xref: /freebsd/sys/netinet/in_pcb.c (revision 2b8331622f0b212cf3bb4fc4914a501e5321d506)
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 #ifdef INET
864 /*
865  * Set up a bind operation on a PCB, performing port allocation
866  * as required, but do not actually modify the PCB. Callers can
867  * either complete the bind by setting inp_laddr/inp_lport and
868  * calling in_pcbinshash(), or they can just use the resulting
869  * port and address to authorise the sending of a once-off packet.
870  *
871  * On error, the values of *laddrp and *lportp are not changed.
872  */
873 int
874 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp,
875     u_short *lportp, struct ucred *cred)
876 {
877 	struct socket *so = inp->inp_socket;
878 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
879 	struct in_addr laddr;
880 	u_short lport = 0;
881 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
882 	int error;
883 
884 	/*
885 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
886 	 * so that we don't have to add to the (already messy) code below.
887 	 */
888 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
889 
890 	/*
891 	 * No state changes, so read locks are sufficient here.
892 	 */
893 	INP_LOCK_ASSERT(inp);
894 	INP_HASH_LOCK_ASSERT(pcbinfo);
895 
896 	laddr.s_addr = *laddrp;
897 	if (sin != NULL && laddr.s_addr != INADDR_ANY)
898 		return (EINVAL);
899 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
900 		lookupflags = INPLOOKUP_WILDCARD;
901 	if (sin == NULL) {
902 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
903 			return (error);
904 	} else {
905 		KASSERT(sin->sin_family == AF_INET,
906 		    ("%s: invalid family for address %p", __func__, sin));
907 		KASSERT(sin->sin_len == sizeof(*sin),
908 		    ("%s: invalid length for address %p", __func__, sin));
909 
910 		error = prison_local_ip4(cred, &sin->sin_addr);
911 		if (error)
912 			return (error);
913 		if (sin->sin_port != *lportp) {
914 			/* Don't allow the port to change. */
915 			if (*lportp != 0)
916 				return (EINVAL);
917 			lport = sin->sin_port;
918 		}
919 		/* NB: lport is left as 0 if the port isn't being changed. */
920 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
921 			/*
922 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
923 			 * allow complete duplication of binding if
924 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
925 			 * and a multicast address is bound on both
926 			 * new and duplicated sockets.
927 			 */
928 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
929 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
930 			/*
931 			 * XXX: How to deal with SO_REUSEPORT_LB here?
932 			 * Treat same as SO_REUSEPORT for now.
933 			 */
934 			if ((so->so_options &
935 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
936 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
937 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
938 			sin->sin_port = 0;		/* yech... */
939 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
940 			/*
941 			 * Is the address a local IP address?
942 			 * If INP_BINDANY is set, then the socket may be bound
943 			 * to any endpoint address, local or not.
944 			 */
945 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
946 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
947 				return (EADDRNOTAVAIL);
948 		}
949 		laddr = sin->sin_addr;
950 		if (lport) {
951 			struct inpcb *t;
952 
953 			/* GROSS */
954 			if (ntohs(lport) <= V_ipport_reservedhigh &&
955 			    ntohs(lport) >= V_ipport_reservedlow &&
956 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
957 				return (EACCES);
958 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
959 			    priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
960 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
961 				    lport, INPLOOKUP_WILDCARD, cred);
962 	/*
963 	 * XXX
964 	 * This entire block sorely needs a rewrite.
965 	 */
966 				if (t != NULL &&
967 				    (so->so_type != SOCK_STREAM ||
968 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
969 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
970 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
971 				     (t->inp_flags2 & INP_REUSEPORT) ||
972 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
973 				    (inp->inp_cred->cr_uid !=
974 				     t->inp_cred->cr_uid))
975 					return (EADDRINUSE);
976 			}
977 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
978 			    lport, lookupflags, cred);
979 			if (t != NULL && (reuseport & inp_so_options(t)) == 0 &&
980 			    (reuseport_lb & inp_so_options(t)) == 0) {
981 #ifdef INET6
982 				if (ntohl(sin->sin_addr.s_addr) !=
983 				    INADDR_ANY ||
984 				    ntohl(t->inp_laddr.s_addr) !=
985 				    INADDR_ANY ||
986 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
987 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
988 #endif
989 						return (EADDRINUSE);
990 			}
991 		}
992 	}
993 	if (*lportp != 0)
994 		lport = *lportp;
995 	if (lport == 0) {
996 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
997 		if (error != 0)
998 			return (error);
999 	}
1000 	*laddrp = laddr.s_addr;
1001 	*lportp = lport;
1002 	return (0);
1003 }
1004 
1005 /*
1006  * Connect from a socket to a specified address.
1007  * Both address and port must be specified in argument sin.
1008  * If don't have a local address for this socket yet,
1009  * then pick one.
1010  */
1011 int
1012 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1013     bool rehash)
1014 {
1015 	u_short lport, fport;
1016 	in_addr_t laddr, faddr;
1017 	int anonport, error;
1018 
1019 	INP_WLOCK_ASSERT(inp);
1020 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1021 
1022 	lport = inp->inp_lport;
1023 	laddr = inp->inp_laddr.s_addr;
1024 	anonport = (lport == 0);
1025 	error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1026 	    cred);
1027 	if (error)
1028 		return (error);
1029 
1030 	/* Do the initial binding of the local address if required. */
1031 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1032 		KASSERT(rehash == true,
1033 		    ("Rehashing required for unbound inps"));
1034 		inp->inp_lport = lport;
1035 		inp->inp_laddr.s_addr = laddr;
1036 		if (in_pcbinshash(inp) != 0) {
1037 			inp->inp_laddr.s_addr = INADDR_ANY;
1038 			inp->inp_lport = 0;
1039 			return (EAGAIN);
1040 		}
1041 	}
1042 
1043 	/* Commit the remaining changes. */
1044 	inp->inp_lport = lport;
1045 	inp->inp_laddr.s_addr = laddr;
1046 	inp->inp_faddr.s_addr = faddr;
1047 	inp->inp_fport = fport;
1048 	if (rehash) {
1049 		in_pcbrehash(inp);
1050 	} else {
1051 		in_pcbinshash(inp);
1052 	}
1053 
1054 	if (anonport)
1055 		inp->inp_flags |= INP_ANONPORT;
1056 	return (0);
1057 }
1058 
1059 /*
1060  * Do proper source address selection on an unbound socket in case
1061  * of connect. Take jails into account as well.
1062  */
1063 int
1064 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1065     struct ucred *cred)
1066 {
1067 	struct ifaddr *ifa;
1068 	struct sockaddr *sa;
1069 	struct sockaddr_in *sin, dst;
1070 	struct nhop_object *nh;
1071 	int error;
1072 
1073 	NET_EPOCH_ASSERT();
1074 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1075 
1076 	/*
1077 	 * Bypass source address selection and use the primary jail IP
1078 	 * if requested.
1079 	 */
1080 	if (!prison_saddrsel_ip4(cred, laddr))
1081 		return (0);
1082 
1083 	error = 0;
1084 
1085 	nh = NULL;
1086 	bzero(&dst, sizeof(dst));
1087 	sin = &dst;
1088 	sin->sin_family = AF_INET;
1089 	sin->sin_len = sizeof(struct sockaddr_in);
1090 	sin->sin_addr.s_addr = faddr->s_addr;
1091 
1092 	/*
1093 	 * If route is known our src addr is taken from the i/f,
1094 	 * else punt.
1095 	 *
1096 	 * Find out route to destination.
1097 	 */
1098 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1099 		nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1100 		    0, NHR_NONE, 0);
1101 
1102 	/*
1103 	 * If we found a route, use the address corresponding to
1104 	 * the outgoing interface.
1105 	 *
1106 	 * Otherwise assume faddr is reachable on a directly connected
1107 	 * network and try to find a corresponding interface to take
1108 	 * the source address from.
1109 	 */
1110 	if (nh == NULL || nh->nh_ifp == NULL) {
1111 		struct in_ifaddr *ia;
1112 		struct ifnet *ifp;
1113 
1114 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1115 					inp->inp_socket->so_fibnum));
1116 		if (ia == NULL) {
1117 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1118 						inp->inp_socket->so_fibnum));
1119 		}
1120 		if (ia == NULL) {
1121 			error = ENETUNREACH;
1122 			goto done;
1123 		}
1124 
1125 		if (!prison_flag(cred, PR_IP4)) {
1126 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1127 			goto done;
1128 		}
1129 
1130 		ifp = ia->ia_ifp;
1131 		ia = NULL;
1132 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1133 			sa = ifa->ifa_addr;
1134 			if (sa->sa_family != AF_INET)
1135 				continue;
1136 			sin = (struct sockaddr_in *)sa;
1137 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1138 				ia = (struct in_ifaddr *)ifa;
1139 				break;
1140 			}
1141 		}
1142 		if (ia != NULL) {
1143 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1144 			goto done;
1145 		}
1146 
1147 		/* 3. As a last resort return the 'default' jail address. */
1148 		error = prison_get_ip4(cred, laddr);
1149 		goto done;
1150 	}
1151 
1152 	/*
1153 	 * If the outgoing interface on the route found is not
1154 	 * a loopback interface, use the address from that interface.
1155 	 * In case of jails do those three steps:
1156 	 * 1. check if the interface address belongs to the jail. If so use it.
1157 	 * 2. check if we have any address on the outgoing interface
1158 	 *    belonging to this jail. If so use it.
1159 	 * 3. as a last resort return the 'default' jail address.
