL ||
1081 					    prison_check_ip4(cred,
1082 					    &ia->ia_addr.sin_addr) == 0))
1083 						break;
1084 				}
1085 				if (ia == NULL)
1086 					error = EADDRNOTAVAIL;
1087 				else {
1088 					laddr = ia->ia_addr.sin_addr;
1089 					error = 0;
1090 				}
1091 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1092 			}
1093 		}
1094 		if (error)
1095 			return (error);
1096 	}
1097 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1098 	    laddr, lport, 0, NULL);
1099 	if (oinp != NULL) {
1100 		if (oinpp != NULL)
1101 			*oinpp = oinp;
1102 		return (EADDRINUSE);
1103 	}
1104 	if (lport == 0) {
1105 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1106 		    cred);
1107 		if (error)
1108 			return (error);
1109 	}
1110 	*laddrp = laddr.s_addr;
1111 	*lportp = lport;
1112 	*faddrp = faddr.s_addr;
1113 	*fportp = fport;
1114 	return (0);
1115 }
1116 
1117 void
1118 in_pcbdisconnect(struct inpcb *inp)
1119 {
1120 
1121 	INP_WLOCK_ASSERT(inp);
1122 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1123 
1124 	inp->inp_faddr.s_addr = INADDR_ANY;
1125 	inp->inp_fport = 0;
1126 	in_pcbrehash(inp);
1127 }
1128 #endif /* INET */
1129 
1130 /*
1131  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1132  * For most protocols, this will be invoked immediately prior to calling
1133  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1134  * socket, in which case in_pcbfree() is deferred.
1135  */
1136 void
1137 in_pcbdetach(struct inpcb *inp)
1138 {
1139 
1140 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1141 
1142 	inp->inp_socket->so_pcb = NULL;
1143 	inp->inp_socket = NULL;
1144 }
1145 
1146 /*
1147  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1148  * stability of an inpcb pointer despite the inpcb lock being released.  This
1149  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1150  * but where the inpcb lock may already held, or when acquiring a reference
1151  * via a pcbgroup.
1152  *
1153  * in_pcbref() should be used only to provide brief memory stability, and
1154  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1155  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1156  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1157  * lock and rele are the *only* safe operations that may be performed on the
1158  * inpcb.
1159  *
1160  * While the inpcb will not be freed, releasing the inpcb lock means that the
1161  * connection's state may change, so the caller should be careful to
1162  * revalidate any cached state on reacquiring the lock.  Drop the reference
1163  * using in_pcbrele().
1164  */
1165 void
1166 in_pcbref(struct inpcb *inp)
1167 {
1168 
1169 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1170 
1171 	refcount_acquire(&inp->inp_refcount);
1172 }
1173 
1174 /*
1175  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1176  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1177  * return a flag indicating whether or not the inpcb remains valid.  If it is
1178  * valid, we return with the inpcb lock held.
1179  *
1180  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1181  * reference on an inpcb.  Historically more work was done here (actually, in
1182  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1183  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1184  * about memory stability (and continued use of the write lock).
1185  */
1186 int
1187 in_pcbrele_rlocked(struct inpcb *inp)
1188 {
1189 	struct inpcbinfo *pcbinfo;
1190 
1191 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1192 
1193 	INP_RLOCK_ASSERT(inp);
1194 
1195 	if (refcount_release(&inp->inp_refcount) == 0) {
1196 		/*
1197 		 * If the inpcb has been freed, let the caller know, even if
1198 		 * this isn't the last reference.
1199 		 */
1200 		if (inp->inp_flags2 & INP_FREED) {
1201 			INP_RUNLOCK(inp);
1202 			return (1);
1203 		}
1204 		return (0);
1205 	}
1206 
1207 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1208 
1209 	INP_RUNLOCK(inp);
1210 	pcbinfo = inp->inp_pcbinfo;
1211 	uma_zfree(pcbinfo->ipi_zone, inp);
1212 	return (1);
1213 }
1214 
1215 int
1216 in_pcbrele_wlocked(struct inpcb *inp)
1217 {
1218 	struct inpcbinfo *pcbinfo;
1219 
1220 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1221 
1222 	INP_WLOCK_ASSERT(inp);
1223 
1224 	if (refcount_release(&inp->inp_refcount) == 0) {
1225 		/*
1226 		 * If the inpcb has been freed, let the caller know, even if
1227 		 * this isn't the last reference.
1228 		 */
1229 		if (inp->inp_flags2 & INP_FREED) {
1230 			INP_WUNLOCK(inp);
1231 			return (1);
1232 		}
1233 		return (0);
1234 	}
1235 
1236 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1237 
1238 	INP_WUNLOCK(inp);
1239 	pcbinfo = inp->inp_pcbinfo;
1240 	uma_zfree(pcbinfo->ipi_zone, inp);
1241 	return (1);
1242 }
1243 
1244 /*
1245  * Temporary wrapper.
1246  */
1247 int
1248 in_pcbrele(struct inpcb *inp)
1249 {
1250 
1251 	return (in_pcbrele_wlocked(inp));
1252 }
1253 
1254 /*
1255  * Unconditionally schedule an inpcb to be freed by decrementing its
1256  * reference count, which should occur only after the inpcb has been detached
1257  * from its socket.  If another thread holds a temporary reference (acquired
1258  * using in_pcbref()) then the free is deferred until that reference is
1259  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1260  * work, including removal from global lists, is done in this context, where
1261  * the pcbinfo lock is held.
