xref: /freebsd/sys/netinet/in_pcb.c (revision d746ab215cc85d8f7ab05c5f866c338782c390ec)
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_route.h"
49 #include "opt_rss.h"
50 
51 #include <sys/param.h>
52 #include <sys/hash.h>
53 #include <sys/systm.h>
54 #include <sys/libkern.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/callout.h>
59 #include <sys/eventhandler.h>
60 #include <sys/domain.h>
61 #include <sys/protosw.h>
62 #include <sys/smp.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sockio.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/refcount.h>
69 #include <sys/jail.h>
70 #include <sys/kernel.h>
71 #include <sys/sysctl.h>
72 
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif
76 
77 #include <vm/uma.h>
78 #include <vm/vm.h>
79 
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/if_types.h>
83 #include <net/if_llatbl.h>
84 #include <net/route.h>
85 #include <net/rss_config.h>
86 #include <net/vnet.h>
87 
88 #if defined(INET) || defined(INET6)
89 #include <netinet/in.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_pcb_var.h>
92 #ifdef INET
93 #include <netinet/in_var.h>
94 #include <netinet/in_fib.h>
95 #endif
96 #include <netinet/ip_var.h>
97 #include <netinet/tcp_var.h>
98 #ifdef TCPHPTS
99 #include <netinet/tcp_hpts.h>
100 #endif
101 #include <netinet/udp.h>
102 #include <netinet/udp_var.h>
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #include <netinet6/in6_pcb.h>
106 #include <netinet6/in6_var.h>
107 #include <netinet6/ip6_var.h>
108 #endif /* INET6 */
109 #include <net/route/nhop.h>
110 #endif
111 
112 #include <netipsec/ipsec_support.h>
113 
114 #include <security/mac/mac_framework.h>
115 
116 #define	INPCBLBGROUP_SIZMIN	8
117 #define	INPCBLBGROUP_SIZMAX	256
118 #define	INP_FREED	0x00000200	/* See in_pcb.h. */
119 
120 static struct callout	ipport_tick_callout;
121 
122 /*
123  * These configure the range of local port addresses assigned to
124  * "unspecified" outgoing connections/packets/whatever.
125  */
126 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
127 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
128 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
129 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
130 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
131 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
132 
133 /*
134  * Reserved ports accessible only to root. There are significant
135  * security considerations that must be accounted for when changing these,
136  * but the security benefits can be great. Please be careful.
137  */
138 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
139 VNET_DEFINE(int, ipport_reservedlow);
140 
141 /* Variables dealing with random ephemeral port allocation. */
142 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
143 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
144 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
145 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
146 VNET_DEFINE(int, ipport_tcpallocs);
147 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
148 
149 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
150 
151 #ifdef INET
152 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
153 			    struct in_addr faddr, u_int fport_arg,
154 			    struct in_addr laddr, u_int lport_arg,
155 			    int lookupflags, struct ifnet *ifp,
156 			    uint8_t numa_domain);
157 
158 #define RANGECHK(var, min, max) \
159 	if ((var) < (min)) { (var) = (min); } \
160 	else if ((var) > (max)) { (var) = (max); }
161 
162 static int
163 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
164 {
165 	int error;
166 
167 	error = sysctl_handle_int(oidp, arg1, arg2, req);
168 	if (error == 0) {
169 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
170 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
171 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
172 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
173 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
174 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
175 	}
176 	return (error);
177 }
178 
179 #undef RANGECHK
180 
181 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
182     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
183     "IP Ports");
184 
185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
186     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
187     &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
188     "");
189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
190     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
191     &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
192     "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
194     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
195     &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
196     "");
197 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
198     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
199     &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
200     "");
201 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
202     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
203     &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
204     "");
205 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
206     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
207     &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
208     "");
209 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
210 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
211 	&VNET_NAME(ipport_reservedhigh), 0, "");
212 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
213 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
214 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
215 	CTLFLAG_VNET | CTLFLAG_RW,
216 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
217 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
218 	CTLFLAG_VNET | CTLFLAG_RW,
219 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
220 	"allocations before switching to a sequential one");
221 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
222 	CTLFLAG_VNET | CTLFLAG_RW,
223 	&VNET_NAME(ipport_randomtime), 0,
224 	"Minimum time to keep sequential port "
225 	"allocation before switching to a random one");
226 
227 #ifdef RATELIMIT
228 counter_u64_t rate_limit_new;
229 counter_u64_t rate_limit_chg;
230 counter_u64_t rate_limit_active;
231 counter_u64_t rate_limit_alloc_fail;
232 counter_u64_t rate_limit_set_ok;
233 
234 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
235     "IP Rate Limiting");
236 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
237     &rate_limit_active, "Active rate limited connections");
238 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
239    &rate_limit_alloc_fail, "Rate limited connection failures");
240 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
241    &rate_limit_set_ok, "Rate limited setting succeeded");
242 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD,
243    &rate_limit_new, "Total Rate limit new attempts");
244 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD,
245    &rate_limit_chg, "Total Rate limited change attempts");
246 
247 #endif /* RATELIMIT */
248 
249 #endif /* INET */
250 
251 VNET_DEFINE(uint32_t, in_pcbhashseed);
252 static void
253 in_pcbhashseed_init(void)
254 {
255 
256 	V_in_pcbhashseed = arc4random();
257 }
258 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
259     in_pcbhashseed_init, 0);
260 
261 /*
262  * in_pcb.c: manage the Protocol Control Blocks.
263  *
264  * NOTE: It is assumed that most of these functions will be called with
265  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
266  * functions often modify hash chains or addresses in pcbs.
267  */
268 
269 static struct inpcblbgroup *
270 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
271     uint16_t port, const union in_dependaddr *addr, int size,
272     uint8_t numa_domain)
273 {
274 	struct inpcblbgroup *grp;
275 	size_t bytes;
276 
277 	bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
278 	grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
279 	if (!grp)
280 		return (NULL);
281 	grp->il_vflag = vflag;
282 	grp->il_lport = port;
283 	grp->il_numa_domain = numa_domain;
284 	grp->il_dependladdr = *addr;
285 	grp->il_inpsiz = size;
286 	CK_LIST_INSERT_HEAD(hdr, grp, il_list);
287 	return (grp);
288 }
289 
290 static void
291 in_pcblbgroup_free_deferred(epoch_context_t ctx)
292 {
293 	struct inpcblbgroup *grp;
294 
295 	grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
296 	free(grp, M_PCB);
297 }
298 
299 static void
300 in_pcblbgroup_free(struct inpcblbgroup *grp)
301 {
302 
303 	CK_LIST_REMOVE(grp, il_list);
304 	NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
305 }
306 
307 static struct inpcblbgroup *
308 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
309     struct inpcblbgroup *old_grp, int size)
310 {
311 	struct inpcblbgroup *grp;
312 	int i;
313 
314 	grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
315 	    old_grp->il_lport, &old_grp->il_dependladdr, size,
316 	    old_grp->il_numa_domain);
317 	if (grp == NULL)
318 		return (NULL);
319 
320 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
321 	    ("invalid new local group size %d and old local group count %d",
322 	     grp->il_inpsiz, old_grp->il_inpcnt));
323 
324 	for (i = 0; i < old_grp->il_inpcnt; ++i)
325 		grp->il_inp[i] = old_grp->il_inp[i];
326 	grp->il_inpcnt = old_grp->il_inpcnt;
327 	in_pcblbgroup_free(old_grp);
328 	return (grp);
329 }
330 
331 /*
332  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
333  * and shrink group if possible.
334  */
335 static void
336 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
337     int i)
338 {
339 	struct inpcblbgroup *grp, *new_grp;
340 
341 	grp = *grpp;
342 	for (; i + 1 < grp->il_inpcnt; ++i)
343 		grp->il_inp[i] = grp->il_inp[i + 1];
344 	grp->il_inpcnt--;
345 
346 	if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
347 	    grp->il_inpcnt <= grp->il_inpsiz / 4) {
348 		/* Shrink this group. */
349 		new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
350 		if (new_grp != NULL)
351 			*grpp = new_grp;
352 	}
353 }
354 
355 /*
356  * Add PCB to load balance group for SO_REUSEPORT_LB option.
357  */
358 static int
359 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain)
360 {
361 	const static struct timeval interval = { 60, 0 };
362 	static struct timeval lastprint;
363 	struct inpcbinfo *pcbinfo;
364 	struct inpcblbgrouphead *hdr;
365 	struct inpcblbgroup *grp;
366 	uint32_t idx;
367 
368 	pcbinfo = inp->inp_pcbinfo;
369 
370 	INP_WLOCK_ASSERT(inp);
371 	INP_HASH_WLOCK_ASSERT(pcbinfo);
372 
373 	/*
374 	 * Don't allow jailed socket to join local group.
375 	 */
376 	if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
377 		return (0);
378 
379 #ifdef INET6
380 	/*
381 	 * Don't allow IPv4 mapped INET6 wild socket.
382 	 */
383 	if ((inp->inp_vflag & INP_IPV4) &&
384 	    inp->inp_laddr.s_addr == INADDR_ANY &&
385 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
386 		return (0);
387 	}
388 #endif
389 
390 	idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
391 	hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
392 	CK_LIST_FOREACH(grp, hdr, il_list) {
393 		if (grp->il_vflag == inp->inp_vflag &&
394 		    grp->il_lport == inp->inp_lport &&
395 		    grp->il_numa_domain == numa_domain &&
396 		    memcmp(&grp->il_dependladdr,
397 		    &inp->inp_inc.inc_ie.ie_dependladdr,
398 		    sizeof(grp->il_dependladdr)) == 0)
399 			break;
400 	}
401 	if (grp == NULL) {
402 		/* Create new load balance group. */
403 		grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
404 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
405 		    INPCBLBGROUP_SIZMIN, numa_domain);
406 		if (grp == NULL)
407 			return (ENOBUFS);
408 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
409 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
410 			if (ratecheck(&lastprint, &interval))
411 				printf("lb group port %d, limit reached\n",
412 				    ntohs(grp->il_lport));
413 			return (0);
414 		}
415 
416 		/* Expand this local group. */
417 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
418 		if (grp == NULL)
419 			return (ENOBUFS);
420 	}
421 
422 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
423 	    ("invalid local group size %d and count %d", grp->il_inpsiz,
424 	    grp->il_inpcnt));
425 
426 	grp->il_inp[grp->il_inpcnt] = inp;
427 	grp->il_inpcnt++;
428 	return (0);
429 }
430 
431 /*
432  * Remove PCB from load balance group.
433  */
434 static void
435 in_pcbremlbgrouphash(struct inpcb *inp)
436 {
437 	struct inpcbinfo *pcbinfo;
438 	struct inpcblbgrouphead *hdr;
439 	struct inpcblbgroup *grp;
440 	int i;
441 
442 	pcbinfo = inp->inp_pcbinfo;
443 
444 	INP_WLOCK_ASSERT(inp);
445 	INP_HASH_WLOCK_ASSERT(pcbinfo);
446 
447 	hdr = &pcbinfo->ipi_lbgrouphashbase[
448 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
449 	CK_LIST_FOREACH(grp, hdr, il_list) {
450 		for (i = 0; i < grp->il_inpcnt; ++i) {
451 			if (grp->il_inp[i] != inp)
452 				continue;
453 
454 			if (grp->il_inpcnt == 1) {
455 				/* We are the last, free this local group. */
456 				in_pcblbgroup_free(grp);
457 			} else {
458 				/* Pull up inpcbs, shrink group if possible. */
459 				in_pcblbgroup_reorder(hdr, &grp, i);
460 			}
461 			return;
462 		}
463 	}
464 }
465 
466 int
467 in_pcblbgroup_numa(struct inpcb *inp, int arg)
468 {
469 	struct inpcbinfo *pcbinfo;
470 	struct inpcblbgrouphead *hdr;
471 	struct inpcblbgroup *grp;
472 	int err, i;
473 	uint8_t numa_domain;
474 
475 	switch (arg) {
476 	case TCP_REUSPORT_LB_NUMA_NODOM:
477 		numa_domain = M_NODOM;
478 		break;
479 	case TCP_REUSPORT_LB_NUMA_CURDOM:
480 		numa_domain = PCPU_GET(domain);
481 		break;
482 	default:
483 		if (arg < 0 || arg >= vm_ndomains)
484 			return (EINVAL);
485 		numa_domain = arg;
486 	}
487 
488 	err = 0;
489 	pcbinfo = inp->inp_pcbinfo;
490 	INP_WLOCK_ASSERT(inp);
491 	INP_HASH_WLOCK(pcbinfo);
492 	hdr = &pcbinfo->ipi_lbgrouphashbase[
493 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
494 	CK_LIST_FOREACH(grp, hdr, il_list) {
495 		for (i = 0; i < grp->il_inpcnt; ++i) {
496 			if (grp->il_inp[i] != inp)
497 				continue;
498 
499 			if (grp->il_numa_domain == numa_domain) {
500 				goto abort_with_hash_wlock;
501 			}
502 
503 			/* Remove it from the old group. */
504 			in_pcbremlbgrouphash(inp);
505 
506 			/* Add it to the new group based on numa domain. */
507 			in_pcbinslbgrouphash(inp, numa_domain);
508 			goto abort_with_hash_wlock;
509 		}
510 	}
511 	err = ENOENT;
512 abort_with_hash_wlock:
513 	INP_HASH_WUNLOCK(pcbinfo);
514 	return (err);
515 }
516 
517 /* Make sure it is safe to use hashinit(9) on CK_LIST. */
518 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb)));
519 
520 /*
521  * Initialize an inpcbinfo - a per-VNET instance of connections db.
