xref: /freebsd/sys/netinet/in_pcb.c (revision 08c4a937a6685f05667996228898521fc453f8f3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993, 1995
5  *	The Regents of the University of California.
6  * Copyright (c) 2007-2009 Robert N. M. Watson
7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
8  * All rights reserved.
9  *
10  * Portions of this software were developed by Robert N. M. Watson under
11  * contract to Juniper Networks, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)in_pcb.c	8.4 (Berkeley) 5/24/95
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_ddb.h"
44 #include "opt_ipsec.h"
45 #include "opt_inet.h"
46 #include "opt_inet6.h"
47 #include "opt_ratelimit.h"
48 #include "opt_pcbgroup.h"
49 #include "opt_rss.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/callout.h>
57 #include <sys/eventhandler.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/rmlock.h>
61 #include <sys/smp.h>
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <sys/sockio.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/refcount.h>
68 #include <sys/jail.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #include <vm/uma.h>
77 
78 #include <net/if.h>
79 #include <net/if_var.h>
80 #include <net/if_types.h>
81 #include <net/if_llatbl.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
84 #include <net/vnet.h>
85 
86 #if defined(INET) || defined(INET6)
87 #include <netinet/in.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp_var.h>
91 #ifdef TCPHPTS
92 #include <netinet/tcp_hpts.h>
93 #endif
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
96 #endif
97 #ifdef INET
98 #include <netinet/in_var.h>
99 #endif
100 #ifdef INET6
101 #include <netinet/ip6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #endif /* INET6 */
106 
107 #include <netipsec/ipsec_support.h>
108 
109 #include <security/mac/mac_framework.h>
110 
111 #define	INPCBLBGROUP_SIZMIN	8
112 #define	INPCBLBGROUP_SIZMAX	256
113 
114 static struct callout	ipport_tick_callout;
115 
116 /*
117  * These configure the range of local port addresses assigned to
118  * "unspecified" outgoing connections/packets/whatever.
119  */
120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;	/* 1023 */
121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;	/* 600 */
122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;	/* 10000 */
123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;	/* 65535 */
124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;	/* 49152 */
125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;	/* 65535 */
126 
127 /*
128  * Reserved ports accessible only to root. There are significant
129  * security considerations that must be accounted for when changing these,
130  * but the security benefits can be great. Please be careful.
131  */
132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;	/* 1023 */
133 VNET_DEFINE(int, ipport_reservedlow);
134 
135 /* Variables dealing with random ephemeral port allocation. */
136 VNET_DEFINE(int, ipport_randomized) = 1;	/* user controlled via sysctl */
137 VNET_DEFINE(int, ipport_randomcps) = 10;	/* user controlled via sysctl */
138 VNET_DEFINE(int, ipport_randomtime) = 45;	/* user controlled via sysctl */
139 VNET_DEFINE(int, ipport_stoprandom);		/* toggled by ipport_tick */
140 VNET_DEFINE(int, ipport_tcpallocs);
141 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
142 
143 #define	V_ipport_tcplastcount		VNET(ipport_tcplastcount)
144 
145 static void	in_pcbremlists(struct inpcb *inp);
146 #ifdef INET
147 static struct inpcb	*in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
148 			    struct in_addr faddr, u_int fport_arg,
149 			    struct in_addr laddr, u_int lport_arg,
150 			    int lookupflags, struct ifnet *ifp);
151 
152 #define RANGECHK(var, min, max) \
153 	if ((var) < (min)) { (var) = (min); } \
154 	else if ((var) > (max)) { (var) = (max); }
155 
156 static int
157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
158 {
159 	int error;
160 
161 	error = sysctl_handle_int(oidp, arg1, arg2, req);
162 	if (error == 0) {
163 		RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
164 		RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
165 		RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
166 		RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
167 		RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
168 		RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
169 	}
170 	return (error);
171 }
172 
173 #undef RANGECHK
174 
175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
176     "IP Ports");
177 
178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
179 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
180 	&VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
182 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
183 	&VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
185 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
186 	&VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
188 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
189 	&VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
191 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
192 	&VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
194 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
195 	&VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
197 	CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
198 	&VNET_NAME(ipport_reservedhigh), 0, "");
199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
200 	CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
202 	CTLFLAG_VNET | CTLFLAG_RW,
203 	&VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
205 	CTLFLAG_VNET | CTLFLAG_RW,
206 	&VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
207 	"allocations before switching to a sequental one");
208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
209 	CTLFLAG_VNET | CTLFLAG_RW,
210 	&VNET_NAME(ipport_randomtime), 0,
211 	"Minimum time to keep sequental port "
212 	"allocation before switching to a random one");
213 #endif /* INET */
214 
215 /*
216  * in_pcb.c: manage the Protocol Control Blocks.
