xref: /freebsd/sys/netinet/in_pcb.c (revision 3dd5524264095ed8612c28908e13f80668eff2f9)
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  * Copyright (c) 2021-2022 Gleb Smirnoff <glebius@FreeBSD.org>
9  * All rights reserved.
10  *
11  * Portions of this software were developed by Robert N. M. Watson under
12  * contract to Juniper Networks, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_ipsec.h"
46 #include "opt_inet.h"
47 #include "opt_inet6.h"
48 #include "opt_ratelimit.h"
49 #include "opt_route.h"
50 #include "opt_rss.h"
51 
52 #include <sys/param.h>
53 #include <sys/hash.h>
54 #include <sys/systm.h>
55 #include <sys/libkern.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.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_private.h>
83 #include <net/if_types.h>
84 #include <net/if_llatbl.h>
85 #include <net/route.h>
86 #include <net/rss_config.h>
87 #include <net/vnet.h>
88 
89 #if defined(INET) || defined(INET6)
90 #include <netinet/in.h>
91 #include <netinet/in_pcb.h>
92 #include <netinet/in_pcb_var.h>
93 #include <netinet/tcp.h>
94 #ifdef INET
95 #include <netinet/in_var.h>
96 #include <netinet/in_fib.h>
97 #endif
98 #include <netinet/ip_var.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #endif /* INET6 */
105 #include <net/route/nhop.h>
106 #endif
107 
108 #include <netipsec/ipsec_support.h>
109 
110 #include <security/mac/mac_framework.h>
111 
112 #define	INPCBLBGROUP_SIZMIN	8
113 #define	INPCBLBGROUP_SIZMAX	256
114 #define	INP_FREED	0x00000200	/* See in_pcb.h. */
115 
116 /*
117  * These configure the range of local port addresses assigned to
118  * "unspecified" outgoing connections/packets/whatever.
119  */
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
126 
127 /*
128  * Reserved ports accessible only to root. There are significant
129  * security considerations that must be accounted for when changing these,
130  * but the security benefits can be great. Please be careful.
131  */
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
134 
135 /* Enable random ephemeral port allocation by default. */
136 VNET_DEFINE(int, ipport_randomized) = 1;
137 
138 #ifdef INET
139 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
140 			    struct in_addr faddr, u_int fport_arg,
141 			    struct in_addr laddr, u_int lport_arg,
142 			    int lookupflags, uint8_t numa_domain);
143 
144 #define RANGECHK(var, min, max) \
145 	if ((var) < (min)) { (var) = (min); } \
146 	else if ((var) > (max)) { (var) = (max); }
147 
148 static int
149 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
150 {
151 	int error;
152 
153 	error = sysctl_handle_int(oidp, arg1, arg2, req);
154 	if (error == 0) {
155 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
156 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
157 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
158 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
159 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
160 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
161 	}
162 	return (error);
163 }
164 
165 #undef RANGECHK
166 
167 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
168     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
169     "IP Ports");
170 
171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
172     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
173     &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
174     "");
175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
176     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
177     &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
178     "");
179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
180     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
181     &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
182     "");
183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
184     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
185     &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
186     "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
188     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
189     &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
190     "");
191 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
192     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
193     &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
194     "");
195 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
196 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
197 	&VNET_NAME(ipport_reservedhigh), 0, "");
198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
199 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
201 	CTLFLAG_VNET | CTLFLAG_RW,
202 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
203 
204 #ifdef RATELIMIT
205 counter_u64_t rate_limit_new;
206 counter_u64_t rate_limit_chg;
207 counter_u64_t rate_limit_active;
208 counter_u64_t rate_limit_alloc_fail;
209 counter_u64_t rate_limit_set_ok;
210 
211 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
212     "IP Rate Limiting");
213 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
214     &rate_limit_active, "Active rate limited connections");
215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
216    &rate_limit_alloc_fail, "Rate limited connection failures");
217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
218    &rate_limit_set_ok, "Rate limited setting succeeded");
219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD,
220    &rate_limit_new, "Total Rate limit new attempts");
221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD,
222    &rate_limit_chg, "Total Rate limited change attempts");
223 
224 #endif /* RATELIMIT */
225 
226 #endif /* INET */
227 
228 VNET_DEFINE(uint32_t, in_pcbhashseed);
229 static void
230 in_pcbhashseed_init(void)
231 {
232 
233 	V_in_pcbhashseed = arc4random();
234 }
235 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
236     in_pcbhashseed_init, 0);
237 
238 static void in_pcbremhash(struct inpcb *);
239 
240 /*
241  * in_pcb.c: manage the Protocol Control Blocks.
242  *
243  * NOTE: It is assumed that most of these functions will be called with
244  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
245  * functions often modify hash chains or addresses in pcbs.
246  */
247 
248 static struct inpcblbgroup *
249 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred,
250     u_char vflag, uint16_t port, const union in_dependaddr *addr, int size,
251     uint8_t numa_domain)
252 {
253 	struct inpcblbgroup *grp;
254 	size_t bytes;
255 
256 	bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
257 	grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
258 	if (grp == NULL)
259 		return (NULL);
260 	grp->il_cred = crhold(cred);
261 	grp->il_vflag = vflag;
262 	grp->il_lport = port;
263 	grp->il_numa_domain = numa_domain;
264 	grp->il_dependladdr = *addr;
265 	grp->il_inpsiz = size;
266 	CK_LIST_INSERT_HEAD(hdr, grp, il_list);
267 	return (grp);
268 }
269 
270 static void
271 in_pcblbgroup_free_deferred(epoch_context_t ctx)
272 {
273 	struct inpcblbgroup *grp;
274 
275 	grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
276 	crfree(grp->il_cred);
277 	free(grp, M_PCB);
278 }
279 
280 static void
281 in_pcblbgroup_free(struct inpcblbgroup *grp)
282 {
283 
284 	CK_LIST_REMOVE(grp, il_list);
285 	NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
286 }
287 
288 static struct inpcblbgroup *
289 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
290     struct inpcblbgroup *old_grp, int size)
291 {
292 	struct inpcblbgroup *grp;
293 	int i;
294 
295 	grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag,
296 	    old_grp->il_lport, &old_grp->il_dependladdr, size,
297 	    old_grp->il_numa_domain);
298 	if (grp == NULL)
299 		return (NULL);
300 
301 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
302 	    ("invalid new local group size %d and old local group count %d",
303 	     grp->il_inpsiz, old_grp->il_inpcnt));
304 
305 	for (i = 0; i < old_grp->il_inpcnt; ++i)
306 		grp->il_inp[i] = old_grp->il_inp[i];
307 	grp->il_inpcnt = old_grp->il_inpcnt;
308 	in_pcblbgroup_free(old_grp);
309 	return (grp);
310 }
311 
312 /*
313  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
314  * and shrink group if possible.
315  */
316 static void
317 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
318     int i)
319 {
320 	struct inpcblbgroup *grp, *new_grp;
321 
322 	grp = *grpp;
323 	for (; i + 1 < grp->il_inpcnt; ++i)
324 		grp->il_inp[i] = grp->il_inp[i + 1];
325 	grp->il_inpcnt--;
326 
327 	if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
328 	    grp->il_inpcnt <= grp->il_inpsiz / 4) {
329 		/* Shrink this group. */
330 		new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
331 		if (new_grp != NULL)
332 			*grpp = new_grp;
333 	}
334 }
335 
336 /*
337  * Add PCB to load balance group for SO_REUSEPORT_LB option.
338  */
339 static int
340 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain)
341 {
342 	const static struct timeval interval = { 60, 0 };
343 	static struct timeval lastprint;
344 	struct inpcbinfo *pcbinfo;
345 	struct inpcblbgrouphead *hdr;
346 	struct inpcblbgroup *grp;
347 	uint32_t idx;
348 
349 	pcbinfo = inp->inp_pcbinfo;
350 
351 	INP_WLOCK_ASSERT(inp);
352 	INP_HASH_WLOCK_ASSERT(pcbinfo);
353 
354 #ifdef INET6
355 	/*
356 	 * Don't allow IPv4 mapped INET6 wild socket.
357 	 */
358 	if ((inp->inp_vflag & INP_IPV4) &&
359 	    inp->inp_laddr.s_addr == INADDR_ANY &&
360 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
361 		return (0);
362 	}
363 #endif
364 
365 	idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
366 	hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
367 	CK_LIST_FOREACH(grp, hdr, il_list) {
368 		if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison &&
369 		    grp->il_vflag == inp->inp_vflag &&
370 		    grp->il_lport == inp->inp_lport &&
371 		    grp->il_numa_domain == numa_domain &&
372 		    memcmp(&grp->il_dependladdr,
373 		    &inp->inp_inc.inc_ie.ie_dependladdr,
374 		    sizeof(grp->il_dependladdr)) == 0) {
375 			break;
376 		}
377 	}
378 	if (grp == NULL) {
379 		/* Create new load balance group. */
380 		grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag,
381 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
382 		    INPCBLBGROUP_SIZMIN, numa_domain);
383 		if (grp == NULL)
384 			return (ENOBUFS);
385 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
386 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
387 			if (ratecheck(&lastprint, &interval))
388 				printf("lb group port %d, limit reached\n",
389 				    ntohs(grp->il_lport));
390 			return (0);
391 		}
392 
393 		/* Expand this local group. */
394 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
395 		if (grp == NULL)
396 			return (ENOBUFS);
397 	}
398 
399 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
400 	    ("invalid local group size %d and count %d", grp->il_inpsiz,
401 	    grp->il_inpcnt));
402 
403 	grp->il_inp[grp->il_inpcnt] = inp;
404 	grp->il_inpcnt++;
405 	return (0);
406 }
407 
408 /*
409  * Remove PCB from load balance group.
410  */
411 static void
412 in_pcbremlbgrouphash(struct inpcb *inp)
413 {
414 	struct inpcbinfo *pcbinfo;
415 	struct inpcblbgrouphead *hdr;
416 	struct inpcblbgroup *grp;
417 	int i;
418 
419 	pcbinfo = inp->inp_pcbinfo;
420 
421 	INP_WLOCK_ASSERT(inp);
422 	INP_HASH_WLOCK_ASSERT(pcbinfo);
423 
424 	hdr = &pcbinfo->ipi_lbgrouphashbase[
425 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
426 	CK_LIST_FOREACH(grp, hdr, il_list) {
427 		for (i = 0; i < grp->il_inpcnt; ++i) {
428 			if (grp->il_inp[i] != inp)
429 				continue;
430 
431 			if (grp->il_inpcnt == 1) {
432 				/* We are the last, free this local group. */
433 				in_pcblbgroup_free(grp);
434 			} else {
435 				/* Pull up inpcbs, shrink group if possible. */
436 				in_pcblbgroup_reorder(hdr, &grp, i);
437 			}
438 			return;
439 		}
440 	}
441 }
442 
443 int
444 in_pcblbgroup_numa(struct inpcb *inp, int arg)
445 {
446 	struct inpcbinfo *pcbinfo;
447 	struct inpcblbgrouphead *hdr;
448 	struct inpcblbgroup *grp;
449 	int err, i;
450 	uint8_t numa_domain;
451 
452 	switch (arg) {
453 	case TCP_REUSPORT_LB_NUMA_NODOM:
454 		numa_domain = M_NODOM;
455 		break;
456 	case TCP_REUSPORT_LB_NUMA_CURDOM:
457 		numa_domain = PCPU_GET(domain);
458 		break;
459 	default:
460 		if (arg < 0 || arg >= vm_ndomains)
461 			return (EINVAL);
462 		numa_domain = arg;
463 	}
464 
465 	err = 0;
466 	pcbinfo = inp->inp_pcbinfo;
467 	INP_WLOCK_ASSERT(inp);
468 	INP_HASH_WLOCK(pcbinfo);
469 	hdr = &pcbinfo->ipi_lbgrouphashbase[
470 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
471 	CK_LIST_FOREACH(grp, hdr, il_list) {
472 		for (i = 0; i < grp->il_inpcnt; ++i) {
473 			if (grp->il_inp[i] != inp)
474 				continue;
475 
476 			if (grp->il_numa_domain == numa_domain) {
477 				goto abort_with_hash_wlock;
478 			}
479 
480 			/* Remove it from the old group. */
481 			in_pcbremlbgrouphash(inp);
482 
483 			/* Add it to the new group based on numa domain. */
484 			in_pcbinslbgrouphash(inp, numa_domain);
485 			goto abort_with_hash_wlock;
486 		}
487 	}
488 	err = ENOENT;
489 abort_with_hash_wlock:
490 	INP_HASH_WUNLOCK(pcbinfo);
491 	return (err);
492 }
493 
494 /* Make sure it is safe to use hashinit(9) on CK_LIST. */
495 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb)));
496 
497 /*
498  * Initialize an inpcbinfo - a per-VNET instance of connections db.
