xref: /freebsd/sys/netinet/in_pcb.c (revision bec6dc30e1a6ec4e96d8b29510d345e3f2a8ea40)
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/smp.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/refcount.h>
68 #include <sys/jail.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #include <vm/uma.h>
77 
78 #include <net/if.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
84 #include <net/vnet.h>
85 
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp_var.h>
91 #ifdef TCPHPTS
92 #include <netinet/tcp_hpts.h>
93 #endif
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
96 #endif
97 #ifdef INET
98 #include <netinet/in_var.h>
99 #endif
100 #ifdef INET6
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #endif /* INET6 */
106 
107 #include <netipsec/ipsec_support.h>
108 
109 #include <security/mac/mac_framework.h>
110 
111 #define	INPCBLBGROUP_SIZMIN	8
112 #define	INPCBLBGROUP_SIZMAX	256
113 
114 static struct callout	ipport_tick_callout;
115 
116 /*
117  * These configure the range of local port addresses assigned to
118  * "unspecified" outgoing connections/packets/whatever.
119  */
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
126 
127 /*
128  * Reserved ports accessible only to root. There are significant
129  * security considerations that must be accounted for when changing these,
130  * but the security benefits can be great. Please be careful.
131  */
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
134 
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 static VNET_DEFINE(int, ipport_tcplastcount);
142 
143 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
144 
145 static void	in_pcbremlists(struct inpcb *inp);
146 #ifdef INET
147 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 			    struct in_addr faddr, u_int fport_arg,
149 			    struct in_addr laddr, u_int lport_arg,
150 			    int lookupflags, struct ifnet *ifp);
151 
152 #define RANGECHK(var, min, max) \
153 	if ((var) < (min)) { (var) = (min); } \
154 	else if ((var) > (max)) { (var) = (max); }
155 
156 static int
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
158 {
159 	int error;
160 
161 	error = sysctl_handle_int(oidp, arg1, arg2, req);
162 	if (error == 0) {
163 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
169 	}
170 	return (error);
171 }
172 
173 #undef RANGECHK
174 
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
176     "IP Ports");
177 
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 	&VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 	CTLFLAG_VNET | CTLFLAG_RW,
203 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 	CTLFLAG_VNET | CTLFLAG_RW,
206 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 	"allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 	CTLFLAG_VNET | CTLFLAG_RW,
210 	&VNET_NAME(ipport_randomtime), 0,
211 	"Minimum time to keep sequental port "
212 	"allocation before switching to a random one");
213 #endif /* INET */
214 
215 /*
216  * in_pcb.c: manage the Protocol Control Blocks.
217  *
218  * NOTE: It is assumed that most of these functions will be called with
219  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220  * functions often modify hash chains or addresses in pcbs.
221  */
222 
223 static struct inpcblbgroup *
224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225     uint16_t port, const union in_dependaddr *addr, int size)
226 {
227 	struct inpcblbgroup *grp;
228 	size_t bytes;
229 
230 	bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 	grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
232 	if (!grp)
233 		return (NULL);
234 	grp->il_vflag = vflag;
235 	grp->il_lport = port;
236 	grp->il_dependladdr = *addr;
237 	grp->il_inpsiz = size;
238 	LIST_INSERT_HEAD(hdr, grp, il_list);
239 	return (grp);
240 }
241 
242 static void
243 in_pcblbgroup_free(struct inpcblbgroup *grp)
244 {
245 
246 	LIST_REMOVE(grp, il_list);
247 	free(grp, M_TEMP);
248 }
249 
250 static struct inpcblbgroup *
251 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
252     struct inpcblbgroup *old_grp, int size)
253 {
254 	struct inpcblbgroup *grp;
255 	int i;
256 
257 	grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
258 	    old_grp->il_lport, &old_grp->il_dependladdr, size);
259 	if (!grp)
260 		return (NULL);
261 
262 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
263 	    ("invalid new local group size %d and old local group count %d",
264 	     grp->il_inpsiz, old_grp->il_inpcnt));
265 
266 	for (i = 0; i < old_grp->il_inpcnt; ++i)
267 		grp->il_inp[i] = old_grp->il_inp[i];
268 	grp->il_inpcnt = old_grp->il_inpcnt;
269 	in_pcblbgroup_free(old_grp);
270 	return (grp);
271 }
272 
273 /*
274  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
275  * and shrink group if possible.
276  */
277 static void
278 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
279     int i)
280 {
281 	struct inpcblbgroup *grp = *grpp;
282 
283 	for (; i + 1 < grp->il_inpcnt; ++i)
284 		grp->il_inp[i] = grp->il_inp[i + 1];
285 	grp->il_inpcnt--;
286 
287 	if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
288 	    grp->il_inpcnt <= (grp->il_inpsiz / 4)) {
289 		/* Shrink this group. */
290 		struct inpcblbgroup *new_grp =
291 		        in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
292 		if (new_grp)
293 			*grpp = new_grp;
294 	}
295 	return;
296 }
297 
298 /*
299  * Add PCB to load balance group for SO_REUSEPORT_LB option.
300  */
301 static int
302 in_pcbinslbgrouphash(struct inpcb *inp)
303 {
304 	struct inpcbinfo *pcbinfo;
305 	struct inpcblbgrouphead *hdr;
306 	struct inpcblbgroup *grp;
307 	uint16_t hashmask, lport;
308 	uint32_t group_index;
309 	struct ucred *cred;
310 	static int limit_logged = 0;
311 
312 	pcbinfo = inp->inp_pcbinfo;
313 
314 	INP_WLOCK_ASSERT(inp);
315 	INP_HASH_WLOCK_ASSERT(pcbinfo);
316 
317 	if (pcbinfo->ipi_lbgrouphashbase == NULL)
318 		return (0);
319 
320 	hashmask = pcbinfo->ipi_lbgrouphashmask;
321 	lport = inp->inp_lport;
322 	group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
323 	hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
324 
325 	/*
326 	 * Don't allow jailed socket to join local group.
327 	 */
328 	if (inp->inp_socket != NULL)
329 		cred = inp->inp_socket->so_cred;
330 	else
331 		cred = NULL;
332 	if (cred != NULL && jailed(cred))
333 		return (0);
334 
335 #ifdef INET6
336 	/*
337 	 * Don't allow IPv4 mapped INET6 wild socket.
338 	 */
339 	if ((inp->inp_vflag & INP_IPV4) &&
340 	    inp->inp_laddr.s_addr == INADDR_ANY &&
341 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
342 		return (0);
343 	}
344 #endif
345 
346 	hdr = &pcbinfo->ipi_lbgrouphashbase[
347 	    INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
348 	        pcbinfo->ipi_lbgrouphashmask)];
349 	LIST_FOREACH(grp, hdr, il_list) {
350 		if (grp->il_vflag == inp->inp_vflag &&
351 		    grp->il_lport == inp->inp_lport &&
352 		    memcmp(&grp->il_dependladdr,
353 		        &inp->inp_inc.inc_ie.ie_dependladdr,
354 		        sizeof(grp->il_dependladdr)) == 0) {
355 			break;
356 		}
357 	}
358 	if (grp == NULL) {
359 		/* Create new load balance group. */
360 		grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
361 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
362 		    INPCBLBGROUP_SIZMIN);
363 		if (!grp)
364 			return (ENOBUFS);
365 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
366 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
367 			if (!limit_logged) {
368 				limit_logged = 1;
369 				printf("lb group port %d, limit reached\n",
370 				    ntohs(grp->il_lport));
371 			}
372 			return (0);
373 		}
374 
375 		/* Expand this local group. */
376 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
377 		if (!grp)
378 			return (ENOBUFS);
379 	}
380 
381 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
382 			("invalid local group size %d and count %d",
383 			 grp->il_inpsiz, grp->il_inpcnt));
384 
385 	grp->il_inp[grp->il_inpcnt] = inp;
386 	grp->il_inpcnt++;
387 	return (0);
388 }
389 
390 /*
391  * Remove PCB from load balance group.
392  */
393 static void
394 in_pcbremlbgrouphash(struct inpcb *inp)
395 {
396 	struct inpcbinfo *pcbinfo;
397 	struct inpcblbgrouphead *hdr;
398 	struct inpcblbgroup *grp;
399 	int i;
400 
401 	pcbinfo = inp->inp_pcbinfo;
402 
403 	INP_WLOCK_ASSERT(inp);
404 	INP_HASH_WLOCK_ASSERT(pcbinfo);
405 
406 	if (pcbinfo->ipi_lbgrouphashbase == NULL)
407 		return;
408 
409 	hdr = &pcbinfo->ipi_lbgrouphashbase[
410 	    INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
411 	        pcbinfo->ipi_lbgrouphashmask)];
412 
413 	LIST_FOREACH(grp, hdr, il_list) {
414 		for (i = 0; i < grp->il_inpcnt; ++i) {
415 			if (grp->il_inp[i] != inp)
416 				continue;
417 
418 			if (grp->il_inpcnt == 1) {
419 				/* We are the last, free this local group. */
420 				in_pcblbgroup_free(grp);
421 			} else {
422 				/* Pull up inpcbs, shrink group if possible. */
423 				in_pcblbgroup_reorder(hdr, &grp, i);
424 			}
425 			return;
426 		}
427 	}
428 }
429 
430 /*
431  * Different protocols initialize their inpcbs differently - giving
432  * different name to the lock.  But they all are disposed the same.
433  */
434 static void
435 inpcb_fini(void *mem, int size)
436 {
437 	struct inpcb *inp = mem;
438 
439 	INP_LOCK_DESTROY(inp);
440 }
441 
442 /*
443  * Initialize an inpcbinfo -- we should be able to reduce the number of
444  * arguments in time.
445  */
446 void
447 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
448     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
449     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
450 {
451 
452 	INP_INFO_LOCK_INIT(pcbinfo, name);
453 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
454 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
455 #ifdef VIMAGE
456 	pcbinfo->ipi_vnet = curvnet;
457 #endif
458 	pcbinfo->ipi_listhead = listhead;
459 	CK_LIST_INIT(pcbinfo->ipi_listhead);
460 	pcbinfo->ipi_count = 0;
461 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
462 	    &pcbinfo->ipi_hashmask);
463 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
464 	    &pcbinfo->ipi_porthashmask);
465 	pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
466 	    &pcbinfo->ipi_lbgrouphashmask);
467 #ifdef PCBGROUP
468 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
469 #endif
470 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
471 	    NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
472 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
473 	uma_zone_set_warning(pcbinfo->ipi_zone,
474 	    "kern.ipc.maxsockets limit reached");
475 }
476 
477 /*
478  * Destroy an inpcbinfo.
479  */
480 void
481 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
482 {
483 
484 	KASSERT(pcbinfo->ipi_count == 0,
485 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
486 
487 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
488 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
489 	    pcbinfo->ipi_porthashmask);
490 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
491 	    pcbinfo->ipi_lbgrouphashmask);
492 #ifdef PCBGROUP
493 	in_pcbgroup_destroy(pcbinfo);
494 #endif
495 	uma_zdestroy(pcbinfo->ipi_zone);
496 	INP_LIST_LOCK_DESTROY(pcbinfo);
497 	INP_HASH_LOCK_DESTROY(pcbinfo);
498 	INP_INFO_LOCK_DESTROY(pcbinfo);
499 }
500 
501 /*
502  * Allocate a PCB and associate it with the socket.
503  * On success return with the PCB locked.
504  */
505 int
506 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
507 {
508 	struct inpcb *inp;
509 	int error;
510 
511 #ifdef INVARIANTS
512 	if (pcbinfo == &V_tcbinfo) {
513 		INP_INFO_RLOCK_ASSERT(pcbinfo);
514 	} else {
515 		INP_INFO_WLOCK_ASSERT(pcbinfo);
516 	}
517 #endif
518 
519 	error = 0;
520 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
521 	if (inp == NULL)
522 		return (ENOBUFS);
523 	bzero(&inp->inp_start_zero, inp_zero_size);
524 	inp->inp_pcbinfo = pcbinfo;
525 	inp->inp_socket = so;
526 	inp->inp_cred = crhold(so->so_cred);
527 	inp->inp_inc.inc_fibnum = so->so_fibnum;
528 #ifdef MAC
529 	error = mac_inpcb_init(inp, M_NOWAIT);
530 	if (error != 0)
531 		goto out;
532 	mac_inpcb_create(so, inp);
533 #endif
534 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
535 	error = ipsec_init_pcbpolicy(inp);
536 	if (error != 0) {
537 #ifdef MAC
538 		mac_inpcb_destroy(inp);
539 #endif
540 		goto out;
541 	}
542 #endif /*IPSEC*/
543 #ifdef INET6
544 	if (INP_SOCKAF(so) == AF_INET6) {
545 		inp->inp_vflag |= INP_IPV6PROTO;
546 		if (V_ip6_v6only)
547 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
548 	}
549 #endif
550 	INP_WLOCK(inp);
551 	INP_LIST_WLOCK(pcbinfo);
552 	CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
553 	pcbinfo->ipi_count++;
554 	so->so_pcb = (caddr_t)inp;
555 #ifdef INET6
556 	if (V_ip6_auto_flowlabel)
557 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
558 #endif
559 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
560 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
561 
562 	/*
563 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
564 	 * to be cleaned up.