217  *
218  * NOTE: It is assumed that most of these functions will be called with
219  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
220  * functions often modify hash chains or addresses in pcbs.
221  */
222 
223 static struct inpcblbgroup *
224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
225     uint16_t port, const union in_dependaddr *addr, int size)
226 {
227 	struct inpcblbgroup *grp;
228 	size_t bytes;
229 
230 	bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
231 	grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
232 	if (!grp)
233 		return (NULL);
234 	grp->il_vflag = vflag;
235 	grp->il_lport = port;
236 	grp->il_dependladdr = *addr;
237 	grp->il_inpsiz = size;
238 	CK_LIST_INSERT_HEAD(hdr, grp, il_list);
239 	return (grp);
240 }
241 
242 static void
243 in_pcblbgroup_free_deferred(epoch_context_t ctx)
244 {
245 	struct inpcblbgroup *grp;
246 
247 	grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
248 	free(grp, M_PCB);
249 }
250 
251 static void
252 in_pcblbgroup_free(struct inpcblbgroup *grp)
253 {
254 
255 	CK_LIST_REMOVE(grp, il_list);
256 	epoch_call(net_epoch_preempt, &grp->il_epoch_ctx,
257 	    in_pcblbgroup_free_deferred);
258 }
259 
260 static struct inpcblbgroup *
261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
262     struct inpcblbgroup *old_grp, int size)
263 {
264 	struct inpcblbgroup *grp;
265 	int i;
266 
267 	grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
268 	    old_grp->il_lport, &old_grp->il_dependladdr, size);
269 	if (grp == NULL)
270 		return (NULL);
271 
272 	KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
273 	    ("invalid new local group size %d and old local group count %d",
274 	     grp->il_inpsiz, old_grp->il_inpcnt));
275 
276 	for (i = 0; i < old_grp->il_inpcnt; ++i)
277 		grp->il_inp[i] = old_grp->il_inp[i];
278 	grp->il_inpcnt = old_grp->il_inpcnt;
279 	in_pcblbgroup_free(old_grp);
280 	return (grp);
281 }
282 
283 /*
284  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
285  * and shrink group if possible.
286  */
287 static void
288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
289     int i)
290 {
291 	struct inpcblbgroup *grp, *new_grp;
292 
293 	grp = *grpp;
294 	for (; i + 1 < grp->il_inpcnt; ++i)
295 		grp->il_inp[i] = grp->il_inp[i + 1];
296 	grp->il_inpcnt--;
297 
298 	if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
299 	    grp->il_inpcnt <= grp->il_inpsiz / 4) {
300 		/* Shrink this group. */
301 		new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
302 		if (new_grp != NULL)
303 			*grpp = new_grp;
304 	}
305 }
306 
307 /*
308  * Add PCB to load balance group for SO_REUSEPORT_LB option.
309  */
310 static int
311 in_pcbinslbgrouphash(struct inpcb *inp)
312 {
313 	const static struct timeval interval = { 60, 0 };
314 	static struct timeval lastprint;
315 	struct inpcbinfo *pcbinfo;
316 	struct inpcblbgrouphead *hdr;
317 	struct inpcblbgroup *grp;
318 	uint32_t idx;
319 
320 	pcbinfo = inp->inp_pcbinfo;
321 
322 	INP_WLOCK_ASSERT(inp);
323 	INP_HASH_WLOCK_ASSERT(pcbinfo);
324 
325 	/*
326 	 * Don't allow jailed socket to join local group.
327 	 */
328 	if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
329 		return (0);
330 
331 #ifdef INET6
332 	/*
333 	 * Don't allow IPv4 mapped INET6 wild socket.