499  */
500 void
501 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor,
502     u_int hash_nelements, u_int porthash_nelements)
503 {
504 
505 	mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF);
506 	mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name,
507 	    NULL, MTX_DEF);
508 #ifdef VIMAGE
509 	pcbinfo->ipi_vnet = curvnet;
510 #endif
511 	CK_LIST_INIT(&pcbinfo->ipi_listhead);
512 	pcbinfo->ipi_count = 0;
513 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
514 	    &pcbinfo->ipi_hashmask);
515 	porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
516 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
517 	    &pcbinfo->ipi_porthashmask);
518 	pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
519 	    &pcbinfo->ipi_lbgrouphashmask);
520 	pcbinfo->ipi_zone = pcbstor->ips_zone;
521 	pcbinfo->ipi_portzone = pcbstor->ips_portzone;
522 	pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
523 }
524 
525 /*
526  * Destroy an inpcbinfo.
527  */
528 void
529 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
530 {
531 
532 	KASSERT(pcbinfo->ipi_count == 0,
533 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
534 
535 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
536 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
537 	    pcbinfo->ipi_porthashmask);
538 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
539 	    pcbinfo->ipi_lbgrouphashmask);
540 	mtx_destroy(&pcbinfo->ipi_hash_lock);
541 	mtx_destroy(&pcbinfo->ipi_lock);
542 }
543 
544 /*
545  * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
546  */
547 static void inpcb_dtor(void *, int, void *);
548 static void inpcb_fini(void *, int);
549 void
550 in_pcbstorage_init(void *arg)
551 {
552 	struct inpcbstorage *pcbstor = arg;
553 
554 	pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
555 	    pcbstor->ips_size, NULL, inpcb_dtor, pcbstor->ips_pcbinit,
556 	    inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR);
557 	pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
558 	    sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
559 	uma_zone_set_smr(pcbstor->ips_portzone,
560 	    uma_zone_get_smr(pcbstor->ips_zone));
561 }
562 
563 /*
564  * Destroy a pcbstorage - used by unloadable protocols.
565  */
566 void
567 in_pcbstorage_destroy(void *arg)
568 {
569 	struct inpcbstorage *pcbstor = arg;
570 
571 	uma_zdestroy(pcbstor->ips_zone);
572 	uma_zdestroy(pcbstor->ips_portzone);
573 }
574 
575 /*
576  * Allocate a PCB and associate it with the socket.
577  * On success return with the PCB locked.
578  */
579 int
580 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
581 {
582 	struct inpcb *inp;
583 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
584 	int error;
585 #endif
586 
587 	inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
588 	if (inp == NULL)
589 		return (ENOBUFS);
590 	bzero(&inp->inp_start_zero, inp_zero_size);
591 #ifdef NUMA
592 	inp->inp_numa_domain = M_NODOM;
593 #endif
594 	inp->inp_pcbinfo = pcbinfo;
595 	inp->inp_socket = so;
596 	inp->inp_cred = crhold(so->so_cred);
597 	inp->inp_inc.inc_fibnum = so->so_fibnum;
598 #ifdef MAC
599 	error = mac_inpcb_init(inp, M_NOWAIT);
600 	if (error != 0)
601 		goto out;
602 	mac_inpcb_create(so, inp);
603 #endif
604 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
605 	error = ipsec_init_pcbpolicy(inp);
606 	if (error != 0) {
607 #ifdef MAC
608 		mac_inpcb_destroy(inp);
609 #endif
610 		goto out;
611 	}
612 #endif /*IPSEC*/
613 #ifdef INET6
614 	if (INP_SOCKAF(so) == AF_INET6) {
615 		inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6;
616 		if (V_ip6_v6only)
617 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
618 #ifdef INET
619 		else
620 			inp->inp_vflag |= INP_IPV4;
621 #endif
622 		if (V_ip6_auto_flowlabel)
623 			inp->inp_flags |= IN6P_AUTOFLOWLABEL;
624 		inp->in6p_hops = -1;	/* use kernel default */
625 	}
626 #endif
627 #if defined(INET) && defined(INET6)
628 	else
629 #endif
630 #ifdef INET
631 		inp->inp_vflag |= INP_IPV4;
632 #endif
633 	/*
634 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
635 	 * to be cleaned up.
636 	 */
637 	inp->inp_route.ro_flags = RT_LLE_CACHE;
638 	refcount_init(&inp->inp_refcount, 1);   /* Reference from socket. */
639 	INP_WLOCK(inp);
640 	INP_INFO_WLOCK(pcbinfo);
641 	pcbinfo->ipi_count++;
642 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
643 	CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
644 	INP_INFO_WUNLOCK(pcbinfo);
645 	so->so_pcb = inp;
646 
647 	return (0);
648 
649 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
650 out:
651 	uma_zfree_smr(pcbinfo->ipi_zone, inp);
652 	return (error);
653 #endif
654 }
655 
656 #ifdef INET
657 int
658 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred)
659 {
660 	int anonport, error;
661 
662 	KASSERT(sin == NULL || sin->sin_family == AF_INET,
663 	    ("%s: invalid address family for %p", __func__, sin));
664 	KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in),
665 	    ("%s: invalid address length for %p", __func__, sin));
666 	INP_WLOCK_ASSERT(inp);
667 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
668 
669 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
670 		return (EINVAL);
671 	anonport = sin == NULL || sin->sin_port == 0;
672 	error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr,
673 	    &inp->inp_lport, cred);
674 	if (error)
675 		return (error);
676 	if (in_pcbinshash(inp) != 0) {
677 		inp->inp_laddr.s_addr = INADDR_ANY;
678 		inp->inp_lport = 0;
679 		return (EAGAIN);
680 	}
681 	if (anonport)
682 		inp->inp_flags |= INP_ANONPORT;
683 	return (0);
684 }
685 #endif
686 
687 #if defined(INET) || defined(INET6)
688 /*
689  * Assign a local port like in_pcb_lport(), but also used with connect()
690  * and a foreign address and port.  If fsa is non-NULL, choose a local port
691  * that is unused with those, otherwise one that is completely unused.
692  * lsa can be NULL for IPv6.
693  */
694 int
695 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
696     struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
697 {
698 	struct inpcbinfo *pcbinfo;
699 	struct inpcb *tmpinp;
700 	unsigned short *lastport;
701 	int count, error;
702 	u_short aux, first, last, lport;
703 #ifdef INET
704 	struct in_addr laddr, faddr;
705 #endif
706 #ifdef INET6
707 	struct in6_addr *laddr6, *faddr6;
708 #endif
709 
710 	pcbinfo = inp->inp_pcbinfo;
711 
712 	/*
713 	 * Because no actual state changes occur here, a global write lock on
714 	 * the pcbinfo isn't required.
715 	 */
716 	INP_LOCK_ASSERT(inp);
717 	INP_HASH_LOCK_ASSERT(pcbinfo);
718 
719 	if (inp->inp_flags & INP_HIGHPORT) {
720 		first = V_ipport_hifirstauto;	/* sysctl */
721 		last  = V_ipport_hilastauto;
722 		lastport = &pcbinfo->ipi_lasthi;
723 	} else if (inp->inp_flags & INP_LOWPORT) {
724 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
725 		if (error)
726 			return (error);
727 		first = V_ipport_lowfirstauto;	/* 1023 */
728 		last  = V_ipport_lowlastauto;	/* 600 */
729 		lastport = &pcbinfo->ipi_lastlow;
730 	} else {
731 		first = V_ipport_firstauto;	/* sysctl */
732 		last  = V_ipport_lastauto;
733 		lastport = &pcbinfo->ipi_lastport;
734 	}
735 
736 	/*
737 	 * Instead of having two loops further down counting up or down
738 	 * make sure that first is always <= last and go with only one
739 	 * code path implementing all logic.
740 	 */
741 	if (first > last) {
742 		aux = first;
743 		first = last;
744 		last = aux;
745 	}
746 
747 #ifdef INET
748 	laddr.s_addr = INADDR_ANY;	/* used by INET6+INET below too */
749 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
750 		if (lsa != NULL)
751 			laddr = ((struct sockaddr_in *)lsa)->sin_addr;
752 		if (fsa != NULL)
753 			faddr = ((struct sockaddr_in *)fsa)->sin_addr;
754 	}
755 #endif
756 #ifdef INET6
757 	laddr6 = NULL;
758 	if ((inp->inp_vflag & INP_IPV6) != 0) {
759 		if (lsa != NULL)
760 			laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
761 		if (fsa != NULL)
762 			faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
763 	}
764 #endif
765 
766 	tmpinp = NULL;
767 	lport = *lportp;
768 
769 	if (V_ipport_randomized)
770 		*lastport = first + (arc4random() % (last - first));
771 
772 	count = last - first;
773 
774 	do {
775 		if (count-- < 0)	/* completely used? */
776 			return (EADDRNOTAVAIL);
777 		++*lastport;
778 		if (*lastport < first || *lastport > last)
779 			*lastport = first;
780 		lport = htons(*lastport);
781 
782 		if (fsa != NULL) {
783 #ifdef INET
784 			if (lsa->sa_family == AF_INET) {
785 				tmpinp = in_pcblookup_hash_locked(pcbinfo,
786 				    faddr, fport, laddr, lport, lookupflags,
787 				    M_NODOM);
788 			}
789 #endif
790 #ifdef INET6
791 			if (lsa->sa_family == AF_INET6) {
792 				tmpinp = in6_pcblookup_hash_locked(pcbinfo,
793 				    faddr6, fport, laddr6, lport, lookupflags,
794 				    M_NODOM);
795 			}
796 #endif
797 		} else {
798 #ifdef INET6
799 			if ((inp->inp_vflag & INP_IPV6) != 0) {
800 				tmpinp = in6_pcblookup_local(pcbinfo,
801 				    &inp->in6p_laddr, lport, lookupflags, cred);
802 #ifdef INET
803 				if (tmpinp == NULL &&
804 				    (inp->inp_vflag & INP_IPV4))
805 					tmpinp = in_pcblookup_local(pcbinfo,
806 					    laddr, lport, lookupflags, cred);
807 #endif
808 			}
809 #endif
810 #if defined(INET) && defined(INET6)
811 			else
812 #endif
813 #ifdef INET
814 				tmpinp = in_pcblookup_local(pcbinfo, laddr,
815 				    lport, lookupflags, cred);
816 #endif
817 		}
818 	} while (tmpinp != NULL);
819 
820 	*lportp = lport;
821 
822 	return (0);
823 }
824 
825 /*
826  * Select a local port (number) to use.
827  */
828 int
829 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
830     struct ucred *cred, int lookupflags)
831 {
832 	struct sockaddr_in laddr;
833 
834 	if (laddrp) {
835 		bzero(&laddr, sizeof(laddr));
836 		laddr.sin_family = AF_INET;
837 		laddr.sin_addr = *laddrp;
838 	}
839 	return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
840 	    NULL, lportp, NULL, 0, cred, lookupflags));
841 }
842 
843 /*
844  * Return cached socket options.
845  */
846 int
847 inp_so_options(const struct inpcb *inp)
848 {
849 	int so_options;
850 
851 	so_options = 0;
852 
853 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
854 		so_options |= SO_REUSEPORT_LB;
855 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
856 		so_options |= SO_REUSEPORT;
857 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
858 		so_options |= SO_REUSEADDR;
859 	return (so_options);
860 }
861 #endif /* INET || INET6 */
862 
863 /*
864  * Check if a new BINDMULTI socket is allowed to be created.
865  *
866  * ni points to the new inp.
867  * oi points to the existing inp.
868  *
869  * This checks whether the existing inp also has BINDMULTI and
870  * whether the credentials match.
871  */
872 int
873 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
874 {
875 	/* Check permissions match */
876 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
877 	    (ni->inp_cred->cr_uid !=
878 	    oi->inp_cred->cr_uid))
879 		return (0);
880 
881 	/* Check the existing inp has BINDMULTI set */
882 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
883 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
884 		return (0);
885 
886 	/*
887 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
888 	 * it is and it matches the checks.
889 	 */
890 	return (1);
891 }
892 
893 #ifdef INET
894 /*
895  * Set up a bind operation on a PCB, performing port allocation
896  * as required, but do not actually modify the PCB. Callers can
897  * either complete the bind by setting inp_laddr/inp_lport and
898  * calling in_pcbinshash(), or they can just use the resulting
899  * port and address to authorise the sending of a once-off packet.