1262  */
1263 void
1264 in_pcbfree(struct inpcb *inp)
1265 {
1266 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1267 
1268 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1269 
1270 #ifdef INVARIANTS
1271 	if (pcbinfo == &V_tcbinfo) {
1272 		INP_INFO_LOCK_ASSERT(pcbinfo);
1273 	} else {
1274 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1275 	}
1276 #endif
1277 	INP_WLOCK_ASSERT(inp);
1278 
1279 	/* XXXRW: Do as much as possible here. */
1280 #ifdef IPSEC
1281 	if (inp->inp_sp != NULL)
1282 		ipsec_delete_pcbpolicy(inp);
1283 #endif
1284 	INP_LIST_WLOCK(pcbinfo);
1285 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1286 	in_pcbremlists(inp);
1287 	INP_LIST_WUNLOCK(pcbinfo);
1288 #ifdef INET6
1289 	if (inp->inp_vflag & INP_IPV6PROTO) {
1290 		ip6_freepcbopts(inp->in6p_outputopts);
1291 		if (inp->in6p_moptions != NULL)
1292 			ip6_freemoptions(inp->in6p_moptions);
1293 	}
1294 #endif
1295 	if (inp->inp_options)
1296 		(void)m_free(inp->inp_options);
1297 #ifdef INET
1298 	if (inp->inp_moptions != NULL)
1299 		inp_freemoptions(inp->inp_moptions);
1300 #endif
1301 	inp->inp_vflag = 0;
1302 	inp->inp_flags2 |= INP_FREED;
1303 	crfree(inp->inp_cred);
1304 #ifdef MAC
1305 	mac_inpcb_destroy(inp);
1306 #endif
1307 	if (!in_pcbrele_wlocked(inp))
1308 		INP_WUNLOCK(inp);
1309 }
1310 
1311 /*
1312  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1313  * port reservation, and preventing it from being returned by inpcb lookups.
1314  *
1315  * It is used by TCP to mark an inpcb as unused and avoid future packet
1316  * delivery or event notification when a socket remains open but TCP has
1317  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1318  * or a RST on the wire, and allows the port binding to be reused while still
1319  * maintaining the invariant that so_pcb always points to a valid inpcb until
1320  * in_pcbdetach().
1321  *
1322  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1323  * in_pcbnotifyall() and in_pcbpurgeif0()?
1324  */
1325 void
1326 in_pcbdrop(struct inpcb *inp)
1327 {
1328 
1329 	INP_WLOCK_ASSERT(inp);
1330 
1331 	/*
1332 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1333 	 * the hash lock...?
1334 	 */
1335 	inp->inp_flags |= INP_DROPPED;
1336 	if (inp->inp_flags & INP_INHASHLIST) {
1337 		struct inpcbport *phd = inp->inp_phd;
1338 
1339 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1340 		LIST_REMOVE(inp, inp_hash);
1341 		LIST_REMOVE(inp, inp_portlist);
1342 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1343 			LIST_REMOVE(phd, phd_hash);
1344 			free(phd, M_PCB);
1345 		}
1346 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1347 		inp->inp_flags &= ~INP_INHASHLIST;
1348 #ifdef PCBGROUP
1349 		in_pcbgroup_remove(inp);
1350 #endif
1351 	}
1352 }
1353 
1354 #ifdef INET
1355 /*
1356  * Common routines to return the socket addresses associated with inpcbs.
1357  */
1358 struct sockaddr *
1359 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1360 {
1361 	struct sockaddr_in *sin;
1362 
1363 	sin = malloc(sizeof *sin, M_SONAME,
1364 		M_WAITOK | M_ZERO);
1365 	sin->sin_family = AF_INET;
1366 	sin->sin_len = sizeof(*sin);
1367 	sin->sin_addr = *addr_p;
1368 	sin->sin_port = port;
1369 
1370 	return (struct sockaddr *)sin;
1371 }
1372 
1373 int
1374 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1375 {
1376 	struct inpcb *inp;
1377 	struct in_addr addr;
1378 	in_port_t port;
1379 
1380 	inp = sotoinpcb(so);
1381 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1382 
1383 	INP_RLOCK(inp);
1384 	port = inp->inp_lport;
1385 	addr = inp->inp_laddr;
1386 	INP_RUNLOCK(inp);
1387 
1388 	*nam = in_sockaddr(port, &addr);
1389 	return 0;
1390 }
1391 
1392 int
1393 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1394 {
1395 	struct inpcb *inp;
1396 	struct in_addr addr;
1397 	in_port_t port;
1398 
1399 	inp = sotoinpcb(so);
1400 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1401 
1402 	INP_RLOCK(inp);
1403 	port = inp->inp_fport;
1404 	addr = inp->inp_faddr;
1405 	INP_RUNLOCK(inp);
1406 
1407 	*nam = in_sockaddr(port, &addr);
1408 	return 0;
1409 }
1410 
1411 void
1412 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1413     struct inpcb *(*notify)(struct inpcb *, int))
1414 {
1415 	struct inpcb *inp, *inp_temp;
1416 
1417 	INP_INFO_WLOCK(pcbinfo);
1418 	LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1419 		INP_WLOCK(inp);
1420 #ifdef INET6
1421 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1422 			INP_WUNLOCK(inp);
1423 			continue;
1424 		}
1425 #endif
1426 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1427 		    inp->inp_socket == NULL) {
1428 			INP_WUNLOCK(inp);
1429 			continue;
1430 		}
1431 		if ((*notify)(inp, errno))
1432 			INP_WUNLOCK(inp);
1433 	}
1434 	INP_INFO_WUNLOCK(pcbinfo);
1435 }
1436 
1437 void
1438 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1439 {
1440 	struct inpcb *inp;
1441 	struct ip_moptions *imo;
1442 	int i, gap;
1443 
1444 	INP_INFO_WLOCK(pcbinfo);
1445 	LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1446 		INP_WLOCK(inp);
1447 		imo = inp->inp_moptions;
1448 		if ((inp->inp_vflag & INP_IPV4) &&
1449 		    imo != NULL) {
1450 			/*
1451 			 * Unselect the outgoing interface if it is being
1452 			 * detached.
1453 			 */
1454 			if (imo->imo_multicast_ifp == ifp)
1455 				imo->imo_multicast_ifp = NULL;
1456 
1457 			/*
1458 			 * Drop multicast group membership if we joined
1459 			 * through the interface being detached.