522  */
523 void
524 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor,
525     u_int hash_nelements, u_int porthash_nelements)
526 {
527 
528 	mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF);
529 	mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name,
530 	    NULL, MTX_DEF);
531 #ifdef VIMAGE
532 	pcbinfo->ipi_vnet = curvnet;
533 #endif
534 	CK_LIST_INIT(&pcbinfo->ipi_listhead);
535 	pcbinfo->ipi_count = 0;
536 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
537 	    &pcbinfo->ipi_hashmask);
538 	porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
539 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
540 	    &pcbinfo->ipi_porthashmask);
541 	pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
542 	    &pcbinfo->ipi_lbgrouphashmask);
543 	pcbinfo->ipi_zone = pcbstor->ips_zone;
544 	pcbinfo->ipi_portzone = pcbstor->ips_portzone;
545 	pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
546 }
547 
548 /*
549  * Destroy an inpcbinfo.
550  */
551 void
552 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
553 {
554 
555 	KASSERT(pcbinfo->ipi_count == 0,
556 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
557 
558 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
559 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
560 	    pcbinfo->ipi_porthashmask);
561 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
562 	    pcbinfo->ipi_lbgrouphashmask);
563 	mtx_destroy(&pcbinfo->ipi_hash_lock);
564 	mtx_destroy(&pcbinfo->ipi_lock);
565 }
566 
567 /*
568  * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
569  */
570 static void inpcb_dtor(void *, int, void *);
571 static void inpcb_fini(void *, int);
572 void
573 in_pcbstorage_init(void *arg)
574 {
575 	struct inpcbstorage *pcbstor = arg;
576 
577 	pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
578 	    sizeof(struct inpcb), NULL, inpcb_dtor, pcbstor->ips_pcbinit,
579 	    inpcb_fini, UMA_ALIGN_PTR, UMA_ZONE_SMR);
580 	pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
581 	    sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
582 	uma_zone_set_smr(pcbstor->ips_portzone,
583 	    uma_zone_get_smr(pcbstor->ips_zone));
584 }
585 
586 /*
587  * Destroy a pcbstorage - used by unloadable protocols.
588  */
589 void
590 in_pcbstorage_destroy(void *arg)
591 {
592 	struct inpcbstorage *pcbstor = arg;
593 
594 	uma_zdestroy(pcbstor->ips_zone);
595 	uma_zdestroy(pcbstor->ips_portzone);
596 }
597 
598 /*
599  * Allocate a PCB and associate it with the socket.
600  * On success return with the PCB locked.
601  */
602 int
603 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
604 {
605 	struct inpcb *inp;
606 	int error;
607 
608 	error = 0;
609 	inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
610 	if (inp == NULL)
611 		return (ENOBUFS);
612 	bzero(&inp->inp_start_zero, inp_zero_size);
613 #ifdef NUMA
614 	inp->inp_numa_domain = M_NODOM;
615 #endif
616 	inp->inp_pcbinfo = pcbinfo;
617 	inp->inp_socket = so;
618 	inp->inp_cred = crhold(so->so_cred);
619 	inp->inp_inc.inc_fibnum = so->so_fibnum;
620 #ifdef MAC
621 	error = mac_inpcb_init(inp, M_NOWAIT);
622 	if (error != 0)
623 		goto out;
624 	mac_inpcb_create(so, inp);
625 #endif
626 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
627 	error = ipsec_init_pcbpolicy(inp);
628 	if (error != 0) {
629 #ifdef MAC
630 		mac_inpcb_destroy(inp);
631 #endif
632 		goto out;
633 	}
634 #endif /*IPSEC*/
635 #ifdef INET6
636 	if (INP_SOCKAF(so) == AF_INET6) {
637 		inp->inp_vflag |= INP_IPV6PROTO;
638 		if (V_ip6_v6only)
639 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
640 	}
641 	if (V_ip6_auto_flowlabel)
642 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
643 #endif
644 	/*
645 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
646 	 * to be cleaned up.
647 	 */
648 	inp->inp_route.ro_flags = RT_LLE_CACHE;
649 #ifdef TCPHPTS
650 	/*
651 	 * If using hpts lets drop a random number in so
652 	 * not all new connections fall on the same CPU.
653 	 */
654 	inp->inp_hpts_cpu = hpts_random_cpu(inp);
655 #endif
656 	refcount_init(&inp->inp_refcount, 1);   /* Reference from socket. */
657 	INP_WLOCK(inp);
658 	INP_INFO_WLOCK(pcbinfo);
659 	pcbinfo->ipi_count++;
660 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
661 	CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
662 	INP_INFO_WUNLOCK(pcbinfo);
663 	so->so_pcb = inp;
664 
665 	return (0);
666 
667 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
668 out:
669 	uma_zfree_smr(pcbinfo->ipi_zone, inp);
670 	return (error);
671 #endif
672 }
673 
674 #ifdef INET
675 int
676 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
677 {
678 	int anonport, error;
679 
680 	KASSERT(nam == NULL || nam->sa_family == AF_INET,
681 	    ("%s: invalid address family for %p", __func__, nam));
682 	KASSERT(nam == NULL || nam->sa_len == sizeof(struct sockaddr_in),
683 	    ("%s: invalid address length for %p", __func__, nam));
684 	INP_WLOCK_ASSERT(inp);
685 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
686 
687 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
688 		return (EINVAL);
689 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
690 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
691 	    &inp->inp_lport, cred);
692 	if (error)
693 		return (error);
694 	if (in_pcbinshash(inp) != 0) {
695 		inp->inp_laddr.s_addr = INADDR_ANY;
696 		inp->inp_lport = 0;
697 		return (EAGAIN);
698 	}
699 	if (anonport)
700 		inp->inp_flags |= INP_ANONPORT;
701 	return (0);
702 }
703 #endif
704 
705 #if defined(INET) || defined(INET6)
706 /*
707  * Assign a local port like in_pcb_lport(), but also used with connect()
708  * and a foreign address and port.  If fsa is non-NULL, choose a local port
709  * that is unused with those, otherwise one that is completely unused.
710  * lsa can be NULL for IPv6.
711  */
712 int
713 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
714     struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
715 {
716 	struct inpcbinfo *pcbinfo;
717 	struct inpcb *tmpinp;
718 	unsigned short *lastport;
719 	int count, dorandom, error;
720 	u_short aux, first, last, lport;
721 #ifdef INET
722 	struct in_addr laddr, faddr;
723 #endif
724 #ifdef INET6
725 	struct in6_addr *laddr6, *faddr6;
726 #endif
727 
728 	pcbinfo = inp->inp_pcbinfo;
729 
730 	/*
731 	 * Because no actual state changes occur here, a global write lock on
732 	 * the pcbinfo isn't required.
733 	 */
734 	INP_LOCK_ASSERT(inp);
735 	INP_HASH_LOCK_ASSERT(pcbinfo);
736 
737 	if (inp->inp_flags & INP_HIGHPORT) {
738 		first = V_ipport_hifirstauto;	/* sysctl */
739 		last  = V_ipport_hilastauto;
740 		lastport = &pcbinfo->ipi_lasthi;
741 	} else if (inp->inp_flags & INP_LOWPORT) {
742 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
743 		if (error)
744 			return (error);
745 		first = V_ipport_lowfirstauto;	/* 1023 */
746 		last  = V_ipport_lowlastauto;	/* 600 */
747 		lastport = &pcbinfo->ipi_lastlow;
748 	} else {
749 		first = V_ipport_firstauto;	/* sysctl */
750 		last  = V_ipport_lastauto;
751 		lastport = &pcbinfo->ipi_lastport;
752 	}
753 	/*
754 	 * For UDP(-Lite), use random port allocation as long as the user
755 	 * allows it.  For TCP (and as of yet unknown) connections,
756 	 * use random port allocation only if the user allows it AND
757 	 * ipport_tick() allows it.
758 	 */
759 	if (V_ipport_randomized &&
760 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
761 		pcbinfo == &V_ulitecbinfo))
762 		dorandom = 1;
763 	else
764 		dorandom = 0;
765 	/*
766 	 * It makes no sense to do random port allocation if
767 	 * we have the only port available.
768 	 */
769 	if (first == last)
770 		dorandom = 0;
771 	/* Make sure to not include UDP(-Lite) packets in the count. */
772 	if (pcbinfo != &V_udbinfo && pcbinfo != &V_ulitecbinfo)
773 		V_ipport_tcpallocs++;
774 	/*
775 	 * Instead of having two loops further down counting up or down
776 	 * make sure that first is always <= last and go with only one
777 	 * code path implementing all logic.
778 	 */
779 	if (first > last) {
780 		aux = first;
781 		first = last;
782 		last = aux;
783 	}
784 
785 #ifdef INET
786 	laddr.s_addr = INADDR_ANY;
787 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
788 		if (lsa != NULL)
789 			laddr = ((struct sockaddr_in *)lsa)->sin_addr;
790 		if (fsa != NULL)
791 			faddr = ((struct sockaddr_in *)fsa)->sin_addr;
792 	}
793 #endif
794 #ifdef INET6
795 	laddr6 = NULL;
796 	if ((inp->inp_vflag & INP_IPV6) != 0) {
797 		if (lsa != NULL)
798 			laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
799 		if (fsa != NULL)
800 			faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
801 	}
802 #endif
803 
804 	tmpinp = NULL;
805 	lport = *lportp;
806 
807 	if (dorandom)
808 		*lastport = first + (arc4random() % (last - first));
809 
810 	count = last - first;
811 
812 	do {
813 		if (count-- < 0)	/* completely used? */
814 			return (EADDRNOTAVAIL);
815 		++*lastport;
816 		if (*lastport < first || *lastport > last)
817 			*lastport = first;
818 		lport = htons(*lastport);
819 
820 		if (fsa != NULL) {
821 #ifdef INET
822 			if (lsa->sa_family == AF_INET) {
823 				tmpinp = in_pcblookup_hash_locked(pcbinfo,
824 				    faddr, fport, laddr, lport, lookupflags,
825 				    NULL, M_NODOM);
826 			}
827 #endif
828 #ifdef INET6
829 			if (lsa->sa_family == AF_INET6) {
830 				tmpinp = in6_pcblookup_hash_locked(pcbinfo,
831 				    faddr6, fport, laddr6, lport, lookupflags,
832 				    NULL, M_NODOM);
833 			}
834 #endif
835 		} else {
836 #ifdef INET6
837 			if ((inp->inp_vflag & INP_IPV6) != 0)
838 				tmpinp = in6_pcblookup_local(pcbinfo,
839 				    &inp->in6p_laddr, lport, lookupflags, cred);
840 #endif
841 #if defined(INET) && defined(INET6)
842 			else
843 #endif
844 #ifdef INET
845 				tmpinp = in_pcblookup_local(pcbinfo, laddr,
846 				    lport, lookupflags, cred);
847 #endif
848 		}
849 	} while (tmpinp != NULL);
850 
851 	*lportp = lport;
852 
853 	return (0);
854 }
855 
856 /*
857  * Select a local port (number) to use.
858  */
859 int
860 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
861     struct ucred *cred, int lookupflags)
862 {
863 	struct sockaddr_in laddr;
864 
865 	if (laddrp) {
866 		bzero(&laddr, sizeof(laddr));
867 		laddr.sin_family = AF_INET;
868 		laddr.sin_addr = *laddrp;
869 	}
870 	return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
871 	    NULL, lportp, NULL, 0, cred, lookupflags));
872 }
873 
874 /*
875  * Return cached socket options.
876  */
877 int
878 inp_so_options(const struct inpcb *inp)
879 {
880 	int so_options;
881 
882 	so_options = 0;
883 
884 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
885 		so_options |= SO_REUSEPORT_LB;
886 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
887 		so_options |= SO_REUSEPORT;
888 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
889 		so_options |= SO_REUSEADDR;
890 	return (so_options);
891 }
892 #endif /* INET || INET6 */
893 
894 /*
895  * Check if a new BINDMULTI socket is allowed to be created.
896  *
897  * ni points to the new inp.
898  * oi points to the exisitng inp.
899  *
900  * This checks whether the existing inp also has BINDMULTI and
901  * whether the credentials match.
902  */
903 int
904 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
905 {
906 	/* Check permissions match */
907 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
908 	    (ni->inp_cred->cr_uid !=
909 	    oi->inp_cred->cr_uid))
910 		return (0);
911 
912 	/* Check the existing inp has BINDMULTI set */
913 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
914 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
915 		return (0);
916 
917 	/*
918 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
919 	 * it is and it matches the checks.
