xref: /freebsd/sys/netinet/in_pcb.c (revision 190cef3d52236565eb22e18b33e9e865ec634aa3)
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 VNET_DEFINE_STATIC(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_PCB);
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 	const static struct timeval interval = { 60, 0 };
305 	static struct timeval lastprint;
306 	struct inpcbinfo *pcbinfo;
307 	struct inpcblbgrouphead *hdr;
308 	struct inpcblbgroup *grp;
309 	uint16_t hashmask, lport;
310 	uint32_t group_index;
311 	struct ucred *cred;
312 
313 	pcbinfo = inp->inp_pcbinfo;
314 
315 	INP_WLOCK_ASSERT(inp);
316 	INP_HASH_WLOCK_ASSERT(pcbinfo);
317 
318 	if (pcbinfo->ipi_lbgrouphashbase == NULL)
319 		return (0);
320 
321 	hashmask = pcbinfo->ipi_lbgrouphashmask;
322 	lport = inp->inp_lport;
323 	group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
324 	hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
325 
326 	/*
327 	 * Don't allow jailed socket to join local group.
328 	 */
329 	if (inp->inp_socket != NULL)
330 		cred = inp->inp_socket->so_cred;
331 	else
332 		cred = NULL;
333 	if (cred != NULL && jailed(cred))
334 		return (0);
335 
336 #ifdef INET6
337 	/*
338 	 * Don't allow IPv4 mapped INET6 wild socket.
339 	 */
340 	if ((inp->inp_vflag & INP_IPV4) &&
341 	    inp->inp_laddr.s_addr == INADDR_ANY &&
342 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
343 		return (0);
344 	}
345 #endif
346 
347 	hdr = &pcbinfo->ipi_lbgrouphashbase[
348 	    INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
349 	        pcbinfo->ipi_lbgrouphashmask)];
350 	LIST_FOREACH(grp, hdr, il_list) {
351 		if (grp->il_vflag == inp->inp_vflag &&
352 		    grp->il_lport == inp->inp_lport &&
353 		    memcmp(&grp->il_dependladdr,
354 		        &inp->inp_inc.inc_ie.ie_dependladdr,
355 		        sizeof(grp->il_dependladdr)) == 0) {
356 			break;
357 		}
358 	}
359 	if (grp == NULL) {
360 		/* Create new load balance group. */
361 		grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
362 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
363 		    INPCBLBGROUP_SIZMIN);
364 		if (!grp)
365 			return (ENOBUFS);
366 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
367 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
368 			if (ratecheck(&lastprint, &interval))
369 				printf("lb group port %d, limit reached\n",
370 				    ntohs(grp->il_lport));
371 			return (0);
372 		}
373 
374 		/* Expand this local group. */
375 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
376 		if (!grp)
377 			return (ENOBUFS);
378 	}
379 
380 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
381 			("invalid local group size %d and count %d",
382 			 grp->il_inpsiz, grp->il_inpcnt));
383 
384 	grp->il_inp[grp->il_inpcnt] = inp;
385 	grp->il_inpcnt++;
386 	return (0);
387 }
388 
389 /*
390  * Remove PCB from load balance group.
391  */
392 static void
393 in_pcbremlbgrouphash(struct inpcb *inp)
394 {
395 	struct inpcbinfo *pcbinfo;
396 	struct inpcblbgrouphead *hdr;
397 	struct inpcblbgroup *grp;
398 	int i;
399 
400 	pcbinfo = inp->inp_pcbinfo;
401 
402 	INP_WLOCK_ASSERT(inp);
403 	INP_HASH_WLOCK_ASSERT(pcbinfo);
404 
405 	if (pcbinfo->ipi_lbgrouphashbase == NULL)
406 		return;
407 
408 	hdr = &pcbinfo->ipi_lbgrouphashbase[
409 	    INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
410 	        pcbinfo->ipi_lbgrouphashmask)];
411 
412 	LIST_FOREACH(grp, hdr, il_list) {
413 		for (i = 0; i < grp->il_inpcnt; ++i) {
414 			if (grp->il_inp[i] != inp)
415 				continue;
416 
417 			if (grp->il_inpcnt == 1) {
418 				/* We are the last, free this local group. */
419 				in_pcblbgroup_free(grp);
420 			} else {
421 				/* Pull up inpcbs, shrink group if possible. */
422 				in_pcblbgroup_reorder(hdr, &grp, i);
423 			}
424 			return;
425 		}
426 	}
427 }
428 
429 /*
430  * Different protocols initialize their inpcbs differently - giving
431  * different name to the lock.  But they all are disposed the same.
432  */
433 static void
434 inpcb_fini(void *mem, int size)
435 {
436 	struct inpcb *inp = mem;
437 
438 	INP_LOCK_DESTROY(inp);
439 }
440 
441 /*
442  * Initialize an inpcbinfo -- we should be able to reduce the number of
443  * arguments in time.
444  */
445 void
446 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
447     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
448     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
449 {
450 
451 	INP_INFO_LOCK_INIT(pcbinfo, name);
452 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
453 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
454 #ifdef VIMAGE
455 	pcbinfo->ipi_vnet = curvnet;
456 #endif
457 	pcbinfo->ipi_listhead = listhead;
458 	CK_LIST_INIT(pcbinfo->ipi_listhead);
459 	pcbinfo->ipi_count = 0;
460 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
461 	    &pcbinfo->ipi_hashmask);
462 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
463 	    &pcbinfo->ipi_porthashmask);
464 	pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
465 	    &pcbinfo->ipi_lbgrouphashmask);
466 #ifdef PCBGROUP
467 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
468 #endif
469 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
470 	    NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
471 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
472 	uma_zone_set_warning(pcbinfo->ipi_zone,
473 	    "kern.ipc.maxsockets limit reached");
474 }
475 
476 /*
477  * Destroy an inpcbinfo.
478  */
479 void
480 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
481 {
482 
483 	KASSERT(pcbinfo->ipi_count == 0,
484 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
485 
486 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
487 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
488 	    pcbinfo->ipi_porthashmask);
489 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
490 	    pcbinfo->ipi_lbgrouphashmask);
491 #ifdef PCBGROUP
492 	in_pcbgroup_destroy(pcbinfo);
493 #endif
494 	uma_zdestroy(pcbinfo->ipi_zone);
495 	INP_LIST_LOCK_DESTROY(pcbinfo);
496 	INP_HASH_LOCK_DESTROY(pcbinfo);
497 	INP_INFO_LOCK_DESTROY(pcbinfo);
498 }
499 
500 /*
501  * Allocate a PCB and associate it with the socket.
502  * On success return with the PCB locked.
503  */
504 int
505 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
506 {
507 	struct inpcb *inp;
508 	int error;
509 
510 #ifdef INVARIANTS
511 	if (pcbinfo == &V_tcbinfo) {
512 		INP_INFO_RLOCK_ASSERT(pcbinfo);
513 	} else {
514 		INP_INFO_WLOCK_ASSERT(pcbinfo);
515 	}
516 #endif
517 
518 	error = 0;
519 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
520 	if (inp == NULL)
521 		return (ENOBUFS);
522 	bzero(&inp->inp_start_zero, inp_zero_size);
523 	inp->inp_pcbinfo = pcbinfo;
524 	inp->inp_socket = so;
525 	inp->inp_cred = crhold(so->so_cred);
526 	inp->inp_inc.inc_fibnum = so->so_fibnum;
527 #ifdef MAC
528 	error = mac_inpcb_init(inp, M_NOWAIT);
529 	if (error != 0)
530 		goto out;
531 	mac_inpcb_create(so, inp);
532 #endif
533 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
534 	error = ipsec_init_pcbpolicy(inp);
535 	if (error != 0) {
536 #ifdef MAC
537 		mac_inpcb_destroy(inp);
538 #endif
539 		goto out;
540 	}
541 #endif /*IPSEC*/
542 #ifdef INET6
543 	if (INP_SOCKAF(so) == AF_INET6) {
544 		inp->inp_vflag |= INP_IPV6PROTO;
545 		if (V_ip6_v6only)
546 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
547 	}
548 #endif
549 	INP_WLOCK(inp);
550 	INP_LIST_WLOCK(pcbinfo);
551 	CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
552 	pcbinfo->ipi_count++;
553 	so->so_pcb = (caddr_t)inp;
554 #ifdef INET6
555 	if (V_ip6_auto_flowlabel)
556 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
557 #endif
558 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
559 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
560 
561 	/*
562 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
563 	 * to be cleaned up.
564 	 */
565 	inp->inp_route.ro_flags = RT_LLE_CACHE;
566 	INP_LIST_WUNLOCK(pcbinfo);
567 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
568 out:
569 	if (error != 0) {
570 		crfree(inp->inp_cred);
571 		uma_zfree(pcbinfo->ipi_zone, inp);
572 	}
573 #endif
574 	return (error);
575 }
576 
577 #ifdef INET
578 int
579 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
580 {
581 	int anonport, error;
582 
583 	INP_WLOCK_ASSERT(inp);
584 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
585 
586 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
587 		return (EINVAL);
588 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
589 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
590 	    &inp->inp_lport, cred);
591 	if (error)
592 		return (error);
593 	if (in_pcbinshash(inp) != 0) {
594 		inp->inp_laddr.s_addr = INADDR_ANY;
595 		inp->inp_lport = 0;
596 		return (EAGAIN);
597 	}
598 	if (anonport)
599 		inp->inp_flags |= INP_ANONPORT;
600 	return (0);
601 }
602 #endif
603 
604 /*
605  * Select a local port (number) to use.
606  */
607 #if defined(INET) || defined(INET6)
608 int
609 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
610     struct ucred *cred, int lookupflags)
611 {
612 	struct inpcbinfo *pcbinfo;
613 	struct inpcb *tmpinp;
614 	unsigned short *lastport;
615 	int count, dorandom, error;
616 	u_short aux, first, last, lport;
617 #ifdef INET
618 	struct in_addr laddr;
619 #endif
620 
621 	pcbinfo = inp->inp_pcbinfo;
622 
623 	/*
624 	 * Because no actual state changes occur here, a global write lock on
625 	 * the pcbinfo isn't required.
626 	 */
627 	INP_LOCK_ASSERT(inp);
628 	INP_HASH_LOCK_ASSERT(pcbinfo);
629 
630 	if (inp->inp_flags & INP_HIGHPORT) {
631 		first = V_ipport_hifirstauto;	/* sysctl */
632 		last  = V_ipport_hilastauto;
633 		lastport = &pcbinfo->ipi_lasthi;
634 	} else if (inp->inp_flags & INP_LOWPORT) {
635 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
636 		if (error)
637 			return (error);
638 		first = V_ipport_lowfirstauto;	/* 1023 */
639 		last  = V_ipport_lowlastauto;	/* 600 */
640 		lastport = &pcbinfo->ipi_lastlow;
641 	} else {
642 		first = V_ipport_firstauto;	/* sysctl */
643 		last  = V_ipport_lastauto;
644 		lastport = &pcbinfo->ipi_lastport;
645 	}
646 	/*
647 	 * For UDP(-Lite), use random port allocation as long as the user
648 	 * allows it.  For TCP (and as of yet unknown) connections,
649 	 * use random port allocation only if the user allows it AND
650 	 * ipport_tick() allows it.