900  *
901  * On error, the values of *laddrp and *lportp are not changed.
902  */
903 int
904 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp,
905     u_short *lportp, struct ucred *cred)
906 {
907 	struct socket *so = inp->inp_socket;
908 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
909 	struct in_addr laddr;
910 	u_short lport = 0;
911 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
912 	int error;
913 
914 	/*
915 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
916 	 * so that we don't have to add to the (already messy) code below.
917 	 */
918 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
919 
920 	/*
921 	 * No state changes, so read locks are sufficient here.
922 	 */
923 	INP_LOCK_ASSERT(inp);
924 	INP_HASH_LOCK_ASSERT(pcbinfo);
925 
926 	laddr.s_addr = *laddrp;
927 	if (sin != NULL && laddr.s_addr != INADDR_ANY)
928 		return (EINVAL);
929 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
930 		lookupflags = INPLOOKUP_WILDCARD;
931 	if (sin == NULL) {
932 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
933 			return (error);
934 	} else {
935 		KASSERT(sin->sin_family == AF_INET,
936 		    ("%s: invalid family for address %p", __func__, sin));
937 		KASSERT(sin->sin_len == sizeof(*sin),
938 		    ("%s: invalid length for address %p", __func__, sin));
939 
940 		error = prison_local_ip4(cred, &sin->sin_addr);
941 		if (error)
942 			return (error);
943 		if (sin->sin_port != *lportp) {
944 			/* Don't allow the port to change. */
945 			if (*lportp != 0)
946 				return (EINVAL);
947 			lport = sin->sin_port;
948 		}
949 		/* NB: lport is left as 0 if the port isn't being changed. */
950 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
951 			/*
952 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
953 			 * allow complete duplication of binding if
954 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
955 			 * and a multicast address is bound on both
956 			 * new and duplicated sockets.
957 			 */
958 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
959 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
960 			/*
961 			 * XXX: How to deal with SO_REUSEPORT_LB here?
962 			 * Treat same as SO_REUSEPORT for now.
963 			 */
964 			if ((so->so_options &
965 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
966 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
967 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
968 			sin->sin_port = 0;		/* yech... */
969 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
970 			/*
971 			 * Is the address a local IP address?
972 			 * If INP_BINDANY is set, then the socket may be bound
973 			 * to any endpoint address, local or not.
974 			 */
975 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
976 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
977 				return (EADDRNOTAVAIL);
978 		}
979 		laddr = sin->sin_addr;
980 		if (lport) {
981 			struct inpcb *t;
982 
983 			/* GROSS */
984 			if (ntohs(lport) <= V_ipport_reservedhigh &&
985 			    ntohs(lport) >= V_ipport_reservedlow &&
986 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
987 				return (EACCES);
988 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
989 			    priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
990 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
991 				    lport, INPLOOKUP_WILDCARD, cred);
992 	/*
993 	 * XXX
994 	 * This entire block sorely needs a rewrite.
995 	 */
996 				if (t &&
997 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
998 				    (so->so_type != SOCK_STREAM ||
999 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
1000 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
1001 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
1002 				     (t->inp_flags2 & INP_REUSEPORT) ||
1003 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
1004 				    (inp->inp_cred->cr_uid !=
1005 				     t->inp_cred->cr_uid))
1006 					return (EADDRINUSE);
1007 
1008 				/*
1009 				 * If the socket is a BINDMULTI socket, then
1010 				 * the credentials need to match and the
1011 				 * original socket also has to have been bound
1012 				 * with BINDMULTI.
1013 				 */
1014 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1015 					return (EADDRINUSE);
1016 			}
1017 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
1018 			    lport, lookupflags, cred);
1019 			if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
1020 			    (reuseport & inp_so_options(t)) == 0 &&
1021 			    (reuseport_lb & inp_so_options(t)) == 0) {
1022 #ifdef INET6
1023 				if (ntohl(sin->sin_addr.s_addr) !=
1024 				    INADDR_ANY ||
1025 				    ntohl(t->inp_laddr.s_addr) !=
1026 				    INADDR_ANY ||
1027 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
1028 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
1029 #endif
1030 						return (EADDRINUSE);
1031 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1032 					return (EADDRINUSE);
1033 			}
1034 		}
1035 	}
1036 	if (*lportp != 0)
1037 		lport = *lportp;
1038 	if (lport == 0) {
1039 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1040 		if (error != 0)
1041 			return (error);
1042 	}
1043 	*laddrp = laddr.s_addr;
1044 	*lportp = lport;
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Connect from a socket to a specified address.
1050  * Both address and port must be specified in argument sin.
1051  * If don't have a local address for this socket yet,
1052  * then pick one.
1053  */
1054 int
1055 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1056     bool rehash)
1057 {
1058 	u_short lport, fport;
1059 	in_addr_t laddr, faddr;
1060 	int anonport, error;
1061 
1062 	INP_WLOCK_ASSERT(inp);
1063 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1064 
1065 	lport = inp->inp_lport;
1066 	laddr = inp->inp_laddr.s_addr;
1067 	anonport = (lport == 0);
1068 	error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1069 	    cred);
1070 	if (error)
1071 		return (error);
1072 
1073 	/* Do the initial binding of the local address if required. */
1074 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1075 		KASSERT(rehash == true,
1076 		    ("Rehashing required for unbound inps"));
1077 		inp->inp_lport = lport;
1078 		inp->inp_laddr.s_addr = laddr;
1079 		if (in_pcbinshash(inp) != 0) {
1080 			inp->inp_laddr.s_addr = INADDR_ANY;
1081 			inp->inp_lport = 0;
1082 			return (EAGAIN);
1083 		}
1084 	}
1085 
1086 	/* Commit the remaining changes. */
1087 	inp->inp_lport = lport;
1088 	inp->inp_laddr.s_addr = laddr;
1089 	inp->inp_faddr.s_addr = faddr;
1090 	inp->inp_fport = fport;
1091 	if (rehash) {
1092 		in_pcbrehash(inp);
1093 	} else {
1094 		in_pcbinshash(inp);
1095 	}
1096 
1097 	if (anonport)
1098 		inp->inp_flags |= INP_ANONPORT;
1099 	return (0);
1100 }
1101 
1102 /*
1103  * Do proper source address selection on an unbound socket in case
1104  * of connect. Take jails into account as well.
1105  */
1106 int
1107 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1108     struct ucred *cred)
1109 {
1110 	struct ifaddr *ifa;
1111 	struct sockaddr *sa;
1112 	struct sockaddr_in *sin, dst;
1113 	struct nhop_object *nh;
1114 	int error;
1115 
1116 	NET_EPOCH_ASSERT();
1117 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1118 
1119 	/*
1120 	 * Bypass source address selection and use the primary jail IP
1121 	 * if requested.
1122 	 */
1123 	if (!prison_saddrsel_ip4(cred, laddr))
1124 		return (0);
1125 
1126 	error = 0;
1127 
1128 	nh = NULL;
1129 	bzero(&dst, sizeof(dst));
1130 	sin = &dst;
1131 	sin->sin_family = AF_INET;
1132 	sin->sin_len = sizeof(struct sockaddr_in);
1133 	sin->sin_addr.s_addr = faddr->s_addr;
1134 
1135 	/*
1136 	 * If route is known our src addr is taken from the i/f,
1137 	 * else punt.
1138 	 *
1139 	 * Find out route to destination.
1140 	 */
1141 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1142 		nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1143 		    0, NHR_NONE, 0);
1144 
1145 	/*
1146 	 * If we found a route, use the address corresponding to
1147 	 * the outgoing interface.
1148 	 *
1149 	 * Otherwise assume faddr is reachable on a directly connected
1150 	 * network and try to find a corresponding interface to take
1151 	 * the source address from.
1152 	 */
1153 	if (nh == NULL || nh->nh_ifp == NULL) {
1154 		struct in_ifaddr *ia;
1155 		struct ifnet *ifp;
1156 
1157 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1158 					inp->inp_socket->so_fibnum));
1159 		if (ia == NULL) {
1160 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1161 						inp->inp_socket->so_fibnum));
1162 		}
1163 		if (ia == NULL) {
1164 			error = ENETUNREACH;
1165 			goto done;
1166 		}
1167 
1168 		if (!prison_flag(cred, PR_IP4)) {
1169 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1170 			goto done;
1171 		}
1172 
1173 		ifp = ia->ia_ifp;
1174 		ia = NULL;
1175 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1176 			sa = ifa->ifa_addr;
1177 			if (sa->sa_family != AF_INET)
1178 				continue;
1179 			sin = (struct sockaddr_in *)sa;
1180 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1181 				ia = (struct in_ifaddr *)ifa;
1182 				break;
1183 			}
1184 		}
1185 		if (ia != NULL) {
1186 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1187 			goto done;
1188 		}
1189 
1190 		/* 3. As a last resort return the 'default' jail address. */
1191 		error = prison_get_ip4(cred, laddr);
1192 		goto done;
1193 	}
1194 
1195 	/*
1196 	 * If the outgoing interface on the route found is not
1197 	 * a loopback interface, use the address from that interface.
1198 	 * In case of jails do those three steps:
1199 	 * 1. check if the interface address belongs to the jail. If so use it.
1200 	 * 2. check if we have any address on the outgoing interface
1201 	 *    belonging to this jail. If so use it.
1202 	 * 3. as a last resort return the 'default' jail address.
1203 	 */
1204 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1205 		struct in_ifaddr *ia;
1206 		struct ifnet *ifp;
1207 
1208 		/* If not jailed, use the default returned. */
1209 		if (!prison_flag(cred, PR_IP4)) {
1210 			ia = (struct in_ifaddr *)nh->nh_ifa;
1211 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1212 			goto done;
1213 		}
1214 
1215 		/* Jailed. */
1216 		/* 1. Check if the iface address belongs to the jail. */
1217 		sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1218 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1219 			ia = (struct in_ifaddr *)nh->nh_ifa;
1220 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1221 			goto done;
1222 		}
1223 
1224 		/*
1225 		 * 2. Check if we have any address on the outgoing interface
1226 		 *    belonging to this jail.
1227 		 */
1228 		ia = NULL;
1229 		ifp = nh->nh_ifp;
1230 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1231 			sa = ifa->ifa_addr;
1232 			if (sa->sa_family != AF_INET)
1233 				continue;
1234 			sin = (struct sockaddr_in *)sa;
1235 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1236 				ia = (struct in_ifaddr *)ifa;
1237 				break;
1238 			}
1239 		}
1240 		if (ia != NULL) {
1241 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1242 			goto done;
1243 		}
1244 
1245 		/* 3. As a last resort return the 'default' jail address. */
1246 		error = prison_get_ip4(cred, laddr);
1247 		goto done;
1248 	}
1249 
1250 	/*
1251 	 * The outgoing interface is marked with 'loopback net', so a route
1252 	 * to ourselves is here.
1253 	 * Try to find the interface of the destination address and then
1254 	 * take the address from there. That interface is not necessarily
1255 	 * a loopback interface.
1256 	 * In case of jails, check that it is an address of the jail
1257 	 * and if we cannot find, fall back to the 'default' jail address.
1258 	 */
1259 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1260 		struct in_ifaddr *ia;
1261 
1262 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1263 					inp->inp_socket->so_fibnum));
1264 		if (ia == NULL)
1265 			ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1266 						inp->inp_socket->so_fibnum));
1267 		if (ia == NULL)
1268 			ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1269 
1270 		if (!prison_flag(cred, PR_IP4)) {
1271 			if (ia == NULL) {
1272 				error = ENETUNREACH;
1273 				goto done;
1274 			}
1275 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1276 			goto done;
1277 		}
1278 
1279 		/* Jailed. */
1280 		if (ia != NULL) {
1281 			struct ifnet *ifp;
1282 
1283 			ifp = ia->ia_ifp;
1284 			ia = NULL;
1285 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1286 				sa = ifa->ifa_addr;
1287 				if (sa->sa_family != AF_INET)
1288 					continue;
1289 				sin = (struct sockaddr_in *)sa;
1290 				if (prison_check_ip4(cred,
1291 				    &sin->sin_addr) == 0) {
1292 					ia = (struct in_ifaddr *)ifa;
1293 					break;
1294 				}
1295 			}
1296 			if (ia != NULL) {
1297 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1298 				goto done;
1299 			}
1300 		}
1301 
1302 		/* 3. As a last resort return the 'default' jail address. */
1303 		error = prison_get_ip4(cred, laddr);
1304 		goto done;
1305 	}
1306 
1307 done:
1308 	return (error);
1309 }
1310 
1311 /*
1312  * Set up for a connect from a socket to the specified address.