1460 			 */
1461 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1462 			    i++) {
1463 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1464 					in_delmulti(imo->imo_membership[i]);
1465 					gap++;
1466 				} else if (gap != 0)
1467 					imo->imo_membership[i - gap] =
1468 					    imo->imo_membership[i];
1469 			}
1470 			imo->imo_num_memberships -= gap;
1471 		}
1472 		INP_WUNLOCK(inp);
1473 	}
1474 	INP_INFO_WUNLOCK(pcbinfo);
1475 }
1476 
1477 /*
1478  * Lookup a PCB based on the local address and port.  Caller must hold the
1479  * hash lock.  No inpcb locks or references are acquired.
1480  */
1481 #define INP_LOOKUP_MAPPED_PCB_COST	3
1482 struct inpcb *
1483 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1484     u_short lport, int lookupflags, struct ucred *cred)
1485 {
1486 	struct inpcb *inp;
1487 #ifdef INET6
1488 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1489 #else
1490 	int matchwild = 3;
1491 #endif
1492 	int wildcard;
1493 
1494 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1495 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1496 
1497 	INP_HASH_LOCK_ASSERT(pcbinfo);
1498 
1499 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1500 		struct inpcbhead *head;
1501 		/*
1502 		 * Look for an unconnected (wildcard foreign addr) PCB that
1503 		 * matches the local address and port we're looking for.
1504 		 */
1505 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1506 		    0, pcbinfo->ipi_hashmask)];
1507 		LIST_FOREACH(inp, head, inp_hash) {
1508 #ifdef INET6
1509 			/* XXX inp locking */
1510 			if ((inp->inp_vflag & INP_IPV4) == 0)
1511 				continue;
1512 #endif
1513 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1514 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1515 			    inp->inp_lport == lport) {
1516 				/*
1517 				 * Found?
1518 				 */
1519 				if (cred == NULL ||
1520 				    prison_equal_ip4(cred->cr_prison,
1521 					inp->inp_cred->cr_prison))
1522 					return (inp);
1523 			}
1524 		}
1525 		/*
1526 		 * Not found.
1527 		 */
1528 		return (NULL);
1529 	} else {
1530 		struct inpcbporthead *porthash;
1531 		struct inpcbport *phd;
1532 		struct inpcb *match = NULL;
1533 		/*
1534 		 * Best fit PCB lookup.
1535 		 *
1536 		 * First see if this local port is in use by looking on the
1537 		 * port hash list.
1538 		 */
1539 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1540 		    pcbinfo->ipi_porthashmask)];
1541 		LIST_FOREACH(phd, porthash, phd_hash) {
1542 			if (phd->phd_port == lport)
1543 				break;
1544 		}
1545 		if (phd != NULL) {
1546 			/*
1547 			 * Port is in use by one or more PCBs. Look for best
1548 			 * fit.
1549 			 */
1550 			LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1551 				wildcard = 0;
1552 				if (cred != NULL &&
1553 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1554 					cred->cr_prison))
1555 					continue;
1556 #ifdef INET6
1557 				/* XXX inp locking */
1558 				if ((inp->inp_vflag & INP_IPV4) == 0)
1559 					continue;
1560 				/*
1561 				 * We never select the PCB that has
1562 				 * INP_IPV6 flag and is bound to :: if
1563 				 * we have another PCB which is bound
1564 				 * to 0.0.0.0.  If a PCB has the
1565 				 * INP_IPV6 flag, then we set its cost
1566 				 * higher than IPv4 only PCBs.
1567 				 *
1568 				 * Note that the case only happens
1569 				 * when a socket is bound to ::, under
1570 				 * the condition that the use of the
1571 				 * mapped address is allowed.
1572 				 */
1573 				if ((inp->inp_vflag & INP_IPV6) != 0)
1574 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1575 #endif
1576 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1577 					wildcard++;
1578 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1579 					if (laddr.s_addr == INADDR_ANY)
1580 						wildcard++;
1581 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1582 						continue;
1583 				} else {
1584 					if (laddr.s_addr != INADDR_ANY)
1585 						wildcard++;
1586 				}
1587 				if (wildcard < matchwild) {
1588 					match = inp;
1589 					matchwild = wildcard;
1590 					if (matchwild == 0)
1591 						break;
1592 				}
1593 			}
1594 		}
1595 		return (match);
1596 	}
1597 }
1598 #undef INP_LOOKUP_MAPPED_PCB_COST
1599 
1600 #ifdef PCBGROUP
1601 /*
1602  * Lookup PCB in hash list, using pcbgroup tables.
1603  */
1604 static struct inpcb *
1605 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1606     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1607     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1608 {
1609 	struct inpcbhead *head;
1610 	struct inpcb *inp, *tmpinp;
1611 	u_short fport = fport_arg, lport = lport_arg;
1612 
1613 	/*
1614 	 * First look for an exact match.
1615 	 */
1616 	tmpinp = NULL;
1617 	INP_GROUP_LOCK(pcbgroup);
1618 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1619 	    pcbgroup->ipg_hashmask)];
1620 	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1621 #ifdef INET6
1622 		/* XXX inp locking */
1623 		if ((inp->inp_vflag & INP_IPV4) == 0)
1624 			continue;
1625 #endif
1626 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1627 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1628 		    inp->inp_fport == fport &&
1629 		    inp->inp_lport == lport) {
1630 			/*
1631 			 * XXX We should be able to directly return
1632 			 * the inp here, without any checks.
1633 			 * Well unless both bound with SO_REUSEPORT?
1634 			 */
1635 			if (prison_flag(inp->inp_cred, PR_IP4))
1636 				goto found;
1637 			if (tmpinp == NULL)
1638 				tmpinp = inp;
1639 		}
1640 	}
1641 	if (tmpinp != NULL) {
1642 		inp = tmpinp;
1643 		goto found;
1644 	}
1645 
1646 #ifdef	RSS
1647 	/*
1648 	 * For incoming connections, we may wish to do a wildcard
1649 	 * match for an RSS-local socket.