565 	 */
566 	inp->inp_route.ro_flags = RT_LLE_CACHE;
567 	INP_LIST_WUNLOCK(pcbinfo);
568 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
569 out:
570 	if (error != 0) {
571 		crfree(inp->inp_cred);
572 		uma_zfree(pcbinfo->ipi_zone, inp);
573 	}
574 #endif
575 	return (error);
576 }
577 
578 #ifdef INET
579 int
580 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
581 {
582 	int anonport, error;
583 
584 	INP_WLOCK_ASSERT(inp);
585 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
586 
587 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
588 		return (EINVAL);
589 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
590 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
591 	    &inp->inp_lport, cred);
592 	if (error)
593 		return (error);
594 	if (in_pcbinshash(inp) != 0) {
595 		inp->inp_laddr.s_addr = INADDR_ANY;
596 		inp->inp_lport = 0;
597 		return (EAGAIN);
598 	}
599 	if (anonport)
600 		inp->inp_flags |= INP_ANONPORT;
601 	return (0);
602 }
603 #endif
604 
605 /*
606  * Select a local port (number) to use.
607  */
608 #if defined(INET) || defined(INET6)
609 int
610 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
611     struct ucred *cred, int lookupflags)
612 {
613 	struct inpcbinfo *pcbinfo;
614 	struct inpcb *tmpinp;
615 	unsigned short *lastport;
616 	int count, dorandom, error;
617 	u_short aux, first, last, lport;
618 #ifdef INET
619 	struct in_addr laddr;
620 #endif
621 
622 	pcbinfo = inp->inp_pcbinfo;
623 
624 	/*
625 	 * Because no actual state changes occur here, a global write lock on
626 	 * the pcbinfo isn't required.
627 	 */
628 	INP_LOCK_ASSERT(inp);
629 	INP_HASH_LOCK_ASSERT(pcbinfo);
630 
631 	if (inp->inp_flags & INP_HIGHPORT) {
632 		first = V_ipport_hifirstauto;	/* sysctl */
633 		last  = V_ipport_hilastauto;
634 		lastport = &pcbinfo->ipi_lasthi;
635 	} else if (inp->inp_flags & INP_LOWPORT) {
636 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
637 		if (error)
638 			return (error);
639 		first = V_ipport_lowfirstauto;	/* 1023 */
640 		last  = V_ipport_lowlastauto;	/* 600 */
641 		lastport = &pcbinfo->ipi_lastlow;
642 	} else {
643 		first = V_ipport_firstauto;	/* sysctl */
644 		last  = V_ipport_lastauto;
645 		lastport = &pcbinfo->ipi_lastport;
646 	}
647 	/*
648 	 * For UDP(-Lite), use random port allocation as long as the user
649 	 * allows it.  For TCP (and as of yet unknown) connections,
650 	 * use random port allocation only if the user allows it AND
651 	 * ipport_tick() allows it.
652 	 */
653 	if (V_ipport_randomized &&
654 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
655 		pcbinfo == &V_ulitecbinfo))
656 		dorandom = 1;
657 	else
658 		dorandom = 0;
659 	/*
660 	 * It makes no sense to do random port allocation if
661 	 * we have the only port available.
662 	 */
663 	if (first == last)
664 		dorandom = 0;
665 	/* Make sure to not include UDP(-Lite) packets in the count. */
666 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
667 		V_ipport_tcpallocs++;
668 	/*
669 	 * Instead of having two loops further down counting up or down
670 	 * make sure that first is always <= last and go with only one
671 	 * code path implementing all logic.
672 	 */
673 	if (first > last) {
674 		aux = first;
675 		first = last;
676 		last = aux;
677 	}
678 
679 #ifdef INET
680 	/* Make the compiler happy. */
681 	laddr.s_addr = 0;
682 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
683 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
684 		    __func__, inp));
685 		laddr = *laddrp;
686 	}
687 #endif
688 	tmpinp = NULL;	/* Make compiler happy. */
689 	lport = *lportp;
690 
691 	if (dorandom)
692 		*lastport = first + (arc4random() % (last - first));
693 
694 	count = last - first;
695 
696 	do {
697 		if (count-- < 0)	/* completely used? */
698 			return (EADDRNOTAVAIL);
699 		++*lastport;
700 		if (*lastport < first || *lastport > last)
701 			*lastport = first;
702 		lport = htons(*lastport);
703 
704 #ifdef INET6
705 		if ((inp->inp_vflag & INP_IPV6) != 0)
706 			tmpinp = in6_pcblookup_local(pcbinfo,
707 			    &inp->in6p_laddr, lport, lookupflags, cred);
708 #endif
709 #if defined(INET) && defined(INET6)
710 		else
711 #endif
712 #ifdef INET
713 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
714 			    lport, lookupflags, cred);
715 #endif
716 	} while (tmpinp != NULL);
717 
718 #ifdef INET
719 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
720 		laddrp->s_addr = laddr.s_addr;
721 #endif
722 	*lportp = lport;
723 
724 	return (0);
725 }
726 
727 /*
728  * Return cached socket options.
729  */
730 int
731 inp_so_options(const struct inpcb *inp)
732 {
733 	int so_options;
734 
735 	so_options = 0;
736 
737 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
738 		so_options |= SO_REUSEPORT_LB;
739 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
740 		so_options |= SO_REUSEPORT;
741 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
742 		so_options |= SO_REUSEADDR;
743 	return (so_options);
744 }
745 #endif /* INET || INET6 */
746 
747 /*
748  * Check if a new BINDMULTI socket is allowed to be created.
749  *
750  * ni points to the new inp.
751  * oi points to the exisitng inp.
752  *
753  * This checks whether the existing inp also has BINDMULTI and
754  * whether the credentials match.
755  */
756 int
757 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
758 {
759 	/* Check permissions match */
760 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
761 	    (ni->inp_cred->cr_uid !=
762 	    oi->inp_cred->cr_uid))
763 		return (0);
764 
765 	/* Check the existing inp has BINDMULTI set */
766 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
767 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
768 		return (0);
769 
770 	/*
771 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
772 	 * it is and it matches the checks.
773 	 */
774 	return (1);
775 }
776 
777 #ifdef INET
778 /*
779  * Set up a bind operation on a PCB, performing port allocation
780  * as required, but do not actually modify the PCB. Callers can
781  * either complete the bind by setting inp_laddr/inp_lport and
782  * calling in_pcbinshash(), or they can just use the resulting
783  * port and address to authorise the sending of a once-off packet.
784  *
785  * On error, the values of *laddrp and *lportp are not changed.
786  */
787 int
788 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
789     u_short *lportp, struct ucred *cred)
790 {
791 	struct socket *so = inp->inp_socket;
792 	struct sockaddr_in *sin;
793 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
794 	struct in_addr laddr;
795 	u_short lport = 0;
796 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
797 	int error;
798 
799 	/*
800 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
801 	 * so that we don't have to add to the (already messy) code below.
802 	 */
803 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
804 
805 	/*
806 	 * No state changes, so read locks are sufficient here.
807 	 */
808 	INP_LOCK_ASSERT(inp);
809 	INP_HASH_LOCK_ASSERT(pcbinfo);
810 
811 	if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
812 		return (EADDRNOTAVAIL);
813 	laddr.s_addr = *laddrp;
814 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
815 		return (EINVAL);
816 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
817 		lookupflags = INPLOOKUP_WILDCARD;
818 	if (nam == NULL) {
819 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
820 			return (error);
821 	} else {
822 		sin = (struct sockaddr_in *)nam;
823 		if (nam->sa_len != sizeof (*sin))
824 			return (EINVAL);
825 #ifdef notdef
826 		/*
827 		 * We should check the family, but old programs
828 		 * incorrectly fail to initialize it.
829 		 */
830 		if (sin->sin_family != AF_INET)
831 			return (EAFNOSUPPORT);
832 #endif
833 		error = prison_local_ip4(cred, &sin->sin_addr);
834 		if (error)
835 			return (error);
836 		if (sin->sin_port != *lportp) {
837 			/* Don't allow the port to change. */
838 			if (*lportp != 0)
839 				return (EINVAL);
840 			lport = sin->sin_port;
841 		}
842 		/* NB: lport is left as 0 if the port isn't being changed. */
843 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
844 			/*
845 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
846 			 * allow complete duplication of binding if
847 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
848 			 * and a multicast address is bound on both
849 			 * new and duplicated sockets.
850 			 */
851 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
852 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
853 			/*
854 			 * XXX: How to deal with SO_REUSEPORT_LB here?
855 			 * Treat same as SO_REUSEPORT for now.
856 			 */
857 			if ((so->so_options &
858 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
859 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
860 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
861 			sin->sin_port = 0;		/* yech... */
862 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
863 			/*
864 			 * Is the address a local IP address?
865 			 * If INP_BINDANY is set, then the socket may be bound
866 			 * to any endpoint address, local or not.
867 			 */
868 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
869 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
870 				return (EADDRNOTAVAIL);
871 		}
872 		laddr = sin->sin_addr;
873 		if (lport) {
874 			struct inpcb *t;
875 			struct tcptw *tw;
876 
877 			/* GROSS */
878 			if (ntohs(lport) <= V_ipport_reservedhigh &&
879 			    ntohs(lport) >= V_ipport_reservedlow &&
880 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
881 			    0))
882 				return (EACCES);
883 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
884 			    priv_check_cred(inp->inp_cred,
885 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
886 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
887 				    lport, INPLOOKUP_WILDCARD, cred);
888 	/*
889 	 * XXX
890 	 * This entire block sorely needs a rewrite.
891 	 */
892 				if (t &&
893 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
894 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
895 				    (so->so_type != SOCK_STREAM ||
896 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
897 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
898 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
899 				     (t->inp_flags2 & INP_REUSEPORT) ||
900 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
901 				    (inp->inp_cred->cr_uid !=
902 				     t->inp_cred->cr_uid))
903 					return (EADDRINUSE);
904 
905 				/*
906 				 * If the socket is a BINDMULTI socket, then
907 				 * the credentials need to match and the
908 				 * original socket also has to have been bound
909 				 * with BINDMULTI.
910 				 */
911 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
912 					return (EADDRINUSE);
913 			}
914 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
915 			    lport, lookupflags, cred);
916 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
917 				/*
918 				 * XXXRW: If an incpb has had its timewait
919 				 * state recycled, we treat the address as
920 				 * being in use (for now).  This is better
921 				 * than a panic, but not desirable.
922 				 */
923 				tw = intotw(t);
924 				if (tw == NULL ||
925 				    ((reuseport & tw->tw_so_options) == 0 &&
926 					(reuseport_lb &
927 				            tw->tw_so_options) == 0)) {
928 					return (EADDRINUSE);
929 				}
930 			} else if (t &&
931 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
932 				   (reuseport & inp_so_options(t)) == 0 &&
933 				   (reuseport_lb & inp_so_options(t)) == 0) {
934 #ifdef INET6
935 				if (ntohl(sin->sin_addr.s_addr) !=
936 				    INADDR_ANY ||
937 				    ntohl(t->inp_laddr.s_addr) !=
938 				    INADDR_ANY ||
939 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
940 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
941 #endif
942 						return (EADDRINUSE);
943 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
944 					return (EADDRINUSE);
945 			}
946 		}
947 	}
948 	if (*lportp != 0)
949 		lport = *lportp;
950 	if (lport == 0) {
951 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
952 		if (error != 0)
953 			return (error);
954 
955 	}
956 	*laddrp = laddr.s_addr;
957 	*lportp = lport;
958 	return (0);
959 }
960 
961 /*
962  * Connect from a socket to a specified address.
963  * Both address and port must be specified in argument sin.
964  * If don't have a local address for this socket yet,
965  * then pick one.