1313  * On entry, *laddrp and *lportp should contain the current local
1314  * address and port for the PCB; these are updated to the values
1315  * that should be placed in inp_laddr and inp_lport to complete
1316  * the connect.
1317  *
1318  * On success, *faddrp and *fportp will be set to the remote address
1319  * and port. These are not updated in the error case.
1320  */
1321 int
1322 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1323     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1324     struct ucred *cred)
1325 {
1326 	struct in_ifaddr *ia;
1327 	struct in_addr laddr, faddr;
1328 	u_short lport, fport;
1329 	int error;
1330 
1331 	KASSERT(sin->sin_family == AF_INET,
1332 	    ("%s: invalid address family for %p", __func__, sin));
1333 	KASSERT(sin->sin_len == sizeof(*sin),
1334 	    ("%s: invalid address length for %p", __func__, sin));
1335 
1336 	/*
1337 	 * Because a global state change doesn't actually occur here, a read
1338 	 * lock is sufficient.
1339 	 */
1340 	NET_EPOCH_ASSERT();
1341 	INP_LOCK_ASSERT(inp);
1342 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1343 
1344 	if (sin->sin_port == 0)
1345 		return (EADDRNOTAVAIL);
1346 	laddr.s_addr = *laddrp;
1347 	lport = *lportp;
1348 	faddr = sin->sin_addr;
1349 	fport = sin->sin_port;
1350 #ifdef ROUTE_MPATH
1351 	if (CALC_FLOWID_OUTBOUND) {
1352 		uint32_t hash_val, hash_type;
1353 
1354 		hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1355 		    inp->inp_socket->so_proto->pr_protocol, &hash_type);
1356 
1357 		inp->inp_flowid = hash_val;
1358 		inp->inp_flowtype = hash_type;
1359 	}
1360 #endif
1361 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1362 		/*
1363 		 * If the destination address is INADDR_ANY,
1364 		 * use the primary local address.
1365 		 * If the supplied address is INADDR_BROADCAST,
1366 		 * and the primary interface supports broadcast,
1367 		 * choose the broadcast address for that interface.
1368 		 */
1369 		if (faddr.s_addr == INADDR_ANY) {
1370 			faddr =
1371 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1372 			if ((error = prison_get_ip4(cred, &faddr)) != 0)
1373 				return (error);
1374 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1375 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1376 			    IFF_BROADCAST)
1377 				faddr = satosin(&CK_STAILQ_FIRST(
1378 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1379 		}
1380 	}
1381 	if (laddr.s_addr == INADDR_ANY) {
1382 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1383 		/*
1384 		 * If the destination address is multicast and an outgoing
1385 		 * interface has been set as a multicast option, prefer the
1386 		 * address of that interface as our source address.
1387 		 */
1388 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1389 		    inp->inp_moptions != NULL) {
1390 			struct ip_moptions *imo;
1391 			struct ifnet *ifp;
1392 
1393 			imo = inp->inp_moptions;
1394 			if (imo->imo_multicast_ifp != NULL) {
1395 				ifp = imo->imo_multicast_ifp;
1396 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1397 					if (ia->ia_ifp == ifp &&
1398 					    prison_check_ip4(cred,
1399 					    &ia->ia_addr.sin_addr) == 0)
1400 						break;
1401 				}
1402 				if (ia == NULL)
1403 					error = EADDRNOTAVAIL;
1404 				else {
1405 					laddr = ia->ia_addr.sin_addr;
1406 					error = 0;
1407 				}
1408 			}
1409 		}
1410 		if (error)
1411 			return (error);
1412 	}
1413 
1414 	if (lport != 0) {
1415 		if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1416 		    fport, laddr, lport, 0, M_NODOM) != NULL)
1417 			return (EADDRINUSE);
1418 	} else {
1419 		struct sockaddr_in lsin, fsin;
1420 
1421 		bzero(&lsin, sizeof(lsin));
1422 		bzero(&fsin, sizeof(fsin));
1423 		lsin.sin_family = AF_INET;
1424 		lsin.sin_addr = laddr;
1425 		fsin.sin_family = AF_INET;
1426 		fsin.sin_addr = faddr;
1427 		error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1428 		    &lport, (struct sockaddr *)& fsin, fport, cred,
1429 		    INPLOOKUP_WILDCARD);
1430 		if (error)
1431 			return (error);
1432 	}
1433 	*laddrp = laddr.s_addr;
1434 	*lportp = lport;
1435 	*faddrp = faddr.s_addr;
1436 	*fportp = fport;
1437 	return (0);
1438 }
1439 
1440 void
1441 in_pcbdisconnect(struct inpcb *inp)
1442 {
1443 
1444 	INP_WLOCK_ASSERT(inp);
1445 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1446 
1447 	inp->inp_laddr.s_addr = INADDR_ANY;
1448 	inp->inp_faddr.s_addr = INADDR_ANY;
1449 	inp->inp_fport = 0;
1450 	in_pcbrehash(inp);
1451 }
1452 #endif /* INET */
1453 
1454 /*
1455  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1456  * For most protocols, this will be invoked immediately prior to calling
1457  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1458  * socket, in which case in_pcbfree() is deferred.
1459  */
1460 void
1461 in_pcbdetach(struct inpcb *inp)
1462 {
1463 
1464 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1465 
1466 #ifdef RATELIMIT
1467 	if (inp->inp_snd_tag != NULL)
1468 		in_pcbdetach_txrtlmt(inp);
1469 #endif
1470 	inp->inp_socket->so_pcb = NULL;
1471 	inp->inp_socket = NULL;
1472 }
1473 
1474 /*
1475  * inpcb hash lookups are protected by SMR section.
1476  *
1477  * Once desired pcb has been found, switching from SMR section to a pcb
1478  * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1479  * here because SMR is a critical section.
1480  * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1481  */
1482 static inline void
1483 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1484 {
1485 
1486 	lock == INPLOOKUP_RLOCKPCB ?
1487 	    rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1488 }
1489 
1490 static inline void
1491 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1492 {
1493 
1494 	lock == INPLOOKUP_RLOCKPCB ?
1495 	    rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1496 }
1497 
1498 static inline int
1499 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1500 {
1501 
1502 	return (lock == INPLOOKUP_RLOCKPCB ?
1503 	    rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1504 }
1505 
1506 static inline bool
1507 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1508 {
1509 
1510 	return (lock == INPLOOKUP_RLOCKPCB ?
1511 	    in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1512 }
1513 
1514 static inline bool
1515 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1516 {
1517 
1518 	MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1519 	SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1520 
1521 	if (__predict_true(inp_trylock(inp, lock))) {
1522 		if (__predict_false(inp->inp_flags & ignflags)) {
1523 			smr_exit(inp->inp_pcbinfo->ipi_smr);
1524 			inp_unlock(inp, lock);
1525 			return (false);
1526 		}
1527 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1528 		return (true);
1529 	}
1530 
1531 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1532 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1533 		inp_lock(inp, lock);
1534 		if (__predict_false(in_pcbrele(inp, lock)))
1535 			return (false);
1536 		/*
1537 		 * inp acquired through refcount & lock for sure didn't went
1538 		 * through uma_zfree().  However, it may have already went
1539 		 * through in_pcbfree() and has another reference, that
1540 		 * prevented its release by our in_pcbrele().
1541 		 */
1542 		if (__predict_false(inp->inp_flags & ignflags)) {
1543 			inp_unlock(inp, lock);
1544 			return (false);
1545 		}
1546 		return (true);
1547 	} else {
1548 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1549 		return (false);
1550 	}
1551 }
1552 
1553 bool
1554 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1555 {
1556 
1557 	/*
1558 	 * in_pcblookup() family of functions ignore not only freed entries,
1559 	 * that may be found due to lockless access to the hash, but dropped
1560 	 * entries, too.
1561 	 */
1562 	return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1563 }
1564 
1565 /*
1566  * inp_next() - inpcb hash/list traversal iterator
1567  *
1568  * Requires initialized struct inpcb_iterator for context.
1569  * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1570  *
1571  * - Iterator can have either write-lock or read-lock semantics, that can not
1572  *   be changed later.
1573  * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1574  *   a single hash slot.  Note: only rip_input() does the latter.
1575  * - Iterator may have optional bool matching function.  The matching function
1576  *   will be executed for each inpcb in the SMR context, so it can not acquire
1577  *   locks and can safely access only immutable fields of inpcb.
1578  *
1579  * A fresh initialized iterator has NULL inpcb in its context and that
1580  * means that inp_next() call would return the very first inpcb on the list
1581  * locked with desired semantic.  In all following calls the context pointer
1582  * shall hold the current inpcb pointer.  The KPI user is not supposed to
1583  * unlock the current inpcb!  Upon end of traversal inp_next() will return NULL
1584  * and write NULL to its context.  After end of traversal an iterator can be
1585  * reused.
1586  *
1587  * List traversals have the following features/constraints:
1588  * - New entries won't be seen, as they are always added to the head of a list.
1589  * - Removed entries won't stop traversal as long as they are not added to
1590  *   a different list. This is violated by in_pcbrehash().
1591  */
1592 #define	II_LIST_FIRST(ipi, hash)					\
1593 		(((hash) == INP_ALL_LIST) ?				\
1594 		    CK_LIST_FIRST(&(ipi)->ipi_listhead) :		\
1595 		    CK_LIST_FIRST(&(ipi)->ipi_hashbase[(hash)]))
1596 #define	II_LIST_NEXT(inp, hash)						\
1597 		(((hash) == INP_ALL_LIST) ?				\
1598 		    CK_LIST_NEXT((inp), inp_list) :			\
1599 		    CK_LIST_NEXT((inp), inp_hash))
1600 #define	II_LOCK_ASSERT(inp, lock)					\
1601 		rw_assert(&(inp)->inp_lock,				\
1602 		    (lock) == INPLOOKUP_RLOCKPCB ?  RA_RLOCKED : RA_WLOCKED )
1603 struct inpcb *
1604 inp_next(struct inpcb_iterator *ii)
1605 {
1606 	const struct inpcbinfo *ipi = ii->ipi;
1607 	inp_match_t *match = ii->match;
1608 	void *ctx = ii->ctx;
1609 	inp_lookup_t lock = ii->lock;
1610 	int hash = ii->hash;
1611 	struct inpcb *inp;
1612 
1613 	if (ii->inp == NULL) {		/* First call. */
1614 		smr_enter(ipi->ipi_smr);
1615 		/* This is unrolled CK_LIST_FOREACH(). */
1616 		for (inp = II_LIST_FIRST(ipi, hash);
1617 		    inp != NULL;
1618 		    inp = II_LIST_NEXT(inp, hash)) {
1619 			if (match != NULL && (match)(inp, ctx) == false)
1620 				continue;
1621 			if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1622 				break;
1623 			else {
1624 				smr_enter(ipi->ipi_smr);
1625 				MPASS(inp != II_LIST_FIRST(ipi, hash));
1626 				inp = II_LIST_FIRST(ipi, hash);
1627 				if (inp == NULL)
1628 					break;
1629 			}
1630 		}
1631 
1632 		if (inp == NULL)
1633 			smr_exit(ipi->ipi_smr);
1634 		else
1635 			ii->inp = inp;
1636 
1637 		return (inp);
1638 	}
1639 
1640 	/* Not a first call. */
1641 	smr_enter(ipi->ipi_smr);
1642 restart:
1643 	inp = ii->inp;
1644 	II_LOCK_ASSERT(inp, lock);
1645 next:
1646 	inp = II_LIST_NEXT(inp, hash);
1647 	if (inp == NULL) {
1648 		smr_exit(ipi->ipi_smr);
1649 		goto found;
1650 	}
1651 
1652 	if (match != NULL && (match)(inp, ctx) == false)
1653 		goto next;
1654 
1655 	if (__predict_true(inp_trylock(inp, lock))) {
1656 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1657 			/*
1658 			 * Entries are never inserted in middle of a list, thus
1659 			 * as long as we are in SMR, we can continue traversal.
1660 			 * Jump to 'restart' should yield in the same result,
1661 			 * but could produce unnecessary looping.  Could this
1662 			 * looping be unbound?
1663 			 */
1664 			inp_unlock(inp, lock);
1665 			goto next;
1666 		} else {
1667 			smr_exit(ipi->ipi_smr);
1668 			goto found;
1669 		}
1670 	}
1671 
1672 	/*
1673 	 * Can't obtain lock immediately, thus going hard.  Once we exit the
1674 	 * SMR section we can no longer jump to 'next', and our only stable
1675 	 * anchoring point is ii->inp, which we keep locked for this case, so
1676 	 * we jump to 'restart'.