1160 	 */
1161 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1162 		struct in_ifaddr *ia;
1163 		struct ifnet *ifp;
1164 
1165 		/* If not jailed, use the default returned. */
1166 		if (!prison_flag(cred, PR_IP4)) {
1167 			ia = (struct in_ifaddr *)nh->nh_ifa;
1168 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1169 			goto done;
1170 		}
1171 
1172 		/* Jailed. */
1173 		/* 1. Check if the iface address belongs to the jail. */
1174 		sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1175 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1176 			ia = (struct in_ifaddr *)nh->nh_ifa;
1177 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1178 			goto done;
1179 		}
1180 
1181 		/*
1182 		 * 2. Check if we have any address on the outgoing interface
1183 		 *    belonging to this jail.
1184 		 */
1185 		ia = NULL;
1186 		ifp = nh->nh_ifp;
1187 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1188 			sa = ifa->ifa_addr;
1189 			if (sa->sa_family != AF_INET)
1190 				continue;
1191 			sin = (struct sockaddr_in *)sa;
1192 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1193 				ia = (struct in_ifaddr *)ifa;
1194 				break;
1195 			}
1196 		}
1197 		if (ia != NULL) {
1198 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1199 			goto done;
1200 		}
1201 
1202 		/* 3. As a last resort return the 'default' jail address. */
1203 		error = prison_get_ip4(cred, laddr);
1204 		goto done;
1205 	}
1206 
1207 	/*
1208 	 * The outgoing interface is marked with 'loopback net', so a route
1209 	 * to ourselves is here.
1210 	 * Try to find the interface of the destination address and then
1211 	 * take the address from there. That interface is not necessarily
1212 	 * a loopback interface.
1213 	 * In case of jails, check that it is an address of the jail
1214 	 * and if we cannot find, fall back to the 'default' jail address.
1215 	 */
1216 	if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1217 		struct in_ifaddr *ia;
1218 
1219 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1220 					inp->inp_socket->so_fibnum));
1221 		if (ia == NULL)
1222 			ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1223 						inp->inp_socket->so_fibnum));
1224 		if (ia == NULL)
1225 			ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1226 
1227 		if (!prison_flag(cred, PR_IP4)) {
1228 			if (ia == NULL) {
1229 				error = ENETUNREACH;
1230 				goto done;
1231 			}
1232 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1233 			goto done;
1234 		}
1235 
1236 		/* Jailed. */
1237 		if (ia != NULL) {
1238 			struct ifnet *ifp;
1239 
1240 			ifp = ia->ia_ifp;
1241 			ia = NULL;
1242 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1243 				sa = ifa->ifa_addr;
1244 				if (sa->sa_family != AF_INET)
1245 					continue;
1246 				sin = (struct sockaddr_in *)sa;
1247 				if (prison_check_ip4(cred,
1248 				    &sin->sin_addr) == 0) {
1249 					ia = (struct in_ifaddr *)ifa;
1250 					break;
1251 				}
1252 			}
1253 			if (ia != NULL) {
1254 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1255 				goto done;
1256 			}
1257 		}
1258 
1259 		/* 3. As a last resort return the 'default' jail address. */
1260 		error = prison_get_ip4(cred, laddr);
1261 		goto done;
1262 	}
1263 
1264 done:
1265 	if (error == 0 && laddr->s_addr == INADDR_ANY)
1266 		return (EHOSTUNREACH);
1267 	return (error);
1268 }
1269 
1270 /*
1271  * Set up for a connect from a socket to the specified address.
1272  * On entry, *laddrp and *lportp should contain the current local
1273  * address and port for the PCB; these are updated to the values
1274  * that should be placed in inp_laddr and inp_lport to complete
1275  * the connect.
1276  *
1277  * On success, *faddrp and *fportp will be set to the remote address
1278  * and port. These are not updated in the error case.
1279  */
1280 int
1281 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1282     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1283     struct ucred *cred)
1284 {
1285 	struct in_ifaddr *ia;
1286 	struct in_addr laddr, faddr;
1287 	u_short lport, fport;
1288 	int error;
1289 
1290 	KASSERT(sin->sin_family == AF_INET,
1291 	    ("%s: invalid address family for %p", __func__, sin));
1292 	KASSERT(sin->sin_len == sizeof(*sin),
1293 	    ("%s: invalid address length for %p", __func__, sin));
1294 
1295 	/*
1296 	 * Because a global state change doesn't actually occur here, a read
1297 	 * lock is sufficient.
1298 	 */
1299 	NET_EPOCH_ASSERT();
1300 	INP_LOCK_ASSERT(inp);
1301 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1302 
1303 	if (sin->sin_port == 0)
1304 		return (EADDRNOTAVAIL);
1305 	laddr.s_addr = *laddrp;
1306 	lport = *lportp;
1307 	faddr = sin->sin_addr;
1308 	fport = sin->sin_port;
1309 #ifdef ROUTE_MPATH
1310 	if (CALC_FLOWID_OUTBOUND) {
1311 		uint32_t hash_val, hash_type;
1312 
1313 		hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1314 		    inp->inp_socket->so_proto->pr_protocol, &hash_type);
1315 
1316 		inp->inp_flowid = hash_val;
1317 		inp->inp_flowtype = hash_type;
1318 	}
1319 #endif
1320 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1321 		/*
1322 		 * If the destination address is INADDR_ANY,
1323 		 * use the primary local address.
1324 		 * If the supplied address is INADDR_BROADCAST,
1325 		 * and the primary interface supports broadcast,
1326 		 * choose the broadcast address for that interface.
1327 		 */
1328 		if (faddr.s_addr == INADDR_ANY) {
1329 			faddr =
1330 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1331 			if ((error = prison_get_ip4(cred, &faddr)) != 0)
1332 				return (error);
1333 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1334 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1335 			    IFF_BROADCAST)
1336 				faddr = satosin(&CK_STAILQ_FIRST(
1337 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1338 		}
1339 	}
1340 	if (laddr.s_addr == INADDR_ANY) {
1341 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1342 		/*
1343 		 * If the destination address is multicast and an outgoing
1344 		 * interface has been set as a multicast option, prefer the
1345 		 * address of that interface as our source address.
1346 		 */
1347 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1348 		    inp->inp_moptions != NULL) {
1349 			struct ip_moptions *imo;
1350 			struct ifnet *ifp;
1351 
1352 			imo = inp->inp_moptions;
1353 			if (imo->imo_multicast_ifp != NULL) {
1354 				ifp = imo->imo_multicast_ifp;
1355 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1356 					if (ia->ia_ifp == ifp &&
1357 					    prison_check_ip4(cred,
1358 					    &ia->ia_addr.sin_addr) == 0)
1359 						break;
1360 				}
1361 				if (ia == NULL)
1362 					error = EADDRNOTAVAIL;
1363 				else {
1364 					laddr = ia->ia_addr.sin_addr;
1365 					error = 0;
1366 				}
1367 			}
1368 		}
1369 		if (error)
1370 			return (error);
1371 	}
1372 
1373 	if (lport != 0) {
1374 		if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1375 		    fport, laddr, lport, 0, M_NODOM) != NULL)
1376 			return (EADDRINUSE);
1377 	} else {
1378 		struct sockaddr_in lsin, fsin;
1379 
1380 		bzero(&lsin, sizeof(lsin));
1381 		bzero(&fsin, sizeof(fsin));
1382 		lsin.sin_family = AF_INET;
1383 		lsin.sin_addr = laddr;
1384 		fsin.sin_family = AF_INET;
1385 		fsin.sin_addr = faddr;
1386 		error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1387 		    &lport, (struct sockaddr *)& fsin, fport, cred,
1388 		    INPLOOKUP_WILDCARD);
1389 		if (error)
1390 			return (error);
1391 	}
1392 	*laddrp = laddr.s_addr;
1393 	*lportp = lport;
1394 	*faddrp = faddr.s_addr;
1395 	*fportp = fport;
1396 	return (0);
1397 }
1398 
1399 void
1400 in_pcbdisconnect(struct inpcb *inp)
1401 {
1402 
1403 	INP_WLOCK_ASSERT(inp);
1404 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1405 
1406 	inp->inp_laddr.s_addr = INADDR_ANY;
1407 	inp->inp_faddr.s_addr = INADDR_ANY;
1408 	inp->inp_fport = 0;
1409 	in_pcbrehash(inp);
1410 }
1411 #endif /* INET */
1412 
1413 /*
1414  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1415  * For most protocols, this will be invoked immediately prior to calling
1416  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1417  * socket, in which case in_pcbfree() is deferred.
1418  */
1419 void
1420 in_pcbdetach(struct inpcb *inp)
1421 {
1422 
1423 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1424 
1425 #ifdef RATELIMIT
1426 	if (inp->inp_snd_tag != NULL)
1427 		in_pcbdetach_txrtlmt(inp);
1428 #endif
1429 	inp->inp_socket->so_pcb = NULL;
1430 	inp->inp_socket = NULL;
1431 }
1432 
1433 /*
1434  * inpcb hash lookups are protected by SMR section.
1435  *
1436  * Once desired pcb has been found, switching from SMR section to a pcb
1437  * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1438  * here because SMR is a critical section.
1439  * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1440  */
1441 static inline void
1442 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1443 {
1444 
1445 	lock == INPLOOKUP_RLOCKPCB ?
1446 	    rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1447 }
1448 
1449 static inline void
1450 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1451 {
1452 
1453 	lock == INPLOOKUP_RLOCKPCB ?