651 	 */
652 	if (V_ipport_randomized &&
653 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
654 		pcbinfo == &V_ulitecbinfo))
655 		dorandom = 1;
656 	else
657 		dorandom = 0;
658 	/*
659 	 * It makes no sense to do random port allocation if
660 	 * we have the only port available.
661 	 */
662 	if (first == last)
663 		dorandom = 0;
664 	/* Make sure to not include UDP(-Lite) packets in the count. */
665 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
666 		V_ipport_tcpallocs++;
667 	/*
668 	 * Instead of having two loops further down counting up or down
669 	 * make sure that first is always <= last and go with only one
670 	 * code path implementing all logic.
671 	 */
672 	if (first > last) {
673 		aux = first;
674 		first = last;
675 		last = aux;
676 	}
677 
678 #ifdef INET
679 	/* Make the compiler happy. */
680 	laddr.s_addr = 0;
681 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
682 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
683 		    __func__, inp));
684 		laddr = *laddrp;
685 	}
686 #endif
687 	tmpinp = NULL;	/* Make compiler happy. */
688 	lport = *lportp;
689 
690 	if (dorandom)
691 		*lastport = first + (arc4random() % (last - first));
692 
693 	count = last - first;
694 
695 	do {
696 		if (count-- < 0)	/* completely used? */
697 			return (EADDRNOTAVAIL);
698 		++*lastport;
699 		if (*lastport < first || *lastport > last)
700 			*lastport = first;
701 		lport = htons(*lastport);
702 
703 #ifdef INET6
704 		if ((inp->inp_vflag & INP_IPV6) != 0)
705 			tmpinp = in6_pcblookup_local(pcbinfo,
706 			    &inp->in6p_laddr, lport, lookupflags, cred);
707 #endif
708 #if defined(INET) && defined(INET6)
709 		else
710 #endif
711 #ifdef INET
712 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
713 			    lport, lookupflags, cred);
714 #endif
715 	} while (tmpinp != NULL);
716 
717 #ifdef INET
718 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
719 		laddrp->s_addr = laddr.s_addr;
720 #endif
721 	*lportp = lport;
722 
723 	return (0);
724 }
725 
726 /*
727  * Return cached socket options.
728  */
729 int
730 inp_so_options(const struct inpcb *inp)
731 {
732 	int so_options;
733 
734 	so_options = 0;
735 
736 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
737 		so_options |= SO_REUSEPORT_LB;
738 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
739 		so_options |= SO_REUSEPORT;
740 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
741 		so_options |= SO_REUSEADDR;
742 	return (so_options);
743 }
744 #endif /* INET || INET6 */
745 
746 /*
747  * Check if a new BINDMULTI socket is allowed to be created.
748  *
749  * ni points to the new inp.
750  * oi points to the exisitng inp.
751  *
752  * This checks whether the existing inp also has BINDMULTI and
753  * whether the credentials match.
754  */
755 int
756 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
757 {
758 	/* Check permissions match */
759 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
760 	    (ni->inp_cred->cr_uid !=
761 	    oi->inp_cred->cr_uid))
762 		return (0);
763 
764 	/* Check the existing inp has BINDMULTI set */
765 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
766 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
767 		return (0);
768 
769 	/*
770 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
771 	 * it is and it matches the checks.
772 	 */
773 	return (1);
774 }
775 
776 #ifdef INET
777 /*
778  * Set up a bind operation on a PCB, performing port allocation
779  * as required, but do not actually modify the PCB. Callers can
780  * either complete the bind by setting inp_laddr/inp_lport and
781  * calling in_pcbinshash(), or they can just use the resulting
782  * port and address to authorise the sending of a once-off packet.
783  *
784  * On error, the values of *laddrp and *lportp are not changed.
785  */
786 int
787 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
788     u_short *lportp, struct ucred *cred)
789 {
790 	struct socket *so = inp->inp_socket;
791 	struct sockaddr_in *sin;
792 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
793 	struct in_addr laddr;
794 	u_short lport = 0;
795 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
796 	int error;
797 
798 	/*
799 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
800 	 * so that we don't have to add to the (already messy) code below.
801 	 */
802 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
803 
804 	/*
805 	 * No state changes, so read locks are sufficient here.
806 	 */
807 	INP_LOCK_ASSERT(inp);
808 	INP_HASH_LOCK_ASSERT(pcbinfo);
809 
810 	if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
811 		return (EADDRNOTAVAIL);
812 	laddr.s_addr = *laddrp;
813 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
814 		return (EINVAL);
815 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
816 		lookupflags = INPLOOKUP_WILDCARD;
817 	if (nam == NULL) {
818 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
819 			return (error);
820 	} else {
821 		sin = (struct sockaddr_in *)nam;
822 		if (nam->sa_len != sizeof (*sin))
823 			return (EINVAL);
824 #ifdef notdef
825 		/*
826 		 * We should check the family, but old programs
827 		 * incorrectly fail to initialize it.
828 		 */
829 		if (sin->sin_family != AF_INET)
830 			return (EAFNOSUPPORT);
831 #endif
832 		error = prison_local_ip4(cred, &sin->sin_addr);
833 		if (error)
834 			return (error);
835 		if (sin->sin_port != *lportp) {
836 			/* Don't allow the port to change. */
837 			if (*lportp != 0)
838 				return (EINVAL);
839 			lport = sin->sin_port;
840 		}
841 		/* NB: lport is left as 0 if the port isn't being changed. */
842 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
843 			/*
844 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
845 			 * allow complete duplication of binding if
846 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
847 			 * and a multicast address is bound on both
848 			 * new and duplicated sockets.
849 			 */
850 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
851 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
852 			/*
853 			 * XXX: How to deal with SO_REUSEPORT_LB here?
854 			 * Treat same as SO_REUSEPORT for now.
855 			 */
856 			if ((so->so_options &
857 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
858 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
859 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
860 			sin->sin_port = 0;		/* yech... */
861 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
862 			/*
863 			 * Is the address a local IP address?
864 			 * If INP_BINDANY is set, then the socket may be bound
865 			 * to any endpoint address, local or not.
866 			 */
867 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
868 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
869 				return (EADDRNOTAVAIL);
870 		}
871 		laddr = sin->sin_addr;
872 		if (lport) {
873 			struct inpcb *t;
874 			struct tcptw *tw;
875 
876 			/* GROSS */
877 			if (ntohs(lport) <= V_ipport_reservedhigh &&
878 			    ntohs(lport) >= V_ipport_reservedlow &&
879 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
880 			    0))
881 				return (EACCES);
882 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
883 			    priv_check_cred(inp->inp_cred,
884 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
885 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
886 				    lport, INPLOOKUP_WILDCARD, cred);
887 	/*
888 	 * XXX
889 	 * This entire block sorely needs a rewrite.
890 	 */
891 				if (t &&
892 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
893 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
894 				    (so->so_type != SOCK_STREAM ||
895 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
896 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
897 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
898 				     (t->inp_flags2 & INP_REUSEPORT) ||
899 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
900 				    (inp->inp_cred->cr_uid !=
901 				     t->inp_cred->cr_uid))
902 					return (EADDRINUSE);
903 
904 				/*
905 				 * If the socket is a BINDMULTI socket, then
906 				 * the credentials need to match and the
907 				 * original socket also has to have been bound
908 				 * with BINDMULTI.
909 				 */
910 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
911 					return (EADDRINUSE);
912 			}
913 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
914 			    lport, lookupflags, cred);
915 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
916 				/*
917 				 * XXXRW: If an incpb has had its timewait
918 				 * state recycled, we treat the address as
919 				 * being in use (for now).  This is better
920 				 * than a panic, but not desirable.
921 				 */
922 				tw = intotw(t);
923 				if (tw == NULL ||
924 				    ((reuseport & tw->tw_so_options) == 0 &&
925 					(reuseport_lb &
926 				            tw->tw_so_options) == 0)) {
927 					return (EADDRINUSE);
928 				}
929 			} else if (t &&
930 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
931 				   (reuseport & inp_so_options(t)) == 0 &&
932 				   (reuseport_lb & inp_so_options(t)) == 0) {
933 #ifdef INET6
934 				if (ntohl(sin->sin_addr.s_addr) !=
935 				    INADDR_ANY ||
936 				    ntohl(t->inp_laddr.s_addr) !=
937 				    INADDR_ANY ||
938 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
939 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
940 #endif
941 						return (EADDRINUSE);
942 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
943 					return (EADDRINUSE);
944 			}
945 		}
946 	}
947 	if (*lportp != 0)
948 		lport = *lportp;
949 	if (lport == 0) {
950 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
951 		if (error != 0)
952 			return (error);
953 
954 	}
955 	*laddrp = laddr.s_addr;
956 	*lportp = lport;
957 	return (0);
958 }
959 
960 /*
961  * Connect from a socket to a specified address.
962  * Both address and port must be specified in argument sin.
963  * If don't have a local address for this socket yet,
964  * then pick one.
965  */
966 int
967 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
968     struct ucred *cred, struct mbuf *m)
969 {
970 	u_short lport, fport;
971 	in_addr_t laddr, faddr;
972 	int anonport, error;
973 
974 	INP_WLOCK_ASSERT(inp);
975 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
976 
977 	lport = inp->inp_lport;
978 	laddr = inp->inp_laddr.s_addr;
979 	anonport = (lport == 0);
980 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
981 	    NULL, cred);
982 	if (error)
983 		return (error);
984 
985 	/* Do the initial binding of the local address if required. */
986 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
987 		inp->inp_lport = lport;
988 		inp->inp_laddr.s_addr = laddr;
989 		if (in_pcbinshash(inp) != 0) {
990 			inp->inp_laddr.s_addr = INADDR_ANY;
991 			inp->inp_lport = 0;
992 			return (EAGAIN);
993 		}
994 	}
995 
996 	/* Commit the remaining changes. */
997 	inp->inp_lport = lport;
998 	inp->inp_laddr.s_addr = laddr;
999 	inp->inp_faddr.s_addr = faddr;
1000 	inp->inp_fport = fport;
1001 	in_pcbrehash_mbuf(inp, m);
1002 
1003 	if (anonport)
1004 		inp->inp_flags |= INP_ANONPORT;
1005 	return (0);
1006 }
1007 
1008 int
1009 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1010 {
1011 
1012 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1013 }
1014 
1015 /*
1016  * Do proper source address selection on an unbound socket in case
1017  * of connect. Take jails into account as well.
