xref: /freebsd/sys/netinet/in_pcb.c (revision f10421e96d0b7f2de657cabedc1e2884a429c76d)
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 #ifdef INET
90 #include <netinet/in_var.h>
91 #endif
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp_var.h>
94 #ifdef TCPHPTS
95 #include <netinet/tcp_hpts.h>
96 #endif
97 #include <netinet/udp.h>
98 #include <netinet/udp_var.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #endif /* INET6 */
105 #endif
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 
214 #ifdef RATELIMIT
215 counter_u64_t rate_limit_active;
216 counter_u64_t rate_limit_alloc_fail;
217 counter_u64_t rate_limit_set_ok;
218 
219 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD, 0,
220     "IP Rate Limiting");
221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
222     &rate_limit_active, "Active rate limited connections");
223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
224    &rate_limit_alloc_fail, "Rate limited connection failures");
225 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
226    &rate_limit_set_ok, "Rate limited setting succeeded");
227 #endif /* RATELIMIT */
228 
229 #endif /* INET */
230 
231 /*
232  * in_pcb.c: manage the Protocol Control Blocks.
233  *
234  * NOTE: It is assumed that most of these functions will be called with
235  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
236  * functions often modify hash chains or addresses in pcbs.
237  */
238 
239 static struct inpcblbgroup *
240 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
241     uint16_t port, const union in_dependaddr *addr, int size)
242 {
243 	struct inpcblbgroup *grp;
244 	size_t bytes;
245 
246 	bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
247 	grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
248 	if (!grp)
249 		return (NULL);
250 	grp->il_vflag = vflag;
251 	grp->il_lport = port;
252 	grp->il_dependladdr = *addr;
253 	grp->il_inpsiz = size;
254 	CK_LIST_INSERT_HEAD(hdr, grp, il_list);
255 	return (grp);
256 }
257 
258 static void
259 in_pcblbgroup_free_deferred(epoch_context_t ctx)
260 {
261 	struct inpcblbgroup *grp;
262 
263 	grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
264 	free(grp, M_PCB);
265 }
266 
267 static void
268 in_pcblbgroup_free(struct inpcblbgroup *grp)
269 {
270 
271 	CK_LIST_REMOVE(grp, il_list);
272 	epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
273 	    in_pcblbgroup_free_deferred);
274 }
275 
276 static struct inpcblbgroup *
277 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
278     struct inpcblbgroup *old_grp, int size)
279 {
280 	struct inpcblbgroup *grp;
281 	int i;
282 
283 	grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
284 	    old_grp->il_lport, &old_grp->il_dependladdr, size);
285 	if (grp == NULL)
286 		return (NULL);
287 
288 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
289 	    ("invalid new local group size %d and old local group count %d",
290 	     grp->il_inpsiz, old_grp->il_inpcnt));
291 
292 	for (i = 0; i < old_grp->il_inpcnt; ++i)
293 		grp->il_inp[i] = old_grp->il_inp[i];
294 	grp->il_inpcnt = old_grp->il_inpcnt;
295 	in_pcblbgroup_free(old_grp);
296 	return (grp);
297 }
298 
299 /*
300  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
301  * and shrink group if possible.
302  */
303 static void
304 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
305     int i)
306 {
307 	struct inpcblbgroup *grp, *new_grp;
308 
309 	grp = *grpp;
310 	for (; i + 1 < grp->il_inpcnt; ++i)
311 		grp->il_inp[i] = grp->il_inp[i + 1];
312 	grp->il_inpcnt--;
313 
314 	if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
315 	    grp->il_inpcnt <= grp->il_inpsiz / 4) {
316 		/* Shrink this group. */
317 		new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
318 		if (new_grp != NULL)
319 			*grpp = new_grp;
320 	}
321 }
322 
323 /*
324  * Add PCB to load balance group for SO_REUSEPORT_LB option.
325  */
326 static int
327 in_pcbinslbgrouphash(struct inpcb *inp)
328 {
329 	const static struct timeval interval = { 60, 0 };
330 	static struct timeval lastprint;
331 	struct inpcbinfo *pcbinfo;
332 	struct inpcblbgrouphead *hdr;
333 	struct inpcblbgroup *grp;
334 	uint32_t idx;
335 
336 	pcbinfo = inp->inp_pcbinfo;
337 
338 	INP_WLOCK_ASSERT(inp);
339 	INP_HASH_WLOCK_ASSERT(pcbinfo);
340 
341 	/*
342 	 * Don't allow jailed socket to join local group.
343 	 */
344 	if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
345 		return (0);
346 
347 #ifdef INET6
348 	/*
349 	 * Don't allow IPv4 mapped INET6 wild socket.
350 	 */
351 	if ((inp->inp_vflag & INP_IPV4) &&
352 	    inp->inp_laddr.s_addr == INADDR_ANY &&
353 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
354 		return (0);
355 	}
356 #endif
357 
358 	idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
359 	hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
360 	CK_LIST_FOREACH(grp, hdr, il_list) {
361 		if (grp->il_vflag == inp->inp_vflag &&
362 		    grp->il_lport == inp->inp_lport &&
363 		    memcmp(&grp->il_dependladdr,
364 		    &inp->inp_inc.inc_ie.ie_dependladdr,
365 		    sizeof(grp->il_dependladdr)) == 0)
366 			break;
367 	}
368 	if (grp == NULL) {
369 		/* Create new load balance group. */
370 		grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
371 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
372 		    INPCBLBGROUP_SIZMIN);
373 		if (grp == NULL)
374 			return (ENOBUFS);
375 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
376 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
377 			if (ratecheck(&lastprint, &interval))
378 				printf("lb group port %d, limit reached\n",
379 				    ntohs(grp->il_lport));
380 			return (0);
381 		}
382 
383 		/* Expand this local group. */
384 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
385 		if (grp == NULL)
386 			return (ENOBUFS);
387 	}
388 
389 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
390 	    ("invalid local group size %d and count %d", grp->il_inpsiz,
391 	    grp->il_inpcnt));
392 
393 	grp->il_inp[grp->il_inpcnt] = inp;
394 	grp->il_inpcnt++;
395 	return (0);
396 }
397 
398 /*
399  * Remove PCB from load balance group.
400  */
401 static void
402 in_pcbremlbgrouphash(struct inpcb *inp)
403 {
404 	struct inpcbinfo *pcbinfo;
405 	struct inpcblbgrouphead *hdr;
406 	struct inpcblbgroup *grp;
407 	int i;
408 
409 	pcbinfo = inp->inp_pcbinfo;
410 
411 	INP_WLOCK_ASSERT(inp);
412 	INP_HASH_WLOCK_ASSERT(pcbinfo);
413 
414 	hdr = &pcbinfo->ipi_lbgrouphashbase[
415 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
416 	CK_LIST_FOREACH(grp, hdr, il_list) {
417 		for (i = 0; i < grp->il_inpcnt; ++i) {
418 			if (grp->il_inp[i] != inp)
419 				continue;
420 
421 			if (grp->il_inpcnt == 1) {
422 				/* We are the last, free this local group. */
423 				in_pcblbgroup_free(grp);
424 			} else {
425 				/* Pull up inpcbs, shrink group if possible. */
426 				in_pcblbgroup_reorder(hdr, &grp, i);
427 			}
428 			return;
429 		}
430 	}
431 }
432 
433 /*
434  * Different protocols initialize their inpcbs differently - giving
435  * different name to the lock.  But they all are disposed the same.
436  */
437 static void
438 inpcb_fini(void *mem, int size)
439 {
440 	struct inpcb *inp = mem;
441 
442 	INP_LOCK_DESTROY(inp);
443 }
444 
445 /*
446  * Initialize an inpcbinfo -- we should be able to reduce the number of
447  * arguments in time.
448  */
449 void
450 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
451     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
452     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
453 {
454 
455 	porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
456 
457 	INP_INFO_LOCK_INIT(pcbinfo, name);
458 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
459 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
460 #ifdef VIMAGE
461 	pcbinfo->ipi_vnet = curvnet;
462 #endif
463 	pcbinfo->ipi_listhead = listhead;
464 	CK_LIST_INIT(pcbinfo->ipi_listhead);
465 	pcbinfo->ipi_count = 0;
466 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
467 	    &pcbinfo->ipi_hashmask);
468 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
469 	    &pcbinfo->ipi_porthashmask);
470 	pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
471 	    &pcbinfo->ipi_lbgrouphashmask);
472 #ifdef PCBGROUP
473 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
474 #endif
475 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
476 	    NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
477 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
478 	uma_zone_set_warning(pcbinfo->ipi_zone,
479 	    "kern.ipc.maxsockets limit reached");
480 }
481 
482 /*
483  * Destroy an inpcbinfo.
484  */
485 void
486 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
487 {
488 
489 	KASSERT(pcbinfo->ipi_count == 0,
490 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
491 
492 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
493 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
494 	    pcbinfo->ipi_porthashmask);
495 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
496 	    pcbinfo->ipi_lbgrouphashmask);
497 #ifdef PCBGROUP
498 	in_pcbgroup_destroy(pcbinfo);
499 #endif
500 	uma_zdestroy(pcbinfo->ipi_zone);
501 	INP_LIST_LOCK_DESTROY(pcbinfo);
502 	INP_HASH_LOCK_DESTROY(pcbinfo);
503 	INP_INFO_LOCK_DESTROY(pcbinfo);
504 }
505 
506 /*
507  * Allocate a PCB and associate it with the socket.
508  * On success return with the PCB locked.
509  */
510 int
511 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
512 {
513 	struct inpcb *inp;
514 	int error;
515 
516 #ifdef INVARIANTS
517 	if (pcbinfo == &V_tcbinfo) {
518 		NET_EPOCH_ASSERT();
519 	} else {
520 		INP_INFO_WLOCK_ASSERT(pcbinfo);
521 	}
522 #endif
523 
524 	error = 0;
525 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
526 	if (inp == NULL)
527 		return (ENOBUFS);
528 	bzero(&inp->inp_start_zero, inp_zero_size);
529 #ifdef NUMA
530 	inp->inp_numa_domain = M_NODOM;
531 #endif
532 	inp->inp_pcbinfo = pcbinfo;
533 	inp->inp_socket = so;
534 	inp->inp_cred = crhold(so->so_cred);
535 	inp->inp_inc.inc_fibnum = so->so_fibnum;
536 #ifdef MAC
537 	error = mac_inpcb_init(inp, M_NOWAIT);
538 	if (error != 0)
539 		goto out;
540 	mac_inpcb_create(so, inp);
541 #endif
542 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
543 	error = ipsec_init_pcbpolicy(inp);
544 	if (error != 0) {
545 #ifdef MAC
546 		mac_inpcb_destroy(inp);
547 #endif
548 		goto out;
549 	}
550 #endif /*IPSEC*/
551 #ifdef INET6
552 	if (INP_SOCKAF(so) == AF_INET6) {
553 		inp->inp_vflag |= INP_IPV6PROTO;
554 		if (V_ip6_v6only)
555 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
556 	}
557 #endif
558 	INP_WLOCK(inp);
559 	INP_LIST_WLOCK(pcbinfo);
560 	CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
561 	pcbinfo->ipi_count++;
562 	so->so_pcb = (caddr_t)inp;
563 #ifdef INET6
564 	if (V_ip6_auto_flowlabel)
565 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
566 #endif
567 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
568 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
569 
570 	/*
571 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
572 	 * to be cleaned up.
573 	 */
574 	inp->inp_route.ro_flags = RT_LLE_CACHE;
575 	INP_LIST_WUNLOCK(pcbinfo);
576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
577 out:
578 	if (error != 0) {
579 		crfree(inp->inp_cred);
580 		uma_zfree(pcbinfo->ipi_zone, inp);
581 	}
582 #endif
583 	return (error);
584 }
585 
586 #ifdef INET
587 int
588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
589 {
590 	int anonport, error;
591 
592 	INP_WLOCK_ASSERT(inp);
593 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
594 
595 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
596 		return (EINVAL);
597 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
598 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
599 	    &inp->inp_lport, cred);
600 	if (error)
601 		return (error);
602 	if (in_pcbinshash(inp) != 0) {
603 		inp->inp_laddr.s_addr = INADDR_ANY;
604 		inp->inp_lport = 0;
605 		return (EAGAIN);
606 	}
607 	if (anonport)
608 		inp->inp_flags |= INP_ANONPORT;
609 	return (0);
610 }
611 #endif
612 
613 /*
614  * Select a local port (number) to use.