966  */
967 int
968 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
969     struct ucred *cred, struct mbuf *m)
970 {
971 	u_short lport, fport;
972 	in_addr_t laddr, faddr;
973 	int anonport, error;
974 
975 	INP_WLOCK_ASSERT(inp);
976 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
977 
978 	lport = inp->inp_lport;
979 	laddr = inp->inp_laddr.s_addr;
980 	anonport = (lport == 0);
981 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
982 	    NULL, cred);
983 	if (error)
984 		return (error);
985 
986 	/* Do the initial binding of the local address if required. */
987 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
988 		inp->inp_lport = lport;
989 		inp->inp_laddr.s_addr = laddr;
990 		if (in_pcbinshash(inp) != 0) {
991 			inp->inp_laddr.s_addr = INADDR_ANY;
992 			inp->inp_lport = 0;
993 			return (EAGAIN);
994 		}
995 	}
996 
997 	/* Commit the remaining changes. */
998 	inp->inp_lport = lport;
999 	inp->inp_laddr.s_addr = laddr;
1000 	inp->inp_faddr.s_addr = faddr;
1001 	inp->inp_fport = fport;
1002 	in_pcbrehash_mbuf(inp, m);
1003 
1004 	if (anonport)
1005 		inp->inp_flags |= INP_ANONPORT;
1006 	return (0);
1007 }
1008 
1009 int
1010 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1011 {
1012 
1013 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1014 }
1015 
1016 /*
1017  * Do proper source address selection on an unbound socket in case
1018  * of connect. Take jails into account as well.
1019  */
1020 int
1021 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1022     struct ucred *cred)
1023 {
1024 	struct ifaddr *ifa;
1025 	struct sockaddr *sa;
1026 	struct sockaddr_in *sin;
1027 	struct route sro;
1028 	int error;
1029 
1030 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1031 	/*
1032 	 * Bypass source address selection and use the primary jail IP
1033 	 * if requested.
1034 	 */
1035 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1036 		return (0);
1037 
1038 	error = 0;
1039 	bzero(&sro, sizeof(sro));
1040 
1041 	sin = (struct sockaddr_in *)&sro.ro_dst;
1042 	sin->sin_family = AF_INET;
1043 	sin->sin_len = sizeof(struct sockaddr_in);
1044 	sin->sin_addr.s_addr = faddr->s_addr;
1045 
1046 	/*
1047 	 * If route is known our src addr is taken from the i/f,
1048 	 * else punt.
1049 	 *
1050 	 * Find out route to destination.
1051 	 */
1052 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1053 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1054 
1055 	/*
1056 	 * If we found a route, use the address corresponding to
1057 	 * the outgoing interface.
1058 	 *
1059 	 * Otherwise assume faddr is reachable on a directly connected
1060 	 * network and try to find a corresponding interface to take
1061 	 * the source address from.
1062 	 */
1063 	NET_EPOCH_ENTER();
1064 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1065 		struct in_ifaddr *ia;
1066 		struct ifnet *ifp;
1067 
1068 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1069 					inp->inp_socket->so_fibnum));
1070 		if (ia == NULL) {
1071 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1072 						inp->inp_socket->so_fibnum));
1073 
1074 		}
1075 		if (ia == NULL) {
1076 			error = ENETUNREACH;
1077 			goto done;
1078 		}
1079 
1080 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1081 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1082 			goto done;
1083 		}
1084 
1085 		ifp = ia->ia_ifp;
1086 		ia = NULL;
1087 		IF_ADDR_RLOCK(ifp);
1088 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1089 
1090 			sa = ifa->ifa_addr;
1091 			if (sa->sa_family != AF_INET)
1092 				continue;
1093 			sin = (struct sockaddr_in *)sa;
1094 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1095 				ia = (struct in_ifaddr *)ifa;
1096 				break;
1097 			}
1098 		}
1099 		if (ia != NULL) {
1100 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1101 			IF_ADDR_RUNLOCK(ifp);
1102 			goto done;
1103 		}
1104 		IF_ADDR_RUNLOCK(ifp);
1105 
1106 		/* 3. As a last resort return the 'default' jail address. */
1107 		error = prison_get_ip4(cred, laddr);
1108 		goto done;
1109 	}
1110 
1111 	/*
1112 	 * If the outgoing interface on the route found is not
1113 	 * a loopback interface, use the address from that interface.
1114 	 * In case of jails do those three steps:
1115 	 * 1. check if the interface address belongs to the jail. If so use it.
1116 	 * 2. check if we have any address on the outgoing interface
1117 	 *    belonging to this jail. If so use it.
1118 	 * 3. as a last resort return the 'default' jail address.
1119 	 */
1120 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1121 		struct in_ifaddr *ia;
1122 		struct ifnet *ifp;
1123 
1124 		/* If not jailed, use the default returned. */
1125 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1126 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1127 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1128 			goto done;
1129 		}
1130 
1131 		/* Jailed. */
1132 		/* 1. Check if the iface address belongs to the jail. */
1133 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1134 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1135 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1136 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1137 			goto done;
1138 		}
1139 
1140 		/*
1141 		 * 2. Check if we have any address on the outgoing interface
1142 		 *    belonging to this jail.
1143 		 */
1144 		ia = NULL;
1145 		ifp = sro.ro_rt->rt_ifp;
1146 		IF_ADDR_RLOCK(ifp);
1147 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1148 			sa = ifa->ifa_addr;
1149 			if (sa->sa_family != AF_INET)
1150 				continue;
1151 			sin = (struct sockaddr_in *)sa;
1152 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1153 				ia = (struct in_ifaddr *)ifa;
1154 				break;
1155 			}
1156 		}
1157 		if (ia != NULL) {
1158 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1159 			IF_ADDR_RUNLOCK(ifp);
1160 			goto done;
1161 		}
1162 		IF_ADDR_RUNLOCK(ifp);
1163 
1164 		/* 3. As a last resort return the 'default' jail address. */
1165 		error = prison_get_ip4(cred, laddr);
1166 		goto done;
1167 	}
1168 
1169 	/*
1170 	 * The outgoing interface is marked with 'loopback net', so a route
1171 	 * to ourselves is here.
1172 	 * Try to find the interface of the destination address and then
1173 	 * take the address from there. That interface is not necessarily
1174 	 * a loopback interface.
1175 	 * In case of jails, check that it is an address of the jail
1176 	 * and if we cannot find, fall back to the 'default' jail address.
1177 	 */
1178 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1179 		struct sockaddr_in sain;
1180 		struct in_ifaddr *ia;
1181 
1182 		bzero(&sain, sizeof(struct sockaddr_in));
1183 		sain.sin_family = AF_INET;
1184 		sain.sin_len = sizeof(struct sockaddr_in);
1185 		sain.sin_addr.s_addr = faddr->s_addr;
1186 
1187 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1188 					inp->inp_socket->so_fibnum));
1189 		if (ia == NULL)
1190 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1191 						inp->inp_socket->so_fibnum));
1192 		if (ia == NULL)
1193 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1194 
1195 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1196 			if (ia == NULL) {
1197 				error = ENETUNREACH;
1198 				goto done;
1199 			}
1200 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1201 			goto done;
1202 		}
1203 
1204 		/* Jailed. */
1205 		if (ia != NULL) {
1206 			struct ifnet *ifp;
1207 
1208 			ifp = ia->ia_ifp;
1209 			ia = NULL;
1210 			IF_ADDR_RLOCK(ifp);
1211 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1212 
1213 				sa = ifa->ifa_addr;
1214 				if (sa->sa_family != AF_INET)
1215 					continue;
1216 				sin = (struct sockaddr_in *)sa;
1217 				if (prison_check_ip4(cred,
1218 				    &sin->sin_addr) == 0) {
1219 					ia = (struct in_ifaddr *)ifa;
1220 					break;
1221 				}
1222 			}
1223 			if (ia != NULL) {
1224 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1225 				IF_ADDR_RUNLOCK(ifp);
1226 				goto done;
1227 			}
1228 			IF_ADDR_RUNLOCK(ifp);
1229 		}
1230 
1231 		/* 3. As a last resort return the 'default' jail address. */
1232 		error = prison_get_ip4(cred, laddr);
1233 		goto done;
1234 	}
1235 
1236 done:
1237 	NET_EPOCH_EXIT();
1238 	if (sro.ro_rt != NULL)
1239 		RTFREE(sro.ro_rt);
1240 	return (error);
1241 }
1242 
1243 /*
1244  * Set up for a connect from a socket to the specified address.
1245  * On entry, *laddrp and *lportp should contain the current local
1246  * address and port for the PCB; these are updated to the values
1247  * that should be placed in inp_laddr and inp_lport to complete
1248  * the connect.
1249  *
1250  * On success, *faddrp and *fportp will be set to the remote address
1251  * and port. These are not updated in the error case.
1252  *
1253  * If the operation fails because the connection already exists,
1254  * *oinpp will be set to the PCB of that connection so that the
1255  * caller can decide to override it. In all other cases, *oinpp
1256  * is set to NULL.
1257  */
1258 int
1259 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1260     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1261     struct inpcb **oinpp, struct ucred *cred)
1262 {
1263 	struct rm_priotracker in_ifa_tracker;
1264 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1265 	struct in_ifaddr *ia;
1266 	struct inpcb *oinp;
1267 	struct in_addr laddr, faddr;
1268 	u_short lport, fport;
1269 	int error;
1270 
1271 	/*
1272 	 * Because a global state change doesn't actually occur here, a read
1273 	 * lock is sufficient.
1274 	 */
1275 	INP_LOCK_ASSERT(inp);
1276 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1277 
1278 	if (oinpp != NULL)
1279 		*oinpp = NULL;
1280 	if (nam->sa_len != sizeof (*sin))
1281 		return (EINVAL);
1282 	if (sin->sin_family != AF_INET)
1283 		return (EAFNOSUPPORT);
1284 	if (sin->sin_port == 0)
1285 		return (EADDRNOTAVAIL);
1286 	laddr.s_addr = *laddrp;
1287 	lport = *lportp;
1288 	faddr = sin->sin_addr;
1289 	fport = sin->sin_port;
1290 
1291 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1292 		/*
1293 		 * If the destination address is INADDR_ANY,
1294 		 * use the primary local address.
1295 		 * If the supplied address is INADDR_BROADCAST,
1296 		 * and the primary interface supports broadcast,
1297 		 * choose the broadcast address for that interface.
1298 		 */
1299 		if (faddr.s_addr == INADDR_ANY) {
1300 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1301 			faddr =
1302 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1303 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1304 			if (cred != NULL &&
1305 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1306 				return (error);
1307 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1308 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1309 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1310 			    IFF_BROADCAST)
1311 				faddr = satosin(&CK_STAILQ_FIRST(
1312 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1313 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1314 		}
1315 	}
1316 	if (laddr.s_addr == INADDR_ANY) {
1317 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1318 		/*
1319 		 * If the destination address is multicast and an outgoing
1320 		 * interface has been set as a multicast option, prefer the
1321 		 * address of that interface as our source address.
1322 		 */
1323 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1324 		    inp->inp_moptions != NULL) {
1325 			struct ip_moptions *imo;
1326 			struct ifnet *ifp;
1327 
1328 			imo = inp->inp_moptions;
1329 			if (imo->imo_multicast_ifp != NULL) {
1330 				ifp = imo->imo_multicast_ifp;
1331 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1332 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1333 					if ((ia->ia_ifp == ifp) &&
1334 					    (cred == NULL ||
1335 					    prison_check_ip4(cred,
1336 					    &ia->ia_addr.sin_addr) == 0))
1337 						break;
1338 				}
1339 				if (ia == NULL)
1340 					error = EADDRNOTAVAIL;
1341 				else {
1342 					laddr = ia->ia_addr.sin_addr;
1343 					error = 0;
1344 				}
1345 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1346 			}
1347 		}
1348 		if (error)
1349 			return (error);
1350 	}
1351 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1352 	    laddr, lport, 0, NULL);
1353 	if (oinp != NULL) {
1354 		if (oinpp != NULL)
1355 			*oinpp = oinp;
1356 		return (EADDRINUSE);
1357 	}
1358 	if (lport == 0) {
1359 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1360 		    cred);
1361 		if (error)
1362 			return (error);
1363 	}
1364 	*laddrp = laddr.s_addr;
1365 	*lportp = lport;
1366 	*faddrp = faddr.s_addr;
1367 	*fportp = fport;
1368 	return (0);
1369 }
1370 
1371 void
1372 in_pcbdisconnect(struct inpcb *inp)
1373 {
1374 
1375 	INP_WLOCK_ASSERT(inp);
1376 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1377 
1378 	inp->inp_faddr.s_addr = INADDR_ANY;
1379 	inp->inp_fport = 0;
1380 	in_pcbrehash(inp);
1381 }
1382 #endif /* INET */
1383 
1384 /*
1385  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1386  * For most protocols, this will be invoked immediately prior to calling
1387  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1388  * socket, in which case in_pcbfree() is deferred.