1018  */
1019 int
1020 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1021     struct ucred *cred)
1022 {
1023 	struct ifaddr *ifa;
1024 	struct sockaddr *sa;
1025 	struct sockaddr_in *sin;
1026 	struct route sro;
1027 	int error;
1028 
1029 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1030 	/*
1031 	 * Bypass source address selection and use the primary jail IP
1032 	 * if requested.
1033 	 */
1034 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1035 		return (0);
1036 
1037 	error = 0;
1038 	bzero(&sro, sizeof(sro));
1039 
1040 	sin = (struct sockaddr_in *)&sro.ro_dst;
1041 	sin->sin_family = AF_INET;
1042 	sin->sin_len = sizeof(struct sockaddr_in);
1043 	sin->sin_addr.s_addr = faddr->s_addr;
1044 
1045 	/*
1046 	 * If route is known our src addr is taken from the i/f,
1047 	 * else punt.
1048 	 *
1049 	 * Find out route to destination.
1050 	 */
1051 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1052 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1053 
1054 	/*
1055 	 * If we found a route, use the address corresponding to
1056 	 * the outgoing interface.
1057 	 *
1058 	 * Otherwise assume faddr is reachable on a directly connected
1059 	 * network and try to find a corresponding interface to take
1060 	 * the source address from.
1061 	 */
1062 	NET_EPOCH_ENTER();
1063 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1064 		struct in_ifaddr *ia;
1065 		struct ifnet *ifp;
1066 
1067 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1068 					inp->inp_socket->so_fibnum));
1069 		if (ia == NULL) {
1070 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1071 						inp->inp_socket->so_fibnum));
1072 
1073 		}
1074 		if (ia == NULL) {
1075 			error = ENETUNREACH;
1076 			goto done;
1077 		}
1078 
1079 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1080 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1081 			goto done;
1082 		}
1083 
1084 		ifp = ia->ia_ifp;
1085 		ia = NULL;
1086 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1087 
1088 			sa = ifa->ifa_addr;
1089 			if (sa->sa_family != AF_INET)
1090 				continue;
1091 			sin = (struct sockaddr_in *)sa;
1092 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1093 				ia = (struct in_ifaddr *)ifa;
1094 				break;
1095 			}
1096 		}
1097 		if (ia != NULL) {
1098 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1099 			goto done;
1100 		}
1101 
1102 		/* 3. As a last resort return the 'default' jail address. */
1103 		error = prison_get_ip4(cred, laddr);
1104 		goto done;
1105 	}
1106 
1107 	/*
1108 	 * If the outgoing interface on the route found is not
1109 	 * a loopback interface, use the address from that interface.
1110 	 * In case of jails do those three steps:
1111 	 * 1. check if the interface address belongs to the jail. If so use it.
1112 	 * 2. check if we have any address on the outgoing interface
1113 	 *    belonging to this jail. If so use it.
1114 	 * 3. as a last resort return the 'default' jail address.
1115 	 */
1116 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1117 		struct in_ifaddr *ia;
1118 		struct ifnet *ifp;
1119 
1120 		/* If not jailed, use the default returned. */
1121 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1122 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1123 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1124 			goto done;
1125 		}
1126 
1127 		/* Jailed. */
1128 		/* 1. Check if the iface address belongs to the jail. */
1129 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1130 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1131 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1132 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1133 			goto done;
1134 		}
1135 
1136 		/*
1137 		 * 2. Check if we have any address on the outgoing interface
1138 		 *    belonging to this jail.
1139 		 */
1140 		ia = NULL;
1141 		ifp = sro.ro_rt->rt_ifp;
1142 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1143 			sa = ifa->ifa_addr;
1144 			if (sa->sa_family != AF_INET)
1145 				continue;
1146 			sin = (struct sockaddr_in *)sa;
1147 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1148 				ia = (struct in_ifaddr *)ifa;
1149 				break;
1150 			}
1151 		}
1152 		if (ia != NULL) {
1153 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1154 			goto done;
1155 		}
1156 
1157 		/* 3. As a last resort return the 'default' jail address. */
1158 		error = prison_get_ip4(cred, laddr);
1159 		goto done;
1160 	}
1161 
1162 	/*
1163 	 * The outgoing interface is marked with 'loopback net', so a route
1164 	 * to ourselves is here.
1165 	 * Try to find the interface of the destination address and then
1166 	 * take the address from there. That interface is not necessarily
1167 	 * a loopback interface.
1168 	 * In case of jails, check that it is an address of the jail
1169 	 * and if we cannot find, fall back to the 'default' jail address.
1170 	 */
1171 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1172 		struct sockaddr_in sain;
1173 		struct in_ifaddr *ia;
1174 
1175 		bzero(&sain, sizeof(struct sockaddr_in));
1176 		sain.sin_family = AF_INET;
1177 		sain.sin_len = sizeof(struct sockaddr_in);
1178 		sain.sin_addr.s_addr = faddr->s_addr;
1179 
1180 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1181 					inp->inp_socket->so_fibnum));
1182 		if (ia == NULL)
1183 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1184 						inp->inp_socket->so_fibnum));
1185 		if (ia == NULL)
1186 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1187 
1188 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1189 			if (ia == NULL) {
1190 				error = ENETUNREACH;
1191 				goto done;
1192 			}
1193 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1194 			goto done;
1195 		}
1196 
1197 		/* Jailed. */
1198 		if (ia != NULL) {
1199 			struct ifnet *ifp;
1200 
1201 			ifp = ia->ia_ifp;
1202 			ia = NULL;
1203 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1204 				sa = ifa->ifa_addr;
1205 				if (sa->sa_family != AF_INET)
1206 					continue;
1207 				sin = (struct sockaddr_in *)sa;
1208 				if (prison_check_ip4(cred,
1209 				    &sin->sin_addr) == 0) {
1210 					ia = (struct in_ifaddr *)ifa;
1211 					break;
1212 				}
1213 			}
1214 			if (ia != NULL) {
1215 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1216 				goto done;
1217 			}
1218 		}
1219 
1220 		/* 3. As a last resort return the 'default' jail address. */
1221 		error = prison_get_ip4(cred, laddr);
1222 		goto done;
1223 	}
1224 
1225 done:
1226 	NET_EPOCH_EXIT();
1227 	if (sro.ro_rt != NULL)
1228 		RTFREE(sro.ro_rt);
1229 	return (error);
1230 }
1231 
1232 /*
1233  * Set up for a connect from a socket to the specified address.
1234  * On entry, *laddrp and *lportp should contain the current local
1235  * address and port for the PCB; these are updated to the values
1236  * that should be placed in inp_laddr and inp_lport to complete
1237  * the connect.
1238  *
1239  * On success, *faddrp and *fportp will be set to the remote address
1240  * and port. These are not updated in the error case.
1241  *
1242  * If the operation fails because the connection already exists,
1243  * *oinpp will be set to the PCB of that connection so that the
1244  * caller can decide to override it. In all other cases, *oinpp
1245  * is set to NULL.
1246  */
1247 int
1248 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1249     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1250     struct inpcb **oinpp, struct ucred *cred)
1251 {
1252 	struct rm_priotracker in_ifa_tracker;
1253 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1254 	struct in_ifaddr *ia;
1255 	struct inpcb *oinp;
1256 	struct in_addr laddr, faddr;
1257 	u_short lport, fport;
1258 	int error;
1259 
1260 	/*
1261 	 * Because a global state change doesn't actually occur here, a read
1262 	 * lock is sufficient.
1263 	 */
1264 	INP_LOCK_ASSERT(inp);
1265 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1266 
1267 	if (oinpp != NULL)
1268 		*oinpp = NULL;
1269 	if (nam->sa_len != sizeof (*sin))
1270 		return (EINVAL);
1271 	if (sin->sin_family != AF_INET)
1272 		return (EAFNOSUPPORT);
1273 	if (sin->sin_port == 0)
1274 		return (EADDRNOTAVAIL);
1275 	laddr.s_addr = *laddrp;
1276 	lport = *lportp;
1277 	faddr = sin->sin_addr;
1278 	fport = sin->sin_port;
1279 
1280 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1281 		/*
1282 		 * If the destination address is INADDR_ANY,
1283 		 * use the primary local address.
1284 		 * If the supplied address is INADDR_BROADCAST,
1285 		 * and the primary interface supports broadcast,
1286 		 * choose the broadcast address for that interface.
1287 		 */
1288 		if (faddr.s_addr == INADDR_ANY) {
1289 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1290 			faddr =
1291 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1292 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1293 			if (cred != NULL &&
1294 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1295 				return (error);
1296 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1297 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1298 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1299 			    IFF_BROADCAST)
1300 				faddr = satosin(&CK_STAILQ_FIRST(
1301 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1302 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1303 		}
1304 	}
1305 	if (laddr.s_addr == INADDR_ANY) {
1306 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1307 		/*
1308 		 * If the destination address is multicast and an outgoing
1309 		 * interface has been set as a multicast option, prefer the
1310 		 * address of that interface as our source address.
1311 		 */
1312 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1313 		    inp->inp_moptions != NULL) {
1314 			struct ip_moptions *imo;
1315 			struct ifnet *ifp;
1316 
1317 			imo = inp->inp_moptions;
1318 			if (imo->imo_multicast_ifp != NULL) {
1319 				ifp = imo->imo_multicast_ifp;
1320 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1321 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1322 					if ((ia->ia_ifp == ifp) &&
1323 					    (cred == NULL ||
1324 					    prison_check_ip4(cred,
1325 					    &ia->ia_addr.sin_addr) == 0))
1326 						break;
1327 				}
1328 				if (ia == NULL)
1329 					error = EADDRNOTAVAIL;
1330 				else {
1331 					laddr = ia->ia_addr.sin_addr;
1332 					error = 0;
1333 				}
1334 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1335 			}
1336 		}
1337 		if (error)
1338 			return (error);
1339 	}
1340 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1341 	    laddr, lport, 0, NULL);
1342 	if (oinp != NULL) {
1343 		if (oinpp != NULL)
1344 			*oinpp = oinp;
1345 		return (EADDRINUSE);
1346 	}
1347 	if (lport == 0) {
1348 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1349 		    cred);
1350 		if (error)
1351 			return (error);
1352 	}
1353 	*laddrp = laddr.s_addr;
1354 	*lportp = lport;
1355 	*faddrp = faddr.s_addr;
1356 	*fportp = fport;
1357 	return (0);
1358 }
1359 
1360 void
1361 in_pcbdisconnect(struct inpcb *inp)
1362 {
1363 
1364 	INP_WLOCK_ASSERT(inp);
1365 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1366 
1367 	inp->inp_faddr.s_addr = INADDR_ANY;
1368 	inp->inp_fport = 0;
1369 	in_pcbrehash(inp);
1370 }
1371 #endif /* INET */
1372 
1373 /*
1374  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1375  * For most protocols, this will be invoked immediately prior to calling
1376  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1377  * socket, in which case in_pcbfree() is deferred.