615  */
616 #if defined(INET) || defined(INET6)
617 int
618 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
619     struct ucred *cred, int lookupflags)
620 {
621 	struct inpcbinfo *pcbinfo;
622 	struct inpcb *tmpinp;
623 	unsigned short *lastport;
624 	int count, dorandom, error;
625 	u_short aux, first, last, lport;
626 #ifdef INET
627 	struct in_addr laddr;
628 #endif
629 
630 	pcbinfo = inp->inp_pcbinfo;
631 
632 	/*
633 	 * Because no actual state changes occur here, a global write lock on
634 	 * the pcbinfo isn't required.
635 	 */
636 	INP_LOCK_ASSERT(inp);
637 	INP_HASH_LOCK_ASSERT(pcbinfo);
638 
639 	if (inp->inp_flags & INP_HIGHPORT) {
640 		first = V_ipport_hifirstauto;	/* sysctl */
641 		last  = V_ipport_hilastauto;
642 		lastport = &pcbinfo->ipi_lasthi;
643 	} else if (inp->inp_flags & INP_LOWPORT) {
644 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
645 		if (error)
646 			return (error);
647 		first = V_ipport_lowfirstauto;	/* 1023 */
648 		last  = V_ipport_lowlastauto;	/* 600 */
649 		lastport = &pcbinfo->ipi_lastlow;
650 	} else {
651 		first = V_ipport_firstauto;	/* sysctl */
652 		last  = V_ipport_lastauto;
653 		lastport = &pcbinfo->ipi_lastport;
654 	}
655 	/*
656 	 * For UDP(-Lite), use random port allocation as long as the user
657 	 * allows it.  For TCP (and as of yet unknown) connections,
658 	 * use random port allocation only if the user allows it AND
659 	 * ipport_tick() allows it.
660 	 */
661 	if (V_ipport_randomized &&
662 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
663 		pcbinfo == &V_ulitecbinfo))
664 		dorandom = 1;
665 	else
666 		dorandom = 0;
667 	/*
668 	 * It makes no sense to do random port allocation if
669 	 * we have the only port available.
670 	 */
671 	if (first == last)
672 		dorandom = 0;
673 	/* Make sure to not include UDP(-Lite) packets in the count. */
674 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
675 		V_ipport_tcpallocs++;
676 	/*
677 	 * Instead of having two loops further down counting up or down
678 	 * make sure that first is always <= last and go with only one
679 	 * code path implementing all logic.
680 	 */
681 	if (first > last) {
682 		aux = first;
683 		first = last;
684 		last = aux;
685 	}
686 
687 #ifdef INET
688 	/* Make the compiler happy. */
689 	laddr.s_addr = 0;
690 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
691 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
692 		    __func__, inp));
693 		laddr = *laddrp;
694 	}
695 #endif
696 	tmpinp = NULL;	/* Make compiler happy. */
697 	lport = *lportp;
698 
699 	if (dorandom)
700 		*lastport = first + (arc4random() % (last - first));
701 
702 	count = last - first;
703 
704 	do {
705 		if (count-- < 0)	/* completely used? */
706 			return (EADDRNOTAVAIL);
707 		++*lastport;
708 		if (*lastport < first || *lastport > last)
709 			*lastport = first;
710 		lport = htons(*lastport);
711 
712 #ifdef INET6
713 		if ((inp->inp_vflag & INP_IPV6) != 0)
714 			tmpinp = in6_pcblookup_local(pcbinfo,
715 			    &inp->in6p_laddr, lport, lookupflags, cred);
716 #endif
717 #if defined(INET) && defined(INET6)
718 		else
719 #endif
720 #ifdef INET
721 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
722 			    lport, lookupflags, cred);
723 #endif
724 	} while (tmpinp != NULL);
725 
726 #ifdef INET
727 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
728 		laddrp->s_addr = laddr.s_addr;
729 #endif
730 	*lportp = lport;
731 
732 	return (0);
733 }
734 
735 /*
736  * Return cached socket options.
737  */
738 int
739 inp_so_options(const struct inpcb *inp)
740 {
741 	int so_options;
742 
743 	so_options = 0;
744 
745 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
746 		so_options |= SO_REUSEPORT_LB;
747 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
748 		so_options |= SO_REUSEPORT;
749 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
750 		so_options |= SO_REUSEADDR;
751 	return (so_options);
752 }
753 #endif /* INET || INET6 */
754 
755 /*
756  * Check if a new BINDMULTI socket is allowed to be created.
757  *
758  * ni points to the new inp.
759  * oi points to the exisitng inp.
760  *
761  * This checks whether the existing inp also has BINDMULTI and
762  * whether the credentials match.
763  */
764 int
765 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
766 {
767 	/* Check permissions match */
768 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
769 	    (ni->inp_cred->cr_uid !=
770 	    oi->inp_cred->cr_uid))
771 		return (0);
772 
773 	/* Check the existing inp has BINDMULTI set */
774 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
775 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
776 		return (0);
777 
778 	/*
779 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
780 	 * it is and it matches the checks.
781 	 */
782 	return (1);
783 }
784 
785 #ifdef INET
786 /*
787  * Set up a bind operation on a PCB, performing port allocation
788  * as required, but do not actually modify the PCB. Callers can
789  * either complete the bind by setting inp_laddr/inp_lport and
790  * calling in_pcbinshash(), or they can just use the resulting
791  * port and address to authorise the sending of a once-off packet.
792  *
793  * On error, the values of *laddrp and *lportp are not changed.
794  */
795 int
796 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
797     u_short *lportp, struct ucred *cred)
798 {
799 	struct socket *so = inp->inp_socket;
800 	struct sockaddr_in *sin;
801 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
802 	struct in_addr laddr;
803 	u_short lport = 0;
804 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
805 	int error;
806 
807 	/*
808 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
809 	 * so that we don't have to add to the (already messy) code below.
810 	 */
811 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
812 
813 	/*
814 	 * No state changes, so read locks are sufficient here.
815 	 */
816 	INP_LOCK_ASSERT(inp);
817 	INP_HASH_LOCK_ASSERT(pcbinfo);
818 
819 	if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
820 		return (EADDRNOTAVAIL);
821 	laddr.s_addr = *laddrp;
822 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
823 		return (EINVAL);
824 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
825 		lookupflags = INPLOOKUP_WILDCARD;
826 	if (nam == NULL) {
827 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
828 			return (error);
829 	} else {
830 		sin = (struct sockaddr_in *)nam;
831 		if (nam->sa_len != sizeof (*sin))
832 			return (EINVAL);
833 #ifdef notdef
834 		/*
835 		 * We should check the family, but old programs
836 		 * incorrectly fail to initialize it.
837 		 */
838 		if (sin->sin_family != AF_INET)
839 			return (EAFNOSUPPORT);
840 #endif
841 		error = prison_local_ip4(cred, &sin->sin_addr);
842 		if (error)
843 			return (error);
844 		if (sin->sin_port != *lportp) {
845 			/* Don't allow the port to change. */
846 			if (*lportp != 0)
847 				return (EINVAL);
848 			lport = sin->sin_port;
849 		}
850 		/* NB: lport is left as 0 if the port isn't being changed. */
851 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
852 			/*
853 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
854 			 * allow complete duplication of binding if
855 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
856 			 * and a multicast address is bound on both
857 			 * new and duplicated sockets.
858 			 */
859 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
860 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
861 			/*
862 			 * XXX: How to deal with SO_REUSEPORT_LB here?
863 			 * Treat same as SO_REUSEPORT for now.
864 			 */
865 			if ((so->so_options &
866 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
867 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
868 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
869 			sin->sin_port = 0;		/* yech... */
870 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
871 			/*
872 			 * Is the address a local IP address?
873 			 * If INP_BINDANY is set, then the socket may be bound
874 			 * to any endpoint address, local or not.
875 			 */
876 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
877 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
878 				return (EADDRNOTAVAIL);
879 		}
880 		laddr = sin->sin_addr;
881 		if (lport) {
882 			struct inpcb *t;
883 			struct tcptw *tw;
884 
885 			/* GROSS */
886 			if (ntohs(lport) <= V_ipport_reservedhigh &&
887 			    ntohs(lport) >= V_ipport_reservedlow &&
888 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
889 				return (EACCES);
890 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
891 			    priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
892 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
893 				    lport, INPLOOKUP_WILDCARD, cred);
894 	/*
895 	 * XXX
896 	 * This entire block sorely needs a rewrite.
897 	 */
898 				if (t &&
899 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
900 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
901 				    (so->so_type != SOCK_STREAM ||
902 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
903 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
904 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
905 				     (t->inp_flags2 & INP_REUSEPORT) ||
906 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
907 				    (inp->inp_cred->cr_uid !=
908 				     t->inp_cred->cr_uid))
909 					return (EADDRINUSE);
910 
911 				/*
912 				 * If the socket is a BINDMULTI socket, then
913 				 * the credentials need to match and the
914 				 * original socket also has to have been bound
915 				 * with BINDMULTI.
916 				 */
917 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
918 					return (EADDRINUSE);
919 			}
920 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
921 			    lport, lookupflags, cred);
922 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
923 				/*
924 				 * XXXRW: If an incpb has had its timewait
925 				 * state recycled, we treat the address as
926 				 * being in use (for now).  This is better
927 				 * than a panic, but not desirable.
928 				 */
929 				tw = intotw(t);
930 				if (tw == NULL ||
931 				    ((reuseport & tw->tw_so_options) == 0 &&
932 					(reuseport_lb &
933 				            tw->tw_so_options) == 0)) {
934 					return (EADDRINUSE);
935 				}
936 			} else if (t &&
937 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
938 				   (reuseport & inp_so_options(t)) == 0 &&
939 				   (reuseport_lb & inp_so_options(t)) == 0) {
940 #ifdef INET6
941 				if (ntohl(sin->sin_addr.s_addr) !=
942 				    INADDR_ANY ||
943 				    ntohl(t->inp_laddr.s_addr) !=
944 				    INADDR_ANY ||
945 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
946 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
947 #endif
948 						return (EADDRINUSE);
949 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
950 					return (EADDRINUSE);
951 			}
952 		}
953 	}
954 	if (*lportp != 0)
955 		lport = *lportp;
956 	if (lport == 0) {
957 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
958 		if (error != 0)
959 			return (error);
960 
961 	}
962 	*laddrp = laddr.s_addr;
963 	*lportp = lport;
964 	return (0);
965 }
966 
967 /*
968  * Connect from a socket to a specified address.
969  * Both address and port must be specified in argument sin.
970  * If don't have a local address for this socket yet,
971  * then pick one.
972  */
973 int
974 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
975     struct ucred *cred, struct mbuf *m, bool rehash)
976 {
977 	u_short lport, fport;
978 	in_addr_t laddr, faddr;
979 	int anonport, error;
980 
981 	INP_WLOCK_ASSERT(inp);
982 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
983 
984 	lport = inp->inp_lport;
985 	laddr = inp->inp_laddr.s_addr;
986 	anonport = (lport == 0);
987 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
988 	    NULL, cred);
989 	if (error)
990 		return (error);
991 
992 	/* Do the initial binding of the local address if required. */
993 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
994 		KASSERT(rehash == true,
995 		    ("Rehashing required for unbound inps"));
996 		inp->inp_lport = lport;
997 		inp->inp_laddr.s_addr = laddr;
998 		if (in_pcbinshash(inp) != 0) {
999 			inp->inp_laddr.s_addr = INADDR_ANY;
1000 			inp->inp_lport = 0;
1001 			return (EAGAIN);
1002 		}
1003 	}
1004 
1005 	/* Commit the remaining changes. */
1006 	inp->inp_lport = lport;
1007 	inp->inp_laddr.s_addr = laddr;
1008 	inp->inp_faddr.s_addr = faddr;
1009 	inp->inp_fport = fport;
1010 	if (rehash) {
1011 		in_pcbrehash_mbuf(inp, m);
1012 	} else {
1013 		in_pcbinshash_mbuf(inp, m);
1014 	}
1015 
1016 	if (anonport)
1017 		inp->inp_flags |= INP_ANONPORT;
1018 	return (0);
1019 }
1020 
1021 int
1022 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1023 {
1024 
1025 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1026 }
1027 
1028 /*
1029  * Do proper source address selection on an unbound socket in case
1030  * of connect. Take jails into account as well.