1454 	    rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1455 }
1456 
1457 static inline int
1458 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1459 {
1460 
1461 	return (lock == INPLOOKUP_RLOCKPCB ?
1462 	    rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1463 }
1464 
1465 static inline bool
1466 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1467 {
1468 
1469 	return (lock == INPLOOKUP_RLOCKPCB ?
1470 	    in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1471 }
1472 
1473 static inline bool
1474 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1475 {
1476 
1477 	MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1478 	SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1479 
1480 	if (__predict_true(inp_trylock(inp, lock))) {
1481 		if (__predict_false(inp->inp_flags & ignflags)) {
1482 			smr_exit(inp->inp_pcbinfo->ipi_smr);
1483 			inp_unlock(inp, lock);
1484 			return (false);
1485 		}
1486 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1487 		return (true);
1488 	}
1489 
1490 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1491 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1492 		inp_lock(inp, lock);
1493 		if (__predict_false(in_pcbrele(inp, lock)))
1494 			return (false);
1495 		/*
1496 		 * inp acquired through refcount & lock for sure didn't went
1497 		 * through uma_zfree().  However, it may have already went
1498 		 * through in_pcbfree() and has another reference, that
1499 		 * prevented its release by our in_pcbrele().
1500 		 */
1501 		if (__predict_false(inp->inp_flags & ignflags)) {
1502 			inp_unlock(inp, lock);
1503 			return (false);
1504 		}
1505 		return (true);
1506 	} else {
1507 		smr_exit(inp->inp_pcbinfo->ipi_smr);
1508 		return (false);
1509 	}
1510 }
1511 
1512 bool
1513 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1514 {
1515 
1516 	/*
1517 	 * in_pcblookup() family of functions ignore not only freed entries,
1518 	 * that may be found due to lockless access to the hash, but dropped
1519 	 * entries, too.
1520 	 */
1521 	return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1522 }
1523 
1524 /*
1525  * inp_next() - inpcb hash/list traversal iterator
1526  *
1527  * Requires initialized struct inpcb_iterator for context.
1528  * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1529  *
1530  * - Iterator can have either write-lock or read-lock semantics, that can not
1531  *   be changed later.
1532  * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1533  *   a single hash slot.  Note: only rip_input() does the latter.
1534  * - Iterator may have optional bool matching function.  The matching function
1535  *   will be executed for each inpcb in the SMR context, so it can not acquire
1536  *   locks and can safely access only immutable fields of inpcb.
1537  *
1538  * A fresh initialized iterator has NULL inpcb in its context and that
1539  * means that inp_next() call would return the very first inpcb on the list
1540  * locked with desired semantic.  In all following calls the context pointer
1541  * shall hold the current inpcb pointer.  The KPI user is not supposed to
1542  * unlock the current inpcb!  Upon end of traversal inp_next() will return NULL
1543  * and write NULL to its context.  After end of traversal an iterator can be
1544  * reused.
1545  *
1546  * List traversals have the following features/constraints:
1547  * - New entries won't be seen, as they are always added to the head of a list.
1548  * - Removed entries won't stop traversal as long as they are not added to
1549  *   a different list. This is violated by in_pcbrehash().
1550  */
1551 #define	II_LIST_FIRST(ipi, hash)					\
1552 		(((hash) == INP_ALL_LIST) ?				\
1553 		    CK_LIST_FIRST(&(ipi)->ipi_listhead) :		\
1554 		    CK_LIST_FIRST(&(ipi)->ipi_hashbase[(hash)]))
1555 #define	II_LIST_NEXT(inp, hash)						\
1556 		(((hash) == INP_ALL_LIST) ?				\
1557 		    CK_LIST_NEXT((inp), inp_list) :			\
1558 		    CK_LIST_NEXT((inp), inp_hash))
1559 #define	II_LOCK_ASSERT(inp, lock)					\
1560 		rw_assert(&(inp)->inp_lock,				\
1561 		    (lock) == INPLOOKUP_RLOCKPCB ?  RA_RLOCKED : RA_WLOCKED )
1562 struct inpcb *
1563 inp_next(struct inpcb_iterator *ii)
1564 {
1565 	const struct inpcbinfo *ipi = ii->ipi;
1566 	inp_match_t *match = ii->match;
1567 	void *ctx = ii->ctx;
1568 	inp_lookup_t lock = ii->lock;
1569 	int hash = ii->hash;
1570 	struct inpcb *inp;
1571 
1572 	if (ii->inp == NULL) {		/* First call. */
1573 		smr_enter(ipi->ipi_smr);
1574 		/* This is unrolled CK_LIST_FOREACH(). */
1575 		for (inp = II_LIST_FIRST(ipi, hash);
1576 		    inp != NULL;
1577 		    inp = II_LIST_NEXT(inp, hash)) {
1578 			if (match != NULL && (match)(inp, ctx) == false)
1579 				continue;
1580 			if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1581 				break;
1582 			else {
1583 				smr_enter(ipi->ipi_smr);
1584 				MPASS(inp != II_LIST_FIRST(ipi, hash));
1585 				inp = II_LIST_FIRST(ipi, hash);
1586 				if (inp == NULL)
1587 					break;
1588 			}
1589 		}
1590 
1591 		if (inp == NULL)
1592 			smr_exit(ipi->ipi_smr);
1593 		else
1594 			ii->inp = inp;
1595 
1596 		return (inp);
1597 	}
1598 
1599 	/* Not a first call. */
1600 	smr_enter(ipi->ipi_smr);
1601 restart:
1602 	inp = ii->inp;
1603 	II_LOCK_ASSERT(inp, lock);
1604 next:
1605 	inp = II_LIST_NEXT(inp, hash);
1606 	if (inp == NULL) {
1607 		smr_exit(ipi->ipi_smr);
1608 		goto found;
1609 	}
1610 
1611 	if (match != NULL && (match)(inp, ctx) == false)
1612 		goto next;
1613 
1614 	if (__predict_true(inp_trylock(inp, lock))) {
1615 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1616 			/*
1617 			 * Entries are never inserted in middle of a list, thus
1618 			 * as long as we are in SMR, we can continue traversal.
1619 			 * Jump to 'restart' should yield in the same result,
1620 			 * but could produce unnecessary looping.  Could this
1621 			 * looping be unbound?
1622 			 */
1623 			inp_unlock(inp, lock);
1624 			goto next;
1625 		} else {
1626 			smr_exit(ipi->ipi_smr);
1627 			goto found;
1628 		}
1629 	}
1630 
1631 	/*
1632 	 * Can't obtain lock immediately, thus going hard.  Once we exit the
1633 	 * SMR section we can no longer jump to 'next', and our only stable
1634 	 * anchoring point is ii->inp, which we keep locked for this case, so
1635 	 * we jump to 'restart'.
1636 	 */
1637 	if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1638 		smr_exit(ipi->ipi_smr);
1639 		inp_lock(inp, lock);
1640 		if (__predict_false(in_pcbrele(inp, lock))) {
1641 			smr_enter(ipi->ipi_smr);
1642 			goto restart;
1643 		}
1644 		/*
1645 		 * See comment in inp_smr_lock().
1646 		 */
1647 		if (__predict_false(inp->inp_flags & INP_FREED)) {
1648 			inp_unlock(inp, lock);
1649 			smr_enter(ipi->ipi_smr);
1650 			goto restart;
1651 		}
1652 	} else
1653 		goto next;
1654 
1655 found:
1656 	inp_unlock(ii->inp, lock);
1657 	ii->inp = inp;
1658 
1659 	return (ii->inp);
1660 }
1661 
1662 /*
1663  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1664  * stability of an inpcb pointer despite the inpcb lock being released or
1665  * SMR section exited.
1666  *
1667  * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1668  */
1669 void
1670 in_pcbref(struct inpcb *inp)
1671 {
1672 	u_int old __diagused;
1673 
1674 	old = refcount_acquire(&inp->inp_refcount);
1675 	KASSERT(old > 0, ("%s: refcount 0", __func__));
1676 }
1677 
1678 /*
1679  * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1680  * freeing the pcb, if the reference was very last.
1681  */
1682 bool
1683 in_pcbrele_rlocked(struct inpcb *inp)
1684 {
1685 
1686 	INP_RLOCK_ASSERT(inp);
1687 
1688 	if (!refcount_release(&inp->inp_refcount))
1689 		return (false);
1690 
1691 	MPASS(inp->inp_flags & INP_FREED);
1692 	MPASS(inp->inp_socket == NULL);
1693 	MPASS(inp->inp_in_hpts == 0);
1694 	INP_RUNLOCK(inp);
1695 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1696 	return (true);
1697 }
1698 
1699 bool
1700 in_pcbrele_wlocked(struct inpcb *inp)
1701 {
1702 
1703 	INP_WLOCK_ASSERT(inp);
1704 
1705 	if (!refcount_release(&inp->inp_refcount))
1706 		return (false);
1707 
1708 	MPASS(inp->inp_flags & INP_FREED);
1709 	MPASS(inp->inp_socket == NULL);
1710 	MPASS(inp->inp_in_hpts == 0);
1711 	INP_WUNLOCK(inp);
1712 	uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1713 	return (true);
1714 }
1715 
1716 /*
1717  * Unconditionally schedule an inpcb to be freed by decrementing its
1718  * reference count, which should occur only after the inpcb has been detached
1719  * from its socket.  If another thread holds a temporary reference (acquired
1720  * using in_pcbref()) then the free is deferred until that reference is
1721  * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1722  *  Almost all work, including removal from global lists, is done in this
1723  * context, where the pcbinfo lock is held.