920 	 */
921 	return (1);
922 }
923 
924 #ifdef INET
925 /*
926  * Set up a bind operation on a PCB, performing port allocation
927  * as required, but do not actually modify the PCB. Callers can
928  * either complete the bind by setting inp_laddr/inp_lport and
929  * calling in_pcbinshash(), or they can just use the resulting
930  * port and address to authorise the sending of a once-off packet.
931  *
932  * On error, the values of *laddrp and *lportp are not changed.
933  */
934 int
935 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
936     u_short *lportp, struct ucred *cred)
937 {
938 	struct socket *so = inp->inp_socket;
939 	struct sockaddr_in *sin;
940 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
941 	struct in_addr laddr;
942 	u_short lport = 0;
943 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
944 	int error;
945 
946 	/*
947 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
948 	 * so that we don't have to add to the (already messy) code below.
949 	 */
950 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
951 
952 	/*
953 	 * No state changes, so read locks are sufficient here.
954 	 */
955 	INP_LOCK_ASSERT(inp);
956 	INP_HASH_LOCK_ASSERT(pcbinfo);
957 
958 	laddr.s_addr = *laddrp;
959 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
960 		return (EINVAL);
961 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
962 		lookupflags = INPLOOKUP_WILDCARD;
963 	if (nam == NULL) {
964 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
965 			return (error);
966 	} else {
967 		sin = (struct sockaddr_in *)nam;
968 		KASSERT(sin->sin_family == AF_INET,
969 		    ("%s: invalid family for address %p", __func__, sin));
970 		KASSERT(sin->sin_len == sizeof(*sin),
971 		    ("%s: invalid length for address %p", __func__, sin));
972 
973 		error = prison_local_ip4(cred, &sin->sin_addr);
974 		if (error)
975 			return (error);
976 		if (sin->sin_port != *lportp) {
977 			/* Don't allow the port to change. */
978 			if (*lportp != 0)
979 				return (EINVAL);
980 			lport = sin->sin_port;
981 		}
982 		/* NB: lport is left as 0 if the port isn't being changed. */
983 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
984 			/*
985 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
986 			 * allow complete duplication of binding if
987 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
988 			 * and a multicast address is bound on both
989 			 * new and duplicated sockets.
990 			 */
991 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
992 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
993 			/*
994 			 * XXX: How to deal with SO_REUSEPORT_LB here?
995 			 * Treat same as SO_REUSEPORT for now.
996 			 */
997 			if ((so->so_options &
998 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
999 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
1000 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
1001 			sin->sin_port = 0;		/* yech... */
1002 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
1003 			/*
1004 			 * Is the address a local IP address?
1005 			 * If INP_BINDANY is set, then the socket may be bound
1006 			 * to any endpoint address, local or not.
1007 			 */
1008 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
1009 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
1010 				return (EADDRNOTAVAIL);
1011 		}
1012 		laddr = sin->sin_addr;
1013 		if (lport) {
1014 			struct inpcb *t;
1015 			struct tcptw *tw;
1016 
1017 			/* GROSS */
1018 			if (ntohs(lport) <= V_ipport_reservedhigh &&
1019 			    ntohs(lport) >= V_ipport_reservedlow &&
1020 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
1021 				return (EACCES);
1022 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
1023 			    priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
1024 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
1025 				    lport, INPLOOKUP_WILDCARD, cred);
1026 	/*
1027 	 * XXX
1028 	 * This entire block sorely needs a rewrite.
1029 	 */
1030 				if (t &&
1031 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
1032 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
1033 				    (so->so_type != SOCK_STREAM ||
1034 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
1035 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
1036 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
1037 				     (t->inp_flags2 & INP_REUSEPORT) ||
1038 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
1039 				    (inp->inp_cred->cr_uid !=
1040 				     t->inp_cred->cr_uid))
1041 					return (EADDRINUSE);
1042 
1043 				/*
1044 				 * If the socket is a BINDMULTI socket, then
1045 				 * the credentials need to match and the
1046 				 * original socket also has to have been bound
1047 				 * with BINDMULTI.
1048 				 */
1049 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1050 					return (EADDRINUSE);
1051 			}
1052 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
1053 			    lport, lookupflags, cred);
1054 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
1055 				/*
1056 				 * XXXRW: If an incpb has had its timewait
1057 				 * state recycled, we treat the address as
1058 				 * being in use (for now).  This is better
1059 				 * than a panic, but not desirable.
1060 				 */
1061 				tw = intotw(t);
1062 				if (tw == NULL ||
1063 				    ((reuseport & tw->tw_so_options) == 0 &&
1064 					(reuseport_lb &
1065 				            tw->tw_so_options) == 0)) {
1066 					return (EADDRINUSE);
1067 				}
1068 			} else if (t &&
1069 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
1070 				   (reuseport & inp_so_options(t)) == 0 &&
1071 				   (reuseport_lb & inp_so_options(t)) == 0) {
1072 #ifdef INET6
1073 				if (ntohl(sin->sin_addr.s_addr) !=
1074 				    INADDR_ANY ||
1075 				    ntohl(t->inp_laddr.s_addr) !=
1076 				    INADDR_ANY ||
1077 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
1078 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
1079 #endif
1080 						return (EADDRINUSE);
1081 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1082 					return (EADDRINUSE);
1083 			}
1084 		}
1085 	}
1086 	if (*lportp != 0)
1087 		lport = *lportp;
1088 	if (lport == 0) {
1089 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1090 		if (error != 0)
1091 			return (error);
1092 	}
1093 	*laddrp = laddr.s_addr;
1094 	*lportp = lport;
1095 	return (0);
1096 }
1097 
1098 /*
1099  * Connect from a socket to a specified address.
1100  * Both address and port must be specified in argument sin.
1101  * If don't have a local address for this socket yet,
1102  * then pick one.
1103  */
1104 int
1105 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred,
1106     bool rehash)
1107 {
1108 	u_short lport, fport;
1109 	in_addr_t laddr, faddr;
1110 	int anonport, error;
1111 
1112 	INP_WLOCK_ASSERT(inp);
1113 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1114 
1115 	lport = inp->inp_lport;
1116 	laddr = inp->inp_laddr.s_addr;
1117 	anonport = (lport == 0);
1118 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
1119 	    NULL, cred);
1120 	if (error)
1121 		return (error);
1122 
1123 	/* Do the initial binding of the local address if required. */
1124 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1125 		KASSERT(rehash == true,
1126 		    ("Rehashing required for unbound inps"));
1127 		inp->inp_lport = lport;
1128 		inp->inp_laddr.s_addr = laddr;
1129 		if (in_pcbinshash(inp) != 0) {
1130 			inp->inp_laddr.s_addr = INADDR_ANY;
1131 			inp->inp_lport = 0;
1132 			return (EAGAIN);
1133 		}
1134 	}
1135 
1136 	/* Commit the remaining changes. */
1137 	inp->inp_lport = lport;
1138 	inp->inp_laddr.s_addr = laddr;
1139 	inp->inp_faddr.s_addr = faddr;
1140 	inp->inp_fport = fport;
1141 	if (rehash) {
1142 		in_pcbrehash(inp);
1143 	} else {
1144 		in_pcbinshash(inp);
1145 	}
1146 
1147 	if (anonport)
1148 		inp->inp_flags |= INP_ANONPORT;
1149 	return (0);
1150 }
1151 
1152 /*
1153  * Do proper source address selection on an unbound socket in case
1154  * of connect. Take jails into account as well.
1155  */
1156 int
1157 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1158     struct ucred *cred)
1159 {
1160 	struct ifaddr *ifa;
1161 	struct sockaddr *sa;
1162 	struct sockaddr_in *sin, dst;
1163 	struct nhop_object *nh;
1164 	int error;
1165 
1166 	NET_EPOCH_ASSERT();
1167 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1168 	/*
1169 	 * Bypass source address selection and use the primary jail IP
1170 	 * if requested.
1171 	 */
1172 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1173 		return (0);
1174 
1175 	error = 0;
1176 
1177 	nh = NULL;
1178 	bzero(&dst, sizeof(dst));
1179 	sin = &dst;
1180 	sin->sin_family = AF_INET;
1181 	sin->sin_len = sizeof(struct sockaddr_in);
1182 	sin->sin_addr.s_addr = faddr->s_addr;
1183 
1184 	/*
1185 	 * If route is known our src addr is taken from the i/f,
1186 	 * else punt.
1187 	 *
1188 	 * Find out route to destination.
1189 	 */
1190 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1191 		nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1192 		    0, NHR_NONE, 0);
1193 
1194 	/*
1195 	 * If we found a route, use the address corresponding to
1196 	 * the outgoing interface.
1197 	 *
1198 	 * Otherwise assume faddr is reachable on a directly connected
1199 	 * network and try to find a corresponding interface to take
1200 	 * the source address from.
1201 	 */
1202 	if (nh == NULL || nh->nh_ifp == NULL) {
1203 		struct in_ifaddr *ia;
1204 		struct ifnet *ifp;
1205 
1206 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1207 					inp->inp_socket->so_fibnum));
1208 		if (ia == NULL) {
1209 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1210 						inp->inp_socket->so_fibnum));
1211 		}
1212 		if (ia == NULL) {
1213 			error = ENETUNREACH;
1214 			goto done;
1215 		}
1216 
1217 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1218 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1219 			goto done;
1220 		}
1221 
1222 		ifp = ia->ia_ifp;
1223 		ia = NULL;
1224 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1225 			sa = ifa->ifa_addr;
1226 			if (sa->sa_family != AF_INET)
1227 				continue;
1228 			sin = (struct sockaddr_in *)sa;
1229 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1230 				ia = (struct in_ifaddr *)ifa;
1231 				break;
1232 			}
1233 		}
1234 		if (ia != NULL) {
1235 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1236 			goto done;
1237 		}
1238 
1239 		/* 3. As a last resort return the 'default' jail address. */
1240 		error = prison_get_ip4(cred, laddr);
1241 		goto done;
1242 	}
1243 
1244 	/*
1245 	 * If the outgoing interface on the route found is not
1246 	 * a loopback interface, use the address from that interface.
1247 	 * In case of jails do those three steps:
1248 	 * 1. check if the interface address belongs to the jail. If so use it.
1249 	 * 2. check if we have any address on the outgoing interface
1250 	 *    belonging to this jail. If so use it.
1251 	 * 3. as a last resort return the 'default' jail address.
1252 	 */
1253 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1254 		struct in_ifaddr *ia;
1255 		struct ifnet *ifp;
1256 
1257 		/* If not jailed, use the default returned. */
1258 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1259 			ia = (struct in_ifaddr *)nh->nh_ifa;
1260 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1261 			goto done;
1262 		}
1263 
1264 		/* Jailed. */
1265 		/* 1. Check if the iface address belongs to the jail. */
1266 		sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1267 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1268 			ia = (struct in_ifaddr *)nh->nh_ifa;
1269 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1270 			goto done;
1271 		}
1272 
1273 		/*
1274 		 * 2. Check if we have any address on the outgoing interface
1275 		 *    belonging to this jail.
1276 		 */
1277 		ia = NULL;
1278 		ifp = nh->nh_ifp;
1279 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1280 			sa = ifa->ifa_addr;
1281 			if (sa->sa_family != AF_INET)
1282 				continue;
1283 			sin = (struct sockaddr_in *)sa;
1284 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1285 				ia = (struct in_ifaddr *)ifa;
1286 				break;
1287 			}
1288 		}
1289 		if (ia != NULL) {
1290 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1291 			goto done;
1292 		}
1293 
1294 		/* 3. As a last resort return the 'default' jail address. */
1295 		error = prison_get_ip4(cred, laddr);
1296 		goto done;
1297 	}
1298 
1299 	/*
1300 	 * The outgoing interface is marked with 'loopback net', so a route
1301 	 * to ourselves is here.
1302 	 * Try to find the interface of the destination address and then
1303 	 * take the address from there. That interface is not necessarily
1304 	 * a loopback interface.
1305 	 * In case of jails, check that it is an address of the jail
1306 	 * and if we cannot find, fall back to the 'default' jail address.
1307 	 */
1308 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1309 		struct in_ifaddr *ia;
1310 
1311 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1312 					inp->inp_socket->so_fibnum));
1313 		if (ia == NULL)
1314 			ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1315 						inp->inp_socket->so_fibnum));
1316 		if (ia == NULL)
1317 			ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1318 
1319 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1320 			if (ia == NULL) {
1321 				error = ENETUNREACH;
1322 				goto done;
1323 			}
1324 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1325 			goto done;
1326 		}
1327 
1328 		/* Jailed. */
1329 		if (ia != NULL) {
1330 			struct ifnet *ifp;
1331 
1332 			ifp = ia->ia_ifp;
1333 			ia = NULL;
1334 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1335 				sa = ifa->ifa_addr;
1336 				if (sa->sa_family != AF_INET)
1337 					continue;
1338 				sin = (struct sockaddr_in *)sa;
1339 				if (prison_check_ip4(cred,
1340 				    &sin->sin_addr) == 0) {
1341 					ia = (struct in_ifaddr *)ifa;
1342 					break;
1343 				}
1344 			}
1345 			if (ia != NULL) {
1346 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1347 				goto done;
1348 			}
1349 		}
1350 
1351 		/* 3. As a last resort return the 'default' jail address. */
1352 		error = prison_get_ip4(cred, laddr);
1353 		goto done;
1354 	}
1355 
1356 done:
1357 	return (error);
1358 }
1359 
1360 /*
1361  * Set up for a connect from a socket to the specified address.