1031  */
1032 int
1033 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1034     struct ucred *cred)
1035 {
1036 	struct ifaddr *ifa;
1037 	struct sockaddr *sa;
1038 	struct sockaddr_in *sin;
1039 	struct route sro;
1040 	struct epoch_tracker et;
1041 	int error;
1042 
1043 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1044 	/*
1045 	 * Bypass source address selection and use the primary jail IP
1046 	 * if requested.
1047 	 */
1048 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1049 		return (0);
1050 
1051 	error = 0;
1052 	bzero(&sro, sizeof(sro));
1053 
1054 	sin = (struct sockaddr_in *)&sro.ro_dst;
1055 	sin->sin_family = AF_INET;
1056 	sin->sin_len = sizeof(struct sockaddr_in);
1057 	sin->sin_addr.s_addr = faddr->s_addr;
1058 
1059 	/*
1060 	 * If route is known our src addr is taken from the i/f,
1061 	 * else punt.
1062 	 *
1063 	 * Find out route to destination.
1064 	 */
1065 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1066 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1067 
1068 	/*
1069 	 * If we found a route, use the address corresponding to
1070 	 * the outgoing interface.
1071 	 *
1072 	 * Otherwise assume faddr is reachable on a directly connected
1073 	 * network and try to find a corresponding interface to take
1074 	 * the source address from.
1075 	 */
1076 	NET_EPOCH_ENTER(et);
1077 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1078 		struct in_ifaddr *ia;
1079 		struct ifnet *ifp;
1080 
1081 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1082 					inp->inp_socket->so_fibnum));
1083 		if (ia == NULL) {
1084 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1085 						inp->inp_socket->so_fibnum));
1086 
1087 		}
1088 		if (ia == NULL) {
1089 			error = ENETUNREACH;
1090 			goto done;
1091 		}
1092 
1093 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1094 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1095 			goto done;
1096 		}
1097 
1098 		ifp = ia->ia_ifp;
1099 		ia = NULL;
1100 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1101 
1102 			sa = ifa->ifa_addr;
1103 			if (sa->sa_family != AF_INET)
1104 				continue;
1105 			sin = (struct sockaddr_in *)sa;
1106 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1107 				ia = (struct in_ifaddr *)ifa;
1108 				break;
1109 			}
1110 		}
1111 		if (ia != NULL) {
1112 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1113 			goto done;
1114 		}
1115 
1116 		/* 3. As a last resort return the 'default' jail address. */
1117 		error = prison_get_ip4(cred, laddr);
1118 		goto done;
1119 	}
1120 
1121 	/*
1122 	 * If the outgoing interface on the route found is not
1123 	 * a loopback interface, use the address from that interface.
1124 	 * In case of jails do those three steps:
1125 	 * 1. check if the interface address belongs to the jail. If so use it.
1126 	 * 2. check if we have any address on the outgoing interface
1127 	 *    belonging to this jail. If so use it.
1128 	 * 3. as a last resort return the 'default' jail address.
1129 	 */
1130 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1131 		struct in_ifaddr *ia;
1132 		struct ifnet *ifp;
1133 
1134 		/* If not jailed, use the default returned. */
1135 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1136 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1137 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1138 			goto done;
1139 		}
1140 
1141 		/* Jailed. */
1142 		/* 1. Check if the iface address belongs to the jail. */
1143 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1144 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1145 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1146 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1147 			goto done;
1148 		}
1149 
1150 		/*
1151 		 * 2. Check if we have any address on the outgoing interface
1152 		 *    belonging to this jail.
1153 		 */
1154 		ia = NULL;
1155 		ifp = sro.ro_rt->rt_ifp;
1156 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1157 			sa = ifa->ifa_addr;
1158 			if (sa->sa_family != AF_INET)
1159 				continue;
1160 			sin = (struct sockaddr_in *)sa;
1161 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1162 				ia = (struct in_ifaddr *)ifa;
1163 				break;
1164 			}
1165 		}
1166 		if (ia != NULL) {
1167 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1168 			goto done;
1169 		}
1170 
1171 		/* 3. As a last resort return the 'default' jail address. */
1172 		error = prison_get_ip4(cred, laddr);
1173 		goto done;
1174 	}
1175 
1176 	/*
1177 	 * The outgoing interface is marked with 'loopback net', so a route
1178 	 * to ourselves is here.
1179 	 * Try to find the interface of the destination address and then
1180 	 * take the address from there. That interface is not necessarily
1181 	 * a loopback interface.
1182 	 * In case of jails, check that it is an address of the jail
1183 	 * and if we cannot find, fall back to the 'default' jail address.
1184 	 */
1185 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1186 		struct sockaddr_in sain;
1187 		struct in_ifaddr *ia;
1188 
1189 		bzero(&sain, sizeof(struct sockaddr_in));
1190 		sain.sin_family = AF_INET;
1191 		sain.sin_len = sizeof(struct sockaddr_in);
1192 		sain.sin_addr.s_addr = faddr->s_addr;
1193 
1194 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1195 					inp->inp_socket->so_fibnum));
1196 		if (ia == NULL)
1197 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1198 						inp->inp_socket->so_fibnum));
1199 		if (ia == NULL)
1200 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1201 
1202 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1203 			if (ia == NULL) {
1204 				error = ENETUNREACH;
1205 				goto done;
1206 			}
1207 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1208 			goto done;
1209 		}
1210 
1211 		/* Jailed. */
1212 		if (ia != NULL) {
1213 			struct ifnet *ifp;
1214 
1215 			ifp = ia->ia_ifp;
1216 			ia = NULL;
1217 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1218 				sa = ifa->ifa_addr;
1219 				if (sa->sa_family != AF_INET)
1220 					continue;
1221 				sin = (struct sockaddr_in *)sa;
1222 				if (prison_check_ip4(cred,
1223 				    &sin->sin_addr) == 0) {
1224 					ia = (struct in_ifaddr *)ifa;
1225 					break;
1226 				}
1227 			}
1228 			if (ia != NULL) {
1229 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1230 				goto done;
1231 			}
1232 		}
1233 
1234 		/* 3. As a last resort return the 'default' jail address. */
1235 		error = prison_get_ip4(cred, laddr);
1236 		goto done;
1237 	}
1238 
1239 done:
1240 	NET_EPOCH_EXIT(et);
1241 	if (sro.ro_rt != NULL)
1242 		RTFREE(sro.ro_rt);
1243 	return (error);
1244 }
1245 
1246 /*
1247  * Set up for a connect from a socket to the specified address.
1248  * On entry, *laddrp and *lportp should contain the current local
1249  * address and port for the PCB; these are updated to the values
1250  * that should be placed in inp_laddr and inp_lport to complete
1251  * the connect.
1252  *
1253  * On success, *faddrp and *fportp will be set to the remote address
1254  * and port. These are not updated in the error case.
1255  *
1256  * If the operation fails because the connection already exists,
1257  * *oinpp will be set to the PCB of that connection so that the
1258  * caller can decide to override it. In all other cases, *oinpp
1259  * is set to NULL.
1260  */
1261 int
1262 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1263     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1264     struct inpcb **oinpp, struct ucred *cred)
1265 {
1266 	struct rm_priotracker in_ifa_tracker;
1267 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1268 	struct in_ifaddr *ia;
1269 	struct inpcb *oinp;
1270 	struct in_addr laddr, faddr;
1271 	u_short lport, fport;
1272 	int error;
1273 
1274 	/*
1275 	 * Because a global state change doesn't actually occur here, a read
1276 	 * lock is sufficient.
1277 	 */
1278 	INP_LOCK_ASSERT(inp);
1279 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1280 
1281 	if (oinpp != NULL)
1282 		*oinpp = NULL;
1283 	if (nam->sa_len != sizeof (*sin))
1284 		return (EINVAL);
1285 	if (sin->sin_family != AF_INET)
1286 		return (EAFNOSUPPORT);
1287 	if (sin->sin_port == 0)
1288 		return (EADDRNOTAVAIL);
1289 	laddr.s_addr = *laddrp;
1290 	lport = *lportp;
1291 	faddr = sin->sin_addr;
1292 	fport = sin->sin_port;
1293 
1294 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1295 		/*
1296 		 * If the destination address is INADDR_ANY,
1297 		 * use the primary local address.
1298 		 * If the supplied address is INADDR_BROADCAST,
1299 		 * and the primary interface supports broadcast,
1300 		 * choose the broadcast address for that interface.
1301 		 */
1302 		if (faddr.s_addr == INADDR_ANY) {
1303 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1304 			faddr =
1305 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1306 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1307 			if (cred != NULL &&
1308 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1309 				return (error);
1310 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1311 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1312 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1313 			    IFF_BROADCAST)
1314 				faddr = satosin(&CK_STAILQ_FIRST(
1315 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1316 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1317 		}
1318 	}
1319 	if (laddr.s_addr == INADDR_ANY) {
1320 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1321 		/*
1322 		 * If the destination address is multicast and an outgoing
1323 		 * interface has been set as a multicast option, prefer the
1324 		 * address of that interface as our source address.
1325 		 */
1326 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1327 		    inp->inp_moptions != NULL) {
1328 			struct ip_moptions *imo;
1329 			struct ifnet *ifp;
1330 
1331 			imo = inp->inp_moptions;
1332 			if (imo->imo_multicast_ifp != NULL) {
1333 				ifp = imo->imo_multicast_ifp;
1334 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1335 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1336 					if ((ia->ia_ifp == ifp) &&
1337 					    (cred == NULL ||
1338 					    prison_check_ip4(cred,
1339 					    &ia->ia_addr.sin_addr) == 0))
1340 						break;
1341 				}
1342 				if (ia == NULL)
1343 					error = EADDRNOTAVAIL;
1344 				else {
1345 					laddr = ia->ia_addr.sin_addr;
1346 					error = 0;
1347 				}
1348 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1349 			}
1350 		}
1351 		if (error)
1352 			return (error);
1353 	}
1354 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1355 	    laddr, lport, 0, NULL);
1356 	if (oinp != NULL) {
1357 		if (oinpp != NULL)
1358 			*oinpp = oinp;
1359 		return (EADDRINUSE);
1360 	}
1361 	if (lport == 0) {
1362 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1363 		    cred);
1364 		if (error)
1365 			return (error);
1366 	}
1367 	*laddrp = laddr.s_addr;
1368 	*lportp = lport;
1369 	*faddrp = faddr.s_addr;
1370 	*fportp = fport;
1371 	return (0);
1372 }
1373 
1374 void
1375 in_pcbdisconnect(struct inpcb *inp)
1376 {
1377 
1378 	INP_WLOCK_ASSERT(inp);
1379 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1380 
1381 	inp->inp_faddr.s_addr = INADDR_ANY;
1382 	inp->inp_fport = 0;
1383 	in_pcbrehash(inp);
1384 }
1385 #endif /* INET */
1386 
1387 /*
1388  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1389  * For most protocols, this will be invoked immediately prior to calling
1390  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1391  * socket, in which case in_pcbfree() is deferred.
1392  */
1393 void
1394 in_pcbdetach(struct inpcb *inp)
1395 {
1396 
1397 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1398 
1399 #ifdef RATELIMIT
1400 	if (inp->inp_snd_tag != NULL)
1401 		in_pcbdetach_txrtlmt(inp);
1402 #endif
1403 	inp->inp_socket->so_pcb = NULL;
1404 	inp->inp_socket = NULL;
1405 }
1406 
1407 /*
1408  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1409  * stability of an inpcb pointer despite the inpcb lock being released.  This
1410  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1411  * but where the inpcb lock may already held, or when acquiring a reference
1412  * via a pcbgroup.
1413  *
1414  * in_pcbref() should be used only to provide brief memory stability, and
1415  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1416  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1417  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1418  * lock and rele are the *only* safe operations that may be performed on the
1419  * inpcb.
1420  *
1421  * While the inpcb will not be freed, releasing the inpcb lock means that the
1422  * connection's state may change, so the caller should be careful to
1423  * revalidate any cached state on reacquiring the lock.  Drop the reference
1424  * using in_pcbrele().