1677 	 */
1678 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1679 		smr_exit(ipi->ipi_smr);
1680 		inp_lock(inp, lock);
1681 		if (__predict_false(in_pcbrele(inp, lock))) {
1682 			smr_enter(ipi->ipi_smr);
1683 			goto restart;
1684 		}
1685 		/*
1686 		 * See comment in inp_smr_lock().
1687 		 */
1688 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1689 			inp_unlock(inp, lock);
1690 			smr_enter(ipi->ipi_smr);
1691 			goto restart;
1692 		}
1693 	} else
1694 		goto next;
1695 
1696 found:
1697 	inp_unlock(ii->inp, lock);
1698 	ii->inp = inp;
1699 
1700 	return (ii->inp);
1701 }
1702 
1703 /*
1704  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1705  * stability of an inpcb pointer despite the inpcb lock being released or
1706  * SMR section exited.
1707  *
1708  * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1709  */
1710 void
1711 in_pcbref(struct inpcb *inp)
1712 {
1713 	u_int old __diagused;
1714 
1715 	old = refcount_acquire(&inp->inp_refcount);
1716 	KASSERT(old > 0, ("%s: refcount 0", __func__));
1717 }
1718 
1719 /*
1720  * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1721  * freeing the pcb, if the reference was very last.
1722  */
1723 bool
1724 in_pcbrele_rlocked(struct inpcb *inp)
1725 {
1726 
1727 	INP_RLOCK_ASSERT(inp);
1728 
1729 	if (!refcount_release(&inp->inp_refcount))
1730 		return (false);
1731 
1732 	MPASS(inp->inp_flags & INP_FREED);
1733 	MPASS(inp->inp_socket == NULL);
1734 	MPASS(inp->inp_in_hpts == 0);
1735 	INP_RUNLOCK(inp);
1736 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1737 	return (true);
1738 }
1739 
1740 bool
1741 in_pcbrele_wlocked(struct inpcb *inp)
1742 {
1743 
1744 	INP_WLOCK_ASSERT(inp);
1745 
1746 	if (!refcount_release(&inp->inp_refcount))
1747 		return (false);
1748 
1749 	MPASS(inp->inp_flags & INP_FREED);
1750 	MPASS(inp->inp_socket == NULL);
1751 	MPASS(inp->inp_in_hpts == 0);
1752 	INP_WUNLOCK(inp);
1753 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1754 	return (true);
1755 }
1756 
1757 /*
1758  * Unconditionally schedule an inpcb to be freed by decrementing its
1759  * reference count, which should occur only after the inpcb has been detached
1760  * from its socket.  If another thread holds a temporary reference (acquired
1761  * using in_pcbref()) then the free is deferred until that reference is
1762  * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1763  *  Almost all work, including removal from global lists, is done in this
1764  * context, where the pcbinfo lock is held.
1765  */
1766 void
1767 in_pcbfree(struct inpcb *inp)
1768 {
1769 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1770 #ifdef INET
1771 	struct ip_moptions *imo;
1772 #endif
1773 #ifdef INET6
1774 	struct ip6_moptions *im6o;
1775 #endif
1776 
1777 	INP_WLOCK_ASSERT(inp);
1778 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1779 	KASSERT((inp->inp_flags & INP_FREED) == 0,
1780 	    ("%s: called twice for pcb %p", __func__, inp));
1781 
1782 	inp->inp_flags |= INP_FREED;
1783 	INP_INFO_WLOCK(pcbinfo);
1784 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1785 	pcbinfo->ipi_count--;
1786 	CK_LIST_REMOVE(inp, inp_list);
1787 	INP_INFO_WUNLOCK(pcbinfo);
1788 
1789 	if (inp->inp_flags & INP_INHASHLIST)
1790 		in_pcbremhash(inp);
1791 
1792 	RO_INVALIDATE_CACHE(&inp->inp_route);
1793 #ifdef MAC
1794 	mac_inpcb_destroy(inp);
1795 #endif
1796 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1797 	if (inp->inp_sp != NULL)
1798 		ipsec_delete_pcbpolicy(inp);
1799 #endif
1800 #ifdef INET
1801 	if (inp->inp_options)
1802 		(void)m_free(inp->inp_options);
1803 	imo = inp->inp_moptions;
1804 #endif
1805 #ifdef INET6
1806 	if (inp->inp_vflag & INP_IPV6PROTO) {
1807 		ip6_freepcbopts(inp->in6p_outputopts);
1808 		im6o = inp->in6p_moptions;
1809 	} else
1810 		im6o = NULL;
1811 #endif
1812 
1813 	if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1814 		INP_WUNLOCK(inp);
1815 	}
1816 #ifdef INET6
1817 	ip6_freemoptions(im6o);
1818 #endif
1819 #ifdef INET
1820 	inp_freemoptions(imo);
1821 #endif
1822 	/* Destruction is finalized in inpcb_dtor(). */
1823 }
1824 
1825 static void
1826 inpcb_dtor(void *mem, int size, void *arg)
1827 {
1828 	struct inpcb *inp = mem;
1829 
1830 	crfree(inp->inp_cred);
1831 #ifdef INVARIANTS
1832 	inp->inp_cred = NULL;
1833 #endif
1834 }
1835 
1836 /*
1837  * Different protocols initialize their inpcbs differently - giving
1838  * different name to the lock.  But they all are disposed the same.
1839  */
1840 static void
1841 inpcb_fini(void *mem, int size)
1842 {
1843 	struct inpcb *inp = mem;
1844 
1845 	INP_LOCK_DESTROY(inp);
1846 }
1847 
1848 /*
1849  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1850  * port reservation, and preventing it from being returned by inpcb lookups.
1851  *
1852  * It is used by TCP to mark an inpcb as unused and avoid future packet
1853  * delivery or event notification when a socket remains open but TCP has
1854  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1855  * or a RST on the wire, and allows the port binding to be reused while still
1856  * maintaining the invariant that so_pcb always points to a valid inpcb until
1857  * in_pcbdetach().
1858  *
1859  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1860  * in_pcbnotifyall() and in_pcbpurgeif0()?
1861  */
1862 void
1863 in_pcbdrop(struct inpcb *inp)
1864 {
1865 
1866 	INP_WLOCK_ASSERT(inp);
1867 #ifdef INVARIANTS
1868 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1869 		MPASS(inp->inp_refcount > 1);
1870 #endif
1871 
1872 	inp->inp_flags |= INP_DROPPED;
1873 	if (inp->inp_flags & INP_INHASHLIST)
1874 		in_pcbremhash(inp);
1875 }
1876 
1877 #ifdef INET
1878 /*
1879  * Common routines to return the socket addresses associated with inpcbs.
1880  */
1881 struct sockaddr *
1882 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1883 {
1884 	struct sockaddr_in *sin;
1885 
1886 	sin = malloc(sizeof *sin, M_SONAME,
1887 		M_WAITOK | M_ZERO);
1888 	sin->sin_family = AF_INET;
1889 	sin->sin_len = sizeof(*sin);
1890 	sin->sin_addr = *addr_p;
1891 	sin->sin_port = port;
1892 
1893 	return (struct sockaddr *)sin;
1894 }
1895 
1896 int
1897 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1898 {
1899 	struct inpcb *inp;
1900 	struct in_addr addr;
1901 	in_port_t port;
1902 
1903 	inp = sotoinpcb(so);
1904 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1905 
1906 	INP_RLOCK(inp);
1907 	port = inp->inp_lport;
1908 	addr = inp->inp_laddr;
1909 	INP_RUNLOCK(inp);
1910 
1911 	*nam = in_sockaddr(port, &addr);
1912 	return 0;
1913 }
1914 
1915 int
1916 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1917 {
1918 	struct inpcb *inp;
1919 	struct in_addr addr;
1920 	in_port_t port;
1921 
1922 	inp = sotoinpcb(so);
1923 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1924 
1925 	INP_RLOCK(inp);
1926 	port = inp->inp_fport;
1927 	addr = inp->inp_faddr;
1928 	INP_RUNLOCK(inp);
1929 
1930 	*nam = in_sockaddr(port, &addr);
1931 	return 0;
1932 }
1933 
1934 void
1935 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1936     struct inpcb *(*notify)(struct inpcb *, int))
1937 {
1938 	struct inpcb *inp, *inp_temp;
1939 
1940 	INP_INFO_WLOCK(pcbinfo);
1941 	CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1942 		INP_WLOCK(inp);
1943 #ifdef INET6
1944 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1945 			INP_WUNLOCK(inp);
1946 			continue;
1947 		}
1948 #endif
1949 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1950 		    inp->inp_socket == NULL) {
1951 			INP_WUNLOCK(inp);
1952 			continue;
1953 		}
1954 		if ((*notify)(inp, errno))
1955 			INP_WUNLOCK(inp);
1956 	}
1957 	INP_INFO_WUNLOCK(pcbinfo);
1958 }
1959 
1960 static bool
1961 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1962 {
1963 
1964 	if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1965 		return (true);
1966 	else
1967 		return (false);
1968 }
1969 
1970 void
1971 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1972 {
1973 	struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1974 	    inp_v4_multi_match, NULL);
1975 	struct inpcb *inp;
1976 	struct in_multi *inm;
1977 	struct in_mfilter *imf;
1978 	struct ip_moptions *imo;
1979 
1980 	IN_MULTI_LOCK_ASSERT();
1981 
1982 	while ((inp = inp_next(&inpi)) != NULL) {
1983 		INP_WLOCK_ASSERT(inp);
1984 
1985 		imo = inp->inp_moptions;
1986 		/*
1987 		 * Unselect the outgoing interface if it is being
1988 		 * detached.
1989 		 */
1990 		if (imo->imo_multicast_ifp == ifp)
1991 			imo->imo_multicast_ifp = NULL;
1992 
1993 		/*
1994 		 * Drop multicast group membership if we joined
1995 		 * through the interface being detached.
1996 		 *
1997 		 * XXX This can all be deferred to an epoch_call
1998 		 */
1999 restart:
2000 		IP_MFILTER_FOREACH(imf, &imo->imo_head) {
2001 			if ((inm = imf->imf_inm) == NULL)
2002 				continue;
2003 			if (inm->inm_ifp != ifp)
2004 				continue;
2005 			ip_mfilter_remove(&imo->imo_head, imf);
2006 			in_leavegroup_locked(inm, NULL);
2007 			ip_mfilter_free(imf);
2008 			goto restart;
2009 		}
2010 	}
2011 }
2012 
2013 /*
2014  * Lookup a PCB based on the local address and port.  Caller must hold the
2015  * hash lock.  No inpcb locks or references are acquired.
2016  */
2017 #define INP_LOOKUP_MAPPED_PCB_COST	3
2018 struct inpcb *
2019 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
2020     u_short lport, int lookupflags, struct ucred *cred)
2021 {
2022 	struct inpcb *inp;
2023 #ifdef INET6
2024 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
2025 #else
2026 	int matchwild = 3;
2027 #endif
2028 	int wildcard;
2029 
2030 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2031 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2032 	INP_HASH_LOCK_ASSERT(pcbinfo);
2033 
2034 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
2035 		struct inpcbhead *head;
2036 		/*
2037 		 * Look for an unconnected (wildcard foreign addr) PCB that
2038 		 * matches the local address and port we're looking for.
2039 		 */
2040 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2041 		    pcbinfo->ipi_hashmask)];
2042 		CK_LIST_FOREACH(inp, head, inp_hash) {
2043 #ifdef INET6
2044 			/* XXX inp locking */
2045 			if ((inp->inp_vflag & INP_IPV4) == 0)
2046 				continue;
2047 #endif
2048 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
2049 			    inp->inp_laddr.s_addr == laddr.s_addr &&
2050 			    inp->inp_lport == lport) {
2051 				/*
2052 				 * Found?
2053 				 */
2054 				if (prison_equal_ip4(cred->cr_prison,
2055 				    inp->inp_cred->cr_prison))
2056 					return (inp);
2057 			}
2058 		}
2059 		/*
2060 		 * Not found.
2061 		 */
2062 		return (NULL);
2063 	} else {
2064 		struct inpcbporthead *porthash;
2065 		struct inpcbport *phd;
2066 		struct inpcb *match = NULL;
2067 		/*
2068 		 * Best fit PCB lookup.
2069 		 *
2070 		 * First see if this local port is in use by looking on the
2071 		 * port hash list.