1389  */
1390 void
1391 in_pcbdetach(struct inpcb *inp)
1392 {
1393 
1394 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1395 
1396 #ifdef RATELIMIT
1397 	if (inp->inp_snd_tag != NULL)
1398 		in_pcbdetach_txrtlmt(inp);
1399 #endif
1400 	inp->inp_socket->so_pcb = NULL;
1401 	inp->inp_socket = NULL;
1402 }
1403 
1404 /*
1405  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1406  * stability of an inpcb pointer despite the inpcb lock being released.  This
1407  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1408  * but where the inpcb lock may already held, or when acquiring a reference
1409  * via a pcbgroup.
1410  *
1411  * in_pcbref() should be used only to provide brief memory stability, and
1412  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1413  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1414  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1415  * lock and rele are the *only* safe operations that may be performed on the
1416  * inpcb.
1417  *
1418  * While the inpcb will not be freed, releasing the inpcb lock means that the
1419  * connection's state may change, so the caller should be careful to
1420  * revalidate any cached state on reacquiring the lock.  Drop the reference
1421  * using in_pcbrele().
1422  */
1423 void
1424 in_pcbref(struct inpcb *inp)
1425 {
1426 
1427 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1428 
1429 	refcount_acquire(&inp->inp_refcount);
1430 }
1431 
1432 /*
1433  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1434  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1435  * return a flag indicating whether or not the inpcb remains valid.  If it is
1436  * valid, we return with the inpcb lock held.
1437  *
1438  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1439  * reference on an inpcb.  Historically more work was done here (actually, in
1440  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1441  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1442  * about memory stability (and continued use of the write lock).
1443  */
1444 int
1445 in_pcbrele_rlocked(struct inpcb *inp)
1446 {
1447 	struct inpcbinfo *pcbinfo;
1448 
1449 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1450 
1451 	INP_RLOCK_ASSERT(inp);
1452 
1453 	if (refcount_release(&inp->inp_refcount) == 0) {
1454 		/*
1455 		 * If the inpcb has been freed, let the caller know, even if
1456 		 * this isn't the last reference.
1457 		 */
1458 		if (inp->inp_flags2 & INP_FREED) {
1459 			INP_RUNLOCK(inp);
1460 			return (1);
1461 		}
1462 		return (0);
1463 	}
1464 
1465 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1466 #ifdef TCPHPTS
1467 	if (inp->inp_in_hpts || inp->inp_in_input) {
1468 		struct tcp_hpts_entry *hpts;
1469 		/*
1470 		 * We should not be on the hpts at
1471 		 * this point in any form. we must
1472 		 * get the lock to be sure.
1473 		 */
1474 		hpts = tcp_hpts_lock(inp);
1475 		if (inp->inp_in_hpts)
1476 			panic("Hpts:%p inp:%p at free still on hpts",
1477 			      hpts, inp);
1478 		mtx_unlock(&hpts->p_mtx);
1479 		hpts = tcp_input_lock(inp);
1480 		if (inp->inp_in_input)
1481 			panic("Hpts:%p inp:%p at free still on input hpts",
1482 			      hpts, inp);
1483 		mtx_unlock(&hpts->p_mtx);
1484 	}
1485 #endif
1486 	INP_RUNLOCK(inp);
1487 	pcbinfo = inp->inp_pcbinfo;
1488 	uma_zfree(pcbinfo->ipi_zone, inp);
1489 	return (1);
1490 }
1491 
1492 int
1493 in_pcbrele_wlocked(struct inpcb *inp)
1494 {
1495 	struct inpcbinfo *pcbinfo;
1496 
1497 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1498 
1499 	INP_WLOCK_ASSERT(inp);
1500 
1501 	if (refcount_release(&inp->inp_refcount) == 0) {
1502 		/*
1503 		 * If the inpcb has been freed, let the caller know, even if
1504 		 * this isn't the last reference.
1505 		 */
1506 		if (inp->inp_flags2 & INP_FREED) {
1507 			INP_WUNLOCK(inp);
1508 			return (1);
1509 		}
1510 		return (0);
1511 	}
1512 
1513 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1514 #ifdef TCPHPTS
1515 	if (inp->inp_in_hpts || inp->inp_in_input) {
1516 		struct tcp_hpts_entry *hpts;
1517 		/*
1518 		 * We should not be on the hpts at
1519 		 * this point in any form. we must
1520 		 * get the lock to be sure.
1521 		 */
1522 		hpts = tcp_hpts_lock(inp);
1523 		if (inp->inp_in_hpts)
1524 			panic("Hpts:%p inp:%p at free still on hpts",
1525 			      hpts, inp);
1526 		mtx_unlock(&hpts->p_mtx);
1527 		hpts = tcp_input_lock(inp);
1528 		if (inp->inp_in_input)
1529 			panic("Hpts:%p inp:%p at free still on input hpts",
1530 			      hpts, inp);
1531 		mtx_unlock(&hpts->p_mtx);
1532 	}
1533 #endif
1534 	INP_WUNLOCK(inp);
1535 	pcbinfo = inp->inp_pcbinfo;
1536 	uma_zfree(pcbinfo->ipi_zone, inp);
1537 	return (1);
1538 }
1539 
1540 /*
1541  * Temporary wrapper.
1542  */
1543 int
1544 in_pcbrele(struct inpcb *inp)
1545 {
1546 
1547 	return (in_pcbrele_wlocked(inp));
1548 }
1549 
1550 void
1551 in_pcblist_rele_rlocked(epoch_context_t ctx)
1552 {
1553 	struct in_pcblist *il;
1554 	struct inpcb *inp;
1555 	struct inpcbinfo *pcbinfo;
1556 	int i, n;
1557 
1558 	il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1559 	pcbinfo = il->il_pcbinfo;
1560 	n = il->il_count;
1561 	INP_INFO_WLOCK(pcbinfo);
1562 	for (i = 0; i < n; i++) {
1563 		inp = il->il_inp_list[i];
1564 		INP_RLOCK(inp);
1565 		if (!in_pcbrele_rlocked(inp))
1566 			INP_RUNLOCK(inp);
1567 	}
1568 	INP_INFO_WUNLOCK(pcbinfo);
1569 	free(il, M_TEMP);
1570 }
1571 
1572 static void
1573 inpcbport_free(epoch_context_t ctx)
1574 {
1575 	struct inpcbport *phd;
1576 
1577 	phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1578 	free(phd, M_PCB);
1579 }
1580 
1581 static void
1582 in_pcbfree_deferred(epoch_context_t ctx)
1583 {
1584 	struct inpcb *inp;
1585 	int released __unused;
1586 
1587 	inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1588 
1589 	INP_WLOCK(inp);
1590 #ifdef INET
1591 	inp_freemoptions(inp->inp_moptions);
1592 	inp->inp_moptions = NULL;
1593 #endif
1594 	/* XXXRW: Do as much as possible here. */
1595 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1596 	if (inp->inp_sp != NULL)
1597 		ipsec_delete_pcbpolicy(inp);
1598 #endif
1599 #ifdef INET6
1600 	if (inp->inp_vflag & INP_IPV6PROTO) {
1601 		ip6_freepcbopts(inp->in6p_outputopts);
1602 		ip6_freemoptions(inp->in6p_moptions);
1603 		inp->in6p_moptions = NULL;
1604 	}
1605 #endif
1606 	if (inp->inp_options)
1607 		(void)m_free(inp->inp_options);
1608 	inp->inp_vflag = 0;
1609 	crfree(inp->inp_cred);
1610 #ifdef MAC
1611 	mac_inpcb_destroy(inp);
1612 #endif
1613 	released = in_pcbrele_wlocked(inp);
1614 	MPASS(released);
1615 }
1616 
1617 /*
1618  * Unconditionally schedule an inpcb to be freed by decrementing its
1619  * reference count, which should occur only after the inpcb has been detached
1620  * from its socket.  If another thread holds a temporary reference (acquired
1621  * using in_pcbref()) then the free is deferred until that reference is
1622  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1623  * work, including removal from global lists, is done in this context, where
1624  * the pcbinfo lock is held.
1625  */
1626 void
1627 in_pcbfree(struct inpcb *inp)
1628 {
1629 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1630 
1631 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1632 	KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1633 	    ("%s: called twice for pcb %p", __func__, inp));
1634 	if (inp->inp_flags2 & INP_FREED) {
1635 		INP_WUNLOCK(inp);
1636 		return;
1637 	}
1638 
1639 #ifdef INVARIANTS
1640 	if (pcbinfo == &V_tcbinfo) {
1641 		INP_INFO_LOCK_ASSERT(pcbinfo);
1642 	} else {
1643 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1644 	}
1645 #endif
1646 	INP_WLOCK_ASSERT(inp);
1647 	INP_LIST_WLOCK(pcbinfo);
1648 	in_pcbremlists(inp);
1649 	INP_LIST_WUNLOCK(pcbinfo);
1650 	RO_INVALIDATE_CACHE(&inp->inp_route);
1651 	/* mark as destruction in progress */
1652 	inp->inp_flags2 |= INP_FREED;
1653 	INP_WUNLOCK(inp);
1654 	epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1655 }
1656 
1657 /*
1658  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1659  * port reservation, and preventing it from being returned by inpcb lookups.
1660  *
1661  * It is used by TCP to mark an inpcb as unused and avoid future packet
1662  * delivery or event notification when a socket remains open but TCP has
1663  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1664  * or a RST on the wire, and allows the port binding to be reused while still
1665  * maintaining the invariant that so_pcb always points to a valid inpcb until
1666  * in_pcbdetach().
1667  *
1668  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1669  * in_pcbnotifyall() and in_pcbpurgeif0()?
1670  */
1671 void
1672 in_pcbdrop(struct inpcb *inp)
1673 {
1674 
1675 	INP_WLOCK_ASSERT(inp);
1676 
1677 	/*
1678 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1679 	 * the hash lock...?
1680 	 */
1681 	inp->inp_flags |= INP_DROPPED;
1682 	if (inp->inp_flags & INP_INHASHLIST) {
1683 		struct inpcbport *phd = inp->inp_phd;
1684 
1685 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1686 		in_pcbremlbgrouphash(inp);
1687 		CK_LIST_REMOVE(inp, inp_hash);
1688 		CK_LIST_REMOVE(inp, inp_portlist);
1689 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1690 			CK_LIST_REMOVE(phd, phd_hash);
1691 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1692 		}
1693 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1694 		inp->inp_flags &= ~INP_INHASHLIST;
1695 #ifdef PCBGROUP
1696 		in_pcbgroup_remove(inp);
1697 #endif
1698 	}
1699 }
1700 
1701 #ifdef INET
1702 /*
1703  * Common routines to return the socket addresses associated with inpcbs.
1704  */
1705 struct sockaddr *
1706 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1707 {
1708 	struct sockaddr_in *sin;
1709 
1710 	sin = malloc(sizeof *sin, M_SONAME,
1711 		M_WAITOK | M_ZERO);
1712 	sin->sin_family = AF_INET;
1713 	sin->sin_len = sizeof(*sin);
1714 	sin->sin_addr = *addr_p;
1715 	sin->sin_port = port;
1716 
1717 	return (struct sockaddr *)sin;
1718 }
1719 
1720 int
1721 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1722 {
1723 	struct inpcb *inp;
1724 	struct in_addr addr;
1725 	in_port_t port;
1726 
1727 	inp = sotoinpcb(so);
1728 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1729 
1730 	INP_RLOCK(inp);
1731 	port = inp->inp_lport;
1732 	addr = inp->inp_laddr;
1733 	INP_RUNLOCK(inp);
1734 
1735 	*nam = in_sockaddr(port, &addr);
1736 	return 0;
1737 }
1738 
1739 int
1740 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1741 {
1742 	struct inpcb *inp;
1743 	struct in_addr addr;
1744 	in_port_t port;
1745 
1746 	inp = sotoinpcb(so);
1747 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1748 
1749 	INP_RLOCK(inp);
1750 	port = inp->inp_fport;
1751 	addr = inp->inp_faddr;
1752 	INP_RUNLOCK(inp);
1753 
1754 	*nam = in_sockaddr(port, &addr);
1755 	return 0;
1756 }
1757 
1758 void
1759 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1760     struct inpcb *(*notify)(struct inpcb *, int))
1761 {
1762 	struct inpcb *inp, *inp_temp;
1763 
1764 	INP_INFO_WLOCK(pcbinfo);
1765 	CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1766 		INP_WLOCK(inp);
1767 #ifdef INET6
1768 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1769 			INP_WUNLOCK(inp);
1770 			continue;
1771 		}
1772 #endif
1773 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1774 		    inp->inp_socket == NULL) {
1775 			INP_WUNLOCK(inp);
1776 			continue;
1777 		}
1778 		if ((*notify)(inp, errno))
1779 			INP_WUNLOCK(inp);
1780 	}
1781 	INP_INFO_WUNLOCK(pcbinfo);
1782 }
1783 
1784 void
1785 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1786 {
1787 	struct inpcb *inp;
1788 	struct ip_moptions *imo;
1789 	int i, gap;
1790 
1791 	INP_INFO_WLOCK(pcbinfo);
1792 	CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1793 		INP_WLOCK(inp);
1794 		imo = inp->inp_moptions;
1795 		if ((inp->inp_vflag & INP_IPV4) &&
1796 		    imo != NULL) {
1797 			/*
1798 			 * Unselect the outgoing interface if it is being
1799 			 * detached.