1425  */
1426 void
1427 in_pcbref(struct inpcb *inp)
1428 {
1429 
1430 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1431 
1432 	refcount_acquire(&inp->inp_refcount);
1433 }
1434 
1435 /*
1436  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1437  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1438  * return a flag indicating whether or not the inpcb remains valid.  If it is
1439  * valid, we return with the inpcb lock held.
1440  *
1441  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1442  * reference on an inpcb.  Historically more work was done here (actually, in
1443  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1444  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1445  * about memory stability (and continued use of the write lock).
1446  */
1447 int
1448 in_pcbrele_rlocked(struct inpcb *inp)
1449 {
1450 	struct inpcbinfo *pcbinfo;
1451 
1452 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1453 
1454 	INP_RLOCK_ASSERT(inp);
1455 
1456 	if (refcount_release(&inp->inp_refcount) == 0) {
1457 		/*
1458 		 * If the inpcb has been freed, let the caller know, even if
1459 		 * this isn't the last reference.
1460 		 */
1461 		if (inp->inp_flags2 & INP_FREED) {
1462 			INP_RUNLOCK(inp);
1463 			return (1);
1464 		}
1465 		return (0);
1466 	}
1467 
1468 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1469 #ifdef TCPHPTS
1470 	if (inp->inp_in_hpts || inp->inp_in_input) {
1471 		struct tcp_hpts_entry *hpts;
1472 		/*
1473 		 * We should not be on the hpts at
1474 		 * this point in any form. we must
1475 		 * get the lock to be sure.
1476 		 */
1477 		hpts = tcp_hpts_lock(inp);
1478 		if (inp->inp_in_hpts)
1479 			panic("Hpts:%p inp:%p at free still on hpts",
1480 			      hpts, inp);
1481 		mtx_unlock(&hpts->p_mtx);
1482 		hpts = tcp_input_lock(inp);
1483 		if (inp->inp_in_input)
1484 			panic("Hpts:%p inp:%p at free still on input hpts",
1485 			      hpts, inp);
1486 		mtx_unlock(&hpts->p_mtx);
1487 	}
1488 #endif
1489 	INP_RUNLOCK(inp);
1490 	pcbinfo = inp->inp_pcbinfo;
1491 	uma_zfree(pcbinfo->ipi_zone, inp);
1492 	return (1);
1493 }
1494 
1495 int
1496 in_pcbrele_wlocked(struct inpcb *inp)
1497 {
1498 	struct inpcbinfo *pcbinfo;
1499 
1500 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1501 
1502 	INP_WLOCK_ASSERT(inp);
1503 
1504 	if (refcount_release(&inp->inp_refcount) == 0) {
1505 		/*
1506 		 * If the inpcb has been freed, let the caller know, even if
1507 		 * this isn't the last reference.
1508 		 */
1509 		if (inp->inp_flags2 & INP_FREED) {
1510 			INP_WUNLOCK(inp);
1511 			return (1);
1512 		}
1513 		return (0);
1514 	}
1515 
1516 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1517 #ifdef TCPHPTS
1518 	if (inp->inp_in_hpts || inp->inp_in_input) {
1519 		struct tcp_hpts_entry *hpts;
1520 		/*
1521 		 * We should not be on the hpts at
1522 		 * this point in any form. we must
1523 		 * get the lock to be sure.
1524 		 */
1525 		hpts = tcp_hpts_lock(inp);
1526 		if (inp->inp_in_hpts)
1527 			panic("Hpts:%p inp:%p at free still on hpts",
1528 			      hpts, inp);
1529 		mtx_unlock(&hpts->p_mtx);
1530 		hpts = tcp_input_lock(inp);
1531 		if (inp->inp_in_input)
1532 			panic("Hpts:%p inp:%p at free still on input hpts",
1533 			      hpts, inp);
1534 		mtx_unlock(&hpts->p_mtx);
1535 	}
1536 #endif
1537 	INP_WUNLOCK(inp);
1538 	pcbinfo = inp->inp_pcbinfo;
1539 	uma_zfree(pcbinfo->ipi_zone, inp);
1540 	return (1);
1541 }
1542 
1543 /*
1544  * Temporary wrapper.
1545  */
1546 int
1547 in_pcbrele(struct inpcb *inp)
1548 {
1549 
1550 	return (in_pcbrele_wlocked(inp));
1551 }
1552 
1553 void
1554 in_pcblist_rele_rlocked(epoch_context_t ctx)
1555 {
1556 	struct in_pcblist *il;
1557 	struct inpcb *inp;
1558 	struct inpcbinfo *pcbinfo;
1559 	int i, n;
1560 
1561 	il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1562 	pcbinfo = il->il_pcbinfo;
1563 	n = il->il_count;
1564 	INP_INFO_WLOCK(pcbinfo);
1565 	for (i = 0; i < n; i++) {
1566 		inp = il->il_inp_list[i];
1567 		INP_RLOCK(inp);
1568 		if (!in_pcbrele_rlocked(inp))
1569 			INP_RUNLOCK(inp);
1570 	}
1571 	INP_INFO_WUNLOCK(pcbinfo);
1572 	free(il, M_TEMP);
1573 }
1574 
1575 static void
1576 inpcbport_free(epoch_context_t ctx)
1577 {
1578 	struct inpcbport *phd;
1579 
1580 	phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1581 	free(phd, M_PCB);
1582 }
1583 
1584 static void
1585 in_pcbfree_deferred(epoch_context_t ctx)
1586 {
1587 	struct inpcb *inp;
1588 	int released __unused;
1589 
1590 	inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1591 
1592 	INP_WLOCK(inp);
1593 	CURVNET_SET(inp->inp_vnet);
1594 #ifdef INET
1595 	struct ip_moptions *imo = inp->inp_moptions;
1596 	inp->inp_moptions = NULL;
1597 #endif
1598 	/* XXXRW: Do as much as possible here. */
1599 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1600 	if (inp->inp_sp != NULL)
1601 		ipsec_delete_pcbpolicy(inp);
1602 #endif
1603 #ifdef INET6
1604 	struct ip6_moptions *im6o = NULL;
1605 	if (inp->inp_vflag & INP_IPV6PROTO) {
1606 		ip6_freepcbopts(inp->in6p_outputopts);
1607 		im6o = inp->in6p_moptions;
1608 		inp->in6p_moptions = NULL;
1609 	}
1610 #endif
1611 	if (inp->inp_options)
1612 		(void)m_free(inp->inp_options);
1613 	inp->inp_vflag = 0;
1614 	crfree(inp->inp_cred);
1615 #ifdef MAC
1616 	mac_inpcb_destroy(inp);
1617 #endif
1618 	released = in_pcbrele_wlocked(inp);
1619 	MPASS(released);
1620 #ifdef INET6
1621 	ip6_freemoptions(im6o);
1622 #endif
1623 #ifdef INET
1624 	inp_freemoptions(imo);
1625 #endif
1626 	CURVNET_RESTORE();
1627 }
1628 
1629 /*
1630  * Unconditionally schedule an inpcb to be freed by decrementing its
1631  * reference count, which should occur only after the inpcb has been detached
1632  * from its socket.  If another thread holds a temporary reference (acquired
1633  * using in_pcbref()) then the free is deferred until that reference is
1634  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1635  * work, including removal from global lists, is done in this context, where
1636  * the pcbinfo lock is held.
1637  */
1638 void
1639 in_pcbfree(struct inpcb *inp)
1640 {
1641 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1642 
1643 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1644 	KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1645 	    ("%s: called twice for pcb %p", __func__, inp));
1646 	if (inp->inp_flags2 & INP_FREED) {
1647 		INP_WUNLOCK(inp);
1648 		return;
1649 	}
1650 
1651 #ifdef INVARIANTS
1652 	if (pcbinfo == &V_tcbinfo) {
1653 		INP_INFO_LOCK_ASSERT(pcbinfo);
1654 	} else {
1655 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1656 	}
1657 #endif
1658 	INP_WLOCK_ASSERT(inp);
1659 	INP_LIST_WLOCK(pcbinfo);
1660 	in_pcbremlists(inp);
1661 	INP_LIST_WUNLOCK(pcbinfo);
1662 	RO_INVALIDATE_CACHE(&inp->inp_route);
1663 	/* mark as destruction in progress */
1664 	inp->inp_flags2 |= INP_FREED;
1665 	INP_WUNLOCK(inp);
1666 	epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1667 }
1668 
1669 /*
1670  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1671  * port reservation, and preventing it from being returned by inpcb lookups.
1672  *
1673  * It is used by TCP to mark an inpcb as unused and avoid future packet
1674  * delivery or event notification when a socket remains open but TCP has
1675  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1676  * or a RST on the wire, and allows the port binding to be reused while still
1677  * maintaining the invariant that so_pcb always points to a valid inpcb until
1678  * in_pcbdetach().
1679  *
1680  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1681  * in_pcbnotifyall() and in_pcbpurgeif0()?
1682  */
1683 void
1684 in_pcbdrop(struct inpcb *inp)
1685 {
1686 
1687 	INP_WLOCK_ASSERT(inp);
1688 #ifdef INVARIANTS
1689 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1690 		MPASS(inp->inp_refcount > 1);
1691 #endif
1692 
1693 	/*
1694 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1695 	 * the hash lock...?
1696 	 */
1697 	inp->inp_flags |= INP_DROPPED;
1698 	if (inp->inp_flags & INP_INHASHLIST) {
1699 		struct inpcbport *phd = inp->inp_phd;
1700 
1701 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1702 		in_pcbremlbgrouphash(inp);
1703 		CK_LIST_REMOVE(inp, inp_hash);
1704 		CK_LIST_REMOVE(inp, inp_portlist);
1705 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1706 			CK_LIST_REMOVE(phd, phd_hash);
1707 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1708 		}
1709 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1710 		inp->inp_flags &= ~INP_INHASHLIST;
1711 #ifdef PCBGROUP
1712 		in_pcbgroup_remove(inp);
1713 #endif
1714 	}
1715 }
1716 
1717 #ifdef INET
1718 /*
1719  * Common routines to return the socket addresses associated with inpcbs.
1720  */
1721 struct sockaddr *
1722 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1723 {
1724 	struct sockaddr_in *sin;
1725 
1726 	sin = malloc(sizeof *sin, M_SONAME,
1727 		M_WAITOK | M_ZERO);
1728 	sin->sin_family = AF_INET;
1729 	sin->sin_len = sizeof(*sin);
1730 	sin->sin_addr = *addr_p;
1731 	sin->sin_port = port;
1732 
1733 	return (struct sockaddr *)sin;
1734 }
1735 
1736 int
1737 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1738 {
1739 	struct inpcb *inp;
1740 	struct in_addr addr;
1741 	in_port_t port;
1742 
1743 	inp = sotoinpcb(so);
1744 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1745 
1746 	INP_RLOCK(inp);
1747 	port = inp->inp_lport;
1748 	addr = inp->inp_laddr;
1749 	INP_RUNLOCK(inp);
1750 
1751 	*nam = in_sockaddr(port, &addr);
1752 	return 0;
1753 }
1754 
1755 int
1756 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1757 {
1758 	struct inpcb *inp;
1759 	struct in_addr addr;
1760 	in_port_t port;
1761 
1762 	inp = sotoinpcb(so);
1763 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1764 
1765 	INP_RLOCK(inp);
1766 	port = inp->inp_fport;
1767 	addr = inp->inp_faddr;
1768 	INP_RUNLOCK(inp);
1769 
1770 	*nam = in_sockaddr(port, &addr);
1771 	return 0;
1772 }
1773 
1774 void
1775 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1776     struct inpcb *(*notify)(struct inpcb *, int))
1777 {
1778 	struct inpcb *inp, *inp_temp;
1779 
1780 	INP_INFO_WLOCK(pcbinfo);
1781 	CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1782 		INP_WLOCK(inp);
1783 #ifdef INET6
1784 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1785 			INP_WUNLOCK(inp);
1786 			continue;
1787 		}
1788 #endif
1789 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1790 		    inp->inp_socket == NULL) {
1791 			INP_WUNLOCK(inp);
1792 			continue;
1793 		}
1794 		if ((*notify)(inp, errno))
1795 			INP_WUNLOCK(inp);
1796 	}
1797 	INP_INFO_WUNLOCK(pcbinfo);
1798 }
1799 
1800 void
1801 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1802 {
1803 	struct inpcb *inp;
1804 	struct in_multi *inm;
1805 	struct in_mfilter *imf;
1806 	struct ip_moptions *imo;
1807 
1808 	INP_INFO_WLOCK(pcbinfo);
1809 	CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1810 		INP_WLOCK(inp);
1811 		imo = inp->inp_moptions;
1812 		if ((inp->inp_vflag & INP_IPV4) &&
1813 		    imo != NULL) {
1814 			/*
1815 			 * Unselect the outgoing interface if it is being
1816 			 * detached.