1362  * On entry, *laddrp and *lportp should contain the current local
1363  * address and port for the PCB; these are updated to the values
1364  * that should be placed in inp_laddr and inp_lport to complete
1365  * the connect.
1366  *
1367  * On success, *faddrp and *fportp will be set to the remote address
1368  * and port. These are not updated in the error case.
1369  *
1370  * If the operation fails because the connection already exists,
1371  * *oinpp will be set to the PCB of that connection so that the
1372  * caller can decide to override it. In all other cases, *oinpp
1373  * is set to NULL.
1374  */
1375 int
1376 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1377     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1378     struct inpcb **oinpp, struct ucred *cred)
1379 {
1380 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1381 	struct in_ifaddr *ia;
1382 	struct inpcb *oinp;
1383 	struct in_addr laddr, faddr;
1384 	u_short lport, fport;
1385 	int error;
1386 
1387 	KASSERT(sin->sin_family == AF_INET,
1388 	    ("%s: invalid address family for %p", __func__, sin));
1389 	KASSERT(sin->sin_len == sizeof(*sin),
1390 	    ("%s: invalid address length for %p", __func__, sin));
1391 
1392 	/*
1393 	 * Because a global state change doesn't actually occur here, a read
1394 	 * lock is sufficient.
1395 	 */
1396 	NET_EPOCH_ASSERT();
1397 	INP_LOCK_ASSERT(inp);
1398 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1399 
1400 	if (oinpp != NULL)
1401 		*oinpp = NULL;
1402 	if (sin->sin_port == 0)
1403 		return (EADDRNOTAVAIL);
1404 	laddr.s_addr = *laddrp;
1405 	lport = *lportp;
1406 	faddr = sin->sin_addr;
1407 	fport = sin->sin_port;
1408 #ifdef ROUTE_MPATH
1409 	if (CALC_FLOWID_OUTBOUND) {
1410 		uint32_t hash_val, hash_type;
1411 
1412 		hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1413 		    inp->inp_socket->so_proto->pr_protocol, &hash_type);
1414 
1415 		inp->inp_flowid = hash_val;
1416 		inp->inp_flowtype = hash_type;
1417 	}
1418 #endif
1419 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1420 		/*
1421 		 * If the destination address is INADDR_ANY,
1422 		 * use the primary local address.
1423 		 * If the supplied address is INADDR_BROADCAST,
1424 		 * and the primary interface supports broadcast,
1425 		 * choose the broadcast address for that interface.
1426 		 */
1427 		if (faddr.s_addr == INADDR_ANY) {
1428 			faddr =
1429 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1430 			if (cred != NULL &&
1431 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1432 				return (error);
1433 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1434 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1435 			    IFF_BROADCAST)
1436 				faddr = satosin(&CK_STAILQ_FIRST(
1437 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1438 		}
1439 	}
1440 	if (laddr.s_addr == INADDR_ANY) {
1441 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1442 		/*
1443 		 * If the destination address is multicast and an outgoing
1444 		 * interface has been set as a multicast option, prefer the
1445 		 * address of that interface as our source address.
1446 		 */
1447 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1448 		    inp->inp_moptions != NULL) {
1449 			struct ip_moptions *imo;
1450 			struct ifnet *ifp;
1451 
1452 			imo = inp->inp_moptions;
1453 			if (imo->imo_multicast_ifp != NULL) {
1454 				ifp = imo->imo_multicast_ifp;
1455 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1456 					if ((ia->ia_ifp == ifp) &&
1457 					    (cred == NULL ||
1458 					    prison_check_ip4(cred,
1459 					    &ia->ia_addr.sin_addr) == 0))
1460 						break;
1461 				}
1462 				if (ia == NULL)
1463 					error = EADDRNOTAVAIL;
1464 				else {
1465 					laddr = ia->ia_addr.sin_addr;
1466 					error = 0;
1467 				}
1468 			}
1469 		}
1470 		if (error)
1471 			return (error);
1472 	}
1473 
1474 	if (lport != 0) {
1475 		oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1476 		    fport, laddr, lport, 0, NULL, M_NODOM);
1477 		if (oinp != NULL) {
1478 			if (oinpp != NULL)
1479 				*oinpp = oinp;
1480 			return (EADDRINUSE);
1481 		}
1482 	} else {
1483 		struct sockaddr_in lsin, fsin;
1484 
1485 		bzero(&lsin, sizeof(lsin));
1486 		bzero(&fsin, sizeof(fsin));
1487 		lsin.sin_family = AF_INET;
1488 		lsin.sin_addr = laddr;
1489 		fsin.sin_family = AF_INET;
1490 		fsin.sin_addr = faddr;
1491 		error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1492 		    &lport, (struct sockaddr *)& fsin, fport, cred,
1493 		    INPLOOKUP_WILDCARD);
1494 		if (error)
1495 			return (error);
1496 	}
1497 	*laddrp = laddr.s_addr;
1498 	*lportp = lport;
1499 	*faddrp = faddr.s_addr;
1500 	*fportp = fport;
1501 	return (0);
1502 }
1503 
1504 void
1505 in_pcbdisconnect(struct inpcb *inp)
1506 {
1507 
1508 	INP_WLOCK_ASSERT(inp);
1509 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1510 
1511 	inp->inp_faddr.s_addr = INADDR_ANY;
1512 	inp->inp_fport = 0;
1513 	in_pcbrehash(inp);
1514 }
1515 #endif /* INET */
1516 
1517 /*
1518  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1519  * For most protocols, this will be invoked immediately prior to calling
1520  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1521  * socket, in which case in_pcbfree() is deferred.
1522  */
1523 void
1524 in_pcbdetach(struct inpcb *inp)
1525 {
1526 
1527 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1528 
1529 #ifdef RATELIMIT
1530 	if (inp->inp_snd_tag != NULL)
1531 		in_pcbdetach_txrtlmt(inp);
1532 #endif
1533 	inp->inp_socket->so_pcb = NULL;
1534 	inp->inp_socket = NULL;
1535 }
1536 
1537 /*
1538  * inpcb hash lookups are protected by SMR section.
1539  *
1540  * Once desired pcb has been found, switching from SMR section to a pcb
1541  * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1542  * here because SMR is a critical section.
1543  * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1544  */
1545 static inline void
1546 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1547 {
1548 
1549 	lock == INPLOOKUP_RLOCKPCB ?
1550 	    rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1551 }
1552 
1553 static inline void
1554 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1555 {
1556 
1557 	lock == INPLOOKUP_RLOCKPCB ?
1558 	    rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1559 }
1560 
1561 static inline int
1562 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1563 {
1564 
1565 	return (lock == INPLOOKUP_RLOCKPCB ?
1566 	    rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1567 }
1568 
1569 static inline bool
1570 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1571 {
1572 
1573 	return (lock == INPLOOKUP_RLOCKPCB ?
1574 	    in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1575 }
1576 
1577 bool
1578 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1579 {
1580 
1581 	MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1582 	SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1583 
1584 	if (__predict_true(inp_trylock(inp, lock))) {
1585 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1586 			smr_exit(inp->inp_pcbinfo->ipi_smr);
1587 			inp_unlock(inp, lock);
1588 			return (false);
1589 		}
1590 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1591 		return (true);
1592 	}
1593 
1594 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1595 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1596 		inp_lock(inp, lock);
1597 		if (__predict_false(in_pcbrele(inp, lock)))
1598 			return (false);
1599 		/*
1600 		 * inp acquired through refcount & lock for sure didn't went
1601 		 * through uma_zfree().  However, it may have already went
1602 		 * through in_pcbfree() and has another reference, that
1603 		 * prevented its release by our in_pcbrele().
1604 		 */
1605 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1606 			inp_unlock(inp, lock);
1607 			return (false);
1608 		}
1609 		return (true);
1610 	} else {
1611 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1612 		return (false);
1613 	}
1614 }
1615 
1616 /*
1617  * inp_next() - inpcb hash/list traversal iterator
1618  *
1619  * Requires initialized struct inpcb_iterator for context.
1620  * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1621  *
1622  * - Iterator can have either write-lock or read-lock semantics, that can not
1623  *   be changed later.
1624  * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1625  *   a single hash slot.  Note: only rip_input() does the latter.
1626  * - Iterator may have optional bool matching function.  The matching function
1627  *   will be executed for each inpcb in the SMR context, so it can not acquire
1628  *   locks and can safely access only immutable fields of inpcb.
1629  *
1630  * A fresh initialized iterator has NULL inpcb in its context and that
1631  * means that inp_next() call would return the very first inpcb on the list
1632  * locked with desired semantic.  In all following calls the context pointer
1633  * shall hold the current inpcb pointer.  The KPI user is not supposed to
1634  * unlock the current inpcb!  Upon end of traversal inp_next() will return NULL
1635  * and write NULL to its context.  After end of traversal an iterator can be
1636  * reused.
1637  *
1638  * List traversals have the following features/constraints:
1639  * - New entries won't be seen, as they are always added to the head of a list.
1640  * - Removed entries won't stop traversal as long as they are not added to
1641  *   a different list. This is violated by in_pcbrehash().
1642  */
1643 #define	II_LIST_FIRST(ipi, hash)					\
1644 		(((hash) == INP_ALL_LIST) ?				\
1645 		    CK_LIST_FIRST(&(ipi)->ipi_listhead) :		\
1646 		    CK_LIST_FIRST(&(ipi)->ipi_hashbase[(hash)]))
1647 #define	II_LIST_NEXT(inp, hash)						\
1648 		(((hash) == INP_ALL_LIST) ?				\
1649 		    CK_LIST_NEXT((inp), inp_list) :			\
1650 		    CK_LIST_NEXT((inp), inp_hash))
1651 #define	II_LOCK_ASSERT(inp, lock)					\
1652 		rw_assert(&(inp)->inp_lock,				\
1653 		    (lock) == INPLOOKUP_RLOCKPCB ?  RA_RLOCKED : RA_WLOCKED )
1654 struct inpcb *
1655 inp_next(struct inpcb_iterator *ii)
1656 {
1657 	const struct inpcbinfo *ipi = ii->ipi;
1658 	inp_match_t *match = ii->match;
1659 	void *ctx = ii->ctx;
1660 	inp_lookup_t lock = ii->lock;
1661 	int hash = ii->hash;
1662 	struct inpcb *inp;
1663 
1664 	if (ii->inp == NULL) {		/* First call. */
1665 		smr_enter(ipi->ipi_smr);
1666 		/* This is unrolled CK_LIST_FOREACH(). */
1667 		for (inp = II_LIST_FIRST(ipi, hash);
1668 		    inp != NULL;
1669 		    inp = II_LIST_NEXT(inp, hash)) {
1670 			if (match != NULL && (match)(inp, ctx) == false)
1671 				continue;
1672 			if (__predict_true(inp_smr_lock(inp, lock)))
1673 				break;
1674 			else {
1675 				smr_enter(ipi->ipi_smr);
1676 				MPASS(inp != II_LIST_FIRST(ipi, hash));
1677 				inp = II_LIST_FIRST(ipi, hash);
1678 				if (inp == NULL)
1679 					break;
1680 			}
1681 		}
1682 
1683 		if (inp == NULL)
1684 			smr_exit(ipi->ipi_smr);
1685 		else
1686 			ii->inp = inp;
1687 
1688 		return (inp);
1689 	}
1690 
1691 	/* Not a first call. */
1692 	smr_enter(ipi->ipi_smr);
1693 restart:
1694 	inp = ii->inp;
1695 	II_LOCK_ASSERT(inp, lock);
1696 next:
1697 	inp = II_LIST_NEXT(inp, hash);
1698 	if (inp == NULL) {
1699 		smr_exit(ipi->ipi_smr);
1700 		goto found;
1701 	}
1702 
1703 	if (match != NULL && (match)(inp, ctx) == false)
1704 		goto next;
1705 
1706 	if (__predict_true(inp_trylock(inp, lock))) {
1707 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1708 			/*
1709 			 * Entries are never inserted in middle of a list, thus
1710 			 * as long as we are in SMR, we can continue traversal.
1711 			 * Jump to 'restart' should yield in the same result,
1712 			 * but could produce unnecessary looping.  Could this
1713 			 * looping be unbound?