1800 			 */
1801 			if (imo->imo_multicast_ifp == ifp)
1802 				imo->imo_multicast_ifp = NULL;
1803 
1804 			/*
1805 			 * Drop multicast group membership if we joined
1806 			 * through the interface being detached.
1807 			 *
1808 			 * XXX This can all be deferred to an epoch_call
1809 			 */
1810 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1811 			    i++) {
1812 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1813 					IN_MULTI_LOCK_ASSERT();
1814 					in_leavegroup_locked(imo->imo_membership[i], NULL);
1815 					gap++;
1816 				} else if (gap != 0)
1817 					imo->imo_membership[i - gap] =
1818 					    imo->imo_membership[i];
1819 			}
1820 			imo->imo_num_memberships -= gap;
1821 		}
1822 		INP_WUNLOCK(inp);
1823 	}
1824 	INP_INFO_WUNLOCK(pcbinfo);
1825 }
1826 
1827 /*
1828  * Lookup a PCB based on the local address and port.  Caller must hold the
1829  * hash lock.  No inpcb locks or references are acquired.
1830  */
1831 #define INP_LOOKUP_MAPPED_PCB_COST	3
1832 struct inpcb *
1833 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1834     u_short lport, int lookupflags, struct ucred *cred)
1835 {
1836 	struct inpcb *inp;
1837 #ifdef INET6
1838 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1839 #else
1840 	int matchwild = 3;
1841 #endif
1842 	int wildcard;
1843 
1844 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1845 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1846 
1847 	INP_HASH_LOCK_ASSERT(pcbinfo);
1848 
1849 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1850 		struct inpcbhead *head;
1851 		/*
1852 		 * Look for an unconnected (wildcard foreign addr) PCB that
1853 		 * matches the local address and port we're looking for.
1854 		 */
1855 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1856 		    0, pcbinfo->ipi_hashmask)];
1857 		CK_LIST_FOREACH(inp, head, inp_hash) {
1858 #ifdef INET6
1859 			/* XXX inp locking */
1860 			if ((inp->inp_vflag & INP_IPV4) == 0)
1861 				continue;
1862 #endif
1863 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1864 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1865 			    inp->inp_lport == lport) {
1866 				/*
1867 				 * Found?
1868 				 */
1869 				if (cred == NULL ||
1870 				    prison_equal_ip4(cred->cr_prison,
1871 					inp->inp_cred->cr_prison))
1872 					return (inp);
1873 			}
1874 		}
1875 		/*
1876 		 * Not found.
1877 		 */
1878 		return (NULL);
1879 	} else {
1880 		struct inpcbporthead *porthash;
1881 		struct inpcbport *phd;
1882 		struct inpcb *match = NULL;
1883 		/*
1884 		 * Best fit PCB lookup.
1885 		 *
1886 		 * First see if this local port is in use by looking on the
1887 		 * port hash list.
1888 		 */
1889 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1890 		    pcbinfo->ipi_porthashmask)];
1891 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
1892 			if (phd->phd_port == lport)
1893 				break;
1894 		}
1895 		if (phd != NULL) {
1896 			/*
1897 			 * Port is in use by one or more PCBs. Look for best
1898 			 * fit.
1899 			 */
1900 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1901 				wildcard = 0;
1902 				if (cred != NULL &&
1903 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1904 					cred->cr_prison))
1905 					continue;
1906 #ifdef INET6
1907 				/* XXX inp locking */
1908 				if ((inp->inp_vflag & INP_IPV4) == 0)
1909 					continue;
1910 				/*
1911 				 * We never select the PCB that has
1912 				 * INP_IPV6 flag and is bound to :: if
1913 				 * we have another PCB which is bound
1914 				 * to 0.0.0.0.  If a PCB has the
1915 				 * INP_IPV6 flag, then we set its cost
1916 				 * higher than IPv4 only PCBs.
1917 				 *
1918 				 * Note that the case only happens
1919 				 * when a socket is bound to ::, under
1920 				 * the condition that the use of the
1921 				 * mapped address is allowed.
1922 				 */
1923 				if ((inp->inp_vflag & INP_IPV6) != 0)
1924 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1925 #endif
1926 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1927 					wildcard++;
1928 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1929 					if (laddr.s_addr == INADDR_ANY)
1930 						wildcard++;
1931 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1932 						continue;
1933 				} else {
1934 					if (laddr.s_addr != INADDR_ANY)
1935 						wildcard++;
1936 				}
1937 				if (wildcard < matchwild) {
1938 					match = inp;
1939 					matchwild = wildcard;
1940 					if (matchwild == 0)
1941 						break;
1942 				}
1943 			}
1944 		}
1945 		return (match);
1946 	}
1947 }
1948 #undef INP_LOOKUP_MAPPED_PCB_COST
1949 
1950 static struct inpcb *
1951 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1952   const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1953   uint16_t fport, int lookupflags)
1954 {
1955 	struct inpcb *local_wild = NULL;
1956 	const struct inpcblbgrouphead *hdr;
1957 	struct inpcblbgroup *grp;
1958 	struct inpcblbgroup *grp_local_wild;
1959 
1960 	INP_HASH_LOCK_ASSERT(pcbinfo);
1961 
1962 	hdr = &pcbinfo->ipi_lbgrouphashbase[
1963 		  INP_PCBLBGROUP_PORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1964 
1965 	/*
1966 	 * Order of socket selection:
1967 	 * 1. non-wild.
1968 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1969 	 *
1970 	 * NOTE:
1971 	 * - Load balanced group does not contain jailed sockets
1972 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1973 	 */
1974 	LIST_FOREACH(grp, hdr, il_list) {
1975 #ifdef INET6
1976 		if (!(grp->il_vflag & INP_IPV4))
1977 			continue;
1978 #endif
1979 
1980 		if (grp->il_lport == lport) {
1981 
1982 			uint32_t idx = 0;
1983 			int pkt_hash = INP_PCBLBGROUP_PKTHASH(faddr->s_addr,
1984 			    lport, fport);
1985 
1986 			idx = pkt_hash % grp->il_inpcnt;
1987 
1988 			if (grp->il_laddr.s_addr == laddr->s_addr) {
1989 				return (grp->il_inp[idx]);
1990 			} else {
1991 				if (grp->il_laddr.s_addr == INADDR_ANY &&
1992 					(lookupflags & INPLOOKUP_WILDCARD)) {
1993 					local_wild = grp->il_inp[idx];
1994 					grp_local_wild = grp;
1995 				}
1996 			}
1997 		}
1998 	}
1999 	if (local_wild != NULL) {
2000 		return (local_wild);
2001 	}
2002 	return (NULL);
2003 }
2004 
2005 #ifdef PCBGROUP
2006 /*
2007  * Lookup PCB in hash list, using pcbgroup tables.
2008  */
2009 static struct inpcb *
2010 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2011     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2012     u_int lport_arg, int lookupflags, struct ifnet *ifp)
2013 {
2014 	struct inpcbhead *head;
2015 	struct inpcb *inp, *tmpinp;
2016 	u_short fport = fport_arg, lport = lport_arg;
2017 	bool locked;
2018 
2019 	/*
2020 	 * First look for an exact match.
2021 	 */
2022 	tmpinp = NULL;
2023 	INP_GROUP_LOCK(pcbgroup);
2024 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2025 	    pcbgroup->ipg_hashmask)];
2026 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2027 #ifdef INET6
2028 		/* XXX inp locking */
2029 		if ((inp->inp_vflag & INP_IPV4) == 0)
2030 			continue;
2031 #endif
2032 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2033 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2034 		    inp->inp_fport == fport &&
2035 		    inp->inp_lport == lport) {
2036 			/*
2037 			 * XXX We should be able to directly return
2038 			 * the inp here, without any checks.
2039 			 * Well unless both bound with SO_REUSEPORT?
2040 			 */
2041 			if (prison_flag(inp->inp_cred, PR_IP4))
2042 				goto found;
2043 			if (tmpinp == NULL)
2044 				tmpinp = inp;
2045 		}
2046 	}
2047 	if (tmpinp != NULL) {
2048 		inp = tmpinp;
2049 		goto found;
2050 	}
2051 
2052 #ifdef	RSS
2053 	/*
2054 	 * For incoming connections, we may wish to do a wildcard
2055 	 * match for an RSS-local socket.
2056 	 */
2057 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2058 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2059 #ifdef INET6
2060 		struct inpcb *local_wild_mapped = NULL;
2061 #endif
2062 		struct inpcb *jail_wild = NULL;
2063 		struct inpcbhead *head;
2064 		int injail;
2065 
2066 		/*
2067 		 * Order of socket selection - we always prefer jails.
2068 		 *      1. jailed, non-wild.
2069 		 *      2. jailed, wild.
2070 		 *      3. non-jailed, non-wild.
2071 		 *      4. non-jailed, wild.
2072 		 */
2073 
2074 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2075 		    lport, 0, pcbgroup->ipg_hashmask)];
2076 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2077 #ifdef INET6
2078 			/* XXX inp locking */
2079 			if ((inp->inp_vflag & INP_IPV4) == 0)
2080 				continue;
2081 #endif
2082 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2083 			    inp->inp_lport != lport)
2084 				continue;
2085 
2086 			injail = prison_flag(inp->inp_cred, PR_IP4);
2087 			if (injail) {
2088 				if (prison_check_ip4(inp->inp_cred,
2089 				    &laddr) != 0)
2090 					continue;
2091 			} else {
2092 				if (local_exact != NULL)
2093 					continue;
2094 			}
2095 
2096 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2097 				if (injail)
2098 					goto found;
2099 				else
2100 					local_exact = inp;
2101 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2102 #ifdef INET6
2103 				/* XXX inp locking, NULL check */
2104 				if (inp->inp_vflag & INP_IPV6PROTO)
2105 					local_wild_mapped = inp;
2106 				else
2107 #endif
2108 					if (injail)
2109 						jail_wild = inp;
2110 					else
2111 						local_wild = inp;
2112 			}
2113 		} /* LIST_FOREACH */
2114 
2115 		inp = jail_wild;
2116 		if (inp == NULL)
2117 			inp = local_exact;
2118 		if (inp == NULL)
2119 			inp = local_wild;
2120 #ifdef INET6
2121 		if (inp == NULL)
2122 			inp = local_wild_mapped;
2123 #endif
2124 		if (inp != NULL)
2125 			goto found;
2126 	}
2127 #endif
2128 
2129 	/*
2130 	 * Then look for a wildcard match, if requested.
2131 	 */
2132 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2133 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2134 #ifdef INET6
2135 		struct inpcb *local_wild_mapped = NULL;
2136 #endif
2137 		struct inpcb *jail_wild = NULL;
2138 		struct inpcbhead *head;
2139 		int injail;
2140 
2141 		/*
2142 		 * Order of socket selection - we always prefer jails.
2143 		 *      1. jailed, non-wild.
2144 		 *      2. jailed, wild.
2145 		 *      3. non-jailed, non-wild.
2146 		 *      4. non-jailed, wild.