1724  */
1725 void
1726 in_pcbfree(struct inpcb *inp)
1727 {
1728 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1729 #ifdef INET
1730 	struct ip_moptions *imo;
1731 #endif
1732 #ifdef INET6
1733 	struct ip6_moptions *im6o;
1734 #endif
1735 
1736 	INP_WLOCK_ASSERT(inp);
1737 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1738 	KASSERT((inp->inp_flags & INP_FREED) == 0,
1739 	    ("%s: called twice for pcb %p", __func__, inp));
1740 
1741 	inp->inp_flags |= INP_FREED;
1742 	INP_INFO_WLOCK(pcbinfo);
1743 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1744 	pcbinfo->ipi_count--;
1745 	CK_LIST_REMOVE(inp, inp_list);
1746 	INP_INFO_WUNLOCK(pcbinfo);
1747 
1748 	if (inp->inp_flags & INP_INHASHLIST)
1749 		in_pcbremhash(inp);
1750 
1751 	RO_INVALIDATE_CACHE(&inp->inp_route);
1752 #ifdef MAC
1753 	mac_inpcb_destroy(inp);
1754 #endif
1755 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1756 	if (inp->inp_sp != NULL)
1757 		ipsec_delete_pcbpolicy(inp);
1758 #endif
1759 #ifdef INET
1760 	if (inp->inp_options)
1761 		(void)m_free(inp->inp_options);
1762 	imo = inp->inp_moptions;
1763 #endif
1764 #ifdef INET6
1765 	if (inp->inp_vflag & INP_IPV6PROTO) {
1766 		ip6_freepcbopts(inp->in6p_outputopts);
1767 		im6o = inp->in6p_moptions;
1768 	} else
1769 		im6o = NULL;
1770 #endif
1771 
1772 	if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1773 		INP_WUNLOCK(inp);
1774 	}
1775 #ifdef INET6
1776 	ip6_freemoptions(im6o);
1777 #endif
1778 #ifdef INET
1779 	inp_freemoptions(imo);
1780 #endif
1781 	/* Destruction is finalized in inpcb_dtor(). */
1782 }
1783 
1784 static void
1785 inpcb_dtor(void *mem, int size, void *arg)
1786 {
1787 	struct inpcb *inp = mem;
1788 
1789 	crfree(inp->inp_cred);
1790 #ifdef INVARIANTS
1791 	inp->inp_cred = NULL;
1792 #endif
1793 }
1794 
1795 /*
1796  * Different protocols initialize their inpcbs differently - giving
1797  * different name to the lock.  But they all are disposed the same.
1798  */
1799 static void
1800 inpcb_fini(void *mem, int size)
1801 {
1802 	struct inpcb *inp = mem;
1803 
1804 	INP_LOCK_DESTROY(inp);
1805 }
1806 
1807 /*
1808  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1809  * port reservation, and preventing it from being returned by inpcb lookups.
1810  *
1811  * It is used by TCP to mark an inpcb as unused and avoid future packet
1812  * delivery or event notification when a socket remains open but TCP has
1813  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1814  * or a RST on the wire, and allows the port binding to be reused while still
1815  * maintaining the invariant that so_pcb always points to a valid inpcb until
1816  * in_pcbdetach().
1817  *
1818  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1819  * in_pcbnotifyall() and in_pcbpurgeif0()?
1820  */
1821 void
1822 in_pcbdrop(struct inpcb *inp)
1823 {
1824 
1825 	INP_WLOCK_ASSERT(inp);
1826 #ifdef INVARIANTS
1827 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1828 		MPASS(inp->inp_refcount > 1);
1829 #endif
1830 
1831 	inp->inp_flags |= INP_DROPPED;
1832 	if (inp->inp_flags & INP_INHASHLIST)
1833 		in_pcbremhash(inp);
1834 }
1835 
1836 #ifdef INET
1837 /*
1838  * Common routines to return the socket addresses associated with inpcbs.
1839  */
1840 struct sockaddr *
1841 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1842 {
1843 	struct sockaddr_in *sin;
1844 
1845 	sin = malloc(sizeof *sin, M_SONAME,
1846 		M_WAITOK | M_ZERO);
1847 	sin->sin_family = AF_INET;
1848 	sin->sin_len = sizeof(*sin);
1849 	sin->sin_addr = *addr_p;
1850 	sin->sin_port = port;
1851 
1852 	return (struct sockaddr *)sin;
1853 }
1854 
1855 int
1856 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1857 {
1858 	struct inpcb *inp;
1859 	struct in_addr addr;
1860 	in_port_t port;
1861 
1862 	inp = sotoinpcb(so);
1863 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1864 
1865 	INP_RLOCK(inp);
1866 	port = inp->inp_lport;
1867 	addr = inp->inp_laddr;
1868 	INP_RUNLOCK(inp);
1869 
1870 	*nam = in_sockaddr(port, &addr);
1871 	return 0;
1872 }
1873 
1874 int
1875 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1876 {
1877 	struct inpcb *inp;
1878 	struct in_addr addr;
1879 	in_port_t port;
1880 
1881 	inp = sotoinpcb(so);
1882 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1883 
1884 	INP_RLOCK(inp);
1885 	port = inp->inp_fport;
1886 	addr = inp->inp_faddr;
1887 	INP_RUNLOCK(inp);
1888 
1889 	*nam = in_sockaddr(port, &addr);
1890 	return 0;
1891 }
1892 
1893 void
1894 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1895     struct inpcb *(*notify)(struct inpcb *, int))
1896 {
1897 	struct inpcb *inp, *inp_temp;
1898 
1899 	INP_INFO_WLOCK(pcbinfo);
1900 	CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1901 		INP_WLOCK(inp);
1902 #ifdef INET6
1903 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1904 			INP_WUNLOCK(inp);
1905 			continue;
1906 		}
1907 #endif
1908 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1909 		    inp->inp_socket == NULL) {
1910 			INP_WUNLOCK(inp);
1911 			continue;
1912 		}
1913 		if ((*notify)(inp, errno))
1914 			INP_WUNLOCK(inp);
1915 	}
1916 	INP_INFO_WUNLOCK(pcbinfo);
1917 }
1918 
1919 static bool
1920 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1921 {
1922 
1923 	if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1924 		return (true);
1925 	else
1926 		return (false);
1927 }
1928 
1929 void
1930 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1931 {
1932 	struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1933 	    inp_v4_multi_match, NULL);
1934 	struct inpcb *inp;
1935 	struct in_multi *inm;
1936 	struct in_mfilter *imf;
1937 	struct ip_moptions *imo;
1938 
1939 	IN_MULTI_LOCK_ASSERT();
1940 
1941 	while ((inp = inp_next(&inpi)) != NULL) {
1942 		INP_WLOCK_ASSERT(inp);
1943 
1944 		imo = inp->inp_moptions;
1945 		/*
1946 		 * Unselect the outgoing interface if it is being
1947 		 * detached.
1948 		 */
1949 		if (imo->imo_multicast_ifp == ifp)
1950 			imo->imo_multicast_ifp = NULL;
1951 
1952 		/*
1953 		 * Drop multicast group membership if we joined
1954 		 * through the interface being detached.
1955 		 *
1956 		 * XXX This can all be deferred to an epoch_call
1957 		 */
1958 restart:
1959 		IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1960 			if ((inm = imf->imf_inm) == NULL)
1961 				continue;
1962 			if (inm->inm_ifp != ifp)
1963 				continue;
1964 			ip_mfilter_remove(&imo->imo_head, imf);
1965 			in_leavegroup_locked(inm, NULL);
1966 			ip_mfilter_free(imf);
1967 			goto restart;
1968 		}
1969 	}
1970 }
1971 
1972 /*
1973  * Lookup a PCB based on the local address and port.  Caller must hold the
1974  * hash lock.  No inpcb locks or references are acquired.
1975  */
1976 #define INP_LOOKUP_MAPPED_PCB_COST	3
1977 struct inpcb *
1978 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1979     u_short lport, int lookupflags, struct ucred *cred)
1980 {
1981 	struct inpcb *inp;
1982 #ifdef INET6
1983 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1984 #else
1985 	int matchwild = 3;
1986 #endif
1987 	int wildcard;
1988 
1989 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1990 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1991 	INP_HASH_LOCK_ASSERT(pcbinfo);
1992 
1993 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1994 		struct inpcbhead *head;
1995 		/*
1996 		 * Look for an unconnected (wildcard foreign addr) PCB that
1997 		 * matches the local address and port we're looking for.
1998 		 */
1999 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2000 		    pcbinfo->ipi_hashmask)];
2001 		CK_LIST_FOREACH(inp, head, inp_hash) {
2002 #ifdef INET6
2003 			/* XXX inp locking */
2004 			if ((inp->inp_vflag & INP_IPV4) == 0)
2005 				continue;
2006 #endif
2007 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
2008 			    inp->inp_laddr.s_addr == laddr.s_addr &&
2009 			    inp->inp_lport == lport) {
2010 				/*
2011 				 * Found?
2012 				 */
2013 				if (prison_equal_ip4(cred->cr_prison,
2014 				    inp->inp_cred->cr_prison))
2015 					return (inp);
2016 			}
2017 		}
2018 		/*
2019 		 * Not found.
2020 		 */
2021 		return (NULL);
2022 	} else {
2023 		struct inpcbporthead *porthash;
2024 		struct inpcbport *phd;
2025 		struct inpcb *match = NULL;
2026 		/*
2027 		 * Best fit PCB lookup.