1714 			 */
1715 			inp_unlock(inp, lock);
1716 			goto next;
1717 		} else {
1718 			smr_exit(ipi->ipi_smr);
1719 			goto found;
1720 		}
1721 	}
1722 
1723 	/*
1724 	 * Can't obtain lock immediately, thus going hard.  Once we exit the
1725 	 * SMR section we can no longer jump to 'next', and our only stable
1726 	 * anchoring point is ii->inp, which we keep locked for this case, so
1727 	 * we jump to 'restart'.
1728 	 */
1729 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1730 		smr_exit(ipi->ipi_smr);
1731 		inp_lock(inp, lock);
1732 		if (__predict_false(in_pcbrele(inp, lock))) {
1733 			smr_enter(ipi->ipi_smr);
1734 			goto restart;
1735 		}
1736 		/*
1737 		 * See comment in inp_smr_lock().
1738 		 */
1739 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1740 			inp_unlock(inp, lock);
1741 			smr_enter(ipi->ipi_smr);
1742 			goto restart;
1743 		}
1744 	} else
1745 		goto next;
1746 
1747 found:
1748 	inp_unlock(ii->inp, lock);
1749 	ii->inp = inp;
1750 
1751 	return (ii->inp);
1752 }
1753 
1754 /*
1755  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1756  * stability of an inpcb pointer despite the inpcb lock being released or
1757  * SMR section exited.
1758  *
1759  * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1760  */
1761 void
1762 in_pcbref(struct inpcb *inp)
1763 {
1764 	u_int old __diagused;
1765 
1766 	old = refcount_acquire(&inp->inp_refcount);
1767 	KASSERT(old > 0, ("%s: refcount 0", __func__));
1768 }
1769 
1770 /*
1771  * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1772  * freeing the pcb, if the reference was very last.
1773  */
1774 bool
1775 in_pcbrele_rlocked(struct inpcb *inp)
1776 {
1777 
1778 	INP_RLOCK_ASSERT(inp);
1779 
1780 	if (refcount_release(&inp->inp_refcount) == 0)
1781 		return (false);
1782 
1783 	MPASS(inp->inp_flags & INP_FREED);
1784 	MPASS(inp->inp_socket == NULL);
1785 	MPASS(inp->inp_in_hpts == 0);
1786 	INP_RUNLOCK(inp);
1787 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1788 	return (true);
1789 }
1790 
1791 bool
1792 in_pcbrele_wlocked(struct inpcb *inp)
1793 {
1794 
1795 	INP_WLOCK_ASSERT(inp);
1796 
1797 	if (refcount_release(&inp->inp_refcount) == 0)
1798 		return (false);
1799 
1800 	MPASS(inp->inp_flags & INP_FREED);
1801 	MPASS(inp->inp_socket == NULL);
1802 	MPASS(inp->inp_in_hpts == 0);
1803 	INP_WUNLOCK(inp);
1804 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1805 	return (true);
1806 }
1807 
1808 /*
1809  * Unconditionally schedule an inpcb to be freed by decrementing its
1810  * reference count, which should occur only after the inpcb has been detached
1811  * from its socket.  If another thread holds a temporary reference (acquired
1812  * using in_pcbref()) then the free is deferred until that reference is
1813  * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1814  *  Almost all work, including removal from global lists, is done in this
1815  * context, where the pcbinfo lock is held.
1816  */
1817 void
1818 in_pcbfree(struct inpcb *inp)
1819 {
1820 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1821 #ifdef INET
1822 	struct ip_moptions *imo;
1823 #endif
1824 #ifdef INET6
1825 	struct ip6_moptions *im6o;
1826 #endif
1827 
1828 	INP_WLOCK_ASSERT(inp);
1829 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1830 	KASSERT((inp->inp_flags & INP_FREED) == 0,
1831 	    ("%s: called twice for pcb %p", __func__, inp));
1832 
1833 	inp->inp_flags |= INP_FREED;
1834 	INP_INFO_WLOCK(pcbinfo);
1835 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1836 	pcbinfo->ipi_count--;
1837 	CK_LIST_REMOVE(inp, inp_list);
1838 	INP_INFO_WUNLOCK(pcbinfo);
1839 
1840 	if (inp->inp_flags & INP_INHASHLIST) {
1841 		struct inpcbport *phd = inp->inp_phd;
1842 
1843 		INP_HASH_WLOCK(pcbinfo);
1844 		/* XXX: Only do if SO_REUSEPORT_LB set? */
1845 		in_pcbremlbgrouphash(inp);
1846 
1847 		CK_LIST_REMOVE(inp, inp_hash);
1848 		CK_LIST_REMOVE(inp, inp_portlist);
1849 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1850 			CK_LIST_REMOVE(phd, phd_hash);
1851 			uma_zfree_smr(pcbinfo->ipi_portzone, phd);
1852 		}
1853 		INP_HASH_WUNLOCK(pcbinfo);
1854 		inp->inp_flags &= ~INP_INHASHLIST;
1855 	}
1856 
1857 	RO_INVALIDATE_CACHE(&inp->inp_route);
1858 #ifdef MAC
1859 	mac_inpcb_destroy(inp);
1860 #endif
1861 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1862 	if (inp->inp_sp != NULL)
1863 		ipsec_delete_pcbpolicy(inp);
1864 #endif
1865 #ifdef INET
1866 	if (inp->inp_options)
1867 		(void)m_free(inp->inp_options);
1868 	imo = inp->inp_moptions;
1869 #endif
1870 #ifdef INET6
1871 	if (inp->inp_vflag & INP_IPV6PROTO) {
1872 		ip6_freepcbopts(inp->in6p_outputopts);
1873 		im6o = inp->in6p_moptions;
1874 	} else
1875 		im6o = NULL;
1876 #endif
1877 
1878 	if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1879 		INP_WUNLOCK(inp);
1880 	}
1881 #ifdef INET6
1882 	ip6_freemoptions(im6o);
1883 #endif
1884 #ifdef INET
1885 	inp_freemoptions(imo);
1886 #endif
1887 	/* Destruction is finalized in inpcb_dtor(). */
1888 }
1889 
1890 static void
1891 inpcb_dtor(void *mem, int size, void *arg)
1892 {
1893 	struct inpcb *inp = mem;
1894 
1895 	crfree(inp->inp_cred);
1896 #ifdef INVARIANTS
1897 	inp->inp_cred = NULL;
1898 #endif
1899 }
1900 
1901 /*
1902  * Different protocols initialize their inpcbs differently - giving
1903  * different name to the lock.  But they all are disposed the same.
1904  */
1905 static void
1906 inpcb_fini(void *mem, int size)
1907 {
1908 	struct inpcb *inp = mem;
1909 
1910 	INP_LOCK_DESTROY(inp);
1911 }
1912 
1913 /*
1914  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1915  * port reservation, and preventing it from being returned by inpcb lookups.
1916  *
1917  * It is used by TCP to mark an inpcb as unused and avoid future packet
1918  * delivery or event notification when a socket remains open but TCP has
1919  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1920  * or a RST on the wire, and allows the port binding to be reused while still
1921  * maintaining the invariant that so_pcb always points to a valid inpcb until
1922  * in_pcbdetach().
1923  *
1924  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1925  * in_pcbnotifyall() and in_pcbpurgeif0()?
1926  */
1927 void
1928 in_pcbdrop(struct inpcb *inp)
1929 {
1930 
1931 	INP_WLOCK_ASSERT(inp);
1932 #ifdef INVARIANTS
1933 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1934 		MPASS(inp->inp_refcount > 1);
1935 #endif
1936 
1937 	/*
1938 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1939 	 * the hash lock...?
1940 	 */
1941 	inp->inp_flags |= INP_DROPPED;
1942 	if (inp->inp_flags & INP_INHASHLIST) {
1943 		struct inpcbport *phd = inp->inp_phd;
1944 
1945 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1946 		in_pcbremlbgrouphash(inp);
1947 		CK_LIST_REMOVE(inp, inp_hash);
1948 		CK_LIST_REMOVE(inp, inp_portlist);
1949 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1950 			CK_LIST_REMOVE(phd, phd_hash);
1951 			uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
1952 		}
1953 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1954 		inp->inp_flags &= ~INP_INHASHLIST;
1955 	}
1956 }
1957 
1958 #ifdef INET
1959 /*
1960  * Common routines to return the socket addresses associated with inpcbs.
1961  */
1962 struct sockaddr *
1963 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1964 {
1965 	struct sockaddr_in *sin;
1966 
1967 	sin = malloc(sizeof *sin, M_SONAME,
1968 		M_WAITOK | M_ZERO);
1969 	sin->sin_family = AF_INET;
1970 	sin->sin_len = sizeof(*sin);
1971 	sin->sin_addr = *addr_p;
1972 	sin->sin_port = port;
1973 
1974 	return (struct sockaddr *)sin;
1975 }
1976 
1977 int
1978 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1979 {
1980 	struct inpcb *inp;
1981 	struct in_addr addr;
1982 	in_port_t port;
1983 
1984 	inp = sotoinpcb(so);
1985 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1986 
1987 	INP_RLOCK(inp);
1988 	port = inp->inp_lport;
1989 	addr = inp->inp_laddr;
1990 	INP_RUNLOCK(inp);
1991 
1992 	*nam = in_sockaddr(port, &addr);
1993 	return 0;
1994 }
1995 
1996 int
1997 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1998 {
1999 	struct inpcb *inp;
2000 	struct in_addr addr;
2001 	in_port_t port;
2002 
2003 	inp = sotoinpcb(so);
2004 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
2005 
2006 	INP_RLOCK(inp);
2007 	port = inp->inp_fport;
2008 	addr = inp->inp_faddr;
2009 	INP_RUNLOCK(inp);
2010 
2011 	*nam = in_sockaddr(port, &addr);
2012 	return 0;
2013 }
2014 
2015 void
2016 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
2017     struct inpcb *(*notify)(struct inpcb *, int))
2018 {
2019 	struct inpcb *inp, *inp_temp;
2020 
2021 	INP_INFO_WLOCK(pcbinfo);
2022 	CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
2023 		INP_WLOCK(inp);
2024 #ifdef INET6
2025 		if ((inp->inp_vflag & INP_IPV4) == 0) {
2026 			INP_WUNLOCK(inp);
2027 			continue;
2028 		}
2029 #endif
2030 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
2031 		    inp->inp_socket == NULL) {
2032 			INP_WUNLOCK(inp);
2033 			continue;
2034 		}
2035 		if ((*notify)(inp, errno))
2036 			INP_WUNLOCK(inp);
2037 	}
2038 	INP_INFO_WUNLOCK(pcbinfo);
2039 }
2040 
2041 static bool
2042 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
2043 {
2044 
2045 	if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
2046 		return (true);
2047 	else
2048 		return (false);
2049 }
2050 
2051 void
2052 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
2053 {
2054 	struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
2055 	    inp_v4_multi_match, NULL);
2056 	struct inpcb *inp;
2057 	struct in_multi *inm;
2058 	struct in_mfilter *imf;
2059 	struct ip_moptions *imo;
2060 
2061 	IN_MULTI_LOCK_ASSERT();
2062 
2063 	while ((inp = inp_next(&inpi)) != NULL) {
2064 		INP_WLOCK_ASSERT(inp);
2065 
2066 		imo = inp->inp_moptions;
2067 		/*
2068 		 * Unselect the outgoing interface if it is being
2069 		 * detached.
2070 		 */
2071 		if (imo->imo_multicast_ifp == ifp)
2072 			imo->imo_multicast_ifp = NULL;
2073 
2074 		/*
2075 		 * Drop multicast group membership if we joined
2076 		 * through the interface being detached.
2077 		 *
2078 		 * XXX This can all be deferred to an epoch_call
2079 		 */
2080 restart:
2081 		IP_MFILTER_FOREACH(imf, &imo->imo_head) {
2082 			if ((inm = imf->imf_inm) == NULL)
2083 				continue;
2084 			if (inm->inm_ifp != ifp)
2085 				continue;
2086 			ip_mfilter_remove(&imo->imo_head, imf);
2087 			in_leavegroup_locked(inm, NULL);
2088 			ip_mfilter_free(imf);
2089 			goto restart;
2090 		}
2091 	}
2092 }
2093 
2094 /*
2095  * Lookup a PCB based on the local address and port.  Caller must hold the
2096  * hash lock.  No inpcb locks or references are acquired.
2097  */
2098 #define INP_LOOKUP_MAPPED_PCB_COST	3
2099 struct inpcb *
2100 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
2101     u_short lport, int lookupflags, struct ucred *cred)
2102 {
2103 	struct inpcb *inp;
2104 #ifdef INET6
2105 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
2106 #else
2107 	int matchwild = 3;
2108 #endif
2109 	int wildcard;
2110 
2111 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2112 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2113 	INP_HASH_LOCK_ASSERT(pcbinfo);
2114 
2115 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
2116 		struct inpcbhead *head;
2117 		/*
2118 		 * Look for an unconnected (wildcard foreign addr) PCB that
2119 		 * matches the local address and port we're looking for.