1378  */
1379 void
1380 in_pcbdetach(struct inpcb *inp)
1381 {
1382 
1383 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1384 
1385 #ifdef RATELIMIT
1386 	if (inp->inp_snd_tag != NULL)
1387 		in_pcbdetach_txrtlmt(inp);
1388 #endif
1389 	inp->inp_socket->so_pcb = NULL;
1390 	inp->inp_socket = NULL;
1391 }
1392 
1393 /*
1394  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1395  * stability of an inpcb pointer despite the inpcb lock being released.  This
1396  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1397  * but where the inpcb lock may already held, or when acquiring a reference
1398  * via a pcbgroup.
1399  *
1400  * in_pcbref() should be used only to provide brief memory stability, and
1401  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1402  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1403  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1404  * lock and rele are the *only* safe operations that may be performed on the
1405  * inpcb.
1406  *
1407  * While the inpcb will not be freed, releasing the inpcb lock means that the
1408  * connection's state may change, so the caller should be careful to
1409  * revalidate any cached state on reacquiring the lock.  Drop the reference
1410  * using in_pcbrele().
1411  */
1412 void
1413 in_pcbref(struct inpcb *inp)
1414 {
1415 
1416 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1417 
1418 	refcount_acquire(&inp->inp_refcount);
1419 }
1420 
1421 /*
1422  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1423  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1424  * return a flag indicating whether or not the inpcb remains valid.  If it is
1425  * valid, we return with the inpcb lock held.
1426  *
1427  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1428  * reference on an inpcb.  Historically more work was done here (actually, in
1429  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1430  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1431  * about memory stability (and continued use of the write lock).
1432  */
1433 int
1434 in_pcbrele_rlocked(struct inpcb *inp)
1435 {
1436 	struct inpcbinfo *pcbinfo;
1437 
1438 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1439 
1440 	INP_RLOCK_ASSERT(inp);
1441 
1442 	if (refcount_release(&inp->inp_refcount) == 0) {
1443 		/*
1444 		 * If the inpcb has been freed, let the caller know, even if
1445 		 * this isn't the last reference.
1446 		 */
1447 		if (inp->inp_flags2 & INP_FREED) {
1448 			INP_RUNLOCK(inp);
1449 			return (1);
1450 		}
1451 		return (0);
1452 	}
1453 
1454 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1455 #ifdef TCPHPTS
1456 	if (inp->inp_in_hpts || inp->inp_in_input) {
1457 		struct tcp_hpts_entry *hpts;
1458 		/*
1459 		 * We should not be on the hpts at
1460 		 * this point in any form. we must
1461 		 * get the lock to be sure.
1462 		 */
1463 		hpts = tcp_hpts_lock(inp);
1464 		if (inp->inp_in_hpts)
1465 			panic("Hpts:%p inp:%p at free still on hpts",
1466 			      hpts, inp);
1467 		mtx_unlock(&hpts->p_mtx);
1468 		hpts = tcp_input_lock(inp);
1469 		if (inp->inp_in_input)
1470 			panic("Hpts:%p inp:%p at free still on input hpts",
1471 			      hpts, inp);
1472 		mtx_unlock(&hpts->p_mtx);
1473 	}
1474 #endif
1475 	INP_RUNLOCK(inp);
1476 	pcbinfo = inp->inp_pcbinfo;
1477 	uma_zfree(pcbinfo->ipi_zone, inp);
1478 	return (1);
1479 }
1480 
1481 int
1482 in_pcbrele_wlocked(struct inpcb *inp)
1483 {
1484 	struct inpcbinfo *pcbinfo;
1485 
1486 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1487 
1488 	INP_WLOCK_ASSERT(inp);
1489 
1490 	if (refcount_release(&inp->inp_refcount) == 0) {
1491 		/*
1492 		 * If the inpcb has been freed, let the caller know, even if
1493 		 * this isn't the last reference.
1494 		 */
1495 		if (inp->inp_flags2 & INP_FREED) {
1496 			INP_WUNLOCK(inp);
1497 			return (1);
1498 		}
1499 		return (0);
1500 	}
1501 
1502 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1503 #ifdef TCPHPTS
1504 	if (inp->inp_in_hpts || inp->inp_in_input) {
1505 		struct tcp_hpts_entry *hpts;
1506 		/*
1507 		 * We should not be on the hpts at
1508 		 * this point in any form. we must
1509 		 * get the lock to be sure.
1510 		 */
1511 		hpts = tcp_hpts_lock(inp);
1512 		if (inp->inp_in_hpts)
1513 			panic("Hpts:%p inp:%p at free still on hpts",
1514 			      hpts, inp);
1515 		mtx_unlock(&hpts->p_mtx);
1516 		hpts = tcp_input_lock(inp);
1517 		if (inp->inp_in_input)
1518 			panic("Hpts:%p inp:%p at free still on input hpts",
1519 			      hpts, inp);
1520 		mtx_unlock(&hpts->p_mtx);
1521 	}
1522 #endif
1523 	INP_WUNLOCK(inp);
1524 	pcbinfo = inp->inp_pcbinfo;
1525 	uma_zfree(pcbinfo->ipi_zone, inp);
1526 	return (1);
1527 }
1528 
1529 /*
1530  * Temporary wrapper.
1531  */
1532 int
1533 in_pcbrele(struct inpcb *inp)
1534 {
1535 
1536 	return (in_pcbrele_wlocked(inp));
1537 }
1538 
1539 void
1540 in_pcblist_rele_rlocked(epoch_context_t ctx)
1541 {
1542 	struct in_pcblist *il;
1543 	struct inpcb *inp;
1544 	struct inpcbinfo *pcbinfo;
1545 	int i, n;
1546 
1547 	il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1548 	pcbinfo = il->il_pcbinfo;
1549 	n = il->il_count;
1550 	INP_INFO_WLOCK(pcbinfo);
1551 	for (i = 0; i < n; i++) {
1552 		inp = il->il_inp_list[i];
1553 		INP_RLOCK(inp);
1554 		if (!in_pcbrele_rlocked(inp))
1555 			INP_RUNLOCK(inp);
1556 	}
1557 	INP_INFO_WUNLOCK(pcbinfo);
1558 	free(il, M_TEMP);
1559 }
1560 
1561 static void
1562 inpcbport_free(epoch_context_t ctx)
1563 {
1564 	struct inpcbport *phd;
1565 
1566 	phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1567 	free(phd, M_PCB);
1568 }
1569 
1570 static void
1571 in_pcbfree_deferred(epoch_context_t ctx)
1572 {
1573 	struct inpcb *inp;
1574 	int released __unused;
1575 
1576 	inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1577 
1578 	INP_WLOCK(inp);
1579 #ifdef INET
1580 	struct ip_moptions *imo = inp->inp_moptions;
1581 	inp->inp_moptions = NULL;
1582 #endif
1583 	/* XXXRW: Do as much as possible here. */
1584 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1585 	if (inp->inp_sp != NULL)
1586 		ipsec_delete_pcbpolicy(inp);
1587 #endif
1588 #ifdef INET6
1589 	struct ip6_moptions *im6o = NULL;
1590 	if (inp->inp_vflag & INP_IPV6PROTO) {
1591 		ip6_freepcbopts(inp->in6p_outputopts);
1592 		im6o = inp->in6p_moptions;
1593 		inp->in6p_moptions = NULL;
1594 	}
1595 #endif
1596 	if (inp->inp_options)
1597 		(void)m_free(inp->inp_options);
1598 	inp->inp_vflag = 0;
1599 	crfree(inp->inp_cred);
1600 #ifdef MAC
1601 	mac_inpcb_destroy(inp);
1602 #endif
1603 	released = in_pcbrele_wlocked(inp);
1604 	MPASS(released);
1605 #ifdef INET6
1606 	ip6_freemoptions(im6o);
1607 #endif
1608 #ifdef INET
1609 	inp_freemoptions(imo);
1610 #endif
1611 }
1612 
1613 /*
1614  * Unconditionally schedule an inpcb to be freed by decrementing its
1615  * reference count, which should occur only after the inpcb has been detached
1616  * from its socket.  If another thread holds a temporary reference (acquired
1617  * using in_pcbref()) then the free is deferred until that reference is
1618  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1619  * work, including removal from global lists, is done in this context, where
1620  * the pcbinfo lock is held.
1621  */
1622 void
1623 in_pcbfree(struct inpcb *inp)
1624 {
1625 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1626 
1627 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1628 	KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1629 	    ("%s: called twice for pcb %p", __func__, inp));
1630 	if (inp->inp_flags2 & INP_FREED) {
1631 		INP_WUNLOCK(inp);
1632 		return;
1633 	}
1634 
1635 #ifdef INVARIANTS
1636 	if (pcbinfo == &V_tcbinfo) {
1637 		INP_INFO_LOCK_ASSERT(pcbinfo);
1638 	} else {
1639 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1640 	}
1641 #endif
1642 	INP_WLOCK_ASSERT(inp);
1643 	INP_LIST_WLOCK(pcbinfo);
1644 	in_pcbremlists(inp);
1645 	INP_LIST_WUNLOCK(pcbinfo);
1646 	RO_INVALIDATE_CACHE(&inp->inp_route);
1647 	/* mark as destruction in progress */
1648 	inp->inp_flags2 |= INP_FREED;
1649 	INP_WUNLOCK(inp);
1650 	epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1651 }
1652 
1653 /*
1654  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1655  * port reservation, and preventing it from being returned by inpcb lookups.
1656  *
1657  * It is used by TCP to mark an inpcb as unused and avoid future packet
1658  * delivery or event notification when a socket remains open but TCP has
1659  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1660  * or a RST on the wire, and allows the port binding to be reused while still
1661  * maintaining the invariant that so_pcb always points to a valid inpcb until
1662  * in_pcbdetach().
1663  *
1664  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1665  * in_pcbnotifyall() and in_pcbpurgeif0()?