2147 		 */
2148 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2149 		    0, pcbinfo->ipi_wildmask)];
2150 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2151 #ifdef INET6
2152 			/* XXX inp locking */
2153 			if ((inp->inp_vflag & INP_IPV4) == 0)
2154 				continue;
2155 #endif
2156 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2157 			    inp->inp_lport != lport)
2158 				continue;
2159 
2160 			injail = prison_flag(inp->inp_cred, PR_IP4);
2161 			if (injail) {
2162 				if (prison_check_ip4(inp->inp_cred,
2163 				    &laddr) != 0)
2164 					continue;
2165 			} else {
2166 				if (local_exact != NULL)
2167 					continue;
2168 			}
2169 
2170 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2171 				if (injail)
2172 					goto found;
2173 				else
2174 					local_exact = inp;
2175 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2176 #ifdef INET6
2177 				/* XXX inp locking, NULL check */
2178 				if (inp->inp_vflag & INP_IPV6PROTO)
2179 					local_wild_mapped = inp;
2180 				else
2181 #endif
2182 					if (injail)
2183 						jail_wild = inp;
2184 					else
2185 						local_wild = inp;
2186 			}
2187 		} /* LIST_FOREACH */
2188 		inp = jail_wild;
2189 		if (inp == NULL)
2190 			inp = local_exact;
2191 		if (inp == NULL)
2192 			inp = local_wild;
2193 #ifdef INET6
2194 		if (inp == NULL)
2195 			inp = local_wild_mapped;
2196 #endif
2197 		if (inp != NULL)
2198 			goto found;
2199 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2200 	INP_GROUP_UNLOCK(pcbgroup);
2201 	return (NULL);
2202 
2203 found:
2204 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2205 		locked = INP_TRY_WLOCK(inp);
2206 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2207 		locked = INP_TRY_RLOCK(inp);
2208 	else
2209 		panic("%s: locking bug", __func__);
2210 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2211 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2212 			INP_WUNLOCK(inp);
2213 		else
2214 			INP_RUNLOCK(inp);
2215 		return (NULL);
2216 	} else if (!locked)
2217 		in_pcbref(inp);
2218 	INP_GROUP_UNLOCK(pcbgroup);
2219 	if (!locked) {
2220 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2221 			INP_WLOCK(inp);
2222 			if (in_pcbrele_wlocked(inp))
2223 				return (NULL);
2224 		} else {
2225 			INP_RLOCK(inp);
2226 			if (in_pcbrele_rlocked(inp))
2227 				return (NULL);
2228 		}
2229 	}
2230 #ifdef INVARIANTS
2231 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2232 		INP_WLOCK_ASSERT(inp);
2233 	else
2234 		INP_RLOCK_ASSERT(inp);
2235 #endif
2236 	return (inp);
2237 }
2238 #endif /* PCBGROUP */
2239 
2240 /*
2241  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2242  * that the caller has locked the hash list, and will not perform any further
2243  * locking or reference operations on either the hash list or the connection.
2244  */
2245 static struct inpcb *
2246 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2247     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2248     struct ifnet *ifp)
2249 {
2250 	struct inpcbhead *head;
2251 	struct inpcb *inp, *tmpinp;
2252 	u_short fport = fport_arg, lport = lport_arg;
2253 
2254 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2255 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2256 
2257 	INP_HASH_LOCK_ASSERT(pcbinfo);
2258 
2259 	/*
2260 	 * First look for an exact match.
2261 	 */
2262 	tmpinp = NULL;
2263 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2264 	    pcbinfo->ipi_hashmask)];
2265 	CK_LIST_FOREACH(inp, head, inp_hash) {
2266 #ifdef INET6
2267 		/* XXX inp locking */
2268 		if ((inp->inp_vflag & INP_IPV4) == 0)
2269 			continue;
2270 #endif
2271 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2272 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2273 		    inp->inp_fport == fport &&
2274 		    inp->inp_lport == lport) {
2275 			/*
2276 			 * XXX We should be able to directly return
2277 			 * the inp here, without any checks.
2278 			 * Well unless both bound with SO_REUSEPORT?
2279 			 */
2280 			if (prison_flag(inp->inp_cred, PR_IP4))
2281 				return (inp);
2282 			if (tmpinp == NULL)
2283 				tmpinp = inp;
2284 		}
2285 	}
2286 	if (tmpinp != NULL)
2287 		return (tmpinp);
2288 
2289 	/*
2290 	 * Then look in lb group (for wildcard match).
2291 	 */
2292 	if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2293 		(lookupflags & INPLOOKUP_WILDCARD)) {
2294 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2295 		    fport, lookupflags);
2296 		if (inp != NULL) {
2297 			return (inp);
2298 		}
2299 	}
2300 
2301 	/*
2302 	 * Then look for a wildcard match, if requested.
2303 	 */
2304 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2305 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2306 #ifdef INET6
2307 		struct inpcb *local_wild_mapped = NULL;
2308 #endif
2309 		struct inpcb *jail_wild = NULL;
2310 		int injail;
2311 
2312 		/*
2313 		 * Order of socket selection - we always prefer jails.
2314 		 *      1. jailed, non-wild.
2315 		 *      2. jailed, wild.
2316 		 *      3. non-jailed, non-wild.
2317 		 *      4. non-jailed, wild.
2318 		 */
2319 
2320 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2321 		    0, pcbinfo->ipi_hashmask)];
2322 		CK_LIST_FOREACH(inp, head, inp_hash) {
2323 #ifdef INET6
2324 			/* XXX inp locking */
2325 			if ((inp->inp_vflag & INP_IPV4) == 0)
2326 				continue;
2327 #endif
2328 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2329 			    inp->inp_lport != lport)
2330 				continue;
2331 
2332 			injail = prison_flag(inp->inp_cred, PR_IP4);
2333 			if (injail) {
2334 				if (prison_check_ip4(inp->inp_cred,
2335 				    &laddr) != 0)
2336 					continue;
2337 			} else {
2338 				if (local_exact != NULL)
2339 					continue;
2340 			}
2341 
2342 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2343 				if (injail)
2344 					return (inp);
2345 				else
2346 					local_exact = inp;
2347 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2348 #ifdef INET6
2349 				/* XXX inp locking, NULL check */
2350 				if (inp->inp_vflag & INP_IPV6PROTO)
2351 					local_wild_mapped = inp;
2352 				else
2353 #endif
2354 					if (injail)
2355 						jail_wild = inp;
2356 					else
2357 						local_wild = inp;
2358 			}
2359 		} /* LIST_FOREACH */
2360 		if (jail_wild != NULL)
2361 			return (jail_wild);
2362 		if (local_exact != NULL)
2363 			return (local_exact);
2364 		if (local_wild != NULL)
2365 			return (local_wild);
2366 #ifdef INET6
2367 		if (local_wild_mapped != NULL)
2368 			return (local_wild_mapped);
2369 #endif
2370 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2371 
2372 	return (NULL);
2373 }
2374 
2375 /*
2376  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2377  * hash list lock, and will return the inpcb locked (i.e., requires
2378  * INPLOOKUP_LOCKPCB).
2379  */
2380 static struct inpcb *
2381 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2382     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2383     struct ifnet *ifp)
2384 {
2385 	struct inpcb *inp;
2386 
2387 	INP_HASH_RLOCK(pcbinfo);
2388 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2389 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2390 	if (inp != NULL) {
2391 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2392 			INP_WLOCK(inp);
2393 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2394 				INP_WUNLOCK(inp);
2395 				inp = NULL;
2396 			}
2397 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2398 			INP_RLOCK(inp);
2399 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2400 				INP_RUNLOCK(inp);
2401 				inp = NULL;
2402 			}
2403 		} else
2404 			panic("%s: locking bug", __func__);
2405 #ifdef INVARIANTS
2406 		if (inp != NULL) {
2407 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2408 				INP_WLOCK_ASSERT(inp);
2409 			else
2410 				INP_RLOCK_ASSERT(inp);
2411 		}
2412 #endif
2413 	}
2414 	INP_HASH_RUNLOCK(pcbinfo);
2415 	return (inp);
2416 }
2417 
2418 /*
2419  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2420  * from which a pre-calculated hash value may be extracted.
2421  *
2422  * Possibly more of this logic should be in in_pcbgroup.c.
2423  */
2424 struct inpcb *
2425 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2426     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2427 {
2428 #if defined(PCBGROUP) && !defined(RSS)
2429 	struct inpcbgroup *pcbgroup;
2430 #endif
2431 
2432 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2433 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2434 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2435 	    ("%s: LOCKPCB not set", __func__));
2436 
2437 	/*
2438 	 * When not using RSS, use connection groups in preference to the
2439 	 * reservation table when looking up 4-tuples.  When using RSS, just
2440 	 * use the reservation table, due to the cost of the Toeplitz hash
2441 	 * in software.
2442 	 *
2443 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2444 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2445 	 * in software.
2446 	 */
2447 #if defined(PCBGROUP) && !defined(RSS)
2448 	if (in_pcbgroup_enabled(pcbinfo)) {
2449 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2450 		    fport);
2451 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2452 		    laddr, lport, lookupflags, ifp));
2453 	}
2454 #endif
2455 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2456 	    lookupflags, ifp));
2457 }
2458 
2459 struct inpcb *
2460 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2461     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2462     struct ifnet *ifp, struct mbuf *m)
2463 {
2464 #ifdef PCBGROUP
2465 	struct inpcbgroup *pcbgroup;
2466 #endif
2467 
2468 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2469 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2470 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2471 	    ("%s: LOCKPCB not set", __func__));
2472 
2473 #ifdef PCBGROUP
2474 	/*
2475 	 * If we can use a hardware-generated hash to look up the connection
2476 	 * group, use that connection group to find the inpcb.  Otherwise
2477 	 * fall back on a software hash -- or the reservation table if we're
2478 	 * using RSS.
2479 	 *
2480 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2481 	 */
2482 	if (in_pcbgroup_enabled(pcbinfo) &&
2483 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2484 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2485 		    m->m_pkthdr.flowid);
2486 		if (pcbgroup != NULL)
2487 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2488 			    fport, laddr, lport, lookupflags, ifp));
2489 #ifndef RSS
2490 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2491 		    fport);
2492 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2493 		    laddr, lport, lookupflags, ifp));
2494 #endif
2495 	}
2496 #endif
2497 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2498 	    lookupflags, ifp));
2499 }
2500 #endif /* INET */
2501 
2502 /*
2503  * Insert PCB onto various hash lists.
2504  */
2505 static int
2506 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2507 {
2508 	struct inpcbhead *pcbhash;
2509 	struct inpcbporthead *pcbporthash;
2510 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2511 	struct inpcbport *phd;
2512 	u_int32_t hashkey_faddr;
2513 	int so_options;
2514 
2515 	INP_WLOCK_ASSERT(inp);
2516 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2517 
2518 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2519 	    ("in_pcbinshash: INP_INHASHLIST"));
2520 
2521 #ifdef INET6
2522 	if (inp->inp_vflag & INP_IPV6)
2523 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2524 	else
2525 #endif
2526 	hashkey_faddr = inp->inp_faddr.s_addr;
2527 
2528 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2529 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2530 
2531 	pcbporthash = &pcbinfo->ipi_porthashbase[
2532 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2533 
2534 	/*
2535 	 * Add entry to load balance group.
2536 	 * Only do this if SO_REUSEPORT_LB is set.
2537 	 */
2538 	so_options = inp_so_options(inp);
2539 	if (so_options & SO_REUSEPORT_LB) {
2540 		int ret = in_pcbinslbgrouphash(inp);
2541 		if (ret) {
2542 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2543 			return (ret);
2544 		}
2545 	}
2546 
2547 	/*
2548 	 * Go through port list and look for a head for this lport.
2549 	 */
2550 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2551 		if (phd->phd_port == inp->inp_lport)
2552 			break;
2553 	}
2554 	/*
2555 	 * If none exists, malloc one and tack it on.
2556 	 */
2557 	if (phd == NULL) {
2558 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2559 		if (phd == NULL) {
2560 			return (ENOBUFS); /* XXX */
2561 		}
2562 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2563 		phd->phd_port = inp->inp_lport;
2564 		CK_LIST_INIT(&phd->phd_pcblist);
2565 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2566 	}
2567 	inp->inp_phd = phd;
2568 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2569 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2570 	inp->inp_flags |= INP_INHASHLIST;
2571 #ifdef PCBGROUP
2572 	if (do_pcbgroup_update)
2573 		in_pcbgroup_update(inp);
2574 #endif
2575 	return (0);
2576 }
2577 
2578 /*
2579  * For now, there are two public interfaces to insert an inpcb into the hash
2580  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2581  * is used only in the TCP syncache, where in_pcbinshash is called before the
2582  * full 4-tuple is set for the inpcb, and we don't want to install in the
2583  * pcbgroup until later.
2584  *
2585  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2586  * connection groups, and partially initialised inpcbs should not be exposed
2587  * to either reservation hash tables or pcbgroups.