1650 	 */
1651 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1652 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1653 #ifdef INET6
1654 		struct inpcb *local_wild_mapped = NULL;
1655 #endif
1656 		struct inpcb *jail_wild = NULL;
1657 		struct inpcbhead *head;
1658 		int injail;
1659 
1660 		/*
1661 		 * Order of socket selection - we always prefer jails.
1662 		 *      1. jailed, non-wild.
1663 		 *      2. jailed, wild.
1664 		 *      3. non-jailed, non-wild.
1665 		 *      4. non-jailed, wild.
1666 		 */
1667 
1668 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1669 		    lport, 0, pcbgroup->ipg_hashmask)];
1670 		LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1671 #ifdef INET6
1672 			/* XXX inp locking */
1673 			if ((inp->inp_vflag & INP_IPV4) == 0)
1674 				continue;
1675 #endif
1676 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1677 			    inp->inp_lport != lport)
1678 				continue;
1679 
1680 			injail = prison_flag(inp->inp_cred, PR_IP4);
1681 			if (injail) {
1682 				if (prison_check_ip4(inp->inp_cred,
1683 				    &laddr) != 0)
1684 					continue;
1685 			} else {
1686 				if (local_exact != NULL)
1687 					continue;
1688 			}
1689 
1690 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1691 				if (injail)
1692 					goto found;
1693 				else
1694 					local_exact = inp;
1695 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1696 #ifdef INET6
1697 				/* XXX inp locking, NULL check */
1698 				if (inp->inp_vflag & INP_IPV6PROTO)
1699 					local_wild_mapped = inp;
1700 				else
1701 #endif
1702 					if (injail)
1703 						jail_wild = inp;
1704 					else
1705 						local_wild = inp;
1706 			}
1707 		} /* LIST_FOREACH */
1708 
1709 		inp = jail_wild;
1710 		if (inp == NULL)
1711 			inp = local_exact;
1712 		if (inp == NULL)
1713 			inp = local_wild;
1714 #ifdef INET6
1715 		if (inp == NULL)
1716 			inp = local_wild_mapped;
1717 #endif
1718 		if (inp != NULL)
1719 			goto found;
1720 	}
1721 #endif
1722 
1723 	/*
1724 	 * Then look for a wildcard match, if requested.
1725 	 */
1726 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1727 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1728 #ifdef INET6
1729 		struct inpcb *local_wild_mapped = NULL;
1730 #endif
1731 		struct inpcb *jail_wild = NULL;
1732 		struct inpcbhead *head;
1733 		int injail;
1734 
1735 		/*
1736 		 * Order of socket selection - we always prefer jails.
1737 		 *      1. jailed, non-wild.
1738 		 *      2. jailed, wild.
1739 		 *      3. non-jailed, non-wild.
1740 		 *      4. non-jailed, wild.
1741 		 */
1742 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1743 		    0, pcbinfo->ipi_wildmask)];
1744 		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1745 #ifdef INET6
1746 			/* XXX inp locking */
1747 			if ((inp->inp_vflag & INP_IPV4) == 0)
1748 				continue;
1749 #endif
1750 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1751 			    inp->inp_lport != lport)
1752 				continue;
1753 
1754 			injail = prison_flag(inp->inp_cred, PR_IP4);
1755 			if (injail) {
1756 				if (prison_check_ip4(inp->inp_cred,
1757 				    &laddr) != 0)
1758 					continue;
1759 			} else {
1760 				if (local_exact != NULL)
1761 					continue;
1762 			}
1763 
1764 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1765 				if (injail)
1766 					goto found;
1767 				else
1768 					local_exact = inp;
1769 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1770 #ifdef INET6
1771 				/* XXX inp locking, NULL check */
1772 				if (inp->inp_vflag & INP_IPV6PROTO)
1773 					local_wild_mapped = inp;
1774 				else
1775 #endif
1776 					if (injail)
1777 						jail_wild = inp;
1778 					else
1779 						local_wild = inp;
1780 			}
1781 		} /* LIST_FOREACH */
1782 		inp = jail_wild;
1783 		if (inp == NULL)
1784 			inp = local_exact;
1785 		if (inp == NULL)
1786 			inp = local_wild;
1787 #ifdef INET6
1788 		if (inp == NULL)
1789 			inp = local_wild_mapped;
1790 #endif
1791 		if (inp != NULL)
1792 			goto found;
1793 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
1794 	INP_GROUP_UNLOCK(pcbgroup);
1795 	return (NULL);
1796 
1797 found:
1798 	in_pcbref(inp);
1799 	INP_GROUP_UNLOCK(pcbgroup);
1800 	if (lookupflags & INPLOOKUP_WLOCKPCB) {
1801 		INP_WLOCK(inp);
1802 		if (in_pcbrele_wlocked(inp))
1803 			return (NULL);
1804 	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1805 		INP_RLOCK(inp);
1806 		if (in_pcbrele_rlocked(inp))
1807 			return (NULL);
1808 	} else
1809 		panic("%s: locking bug", __func__);
1810 	return (inp);
1811 }
1812 #endif /* PCBGROUP */
1813 
1814 /*
1815  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1816  * that the caller has locked the hash list, and will not perform any further
1817  * locking or reference operations on either the hash list or the connection.
1818  */
1819 static struct inpcb *
1820 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1821     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1822     struct ifnet *ifp)
1823 {
1824 	struct inpcbhead *head;
1825 	struct inpcb *inp, *tmpinp;
1826 	u_short fport = fport_arg, lport = lport_arg;
1827 
1828 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1829 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1830 
1831 	INP_HASH_LOCK_ASSERT(pcbinfo);
1832 
1833 	/*
1834 	 * First look for an exact match.