2120 		 */
2121 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2122 		    pcbinfo->ipi_hashmask)];
2123 		CK_LIST_FOREACH(inp, head, inp_hash) {
2124 #ifdef INET6
2125 			/* XXX inp locking */
2126 			if ((inp->inp_vflag & INP_IPV4) == 0)
2127 				continue;
2128 #endif
2129 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
2130 			    inp->inp_laddr.s_addr == laddr.s_addr &&
2131 			    inp->inp_lport == lport) {
2132 				/*
2133 				 * Found?
2134 				 */
2135 				if (cred == NULL ||
2136 				    prison_equal_ip4(cred->cr_prison,
2137 					inp->inp_cred->cr_prison))
2138 					return (inp);
2139 			}
2140 		}
2141 		/*
2142 		 * Not found.
2143 		 */
2144 		return (NULL);
2145 	} else {
2146 		struct inpcbporthead *porthash;
2147 		struct inpcbport *phd;
2148 		struct inpcb *match = NULL;
2149 		/*
2150 		 * Best fit PCB lookup.
2151 		 *
2152 		 * First see if this local port is in use by looking on the
2153 		 * port hash list.
2154 		 */
2155 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2156 		    pcbinfo->ipi_porthashmask)];
2157 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
2158 			if (phd->phd_port == lport)
2159 				break;
2160 		}
2161 		if (phd != NULL) {
2162 			/*
2163 			 * Port is in use by one or more PCBs. Look for best
2164 			 * fit.
2165 			 */
2166 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2167 				wildcard = 0;
2168 				if (cred != NULL &&
2169 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
2170 					cred->cr_prison))
2171 					continue;
2172 #ifdef INET6
2173 				/* XXX inp locking */
2174 				if ((inp->inp_vflag & INP_IPV4) == 0)
2175 					continue;
2176 				/*
2177 				 * We never select the PCB that has
2178 				 * INP_IPV6 flag and is bound to :: if
2179 				 * we have another PCB which is bound
2180 				 * to 0.0.0.0.  If a PCB has the
2181 				 * INP_IPV6 flag, then we set its cost
2182 				 * higher than IPv4 only PCBs.
2183 				 *
2184 				 * Note that the case only happens
2185 				 * when a socket is bound to ::, under
2186 				 * the condition that the use of the
2187 				 * mapped address is allowed.
2188 				 */
2189 				if ((inp->inp_vflag & INP_IPV6) != 0)
2190 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2191 #endif
2192 				if (inp->inp_faddr.s_addr != INADDR_ANY)
2193 					wildcard++;
2194 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
2195 					if (laddr.s_addr == INADDR_ANY)
2196 						wildcard++;
2197 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
2198 						continue;
2199 				} else {
2200 					if (laddr.s_addr != INADDR_ANY)
2201 						wildcard++;
2202 				}
2203 				if (wildcard < matchwild) {
2204 					match = inp;
2205 					matchwild = wildcard;
2206 					if (matchwild == 0)
2207 						break;
2208 				}
2209 			}
2210 		}
2211 		return (match);
2212 	}
2213 }
2214 #undef INP_LOOKUP_MAPPED_PCB_COST
2215 
2216 static struct inpcb *
2217 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2218     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2219     uint16_t fport, int lookupflags, int numa_domain)
2220 {
2221 	struct inpcb *local_wild, *numa_wild;
2222 	const struct inpcblbgrouphead *hdr;
2223 	struct inpcblbgroup *grp;
2224 	uint32_t idx;
2225 
2226 	INP_HASH_LOCK_ASSERT(pcbinfo);
2227 
2228 	hdr = &pcbinfo->ipi_lbgrouphashbase[
2229 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2230 
2231 	/*
2232 	 * Order of socket selection:
2233 	 * 1. non-wild.
2234 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
2235 	 *
2236 	 * NOTE:
2237 	 * - Load balanced group does not contain jailed sockets
2238 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
2239 	 */
2240 	local_wild = NULL;
2241 	numa_wild = NULL;
2242 	CK_LIST_FOREACH(grp, hdr, il_list) {
2243 #ifdef INET6
2244 		if (!(grp->il_vflag & INP_IPV4))
2245 			continue;
2246 #endif
2247 		if (grp->il_lport != lport)
2248 			continue;
2249 
2250 		idx = INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2251 		    grp->il_inpcnt;
2252 		if (grp->il_laddr.s_addr == laddr->s_addr) {
2253 			if (numa_domain == M_NODOM ||
2254 			    grp->il_numa_domain == numa_domain) {
2255 				return (grp->il_inp[idx]);
2256 			} else {
2257 				numa_wild = grp->il_inp[idx];
2258 			}
2259 		}
2260 		if (grp->il_laddr.s_addr == INADDR_ANY &&
2261 		    (lookupflags & INPLOOKUP_WILDCARD) != 0 &&
2262 		    (local_wild == NULL || numa_domain == M_NODOM ||
2263 			grp->il_numa_domain == numa_domain)) {
2264 			local_wild = grp->il_inp[idx];
2265 		}
2266 	}
2267 	if (numa_wild != NULL)
2268 		return (numa_wild);
2269 
2270 	return (local_wild);
2271 }
2272 
2273 /*
2274  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2275  * that the caller has either locked the hash list, which usually happens
2276  * for bind(2) operations, or is in SMR section, which happens when sorting
2277  * out incoming packets.
2278  */
2279 static struct inpcb *
2280 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2281     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2282     struct ifnet *ifp, uint8_t numa_domain)
2283 {
2284 	struct inpcbhead *head;
2285 	struct inpcb *inp, *tmpinp;
2286 	u_short fport = fport_arg, lport = lport_arg;
2287 
2288 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2289 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2290 	INP_HASH_LOCK_ASSERT(pcbinfo);
2291 
2292 	/*
2293 	 * First look for an exact match.
2294 	 */
2295 	tmpinp = NULL;
2296 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&faddr, lport, fport,
2297 	    pcbinfo->ipi_hashmask)];
2298 	CK_LIST_FOREACH(inp, head, inp_hash) {
2299 #ifdef INET6
2300 		/* XXX inp locking */
2301 		if ((inp->inp_vflag & INP_IPV4) == 0)
2302 			continue;
2303 #endif
2304 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2305 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2306 		    inp->inp_fport == fport &&
2307 		    inp->inp_lport == lport) {
2308 			/*
2309 			 * XXX We should be able to directly return
2310 			 * the inp here, without any checks.
2311 			 * Well unless both bound with SO_REUSEPORT?
2312 			 */
2313 			if (prison_flag(inp->inp_cred, PR_IP4))
2314 				return (inp);
2315 			if (tmpinp == NULL)
2316 				tmpinp = inp;
2317 		}
2318 	}
2319 	if (tmpinp != NULL)
2320 		return (tmpinp);
2321 
2322 	/*
2323 	 * Then look in lb group (for wildcard match).
2324 	 */
2325 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2326 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2327 		    fport, lookupflags, numa_domain);
2328 		if (inp != NULL)
2329 			return (inp);
2330 	}
2331 
2332 	/*
2333 	 * Then look for a wildcard match, if requested.
2334 	 */
2335 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2336 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2337 #ifdef INET6
2338 		struct inpcb *local_wild_mapped = NULL;
2339 #endif
2340 		struct inpcb *jail_wild = NULL;
2341 		int injail;
2342 
2343 		/*
2344 		 * Order of socket selection - we always prefer jails.
2345 		 *      1. jailed, non-wild.
2346 		 *      2. jailed, wild.
2347 		 *      3. non-jailed, non-wild.
2348 		 *      4. non-jailed, wild.
2349 		 */
2350 
2351 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2352 		    pcbinfo->ipi_hashmask)];
2353 		CK_LIST_FOREACH(inp, head, inp_hash) {
2354 #ifdef INET6
2355 			/* XXX inp locking */
2356 			if ((inp->inp_vflag & INP_IPV4) == 0)
2357 				continue;
2358 #endif
2359 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2360 			    inp->inp_lport != lport)
2361 				continue;
2362 
2363 			injail = prison_flag(inp->inp_cred, PR_IP4);
2364 			if (injail) {
2365 				if (prison_check_ip4_locked(
2366 				    inp->inp_cred->cr_prison, &laddr) != 0)
2367 					continue;
2368 			} else {
2369 				if (local_exact != NULL)
2370 					continue;
2371 			}
2372 
2373 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2374 				if (injail)
2375 					return (inp);
2376 				else
2377 					local_exact = inp;
2378 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2379 #ifdef INET6
2380 				/* XXX inp locking, NULL check */
2381 				if (inp->inp_vflag & INP_IPV6PROTO)
2382 					local_wild_mapped = inp;
2383 				else
2384 #endif
2385 					if (injail)
2386 						jail_wild = inp;
2387 					else
2388 						local_wild = inp;
2389 			}
2390 		} /* LIST_FOREACH */
2391 		if (jail_wild != NULL)
2392 			return (jail_wild);
2393 		if (local_exact != NULL)
2394 			return (local_exact);
2395 		if (local_wild != NULL)
2396 			return (local_wild);
2397 #ifdef INET6
2398 		if (local_wild_mapped != NULL)
2399 			return (local_wild_mapped);
2400 #endif
2401 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2402 
2403 	return (NULL);
2404 }
2405 
2406 /*
2407  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2408  * hash list lock, and will return the inpcb locked (i.e., requires
2409  * INPLOOKUP_LOCKPCB).
2410  */
2411 static struct inpcb *
2412 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2413     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2414     struct ifnet *ifp, uint8_t numa_domain)
2415 {
2416 	struct inpcb *inp;
2417 
2418 	smr_enter(pcbinfo->ipi_smr);
2419 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2420 	    lookupflags & INPLOOKUP_WILDCARD, ifp, numa_domain);
2421 	if (inp != NULL) {
2422 		if (__predict_false(inp_smr_lock(inp,
2423 		    (lookupflags & INPLOOKUP_LOCKMASK)) == false))
2424 			inp = NULL;
2425 	} else
2426 		smr_exit(pcbinfo->ipi_smr);
2427 
2428 	return (inp);
2429 }
2430 
2431 /*
2432  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2433  * from which a pre-calculated hash value may be extracted.
2434  */
2435 struct inpcb *
2436 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2437     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2438 {
2439 
2440 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2441 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2442 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2443 	    ("%s: LOCKPCB not set", __func__));
2444 
2445 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2446 	    lookupflags, ifp, M_NODOM));
2447 }
2448 
2449 struct inpcb *
2450 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2451     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2452     struct ifnet *ifp, struct mbuf *m)
2453 {
2454 
2455 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2456 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2457 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2458 	    ("%s: LOCKPCB not set", __func__));
2459 
2460 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2461 	    lookupflags, ifp, m->m_pkthdr.numa_domain));
2462 }
2463 #endif /* INET */
2464 
2465 /*
2466  * Insert PCB onto various hash lists.
2467  */
2468 int
2469 in_pcbinshash(struct inpcb *inp)
2470 {
2471 	struct inpcbhead *pcbhash;
2472 	struct inpcbporthead *pcbporthash;
2473 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2474 	struct inpcbport *phd;
2475 	int so_options;
2476 
2477 	INP_WLOCK_ASSERT(inp);
2478 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2479 
2480 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2481 	    ("in_pcbinshash: INP_INHASHLIST"));
2482 
2483 #ifdef INET6
2484 	if (inp->inp_vflag & INP_IPV6)
2485 		pcbhash = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2486 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2487 	else
2488 #endif
2489 		pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2490 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2491 
2492 	pcbporthash = &pcbinfo->ipi_porthashbase[
2493 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2494 
2495 	/*
2496 	 * Add entry to load balance group.
2497 	 * Only do this if SO_REUSEPORT_LB is set.
2498 	 */
2499 	so_options = inp_so_options(inp);
2500 	if (so_options & SO_REUSEPORT_LB) {
2501 		int ret = in_pcbinslbgrouphash(inp, M_NODOM);
2502 		if (ret) {
2503 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2504 			return (ret);
2505 		}
2506 	}
2507 
2508 	/*
2509 	 * Go through port list and look for a head for this lport.
2510 	 */
2511 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2512 		if (phd->phd_port == inp->inp_lport)
2513 			break;
2514 	}
2515 	/*
2516 	 * If none exists, malloc one and tack it on.
2517 	 */
2518 	if (phd == NULL) {
2519 		phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2520 		if (phd == NULL) {
2521 			return (ENOBUFS); /* XXX */
2522 		}
2523 		phd->phd_port = inp->inp_lport;
2524 		CK_LIST_INIT(&phd->phd_pcblist);
2525 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2526 	}
2527 	inp->inp_phd = phd;
2528 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2529 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2530 	inp->inp_flags |= INP_INHASHLIST;
2531 
2532 	return (0);
2533 }
2534 
2535 /*
2536  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2537  * changed. NOTE: This does not handle the case of the lport changing (the
2538  * hashed port list would have to be updated as well), so the lport must
2539  * not change after in_pcbinshash() has been called.
2540  *
2541  * XXXGL: a race between this function and SMR-protected hash iterator
2542  * will lead to iterator traversing a possibly wrong hash list. However,
2543  * this race should have been here since change from rwlock to epoch.