1666  */
1667 void
1668 in_pcbdrop(struct inpcb *inp)
1669 {
1670 
1671 	INP_WLOCK_ASSERT(inp);
1672 #ifdef INVARIANTS
1673 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1674 		MPASS(inp->inp_refcount > 1);
1675 #endif
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;
1956 	const struct inpcblbgrouphead *hdr;
1957 	struct inpcblbgroup *grp;
1958 	uint32_t idx;
1959 
1960 	INP_HASH_LOCK_ASSERT(pcbinfo);
1961 
1962 	hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport,
1963 	    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 	local_wild = NULL;
1975 	LIST_FOREACH(grp, hdr, il_list) {
1976 #ifdef INET6
1977 		if (!(grp->il_vflag & INP_IPV4))
1978 			continue;
1979 #endif
1980 		if (grp->il_lport != lport)
1981 			continue;
1982 
1983 		idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1984 		    grp->il_inpcnt;
1985 		if (grp->il_laddr.s_addr == laddr->s_addr)
1986 			return (grp->il_inp[idx]);
1987 		if (grp->il_laddr.s_addr == INADDR_ANY &&
1988 		    (lookupflags & INPLOOKUP_WILDCARD) != 0)
1989 			local_wild = grp->il_inp[idx];
1990 	}
1991 	return (local_wild);
1992 }
1993 
1994 #ifdef PCBGROUP
1995 /*
1996  * Lookup PCB in hash list, using pcbgroup tables.
1997  */
1998 static struct inpcb *
1999 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2000     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2001     u_int lport_arg, int lookupflags, struct ifnet *ifp)
2002 {
2003 	struct inpcbhead *head;
2004 	struct inpcb *inp, *tmpinp;
2005 	u_short fport = fport_arg, lport = lport_arg;
2006 	bool locked;
2007 
2008 	/*
2009 	 * First look for an exact match.
2010 	 */
2011 	tmpinp = NULL;
2012 	INP_GROUP_LOCK(pcbgroup);
2013 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2014 	    pcbgroup->ipg_hashmask)];
2015 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2016 #ifdef INET6
2017 		/* XXX inp locking */
2018 		if ((inp->inp_vflag & INP_IPV4) == 0)
2019 			continue;
2020 #endif
2021 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2022 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2023 		    inp->inp_fport == fport &&
2024 		    inp->inp_lport == lport) {
2025 			/*
2026 			 * XXX We should be able to directly return
2027 			 * the inp here, without any checks.
2028 			 * Well unless both bound with SO_REUSEPORT?
2029 			 */
2030 			if (prison_flag(inp->inp_cred, PR_IP4))
2031 				goto found;
2032 			if (tmpinp == NULL)
2033 				tmpinp = inp;
2034 		}
2035 	}
2036 	if (tmpinp != NULL) {
2037 		inp = tmpinp;
2038 		goto found;
2039 	}
2040 
2041 #ifdef	RSS
2042 	/*
2043 	 * For incoming connections, we may wish to do a wildcard
2044 	 * match for an RSS-local socket.
2045 	 */
2046 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2047 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2048 #ifdef INET6
2049 		struct inpcb *local_wild_mapped = NULL;
2050 #endif
2051 		struct inpcb *jail_wild = NULL;
2052 		struct inpcbhead *head;
2053 		int injail;
2054 
2055 		/*
2056 		 * Order of socket selection - we always prefer jails.
2057 		 *      1. jailed, non-wild.
2058 		 *      2. jailed, wild.
2059 		 *      3. non-jailed, non-wild.
2060 		 *      4. non-jailed, wild.
2061 		 */
2062 
2063 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2064 		    lport, 0, pcbgroup->ipg_hashmask)];
2065 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2066 #ifdef INET6
2067 			/* XXX inp locking */
2068 			if ((inp->inp_vflag & INP_IPV4) == 0)
2069 				continue;
2070 #endif
2071 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2072 			    inp->inp_lport != lport)
2073 				continue;
2074 
2075 			injail = prison_flag(inp->inp_cred, PR_IP4);
2076 			if (injail) {
2077 				if (prison_check_ip4(inp->inp_cred,
2078 				    &laddr) != 0)
2079 					continue;
2080 			} else {
2081 				if (local_exact != NULL)
2082 					continue;
2083 			}
2084 
2085 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2086 				if (injail)
2087 					goto found;
2088 				else
2089 					local_exact = inp;
2090 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2091 #ifdef INET6
2092 				/* XXX inp locking, NULL check */
2093 				if (inp->inp_vflag & INP_IPV6PROTO)
2094 					local_wild_mapped = inp;
2095 				else
2096 #endif
2097 					if (injail)
2098 						jail_wild = inp;
2099 					else
2100 						local_wild = inp;
2101 			}
2102 		} /* LIST_FOREACH */
2103 
2104 		inp = jail_wild;
2105 		if (inp == NULL)
2106 			inp = local_exact;
2107 		if (inp == NULL)
2108 			inp = local_wild;
2109 #ifdef INET6
2110 		if (inp == NULL)
2111 			inp = local_wild_mapped;
2112 #endif
2113 		if (inp != NULL)
2114 			goto found;
2115 	}
2116 #endif
2117 
2118 	/*
2119 	 * Then look for a wildcard match, if requested.
2120 	 */
2121 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2122 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2123 #ifdef INET6
2124 		struct inpcb *local_wild_mapped = NULL;
2125 #endif
2126 		struct inpcb *jail_wild = NULL;
2127 		struct inpcbhead *head;
2128 		int injail;
2129 
2130 		/*
2131 		 * Order of socket selection - we always prefer jails.
2132 		 *      1. jailed, non-wild.
2133 		 *      2. jailed, wild.
2134 		 *      3. non-jailed, non-wild.
2135 		 *      4. non-jailed, wild.
2136 		 */
2137 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2138 		    0, pcbinfo->ipi_wildmask)];
2139 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2140 #ifdef INET6
2141 			/* XXX inp locking */
2142 			if ((inp->inp_vflag & INP_IPV4) == 0)
2143 				continue;
2144 #endif
2145 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2146 			    inp->inp_lport != lport)
2147 				continue;
2148 
2149 			injail = prison_flag(inp->inp_cred, PR_IP4);
2150 			if (injail) {
2151 				if (prison_check_ip4(inp->inp_cred,
2152 				    &laddr) != 0)
2153 					continue;
2154 			} else {
2155 				if (local_exact != NULL)
2156 					continue;
2157 			}
2158 
2159 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2160 				if (injail)
2161 					goto found;
2162 				else
2163 					local_exact = inp;
2164 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2165 #ifdef INET6
2166 				/* XXX inp locking, NULL check */
2167 				if (inp->inp_vflag & INP_IPV6PROTO)
2168 					local_wild_mapped = inp;
2169 				else
2170 #endif
2171 					if (injail)
2172 						jail_wild = inp;
2173 					else
2174 						local_wild = inp;
2175 			}
2176 		} /* LIST_FOREACH */
2177 		inp = jail_wild;
2178 		if (inp == NULL)
2179 			inp = local_exact;
2180 		if (inp == NULL)
2181 			inp = local_wild;
2182 #ifdef INET6
2183 		if (inp == NULL)
2184 			inp = local_wild_mapped;
2185 #endif
2186 		if (inp != NULL)
2187 			goto found;
2188 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2189 	INP_GROUP_UNLOCK(pcbgroup);
2190 	return (NULL);
2191 
2192 found:
2193 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2194 		locked = INP_TRY_WLOCK(inp);
2195 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2196 		locked = INP_TRY_RLOCK(inp);
2197 	else
2198 		panic("%s: locking bug", __func__);
2199 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2200 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2201 			INP_WUNLOCK(inp);
2202 		else
2203 			INP_RUNLOCK(inp);
2204 		return (NULL);
2205 	} else if (!locked)
2206 		in_pcbref(inp);
2207 	INP_GROUP_UNLOCK(pcbgroup);
2208 	if (!locked) {
2209 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2210 			INP_WLOCK(inp);
2211 			if (in_pcbrele_wlocked(inp))
2212 				return (NULL);
2213 		} else {
2214 			INP_RLOCK(inp);
2215 			if (in_pcbrele_rlocked(inp))
2216 				return (NULL);
2217 		}
2218 	}
2219 #ifdef INVARIANTS
2220 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2221 		INP_WLOCK_ASSERT(inp);
2222 	else
2223 		INP_RLOCK_ASSERT(inp);
2224 #endif
2225 	return (inp);
2226 }
2227 #endif /* PCBGROUP */
2228 
2229 /*
2230  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2231  * that the caller has locked the hash list, and will not perform any further
2232  * locking or reference operations on either the hash list or the connection.
2233  */
2234 static struct inpcb *
2235 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2236     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2237     struct ifnet *ifp)
2238 {
2239 	struct inpcbhead *head;
2240 	struct inpcb *inp, *tmpinp;
2241 	u_short fport = fport_arg, lport = lport_arg;
2242 
2243 #ifdef INVARIANTS
2244 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2245 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2246 	if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2247 		MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2248 #endif
2249 	/*
2250 	 * First look for an exact match.
2251 	 */
2252 	tmpinp = NULL;
2253 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2254 	    pcbinfo->ipi_hashmask)];
2255 	CK_LIST_FOREACH(inp, head, inp_hash) {
2256 #ifdef INET6
2257 		/* XXX inp locking */
2258 		if ((inp->inp_vflag & INP_IPV4) == 0)
2259 			continue;
2260 #endif
2261 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2262 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2263 		    inp->inp_fport == fport &&
2264 		    inp->inp_lport == lport) {
2265 			/*
2266 			 * XXX We should be able to directly return
2267 			 * the inp here, without any checks.
2268 			 * Well unless both bound with SO_REUSEPORT?
2269 			 */
2270 			if (prison_flag(inp->inp_cred, PR_IP4))
2271 				return (inp);
2272 			if (tmpinp == NULL)
2273 				tmpinp = inp;
2274 		}
2275 	}
2276 	if (tmpinp != NULL)
2277 		return (tmpinp);
2278 
2279 	/*
2280 	 * Then look in lb group (for wildcard match).
2281 	 */
2282 	if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2283 		(lookupflags & INPLOOKUP_WILDCARD)) {
2284 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2285 		    fport, lookupflags);
2286 		if (inp != NULL) {
2287 			return (inp);
2288 		}
2289 	}
2290 
2291 	/*
2292 	 * Then look for a wildcard match, if requested.
2293 	 */
2294 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2295 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2296 #ifdef INET6
2297 		struct inpcb *local_wild_mapped = NULL;
2298 #endif
2299 		struct inpcb *jail_wild = NULL;
2300 		int injail;
2301 
2302 		/*
2303 		 * Order of socket selection - we always prefer jails.
2304 		 *      1. jailed, non-wild.
2305 		 *      2. jailed, wild.
2306 		 *      3. non-jailed, non-wild.
2307 		 *      4. non-jailed, wild.