1835 	 */
1836 	tmpinp = NULL;
1837 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1838 	    pcbinfo->ipi_hashmask)];
1839 	LIST_FOREACH(inp, head, inp_hash) {
1840 #ifdef INET6
1841 		/* XXX inp locking */
1842 		if ((inp->inp_vflag & INP_IPV4) == 0)
1843 			continue;
1844 #endif
1845 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
1846 		    inp->inp_laddr.s_addr == laddr.s_addr &&
1847 		    inp->inp_fport == fport &&
1848 		    inp->inp_lport == lport) {
1849 			/*
1850 			 * XXX We should be able to directly return
1851 			 * the inp here, without any checks.
1852 			 * Well unless both bound with SO_REUSEPORT?
1853 			 */
1854 			if (prison_flag(inp->inp_cred, PR_IP4))
1855 				return (inp);
1856 			if (tmpinp == NULL)
1857 				tmpinp = inp;
1858 		}
1859 	}
1860 	if (tmpinp != NULL)
1861 		return (tmpinp);
1862 
1863 	/*
1864 	 * Then look for a wildcard match, if requested.
1865 	 */
1866 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1867 		struct inpcb *local_wild = NULL, *local_exact = NULL;
1868 #ifdef INET6
1869 		struct inpcb *local_wild_mapped = NULL;
1870 #endif
1871 		struct inpcb *jail_wild = NULL;
1872 		int injail;
1873 
1874 		/*
1875 		 * Order of socket selection - we always prefer jails.
1876 		 *      1. jailed, non-wild.
1877 		 *      2. jailed, wild.
1878 		 *      3. non-jailed, non-wild.
1879 		 *      4. non-jailed, wild.
1880 		 */
1881 
1882 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1883 		    0, pcbinfo->ipi_hashmask)];
1884 		LIST_FOREACH(inp, head, inp_hash) {
1885 #ifdef INET6
1886 			/* XXX inp locking */
1887 			if ((inp->inp_vflag & INP_IPV4) == 0)
1888 				continue;
1889 #endif
1890 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
1891 			    inp->inp_lport != lport)
1892 				continue;
1893 
1894 			injail = prison_flag(inp->inp_cred, PR_IP4);
1895 			if (injail) {
1896 				if (prison_check_ip4(inp->inp_cred,
1897 				    &laddr) != 0)
1898 					continue;
1899 			} else {
1900 				if (local_exact != NULL)
1901 					continue;
1902 			}
1903 
1904 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
1905 				if (injail)
1906 					return (inp);
1907 				else
1908 					local_exact = inp;
1909 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1910 #ifdef INET6
1911 				/* XXX inp locking, NULL check */
1912 				if (inp->inp_vflag & INP_IPV6PROTO)
1913 					local_wild_mapped = inp;
1914 				else
1915 #endif
1916 					if (injail)
1917 						jail_wild = inp;
1918 					else
1919 						local_wild = inp;
1920 			}
1921 		} /* LIST_FOREACH */
1922 		if (jail_wild != NULL)
1923 			return (jail_wild);
1924 		if (local_exact != NULL)
1925 			return (local_exact);
1926 		if (local_wild != NULL)
1927 			return (local_wild);
1928 #ifdef INET6
1929 		if (local_wild_mapped != NULL)
1930 			return (local_wild_mapped);
1931 #endif
1932 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
1933 
1934 	return (NULL);
1935 }
1936 
1937 /*
1938  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
1939  * hash list lock, and will return the inpcb locked (i.e., requires
1940  * INPLOOKUP_LOCKPCB).
1941  */
1942 static struct inpcb *
1943 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1944     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
1945     struct ifnet *ifp)
1946 {
1947 	struct inpcb *inp;
1948 
1949 	INP_HASH_RLOCK(pcbinfo);
1950 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
1951 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
1952 	if (inp != NULL) {
1953 		in_pcbref(inp);
1954 		INP_HASH_RUNLOCK(pcbinfo);
1955 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
1956 			INP_WLOCK(inp);
1957 			if (in_pcbrele_wlocked(inp))
1958 				return (NULL);
1959 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
1960 			INP_RLOCK(inp);
1961 			if (in_pcbrele_rlocked(inp))
1962 				return (NULL);
1963 		} else
1964 			panic("%s: locking bug", __func__);
1965 	} else
1966 		INP_HASH_RUNLOCK(pcbinfo);
1967 	return (inp);
1968 }
1969 
1970 /*
1971  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
1972  * from which a pre-calculated hash value may be extracted.
1973  *
1974  * Possibly more of this logic should be in in_pcbgroup.c.
1975  */
1976 struct inpcb *
1977 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
1978     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
1979 {
1980 #if defined(PCBGROUP) && !defined(RSS)
1981 	struct inpcbgroup *pcbgroup;
1982 #endif
1983 
1984 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
1985 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1986 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
1987 	    ("%s: LOCKPCB not set", __func__));
1988 
1989 	/*
1990 	 * When not using RSS, use connection groups in preference to the
1991 	 * reservation table when looking up 4-tuples.  When using RSS, just
1992 	 * use the reservation table, due to the cost of the Toeplitz hash
1993 	 * in software.
1994 	 *
1995 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
1996 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
1997 	 * in software.
1998 	 */
1999 #if defined(PCBGROUP) && !defined(RSS)
2000 	if (in_pcbgroup_enabled(pcbinfo)) {
2001 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2002 		    fport);
2003 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2004 		    laddr, lport, lookupflags, ifp));
2005 	}
2006 #endif
2007 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2008 	    lookupflags, ifp));
2009 }
2010 
2011 struct inpcb *
2012 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2013     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2014     struct ifnet *ifp, struct mbuf *m)
2015 {
2016 #ifdef PCBGROUP
2017 	struct inpcbgroup *pcbgroup;
2018 #endif
2019 
2020 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2021 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2022 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2023 	    ("%s: LOCKPCB not set", __func__));
2024 
2025 #ifdef PCBGROUP
2026 	/*
2027 	 * If we can use a hardware-generated hash to look up the connection
2028 	 * group, use that connection group to find the inpcb.  Otherwise
2029 	 * fall back on a software hash -- or the reservation table if we're
2030 	 * using RSS.