2028 		 *
2029 		 * First see if this local port is in use by looking on the
2030 		 * port hash list.
2031 		 */
2032 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2033 		    pcbinfo->ipi_porthashmask)];
2034 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
2035 			if (phd->phd_port == lport)
2036 				break;
2037 		}
2038 		if (phd != NULL) {
2039 			/*
2040 			 * Port is in use by one or more PCBs. Look for best
2041 			 * fit.
2042 			 */
2043 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2044 				wildcard = 0;
2045 				if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2046 				    cred->cr_prison))
2047 					continue;
2048 #ifdef INET6
2049 				/* XXX inp locking */
2050 				if ((inp->inp_vflag & INP_IPV4) == 0)
2051 					continue;
2052 				/*
2053 				 * We never select the PCB that has
2054 				 * INP_IPV6 flag and is bound to :: if
2055 				 * we have another PCB which is bound
2056 				 * to 0.0.0.0.  If a PCB has the
2057 				 * INP_IPV6 flag, then we set its cost
2058 				 * higher than IPv4 only PCBs.
2059 				 *
2060 				 * Note that the case only happens
2061 				 * when a socket is bound to ::, under
2062 				 * the condition that the use of the
2063 				 * mapped address is allowed.
2064 				 */
2065 				if ((inp->inp_vflag & INP_IPV6) != 0)
2066 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2067 #endif
2068 				if (inp->inp_faddr.s_addr != INADDR_ANY)
2069 					wildcard++;
2070 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
2071 					if (laddr.s_addr == INADDR_ANY)
2072 						wildcard++;
2073 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
2074 						continue;
2075 				} else {
2076 					if (laddr.s_addr != INADDR_ANY)
2077 						wildcard++;
2078 				}
2079 				if (wildcard < matchwild) {
2080 					match = inp;
2081 					matchwild = wildcard;
2082 					if (matchwild == 0)
2083 						break;
2084 				}
2085 			}
2086 		}
2087 		return (match);
2088 	}
2089 }
2090 #undef INP_LOOKUP_MAPPED_PCB_COST
2091 
2092 static bool
2093 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2094 {
2095 	return (domain == M_NODOM || domain == grp->il_numa_domain);
2096 }
2097 
2098 static struct inpcb *
2099 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2100     const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr,
2101     uint16_t lport, int domain)
2102 {
2103 	const struct inpcblbgrouphead *hdr;
2104 	struct inpcblbgroup *grp;
2105 	struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2106 
2107 	INP_HASH_LOCK_ASSERT(pcbinfo);
2108 
2109 	hdr = &pcbinfo->ipi_lbgrouphashbase[
2110 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2111 
2112 	/*
2113 	 * Search for an LB group match based on the following criteria:
2114 	 * - prefer jailed groups to non-jailed groups
2115 	 * - prefer exact source address matches to wildcard matches
2116 	 * - prefer groups bound to the specified NUMA domain
2117 	 */
2118 	jail_exact = jail_wild = local_exact = local_wild = NULL;
2119 	CK_LIST_FOREACH(grp, hdr, il_list) {
2120 		bool injail;
2121 
2122 #ifdef INET6
2123 		if (!(grp->il_vflag & INP_IPV4))
2124 			continue;
2125 #endif
2126 		if (grp->il_lport != lport)
2127 			continue;
2128 
2129 		injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2130 		if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2131 		    laddr) != 0)
2132 			continue;
2133 
2134 		if (grp->il_laddr.s_addr == laddr->s_addr) {
2135 			if (injail) {
2136 				jail_exact = grp;
2137 				if (in_pcblookup_lb_numa_match(grp, domain))
2138 					/* This is a perfect match. */
2139 					goto out;
2140 			} else if (local_exact == NULL ||
2141 			    in_pcblookup_lb_numa_match(grp, domain)) {
2142 				local_exact = grp;
2143 			}
2144 		} else if (grp->il_laddr.s_addr == INADDR_ANY) {
2145 			if (injail) {
2146 				if (jail_wild == NULL ||
2147 				    in_pcblookup_lb_numa_match(grp, domain))
2148 					jail_wild = grp;
2149 			} else if (local_wild == NULL ||
2150 			    in_pcblookup_lb_numa_match(grp, domain)) {
2151 				local_wild = grp;
2152 			}
2153 		}
2154 	}
2155 
2156 	if (jail_exact != NULL)
2157 		grp = jail_exact;
2158 	else if (jail_wild != NULL)
2159 		grp = jail_wild;
2160 	else if (local_exact != NULL)
2161 		grp = local_exact;
2162 	else
2163 		grp = local_wild;
2164 	if (grp == NULL)
2165 		return (NULL);
2166 out:
2167 	return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2168 	    grp->il_inpcnt]);
2169 }
2170 
2171 static struct inpcb *
2172 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2173     u_short fport, struct in_addr laddr, u_short lport)
2174 {
2175 	struct inpcbhead *head;
2176 	struct inpcb *inp, *match;
2177 
2178 	INP_HASH_LOCK_ASSERT(pcbinfo);
2179 
2180 	match = NULL;
2181 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&faddr, lport, fport,
2182 	    pcbinfo->ipi_hashmask)];
2183 	CK_LIST_FOREACH(inp, head, inp_hash) {
2184 #ifdef INET6
2185 		/* XXX inp locking */
2186 		if ((inp->inp_vflag & INP_IPV4) == 0)
2187 			continue;
2188 #endif
2189 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2190 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2191 		    inp->inp_fport == fport &&
2192 		    inp->inp_lport == lport)
2193 			return (inp);
2194 	}
2195 	return (match);
2196 }
2197 
2198 static struct inpcb *
2199 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2200     u_short fport, struct in_addr laddr, u_short lport)
2201 {
2202 	struct inpcbhead *head;
2203 	struct inpcb *inp, *local_wild, *local_exact, *jail_wild;
2204 #ifdef INET6
2205 	struct inpcb *local_wild_mapped;
2206 #endif
2207 
2208 	INP_HASH_LOCK_ASSERT(pcbinfo);
2209 
2210 	/*
2211 	 * Order of socket selection - we always prefer jails.
2212 	 *      1. jailed, non-wild.
2213 	 *      2. jailed, wild.
2214 	 *      3. non-jailed, non-wild.
2215 	 *      4. non-jailed, wild.
2216 	 */
2217 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2218 	    pcbinfo->ipi_hashmask)];
2219 	local_wild = local_exact = jail_wild = NULL;
2220 #ifdef INET6
2221 	local_wild_mapped = NULL;
2222 #endif
2223 	CK_LIST_FOREACH(inp, head, inp_hash) {
2224 		bool injail;
2225 
2226 #ifdef INET6
2227 		/* XXX inp locking */
2228 		if ((inp->inp_vflag & INP_IPV4) == 0)
2229 			continue;
2230 #endif
2231 		if (inp->inp_faddr.s_addr != INADDR_ANY ||
2232 		    inp->inp_lport != lport)
2233 			continue;
2234 
2235 		injail = prison_flag(inp->inp_cred, PR_IP4) != 0;
2236 		if (injail) {
2237 			if (prison_check_ip4_locked(inp->inp_cred->cr_prison,
2238 			    &laddr) != 0)
2239 				continue;
2240 		} else {
2241 			if (local_exact != NULL)
2242 				continue;
2243 		}
2244 
2245 		if (inp->inp_laddr.s_addr == laddr.s_addr) {
2246 			if (injail)
2247 				return (inp);
2248 			local_exact = inp;
2249 		} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2250 #ifdef INET6
2251 			/* XXX inp locking, NULL check */
2252 			if (inp->inp_vflag & INP_IPV6PROTO)
2253 				local_wild_mapped = inp;
2254 			else
2255 #endif
2256 				if (injail)
2257 					jail_wild = inp;
2258 				else
2259 					local_wild = inp;
2260 		}
2261 	}
2262 	if (jail_wild != NULL)
2263 		return (jail_wild);
2264 	if (local_exact != NULL)
2265 		return (local_exact);
2266 	if (local_wild != NULL)
2267 		return (local_wild);
2268 #ifdef INET6
2269 	if (local_wild_mapped != NULL)
2270 		return (local_wild_mapped);
2271 #endif
2272 	return (NULL);
2273 }
2274 
2275 /*
2276  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2277  * that the caller has either locked the hash list, which usually happens
2278  * for bind(2) operations, or is in SMR section, which happens when sorting
2279  * out incoming packets.
2280  */
2281 static struct inpcb *
2282 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2283     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2284     uint8_t numa_domain)
2285 {
2286 	struct inpcb *inp;
2287 	const u_short fport = fport_arg, lport = lport_arg;
2288 
2289 	KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0,
2290 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2291 	KASSERT(faddr.s_addr != INADDR_ANY,
2292 	    ("%s: invalid foreign address", __func__));
2293 	KASSERT(laddr.s_addr != INADDR_ANY,
2294 	    ("%s: invalid local address", __func__));
2295 	INP_HASH_LOCK_ASSERT(pcbinfo);
2296 
2297 	inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2298 	if (inp != NULL)
2299 		return (inp);
2300 
2301 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2302 		inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, &laddr,
2303 		    lport, numa_domain);
2304 		if (inp == NULL) {
2305 			inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr,
2306 			    fport, laddr, lport);
2307 		}
2308 	}
2309 
2310 	return (inp);
2311 }
2312 
2313 static struct inpcb *
2314 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2315     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2316     uint8_t numa_domain)
2317 {
2318 	struct inpcb *inp;
2319 
2320 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2321 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2322 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2323 	    ("%s: LOCKPCB not set", __func__));
2324 
2325 	smr_enter(pcbinfo->ipi_smr);
2326 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2327 	    lookupflags & INPLOOKUP_WILDCARD, numa_domain);
2328 	if (inp != NULL) {
2329 		if (__predict_false(inp_smr_lock(inp,
2330 		    (lookupflags & INPLOOKUP_LOCKMASK)) == false))
2331 			inp = NULL;
2332 	} else
2333 		smr_exit(pcbinfo->ipi_smr);
2334 
2335 	return (inp);
2336 }
2337 
2338 /*
2339  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2340  * from which a pre-calculated hash value may be extracted.