1817 			 */
1818 			if (imo->imo_multicast_ifp == ifp)
1819 				imo->imo_multicast_ifp = NULL;
1820 
1821 			/*
1822 			 * Drop multicast group membership if we joined
1823 			 * through the interface being detached.
1824 			 *
1825 			 * XXX This can all be deferred to an epoch_call
1826 			 */
1827 restart:
1828 			IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1829 				if ((inm = imf->imf_inm) == NULL)
1830 					continue;
1831 				if (inm->inm_ifp != ifp)
1832 					continue;
1833 				ip_mfilter_remove(&imo->imo_head, imf);
1834 				IN_MULTI_LOCK_ASSERT();
1835 				in_leavegroup_locked(inm, NULL);
1836 				ip_mfilter_free(imf);
1837 				goto restart;
1838 			}
1839 		}
1840 		INP_WUNLOCK(inp);
1841 	}
1842 	INP_INFO_WUNLOCK(pcbinfo);
1843 }
1844 
1845 /*
1846  * Lookup a PCB based on the local address and port.  Caller must hold the
1847  * hash lock.  No inpcb locks or references are acquired.
1848  */
1849 #define INP_LOOKUP_MAPPED_PCB_COST	3
1850 struct inpcb *
1851 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1852     u_short lport, int lookupflags, struct ucred *cred)
1853 {
1854 	struct inpcb *inp;
1855 #ifdef INET6
1856 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1857 #else
1858 	int matchwild = 3;
1859 #endif
1860 	int wildcard;
1861 
1862 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1863 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1864 
1865 	INP_HASH_LOCK_ASSERT(pcbinfo);
1866 
1867 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1868 		struct inpcbhead *head;
1869 		/*
1870 		 * Look for an unconnected (wildcard foreign addr) PCB that
1871 		 * matches the local address and port we're looking for.
1872 		 */
1873 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1874 		    0, pcbinfo->ipi_hashmask)];
1875 		CK_LIST_FOREACH(inp, head, inp_hash) {
1876 #ifdef INET6
1877 			/* XXX inp locking */
1878 			if ((inp->inp_vflag & INP_IPV4) == 0)
1879 				continue;
1880 #endif
1881 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1882 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1883 			    inp->inp_lport == lport) {
1884 				/*
1885 				 * Found?
1886 				 */
1887 				if (cred == NULL ||
1888 				    prison_equal_ip4(cred->cr_prison,
1889 					inp->inp_cred->cr_prison))
1890 					return (inp);
1891 			}
1892 		}
1893 		/*
1894 		 * Not found.
1895 		 */
1896 		return (NULL);
1897 	} else {
1898 		struct inpcbporthead *porthash;
1899 		struct inpcbport *phd;
1900 		struct inpcb *match = NULL;
1901 		/*
1902 		 * Best fit PCB lookup.
1903 		 *
1904 		 * First see if this local port is in use by looking on the
1905 		 * port hash list.
1906 		 */
1907 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1908 		    pcbinfo->ipi_porthashmask)];
1909 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
1910 			if (phd->phd_port == lport)
1911 				break;
1912 		}
1913 		if (phd != NULL) {
1914 			/*
1915 			 * Port is in use by one or more PCBs. Look for best
1916 			 * fit.
1917 			 */
1918 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1919 				wildcard = 0;
1920 				if (cred != NULL &&
1921 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1922 					cred->cr_prison))
1923 					continue;
1924 #ifdef INET6
1925 				/* XXX inp locking */
1926 				if ((inp->inp_vflag & INP_IPV4) == 0)
1927 					continue;
1928 				/*
1929 				 * We never select the PCB that has
1930 				 * INP_IPV6 flag and is bound to :: if
1931 				 * we have another PCB which is bound
1932 				 * to 0.0.0.0.  If a PCB has the
1933 				 * INP_IPV6 flag, then we set its cost
1934 				 * higher than IPv4 only PCBs.
1935 				 *
1936 				 * Note that the case only happens
1937 				 * when a socket is bound to ::, under
1938 				 * the condition that the use of the
1939 				 * mapped address is allowed.
1940 				 */
1941 				if ((inp->inp_vflag & INP_IPV6) != 0)
1942 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1943 #endif
1944 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1945 					wildcard++;
1946 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1947 					if (laddr.s_addr == INADDR_ANY)
1948 						wildcard++;
1949 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1950 						continue;
1951 				} else {
1952 					if (laddr.s_addr != INADDR_ANY)
1953 						wildcard++;
1954 				}
1955 				if (wildcard < matchwild) {
1956 					match = inp;
1957 					matchwild = wildcard;
1958 					if (matchwild == 0)
1959 						break;
1960 				}
1961 			}
1962 		}
1963 		return (match);
1964 	}
1965 }
1966 #undef INP_LOOKUP_MAPPED_PCB_COST
1967 
1968 static struct inpcb *
1969 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1970     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1971     uint16_t fport, int lookupflags)
1972 {
1973 	struct inpcb *local_wild;
1974 	const struct inpcblbgrouphead *hdr;
1975 	struct inpcblbgroup *grp;
1976 	uint32_t idx;
1977 
1978 	INP_HASH_LOCK_ASSERT(pcbinfo);
1979 
1980 	hdr = &pcbinfo->ipi_lbgrouphashbase[
1981 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1982 
1983 	/*
1984 	 * Order of socket selection:
1985 	 * 1. non-wild.
1986 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1987 	 *
1988 	 * NOTE:
1989 	 * - Load balanced group does not contain jailed sockets
1990 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1991 	 */
1992 	local_wild = NULL;
1993 	CK_LIST_FOREACH(grp, hdr, il_list) {
1994 #ifdef INET6
1995 		if (!(grp->il_vflag & INP_IPV4))
1996 			continue;
1997 #endif
1998 		if (grp->il_lport != lport)
1999 			continue;
2000 
2001 		idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2002 		    grp->il_inpcnt;
2003 		if (grp->il_laddr.s_addr == laddr->s_addr)
2004 			return (grp->il_inp[idx]);
2005 		if (grp->il_laddr.s_addr == INADDR_ANY &&
2006 		    (lookupflags & INPLOOKUP_WILDCARD) != 0)
2007 			local_wild = grp->il_inp[idx];
2008 	}
2009 	return (local_wild);
2010 }
2011 
2012 #ifdef PCBGROUP
2013 /*
2014  * Lookup PCB in hash list, using pcbgroup tables.
2015  */
2016 static struct inpcb *
2017 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2018     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2019     u_int lport_arg, int lookupflags, struct ifnet *ifp)
2020 {
2021 	struct inpcbhead *head;
2022 	struct inpcb *inp, *tmpinp;
2023 	u_short fport = fport_arg, lport = lport_arg;
2024 	bool locked;
2025 
2026 	/*
2027 	 * First look for an exact match.
2028 	 */
2029 	tmpinp = NULL;
2030 	INP_GROUP_LOCK(pcbgroup);
2031 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2032 	    pcbgroup->ipg_hashmask)];
2033 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2034 #ifdef INET6
2035 		/* XXX inp locking */
2036 		if ((inp->inp_vflag & INP_IPV4) == 0)
2037 			continue;
2038 #endif
2039 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2040 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2041 		    inp->inp_fport == fport &&
2042 		    inp->inp_lport == lport) {
2043 			/*
2044 			 * XXX We should be able to directly return
2045 			 * the inp here, without any checks.
2046 			 * Well unless both bound with SO_REUSEPORT?
2047 			 */
2048 			if (prison_flag(inp->inp_cred, PR_IP4))
2049 				goto found;
2050 			if (tmpinp == NULL)
2051 				tmpinp = inp;
2052 		}
2053 	}
2054 	if (tmpinp != NULL) {
2055 		inp = tmpinp;
2056 		goto found;
2057 	}
2058 
2059 #ifdef	RSS
2060 	/*
2061 	 * For incoming connections, we may wish to do a wildcard
2062 	 * match for an RSS-local socket.
2063 	 */
2064 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2065 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2066 #ifdef INET6
2067 		struct inpcb *local_wild_mapped = NULL;
2068 #endif
2069 		struct inpcb *jail_wild = NULL;
2070 		struct inpcbhead *head;
2071 		int injail;
2072 
2073 		/*
2074 		 * Order of socket selection - we always prefer jails.
2075 		 *      1. jailed, non-wild.
2076 		 *      2. jailed, wild.
2077 		 *      3. non-jailed, non-wild.
2078 		 *      4. non-jailed, wild.
2079 		 */
2080 
2081 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2082 		    lport, 0, pcbgroup->ipg_hashmask)];
2083 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2084 #ifdef INET6
2085 			/* XXX inp locking */
2086 			if ((inp->inp_vflag & INP_IPV4) == 0)
2087 				continue;
2088 #endif
2089 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2090 			    inp->inp_lport != lport)
2091 				continue;
2092 
2093 			injail = prison_flag(inp->inp_cred, PR_IP4);
2094 			if (injail) {
2095 				if (prison_check_ip4(inp->inp_cred,
2096 				    &laddr) != 0)
2097 					continue;
2098 			} else {
2099 				if (local_exact != NULL)
2100 					continue;
2101 			}
2102 
2103 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2104 				if (injail)
2105 					goto found;
2106 				else
2107 					local_exact = inp;
2108 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2109 #ifdef INET6
2110 				/* XXX inp locking, NULL check */
2111 				if (inp->inp_vflag & INP_IPV6PROTO)
2112 					local_wild_mapped = inp;
2113 				else
2114 #endif
2115 					if (injail)
2116 						jail_wild = inp;
2117 					else
2118 						local_wild = inp;
2119 			}
2120 		} /* LIST_FOREACH */
2121 
2122 		inp = jail_wild;
2123 		if (inp == NULL)
2124 			inp = local_exact;
2125 		if (inp == NULL)
2126 			inp = local_wild;
2127 #ifdef INET6
2128 		if (inp == NULL)
2129 			inp = local_wild_mapped;
2130 #endif
2131 		if (inp != NULL)
2132 			goto found;
2133 	}
2134 #endif
2135 
2136 	/*
2137 	 * Then look for a wildcard match, if requested.
2138 	 */
2139 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2140 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2141 #ifdef INET6
2142 		struct inpcb *local_wild_mapped = NULL;
2143 #endif
2144 		struct inpcb *jail_wild = NULL;
2145 		struct inpcbhead *head;
2146 		int injail;
2147 
2148 		/*
2149 		 * Order of socket selection - we always prefer jails.
2150 		 *      1. jailed, non-wild.
2151 		 *      2. jailed, wild.
2152 		 *      3. non-jailed, non-wild.
2153 		 *      4. non-jailed, wild.