2072 		 */
2073 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2074 		    pcbinfo->ipi_porthashmask)];
2075 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
2076 			if (phd->phd_port == lport)
2077 				break;
2078 		}
2079 		if (phd != NULL) {
2080 			/*
2081 			 * Port is in use by one or more PCBs. Look for best
2082 			 * fit.
2083 			 */
2084 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2085 				wildcard = 0;
2086 				if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2087 				    cred->cr_prison))
2088 					continue;
2089 #ifdef INET6
2090 				/* XXX inp locking */
2091 				if ((inp->inp_vflag & INP_IPV4) == 0)
2092 					continue;
2093 				/*
2094 				 * We never select the PCB that has
2095 				 * INP_IPV6 flag and is bound to :: if
2096 				 * we have another PCB which is bound
2097 				 * to 0.0.0.0.  If a PCB has the
2098 				 * INP_IPV6 flag, then we set its cost
2099 				 * higher than IPv4 only PCBs.
2100 				 *
2101 				 * Note that the case only happens
2102 				 * when a socket is bound to ::, under
2103 				 * the condition that the use of the
2104 				 * mapped address is allowed.
2105 				 */
2106 				if ((inp->inp_vflag & INP_IPV6) != 0)
2107 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2108 #endif
2109 				if (inp->inp_faddr.s_addr != INADDR_ANY)
2110 					wildcard++;
2111 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
2112 					if (laddr.s_addr == INADDR_ANY)
2113 						wildcard++;
2114 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
2115 						continue;
2116 				} else {
2117 					if (laddr.s_addr != INADDR_ANY)
2118 						wildcard++;
2119 				}
2120 				if (wildcard < matchwild) {
2121 					match = inp;
2122 					matchwild = wildcard;
2123 					if (matchwild == 0)
2124 						break;
2125 				}
2126 			}
2127 		}
2128 		return (match);
2129 	}
2130 }
2131 #undef INP_LOOKUP_MAPPED_PCB_COST
2132 
2133 static bool
2134 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2135 {
2136 	return (domain == M_NODOM || domain == grp->il_numa_domain);
2137 }
2138 
2139 static struct inpcb *
2140 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2141     const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr,
2142     uint16_t lport, int domain)
2143 {
2144 	const struct inpcblbgrouphead *hdr;
2145 	struct inpcblbgroup *grp;
2146 	struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2147 
2148 	INP_HASH_LOCK_ASSERT(pcbinfo);
2149 
2150 	hdr = &pcbinfo->ipi_lbgrouphashbase[
2151 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2152 
2153 	/*
2154 	 * Search for an LB group match based on the following criteria:
2155 	 * - prefer jailed groups to non-jailed groups
2156 	 * - prefer exact source address matches to wildcard matches
2157 	 * - prefer groups bound to the specified NUMA domain
2158 	 */
2159 	jail_exact = jail_wild = local_exact = local_wild = NULL;
2160 	CK_LIST_FOREACH(grp, hdr, il_list) {
2161 		bool injail;
2162 
2163 #ifdef INET6
2164 		if (!(grp->il_vflag & INP_IPV4))
2165 			continue;
2166 #endif
2167 		if (grp->il_lport != lport)
2168 			continue;
2169 
2170 		injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2171 		if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2172 		    laddr) != 0)
2173 			continue;
2174 
2175 		if (grp->il_laddr.s_addr == laddr->s_addr) {
2176 			if (injail) {
2177 				jail_exact = grp;
2178 				if (in_pcblookup_lb_numa_match(grp, domain))
2179 					/* This is a perfect match. */
2180 					goto out;
2181 			} else if (local_exact == NULL ||
2182 			    in_pcblookup_lb_numa_match(grp, domain)) {
2183 				local_exact = grp;
2184 			}
2185 		} else if (grp->il_laddr.s_addr == INADDR_ANY) {
2186 			if (injail) {
2187 				if (jail_wild == NULL ||
2188 				    in_pcblookup_lb_numa_match(grp, domain))
2189 					jail_wild = grp;
2190 			} else if (local_wild == NULL ||
2191 			    in_pcblookup_lb_numa_match(grp, domain)) {
2192 				local_wild = grp;
2193 			}
2194 		}
2195 	}
2196 
2197 	if (jail_exact != NULL)
2198 		grp = jail_exact;
2199 	else if (jail_wild != NULL)
2200 		grp = jail_wild;
2201 	else if (local_exact != NULL)
2202 		grp = local_exact;
2203 	else
2204 		grp = local_wild;
2205 	if (grp == NULL)
2206 		return (NULL);
2207 out:
2208 	return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2209 	    grp->il_inpcnt]);
2210 }
2211 
2212 static struct inpcb *
2213 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2214     u_short fport, struct in_addr laddr, u_short lport)
2215 {
2216 	struct inpcbhead *head;
2217 	struct inpcb *inp, *match;
2218 
2219 	INP_HASH_LOCK_ASSERT(pcbinfo);
2220 
2221 	match = NULL;
2222 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&faddr, lport, fport,
2223 	    pcbinfo->ipi_hashmask)];
2224 	CK_LIST_FOREACH(inp, head, inp_hash) {
2225 #ifdef INET6
2226 		/* XXX inp locking */
2227 		if ((inp->inp_vflag & INP_IPV4) == 0)
2228 			continue;
2229 #endif
2230 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2231 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2232 		    inp->inp_fport == fport &&
2233 		    inp->inp_lport == lport)
2234 			return (inp);
2235 	}
2236 	return (match);
2237 }
2238 
2239 static struct inpcb *
2240 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2241     u_short fport, struct in_addr laddr, u_short lport)
2242 {
2243 	struct inpcbhead *head;
2244 	struct inpcb *inp, *local_wild, *local_exact, *jail_wild;
2245 #ifdef INET6
2246 	struct inpcb *local_wild_mapped;
2247 #endif
2248 
2249 	INP_HASH_LOCK_ASSERT(pcbinfo);
2250 
2251 	/*
2252 	 * Order of socket selection - we always prefer jails.
2253 	 *      1. jailed, non-wild.
2254 	 *      2. jailed, wild.
2255 	 *      3. non-jailed, non-wild.
2256 	 *      4. non-jailed, wild.
2257 	 */
2258 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2259 	    pcbinfo->ipi_hashmask)];
2260 	local_wild = local_exact = jail_wild = NULL;
2261 #ifdef INET6
2262 	local_wild_mapped = NULL;
2263 #endif
2264 	CK_LIST_FOREACH(inp, head, inp_hash) {
2265 		bool injail;
2266 
2267 #ifdef INET6
2268 		/* XXX inp locking */
2269 		if ((inp->inp_vflag & INP_IPV4) == 0)
2270 			continue;
2271 #endif
2272 		if (inp->inp_faddr.s_addr != INADDR_ANY ||
2273 		    inp->inp_lport != lport)
2274 			continue;
2275 
2276 		injail = prison_flag(inp->inp_cred, PR_IP4) != 0;
2277 		if (injail) {
2278 			if (prison_check_ip4_locked(inp->inp_cred->cr_prison,
2279 			    &laddr) != 0)
2280 				continue;
2281 		} else {
2282 			if (local_exact != NULL)
2283 				continue;
2284 		}
2285 
2286 		if (inp->inp_laddr.s_addr == laddr.s_addr) {
2287 			if (injail)
2288 				return (inp);
2289 			local_exact = inp;
2290 		} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2291 #ifdef INET6
2292 			/* XXX inp locking, NULL check */
2293 			if (inp->inp_vflag & INP_IPV6PROTO)
2294 				local_wild_mapped = inp;
2295 			else
2296 #endif
2297 				if (injail)
2298 					jail_wild = inp;
2299 				else
2300 					local_wild = inp;
2301 		}
2302 	}
2303 	if (jail_wild != NULL)
2304 		return (jail_wild);
2305 	if (local_exact != NULL)
2306 		return (local_exact);
2307 	if (local_wild != NULL)
2308 		return (local_wild);
2309 #ifdef INET6
2310 	if (local_wild_mapped != NULL)
2311 		return (local_wild_mapped);
2312 #endif
2313 	return (NULL);
2314 }
2315 
2316 /*
2317  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2318  * that the caller has either locked the hash list, which usually happens
2319  * for bind(2) operations, or is in SMR section, which happens when sorting
2320  * out incoming packets.
2321  */
2322 static struct inpcb *
2323 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2324     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2325     uint8_t numa_domain)
2326 {
2327 	struct inpcb *inp;
2328 	const u_short fport = fport_arg, lport = lport_arg;
2329 
2330 	KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0,
2331 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2332 	KASSERT(faddr.s_addr != INADDR_ANY,
2333 	    ("%s: invalid foreign address", __func__));
2334 	KASSERT(laddr.s_addr != INADDR_ANY,
2335 	    ("%s: invalid local address", __func__));
2336 	INP_HASH_LOCK_ASSERT(pcbinfo);
2337 
2338 	inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2339 	if (inp != NULL)
2340 		return (inp);
2341 
2342 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2343 		inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, &laddr,
2344 		    lport, numa_domain);
2345 		if (inp == NULL) {
2346 			inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr,
2347 			    fport, laddr, lport);
2348 		}
2349 	}
2350 
2351 	return (inp);
2352 }
2353 
2354 static struct inpcb *
2355 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2356     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2357     uint8_t numa_domain)
2358 {
2359 	struct inpcb *inp;
2360 
2361 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2362 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2363 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2364 	    ("%s: LOCKPCB not set", __func__));
2365 
2366 	smr_enter(pcbinfo->ipi_smr);
2367 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2368 	    lookupflags & INPLOOKUP_WILDCARD, numa_domain);
2369 	if (inp != NULL) {
2370 		if (__predict_false(inp_smr_lock(inp,
2371 		    (lookupflags & INPLOOKUP_LOCKMASK)) == false))
2372 			inp = NULL;
2373 	} else
2374 		smr_exit(pcbinfo->ipi_smr);
2375 
2376 	return (inp);
2377 }
2378 
2379 /*
2380  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2381  * from which a pre-calculated hash value may be extracted.
2382  */
2383 struct inpcb *
2384 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2385     struct in_addr laddr, u_int lport, int lookupflags,
2386     struct ifnet *ifp __unused)
2387 {
2388 	return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2389 	    lookupflags, M_NODOM));
2390 }
2391 
2392 struct inpcb *
2393 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2394     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2395     struct ifnet *ifp __unused, struct mbuf *m)
2396 {
2397 	return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2398 	    lookupflags, m->m_pkthdr.numa_domain));
2399 }
2400 #endif /* INET */
2401 
2402 /*
2403  * Insert PCB onto various hash lists.
2404  */
2405 int
2406 in_pcbinshash(struct inpcb *inp)
2407 {
2408 	struct inpcbhead *pcbhash;
2409 	struct inpcbporthead *pcbporthash;
2410 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2411 	struct inpcbport *phd;
2412 
2413 	INP_WLOCK_ASSERT(inp);
2414 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2415 
2416 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2417 	    ("in_pcbinshash: INP_INHASHLIST"));
2418 
2419 #ifdef INET6
2420 	if (inp->inp_vflag & INP_IPV6)
2421 		pcbhash = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2422 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2423 	else
2424 #endif
2425 		pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2426 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2427 
2428 	pcbporthash = &pcbinfo->ipi_porthashbase[
2429 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2430 
2431 	/*
2432 	 * Add entry to load balance group.
2433 	 * Only do this if SO_REUSEPORT_LB is set.
2434 	 */
2435 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2436 		int error = in_pcbinslbgrouphash(inp, M_NODOM);
2437 		if (error != 0)
2438 			return (error);
2439 	}
2440 
2441 	/*
2442 	 * Go through port list and look for a head for this lport.
2443 	 */
2444 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2445 		if (phd->phd_port == inp->inp_lport)
2446 			break;
2447 	}
2448 
2449 	/*
2450 	 * If none exists, malloc one and tack it on.
2451 	 */
2452 	if (phd == NULL) {
2453 		phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2454 		if (phd == NULL) {
2455 			if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2456 				in_pcbremlbgrouphash(inp);
2457 			return (ENOMEM);
2458 		}
2459 		phd->phd_port = inp->inp_lport;
2460 		CK_LIST_INIT(&phd->phd_pcblist);
2461 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2462 	}
2463 	inp->inp_phd = phd;
2464 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2465 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2466 	inp->inp_flags |= INP_INHASHLIST;
2467 
2468 	return (0);
2469 }
2470 
2471 static void
2472 in_pcbremhash(struct inpcb *inp)
2473 {
2474 	struct inpcbport *phd = inp->inp_phd;
2475 
2476 	INP_WLOCK_ASSERT(inp);
2477 	MPASS(inp->inp_flags & INP_INHASHLIST);
2478 
2479 	INP_HASH_WLOCK(inp->inp_pcbinfo);
2480 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2481 		in_pcbremlbgrouphash(inp);
2482 	CK_LIST_REMOVE(inp, inp_hash);
2483 	CK_LIST_REMOVE(inp, inp_portlist);
2484 	if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2485 		CK_LIST_REMOVE(phd, phd_hash);
2486 		uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2487 	}
2488 	INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2489 	inp->inp_flags &= ~INP_INHASHLIST;
2490 }
2491 
2492 /*
2493  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2494  * changed. NOTE: This does not handle the case of the lport changing (the
2495  * hashed port list would have to be updated as well), so the lport must
2496  * not change after in_pcbinshash() has been called.