2341  */
2342 struct inpcb *
2343 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2344     struct in_addr laddr, u_int lport, int lookupflags,
2345     struct ifnet *ifp __unused)
2346 {
2347 	return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2348 	    lookupflags, M_NODOM));
2349 }
2350 
2351 struct inpcb *
2352 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2353     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2354     struct ifnet *ifp __unused, struct mbuf *m)
2355 {
2356 	return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2357 	    lookupflags, m->m_pkthdr.numa_domain));
2358 }
2359 #endif /* INET */
2360 
2361 /*
2362  * Insert PCB onto various hash lists.
2363  */
2364 int
2365 in_pcbinshash(struct inpcb *inp)
2366 {
2367 	struct inpcbhead *pcbhash;
2368 	struct inpcbporthead *pcbporthash;
2369 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2370 	struct inpcbport *phd;
2371 
2372 	INP_WLOCK_ASSERT(inp);
2373 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2374 
2375 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2376 	    ("in_pcbinshash: INP_INHASHLIST"));
2377 
2378 #ifdef INET6
2379 	if (inp->inp_vflag & INP_IPV6)
2380 		pcbhash = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2381 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2382 	else
2383 #endif
2384 		pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2385 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2386 
2387 	pcbporthash = &pcbinfo->ipi_porthashbase[
2388 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2389 
2390 	/*
2391 	 * Add entry to load balance group.
2392 	 * Only do this if SO_REUSEPORT_LB is set.
2393 	 */
2394 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2395 		int error = in_pcbinslbgrouphash(inp, M_NODOM);
2396 		if (error != 0)
2397 			return (error);
2398 	}
2399 
2400 	/*
2401 	 * Go through port list and look for a head for this lport.
2402 	 */
2403 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2404 		if (phd->phd_port == inp->inp_lport)
2405 			break;
2406 	}
2407 
2408 	/*
2409 	 * If none exists, malloc one and tack it on.
2410 	 */
2411 	if (phd == NULL) {
2412 		phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2413 		if (phd == NULL) {
2414 			if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2415 				in_pcbremlbgrouphash(inp);
2416 			return (ENOMEM);
2417 		}
2418 		phd->phd_port = inp->inp_lport;
2419 		CK_LIST_INIT(&phd->phd_pcblist);
2420 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2421 	}
2422 	inp->inp_phd = phd;
2423 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2424 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2425 	inp->inp_flags |= INP_INHASHLIST;
2426 
2427 	return (0);
2428 }
2429 
2430 static void
2431 in_pcbremhash(struct inpcb *inp)
2432 {
2433 	struct inpcbport *phd = inp->inp_phd;
2434 
2435 	INP_WLOCK_ASSERT(inp);
2436 	MPASS(inp->inp_flags & INP_INHASHLIST);
2437 
2438 	INP_HASH_WLOCK(inp->inp_pcbinfo);
2439 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2440 		in_pcbremlbgrouphash(inp);
2441 	CK_LIST_REMOVE(inp, inp_hash);
2442 	CK_LIST_REMOVE(inp, inp_portlist);
2443 	if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2444 		CK_LIST_REMOVE(phd, phd_hash);
2445 		uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2446 	}
2447 	INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2448 	inp->inp_flags &= ~INP_INHASHLIST;
2449 }
2450 
2451 /*
2452  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2453  * changed. NOTE: This does not handle the case of the lport changing (the
2454  * hashed port list would have to be updated as well), so the lport must
2455  * not change after in_pcbinshash() has been called.
2456  *
2457  * XXXGL: a race between this function and SMR-protected hash iterator
2458  * will lead to iterator traversing a possibly wrong hash list. However,
2459  * this race should have been here since change from rwlock to epoch.
2460  */
2461 void
2462 in_pcbrehash(struct inpcb *inp)
2463 {
2464 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2465 	struct inpcbhead *head;
2466 
2467 	INP_WLOCK_ASSERT(inp);
2468 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2469 
2470 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2471 	    ("in_pcbrehash: !INP_INHASHLIST"));
2472 
2473 #ifdef INET6
2474 	if (inp->inp_vflag & INP_IPV6)
2475 		head = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2476 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2477 	else
2478 #endif
2479 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2480 		    inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2481 
2482 	CK_LIST_REMOVE(inp, inp_hash);
2483 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2484 }
2485 
2486 /*
2487  * Check for alternatives when higher level complains
2488  * about service problems.  For now, invalidate cached
2489  * routing information.  If the route was created dynamically
2490  * (by a redirect), time to try a default gateway again.
2491  */
2492 void
2493 in_losing(struct inpcb *inp)
2494 {
2495 
2496 	RO_INVALIDATE_CACHE(&inp->inp_route);
2497 	return;
2498 }
2499 
2500 /*
2501  * A set label operation has occurred at the socket layer, propagate the
2502  * label change into the in_pcb for the socket.
2503  */
2504 void
2505 in_pcbsosetlabel(struct socket *so)
2506 {
2507 #ifdef MAC
2508 	struct inpcb *inp;
2509 
2510 	inp = sotoinpcb(so);
2511 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2512 
2513 	INP_WLOCK(inp);
2514 	SOCK_LOCK(so);
2515 	mac_inpcb_sosetlabel(so, inp);
2516 	SOCK_UNLOCK(so);
2517 	INP_WUNLOCK(inp);
2518 #endif
2519 }
2520 
2521 void
2522 inp_wlock(struct inpcb *inp)
2523 {
2524 
2525 	INP_WLOCK(inp);
2526 }
2527 
2528 void
2529 inp_wunlock(struct inpcb *inp)
2530 {
2531 
2532 	INP_WUNLOCK(inp);
2533 }
2534 
2535 void
2536 inp_rlock(struct inpcb *inp)
2537 {
2538 
2539 	INP_RLOCK(inp);
2540 }
2541 
2542 void
2543 inp_runlock(struct inpcb *inp)
2544 {
2545 
2546 	INP_RUNLOCK(inp);
2547 }
2548 
2549 #ifdef INVARIANT_SUPPORT
2550 void
2551 inp_lock_assert(struct inpcb *inp)
2552 {
2553 
2554 	INP_WLOCK_ASSERT(inp);
2555 }
2556 
2557 void
2558 inp_unlock_assert(struct inpcb *inp)
2559 {
2560 
2561 	INP_UNLOCK_ASSERT(inp);
2562 }
2563 #endif
2564 
2565 void
2566 inp_apply_all(struct inpcbinfo *pcbinfo,
2567     void (*func)(struct inpcb *, void *), void *arg)
2568 {
2569 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2570 	    INPLOOKUP_WLOCKPCB);
2571 	struct inpcb *inp;
2572 
2573 	while ((inp = inp_next(&inpi)) != NULL)
2574 		func(inp, arg);
2575 }
2576 
2577 struct socket *
2578 inp_inpcbtosocket(struct inpcb *inp)
2579 {
2580 
2581 	INP_WLOCK_ASSERT(inp);
2582 	return (inp->inp_socket);
2583 }
2584 
2585 struct tcpcb *
2586 inp_inpcbtotcpcb(struct inpcb *inp)
2587 {
2588 
2589 	INP_WLOCK_ASSERT(inp);
2590 	return ((struct tcpcb *)inp->inp_ppcb);
2591 }
2592 
2593 int
2594 inp_ip_tos_get(const struct inpcb *inp)
2595 {
2596 
2597 	return (inp->inp_ip_tos);
2598 }
2599 
2600 void
2601 inp_ip_tos_set(struct inpcb *inp, int val)
2602 {
2603 
2604 	inp->inp_ip_tos = val;
2605 }
2606 
2607 void
2608 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2609     uint32_t *faddr, uint16_t *fp)
2610 {
2611 
2612 	INP_LOCK_ASSERT(inp);
2613 	*laddr = inp->inp_laddr.s_addr;
2614 	*faddr = inp->inp_faddr.s_addr;
2615 	*lp = inp->inp_lport;
2616 	*fp = inp->inp_fport;
2617 }
2618 
2619 struct inpcb *
2620 so_sotoinpcb(struct socket *so)
2621 {
2622 
2623 	return (sotoinpcb(so));
2624 }
2625 
2626 /*
2627  * Create an external-format (``xinpcb'') structure using the information in
2628  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2629  * reduce the spew of irrelevant information over this interface, to isolate
2630  * user code from changes in the kernel structure, and potentially to provide
2631  * information-hiding if we decide that some of this information should be
2632  * hidden from users.