2154 		 */
2155 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2156 		    0, pcbinfo->ipi_wildmask)];
2157 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2158 #ifdef INET6
2159 			/* XXX inp locking */
2160 			if ((inp->inp_vflag & INP_IPV4) == 0)
2161 				continue;
2162 #endif
2163 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2164 			    inp->inp_lport != lport)
2165 				continue;
2166 
2167 			injail = prison_flag(inp->inp_cred, PR_IP4);
2168 			if (injail) {
2169 				if (prison_check_ip4(inp->inp_cred,
2170 				    &laddr) != 0)
2171 					continue;
2172 			} else {
2173 				if (local_exact != NULL)
2174 					continue;
2175 			}
2176 
2177 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2178 				if (injail)
2179 					goto found;
2180 				else
2181 					local_exact = inp;
2182 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2183 #ifdef INET6
2184 				/* XXX inp locking, NULL check */
2185 				if (inp->inp_vflag & INP_IPV6PROTO)
2186 					local_wild_mapped = inp;
2187 				else
2188 #endif
2189 					if (injail)
2190 						jail_wild = inp;
2191 					else
2192 						local_wild = inp;
2193 			}
2194 		} /* LIST_FOREACH */
2195 		inp = jail_wild;
2196 		if (inp == NULL)
2197 			inp = local_exact;
2198 		if (inp == NULL)
2199 			inp = local_wild;
2200 #ifdef INET6
2201 		if (inp == NULL)
2202 			inp = local_wild_mapped;
2203 #endif
2204 		if (inp != NULL)
2205 			goto found;
2206 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2207 	INP_GROUP_UNLOCK(pcbgroup);
2208 	return (NULL);
2209 
2210 found:
2211 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2212 		locked = INP_TRY_WLOCK(inp);
2213 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2214 		locked = INP_TRY_RLOCK(inp);
2215 	else
2216 		panic("%s: locking bug", __func__);
2217 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2218 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2219 			INP_WUNLOCK(inp);
2220 		else
2221 			INP_RUNLOCK(inp);
2222 		return (NULL);
2223 	} else if (!locked)
2224 		in_pcbref(inp);
2225 	INP_GROUP_UNLOCK(pcbgroup);
2226 	if (!locked) {
2227 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2228 			INP_WLOCK(inp);
2229 			if (in_pcbrele_wlocked(inp))
2230 				return (NULL);
2231 		} else {
2232 			INP_RLOCK(inp);
2233 			if (in_pcbrele_rlocked(inp))
2234 				return (NULL);
2235 		}
2236 	}
2237 #ifdef INVARIANTS
2238 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2239 		INP_WLOCK_ASSERT(inp);
2240 	else
2241 		INP_RLOCK_ASSERT(inp);
2242 #endif
2243 	return (inp);
2244 }
2245 #endif /* PCBGROUP */
2246 
2247 /*
2248  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2249  * that the caller has locked the hash list, and will not perform any further
2250  * locking or reference operations on either the hash list or the connection.
2251  */
2252 static struct inpcb *
2253 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2254     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2255     struct ifnet *ifp)
2256 {
2257 	struct inpcbhead *head;
2258 	struct inpcb *inp, *tmpinp;
2259 	u_short fport = fport_arg, lport = lport_arg;
2260 
2261 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2262 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2263 	INP_HASH_LOCK_ASSERT(pcbinfo);
2264 
2265 	/*
2266 	 * First look for an exact match.
2267 	 */
2268 	tmpinp = NULL;
2269 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2270 	    pcbinfo->ipi_hashmask)];
2271 	CK_LIST_FOREACH(inp, head, inp_hash) {
2272 #ifdef INET6
2273 		/* XXX inp locking */
2274 		if ((inp->inp_vflag & INP_IPV4) == 0)
2275 			continue;
2276 #endif
2277 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2278 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2279 		    inp->inp_fport == fport &&
2280 		    inp->inp_lport == lport) {
2281 			/*
2282 			 * XXX We should be able to directly return
2283 			 * the inp here, without any checks.
2284 			 * Well unless both bound with SO_REUSEPORT?
2285 			 */
2286 			if (prison_flag(inp->inp_cred, PR_IP4))
2287 				return (inp);
2288 			if (tmpinp == NULL)
2289 				tmpinp = inp;
2290 		}
2291 	}
2292 	if (tmpinp != NULL)
2293 		return (tmpinp);
2294 
2295 	/*
2296 	 * Then look in lb group (for wildcard match).
2297 	 */
2298 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2299 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2300 		    fport, lookupflags);
2301 		if (inp != NULL)
2302 			return (inp);
2303 	}
2304 
2305 	/*
2306 	 * Then look for a wildcard match, if requested.
2307 	 */
2308 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2309 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2310 #ifdef INET6
2311 		struct inpcb *local_wild_mapped = NULL;
2312 #endif
2313 		struct inpcb *jail_wild = NULL;
2314 		int injail;
2315 
2316 		/*
2317 		 * Order of socket selection - we always prefer jails.
2318 		 *      1. jailed, non-wild.
2319 		 *      2. jailed, wild.
2320 		 *      3. non-jailed, non-wild.
2321 		 *      4. non-jailed, wild.
2322 		 */
2323 
2324 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2325 		    0, pcbinfo->ipi_hashmask)];
2326 		CK_LIST_FOREACH(inp, head, inp_hash) {
2327 #ifdef INET6
2328 			/* XXX inp locking */
2329 			if ((inp->inp_vflag & INP_IPV4) == 0)
2330 				continue;
2331 #endif
2332 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2333 			    inp->inp_lport != lport)
2334 				continue;
2335 
2336 			injail = prison_flag(inp->inp_cred, PR_IP4);
2337 			if (injail) {
2338 				if (prison_check_ip4(inp->inp_cred,
2339 				    &laddr) != 0)
2340 					continue;
2341 			} else {
2342 				if (local_exact != NULL)
2343 					continue;
2344 			}
2345 
2346 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2347 				if (injail)
2348 					return (inp);
2349 				else
2350 					local_exact = inp;
2351 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2352 #ifdef INET6
2353 				/* XXX inp locking, NULL check */
2354 				if (inp->inp_vflag & INP_IPV6PROTO)
2355 					local_wild_mapped = inp;
2356 				else
2357 #endif
2358 					if (injail)
2359 						jail_wild = inp;
2360 					else
2361 						local_wild = inp;
2362 			}
2363 		} /* LIST_FOREACH */
2364 		if (jail_wild != NULL)
2365 			return (jail_wild);
2366 		if (local_exact != NULL)
2367 			return (local_exact);
2368 		if (local_wild != NULL)
2369 			return (local_wild);
2370 #ifdef INET6
2371 		if (local_wild_mapped != NULL)
2372 			return (local_wild_mapped);
2373 #endif
2374 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2375 
2376 	return (NULL);
2377 }
2378 
2379 /*
2380  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2381  * hash list lock, and will return the inpcb locked (i.e., requires
2382  * INPLOOKUP_LOCKPCB).
2383  */
2384 static struct inpcb *
2385 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2386     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2387     struct ifnet *ifp)
2388 {
2389 	struct inpcb *inp;
2390 
2391 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2392 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2393 	if (inp != NULL) {
2394 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2395 			INP_WLOCK(inp);
2396 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2397 				INP_WUNLOCK(inp);
2398 				inp = NULL;
2399 			}
2400 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2401 			INP_RLOCK(inp);
2402 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2403 				INP_RUNLOCK(inp);
2404 				inp = NULL;
2405 			}
2406 		} else
2407 			panic("%s: locking bug", __func__);
2408 #ifdef INVARIANTS
2409 		if (inp != NULL) {
2410 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2411 				INP_WLOCK_ASSERT(inp);
2412 			else
2413 				INP_RLOCK_ASSERT(inp);
2414 		}
2415 #endif
2416 	}
2417 
2418 	return (inp);
2419 }
2420 
2421 /*
2422  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2423  * from which a pre-calculated hash value may be extracted.
2424  *
2425  * Possibly more of this logic should be in in_pcbgroup.c.
2426  */
2427 struct inpcb *
2428 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2429     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2430 {
2431 #if defined(PCBGROUP) && !defined(RSS)
2432 	struct inpcbgroup *pcbgroup;
2433 #endif
2434 
2435 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2436 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2437 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2438 	    ("%s: LOCKPCB not set", __func__));
2439 
2440 	/*
2441 	 * When not using RSS, use connection groups in preference to the
2442 	 * reservation table when looking up 4-tuples.  When using RSS, just
2443 	 * use the reservation table, due to the cost of the Toeplitz hash
2444 	 * in software.
2445 	 *
2446 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2447 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2448 	 * in software.
2449 	 */
2450 #if defined(PCBGROUP) && !defined(RSS)
2451 	if (in_pcbgroup_enabled(pcbinfo)) {
2452 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2453 		    fport);
2454 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2455 		    laddr, lport, lookupflags, ifp));
2456 	}
2457 #endif
2458 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2459 	    lookupflags, ifp));
2460 }
2461 
2462 struct inpcb *
2463 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2464     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2465     struct ifnet *ifp, struct mbuf *m)
2466 {
2467 #ifdef PCBGROUP
2468 	struct inpcbgroup *pcbgroup;
2469 #endif
2470 
2471 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2472 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2473 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2474 	    ("%s: LOCKPCB not set", __func__));
2475 
2476 #ifdef PCBGROUP
2477 	/*
2478 	 * If we can use a hardware-generated hash to look up the connection
2479 	 * group, use that connection group to find the inpcb.  Otherwise
2480 	 * fall back on a software hash -- or the reservation table if we're
2481 	 * using RSS.
2482 	 *
2483 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2484 	 */
2485 	if (in_pcbgroup_enabled(pcbinfo) &&
2486 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2487 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2488 		    m->m_pkthdr.flowid);
2489 		if (pcbgroup != NULL)
2490 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2491 			    fport, laddr, lport, lookupflags, ifp));
2492 #ifndef RSS
2493 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2494 		    fport);
2495 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2496 		    laddr, lport, lookupflags, ifp));
2497 #endif
2498 	}
2499 #endif
2500 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2501 	    lookupflags, ifp));
2502 }
2503 #endif /* INET */
2504 
2505 /*
2506  * Insert PCB onto various hash lists.
2507  */
2508 static int
2509 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2510 {
2511 	struct inpcbhead *pcbhash;
2512 	struct inpcbporthead *pcbporthash;
2513 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2514 	struct inpcbport *phd;
2515 	u_int32_t hashkey_faddr;
2516 	int so_options;
2517 
2518 	INP_WLOCK_ASSERT(inp);
2519 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2520 
2521 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2522 	    ("in_pcbinshash: INP_INHASHLIST"));
2523 
2524 #ifdef INET6
2525 	if (inp->inp_vflag & INP_IPV6)
2526 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2527 	else
2528 #endif
2529 	hashkey_faddr = inp->inp_faddr.s_addr;
2530 
2531 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2532 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2533 
2534 	pcbporthash = &pcbinfo->ipi_porthashbase[
2535 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2536 
2537 	/*
2538 	 * Add entry to load balance group.
2539 	 * Only do this if SO_REUSEPORT_LB is set.
2540 	 */
2541 	so_options = inp_so_options(inp);
2542 	if (so_options & SO_REUSEPORT_LB) {
2543 		int ret = in_pcbinslbgrouphash(inp);
2544 		if (ret) {
2545 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2546 			return (ret);
2547 		}
2548 	}
2549 
2550 	/*
2551 	 * Go through port list and look for a head for this lport.
2552 	 */
2553 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2554 		if (phd->phd_port == inp->inp_lport)
2555 			break;
2556 	}
2557 	/*
2558 	 * If none exists, malloc one and tack it on.
2559 	 */
2560 	if (phd == NULL) {
2561 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2562 		if (phd == NULL) {
2563 			return (ENOBUFS); /* XXX */
2564 		}
2565 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2566 		phd->phd_port = inp->inp_lport;
2567 		CK_LIST_INIT(&phd->phd_pcblist);
2568 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2569 	}
2570 	inp->inp_phd = phd;
2571 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2572 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2573 	inp->inp_flags |= INP_INHASHLIST;
2574 #ifdef PCBGROUP
2575 	if (m != NULL) {
2576 		in_pcbgroup_update_mbuf(inp, m);
2577 	} else {
2578 		in_pcbgroup_update(inp);
2579 	}
2580 #endif
2581 	return (0);
2582 }
2583 
2584 int
2585 in_pcbinshash(struct inpcb *inp)
2586 {
2587 
2588 	return (in_pcbinshash_internal(inp, NULL));
2589 }
2590 
2591 int
2592 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2593 {
2594 
2595 	return (in_pcbinshash_internal(inp, m));
2596 }
2597 
2598 /*
2599  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2600  * changed. NOTE: This does not handle the case of the lport changing (the
2601  * hashed port list would have to be updated as well), so the lport must
2602  * not change after in_pcbinshash() has been called.
2603  */
2604 void
2605 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2606 {
2607 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2608 	struct inpcbhead *head;
2609 	u_int32_t hashkey_faddr;
2610 
2611 	INP_WLOCK_ASSERT(inp);
2612 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2613 
2614 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2615 	    ("in_pcbrehash: !INP_INHASHLIST"));
2616 
2617 #ifdef INET6
2618 	if (inp->inp_vflag & INP_IPV6)
2619 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2620 	else
2621 #endif
2622 	hashkey_faddr = inp->inp_faddr.s_addr;
2623 
2624 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2625 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2626 
2627 	CK_LIST_REMOVE(inp, inp_hash);
2628 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2629 
2630 #ifdef PCBGROUP
2631 	if (m != NULL)
2632 		in_pcbgroup_update_mbuf(inp, m);
2633 	else
2634 		in_pcbgroup_update(inp);
2635 #endif
2636 }
2637 
2638 void
2639 in_pcbrehash(struct inpcb *inp)
2640 {
2641 
2642 	in_pcbrehash_mbuf(inp, NULL);
2643 }
2644 
2645 /*
2646  * Remove PCB from various lists.