2497  *
2498  * XXXGL: a race between this function and SMR-protected hash iterator
2499  * will lead to iterator traversing a possibly wrong hash list. However,
2500  * this race should have been here since change from rwlock to epoch.
2501  */
2502 void
2503 in_pcbrehash(struct inpcb *inp)
2504 {
2505 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2506 	struct inpcbhead *head;
2507 
2508 	INP_WLOCK_ASSERT(inp);
2509 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2510 
2511 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2512 	    ("in_pcbrehash: !INP_INHASHLIST"));
2513 
2514 #ifdef INET6
2515 	if (inp->inp_vflag & INP_IPV6)
2516 		head = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2517 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2518 	else
2519 #endif
2520 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2521 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2522 
2523 	CK_LIST_REMOVE(inp, inp_hash);
2524 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2525 }
2526 
2527 /*
2528  * Check for alternatives when higher level complains
2529  * about service problems.  For now, invalidate cached
2530  * routing information.  If the route was created dynamically
2531  * (by a redirect), time to try a default gateway again.
2532  */
2533 void
2534 in_losing(struct inpcb *inp)
2535 {
2536 
2537 	RO_INVALIDATE_CACHE(&inp->inp_route);
2538 	return;
2539 }
2540 
2541 /*
2542  * A set label operation has occurred at the socket layer, propagate the
2543  * label change into the in_pcb for the socket.
2544  */
2545 void
2546 in_pcbsosetlabel(struct socket *so)
2547 {
2548 #ifdef MAC
2549 	struct inpcb *inp;
2550 
2551 	inp = sotoinpcb(so);
2552 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2553 
2554 	INP_WLOCK(inp);
2555 	SOCK_LOCK(so);
2556 	mac_inpcb_sosetlabel(so, inp);
2557 	SOCK_UNLOCK(so);
2558 	INP_WUNLOCK(inp);
2559 #endif
2560 }
2561 
2562 void
2563 inp_wlock(struct inpcb *inp)
2564 {
2565 
2566 	INP_WLOCK(inp);
2567 }
2568 
2569 void
2570 inp_wunlock(struct inpcb *inp)
2571 {
2572 
2573 	INP_WUNLOCK(inp);
2574 }
2575 
2576 void
2577 inp_rlock(struct inpcb *inp)
2578 {
2579 
2580 	INP_RLOCK(inp);
2581 }
2582 
2583 void
2584 inp_runlock(struct inpcb *inp)
2585 {
2586 
2587 	INP_RUNLOCK(inp);
2588 }
2589 
2590 #ifdef INVARIANT_SUPPORT
2591 void
2592 inp_lock_assert(struct inpcb *inp)
2593 {
2594 
2595 	INP_WLOCK_ASSERT(inp);
2596 }
2597 
2598 void
2599 inp_unlock_assert(struct inpcb *inp)
2600 {
2601 
2602 	INP_UNLOCK_ASSERT(inp);
2603 }
2604 #endif
2605 
2606 void
2607 inp_apply_all(struct inpcbinfo *pcbinfo,
2608     void (*func)(struct inpcb *, void *), void *arg)
2609 {
2610 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2611 	    INPLOOKUP_WLOCKPCB);
2612 	struct inpcb *inp;
2613 
2614 	while ((inp = inp_next(&inpi)) != NULL)
2615 		func(inp, arg);
2616 }
2617 
2618 struct socket *
2619 inp_inpcbtosocket(struct inpcb *inp)
2620 {
2621 
2622 	INP_WLOCK_ASSERT(inp);
2623 	return (inp->inp_socket);
2624 }
2625 
2626 struct tcpcb *
2627 inp_inpcbtotcpcb(struct inpcb *inp)
2628 {
2629 
2630 	INP_WLOCK_ASSERT(inp);
2631 	return ((struct tcpcb *)inp->inp_ppcb);
2632 }
2633 
2634 int
2635 inp_ip_tos_get(const struct inpcb *inp)
2636 {
2637 
2638 	return (inp->inp_ip_tos);
2639 }
2640 
2641 void
2642 inp_ip_tos_set(struct inpcb *inp, int val)
2643 {
2644 
2645 	inp->inp_ip_tos = val;
2646 }
2647 
2648 void
2649 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2650     uint32_t *faddr, uint16_t *fp)
2651 {
2652 
2653 	INP_LOCK_ASSERT(inp);
2654 	*laddr = inp->inp_laddr.s_addr;
2655 	*faddr = inp->inp_faddr.s_addr;
2656 	*lp = inp->inp_lport;
2657 	*fp = inp->inp_fport;
2658 }
2659 
2660 struct inpcb *
2661 so_sotoinpcb(struct socket *so)
2662 {
2663 
2664 	return (sotoinpcb(so));
2665 }
2666 
2667 /*
2668  * Create an external-format (``xinpcb'') structure using the information in
2669  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2670  * reduce the spew of irrelevant information over this interface, to isolate
2671  * user code from changes in the kernel structure, and potentially to provide
2672  * information-hiding if we decide that some of this information should be
2673  * hidden from users.
2674  */
2675 void
2676 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2677 {
2678 
2679 	bzero(xi, sizeof(*xi));
2680 	xi->xi_len = sizeof(struct xinpcb);
2681 	if (inp->inp_socket)
2682 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2683 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2684 	xi->inp_gencnt = inp->inp_gencnt;
2685 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2686 	xi->inp_flow = inp->inp_flow;
2687 	xi->inp_flowid = inp->inp_flowid;
2688 	xi->inp_flowtype = inp->inp_flowtype;
2689 	xi->inp_flags = inp->inp_flags;
2690 	xi->inp_flags2 = inp->inp_flags2;
2691 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2692 	xi->in6p_cksum = inp->in6p_cksum;
2693 	xi->in6p_hops = inp->in6p_hops;
2694 	xi->inp_ip_tos = inp->inp_ip_tos;
2695 	xi->inp_vflag = inp->inp_vflag;
2696 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2697 	xi->inp_ip_p = inp->inp_ip_p;
2698 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2699 }
2700 
2701 int
2702 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2703     int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2704 {
2705 	struct sockopt sopt;
2706 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2707 	    INPLOOKUP_WLOCKPCB);
2708 	struct inpcb *inp;
2709 	struct sockopt_parameters *params;
2710 	struct socket *so;
2711 	int error;
2712 	char buf[1024];
2713 
2714 	if (req->oldptr != NULL || req->oldlen != 0)
2715 		return (EINVAL);
2716 	if (req->newptr == NULL)
2717 		return (EPERM);
2718 	if (req->newlen > sizeof(buf))
2719 		return (ENOMEM);
2720 	error = SYSCTL_IN(req, buf, req->newlen);
2721 	if (error != 0)
2722 		return (error);
2723 	if (req->newlen < sizeof(struct sockopt_parameters))
2724 		return (EINVAL);
2725 	params = (struct sockopt_parameters *)buf;
2726 	sopt.sopt_level = params->sop_level;
2727 	sopt.sopt_name = params->sop_optname;
2728 	sopt.sopt_dir = SOPT_SET;
2729 	sopt.sopt_val = params->sop_optval;
2730 	sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
2731 	sopt.sopt_td = NULL;
2732 #ifdef INET6
2733 	if (params->sop_inc.inc_flags & INC_ISIPV6) {
2734 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_laddr))
2735 			params->sop_inc.inc6_laddr.s6_addr16[1] =
2736 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2737 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_faddr))
2738 			params->sop_inc.inc6_faddr.s6_addr16[1] =
2739 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2740 	}
2741 #endif
2742 	if (params->sop_inc.inc_lport != htons(0)) {
2743 		if (params->sop_inc.inc_fport == htons(0))
2744 			inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
2745 			    pcbinfo->ipi_hashmask);
2746 		else
2747 #ifdef INET6
2748 			if (params->sop_inc.inc_flags & INC_ISIPV6)
2749 				inpi.hash = INP6_PCBHASH(
2750 				    &params->sop_inc.inc6_faddr,
2751 				    params->sop_inc.inc_lport,
2752 				    params->sop_inc.inc_fport,
2753 				    pcbinfo->ipi_hashmask);
2754 			else
2755 #endif
2756 				inpi.hash = INP_PCBHASH(
2757 				    &params->sop_inc.inc_faddr,
2758 				    params->sop_inc.inc_lport,
2759 				    params->sop_inc.inc_fport,
2760 				    pcbinfo->ipi_hashmask);
2761 	}
2762 	while ((inp = inp_next(&inpi)) != NULL)
2763 		if (inp->inp_gencnt == params->sop_id) {
2764 			if (inp->inp_flags & INP_DROPPED) {
2765 				INP_WUNLOCK(inp);
2766 				return (ECONNRESET);
2767 			}
2768 			so = inp->inp_socket;
2769 			KASSERT(so != NULL, ("inp_socket == NULL"));
2770 			soref(so);
2771 			error = (*ctloutput_set)(inp, &sopt);
2772 			sorele(so);
2773 			break;
2774 		}
2775 	if (inp == NULL)
2776 		error = ESRCH;
2777 	return (error);
2778 }
2779 
2780 #ifdef DDB
2781 static void
2782 db_print_indent(int indent)
2783 {
2784 	int i;
2785 
2786 	for (i = 0; i < indent; i++)
2787 		db_printf(" ");
2788 }
2789 
2790 static void
2791 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2792 {
2793 	char faddr_str[48], laddr_str[48];
2794 
2795 	db_print_indent(indent);
2796 	db_printf("%s at %p\n", name, inc);
2797 
2798 	indent += 2;
2799 
2800 #ifdef INET6
2801 	if (inc->inc_flags & INC_ISIPV6) {
2802 		/* IPv6. */
2803 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2804 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2805 	} else
2806 #endif
2807 	{
2808 		/* IPv4. */
2809 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2810 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2811 	}
2812 	db_print_indent(indent);
2813 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2814 	    ntohs(inc->inc_lport));
2815 	db_print_indent(indent);
2816 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2817 	    ntohs(inc->inc_fport));
2818 }
2819 
2820 static void
2821 db_print_inpflags(int inp_flags)
2822 {
2823 	int comma;
2824 
2825 	comma = 0;
2826 	if (inp_flags & INP_RECVOPTS) {
2827 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2828 		comma = 1;
2829 	}
2830 	if (inp_flags & INP_RECVRETOPTS) {
2831 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2832 		comma = 1;
2833 	}
2834 	if (inp_flags & INP_RECVDSTADDR) {
2835 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2836 		comma = 1;
2837 	}
2838 	if (inp_flags & INP_ORIGDSTADDR) {
2839 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2840 		comma = 1;
2841 	}
2842 	if (inp_flags & INP_HDRINCL) {
2843 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2844 		comma = 1;
2845 	}
2846 	if (inp_flags & INP_HIGHPORT) {
2847 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2848 		comma = 1;
2849 	}
2850 	if (inp_flags & INP_LOWPORT) {
2851 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2852 		comma = 1;
2853 	}
2854 	if (inp_flags & INP_ANONPORT) {
2855 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2856 		comma = 1;
2857 	}
2858 	if (inp_flags & INP_RECVIF) {
2859 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2860 		comma = 1;
2861 	}
2862 	if (inp_flags & INP_MTUDISC) {
2863 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2864 		comma = 1;
2865 	}
2866 	if (inp_flags & INP_RECVTTL) {
2867 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2868 		comma = 1;
2869 	}
2870 	if (inp_flags & INP_DONTFRAG) {
2871 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2872 		comma = 1;
2873 	}
2874 	if (inp_flags & INP_RECVTOS) {
2875 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2876 		comma = 1;
2877 	}
2878 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2879 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2880 		comma = 1;
2881 	}
2882 	if (inp_flags & IN6P_PKTINFO) {
2883 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2884 		comma = 1;
2885 	}
2886 	if (inp_flags & IN6P_HOPLIMIT) {
2887 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2888 		comma = 1;
2889 	}
2890 	if (inp_flags & IN6P_HOPOPTS) {
2891 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2892 		comma = 1;
2893 	}
2894 	if (inp_flags & IN6P_DSTOPTS) {
2895 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2896 		comma = 1;
2897 	}
2898 	if (inp_flags & IN6P_RTHDR) {
2899 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2900 		comma = 1;
2901 	}
2902 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2903 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2904 		comma = 1;
2905 	}
2906 	if (inp_flags & IN6P_TCLASS) {
2907 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2908 		comma = 1;
2909 	}
2910 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2911 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2912 		comma = 1;
2913 	}
2914 	if (inp_flags & INP_ONESBCAST) {
2915 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2916 		comma  = 1;
2917 	}
2918 	if (inp_flags & INP_DROPPED) {
2919 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2920 		comma  = 1;
2921 	}
2922 	if (inp_flags & INP_SOCKREF) {
2923 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2924 		comma  = 1;
2925 	}
2926 	if (inp_flags & IN6P_RFC2292) {
2927 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2928 		comma = 1;
2929 	}
2930 	if (inp_flags & IN6P_MTU) {
2931 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2932 		comma = 1;
2933 	}
2934 }
2935 
2936 static void
2937 db_print_inpvflag(u_char inp_vflag)
2938 {
2939 	int comma;
2940 
2941 	comma = 0;
2942 	if (inp_vflag & INP_IPV4) {
2943 		db_printf("%sINP_IPV4", comma ? ", " : "");
2944 		comma  = 1;
2945 	}
2946 	if (inp_vflag & INP_IPV6) {
2947 		db_printf("%sINP_IPV6", comma ? ", " : "");
2948 		comma  = 1;
2949 	}
2950 	if (inp_vflag & INP_IPV6PROTO) {
2951 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2952 		comma  = 1;
2953 	}
2954 }
2955 
2956 static void
2957 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2958 {
2959 
2960 	db_print_indent(indent);
2961 	db_printf("%s at %p\n", name, inp);
2962 
2963 	indent += 2;
2964 
2965 	db_print_indent(indent);
2966 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2967 
2968 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2969 
2970 	db_print_indent(indent);
2971 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2972 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2973 
2974 	db_print_indent(indent);
2975 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2976 	   inp->inp_label, inp->inp_flags);
2977 	db_print_inpflags(inp->inp_flags);
2978 	db_printf(")\n");
2979 
2980 	db_print_indent(indent);
2981 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2982 	    inp->inp_vflag);
2983 	db_print_inpvflag(inp->inp_vflag);
2984 	db_printf(")\n");
2985 
2986 	db_print_indent(indent);
2987 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2988 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2989 
2990 	db_print_indent(indent);
2991 #ifdef INET6
2992 	if (inp->inp_vflag & INP_IPV6) {
2993 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2994 		    "in6p_moptions: %p\n", inp->in6p_options,
2995 		    inp->in6p_outputopts, inp->in6p_moptions);
2996 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2997 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2998 		    inp->in6p_hops);
2999 	} else
3000 #endif
3001 	{
3002 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3003 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3004 		    inp->inp_options, inp->inp_moptions);
3005 	}
3006 
3007 	db_print_indent(indent);
3008 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3009 	    (uintmax_t)inp->inp_gencnt);
3010 }
3011 
3012 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3013 {
3014 	struct inpcb *inp;
3015 
3016 	if (!have_addr) {
3017 		db_printf("usage: show inpcb <addr>\n");
3018 		return;
3019 	}
3020 	inp = (struct inpcb *)addr;
3021 
3022 	db_print_inpcb(inp, "inpcb", 0);
3023 }
3024 #endif /* DDB */
3025 
3026 #ifdef RATELIMIT
3027 /*
3028  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3029  * if any.