2031 	 *
2032 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2033 	 */
2034 	if (in_pcbgroup_enabled(pcbinfo) &&
2035 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2036 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2037 		    m->m_pkthdr.flowid);
2038 		if (pcbgroup != NULL)
2039 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2040 			    fport, laddr, lport, lookupflags, ifp));
2041 #ifndef RSS
2042 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2043 		    fport);
2044 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2045 		    laddr, lport, lookupflags, ifp));
2046 #endif
2047 	}
2048 #endif
2049 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2050 	    lookupflags, ifp));
2051 }
2052 #endif /* INET */
2053 
2054 /*
2055  * Insert PCB onto various hash lists.
2056  */
2057 static int
2058 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2059 {
2060 	struct inpcbhead *pcbhash;
2061 	struct inpcbporthead *pcbporthash;
2062 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2063 	struct inpcbport *phd;
2064 	u_int32_t hashkey_faddr;
2065 
2066 	INP_WLOCK_ASSERT(inp);
2067 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2068 
2069 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2070 	    ("in_pcbinshash: INP_INHASHLIST"));
2071 
2072 #ifdef INET6
2073 	if (inp->inp_vflag & INP_IPV6)
2074 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2075 	else
2076 #endif
2077 	hashkey_faddr = inp->inp_faddr.s_addr;
2078 
2079 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2080 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2081 
2082 	pcbporthash = &pcbinfo->ipi_porthashbase[
2083 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2084 
2085 	/*
2086 	 * Go through port list and look for a head for this lport.
2087 	 */
2088 	LIST_FOREACH(phd, pcbporthash, phd_hash) {
2089 		if (phd->phd_port == inp->inp_lport)
2090 			break;
2091 	}
2092 	/*
2093 	 * If none exists, malloc one and tack it on.
2094 	 */
2095 	if (phd == NULL) {
2096 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2097 		if (phd == NULL) {
2098 			return (ENOBUFS); /* XXX */
2099 		}
2100 		phd->phd_port = inp->inp_lport;
2101 		LIST_INIT(&phd->phd_pcblist);
2102 		LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2103 	}
2104 	inp->inp_phd = phd;
2105 	LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2106 	LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2107 	inp->inp_flags |= INP_INHASHLIST;
2108 #ifdef PCBGROUP
2109 	if (do_pcbgroup_update)
2110 		in_pcbgroup_update(inp);
2111 #endif
2112 	return (0);
2113 }
2114 
2115 /*
2116  * For now, there are two public interfaces to insert an inpcb into the hash
2117  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2118  * is used only in the TCP syncache, where in_pcbinshash is called before the
2119  * full 4-tuple is set for the inpcb, and we don't want to install in the
2120  * pcbgroup until later.
2121  *
2122  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2123  * connection groups, and partially initialised inpcbs should not be exposed
2124  * to either reservation hash tables or pcbgroups.
2125  */
2126 int
2127 in_pcbinshash(struct inpcb *inp)
2128 {
2129 
2130 	return (in_pcbinshash_internal(inp, 1));
2131 }
2132 
2133 int
2134 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2135 {
2136 
2137 	return (in_pcbinshash_internal(inp, 0));
2138 }
2139 
2140 /*
2141  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2142  * changed. NOTE: This does not handle the case of the lport changing (the
2143  * hashed port list would have to be updated as well), so the lport must
2144  * not change after in_pcbinshash() has been called.
2145  */
2146 void
2147 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2148 {
2149 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2150 	struct inpcbhead *head;
2151 	u_int32_t hashkey_faddr;
2152 
2153 	INP_WLOCK_ASSERT(inp);
2154 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2155 
2156 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2157 	    ("in_pcbrehash: !INP_INHASHLIST"));
2158 
2159 #ifdef INET6
2160 	if (inp->inp_vflag & INP_IPV6)
2161 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2162 	else
2163 #endif
2164 	hashkey_faddr = inp->inp_faddr.s_addr;
2165 
2166 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2167 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2168 
2169 	LIST_REMOVE(inp, inp_hash);
2170 	LIST_INSERT_HEAD(head, inp, inp_hash);
2171 
2172 #ifdef PCBGROUP
2173 	if (m != NULL)
2174 		in_pcbgroup_update_mbuf(inp, m);
2175 	else
2176 		in_pcbgroup_update(inp);
2177 #endif
2178 }
2179 
2180 void
2181 in_pcbrehash(struct inpcb *inp)
2182 {
2183 
2184 	in_pcbrehash_mbuf(inp, NULL);
2185 }
2186 
2187 /*
2188  * Remove PCB from various lists.
2189  */
2190 static void
2191 in_pcbremlists(struct inpcb *inp)
2192 {
2193 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2194 
2195 #ifdef INVARIANTS
2196 	if (pcbinfo == &V_tcbinfo) {
2197 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2198 	} else {
2199 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2200 	}
2201 #endif
2202 
2203 	INP_WLOCK_ASSERT(inp);
2204 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2205 
2206 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2207 	if (inp->inp_flags & INP_INHASHLIST) {
2208 		struct inpcbport *phd = inp->inp_phd;
2209 
2210 		INP_HASH_WLOCK(pcbinfo);
2211 		LIST_REMOVE(inp, inp_hash);
2212 		LIST_REMOVE(inp, inp_portlist);
2213 		if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2214 			LIST_REMOVE(phd, phd_hash);
2215 			free(phd, M_PCB);
2216 		}
2217 		INP_HASH_WUNLOCK(pcbinfo);
2218 		inp->inp_flags &= ~INP_INHASHLIST;
2219 	}
2220 	LIST_REMOVE(inp, inp_list);
2221 	pcbinfo->ipi_count--;
2222 #ifdef PCBGROUP
2223 	in_pcbgroup_remove(inp);
2224 #endif
2225 }
2226 
2227 /*
2228  * A set label operation has occurred at the socket layer, propagate the
2229  * label change into the in_pcb for the socket.