2647  */
2648 static void
2649 in_pcbremlists(struct inpcb *inp)
2650 {
2651 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2652 
2653 #ifdef INVARIANTS
2654 	if (pcbinfo == &V_tcbinfo) {
2655 		NET_EPOCH_ASSERT();
2656 	} else {
2657 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2658 	}
2659 #endif
2660 
2661 	INP_WLOCK_ASSERT(inp);
2662 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2663 
2664 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2665 	if (inp->inp_flags & INP_INHASHLIST) {
2666 		struct inpcbport *phd = inp->inp_phd;
2667 
2668 		INP_HASH_WLOCK(pcbinfo);
2669 
2670 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2671 		in_pcbremlbgrouphash(inp);
2672 
2673 		CK_LIST_REMOVE(inp, inp_hash);
2674 		CK_LIST_REMOVE(inp, inp_portlist);
2675 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2676 			CK_LIST_REMOVE(phd, phd_hash);
2677 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2678 		}
2679 		INP_HASH_WUNLOCK(pcbinfo);
2680 		inp->inp_flags &= ~INP_INHASHLIST;
2681 	}
2682 	CK_LIST_REMOVE(inp, inp_list);
2683 	pcbinfo->ipi_count--;
2684 #ifdef PCBGROUP
2685 	in_pcbgroup_remove(inp);
2686 #endif
2687 }
2688 
2689 /*
2690  * Check for alternatives when higher level complains
2691  * about service problems.  For now, invalidate cached
2692  * routing information.  If the route was created dynamically
2693  * (by a redirect), time to try a default gateway again.
2694  */
2695 void
2696 in_losing(struct inpcb *inp)
2697 {
2698 
2699 	RO_INVALIDATE_CACHE(&inp->inp_route);
2700 	return;
2701 }
2702 
2703 /*
2704  * A set label operation has occurred at the socket layer, propagate the
2705  * label change into the in_pcb for the socket.
2706  */
2707 void
2708 in_pcbsosetlabel(struct socket *so)
2709 {
2710 #ifdef MAC
2711 	struct inpcb *inp;
2712 
2713 	inp = sotoinpcb(so);
2714 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2715 
2716 	INP_WLOCK(inp);
2717 	SOCK_LOCK(so);
2718 	mac_inpcb_sosetlabel(so, inp);
2719 	SOCK_UNLOCK(so);
2720 	INP_WUNLOCK(inp);
2721 #endif
2722 }
2723 
2724 /*
2725  * ipport_tick runs once per second, determining if random port allocation
2726  * should be continued.  If more than ipport_randomcps ports have been
2727  * allocated in the last second, then we return to sequential port
2728  * allocation. We return to random allocation only once we drop below
2729  * ipport_randomcps for at least ipport_randomtime seconds.
2730  */
2731 static void
2732 ipport_tick(void *xtp)
2733 {
2734 	VNET_ITERATOR_DECL(vnet_iter);
2735 
2736 	VNET_LIST_RLOCK_NOSLEEP();
2737 	VNET_FOREACH(vnet_iter) {
2738 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2739 		if (V_ipport_tcpallocs <=
2740 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2741 			if (V_ipport_stoprandom > 0)
2742 				V_ipport_stoprandom--;
2743 		} else
2744 			V_ipport_stoprandom = V_ipport_randomtime;
2745 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2746 		CURVNET_RESTORE();
2747 	}
2748 	VNET_LIST_RUNLOCK_NOSLEEP();
2749 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2750 }
2751 
2752 static void
2753 ip_fini(void *xtp)
2754 {
2755 
2756 	callout_stop(&ipport_tick_callout);
2757 }
2758 
2759 /*
2760  * The ipport_callout should start running at about the time we attach the
2761  * inet or inet6 domains.
2762  */
2763 static void
2764 ipport_tick_init(const void *unused __unused)
2765 {
2766 
2767 	/* Start ipport_tick. */
2768 	callout_init(&ipport_tick_callout, 1);
2769 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2770 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2771 		SHUTDOWN_PRI_DEFAULT);
2772 }
2773 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2774     ipport_tick_init, NULL);
2775 
2776 void
2777 inp_wlock(struct inpcb *inp)
2778 {
2779 
2780 	INP_WLOCK(inp);
2781 }
2782 
2783 void
2784 inp_wunlock(struct inpcb *inp)
2785 {
2786 
2787 	INP_WUNLOCK(inp);
2788 }
2789 
2790 void
2791 inp_rlock(struct inpcb *inp)
2792 {
2793 
2794 	INP_RLOCK(inp);
2795 }
2796 
2797 void
2798 inp_runlock(struct inpcb *inp)
2799 {
2800 
2801 	INP_RUNLOCK(inp);
2802 }
2803 
2804 #ifdef INVARIANT_SUPPORT
2805 void
2806 inp_lock_assert(struct inpcb *inp)
2807 {
2808 
2809 	INP_WLOCK_ASSERT(inp);
2810 }
2811 
2812 void
2813 inp_unlock_assert(struct inpcb *inp)
2814 {
2815 
2816 	INP_UNLOCK_ASSERT(inp);
2817 }
2818 #endif
2819 
2820 void
2821 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2822 {
2823 	struct inpcb *inp;
2824 
2825 	INP_INFO_WLOCK(&V_tcbinfo);
2826 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2827 		INP_WLOCK(inp);
2828 		func(inp, arg);
2829 		INP_WUNLOCK(inp);
2830 	}
2831 	INP_INFO_WUNLOCK(&V_tcbinfo);
2832 }
2833 
2834 struct socket *
2835 inp_inpcbtosocket(struct inpcb *inp)
2836 {
2837 
2838 	INP_WLOCK_ASSERT(inp);
2839 	return (inp->inp_socket);
2840 }
2841 
2842 struct tcpcb *
2843 inp_inpcbtotcpcb(struct inpcb *inp)
2844 {
2845 
2846 	INP_WLOCK_ASSERT(inp);
2847 	return ((struct tcpcb *)inp->inp_ppcb);
2848 }
2849 
2850 int
2851 inp_ip_tos_get(const struct inpcb *inp)
2852 {
2853 
2854 	return (inp->inp_ip_tos);
2855 }
2856 
2857 void
2858 inp_ip_tos_set(struct inpcb *inp, int val)
2859 {
2860 
2861 	inp->inp_ip_tos = val;
2862 }
2863 
2864 void
2865 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2866     uint32_t *faddr, uint16_t *fp)
2867 {
2868 
2869 	INP_LOCK_ASSERT(inp);
2870 	*laddr = inp->inp_laddr.s_addr;
2871 	*faddr = inp->inp_faddr.s_addr;
2872 	*lp = inp->inp_lport;
2873 	*fp = inp->inp_fport;
2874 }
2875 
2876 struct inpcb *
2877 so_sotoinpcb(struct socket *so)
2878 {
2879 
2880 	return (sotoinpcb(so));
2881 }
2882 
2883 struct tcpcb *
2884 so_sototcpcb(struct socket *so)
2885 {
2886 
2887 	return (sototcpcb(so));
2888 }
2889 
2890 /*
2891  * Create an external-format (``xinpcb'') structure using the information in
2892  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2893  * reduce the spew of irrelevant information over this interface, to isolate
2894  * user code from changes in the kernel structure, and potentially to provide
2895  * information-hiding if we decide that some of this information should be
2896  * hidden from users.
2897  */
2898 void
2899 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2900 {
2901 
2902 	bzero(xi, sizeof(*xi));
2903 	xi->xi_len = sizeof(struct xinpcb);
2904 	if (inp->inp_socket)
2905 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2906 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2907 	xi->inp_gencnt = inp->inp_gencnt;
2908 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2909 	xi->inp_flow = inp->inp_flow;
2910 	xi->inp_flowid = inp->inp_flowid;
2911 	xi->inp_flowtype = inp->inp_flowtype;
2912 	xi->inp_flags = inp->inp_flags;
2913 	xi->inp_flags2 = inp->inp_flags2;
2914 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2915 	xi->in6p_cksum = inp->in6p_cksum;
2916 	xi->in6p_hops = inp->in6p_hops;
2917 	xi->inp_ip_tos = inp->inp_ip_tos;
2918 	xi->inp_vflag = inp->inp_vflag;
2919 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2920 	xi->inp_ip_p = inp->inp_ip_p;
2921 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2922 }
2923 
2924 #ifdef DDB
2925 static void
2926 db_print_indent(int indent)
2927 {
2928 	int i;
2929 
2930 	for (i = 0; i < indent; i++)
2931 		db_printf(" ");
2932 }
2933 
2934 static void
2935 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2936 {
2937 	char faddr_str[48], laddr_str[48];
2938 
2939 	db_print_indent(indent);
2940 	db_printf("%s at %p\n", name, inc);
2941 
2942 	indent += 2;
2943 
2944 #ifdef INET6
2945 	if (inc->inc_flags & INC_ISIPV6) {
2946 		/* IPv6. */
2947 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2948 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2949 	} else
2950 #endif
2951 	{
2952 		/* IPv4. */
2953 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2954 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2955 	}
2956 	db_print_indent(indent);
2957 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2958 	    ntohs(inc->inc_lport));
2959 	db_print_indent(indent);
2960 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2961 	    ntohs(inc->inc_fport));
2962 }
2963 
2964 static void
2965 db_print_inpflags(int inp_flags)
2966 {
2967 	int comma;
2968 
2969 	comma = 0;
2970 	if (inp_flags & INP_RECVOPTS) {
2971 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2972 		comma = 1;
2973 	}
2974 	if (inp_flags & INP_RECVRETOPTS) {
2975 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2976 		comma = 1;
2977 	}
2978 	if (inp_flags & INP_RECVDSTADDR) {
2979 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2980 		comma = 1;
2981 	}
2982 	if (inp_flags & INP_ORIGDSTADDR) {
2983 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2984 		comma = 1;
2985 	}
2986 	if (inp_flags & INP_HDRINCL) {
2987 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2988 		comma = 1;
2989 	}
2990 	if (inp_flags & INP_HIGHPORT) {
2991 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2992 		comma = 1;
2993 	}
2994 	if (inp_flags & INP_LOWPORT) {
2995 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2996 		comma = 1;
2997 	}
2998 	if (inp_flags & INP_ANONPORT) {
2999 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
3000 		comma = 1;
3001 	}
3002 	if (inp_flags & INP_RECVIF) {
3003 		db_printf("%sINP_RECVIF", comma ? ", " : "");
3004 		comma = 1;
3005 	}
3006 	if (inp_flags & INP_MTUDISC) {
3007 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
3008 		comma = 1;
3009 	}
3010 	if (inp_flags & INP_RECVTTL) {
3011 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
3012 		comma = 1;
3013 	}
3014 	if (inp_flags & INP_DONTFRAG) {
3015 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3016 		comma = 1;
3017 	}
3018 	if (inp_flags & INP_RECVTOS) {
3019 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3020 		comma = 1;
3021 	}
3022 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3023 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3024 		comma = 1;
3025 	}
3026 	if (inp_flags & IN6P_PKTINFO) {
3027 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3028 		comma = 1;
3029 	}
3030 	if (inp_flags & IN6P_HOPLIMIT) {
3031 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3032 		comma = 1;
3033 	}
3034 	if (inp_flags & IN6P_HOPOPTS) {
3035 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3036 		comma = 1;
3037 	}
3038 	if (inp_flags & IN6P_DSTOPTS) {
3039 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3040 		comma = 1;
3041 	}
3042 	if (inp_flags & IN6P_RTHDR) {
3043 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3044 		comma = 1;
3045 	}
3046 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3047 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3048 		comma = 1;
3049 	}
3050 	if (inp_flags & IN6P_TCLASS) {
3051 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3052 		comma = 1;
3053 	}
3054 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3055 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3056 		comma = 1;
3057 	}
3058 	if (inp_flags & INP_TIMEWAIT) {
3059 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3060 		comma  = 1;
3061 	}
3062 	if (inp_flags & INP_ONESBCAST) {
3063 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3064 		comma  = 1;
3065 	}
3066 	if (inp_flags & INP_DROPPED) {
3067 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3068 		comma  = 1;
3069 	}
3070 	if (inp_flags & INP_SOCKREF) {
3071 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3072 		comma  = 1;
3073 	}
3074 	if (inp_flags & IN6P_RFC2292) {
3075 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3076 		comma = 1;
3077 	}
3078 	if (inp_flags & IN6P_MTU) {
3079 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3080 		comma = 1;
3081 	}
3082 }
3083 
3084 static void
3085 db_print_inpvflag(u_char inp_vflag)
3086 {
3087 	int comma;
3088 
3089 	comma = 0;
3090 	if (inp_vflag & INP_IPV4) {
3091 		db_printf("%sINP_IPV4", comma ? ", " : "");
3092 		comma  = 1;
3093 	}
3094 	if (inp_vflag & INP_IPV6) {
3095 		db_printf("%sINP_IPV6", comma ? ", " : "");
3096 		comma  = 1;
3097 	}
3098 	if (inp_vflag & INP_IPV6PROTO) {
3099 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3100 		comma  = 1;
3101 	}
3102 }
3103 
3104 static void
3105 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3106 {
3107 
3108 	db_print_indent(indent);
3109 	db_printf("%s at %p\n", name, inp);
3110 
3111 	indent += 2;
3112 
3113 	db_print_indent(indent);
3114 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3115 
3116 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3117 
3118 	db_print_indent(indent);
3119 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3120 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3121 
3122 	db_print_indent(indent);
3123 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3124 	   inp->inp_label, inp->inp_flags);
3125 	db_print_inpflags(inp->inp_flags);
3126 	db_printf(")\n");
3127 
3128 	db_print_indent(indent);
3129 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3130 	    inp->inp_vflag);
3131 	db_print_inpvflag(inp->inp_vflag);
3132 	db_printf(")\n");
3133 
3134 	db_print_indent(indent);
3135 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3136 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3137 
3138 	db_print_indent(indent);
3139 #ifdef INET6
3140 	if (inp->inp_vflag & INP_IPV6) {
3141 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3142 		    "in6p_moptions: %p\n", inp->in6p_options,
3143 		    inp->in6p_outputopts, inp->in6p_moptions);
3144 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3145 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3146 		    inp->in6p_hops);
3147 	} else
3148 #endif
3149 	{
3150 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3151 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3152 		    inp->inp_options, inp->inp_moptions);
3153 	}
3154 
3155 	db_print_indent(indent);
3156 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3157 	    (uintmax_t)inp->inp_gencnt);
3158 }
3159 
3160 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3161 {
3162 	struct inpcb *inp;
3163 
3164 	if (!have_addr) {
3165 		db_printf("usage: show inpcb <addr>\n");
3166 		return;
3167 	}
3168 	inp = (struct inpcb *)addr;
3169 
3170 	db_print_inpcb(inp, "inpcb", 0);
3171 }
3172 #endif /* DDB */
3173 
3174 #ifdef RATELIMIT
3175 /*
3176  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3177  * if any.