3030  */
3031 int
3032 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3033 {
3034 	union if_snd_tag_modify_params params = {
3035 		.rate_limit.max_rate = max_pacing_rate,
3036 		.rate_limit.flags = M_NOWAIT,
3037 	};
3038 	struct m_snd_tag *mst;
3039 	int error;
3040 
3041 	mst = inp->inp_snd_tag;
3042 	if (mst == NULL)
3043 		return (EINVAL);
3044 
3045 	if (mst->sw->snd_tag_modify == NULL) {
3046 		error = EOPNOTSUPP;
3047 	} else {
3048 		error = mst->sw->snd_tag_modify(mst, &params);
3049 	}
3050 	return (error);
3051 }
3052 
3053 /*
3054  * Query existing TX rate limit based on the existing
3055  * "inp->inp_snd_tag", if any.
3056  */
3057 int
3058 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3059 {
3060 	union if_snd_tag_query_params params = { };
3061 	struct m_snd_tag *mst;
3062 	int error;
3063 
3064 	mst = inp->inp_snd_tag;
3065 	if (mst == NULL)
3066 		return (EINVAL);
3067 
3068 	if (mst->sw->snd_tag_query == NULL) {
3069 		error = EOPNOTSUPP;
3070 	} else {
3071 		error = mst->sw->snd_tag_query(mst, &params);
3072 		if (error == 0 && p_max_pacing_rate != NULL)
3073 			*p_max_pacing_rate = params.rate_limit.max_rate;
3074 	}
3075 	return (error);
3076 }
3077 
3078 /*
3079  * Query existing TX queue level based on the existing
3080  * "inp->inp_snd_tag", if any.
3081  */
3082 int
3083 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3084 {
3085 	union if_snd_tag_query_params params = { };
3086 	struct m_snd_tag *mst;
3087 	int error;
3088 
3089 	mst = inp->inp_snd_tag;
3090 	if (mst == NULL)
3091 		return (EINVAL);
3092 
3093 	if (mst->sw->snd_tag_query == NULL)
3094 		return (EOPNOTSUPP);
3095 
3096 	error = mst->sw->snd_tag_query(mst, &params);
3097 	if (error == 0 && p_txqueue_level != NULL)
3098 		*p_txqueue_level = params.rate_limit.queue_level;
3099 	return (error);
3100 }
3101 
3102 /*
3103  * Allocate a new TX rate limit send tag from the network interface
3104  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3105  */
3106 int
3107 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3108     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3109 
3110 {
3111 	union if_snd_tag_alloc_params params = {
3112 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3113 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3114 		.rate_limit.hdr.flowid = flowid,
3115 		.rate_limit.hdr.flowtype = flowtype,
3116 		.rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3117 		.rate_limit.max_rate = max_pacing_rate,
3118 		.rate_limit.flags = M_NOWAIT,
3119 	};
3120 	int error;
3121 
3122 	INP_WLOCK_ASSERT(inp);
3123 
3124 	/*
3125 	 * If there is already a send tag, or the INP is being torn
3126 	 * down, allocating a new send tag is not allowed. Else send
3127 	 * tags may leak.
3128 	 */
3129 	if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3130 		return (EINVAL);
3131 
3132 	error = m_snd_tag_alloc(ifp, &params, st);
3133 #ifdef INET
3134 	if (error == 0) {
3135 		counter_u64_add(rate_limit_set_ok, 1);
3136 		counter_u64_add(rate_limit_active, 1);
3137 	} else if (error != EOPNOTSUPP)
3138 		  counter_u64_add(rate_limit_alloc_fail, 1);
3139 #endif
3140 	return (error);
3141 }
3142 
3143 void
3144 in_pcbdetach_tag(struct m_snd_tag *mst)
3145 {
3146 
3147 	m_snd_tag_rele(mst);
3148 #ifdef INET
3149 	counter_u64_add(rate_limit_active, -1);
3150 #endif
3151 }
3152 
3153 /*
3154  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3155  * if any:
3156  */
3157 void
3158 in_pcbdetach_txrtlmt(struct inpcb *inp)
3159 {
3160 	struct m_snd_tag *mst;
3161 
3162 	INP_WLOCK_ASSERT(inp);
3163 
3164 	mst = inp->inp_snd_tag;
3165 	inp->inp_snd_tag = NULL;
3166 
3167 	if (mst == NULL)
3168 		return;
3169 
3170 	m_snd_tag_rele(mst);
3171 #ifdef INET
3172 	counter_u64_add(rate_limit_active, -1);
3173 #endif
3174 }
3175 
3176 int
3177 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3178 {
3179 	int error;
3180 
3181 	/*
3182 	 * If the existing send tag is for the wrong interface due to
3183 	 * a route change, first drop the existing tag.  Set the
3184 	 * CHANGED flag so that we will keep trying to allocate a new
3185 	 * tag if we fail to allocate one this time.
3186 	 */
3187 	if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3188 		in_pcbdetach_txrtlmt(inp);
3189 		inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3190 	}
3191 
3192 	/*
3193 	 * NOTE: When attaching to a network interface a reference is
3194 	 * made to ensure the network interface doesn't go away until
3195 	 * all ratelimit connections are gone. The network interface
3196 	 * pointers compared below represent valid network interfaces,
3197 	 * except when comparing towards NULL.
3198 	 */
3199 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3200 		error = 0;
3201 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3202 		if (inp->inp_snd_tag != NULL)
3203 			in_pcbdetach_txrtlmt(inp);
3204 		error = 0;
3205 	} else if (inp->inp_snd_tag == NULL) {
3206 		/*
3207 		 * In order to utilize packet pacing with RSS, we need
3208 		 * to wait until there is a valid RSS hash before we
3209 		 * can proceed:
3210 		 */
3211 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3212 			error = EAGAIN;
3213 		} else {
3214 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3215 			    mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3216 		}
3217 	} else {
3218 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3219 	}
3220 	if (error == 0 || error == EOPNOTSUPP)
3221 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3222 
3223 	return (error);
3224 }
3225 
3226 /*
3227  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3228  * is set in the fast path and will attach/detach/modify the TX rate
3229  * limit send tag based on the socket's so_max_pacing_rate value.
3230  */
3231 void
3232 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3233 {
3234 	struct socket *socket;
3235 	uint32_t max_pacing_rate;
3236 	bool did_upgrade;
3237 
3238 	if (inp == NULL)
3239 		return;
3240 
3241 	socket = inp->inp_socket;
3242 	if (socket == NULL)
3243 		return;
3244 
3245 	if (!INP_WLOCKED(inp)) {
3246 		/*
3247 		 * NOTE: If the write locking fails, we need to bail
3248 		 * out and use the non-ratelimited ring for the
3249 		 * transmit until there is a new chance to get the
3250 		 * write lock.
3251 		 */
3252 		if (!INP_TRY_UPGRADE(inp))
3253 			return;
3254 		did_upgrade = 1;
3255 	} else {
3256 		did_upgrade = 0;
3257 	}
3258 
3259 	/*
3260 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3261 	 * because atomic updates are not required since the variable
3262 	 * is checked at every mbuf we send. It is assumed that the
3263 	 * variable read itself will be atomic.
3264 	 */
3265 	max_pacing_rate = socket->so_max_pacing_rate;
3266 
3267 	in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3268 
3269 	if (did_upgrade)
3270 		INP_DOWNGRADE(inp);
3271 }
3272 
3273 /*
3274  * Track route changes for TX rate limiting.
3275  */
3276 void
3277 in_pcboutput_eagain(struct inpcb *inp)
3278 {
3279 	bool did_upgrade;
3280 
3281 	if (inp == NULL)
3282 		return;
3283 
3284 	if (inp->inp_snd_tag == NULL)
3285 		return;
3286 
3287 	if (!INP_WLOCKED(inp)) {
3288 		/*
3289 		 * NOTE: If the write locking fails, we need to bail
3290 		 * out and use the non-ratelimited ring for the
3291 		 * transmit until there is a new chance to get the
3292 		 * write lock.
3293 		 */
3294 		if (!INP_TRY_UPGRADE(inp))
3295 			return;
3296 		did_upgrade = 1;
3297 	} else {
3298 		did_upgrade = 0;
3299 	}
3300 
3301 	/* detach rate limiting */
3302 	in_pcbdetach_txrtlmt(inp);
3303 
3304 	/* make sure new mbuf send tag allocation is made */
3305 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3306 
3307 	if (did_upgrade)
3308 		INP_DOWNGRADE(inp);
3309 }
3310 
3311 #ifdef INET
3312 static void
3313 rl_init(void *st)
3314 {
3315 	rate_limit_new = counter_u64_alloc(M_WAITOK);
3316 	rate_limit_chg = counter_u64_alloc(M_WAITOK);
3317 	rate_limit_active = counter_u64_alloc(M_WAITOK);
3318 	rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3319 	rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3320 }
3321 
3322 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3323 #endif
3324 #endif /* RATELIMIT */
3325