2308 		 */
2309 
2310 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2311 		    0, pcbinfo->ipi_hashmask)];
2312 		CK_LIST_FOREACH(inp, head, inp_hash) {
2313 #ifdef INET6
2314 			/* XXX inp locking */
2315 			if ((inp->inp_vflag & INP_IPV4) == 0)
2316 				continue;
2317 #endif
2318 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2319 			    inp->inp_lport != lport)
2320 				continue;
2321 
2322 			injail = prison_flag(inp->inp_cred, PR_IP4);
2323 			if (injail) {
2324 				if (prison_check_ip4(inp->inp_cred,
2325 				    &laddr) != 0)
2326 					continue;
2327 			} else {
2328 				if (local_exact != NULL)
2329 					continue;
2330 			}
2331 
2332 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2333 				if (injail)
2334 					return (inp);
2335 				else
2336 					local_exact = inp;
2337 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2338 #ifdef INET6
2339 				/* XXX inp locking, NULL check */
2340 				if (inp->inp_vflag & INP_IPV6PROTO)
2341 					local_wild_mapped = inp;
2342 				else
2343 #endif
2344 					if (injail)
2345 						jail_wild = inp;
2346 					else
2347 						local_wild = inp;
2348 			}
2349 		} /* LIST_FOREACH */
2350 		if (jail_wild != NULL)
2351 			return (jail_wild);
2352 		if (local_exact != NULL)
2353 			return (local_exact);
2354 		if (local_wild != NULL)
2355 			return (local_wild);
2356 #ifdef INET6
2357 		if (local_wild_mapped != NULL)
2358 			return (local_wild_mapped);
2359 #endif
2360 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2361 
2362 	return (NULL);
2363 }
2364 
2365 /*
2366  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2367  * hash list lock, and will return the inpcb locked (i.e., requires
2368  * INPLOOKUP_LOCKPCB).
2369  */
2370 static struct inpcb *
2371 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2372     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2373     struct ifnet *ifp)
2374 {
2375 	struct inpcb *inp;
2376 
2377 	INP_HASH_RLOCK(pcbinfo);
2378 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2379 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2380 	if (inp != NULL) {
2381 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2382 			INP_WLOCK(inp);
2383 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2384 				INP_WUNLOCK(inp);
2385 				inp = NULL;
2386 			}
2387 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2388 			INP_RLOCK(inp);
2389 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2390 				INP_RUNLOCK(inp);
2391 				inp = NULL;
2392 			}
2393 		} else
2394 			panic("%s: locking bug", __func__);
2395 #ifdef INVARIANTS
2396 		if (inp != NULL) {
2397 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2398 				INP_WLOCK_ASSERT(inp);
2399 			else
2400 				INP_RLOCK_ASSERT(inp);
2401 		}
2402 #endif
2403 	}
2404 	INP_HASH_RUNLOCK(pcbinfo);
2405 	return (inp);
2406 }
2407 
2408 /*
2409  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2410  * from which a pre-calculated hash value may be extracted.
2411  *
2412  * Possibly more of this logic should be in in_pcbgroup.c.
2413  */
2414 struct inpcb *
2415 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2416     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2417 {
2418 #if defined(PCBGROUP) && !defined(RSS)
2419 	struct inpcbgroup *pcbgroup;
2420 #endif
2421 
2422 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2423 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2424 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2425 	    ("%s: LOCKPCB not set", __func__));
2426 
2427 	/*
2428 	 * When not using RSS, use connection groups in preference to the
2429 	 * reservation table when looking up 4-tuples.  When using RSS, just
2430 	 * use the reservation table, due to the cost of the Toeplitz hash
2431 	 * in software.
2432 	 *
2433 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2434 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2435 	 * in software.
2436 	 */
2437 #if defined(PCBGROUP) && !defined(RSS)
2438 	if (in_pcbgroup_enabled(pcbinfo)) {
2439 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2440 		    fport);
2441 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2442 		    laddr, lport, lookupflags, ifp));
2443 	}
2444 #endif
2445 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2446 	    lookupflags, ifp));
2447 }
2448 
2449 struct inpcb *
2450 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2451     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2452     struct ifnet *ifp, struct mbuf *m)
2453 {
2454 #ifdef PCBGROUP
2455 	struct inpcbgroup *pcbgroup;
2456 #endif
2457 
2458 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2459 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2460 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2461 	    ("%s: LOCKPCB not set", __func__));
2462 
2463 #ifdef PCBGROUP
2464 	/*
2465 	 * If we can use a hardware-generated hash to look up the connection
2466 	 * group, use that connection group to find the inpcb.  Otherwise
2467 	 * fall back on a software hash -- or the reservation table if we're
2468 	 * using RSS.
2469 	 *
2470 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2471 	 */
2472 	if (in_pcbgroup_enabled(pcbinfo) &&
2473 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2474 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2475 		    m->m_pkthdr.flowid);
2476 		if (pcbgroup != NULL)
2477 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2478 			    fport, laddr, lport, lookupflags, ifp));
2479 #ifndef RSS
2480 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2481 		    fport);
2482 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2483 		    laddr, lport, lookupflags, ifp));
2484 #endif
2485 	}
2486 #endif
2487 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2488 	    lookupflags, ifp));
2489 }
2490 #endif /* INET */
2491 
2492 /*
2493  * Insert PCB onto various hash lists.
2494  */
2495 static int
2496 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2497 {
2498 	struct inpcbhead *pcbhash;
2499 	struct inpcbporthead *pcbporthash;
2500 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2501 	struct inpcbport *phd;
2502 	u_int32_t hashkey_faddr;
2503 	int so_options;
2504 
2505 	INP_WLOCK_ASSERT(inp);
2506 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2507 
2508 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2509 	    ("in_pcbinshash: INP_INHASHLIST"));
2510 
2511 #ifdef INET6
2512 	if (inp->inp_vflag & INP_IPV6)
2513 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2514 	else
2515 #endif
2516 	hashkey_faddr = inp->inp_faddr.s_addr;
2517 
2518 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2519 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2520 
2521 	pcbporthash = &pcbinfo->ipi_porthashbase[
2522 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2523 
2524 	/*
2525 	 * Add entry to load balance group.
2526 	 * Only do this if SO_REUSEPORT_LB is set.
2527 	 */
2528 	so_options = inp_so_options(inp);
2529 	if (so_options & SO_REUSEPORT_LB) {
2530 		int ret = in_pcbinslbgrouphash(inp);
2531 		if (ret) {
2532 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2533 			return (ret);
2534 		}
2535 	}
2536 
2537 	/*
2538 	 * Go through port list and look for a head for this lport.
2539 	 */
2540 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2541 		if (phd->phd_port == inp->inp_lport)
2542 			break;
2543 	}
2544 	/*
2545 	 * If none exists, malloc one and tack it on.
2546 	 */
2547 	if (phd == NULL) {
2548 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2549 		if (phd == NULL) {
2550 			return (ENOBUFS); /* XXX */
2551 		}
2552 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2553 		phd->phd_port = inp->inp_lport;
2554 		CK_LIST_INIT(&phd->phd_pcblist);
2555 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2556 	}
2557 	inp->inp_phd = phd;
2558 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2559 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2560 	inp->inp_flags |= INP_INHASHLIST;
2561 #ifdef PCBGROUP
2562 	if (do_pcbgroup_update)
2563 		in_pcbgroup_update(inp);
2564 #endif
2565 	return (0);
2566 }
2567 
2568 /*
2569  * For now, there are two public interfaces to insert an inpcb into the hash
2570  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2571  * is used only in the TCP syncache, where in_pcbinshash is called before the
2572  * full 4-tuple is set for the inpcb, and we don't want to install in the
2573  * pcbgroup until later.
2574  *
2575  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2576  * connection groups, and partially initialised inpcbs should not be exposed
2577  * to either reservation hash tables or pcbgroups.
2578  */
2579 int
2580 in_pcbinshash(struct inpcb *inp)
2581 {
2582 
2583 	return (in_pcbinshash_internal(inp, 1));
2584 }
2585 
2586 int
2587 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2588 {
2589 
2590 	return (in_pcbinshash_internal(inp, 0));
2591 }
2592 
2593 /*
2594  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2595  * changed. NOTE: This does not handle the case of the lport changing (the
2596  * hashed port list would have to be updated as well), so the lport must
2597  * not change after in_pcbinshash() has been called.
2598  */
2599 void
2600 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2601 {
2602 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2603 	struct inpcbhead *head;
2604 	u_int32_t hashkey_faddr;
2605 
2606 	INP_WLOCK_ASSERT(inp);
2607 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2608 
2609 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2610 	    ("in_pcbrehash: !INP_INHASHLIST"));
2611 
2612 #ifdef INET6
2613 	if (inp->inp_vflag & INP_IPV6)
2614 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2615 	else
2616 #endif
2617 	hashkey_faddr = inp->inp_faddr.s_addr;
2618 
2619 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2620 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2621 
2622 	CK_LIST_REMOVE(inp, inp_hash);
2623 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2624 
2625 #ifdef PCBGROUP
2626 	if (m != NULL)
2627 		in_pcbgroup_update_mbuf(inp, m);
2628 	else
2629 		in_pcbgroup_update(inp);
2630 #endif
2631 }
2632 
2633 void
2634 in_pcbrehash(struct inpcb *inp)
2635 {
2636 
2637 	in_pcbrehash_mbuf(inp, NULL);
2638 }
2639 
2640 /*
2641  * Remove PCB from various lists.
2642  */
2643 static void
2644 in_pcbremlists(struct inpcb *inp)
2645 {
2646 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2647 
2648 #ifdef INVARIANTS
2649 	if (pcbinfo == &V_tcbinfo) {
2650 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2651 	} else {
2652 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2653 	}
2654 #endif
2655 
2656 	INP_WLOCK_ASSERT(inp);
2657 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2658 
2659 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2660 	if (inp->inp_flags & INP_INHASHLIST) {
2661 		struct inpcbport *phd = inp->inp_phd;
2662 
2663 		INP_HASH_WLOCK(pcbinfo);
2664 
2665 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2666 		in_pcbremlbgrouphash(inp);
2667 
2668 		CK_LIST_REMOVE(inp, inp_hash);
2669 		CK_LIST_REMOVE(inp, inp_portlist);
2670 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2671 			CK_LIST_REMOVE(phd, phd_hash);
2672 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2673 		}
2674 		INP_HASH_WUNLOCK(pcbinfo);
2675 		inp->inp_flags &= ~INP_INHASHLIST;
2676 	}
2677 	CK_LIST_REMOVE(inp, inp_list);
2678 	pcbinfo->ipi_count--;
2679 #ifdef PCBGROUP
2680 	in_pcbgroup_remove(inp);
2681 #endif
2682 }
2683 
2684 /*
2685  * Check for alternatives when higher level complains
2686  * about service problems.  For now, invalidate cached
2687  * routing information.  If the route was created dynamically
2688  * (by a redirect), time to try a default gateway again.
2689  */
2690 void
2691 in_losing(struct inpcb *inp)
2692 {
2693 
2694 	RO_INVALIDATE_CACHE(&inp->inp_route);
2695 	return;
2696 }
2697 
2698 /*
2699  * A set label operation has occurred at the socket layer, propagate the
2700  * label change into the in_pcb for the socket.