3178  */
3179 int
3180 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3181 {
3182 	union if_snd_tag_modify_params params = {
3183 		.rate_limit.max_rate = max_pacing_rate,
3184 		.rate_limit.flags = M_NOWAIT,
3185 	};
3186 	struct m_snd_tag *mst;
3187 	struct ifnet *ifp;
3188 	int error;
3189 
3190 	mst = inp->inp_snd_tag;
3191 	if (mst == NULL)
3192 		return (EINVAL);
3193 
3194 	ifp = mst->ifp;
3195 	if (ifp == NULL)
3196 		return (EINVAL);
3197 
3198 	if (ifp->if_snd_tag_modify == NULL) {
3199 		error = EOPNOTSUPP;
3200 	} else {
3201 		error = ifp->if_snd_tag_modify(mst, &params);
3202 	}
3203 	return (error);
3204 }
3205 
3206 /*
3207  * Query existing TX rate limit based on the existing
3208  * "inp->inp_snd_tag", if any.
3209  */
3210 int
3211 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3212 {
3213 	union if_snd_tag_query_params params = { };
3214 	struct m_snd_tag *mst;
3215 	struct ifnet *ifp;
3216 	int error;
3217 
3218 	mst = inp->inp_snd_tag;
3219 	if (mst == NULL)
3220 		return (EINVAL);
3221 
3222 	ifp = mst->ifp;
3223 	if (ifp == NULL)
3224 		return (EINVAL);
3225 
3226 	if (ifp->if_snd_tag_query == NULL) {
3227 		error = EOPNOTSUPP;
3228 	} else {
3229 		error = ifp->if_snd_tag_query(mst, &params);
3230 		if (error == 0 &&  p_max_pacing_rate != NULL)
3231 			*p_max_pacing_rate = params.rate_limit.max_rate;
3232 	}
3233 	return (error);
3234 }
3235 
3236 /*
3237  * Query existing TX queue level based on the existing
3238  * "inp->inp_snd_tag", if any.
3239  */
3240 int
3241 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3242 {
3243 	union if_snd_tag_query_params params = { };
3244 	struct m_snd_tag *mst;
3245 	struct ifnet *ifp;
3246 	int error;
3247 
3248 	mst = inp->inp_snd_tag;
3249 	if (mst == NULL)
3250 		return (EINVAL);
3251 
3252 	ifp = mst->ifp;
3253 	if (ifp == NULL)
3254 		return (EINVAL);
3255 
3256 	if (ifp->if_snd_tag_query == NULL)
3257 		return (EOPNOTSUPP);
3258 
3259 	error = ifp->if_snd_tag_query(mst, &params);
3260 	if (error == 0 &&  p_txqueue_level != NULL)
3261 		*p_txqueue_level = params.rate_limit.queue_level;
3262 	return (error);
3263 }
3264 
3265 /*
3266  * Allocate a new TX rate limit send tag from the network interface
3267  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3268  */
3269 int
3270 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3271     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3272 
3273 {
3274 	union if_snd_tag_alloc_params params = {
3275 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3276 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3277 		.rate_limit.hdr.flowid = flowid,
3278 		.rate_limit.hdr.flowtype = flowtype,
3279 		.rate_limit.max_rate = max_pacing_rate,
3280 		.rate_limit.flags = M_NOWAIT,
3281 	};
3282 	int error;
3283 
3284 	INP_WLOCK_ASSERT(inp);
3285 
3286 	if (*st != NULL)
3287 		return (EINVAL);
3288 
3289 	if (ifp->if_snd_tag_alloc == NULL) {
3290 		error = EOPNOTSUPP;
3291 	} else {
3292 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3293 
3294 #ifdef INET
3295 		if (error == 0) {
3296 			counter_u64_add(rate_limit_set_ok, 1);
3297 			counter_u64_add(rate_limit_active, 1);
3298 		} else
3299 			counter_u64_add(rate_limit_alloc_fail, 1);
3300 #endif
3301 	}
3302 	return (error);
3303 }
3304 
3305 void
3306 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3307 {
3308 	if (ifp == NULL)
3309 		return;
3310 
3311 	/*
3312 	 * If the device was detached while we still had reference(s)
3313 	 * on the ifp, we assume if_snd_tag_free() was replaced with
3314 	 * stubs.
3315 	 */
3316 	ifp->if_snd_tag_free(mst);
3317 
3318 	/* release reference count on network interface */
3319 	if_rele(ifp);
3320 #ifdef INET
3321 	counter_u64_add(rate_limit_active, -1);
3322 #endif
3323 }
3324 
3325 /*
3326  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3327  * if any:
3328  */
3329 void
3330 in_pcbdetach_txrtlmt(struct inpcb *inp)
3331 {
3332 	struct m_snd_tag *mst;
3333 
3334 	INP_WLOCK_ASSERT(inp);
3335 
3336 	mst = inp->inp_snd_tag;
3337 	inp->inp_snd_tag = NULL;
3338 
3339 	if (mst == NULL)
3340 		return;
3341 
3342 	m_snd_tag_rele(mst);
3343 }
3344 
3345 int
3346 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3347 {
3348 	int error;
3349 
3350 	/*
3351 	 * If the existing send tag is for the wrong interface due to
3352 	 * a route change, first drop the existing tag.  Set the
3353 	 * CHANGED flag so that we will keep trying to allocate a new
3354 	 * tag if we fail to allocate one this time.
3355 	 */
3356 	if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3357 		in_pcbdetach_txrtlmt(inp);
3358 		inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3359 	}
3360 
3361 	/*
3362 	 * NOTE: When attaching to a network interface a reference is
3363 	 * made to ensure the network interface doesn't go away until
3364 	 * all ratelimit connections are gone. The network interface
3365 	 * pointers compared below represent valid network interfaces,
3366 	 * except when comparing towards NULL.
3367 	 */
3368 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3369 		error = 0;
3370 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3371 		if (inp->inp_snd_tag != NULL)
3372 			in_pcbdetach_txrtlmt(inp);
3373 		error = 0;
3374 	} else if (inp->inp_snd_tag == NULL) {
3375 		/*
3376 		 * In order to utilize packet pacing with RSS, we need
3377 		 * to wait until there is a valid RSS hash before we
3378 		 * can proceed:
3379 		 */
3380 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3381 			error = EAGAIN;
3382 		} else {
3383 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3384 			    mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3385 		}
3386 	} else {
3387 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3388 	}
3389 	if (error == 0 || error == EOPNOTSUPP)
3390 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3391 
3392 	return (error);
3393 }
3394 
3395 /*
3396  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3397  * is set in the fast path and will attach/detach/modify the TX rate
3398  * limit send tag based on the socket's so_max_pacing_rate value.
3399  */
3400 void
3401 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3402 {
3403 	struct socket *socket;
3404 	uint32_t max_pacing_rate;
3405 	bool did_upgrade;
3406 	int error;
3407 
3408 	if (inp == NULL)
3409 		return;
3410 
3411 	socket = inp->inp_socket;
3412 	if (socket == NULL)
3413 		return;
3414 
3415 	if (!INP_WLOCKED(inp)) {
3416 		/*
3417 		 * NOTE: If the write locking fails, we need to bail
3418 		 * out and use the non-ratelimited ring for the
3419 		 * transmit until there is a new chance to get the
3420 		 * write lock.
3421 		 */
3422 		if (!INP_TRY_UPGRADE(inp))
3423 			return;
3424 		did_upgrade = 1;
3425 	} else {
3426 		did_upgrade = 0;
3427 	}
3428 
3429 	/*
3430 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3431 	 * because atomic updates are not required since the variable
3432 	 * is checked at every mbuf we send. It is assumed that the
3433 	 * variable read itself will be atomic.
3434 	 */
3435 	max_pacing_rate = socket->so_max_pacing_rate;
3436 
3437 	error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3438 
3439 	if (did_upgrade)
3440 		INP_DOWNGRADE(inp);
3441 }
3442 
3443 /*
3444  * Track route changes for TX rate limiting.
3445  */
3446 void
3447 in_pcboutput_eagain(struct inpcb *inp)
3448 {
3449 	bool did_upgrade;
3450 
3451 	if (inp == NULL)
3452 		return;
3453 
3454 	if (inp->inp_snd_tag == NULL)
3455 		return;
3456 
3457 	if (!INP_WLOCKED(inp)) {
3458 		/*
3459 		 * NOTE: If the write locking fails, we need to bail
3460 		 * out and use the non-ratelimited ring for the
3461 		 * transmit until there is a new chance to get the
3462 		 * write lock.
3463 		 */
3464 		if (!INP_TRY_UPGRADE(inp))
3465 			return;
3466 		did_upgrade = 1;
3467 	} else {
3468 		did_upgrade = 0;
3469 	}
3470 
3471 	/* detach rate limiting */
3472 	in_pcbdetach_txrtlmt(inp);
3473 
3474 	/* make sure new mbuf send tag allocation is made */
3475 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3476 
3477 	if (did_upgrade)
3478 		INP_DOWNGRADE(inp);
3479 }
3480 
3481 #ifdef INET
3482 static void
3483 rl_init(void *st)
3484 {
3485 	rate_limit_active = counter_u64_alloc(M_WAITOK);
3486 	rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3487 	rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3488 }
3489 
3490 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3491 #endif
3492 #endif /* RATELIMIT */
3493