2588  */
2589 int
2590 in_pcbinshash(struct inpcb *inp)
2591 {
2592 
2593 	return (in_pcbinshash_internal(inp, 1));
2594 }
2595 
2596 int
2597 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2598 {
2599 
2600 	return (in_pcbinshash_internal(inp, 0));
2601 }
2602 
2603 /*
2604  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2605  * changed. NOTE: This does not handle the case of the lport changing (the
2606  * hashed port list would have to be updated as well), so the lport must
2607  * not change after in_pcbinshash() has been called.
2608  */
2609 void
2610 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2611 {
2612 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2613 	struct inpcbhead *head;
2614 	u_int32_t hashkey_faddr;
2615 
2616 	INP_WLOCK_ASSERT(inp);
2617 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2618 
2619 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2620 	    ("in_pcbrehash: !INP_INHASHLIST"));
2621 
2622 #ifdef INET6
2623 	if (inp->inp_vflag & INP_IPV6)
2624 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2625 	else
2626 #endif
2627 	hashkey_faddr = inp->inp_faddr.s_addr;
2628 
2629 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2630 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2631 
2632 	CK_LIST_REMOVE(inp, inp_hash);
2633 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2634 
2635 #ifdef PCBGROUP
2636 	if (m != NULL)
2637 		in_pcbgroup_update_mbuf(inp, m);
2638 	else
2639 		in_pcbgroup_update(inp);
2640 #endif
2641 }
2642 
2643 void
2644 in_pcbrehash(struct inpcb *inp)
2645 {
2646 
2647 	in_pcbrehash_mbuf(inp, NULL);
2648 }
2649 
2650 /*
2651  * Remove PCB from various lists.
2652  */
2653 static void
2654 in_pcbremlists(struct inpcb *inp)
2655 {
2656 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2657 
2658 #ifdef INVARIANTS
2659 	if (pcbinfo == &V_tcbinfo) {
2660 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2661 	} else {
2662 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2663 	}
2664 #endif
2665 
2666 	INP_WLOCK_ASSERT(inp);
2667 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2668 
2669 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2670 	if (inp->inp_flags & INP_INHASHLIST) {
2671 		struct inpcbport *phd = inp->inp_phd;
2672 
2673 		INP_HASH_WLOCK(pcbinfo);
2674 
2675 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2676 		in_pcbremlbgrouphash(inp);
2677 
2678 		CK_LIST_REMOVE(inp, inp_hash);
2679 		CK_LIST_REMOVE(inp, inp_portlist);
2680 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2681 			CK_LIST_REMOVE(phd, phd_hash);
2682 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2683 		}
2684 		INP_HASH_WUNLOCK(pcbinfo);
2685 		inp->inp_flags &= ~INP_INHASHLIST;
2686 	}
2687 	CK_LIST_REMOVE(inp, inp_list);
2688 	pcbinfo->ipi_count--;
2689 #ifdef PCBGROUP
2690 	in_pcbgroup_remove(inp);
2691 #endif
2692 }
2693 
2694 /*
2695  * Check for alternatives when higher level complains
2696  * about service problems.  For now, invalidate cached
2697  * routing information.  If the route was created dynamically
2698  * (by a redirect), time to try a default gateway again.
2699  */
2700 void
2701 in_losing(struct inpcb *inp)
2702 {
2703 
2704 	RO_INVALIDATE_CACHE(&inp->inp_route);
2705 	return;
2706 }
2707 
2708 /*
2709  * A set label operation has occurred at the socket layer, propagate the
2710  * label change into the in_pcb for the socket.
2711  */
2712 void
2713 in_pcbsosetlabel(struct socket *so)
2714 {
2715 #ifdef MAC
2716 	struct inpcb *inp;
2717 
2718 	inp = sotoinpcb(so);
2719 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2720 
2721 	INP_WLOCK(inp);
2722 	SOCK_LOCK(so);
2723 	mac_inpcb_sosetlabel(so, inp);
2724 	SOCK_UNLOCK(so);
2725 	INP_WUNLOCK(inp);
2726 #endif
2727 }
2728 
2729 /*
2730  * ipport_tick runs once per second, determining if random port allocation
2731  * should be continued.  If more than ipport_randomcps ports have been
2732  * allocated in the last second, then we return to sequential port
2733  * allocation. We return to random allocation only once we drop below
2734  * ipport_randomcps for at least ipport_randomtime seconds.
2735  */
2736 static void
2737 ipport_tick(void *xtp)
2738 {
2739 	VNET_ITERATOR_DECL(vnet_iter);
2740 
2741 	VNET_LIST_RLOCK_NOSLEEP();
2742 	VNET_FOREACH(vnet_iter) {
2743 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2744 		if (V_ipport_tcpallocs <=
2745 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2746 			if (V_ipport_stoprandom > 0)
2747 				V_ipport_stoprandom--;
2748 		} else
2749 			V_ipport_stoprandom = V_ipport_randomtime;
2750 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2751 		CURVNET_RESTORE();
2752 	}
2753 	VNET_LIST_RUNLOCK_NOSLEEP();
2754 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2755 }
2756 
2757 static void
2758 ip_fini(void *xtp)
2759 {
2760 
2761 	callout_stop(&ipport_tick_callout);
2762 }
2763 
2764 /*
2765  * The ipport_callout should start running at about the time we attach the
2766  * inet or inet6 domains.
2767  */
2768 static void
2769 ipport_tick_init(const void *unused __unused)
2770 {
2771 
2772 	/* Start ipport_tick. */
2773 	callout_init(&ipport_tick_callout, 1);
2774 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2775 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2776 		SHUTDOWN_PRI_DEFAULT);
2777 }
2778 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2779     ipport_tick_init, NULL);
2780 
2781 void
2782 inp_wlock(struct inpcb *inp)
2783 {
2784 
2785 	INP_WLOCK(inp);
2786 }
2787 
2788 void
2789 inp_wunlock(struct inpcb *inp)
2790 {
2791 
2792 	INP_WUNLOCK(inp);
2793 }
2794 
2795 void
2796 inp_rlock(struct inpcb *inp)
2797 {
2798 
2799 	INP_RLOCK(inp);
2800 }
2801 
2802 void
2803 inp_runlock(struct inpcb *inp)
2804 {
2805 
2806 	INP_RUNLOCK(inp);
2807 }
2808 
2809 #ifdef INVARIANT_SUPPORT
2810 void
2811 inp_lock_assert(struct inpcb *inp)
2812 {
2813 
2814 	INP_WLOCK_ASSERT(inp);
2815 }
2816 
2817 void
2818 inp_unlock_assert(struct inpcb *inp)
2819 {
2820 
2821 	INP_UNLOCK_ASSERT(inp);
2822 }
2823 #endif
2824 
2825 void
2826 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2827 {
2828 	struct inpcb *inp;
2829 
2830 	INP_INFO_WLOCK(&V_tcbinfo);
2831 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2832 		INP_WLOCK(inp);
2833 		func(inp, arg);
2834 		INP_WUNLOCK(inp);
2835 	}
2836 	INP_INFO_WUNLOCK(&V_tcbinfo);
2837 }
2838 
2839 struct socket *
2840 inp_inpcbtosocket(struct inpcb *inp)
2841 {
2842 
2843 	INP_WLOCK_ASSERT(inp);
2844 	return (inp->inp_socket);
2845 }
2846 
2847 struct tcpcb *
2848 inp_inpcbtotcpcb(struct inpcb *inp)
2849 {
2850 
2851 	INP_WLOCK_ASSERT(inp);
2852 	return ((struct tcpcb *)inp->inp_ppcb);
2853 }
2854 
2855 int
2856 inp_ip_tos_get(const struct inpcb *inp)
2857 {
2858 
2859 	return (inp->inp_ip_tos);
2860 }
2861 
2862 void
2863 inp_ip_tos_set(struct inpcb *inp, int val)
2864 {
2865 
2866 	inp->inp_ip_tos = val;
2867 }
2868 
2869 void
2870 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2871     uint32_t *faddr, uint16_t *fp)
2872 {
2873 
2874 	INP_LOCK_ASSERT(inp);
2875 	*laddr = inp->inp_laddr.s_addr;
2876 	*faddr = inp->inp_faddr.s_addr;
2877 	*lp = inp->inp_lport;
2878 	*fp = inp->inp_fport;
2879 }
2880 
2881 struct inpcb *
2882 so_sotoinpcb(struct socket *so)
2883 {
2884 
2885 	return (sotoinpcb(so));
2886 }
2887 
2888 struct tcpcb *
2889 so_sototcpcb(struct socket *so)
2890 {
2891 
2892 	return (sototcpcb(so));
2893 }
2894 
2895 /*
2896  * Create an external-format (``xinpcb'') structure using the information in
2897  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2898  * reduce the spew of irrelevant information over this interface, to isolate
2899  * user code from changes in the kernel structure, and potentially to provide
2900  * information-hiding if we decide that some of this information should be
2901  * hidden from users.