2633  */
2634 void
2635 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2636 {
2637 
2638 	bzero(xi, sizeof(*xi));
2639 	xi->xi_len = sizeof(struct xinpcb);
2640 	if (inp->inp_socket)
2641 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2642 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2643 	xi->inp_gencnt = inp->inp_gencnt;
2644 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2645 	xi->inp_flow = inp->inp_flow;
2646 	xi->inp_flowid = inp->inp_flowid;
2647 	xi->inp_flowtype = inp->inp_flowtype;
2648 	xi->inp_flags = inp->inp_flags;
2649 	xi->inp_flags2 = inp->inp_flags2;
2650 	xi->in6p_cksum = inp->in6p_cksum;
2651 	xi->in6p_hops = inp->in6p_hops;
2652 	xi->inp_ip_tos = inp->inp_ip_tos;
2653 	xi->inp_vflag = inp->inp_vflag;
2654 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2655 	xi->inp_ip_p = inp->inp_ip_p;
2656 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2657 }
2658 
2659 int
2660 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2661     int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2662 {
2663 	struct sockopt sopt;
2664 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2665 	    INPLOOKUP_WLOCKPCB);
2666 	struct inpcb *inp;
2667 	struct sockopt_parameters *params;
2668 	struct socket *so;
2669 	int error;
2670 	char buf[1024];
2671 
2672 	if (req->oldptr != NULL || req->oldlen != 0)
2673 		return (EINVAL);
2674 	if (req->newptr == NULL)
2675 		return (EPERM);
2676 	if (req->newlen > sizeof(buf))
2677 		return (ENOMEM);
2678 	error = SYSCTL_IN(req, buf, req->newlen);
2679 	if (error != 0)
2680 		return (error);
2681 	if (req->newlen < sizeof(struct sockopt_parameters))
2682 		return (EINVAL);
2683 	params = (struct sockopt_parameters *)buf;
2684 	sopt.sopt_level = params->sop_level;
2685 	sopt.sopt_name = params->sop_optname;
2686 	sopt.sopt_dir = SOPT_SET;
2687 	sopt.sopt_val = params->sop_optval;
2688 	sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
2689 	sopt.sopt_td = NULL;
2690 #ifdef INET6
2691 	if (params->sop_inc.inc_flags & INC_ISIPV6) {
2692 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_laddr))
2693 			params->sop_inc.inc6_laddr.s6_addr16[1] =
2694 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2695 		if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_faddr))
2696 			params->sop_inc.inc6_faddr.s6_addr16[1] =
2697 			    htons(params->sop_inc.inc6_zoneid & 0xffff);
2698 	}
2699 #endif
2700 	if (params->sop_inc.inc_lport != htons(0)) {
2701 		if (params->sop_inc.inc_fport == htons(0))
2702 			inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
2703 			    pcbinfo->ipi_hashmask);
2704 		else
2705 #ifdef INET6
2706 			if (params->sop_inc.inc_flags & INC_ISIPV6)
2707 				inpi.hash = INP6_PCBHASH(
2708 				    &params->sop_inc.inc6_faddr,
2709 				    params->sop_inc.inc_lport,
2710 				    params->sop_inc.inc_fport,
2711 				    pcbinfo->ipi_hashmask);
2712 			else
2713 #endif
2714 				inpi.hash = INP_PCBHASH(
2715 				    &params->sop_inc.inc_faddr,
2716 				    params->sop_inc.inc_lport,
2717 				    params->sop_inc.inc_fport,
2718 				    pcbinfo->ipi_hashmask);
2719 	}
2720 	while ((inp = inp_next(&inpi)) != NULL)
2721 		if (inp->inp_gencnt == params->sop_id) {
2722 			if (inp->inp_flags & INP_DROPPED) {
2723 				INP_WUNLOCK(inp);
2724 				return (ECONNRESET);
2725 			}
2726 			so = inp->inp_socket;
2727 			KASSERT(so != NULL, ("inp_socket == NULL"));
2728 			soref(so);
2729 			error = (*ctloutput_set)(inp, &sopt);
2730 			sorele(so);
2731 			break;
2732 		}
2733 	if (inp == NULL)
2734 		error = ESRCH;
2735 	return (error);
2736 }
2737 
2738 #ifdef DDB
2739 static void
2740 db_print_indent(int indent)
2741 {
2742 	int i;
2743 
2744 	for (i = 0; i < indent; i++)
2745 		db_printf(" ");
2746 }
2747 
2748 static void
2749 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2750 {
2751 	char faddr_str[48], laddr_str[48];
2752 
2753 	db_print_indent(indent);
2754 	db_printf("%s at %p\n", name, inc);
2755 
2756 	indent += 2;
2757 
2758 #ifdef INET6
2759 	if (inc->inc_flags & INC_ISIPV6) {
2760 		/* IPv6. */
2761 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2762 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2763 	} else
2764 #endif
2765 	{
2766 		/* IPv4. */
2767 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2768 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2769 	}
2770 	db_print_indent(indent);
2771 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2772 	    ntohs(inc->inc_lport));
2773 	db_print_indent(indent);
2774 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2775 	    ntohs(inc->inc_fport));
2776 }
2777 
2778 static void
2779 db_print_inpflags(int inp_flags)
2780 {
2781 	int comma;
2782 
2783 	comma = 0;
2784 	if (inp_flags & INP_RECVOPTS) {
2785 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2786 		comma = 1;
2787 	}
2788 	if (inp_flags & INP_RECVRETOPTS) {
2789 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2790 		comma = 1;
2791 	}
2792 	if (inp_flags & INP_RECVDSTADDR) {
2793 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2794 		comma = 1;
2795 	}
2796 	if (inp_flags & INP_ORIGDSTADDR) {
2797 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2798 		comma = 1;
2799 	}
2800 	if (inp_flags & INP_HDRINCL) {
2801 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2802 		comma = 1;
2803 	}
2804 	if (inp_flags & INP_HIGHPORT) {
2805 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2806 		comma = 1;
2807 	}
2808 	if (inp_flags & INP_LOWPORT) {
2809 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2810 		comma = 1;
2811 	}
2812 	if (inp_flags & INP_ANONPORT) {
2813 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2814 		comma = 1;
2815 	}
2816 	if (inp_flags & INP_RECVIF) {
2817 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2818 		comma = 1;
2819 	}
2820 	if (inp_flags & INP_MTUDISC) {
2821 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2822 		comma = 1;
2823 	}
2824 	if (inp_flags & INP_RECVTTL) {
2825 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2826 		comma = 1;
2827 	}
2828 	if (inp_flags & INP_DONTFRAG) {
2829 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2830 		comma = 1;
2831 	}
2832 	if (inp_flags & INP_RECVTOS) {
2833 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
2834 		comma = 1;
2835 	}
2836 	if (inp_flags & IN6P_IPV6_V6ONLY) {
2837 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2838 		comma = 1;
2839 	}
2840 	if (inp_flags & IN6P_PKTINFO) {
2841 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2842 		comma = 1;
2843 	}
2844 	if (inp_flags & IN6P_HOPLIMIT) {
2845 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2846 		comma = 1;
2847 	}
2848 	if (inp_flags & IN6P_HOPOPTS) {
2849 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2850 		comma = 1;
2851 	}
2852 	if (inp_flags & IN6P_DSTOPTS) {
2853 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2854 		comma = 1;
2855 	}
2856 	if (inp_flags & IN6P_RTHDR) {
2857 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2858 		comma = 1;
2859 	}
2860 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
2861 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2862 		comma = 1;
2863 	}
2864 	if (inp_flags & IN6P_TCLASS) {
2865 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2866 		comma = 1;
2867 	}
2868 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
2869 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2870 		comma = 1;
2871 	}
2872 	if (inp_flags & INP_ONESBCAST) {
2873 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2874 		comma  = 1;
2875 	}
2876 	if (inp_flags & INP_DROPPED) {
2877 		db_printf("%sINP_DROPPED", comma ? ", " : "");
2878 		comma  = 1;
2879 	}
2880 	if (inp_flags & INP_SOCKREF) {
2881 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
2882 		comma  = 1;
2883 	}
2884 	if (inp_flags & IN6P_RFC2292) {
2885 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2886 		comma = 1;
2887 	}
2888 	if (inp_flags & IN6P_MTU) {
2889 		db_printf("IN6P_MTU%s", comma ? ", " : "");
2890 		comma = 1;
2891 	}
2892 }
2893 
2894 static void
2895 db_print_inpvflag(u_char inp_vflag)
2896 {
2897 	int comma;
2898 
2899 	comma = 0;
2900 	if (inp_vflag & INP_IPV4) {
2901 		db_printf("%sINP_IPV4", comma ? ", " : "");
2902 		comma  = 1;
2903 	}
2904 	if (inp_vflag & INP_IPV6) {
2905 		db_printf("%sINP_IPV6", comma ? ", " : "");
2906 		comma  = 1;
2907 	}
2908 	if (inp_vflag & INP_IPV6PROTO) {
2909 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2910 		comma  = 1;
2911 	}
2912 }
2913 
2914 static void
2915 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2916 {
2917 
2918 	db_print_indent(indent);
2919 	db_printf("%s at %p\n", name, inp);
2920 
2921 	indent += 2;
2922 
2923 	db_print_indent(indent);
2924 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2925 
2926 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2927 
2928 	db_print_indent(indent);
2929 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2930 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2931 
2932 	db_print_indent(indent);
2933 	db_printf("inp_label: %p   inp_flags: 0x%x (",
2934 	   inp->inp_label, inp->inp_flags);
2935 	db_print_inpflags(inp->inp_flags);
2936 	db_printf(")\n");
2937 
2938 	db_print_indent(indent);
2939 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2940 	    inp->inp_vflag);
2941 	db_print_inpvflag(inp->inp_vflag);
2942 	db_printf(")\n");
2943 
2944 	db_print_indent(indent);
2945 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2946 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2947 
2948 	db_print_indent(indent);
2949 #ifdef INET6
2950 	if (inp->inp_vflag & INP_IPV6) {
2951 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
2952 		    "in6p_moptions: %p\n", inp->in6p_options,
2953 		    inp->in6p_outputopts, inp->in6p_moptions);
2954 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2955 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2956 		    inp->in6p_hops);
2957 	} else
2958 #endif
2959 	{
2960 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2961 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2962 		    inp->inp_options, inp->inp_moptions);
2963 	}
2964 
2965 	db_print_indent(indent);
2966 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2967 	    (uintmax_t)inp->inp_gencnt);
2968 }
2969 
2970 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2971 {
2972 	struct inpcb *inp;
2973 
2974 	if (!have_addr) {
2975 		db_printf("usage: show inpcb <addr>\n");
2976 		return;
2977 	}
2978 	inp = (struct inpcb *)addr;
2979 
2980 	db_print_inpcb(inp, "inpcb", 0);
2981 }
2982 #endif /* DDB */
2983 
2984 #ifdef RATELIMIT
2985 /*
2986  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2987  * if any.