2701  */
2702 void
2703 in_pcbsosetlabel(struct socket *so)
2704 {
2705 #ifdef MAC
2706 	struct inpcb *inp;
2707 
2708 	inp = sotoinpcb(so);
2709 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2710 
2711 	INP_WLOCK(inp);
2712 	SOCK_LOCK(so);
2713 	mac_inpcb_sosetlabel(so, inp);
2714 	SOCK_UNLOCK(so);
2715 	INP_WUNLOCK(inp);
2716 #endif
2717 }
2718 
2719 /*
2720  * ipport_tick runs once per second, determining if random port allocation
2721  * should be continued.  If more than ipport_randomcps ports have been
2722  * allocated in the last second, then we return to sequential port
2723  * allocation. We return to random allocation only once we drop below
2724  * ipport_randomcps for at least ipport_randomtime seconds.
2725  */
2726 static void
2727 ipport_tick(void *xtp)
2728 {
2729 	VNET_ITERATOR_DECL(vnet_iter);
2730 
2731 	VNET_LIST_RLOCK_NOSLEEP();
2732 	VNET_FOREACH(vnet_iter) {
2733 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2734 		if (V_ipport_tcpallocs <=
2735 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2736 			if (V_ipport_stoprandom > 0)
2737 				V_ipport_stoprandom--;
2738 		} else
2739 			V_ipport_stoprandom = V_ipport_randomtime;
2740 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2741 		CURVNET_RESTORE();
2742 	}
2743 	VNET_LIST_RUNLOCK_NOSLEEP();
2744 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2745 }
2746 
2747 static void
2748 ip_fini(void *xtp)
2749 {
2750 
2751 	callout_stop(&ipport_tick_callout);
2752 }
2753 
2754 /*
2755  * The ipport_callout should start running at about the time we attach the
2756  * inet or inet6 domains.
2757  */
2758 static void
2759 ipport_tick_init(const void *unused __unused)
2760 {
2761 
2762 	/* Start ipport_tick. */
2763 	callout_init(&ipport_tick_callout, 1);
2764 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2765 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2766 		SHUTDOWN_PRI_DEFAULT);
2767 }
2768 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2769     ipport_tick_init, NULL);
2770 
2771 void
2772 inp_wlock(struct inpcb *inp)
2773 {
2774 
2775 	INP_WLOCK(inp);
2776 }
2777 
2778 void
2779 inp_wunlock(struct inpcb *inp)
2780 {
2781 
2782 	INP_WUNLOCK(inp);
2783 }
2784 
2785 void
2786 inp_rlock(struct inpcb *inp)
2787 {
2788 
2789 	INP_RLOCK(inp);
2790 }
2791 
2792 void
2793 inp_runlock(struct inpcb *inp)
2794 {
2795 
2796 	INP_RUNLOCK(inp);
2797 }
2798 
2799 #ifdef INVARIANT_SUPPORT
2800 void
2801 inp_lock_assert(struct inpcb *inp)
2802 {
2803 
2804 	INP_WLOCK_ASSERT(inp);
2805 }
2806 
2807 void
2808 inp_unlock_assert(struct inpcb *inp)
2809 {
2810 
2811 	INP_UNLOCK_ASSERT(inp);
2812 }
2813 #endif
2814 
2815 void
2816 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2817 {
2818 	struct inpcb *inp;
2819 
2820 	INP_INFO_WLOCK(&V_tcbinfo);
2821 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2822 		INP_WLOCK(inp);
2823 		func(inp, arg);
2824 		INP_WUNLOCK(inp);
2825 	}
2826 	INP_INFO_WUNLOCK(&V_tcbinfo);
2827 }
2828 
2829 struct socket *
2830 inp_inpcbtosocket(struct inpcb *inp)
2831 {
2832 
2833 	INP_WLOCK_ASSERT(inp);
2834 	return (inp->inp_socket);
2835 }
2836 
2837 struct tcpcb *
2838 inp_inpcbtotcpcb(struct inpcb *inp)
2839 {
2840 
2841 	INP_WLOCK_ASSERT(inp);
2842 	return ((struct tcpcb *)inp->inp_ppcb);
2843 }
2844 
2845 int
2846 inp_ip_tos_get(const struct inpcb *inp)
2847 {
2848 
2849 	return (inp->inp_ip_tos);
2850 }
2851 
2852 void
2853 inp_ip_tos_set(struct inpcb *inp, int val)
2854 {
2855 
2856 	inp->inp_ip_tos = val;
2857 }
2858 
2859 void
2860 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2861     uint32_t *faddr, uint16_t *fp)
2862 {
2863 
2864 	INP_LOCK_ASSERT(inp);
2865 	*laddr = inp->inp_laddr.s_addr;
2866 	*faddr = inp->inp_faddr.s_addr;
2867 	*lp = inp->inp_lport;
2868 	*fp = inp->inp_fport;
2869 }
2870 
2871 struct inpcb *
2872 so_sotoinpcb(struct socket *so)
2873 {
2874 
2875 	return (sotoinpcb(so));
2876 }
2877 
2878 struct tcpcb *
2879 so_sototcpcb(struct socket *so)
2880 {
2881 
2882 	return (sototcpcb(so));
2883 }
2884 
2885 /*
2886  * Create an external-format (``xinpcb'') structure using the information in
2887  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2888  * reduce the spew of irrelevant information over this interface, to isolate
2889  * user code from changes in the kernel structure, and potentially to provide
2890  * information-hiding if we decide that some of this information should be
2891  * hidden from users.
2892  */
2893 void
2894 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2895 {
2896 
2897 	xi->xi_len = sizeof(struct xinpcb);
2898 	if (inp->inp_socket)
2899 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2900 	else
2901 		bzero(&xi->xi_socket, sizeof(struct xsocket));
2902 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2903 	xi->inp_gencnt = inp->inp_gencnt;
2904 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2905 	xi->inp_flow = inp->inp_flow;
2906 	xi->inp_flowid = inp->inp_flowid;
2907 	xi->inp_flowtype = inp->inp_flowtype;
2908 	xi->inp_flags = inp->inp_flags;
2909 	xi->inp_flags2 = inp->inp_flags2;
2910 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2911 	xi->in6p_cksum = inp->in6p_cksum;
2912 	xi->in6p_hops = inp->in6p_hops;
2913 	xi->inp_ip_tos = inp->inp_ip_tos;
2914 	xi->inp_vflag = inp->inp_vflag;
2915 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2916 	xi->inp_ip_p = inp->inp_ip_p;
2917 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2918 }
2919 
2920 #ifdef DDB
2921 static void
2922 db_print_indent(int indent)
2923 {
2924 	int i;
2925 
2926 	for (i = 0; i < indent; i++)
2927 		db_printf(" ");
2928 }
2929 
2930 static void
2931 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2932 {
2933 	char faddr_str[48], laddr_str[48];
2934 
2935 	db_print_indent(indent);
2936 	db_printf("%s at %p\n", name, inc);
2937 
2938 	indent += 2;
2939 
2940 #ifdef INET6
2941 	if (inc->inc_flags & INC_ISIPV6) {
2942 		/* IPv6. */
2943 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2944 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2945 	} else
2946 #endif
2947 	{
2948 		/* IPv4. */
2949 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2950 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2951 	}
2952 	db_print_indent(indent);
2953 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2954 	    ntohs(inc->inc_lport));
2955 	db_print_indent(indent);
2956 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2957 	    ntohs(inc->inc_fport));
2958 }
2959 
2960 static void
2961 db_print_inpflags(int inp_flags)
2962 {
2963 	int comma;
2964 
2965 	comma = 0;
2966 	if (inp_flags & INP_RECVOPTS) {
2967 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2968 		comma = 1;
2969 	}
2970 	if (inp_flags & INP_RECVRETOPTS) {
2971 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2972 		comma = 1;
2973 	}
2974 	if (inp_flags & INP_RECVDSTADDR) {
2975 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2976 		comma = 1;
2977 	}
2978 	if (inp_flags & INP_ORIGDSTADDR) {
2979 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2980 		comma = 1;
2981 	}
2982 	if (inp_flags & INP_HDRINCL) {
2983 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2984 		comma = 1;
2985 	}
2986 	if (inp_flags & INP_HIGHPORT) {
2987 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2988 		comma = 1;
2989 	}
2990 	if (inp_flags & INP_LOWPORT) {
2991 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2992 		comma = 1;
2993 	}
2994 	if (inp_flags & INP_ANONPORT) {
2995 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2996 		comma = 1;
2997 	}
2998 	if (inp_flags & INP_RECVIF) {
2999 		db_printf("%sINP_RECVIF", comma ? ", " : "");
3000 		comma = 1;
3001 	}
3002 	if (inp_flags & INP_MTUDISC) {
3003 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
3004 		comma = 1;
3005 	}
3006 	if (inp_flags & INP_RECVTTL) {
3007 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
3008 		comma = 1;
3009 	}
3010 	if (inp_flags & INP_DONTFRAG) {
3011 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3012 		comma = 1;
3013 	}
3014 	if (inp_flags & INP_RECVTOS) {
3015 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3016 		comma = 1;
3017 	}
3018 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3019 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3020 		comma = 1;
3021 	}
3022 	if (inp_flags & IN6P_PKTINFO) {
3023 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3024 		comma = 1;
3025 	}
3026 	if (inp_flags & IN6P_HOPLIMIT) {
3027 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3028 		comma = 1;
3029 	}
3030 	if (inp_flags & IN6P_HOPOPTS) {
3031 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3032 		comma = 1;
3033 	}
3034 	if (inp_flags & IN6P_DSTOPTS) {
3035 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3036 		comma = 1;
3037 	}
3038 	if (inp_flags & IN6P_RTHDR) {
3039 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3040 		comma = 1;
3041 	}
3042 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3043 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3044 		comma = 1;
3045 	}
3046 	if (inp_flags & IN6P_TCLASS) {
3047 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3048 		comma = 1;
3049 	}
3050 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3051 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3052 		comma = 1;
3053 	}
3054 	if (inp_flags & INP_TIMEWAIT) {
3055 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3056 		comma  = 1;
3057 	}
3058 	if (inp_flags & INP_ONESBCAST) {
3059 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3060 		comma  = 1;
3061 	}
3062 	if (inp_flags & INP_DROPPED) {
3063 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3064 		comma  = 1;
3065 	}
3066 	if (inp_flags & INP_SOCKREF) {
3067 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3068 		comma  = 1;
3069 	}
3070 	if (inp_flags & IN6P_RFC2292) {
3071 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3072 		comma = 1;
3073 	}
3074 	if (inp_flags & IN6P_MTU) {
3075 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3076 		comma = 1;
3077 	}
3078 }
3079 
3080 static void
3081 db_print_inpvflag(u_char inp_vflag)
3082 {
3083 	int comma;
3084 
3085 	comma = 0;
3086 	if (inp_vflag & INP_IPV4) {
3087 		db_printf("%sINP_IPV4", comma ? ", " : "");
3088 		comma  = 1;
3089 	}
3090 	if (inp_vflag & INP_IPV6) {
3091 		db_printf("%sINP_IPV6", comma ? ", " : "");
3092 		comma  = 1;
3093 	}
3094 	if (inp_vflag & INP_IPV6PROTO) {
3095 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3096 		comma  = 1;
3097 	}
3098 }
3099 
3100 static void
3101 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3102 {
3103 
3104 	db_print_indent(indent);
3105 	db_printf("%s at %p\n", name, inp);
3106 
3107 	indent += 2;
3108 
3109 	db_print_indent(indent);
3110 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3111 
3112 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3113 
3114 	db_print_indent(indent);
3115 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3116 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3117 
3118 	db_print_indent(indent);
3119 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3120 	   inp->inp_label, inp->inp_flags);
3121 	db_print_inpflags(inp->inp_flags);
3122 	db_printf(")\n");
3123 
3124 	db_print_indent(indent);
3125 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3126 	    inp->inp_vflag);
3127 	db_print_inpvflag(inp->inp_vflag);
3128 	db_printf(")\n");
3129 
3130 	db_print_indent(indent);
3131 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3132 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3133 
3134 	db_print_indent(indent);
3135 #ifdef INET6
3136 	if (inp->inp_vflag & INP_IPV6) {
3137 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3138 		    "in6p_moptions: %p\n", inp->in6p_options,
3139 		    inp->in6p_outputopts, inp->in6p_moptions);
3140 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3141 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3142 		    inp->in6p_hops);
3143 	} else
3144 #endif
3145 	{
3146 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3147 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3148 		    inp->inp_options, inp->inp_moptions);
3149 	}
3150 
3151 	db_print_indent(indent);
3152 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3153 	    (uintmax_t)inp->inp_gencnt);
3154 }
3155 
3156 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3157 {
3158 	struct inpcb *inp;
3159 
3160 	if (!have_addr) {
3161 		db_printf("usage: show inpcb <addr>\n");
3162 		return;
3163 	}
3164 	inp = (struct inpcb *)addr;
3165 
3166 	db_print_inpcb(inp, "inpcb", 0);
3167 }
3168 #endif /* DDB */
3169 
3170 #ifdef RATELIMIT
3171 /*
3172  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3173  * if any.