2544  */
2545 void
2546 in_pcbrehash(struct inpcb *inp)
2547 {
2548 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2549 	struct inpcbhead *head;
2550 
2551 	INP_WLOCK_ASSERT(inp);
2552 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2553 
2554 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2555 	    ("in_pcbrehash: !INP_INHASHLIST"));
2556 
2557 #ifdef INET6
2558 	if (inp->inp_vflag & INP_IPV6)
2559 		head = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2560 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2561 	else
2562 #endif
2563 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2564 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2565 
2566 	CK_LIST_REMOVE(inp, inp_hash);
2567 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2568 }
2569 
2570 /*
2571  * Check for alternatives when higher level complains
2572  * about service problems.  For now, invalidate cached
2573  * routing information.  If the route was created dynamically
2574  * (by a redirect), time to try a default gateway again.
2575  */
2576 void
2577 in_losing(struct inpcb *inp)
2578 {
2579 
2580 	RO_INVALIDATE_CACHE(&inp->inp_route);
2581 	return;
2582 }
2583 
2584 /*
2585  * A set label operation has occurred at the socket layer, propagate the
2586  * label change into the in_pcb for the socket.
2587  */
2588 void
2589 in_pcbsosetlabel(struct socket *so)
2590 {
2591 #ifdef MAC
2592 	struct inpcb *inp;
2593 
2594 	inp = sotoinpcb(so);
2595 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2596 
2597 	INP_WLOCK(inp);
2598 	SOCK_LOCK(so);
2599 	mac_inpcb_sosetlabel(so, inp);
2600 	SOCK_UNLOCK(so);
2601 	INP_WUNLOCK(inp);
2602 #endif
2603 }
2604 
2605 /*
2606  * ipport_tick runs once per second, determining if random port allocation
2607  * should be continued.  If more than ipport_randomcps ports have been
2608  * allocated in the last second, then we return to sequential port
2609  * allocation. We return to random allocation only once we drop below
2610  * ipport_randomcps for at least ipport_randomtime seconds.
2611  */
2612 static void
2613 ipport_tick(void *xtp)
2614 {
2615 	VNET_ITERATOR_DECL(vnet_iter);
2616 
2617 	VNET_LIST_RLOCK_NOSLEEP();
2618 	VNET_FOREACH(vnet_iter) {
2619 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2620 		if (V_ipport_tcpallocs - V_ipport_tcplastcount <=
2621 		    V_ipport_randomcps) {
2622 			if (V_ipport_stoprandom > 0)
2623 				V_ipport_stoprandom--;
2624 		} else
2625 			V_ipport_stoprandom = V_ipport_randomtime;
2626 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2627 		CURVNET_RESTORE();
2628 	}
2629 	VNET_LIST_RUNLOCK_NOSLEEP();
2630 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2631 }
2632 
2633 static void
2634 ip_fini(void *xtp)
2635 {
2636 
2637 	callout_stop(&ipport_tick_callout);
2638 }
2639 
2640 /*
2641  * The ipport_callout should start running at about the time we attach the
2642  * inet or inet6 domains.
2643  */
2644 static void
2645 ipport_tick_init(const void *unused __unused)
2646 {
2647 
2648 	/* Start ipport_tick. */
2649 	callout_init(&ipport_tick_callout, 1);
2650 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2651 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2652 		SHUTDOWN_PRI_DEFAULT);
2653 }
2654 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2655     ipport_tick_init, NULL);
2656 
2657 void
2658 inp_wlock(struct inpcb *inp)
2659 {
2660 
2661 	INP_WLOCK(inp);
2662 }
2663 
2664 void
2665 inp_wunlock(struct inpcb *inp)
2666 {
2667 
2668 	INP_WUNLOCK(inp);
2669 }
2670 
2671 void
2672 inp_rlock(struct inpcb *inp)
2673 {
2674 
2675 	INP_RLOCK(inp);
2676 }
2677 
2678 void
2679 inp_runlock(struct inpcb *inp)
2680 {
2681 
2682 	INP_RUNLOCK(inp);
2683 }
2684 
2685 #ifdef INVARIANT_SUPPORT
2686 void
2687 inp_lock_assert(struct inpcb *inp)
2688 {
2689 
2690 	INP_WLOCK_ASSERT(inp);
2691 }
2692 
2693 void
2694 inp_unlock_assert(struct inpcb *inp)
2695 {
2696 
2697 	INP_UNLOCK_ASSERT(inp);
2698 }
2699 #endif
2700 
2701 void
2702 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2703 {
2704 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2705 	    INPLOOKUP_WLOCKPCB);
2706 	struct inpcb *inp;
2707 
2708 	while ((inp = inp_next(&inpi)) != NULL)
2709 		func(inp, arg);
2710 }
2711 
2712 struct socket *
2713 inp_inpcbtosocket(struct inpcb *inp)
2714 {
2715 
2716 	INP_WLOCK_ASSERT(inp);
2717 	return (inp->inp_socket);
2718 }
2719 
2720 struct tcpcb *
2721 inp_inpcbtotcpcb(struct inpcb *inp)
2722 {
2723 
2724 	INP_WLOCK_ASSERT(inp);
2725 	return ((struct tcpcb *)inp->inp_ppcb);
2726 }
2727 
2728 int
2729 inp_ip_tos_get(const struct inpcb *inp)
2730 {
2731 
2732 	return (inp->inp_ip_tos);
2733 }
2734 
2735 void
2736 inp_ip_tos_set(struct inpcb *inp, int val)
2737 {
2738 
2739 	inp->inp_ip_tos = val;
2740 }
2741 
2742 void
2743 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2744     uint32_t *faddr, uint16_t *fp)
2745 {
2746 
2747 	INP_LOCK_ASSERT(inp);
2748 	*laddr = inp->inp_laddr.s_addr;
2749 	*faddr = inp->inp_faddr.s_addr;
2750 	*lp = inp->inp_lport;
2751 	*fp = inp->inp_fport;
2752 }
2753 
2754 struct inpcb *
2755 so_sotoinpcb(struct socket *so)
2756 {
2757 
2758 	return (sotoinpcb(so));
2759 }
2760 
2761 struct tcpcb *
2762 so_sototcpcb(struct socket *so)
2763 {
2764 
2765 	return (sototcpcb(so));
2766 }
2767 
2768 /*
2769  * Create an external-format (``xinpcb'') structure using the information in
2770  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2771  * reduce the spew of irrelevant information over this interface, to isolate
2772  * user code from changes in the kernel structure, and potentially to provide
2773  * information-hiding if we decide that some of this information should be
2774  * hidden from users.
2775  */
2776 void
2777 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2778 {
2779 
2780 	bzero(xi, sizeof(*xi));
2781 	xi->xi_len = sizeof(struct xinpcb);
2782 	if (inp->inp_socket)
2783 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2784 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2785 	xi->inp_gencnt = inp->inp_gencnt;
2786 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2787 	xi->inp_flow = inp->inp_flow;
2788 	xi->inp_flowid = inp->inp_flowid;
2789 	xi->inp_flowtype = inp->inp_flowtype;
2790 	xi->inp_flags = inp->inp_flags;
2791 	xi->inp_flags2 = inp->inp_flags2;
2792 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2793 	xi->in6p_cksum = inp->in6p_cksum;
2794 	xi->in6p_hops = inp->in6p_hops;
2795 	xi->inp_ip_tos = inp->inp_ip_tos;
2796 	xi->inp_vflag = inp->inp_vflag;
2797 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2798 	xi->inp_ip_p = inp->inp_ip_p;
2799 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2800 }
2801 
2802 int
2803 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2804     int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2805 {
2806 	struct sockopt sopt;
2807 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2808 	    INPLOOKUP_WLOCKPCB);
2809 	struct inpcb *inp;
2810 	struct sockopt_parameters *params;
2811 	struct socket *so;
2812 	int error;
2813 	char buf[1024];
2814 
2815 	if (req->oldptr != NULL || req->oldlen != 0)
2816 		return (EINVAL);
2817 	if (req->newptr == NULL)
2818 		return (EPERM);
2819 	if (req->newlen > sizeof(buf))
2820 		return (ENOMEM);
2821 	error = SYSCTL_IN(req, buf, req->newlen);
2822 	if (error != 0)
2823 		return (error);
2824 	if (req->newlen < sizeof(struct sockopt_parameters))
2825 		return (EINVAL);
2826 	params = (struct sockopt_parameters *)buf;
2827 	sopt.sopt_level = params->sop_level;
2828 	sopt.sopt_name = params->sop_optname;
2829 	sopt.sopt_dir = SOPT_SET;
2830 	sopt.sopt_val = params->sop_optval;
2831 	sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
2832 	sopt.sopt_td = NULL;
2833 #ifdef INET6
2834 	if (params->sop_inc.inc_flags & INC_ISIPV6) {
2835 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_laddr))
2836 			params->sop_inc.inc6_laddr.s6_addr16[1] =
2837 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2838 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_faddr))
2839 			params->sop_inc.inc6_faddr.s6_addr16[1] =
2840 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2841 	}
2842 #endif
2843 	if (params->sop_inc.inc_lport != htons(0)) {
2844 		if (params->sop_inc.inc_fport == htons(0))
2845 			inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
2846 			    pcbinfo->ipi_hashmask);
2847 		else
2848 #ifdef INET6
2849 			if (params->sop_inc.inc_flags & INC_ISIPV6)
2850 				inpi.hash = INP6_PCBHASH(
2851 				    &params->sop_inc.inc6_faddr,
2852 				    params->sop_inc.inc_lport,
2853 				    params->sop_inc.inc_fport,
2854 				    pcbinfo->ipi_hashmask);
2855 			else
2856 #endif
2857 				inpi.hash = INP_PCBHASH(
2858 				    &params->sop_inc.inc_faddr,
2859 				    params->sop_inc.inc_lport,
2860 				    params->sop_inc.inc_fport,
2861 				    pcbinfo->ipi_hashmask);
2862 	}
2863 	while ((inp = inp_next(&inpi)) != NULL)
2864 		if (inp->inp_gencnt == params->sop_id) {
2865 			if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2866 				INP_WUNLOCK(inp);
2867 				return (ECONNRESET);
2868 			}
2869 			so = inp->inp_socket;
2870 			KASSERT(so != NULL, ("inp_socket == NULL"));
2871 			soref(so);
2872 			error = (*ctloutput_set)(inp, &sopt);
2873 			sorele(so);
2874 			break;
2875 		}
2876 	if (inp == NULL)
2877 		error = ESRCH;
2878 	return (error);
2879 }
2880 
2881 #ifdef DDB
2882 static void
2883 db_print_indent(int indent)
2884 {
2885 	int i;
2886 
2887 	for (i = 0; i < indent; i++)
2888 		db_printf(" ");
2889 }
2890 
2891 static void
2892 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2893 {
2894 	char faddr_str[48], laddr_str[48];
2895 
2896 	db_print_indent(indent);
2897 	db_printf("%s at %p\n", name, inc);
2898 
2899 	indent += 2;
2900 
2901 #ifdef INET6
2902 	if (inc->inc_flags & INC_ISIPV6) {
2903 		/* IPv6. */
2904 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2905 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2906 	} else
2907 #endif
2908 	{
2909 		/* IPv4. */
2910 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2911 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2912 	}
2913 	db_print_indent(indent);
2914 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2915 	    ntohs(inc->inc_lport));
2916 	db_print_indent(indent);
2917 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2918 	    ntohs(inc->inc_fport));
2919 }
2920 
2921 static void
2922 db_print_inpflags(int inp_flags)
2923 {
2924 	int comma;
2925 
2926 	comma = 0;
2927 	if (inp_flags & INP_RECVOPTS) {
2928 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2929 		comma = 1;
2930 	}
2931 	if (inp_flags & INP_RECVRETOPTS) {
2932 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2933 		comma = 1;
2934 	}
2935 	if (inp_flags & INP_RECVDSTADDR) {
2936 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2937 		comma = 1;
2938 	}
2939 	if (inp_flags & INP_ORIGDSTADDR) {
2940 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2941 		comma = 1;
2942 	}
2943 	if (inp_flags & INP_HDRINCL) {
2944 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2945 		comma = 1;
2946 	}
2947 	if (inp_flags & INP_HIGHPORT) {
2948 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2949 		comma = 1;
2950 	}
2951 	if (inp_flags & INP_LOWPORT) {
2952 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2953 		comma = 1;
2954 	}
2955 	if (inp_flags & INP_ANONPORT) {
2956 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2957 		comma = 1;
2958 	}
2959 	if (inp_flags & INP_RECVIF) {
2960 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2961 		comma = 1;
2962 	}
2963 	if (inp_flags & INP_MTUDISC) {
2964 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2965 		comma = 1;
2966 	}
2967 	if (inp_flags & INP_RECVTTL) {
2968 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2969 		comma = 1;
2970 	}
2971 	if (inp_flags & INP_DONTFRAG) {
2972 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2973 		comma = 1;
2974 	}
2975 	if (inp_flags & INP_RECVTOS) {
2976 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2977 		comma = 1;
2978 	}
2979 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2980 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2981 		comma = 1;
2982 	}
2983 	if (inp_flags & IN6P_PKTINFO) {
2984 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2985 		comma = 1;
2986 	}
2987 	if (inp_flags & IN6P_HOPLIMIT) {
2988 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2989 		comma = 1;