2902  */
2903 void
2904 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2905 {
2906 
2907 	xi->xi_len = sizeof(struct xinpcb);
2908 	if (inp->inp_socket)
2909 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2910 	else
2911 		bzero(&xi->xi_socket, sizeof(struct xsocket));
2912 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2913 	xi->inp_gencnt = inp->inp_gencnt;
2914 	xi->inp_ppcb = inp->inp_ppcb;
2915 	xi->inp_flow = inp->inp_flow;
2916 	xi->inp_flowid = inp->inp_flowid;
2917 	xi->inp_flowtype = inp->inp_flowtype;
2918 	xi->inp_flags = inp->inp_flags;
2919 	xi->inp_flags2 = inp->inp_flags2;
2920 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2921 	xi->in6p_cksum = inp->in6p_cksum;
2922 	xi->in6p_hops = inp->in6p_hops;
2923 	xi->inp_ip_tos = inp->inp_ip_tos;
2924 	xi->inp_vflag = inp->inp_vflag;
2925 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2926 	xi->inp_ip_p = inp->inp_ip_p;
2927 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2928 }
2929 
2930 #ifdef DDB
2931 static void
2932 db_print_indent(int indent)
2933 {
2934 	int i;
2935 
2936 	for (i = 0; i < indent; i++)
2937 		db_printf(" ");
2938 }
2939 
2940 static void
2941 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2942 {
2943 	char faddr_str[48], laddr_str[48];
2944 
2945 	db_print_indent(indent);
2946 	db_printf("%s at %p\n", name, inc);
2947 
2948 	indent += 2;
2949 
2950 #ifdef INET6
2951 	if (inc->inc_flags & INC_ISIPV6) {
2952 		/* IPv6. */
2953 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2954 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2955 	} else
2956 #endif
2957 	{
2958 		/* IPv4. */
2959 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2960 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2961 	}
2962 	db_print_indent(indent);
2963 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2964 	    ntohs(inc->inc_lport));
2965 	db_print_indent(indent);
2966 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2967 	    ntohs(inc->inc_fport));
2968 }
2969 
2970 static void
2971 db_print_inpflags(int inp_flags)
2972 {
2973 	int comma;
2974 
2975 	comma = 0;
2976 	if (inp_flags & INP_RECVOPTS) {
2977 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2978 		comma = 1;
2979 	}
2980 	if (inp_flags & INP_RECVRETOPTS) {
2981 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2982 		comma = 1;
2983 	}
2984 	if (inp_flags & INP_RECVDSTADDR) {
2985 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2986 		comma = 1;
2987 	}
2988 	if (inp_flags & INP_ORIGDSTADDR) {
2989 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2990 		comma = 1;
2991 	}
2992 	if (inp_flags & INP_HDRINCL) {
2993 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2994 		comma = 1;
2995 	}
2996 	if (inp_flags & INP_HIGHPORT) {
2997 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2998 		comma = 1;
2999 	}
3000 	if (inp_flags & INP_LOWPORT) {
3001 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
3002 		comma = 1;
3003 	}
3004 	if (inp_flags & INP_ANONPORT) {
3005 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
3006 		comma = 1;
3007 	}
3008 	if (inp_flags & INP_RECVIF) {
3009 		db_printf("%sINP_RECVIF", comma ? ", " : "");
3010 		comma = 1;
3011 	}
3012 	if (inp_flags & INP_MTUDISC) {
3013 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
3014 		comma = 1;
3015 	}
3016 	if (inp_flags & INP_RECVTTL) {
3017 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
3018 		comma = 1;
3019 	}
3020 	if (inp_flags & INP_DONTFRAG) {
3021 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3022 		comma = 1;
3023 	}
3024 	if (inp_flags & INP_RECVTOS) {
3025 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3026 		comma = 1;
3027 	}
3028 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3029 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3030 		comma = 1;
3031 	}
3032 	if (inp_flags & IN6P_PKTINFO) {
3033 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3034 		comma = 1;
3035 	}
3036 	if (inp_flags & IN6P_HOPLIMIT) {
3037 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3038 		comma = 1;
3039 	}
3040 	if (inp_flags & IN6P_HOPOPTS) {
3041 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3042 		comma = 1;
3043 	}
3044 	if (inp_flags & IN6P_DSTOPTS) {
3045 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3046 		comma = 1;
3047 	}
3048 	if (inp_flags & IN6P_RTHDR) {
3049 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3050 		comma = 1;
3051 	}
3052 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3053 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3054 		comma = 1;
3055 	}
3056 	if (inp_flags & IN6P_TCLASS) {
3057 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3058 		comma = 1;
3059 	}
3060 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3061 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3062 		comma = 1;
3063 	}
3064 	if (inp_flags & INP_TIMEWAIT) {
3065 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3066 		comma  = 1;
3067 	}
3068 	if (inp_flags & INP_ONESBCAST) {
3069 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3070 		comma  = 1;
3071 	}
3072 	if (inp_flags & INP_DROPPED) {
3073 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3074 		comma  = 1;
3075 	}
3076 	if (inp_flags & INP_SOCKREF) {
3077 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3078 		comma  = 1;
3079 	}
3080 	if (inp_flags & IN6P_RFC2292) {
3081 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3082 		comma = 1;
3083 	}
3084 	if (inp_flags & IN6P_MTU) {
3085 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3086 		comma = 1;
3087 	}
3088 }
3089 
3090 static void
3091 db_print_inpvflag(u_char inp_vflag)
3092 {
3093 	int comma;
3094 
3095 	comma = 0;
3096 	if (inp_vflag & INP_IPV4) {
3097 		db_printf("%sINP_IPV4", comma ? ", " : "");
3098 		comma  = 1;
3099 	}
3100 	if (inp_vflag & INP_IPV6) {
3101 		db_printf("%sINP_IPV6", comma ? ", " : "");
3102 		comma  = 1;
3103 	}
3104 	if (inp_vflag & INP_IPV6PROTO) {
3105 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3106 		comma  = 1;
3107 	}
3108 }
3109 
3110 static void
3111 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3112 {
3113 
3114 	db_print_indent(indent);
3115 	db_printf("%s at %p\n", name, inp);
3116 
3117 	indent += 2;
3118 
3119 	db_print_indent(indent);
3120 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3121 
3122 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3123 
3124 	db_print_indent(indent);
3125 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3126 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3127 
3128 	db_print_indent(indent);
3129 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3130 	   inp->inp_label, inp->inp_flags);
3131 	db_print_inpflags(inp->inp_flags);
3132 	db_printf(")\n");
3133 
3134 	db_print_indent(indent);
3135 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3136 	    inp->inp_vflag);
3137 	db_print_inpvflag(inp->inp_vflag);
3138 	db_printf(")\n");
3139 
3140 	db_print_indent(indent);
3141 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3142 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3143 
3144 	db_print_indent(indent);
3145 #ifdef INET6
3146 	if (inp->inp_vflag & INP_IPV6) {
3147 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3148 		    "in6p_moptions: %p\n", inp->in6p_options,
3149 		    inp->in6p_outputopts, inp->in6p_moptions);
3150 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3151 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3152 		    inp->in6p_hops);
3153 	} else
3154 #endif
3155 	{
3156 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3157 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3158 		    inp->inp_options, inp->inp_moptions);
3159 	}
3160 
3161 	db_print_indent(indent);
3162 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3163 	    (uintmax_t)inp->inp_gencnt);
3164 }
3165 
3166 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3167 {
3168 	struct inpcb *inp;
3169 
3170 	if (!have_addr) {
3171 		db_printf("usage: show inpcb <addr>\n");
3172 		return;
3173 	}
3174 	inp = (struct inpcb *)addr;
3175 
3176 	db_print_inpcb(inp, "inpcb", 0);
3177 }
3178 #endif /* DDB */
3179 
3180 #ifdef RATELIMIT
3181 /*
3182  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3183  * if any.
3184  */
3185 int
3186 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3187 {
3188 	union if_snd_tag_modify_params params = {
3189 		.rate_limit.max_rate = max_pacing_rate,
3190 	};
3191 	struct m_snd_tag *mst;
3192 	struct ifnet *ifp;
3193 	int error;
3194 
3195 	mst = inp->inp_snd_tag;
3196 	if (mst == NULL)
3197 		return (EINVAL);
3198 
3199 	ifp = mst->ifp;
3200 	if (ifp == NULL)
3201 		return (EINVAL);
3202 
3203 	if (ifp->if_snd_tag_modify == NULL) {
3204 		error = EOPNOTSUPP;
3205 	} else {
3206 		error = ifp->if_snd_tag_modify(mst, &params);
3207 	}
3208 	return (error);
3209 }
3210 
3211 /*
3212  * Query existing TX rate limit based on the existing
3213  * "inp->inp_snd_tag", if any.
3214  */
3215 int
3216 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3217 {
3218 	union if_snd_tag_query_params params = { };
3219 	struct m_snd_tag *mst;
3220 	struct ifnet *ifp;
3221 	int error;
3222 
3223 	mst = inp->inp_snd_tag;
3224 	if (mst == NULL)
3225 		return (EINVAL);
3226 
3227 	ifp = mst->ifp;
3228 	if (ifp == NULL)
3229 		return (EINVAL);
3230 
3231 	if (ifp->if_snd_tag_query == NULL) {
3232 		error = EOPNOTSUPP;
3233 	} else {
3234 		error = ifp->if_snd_tag_query(mst, &params);
3235 		if (error == 0 &&  p_max_pacing_rate != NULL)
3236 			*p_max_pacing_rate = params.rate_limit.max_rate;
3237 	}
3238 	return (error);
3239 }
3240 
3241 /*
3242  * Query existing TX queue level based on the existing
3243  * "inp->inp_snd_tag", if any.
3244  */
3245 int
3246 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3247 {
3248 	union if_snd_tag_query_params params = { };
3249 	struct m_snd_tag *mst;
3250 	struct ifnet *ifp;
3251 	int error;
3252 
3253 	mst = inp->inp_snd_tag;
3254 	if (mst == NULL)
3255 		return (EINVAL);
3256 
3257 	ifp = mst->ifp;
3258 	if (ifp == NULL)
3259 		return (EINVAL);
3260 
3261 	if (ifp->if_snd_tag_query == NULL)
3262 		return (EOPNOTSUPP);
3263 
3264 	error = ifp->if_snd_tag_query(mst, &params);
3265 	if (error == 0 &&  p_txqueue_level != NULL)
3266 		*p_txqueue_level = params.rate_limit.queue_level;
3267 	return (error);
3268 }
3269 
3270 /*
3271  * Allocate a new TX rate limit send tag from the network interface
3272  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3273  */
3274 int
3275 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3276     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3277 {
3278 	union if_snd_tag_alloc_params params = {
3279 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3280 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3281 		.rate_limit.hdr.flowid = flowid,
3282 		.rate_limit.hdr.flowtype = flowtype,
3283 		.rate_limit.max_rate = max_pacing_rate,
3284 	};
3285 	int error;
3286 
3287 	INP_WLOCK_ASSERT(inp);
3288 
3289 	if (inp->inp_snd_tag != NULL)
3290 		return (EINVAL);
3291 
3292 	if (ifp->if_snd_tag_alloc == NULL) {
3293 		error = EOPNOTSUPP;
3294 	} else {
3295 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3296 
3297 		/*
3298 		 * At success increment the refcount on
3299 		 * the send tag's network interface:
3300 		 */
3301 		if (error == 0)
3302 			if_ref(inp->inp_snd_tag->ifp);
3303 	}
3304 	return (error);
3305 }
3306 
3307 /*
3308  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3309  * if any:
3310  */
3311 void
3312 in_pcbdetach_txrtlmt(struct inpcb *inp)
3313 {
3314 	struct m_snd_tag *mst;
3315 	struct ifnet *ifp;
3316 
3317 	INP_WLOCK_ASSERT(inp);
3318 
3319 	mst = inp->inp_snd_tag;
3320 	inp->inp_snd_tag = NULL;
3321 
3322 	if (mst == NULL)
3323 		return;
3324 
3325 	ifp = mst->ifp;
3326 	if (ifp == NULL)
3327 		return;
3328 
3329 	/*
3330 	 * If the device was detached while we still had reference(s)
3331 	 * on the ifp, we assume if_snd_tag_free() was replaced with
3332 	 * stubs.
3333 	 */
3334 	ifp->if_snd_tag_free(mst);
3335 
3336 	/* release reference count on network interface */
3337 	if_rele(ifp);
3338 }
3339 
3340 /*
3341  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3342  * is set in the fast path and will attach/detach/modify the TX rate
3343  * limit send tag based on the socket's so_max_pacing_rate value.
3344  */
3345 void
3346 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3347 {
3348 	struct socket *socket;
3349 	uint32_t max_pacing_rate;
3350 	bool did_upgrade;
3351 	int error;
3352 
3353 	if (inp == NULL)
3354 		return;
3355 
3356 	socket = inp->inp_socket;
3357 	if (socket == NULL)
3358 		return;
3359 
3360 	if (!INP_WLOCKED(inp)) {
3361 		/*
3362 		 * NOTE: If the write locking fails, we need to bail
3363 		 * out and use the non-ratelimited ring for the
3364 		 * transmit until there is a new chance to get the
3365 		 * write lock.
3366 		 */
3367 		if (!INP_TRY_UPGRADE(inp))
3368 			return;
3369 		did_upgrade = 1;
3370 	} else {
3371 		did_upgrade = 0;
3372 	}
3373 
3374 	/*
3375 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3376 	 * because atomic updates are not required since the variable
3377 	 * is checked at every mbuf we send. It is assumed that the
3378 	 * variable read itself will be atomic.
3379 	 */
3380 	max_pacing_rate = socket->so_max_pacing_rate;
3381 
3382 	/*
3383 	 * NOTE: When attaching to a network interface a reference is
3384 	 * made to ensure the network interface doesn't go away until
3385 	 * all ratelimit connections are gone. The network interface
3386 	 * pointers compared below represent valid network interfaces,
3387 	 * except when comparing towards NULL.
3388 	 */
3389 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3390 		error = 0;
3391 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3392 		if (inp->inp_snd_tag != NULL)
3393 			in_pcbdetach_txrtlmt(inp);
3394 		error = 0;
3395 	} else if (inp->inp_snd_tag == NULL) {
3396 		/*
3397 		 * In order to utilize packet pacing with RSS, we need
3398 		 * to wait until there is a valid RSS hash before we
3399 		 * can proceed:
3400 		 */
3401 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3402 			error = EAGAIN;
3403 		} else {
3404 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3405 			    mb->m_pkthdr.flowid, max_pacing_rate);
3406 		}
3407 	} else {
3408 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3409 	}
3410 	if (error == 0 || error == EOPNOTSUPP)
3411 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3412 	if (did_upgrade)
3413 		INP_DOWNGRADE(inp);
3414 }
3415 
3416 /*
3417  * Track route changes for TX rate limiting.
3418  */
3419 void
3420 in_pcboutput_eagain(struct inpcb *inp)
3421 {
3422 	struct socket *socket;
3423 	bool did_upgrade;
3424 
3425 	if (inp == NULL)
3426 		return;
3427 
3428 	socket = inp->inp_socket;
3429 	if (socket == NULL)
3430 		return;
3431 
3432 	if (inp->inp_snd_tag == NULL)
3433 		return;
3434 
3435 	if (!INP_WLOCKED(inp)) {
3436 		/*
3437 		 * NOTE: If the write locking fails, we need to bail
3438 		 * out and use the non-ratelimited ring for the
3439 		 * transmit until there is a new chance to get the
3440 		 * write lock.
3441 		 */
3442 		if (!INP_TRY_UPGRADE(inp))
3443 			return;
3444 		did_upgrade = 1;
3445 	} else {
3446 		did_upgrade = 0;
3447 	}
3448 
3449 	/* detach rate limiting */
3450 	in_pcbdetach_txrtlmt(inp);
3451 
3452 	/* make sure new mbuf send tag allocation is made */
3453 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3454 
3455 	if (did_upgrade)
3456 		INP_DOWNGRADE(inp);
3457 }
3458 #endif /* RATELIMIT */
3459