3174  */
3175 int
3176 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3177 {
3178 	union if_snd_tag_modify_params params = {
3179 		.rate_limit.max_rate = max_pacing_rate,
3180 	};
3181 	struct m_snd_tag *mst;
3182 	struct ifnet *ifp;
3183 	int error;
3184 
3185 	mst = inp->inp_snd_tag;
3186 	if (mst == NULL)
3187 		return (EINVAL);
3188 
3189 	ifp = mst->ifp;
3190 	if (ifp == NULL)
3191 		return (EINVAL);
3192 
3193 	if (ifp->if_snd_tag_modify == NULL) {
3194 		error = EOPNOTSUPP;
3195 	} else {
3196 		error = ifp->if_snd_tag_modify(mst, &params);
3197 	}
3198 	return (error);
3199 }
3200 
3201 /*
3202  * Query existing TX rate limit based on the existing
3203  * "inp->inp_snd_tag", if any.
3204  */
3205 int
3206 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3207 {
3208 	union if_snd_tag_query_params params = { };
3209 	struct m_snd_tag *mst;
3210 	struct ifnet *ifp;
3211 	int error;
3212 
3213 	mst = inp->inp_snd_tag;
3214 	if (mst == NULL)
3215 		return (EINVAL);
3216 
3217 	ifp = mst->ifp;
3218 	if (ifp == NULL)
3219 		return (EINVAL);
3220 
3221 	if (ifp->if_snd_tag_query == NULL) {
3222 		error = EOPNOTSUPP;
3223 	} else {
3224 		error = ifp->if_snd_tag_query(mst, &params);
3225 		if (error == 0 &&  p_max_pacing_rate != NULL)
3226 			*p_max_pacing_rate = params.rate_limit.max_rate;
3227 	}
3228 	return (error);
3229 }
3230 
3231 /*
3232  * Query existing TX queue level based on the existing
3233  * "inp->inp_snd_tag", if any.
3234  */
3235 int
3236 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3237 {
3238 	union if_snd_tag_query_params params = { };
3239 	struct m_snd_tag *mst;
3240 	struct ifnet *ifp;
3241 	int error;
3242 
3243 	mst = inp->inp_snd_tag;
3244 	if (mst == NULL)
3245 		return (EINVAL);
3246 
3247 	ifp = mst->ifp;
3248 	if (ifp == NULL)
3249 		return (EINVAL);
3250 
3251 	if (ifp->if_snd_tag_query == NULL)
3252 		return (EOPNOTSUPP);
3253 
3254 	error = ifp->if_snd_tag_query(mst, &params);
3255 	if (error == 0 &&  p_txqueue_level != NULL)
3256 		*p_txqueue_level = params.rate_limit.queue_level;
3257 	return (error);
3258 }
3259 
3260 /*
3261  * Allocate a new TX rate limit send tag from the network interface
3262  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3263  */
3264 int
3265 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3266     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3267 {
3268 	union if_snd_tag_alloc_params params = {
3269 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3270 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3271 		.rate_limit.hdr.flowid = flowid,
3272 		.rate_limit.hdr.flowtype = flowtype,
3273 		.rate_limit.max_rate = max_pacing_rate,
3274 	};
3275 	int error;
3276 
3277 	INP_WLOCK_ASSERT(inp);
3278 
3279 	if (inp->inp_snd_tag != NULL)
3280 		return (EINVAL);
3281 
3282 	if (ifp->if_snd_tag_alloc == NULL) {
3283 		error = EOPNOTSUPP;
3284 	} else {
3285 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3286 
3287 		/*
3288 		 * At success increment the refcount on
3289 		 * the send tag's network interface:
3290 		 */
3291 		if (error == 0)
3292 			if_ref(inp->inp_snd_tag->ifp);
3293 	}
3294 	return (error);
3295 }
3296 
3297 /*
3298  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3299  * if any:
3300  */
3301 void
3302 in_pcbdetach_txrtlmt(struct inpcb *inp)
3303 {
3304 	struct m_snd_tag *mst;
3305 	struct ifnet *ifp;
3306 
3307 	INP_WLOCK_ASSERT(inp);
3308 
3309 	mst = inp->inp_snd_tag;
3310 	inp->inp_snd_tag = NULL;
3311 
3312 	if (mst == NULL)
3313 		return;
3314 
3315 	ifp = mst->ifp;
3316 	if (ifp == NULL)
3317 		return;
3318 
3319 	/*
3320 	 * If the device was detached while we still had reference(s)
3321 	 * on the ifp, we assume if_snd_tag_free() was replaced with
3322 	 * stubs.
3323 	 */
3324 	ifp->if_snd_tag_free(mst);
3325 
3326 	/* release reference count on network interface */
3327 	if_rele(ifp);
3328 }
3329 
3330 /*
3331  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3332  * is set in the fast path and will attach/detach/modify the TX rate
3333  * limit send tag based on the socket's so_max_pacing_rate value.
3334  */
3335 void
3336 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3337 {
3338 	struct socket *socket;
3339 	uint32_t max_pacing_rate;
3340 	bool did_upgrade;
3341 	int error;
3342 
3343 	if (inp == NULL)
3344 		return;
3345 
3346 	socket = inp->inp_socket;
3347 	if (socket == NULL)
3348 		return;
3349 
3350 	if (!INP_WLOCKED(inp)) {
3351 		/*
3352 		 * NOTE: If the write locking fails, we need to bail
3353 		 * out and use the non-ratelimited ring for the
3354 		 * transmit until there is a new chance to get the
3355 		 * write lock.
3356 		 */
3357 		if (!INP_TRY_UPGRADE(inp))
3358 			return;
3359 		did_upgrade = 1;
3360 	} else {
3361 		did_upgrade = 0;
3362 	}
3363 
3364 	/*
3365 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3366 	 * because atomic updates are not required since the variable
3367 	 * is checked at every mbuf we send. It is assumed that the
3368 	 * variable read itself will be atomic.
3369 	 */
3370 	max_pacing_rate = socket->so_max_pacing_rate;
3371 
3372 	/*
3373 	 * NOTE: When attaching to a network interface a reference is
3374 	 * made to ensure the network interface doesn't go away until
3375 	 * all ratelimit connections are gone. The network interface
3376 	 * pointers compared below represent valid network interfaces,
3377 	 * except when comparing towards NULL.
3378 	 */
3379 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3380 		error = 0;
3381 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3382 		if (inp->inp_snd_tag != NULL)
3383 			in_pcbdetach_txrtlmt(inp);
3384 		error = 0;
3385 	} else if (inp->inp_snd_tag == NULL) {
3386 		/*
3387 		 * In order to utilize packet pacing with RSS, we need
3388 		 * to wait until there is a valid RSS hash before we
3389 		 * can proceed:
3390 		 */
3391 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3392 			error = EAGAIN;
3393 		} else {
3394 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3395 			    mb->m_pkthdr.flowid, max_pacing_rate);
3396 		}
3397 	} else {
3398 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3399 	}
3400 	if (error == 0 || error == EOPNOTSUPP)
3401 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3402 	if (did_upgrade)
3403 		INP_DOWNGRADE(inp);
3404 }
3405 
3406 /*
3407  * Track route changes for TX rate limiting.
3408  */
3409 void
3410 in_pcboutput_eagain(struct inpcb *inp)
3411 {
3412 	struct socket *socket;
3413 	bool did_upgrade;
3414 
3415 	if (inp == NULL)
3416 		return;
3417 
3418 	socket = inp->inp_socket;
3419 	if (socket == NULL)
3420 		return;
3421 
3422 	if (inp->inp_snd_tag == NULL)
3423 		return;
3424 
3425 	if (!INP_WLOCKED(inp)) {
3426 		/*
3427 		 * NOTE: If the write locking fails, we need to bail
3428 		 * out and use the non-ratelimited ring for the
3429 		 * transmit until there is a new chance to get the
3430 		 * write lock.
3431 		 */
3432 		if (!INP_TRY_UPGRADE(inp))
3433 			return;
3434 		did_upgrade = 1;
3435 	} else {
3436 		did_upgrade = 0;
3437 	}
3438 
3439 	/* detach rate limiting */
3440 	in_pcbdetach_txrtlmt(inp);
3441 
3442 	/* make sure new mbuf send tag allocation is made */
3443 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3444 
3445 	if (did_upgrade)
3446 		INP_DOWNGRADE(inp);
3447 }
3448 #endif /* RATELIMIT */
3449