2230  */
2231 void
2232 in_pcbsosetlabel(struct socket *so)
2233 {
2234 #ifdef MAC
2235 	struct inpcb *inp;
2236 
2237 	inp = sotoinpcb(so);
2238 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2239 
2240 	INP_WLOCK(inp);
2241 	SOCK_LOCK(so);
2242 	mac_inpcb_sosetlabel(so, inp);
2243 	SOCK_UNLOCK(so);
2244 	INP_WUNLOCK(inp);
2245 #endif
2246 }
2247 
2248 /*
2249  * ipport_tick runs once per second, determining if random port allocation
2250  * should be continued.  If more than ipport_randomcps ports have been
2251  * allocated in the last second, then we return to sequential port
2252  * allocation. We return to random allocation only once we drop below
2253  * ipport_randomcps for at least ipport_randomtime seconds.
2254  */
2255 static void
2256 ipport_tick(void *xtp)
2257 {
2258 	VNET_ITERATOR_DECL(vnet_iter);
2259 
2260 	VNET_LIST_RLOCK_NOSLEEP();
2261 	VNET_FOREACH(vnet_iter) {
2262 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2263 		if (V_ipport_tcpallocs <=
2264 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2265 			if (V_ipport_stoprandom > 0)
2266 				V_ipport_stoprandom--;
2267 		} else
2268 			V_ipport_stoprandom = V_ipport_randomtime;
2269 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2270 		CURVNET_RESTORE();
2271 	}
2272 	VNET_LIST_RUNLOCK_NOSLEEP();
2273 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2274 }
2275 
2276 static void
2277 ip_fini(void *xtp)
2278 {
2279 
2280 	callout_stop(&ipport_tick_callout);
2281 }
2282 
2283 /*
2284  * The ipport_callout should start running at about the time we attach the
2285  * inet or inet6 domains.
2286  */
2287 static void
2288 ipport_tick_init(const void *unused __unused)
2289 {
2290 
2291 	/* Start ipport_tick. */
2292 	callout_init(&ipport_tick_callout, 1);
2293 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2294 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2295 		SHUTDOWN_PRI_DEFAULT);
2296 }
2297 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2298     ipport_tick_init, NULL);
2299 
2300 void
2301 inp_wlock(struct inpcb *inp)
2302 {
2303 
2304 	INP_WLOCK(inp);
2305 }
2306 
2307 void
2308 inp_wunlock(struct inpcb *inp)
2309 {
2310 
2311 	INP_WUNLOCK(inp);
2312 }
2313 
2314 void
2315 inp_rlock(struct inpcb *inp)
2316 {
2317 
2318 	INP_RLOCK(inp);
2319 }
2320 
2321 void
2322 inp_runlock(struct inpcb *inp)
2323 {
2324 
2325 	INP_RUNLOCK(inp);
2326 }
2327 
2328 #ifdef INVARIANTS
2329 void
2330 inp_lock_assert(struct inpcb *inp)
2331 {
2332 
2333 	INP_WLOCK_ASSERT(inp);
2334 }
2335 
2336 void
2337 inp_unlock_assert(struct inpcb *inp)
2338 {
2339 
2340 	INP_UNLOCK_ASSERT(inp);
2341 }
2342 #endif
2343 
2344 void
2345 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2346 {
2347 	struct inpcb *inp;
2348 
2349 	INP_INFO_WLOCK(&V_tcbinfo);
2350 	LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2351 		INP_WLOCK(inp);
2352 		func(inp, arg);
2353 		INP_WUNLOCK(inp);
2354 	}
2355 	INP_INFO_WUNLOCK(&V_tcbinfo);
2356 }
2357 
2358 struct socket *
2359 inp_inpcbtosocket(struct inpcb *inp)
2360 {
2361 
2362 	INP_WLOCK_ASSERT(inp);
2363 	return (inp->inp_socket);
2364 }
2365 
2366 struct tcpcb *
2367 inp_inpcbtotcpcb(struct inpcb *inp)
2368 {
2369 
2370 	INP_WLOCK_ASSERT(inp);
2371 	return ((struct tcpcb *)inp->inp_ppcb);
2372 }
2373 
2374 int
2375 inp_ip_tos_get(const struct inpcb *inp)
2376 {
2377 
2378 	return (inp->inp_ip_tos);
2379 }
2380 
2381 void
2382 inp_ip_tos_set(struct inpcb *inp, int val)
2383 {
2384 
2385 	inp->inp_ip_tos = val;
2386 }
2387 
2388 void
2389 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2390     uint32_t *faddr, uint16_t *fp)
2391 {
2392 
2393 	INP_LOCK_ASSERT(inp);
2394 	*laddr = inp->inp_laddr.s_addr;
2395 	*faddr = inp->inp_faddr.s_addr;
2396 	*lp = inp->inp_lport;
2397 	*fp = inp->inp_fport;
2398 }
2399 
2400 struct inpcb *
2401 so_sotoinpcb(struct socket *so)
2402 {
2403 
2404 	return (sotoinpcb(so));
2405 }
2406 
2407 struct tcpcb *
2408 so_sototcpcb(struct socket *so)
2409 {
2410 
2411 	return (sototcpcb(so));
2412 }
2413 
2414 #ifdef DDB
2415 static void
2416 db_print_indent(int indent)
2417 {
2418 	int i;
2419 
2420 	for (i = 0; i < indent; i++)
2421 		db_printf(" ");
2422 }
2423 
2424 static void
2425 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2426 {
2427 	char faddr_str[48], laddr_str[48];
2428 
2429 	db_print_indent(indent);
2430 	db_printf("%s at %p\n", name, inc);
2431 
2432 	indent += 2;
2433 
2434 #ifdef INET6
2435 	if (inc->inc_flags & INC_ISIPV6) {
2436 		/* IPv6. */
2437 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2438 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2439 	} else
2440 #endif
2441 	{
2442 		/* IPv4. */
2443 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2444 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2445 	}
2446 	db_print_indent(indent);
2447 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2448 	    ntohs(inc->inc_lport));
2449 	db_print_indent(indent);
2450 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2451 	    ntohs(inc->inc_fport));
2452 }
2453 
2454 static void
2455 db_print_inpflags(int inp_flags)
2456 {
2457 	int comma;
2458 
2459 	comma = 0;
2460 	if (inp_flags & INP_RECVOPTS) {
2461 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2462 		comma = 1;
2463 	}
2464 	if (inp_flags & INP_RECVRETOPTS) {
2465 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2466 		comma = 1;
2467 	}
2468 	if (inp_flags & INP_RECVDSTADDR) {
2469 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2470 		comma = 1;
2471 	}