2990 	}
2991 	if (inp_flags & IN6P_HOPOPTS) {
2992 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2993 		comma = 1;
2994 	}
2995 	if (inp_flags & IN6P_DSTOPTS) {
2996 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2997 		comma = 1;
2998 	}
2999 	if (inp_flags & IN6P_RTHDR) {
3000 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3001 		comma = 1;
3002 	}
3003 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3004 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3005 		comma = 1;
3006 	}
3007 	if (inp_flags & IN6P_TCLASS) {
3008 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3009 		comma = 1;
3010 	}
3011 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3012 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3013 		comma = 1;
3014 	}
3015 	if (inp_flags & INP_TIMEWAIT) {
3016 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3017 		comma  = 1;
3018 	}
3019 	if (inp_flags & INP_ONESBCAST) {
3020 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3021 		comma  = 1;
3022 	}
3023 	if (inp_flags & INP_DROPPED) {
3024 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3025 		comma  = 1;
3026 	}
3027 	if (inp_flags & INP_SOCKREF) {
3028 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3029 		comma  = 1;
3030 	}
3031 	if (inp_flags & IN6P_RFC2292) {
3032 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3033 		comma = 1;
3034 	}
3035 	if (inp_flags & IN6P_MTU) {
3036 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3037 		comma = 1;
3038 	}
3039 }
3040 
3041 static void
3042 db_print_inpvflag(u_char inp_vflag)
3043 {
3044 	int comma;
3045 
3046 	comma = 0;
3047 	if (inp_vflag & INP_IPV4) {
3048 		db_printf("%sINP_IPV4", comma ? ", " : "");
3049 		comma  = 1;
3050 	}
3051 	if (inp_vflag & INP_IPV6) {
3052 		db_printf("%sINP_IPV6", comma ? ", " : "");
3053 		comma  = 1;
3054 	}
3055 	if (inp_vflag & INP_IPV6PROTO) {
3056 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3057 		comma  = 1;
3058 	}
3059 }
3060 
3061 static void
3062 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3063 {
3064 
3065 	db_print_indent(indent);
3066 	db_printf("%s at %p\n", name, inp);
3067 
3068 	indent += 2;
3069 
3070 	db_print_indent(indent);
3071 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3072 
3073 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3074 
3075 	db_print_indent(indent);
3076 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3077 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3078 
3079 	db_print_indent(indent);
3080 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3081 	   inp->inp_label, inp->inp_flags);
3082 	db_print_inpflags(inp->inp_flags);
3083 	db_printf(")\n");
3084 
3085 	db_print_indent(indent);
3086 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3087 	    inp->inp_vflag);
3088 	db_print_inpvflag(inp->inp_vflag);
3089 	db_printf(")\n");
3090 
3091 	db_print_indent(indent);
3092 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3093 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3094 
3095 	db_print_indent(indent);
3096 #ifdef INET6
3097 	if (inp->inp_vflag & INP_IPV6) {
3098 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3099 		    "in6p_moptions: %p\n", inp->in6p_options,
3100 		    inp->in6p_outputopts, inp->in6p_moptions);
3101 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3102 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3103 		    inp->in6p_hops);
3104 	} else
3105 #endif
3106 	{
3107 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3108 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3109 		    inp->inp_options, inp->inp_moptions);
3110 	}
3111 
3112 	db_print_indent(indent);
3113 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3114 	    (uintmax_t)inp->inp_gencnt);
3115 }
3116 
3117 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3118 {
3119 	struct inpcb *inp;
3120 
3121 	if (!have_addr) {
3122 		db_printf("usage: show inpcb <addr>\n");
3123 		return;
3124 	}
3125 	inp = (struct inpcb *)addr;
3126 
3127 	db_print_inpcb(inp, "inpcb", 0);
3128 }
3129 #endif /* DDB */
3130 
3131 #ifdef RATELIMIT
3132 /*
3133  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3134  * if any.
3135  */
3136 int
3137 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3138 {
3139 	union if_snd_tag_modify_params params = {
3140 		.rate_limit.max_rate = max_pacing_rate,
3141 		.rate_limit.flags = M_NOWAIT,
3142 	};
3143 	struct m_snd_tag *mst;
3144 	int error;
3145 
3146 	mst = inp->inp_snd_tag;
3147 	if (mst == NULL)
3148 		return (EINVAL);
3149 
3150 	if (mst->sw->snd_tag_modify == NULL) {
3151 		error = EOPNOTSUPP;
3152 	} else {
3153 		error = mst->sw->snd_tag_modify(mst, &params);
3154 	}
3155 	return (error);
3156 }
3157 
3158 /*
3159  * Query existing TX rate limit based on the existing
3160  * "inp->inp_snd_tag", if any.
3161  */
3162 int
3163 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3164 {
3165 	union if_snd_tag_query_params params = { };
3166 	struct m_snd_tag *mst;
3167 	int error;
3168 
3169 	mst = inp->inp_snd_tag;
3170 	if (mst == NULL)
3171 		return (EINVAL);
3172 
3173 	if (mst->sw->snd_tag_query == NULL) {
3174 		error = EOPNOTSUPP;
3175 	} else {
3176 		error = mst->sw->snd_tag_query(mst, &params);
3177 		if (error == 0 && p_max_pacing_rate != NULL)
3178 			*p_max_pacing_rate = params.rate_limit.max_rate;
3179 	}
3180 	return (error);
3181 }
3182 
3183 /*
3184  * Query existing TX queue level based on the existing
3185  * "inp->inp_snd_tag", if any.
3186  */
3187 int
3188 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3189 {
3190 	union if_snd_tag_query_params params = { };
3191 	struct m_snd_tag *mst;
3192 	int error;
3193 
3194 	mst = inp->inp_snd_tag;
3195 	if (mst == NULL)
3196 		return (EINVAL);
3197 
3198 	if (mst->sw->snd_tag_query == NULL)
3199 		return (EOPNOTSUPP);
3200 
3201 	error = mst->sw->snd_tag_query(mst, &params);
3202 	if (error == 0 && p_txqueue_level != NULL)
3203 		*p_txqueue_level = params.rate_limit.queue_level;
3204 	return (error);
3205 }
3206 
3207 /*
3208  * Allocate a new TX rate limit send tag from the network interface
3209  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3210  */
3211 int
3212 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3213     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3214 
3215 {
3216 	union if_snd_tag_alloc_params params = {
3217 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3218 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3219 		.rate_limit.hdr.flowid = flowid,
3220 		.rate_limit.hdr.flowtype = flowtype,
3221 		.rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3222 		.rate_limit.max_rate = max_pacing_rate,
3223 		.rate_limit.flags = M_NOWAIT,
3224 	};
3225 	int error;
3226 
3227 	INP_WLOCK_ASSERT(inp);
3228 
3229 	/*
3230 	 * If there is already a send tag, or the INP is being torn
3231 	 * down, allocating a new send tag is not allowed. Else send
3232 	 * tags may leak.
3233 	 */
3234 	if (*st != NULL || (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0)
3235 		return (EINVAL);
3236 
3237 	error = m_snd_tag_alloc(ifp, &params, st);
3238 #ifdef INET
3239 	if (error == 0) {
3240 		counter_u64_add(rate_limit_set_ok, 1);
3241 		counter_u64_add(rate_limit_active, 1);
3242 	} else if (error != EOPNOTSUPP)
3243 		  counter_u64_add(rate_limit_alloc_fail, 1);
3244 #endif
3245 	return (error);
3246 }
3247 
3248 void
3249 in_pcbdetach_tag(struct m_snd_tag *mst)
3250 {
3251 
3252 	m_snd_tag_rele(mst);
3253 #ifdef INET
3254 	counter_u64_add(rate_limit_active, -1);
3255 #endif
3256 }
3257 
3258 /*
3259  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3260  * if any:
3261  */
3262 void
3263 in_pcbdetach_txrtlmt(struct inpcb *inp)
3264 {
3265 	struct m_snd_tag *mst;
3266 
3267 	INP_WLOCK_ASSERT(inp);
3268 
3269 	mst = inp->inp_snd_tag;
3270 	inp->inp_snd_tag = NULL;
3271 
3272 	if (mst == NULL)
3273 		return;
3274 
3275 	m_snd_tag_rele(mst);
3276 #ifdef INET
3277 	counter_u64_add(rate_limit_active, -1);
3278 #endif
3279 }
3280 
3281 int
3282 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3283 {
3284 	int error;
3285 
3286 	/*
3287 	 * If the existing send tag is for the wrong interface due to
3288 	 * a route change, first drop the existing tag.  Set the
3289 	 * CHANGED flag so that we will keep trying to allocate a new
3290 	 * tag if we fail to allocate one this time.
3291 	 */
3292 	if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3293 		in_pcbdetach_txrtlmt(inp);
3294 		inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3295 	}
3296 
3297 	/*
3298 	 * NOTE: When attaching to a network interface a reference is
3299 	 * made to ensure the network interface doesn't go away until
3300 	 * all ratelimit connections are gone. The network interface
3301 	 * pointers compared below represent valid network interfaces,
3302 	 * except when comparing towards NULL.
3303 	 */
3304 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3305 		error = 0;
3306 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3307 		if (inp->inp_snd_tag != NULL)
3308 			in_pcbdetach_txrtlmt(inp);
3309 		error = 0;
3310 	} else if (inp->inp_snd_tag == NULL) {
3311 		/*
3312 		 * In order to utilize packet pacing with RSS, we need
3313 		 * to wait until there is a valid RSS hash before we
3314 		 * can proceed:
3315 		 */
3316 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3317 			error = EAGAIN;
3318 		} else {
3319 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3320 			    mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3321 		}
3322 	} else {
3323 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3324 	}
3325 	if (error == 0 || error == EOPNOTSUPP)
3326 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3327 
3328 	return (error);
3329 }
3330 
3331 /*
3332  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3333  * is set in the fast path and will attach/detach/modify the TX rate
3334  * limit send tag based on the socket's so_max_pacing_rate value.
3335  */
3336 void
3337 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3338 {
3339 	struct socket *socket;
3340 	uint32_t max_pacing_rate;
3341 	bool did_upgrade;
3342 	int error;
3343 
3344 	if (inp == NULL)
3345 		return;
3346 
3347 	socket = inp->inp_socket;
3348 	if (socket == NULL)
3349 		return;
3350 
3351 	if (!INP_WLOCKED(inp)) {
3352 		/*
3353 		 * NOTE: If the write locking fails, we need to bail
3354 		 * out and use the non-ratelimited ring for the
3355 		 * transmit until there is a new chance to get the
3356 		 * write lock.
3357 		 */
3358 		if (!INP_TRY_UPGRADE(inp))
3359 			return;
3360 		did_upgrade = 1;
3361 	} else {
3362 		did_upgrade = 0;
3363 	}
3364 
3365 	/*
3366 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3367 	 * because atomic updates are not required since the variable
3368 	 * is checked at every mbuf we send. It is assumed that the
3369 	 * variable read itself will be atomic.
3370 	 */
3371 	max_pacing_rate = socket->so_max_pacing_rate;
3372 
3373 	error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3374 
3375 	if (did_upgrade)
3376 		INP_DOWNGRADE(inp);
3377 }
3378 
3379 /*
3380  * Track route changes for TX rate limiting.
3381  */
3382 void
3383 in_pcboutput_eagain(struct inpcb *inp)
3384 {
3385 	bool did_upgrade;
3386 
3387 	if (inp == NULL)
3388 		return;
3389 
3390 	if (inp->inp_snd_tag == NULL)
3391 		return;
3392 
3393 	if (!INP_WLOCKED(inp)) {
3394 		/*
3395 		 * NOTE: If the write locking fails, we need to bail
3396 		 * out and use the non-ratelimited ring for the
3397 		 * transmit until there is a new chance to get the
3398 		 * write lock.
3399 		 */
3400 		if (!INP_TRY_UPGRADE(inp))
3401 			return;
3402 		did_upgrade = 1;
3403 	} else {
3404 		did_upgrade = 0;
3405 	}
3406 
3407 	/* detach rate limiting */
3408 	in_pcbdetach_txrtlmt(inp);
3409 
3410 	/* make sure new mbuf send tag allocation is made */
3411 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3412 
3413 	if (did_upgrade)
3414 		INP_DOWNGRADE(inp);
3415 }
3416 
3417 #ifdef INET
3418 static void
3419 rl_init(void *st)
3420 {
3421 	rate_limit_new = counter_u64_alloc(M_WAITOK);
3422 	rate_limit_chg = counter_u64_alloc(M_WAITOK);
3423 	rate_limit_active = counter_u64_alloc(M_WAITOK);
3424 	rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3425 	rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3426 }
3427 
3428 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3429 #endif
3430 #endif /* RATELIMIT */
3431