2988  */
2989 int
2990 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2991 {
2992 	union if_snd_tag_modify_params params = {
2993 		.rate_limit.max_rate = max_pacing_rate,
2994 		.rate_limit.flags = M_NOWAIT,
2995 	};
2996 	struct m_snd_tag *mst;
2997 	int error;
2998 
2999 	mst = inp->inp_snd_tag;
3000 	if (mst == NULL)
3001 		return (EINVAL);
3002 
3003 	if (mst->sw->snd_tag_modify == NULL) {
3004 		error = EOPNOTSUPP;
3005 	} else {
3006 		error = mst->sw->snd_tag_modify(mst, &params);
3007 	}
3008 	return (error);
3009 }
3010 
3011 /*
3012  * Query existing TX rate limit based on the existing
3013  * "inp->inp_snd_tag", if any.
3014  */
3015 int
3016 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3017 {
3018 	union if_snd_tag_query_params params = { };
3019 	struct m_snd_tag *mst;
3020 	int error;
3021 
3022 	mst = inp->inp_snd_tag;
3023 	if (mst == NULL)
3024 		return (EINVAL);
3025 
3026 	if (mst->sw->snd_tag_query == NULL) {
3027 		error = EOPNOTSUPP;
3028 	} else {
3029 		error = mst->sw->snd_tag_query(mst, &params);
3030 		if (error == 0 && p_max_pacing_rate != NULL)
3031 			*p_max_pacing_rate = params.rate_limit.max_rate;
3032 	}
3033 	return (error);
3034 }
3035 
3036 /*
3037  * Query existing TX queue level based on the existing
3038  * "inp->inp_snd_tag", if any.
3039  */
3040 int
3041 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3042 {
3043 	union if_snd_tag_query_params params = { };
3044 	struct m_snd_tag *mst;
3045 	int error;
3046 
3047 	mst = inp->inp_snd_tag;
3048 	if (mst == NULL)
3049 		return (EINVAL);
3050 
3051 	if (mst->sw->snd_tag_query == NULL)
3052 		return (EOPNOTSUPP);
3053 
3054 	error = mst->sw->snd_tag_query(mst, &params);
3055 	if (error == 0 && p_txqueue_level != NULL)
3056 		*p_txqueue_level = params.rate_limit.queue_level;
3057 	return (error);
3058 }
3059 
3060 /*
3061  * Allocate a new TX rate limit send tag from the network interface
3062  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3063  */
3064 int
3065 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3066     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3067 
3068 {
3069 	union if_snd_tag_alloc_params params = {
3070 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3071 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3072 		.rate_limit.hdr.flowid = flowid,
3073 		.rate_limit.hdr.flowtype = flowtype,
3074 		.rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3075 		.rate_limit.max_rate = max_pacing_rate,
3076 		.rate_limit.flags = M_NOWAIT,
3077 	};
3078 	int error;
3079 
3080 	INP_WLOCK_ASSERT(inp);
3081 
3082 	/*
3083 	 * If there is already a send tag, or the INP is being torn
3084 	 * down, allocating a new send tag is not allowed. Else send
3085 	 * tags may leak.
3086 	 */
3087 	if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3088 		return (EINVAL);
3089 
3090 	error = m_snd_tag_alloc(ifp, &params, st);
3091 #ifdef INET
3092 	if (error == 0) {
3093 		counter_u64_add(rate_limit_set_ok, 1);
3094 		counter_u64_add(rate_limit_active, 1);
3095 	} else if (error != EOPNOTSUPP)
3096 		  counter_u64_add(rate_limit_alloc_fail, 1);
3097 #endif
3098 	return (error);
3099 }
3100 
3101 void
3102 in_pcbdetach_tag(struct m_snd_tag *mst)
3103 {
3104 
3105 	m_snd_tag_rele(mst);
3106 #ifdef INET
3107 	counter_u64_add(rate_limit_active, -1);
3108 #endif
3109 }
3110 
3111 /*
3112  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3113  * if any:
3114  */
3115 void
3116 in_pcbdetach_txrtlmt(struct inpcb *inp)
3117 {
3118 	struct m_snd_tag *mst;
3119 
3120 	INP_WLOCK_ASSERT(inp);
3121 
3122 	mst = inp->inp_snd_tag;
3123 	inp->inp_snd_tag = NULL;
3124 
3125 	if (mst == NULL)
3126 		return;
3127 
3128 	m_snd_tag_rele(mst);
3129 #ifdef INET
3130 	counter_u64_add(rate_limit_active, -1);
3131 #endif
3132 }
3133 
3134 int
3135 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3136 {
3137 	int error;
3138 
3139 	/*
3140 	 * If the existing send tag is for the wrong interface due to
3141 	 * a route change, first drop the existing tag.  Set the
3142 	 * CHANGED flag so that we will keep trying to allocate a new
3143 	 * tag if we fail to allocate one this time.
3144 	 */
3145 	if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3146 		in_pcbdetach_txrtlmt(inp);
3147 		inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3148 	}
3149 
3150 	/*
3151 	 * NOTE: When attaching to a network interface a reference is
3152 	 * made to ensure the network interface doesn't go away until
3153 	 * all ratelimit connections are gone. The network interface
3154 	 * pointers compared below represent valid network interfaces,
3155 	 * except when comparing towards NULL.
3156 	 */
3157 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3158 		error = 0;
3159 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3160 		if (inp->inp_snd_tag != NULL)
3161 			in_pcbdetach_txrtlmt(inp);
3162 		error = 0;
3163 	} else if (inp->inp_snd_tag == NULL) {
3164 		/*
3165 		 * In order to utilize packet pacing with RSS, we need
3166 		 * to wait until there is a valid RSS hash before we
3167 		 * can proceed:
3168 		 */
3169 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3170 			error = EAGAIN;
3171 		} else {
3172 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3173 			    mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3174 		}
3175 	} else {
3176 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3177 	}
3178 	if (error == 0 || error == EOPNOTSUPP)
3179 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3180 
3181 	return (error);
3182 }
3183 
3184 /*
3185  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3186  * is set in the fast path and will attach/detach/modify the TX rate
3187  * limit send tag based on the socket's so_max_pacing_rate value.
3188  */
3189 void
3190 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3191 {
3192 	struct socket *socket;
3193 	uint32_t max_pacing_rate;
3194 	bool did_upgrade;
3195 
3196 	if (inp == NULL)
3197 		return;
3198 
3199 	socket = inp->inp_socket;
3200 	if (socket == NULL)
3201 		return;
3202 
3203 	if (!INP_WLOCKED(inp)) {
3204 		/*
3205 		 * NOTE: If the write locking fails, we need to bail
3206 		 * out and use the non-ratelimited ring for the
3207 		 * transmit until there is a new chance to get the
3208 		 * write lock.
3209 		 */
3210 		if (!INP_TRY_UPGRADE(inp))
3211 			return;
3212 		did_upgrade = 1;
3213 	} else {
3214 		did_upgrade = 0;
3215 	}
3216 
3217 	/*
3218 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3219 	 * because atomic updates are not required since the variable
3220 	 * is checked at every mbuf we send. It is assumed that the
3221 	 * variable read itself will be atomic.
3222 	 */
3223 	max_pacing_rate = socket->so_max_pacing_rate;
3224 
3225 	in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3226 
3227 	if (did_upgrade)
3228 		INP_DOWNGRADE(inp);
3229 }
3230 
3231 /*
3232  * Track route changes for TX rate limiting.
3233  */
3234 void
3235 in_pcboutput_eagain(struct inpcb *inp)
3236 {
3237 	bool did_upgrade;
3238 
3239 	if (inp == NULL)
3240 		return;
3241 
3242 	if (inp->inp_snd_tag == 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 	/* detach rate limiting */
3260 	in_pcbdetach_txrtlmt(inp);
3261 
3262 	/* make sure new mbuf send tag allocation is made */
3263 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3264 
3265 	if (did_upgrade)
3266 		INP_DOWNGRADE(inp);
3267 }
3268 
3269 #ifdef INET
3270 static void
3271 rl_init(void *st)
3272 {
3273 	rate_limit_new = counter_u64_alloc(M_WAITOK);
3274 	rate_limit_chg = counter_u64_alloc(M_WAITOK);
3275 	rate_limit_active = counter_u64_alloc(M_WAITOK);
3276 	rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3277 	rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3278 }
3279 
3280 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3281 #endif
3282 #endif /* RATELIMIT */
3283