2472 	if (inp_flags & INP_HDRINCL) {
2473 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2474 		comma = 1;
2475 	}
2476 	if (inp_flags & INP_HIGHPORT) {
2477 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2478 		comma = 1;
2479 	}
2480 	if (inp_flags & INP_LOWPORT) {
2481 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2482 		comma = 1;
2483 	}
2484 	if (inp_flags & INP_ANONPORT) {
2485 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2486 		comma = 1;
2487 	}
2488 	if (inp_flags & INP_RECVIF) {
2489 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2490 		comma = 1;
2491 	}
2492 	if (inp_flags & INP_MTUDISC) {
2493 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2494 		comma = 1;
2495 	}
2496 	if (inp_flags & INP_RECVTTL) {
2497 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2498 		comma = 1;
2499 	}
2500 	if (inp_flags & INP_DONTFRAG) {
2501 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2502 		comma = 1;
2503 	}
2504 	if (inp_flags & INP_RECVTOS) {
2505 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2506 		comma = 1;
2507 	}
2508 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2509 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2510 		comma = 1;
2511 	}
2512 	if (inp_flags & IN6P_PKTINFO) {
2513 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2514 		comma = 1;
2515 	}
2516 	if (inp_flags & IN6P_HOPLIMIT) {
2517 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2518 		comma = 1;
2519 	}
2520 	if (inp_flags & IN6P_HOPOPTS) {
2521 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2522 		comma = 1;
2523 	}
2524 	if (inp_flags & IN6P_DSTOPTS) {
2525 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2526 		comma = 1;
2527 	}
2528 	if (inp_flags & IN6P_RTHDR) {
2529 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2530 		comma = 1;
2531 	}
2532 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2533 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2534 		comma = 1;
2535 	}
2536 	if (inp_flags & IN6P_TCLASS) {
2537 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2538 		comma = 1;
2539 	}
2540 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2541 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2542 		comma = 1;
2543 	}
2544 	if (inp_flags & INP_TIMEWAIT) {
2545 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2546 		comma  = 1;
2547 	}
2548 	if (inp_flags & INP_ONESBCAST) {
2549 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2550 		comma  = 1;
2551 	}
2552 	if (inp_flags & INP_DROPPED) {
2553 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2554 		comma  = 1;
2555 	}
2556 	if (inp_flags & INP_SOCKREF) {
2557 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2558 		comma  = 1;
2559 	}
2560 	if (inp_flags & IN6P_RFC2292) {
2561 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2562 		comma = 1;
2563 	}
2564 	if (inp_flags & IN6P_MTU) {
2565 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2566 		comma = 1;
2567 	}
2568 }
2569 
2570 static void
2571 db_print_inpvflag(u_char inp_vflag)
2572 {
2573 	int comma;
2574 
2575 	comma = 0;
2576 	if (inp_vflag & INP_IPV4) {
2577 		db_printf("%sINP_IPV4", comma ? ", " : "");
2578 		comma  = 1;
2579 	}
2580 	if (inp_vflag & INP_IPV6) {
2581 		db_printf("%sINP_IPV6", comma ? ", " : "");
2582 		comma  = 1;
2583 	}
2584 	if (inp_vflag & INP_IPV6PROTO) {
2585 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2586 		comma  = 1;
2587 	}
2588 }
2589 
2590 static void
2591 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2592 {
2593 
2594 	db_print_indent(indent);
2595 	db_printf("%s at %p\n", name, inp);
2596 
2597 	indent += 2;
2598 
2599 	db_print_indent(indent);
2600 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2601 
2602 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2603 
2604 	db_print_indent(indent);
2605 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2606 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2607 
2608 	db_print_indent(indent);
2609 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2610 	   inp->inp_label, inp->inp_flags);
2611 	db_print_inpflags(inp->inp_flags);
2612 	db_printf(")\n");
2613 
2614 	db_print_indent(indent);
2615 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2616 	    inp->inp_vflag);
2617 	db_print_inpvflag(inp->inp_vflag);
2618 	db_printf(")\n");
2619 
2620 	db_print_indent(indent);
2621 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2622 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2623 
2624 	db_print_indent(indent);
2625 #ifdef INET6
2626 	if (inp->inp_vflag & INP_IPV6) {
2627 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2628 		    "in6p_moptions: %p\n", inp->in6p_options,
2629 		    inp->in6p_outputopts, inp->in6p_moptions);
2630 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2631 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2632 		    inp->in6p_hops);
2633 	} else
2634 #endif
2635 	{
2636 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2637 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2638 		    inp->inp_options, inp->inp_moptions);
2639 	}
2640 
2641 	db_print_indent(indent);
2642 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2643 	    (uintmax_t)inp->inp_gencnt);
2644 }
2645 
2646 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2647 {
2648 	struct inpcb *inp;
2649 
2650 	if (!have_addr) {
2651 		db_printf("usage: show inpcb <addr>\n");
2652 		return;
2653 	}
2654 	inp = (struct inpcb *)addr;
2655 
2656 	db_print_inpcb(inp, "inpcb", 0);
2657 }
2658 #endif /* DDB */
2659