334 	 */
335 	if ((inp->inp_vflag & INP_IPV4) &&
336 	    inp->inp_laddr.s_addr == INADDR_ANY &&
337 	    INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
338 		return (0);
339 	}
340 #endif
341 
342 	idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
343 	hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
344 	CK_LIST_FOREACH(grp, hdr, il_list) {
345 		if (grp->il_vflag == inp->inp_vflag &&
346 		    grp->il_lport == inp->inp_lport &&
347 		    memcmp(&grp->il_dependladdr,
348 		    &inp->inp_inc.inc_ie.ie_dependladdr,
349 		    sizeof(grp->il_dependladdr)) == 0)
350 			break;
351 	}
352 	if (grp == NULL) {
353 		/* Create new load balance group. */
354 		grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
355 		    inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
356 		    INPCBLBGROUP_SIZMIN);
357 		if (grp == NULL)
358 			return (ENOBUFS);
359 	} else if (grp->il_inpcnt == grp->il_inpsiz) {
360 		if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
361 			if (ratecheck(&lastprint, &interval))
362 				printf("lb group port %d, limit reached\n",
363 				    ntohs(grp->il_lport));
364 			return (0);
365 		}
366 
367 		/* Expand this local group. */
368 		grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
369 		if (grp == NULL)
370 			return (ENOBUFS);
371 	}
372 
373 	KASSERT(grp->il_inpcnt < grp->il_inpsiz,
374 	    ("invalid local group size %d and count %d", grp->il_inpsiz,
375 	    grp->il_inpcnt));
376 
377 	grp->il_inp[grp->il_inpcnt] = inp;
378 	grp->il_inpcnt++;
379 	return (0);
380 }
381 
382 /*
383  * Remove PCB from load balance group.
384  */
385 static void
386 in_pcbremlbgrouphash(struct inpcb *inp)
387 {
388 	struct inpcbinfo *pcbinfo;
389 	struct inpcblbgrouphead *hdr;
390 	struct inpcblbgroup *grp;
391 	int i;
392 
393 	pcbinfo = inp->inp_pcbinfo;
394 
395 	INP_WLOCK_ASSERT(inp);
396 	INP_HASH_WLOCK_ASSERT(pcbinfo);
397 
398 	hdr = &pcbinfo->ipi_lbgrouphashbase[
399 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
400 	CK_LIST_FOREACH(grp, hdr, il_list) {
401 		for (i = 0; i < grp->il_inpcnt; ++i) {
402 			if (grp->il_inp[i] != inp)
403 				continue;
404 
405 			if (grp->il_inpcnt == 1) {
406 				/* We are the last, free this local group. */
407 				in_pcblbgroup_free(grp);
408 			} else {
409 				/* Pull up inpcbs, shrink group if possible. */
410 				in_pcblbgroup_reorder(hdr, &grp, i);
411 			}
412 			return;
413 		}
414 	}
415 }
416 
417 /*
418  * Different protocols initialize their inpcbs differently - giving
419  * different name to the lock.  But they all are disposed the same.
420  */
421 static void
422 inpcb_fini(void *mem, int size)
423 {
424 	struct inpcb *inp = mem;
425 
426 	INP_LOCK_DESTROY(inp);
427 }
428 
429 /*
430  * Initialize an inpcbinfo -- we should be able to reduce the number of
431  * arguments in time.
432  */
433 void
434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
435     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
436     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
437 {
438 
439 	porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
440 
441 	INP_INFO_LOCK_INIT(pcbinfo, name);
442 	INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");	/* XXXRW: argument? */
443 	INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
444 #ifdef VIMAGE
445 	pcbinfo->ipi_vnet = curvnet;
446 #endif
447 	pcbinfo->ipi_listhead = listhead;
448 	CK_LIST_INIT(pcbinfo->ipi_listhead);
449 	pcbinfo->ipi_count = 0;
450 	pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
451 	    &pcbinfo->ipi_hashmask);
452 	pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
453 	    &pcbinfo->ipi_porthashmask);
454 	pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
455 	    &pcbinfo->ipi_lbgrouphashmask);
456 #ifdef PCBGROUP
457 	in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
458 #endif
459 	pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
460 	    NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
461 	uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
462 	uma_zone_set_warning(pcbinfo->ipi_zone,
463 	    "kern.ipc.maxsockets limit reached");
464 }
465 
466 /*
467  * Destroy an inpcbinfo.
468  */
469 void
470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
471 {
472 
473 	KASSERT(pcbinfo->ipi_count == 0,
474 	    ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
475 
476 	hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
477 	hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
478 	    pcbinfo->ipi_porthashmask);
479 	hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
480 	    pcbinfo->ipi_lbgrouphashmask);
481 #ifdef PCBGROUP
482 	in_pcbgroup_destroy(pcbinfo);
483 #endif
484 	uma_zdestroy(pcbinfo->ipi_zone);
485 	INP_LIST_LOCK_DESTROY(pcbinfo);
486 	INP_HASH_LOCK_DESTROY(pcbinfo);
487 	INP_INFO_LOCK_DESTROY(pcbinfo);
488 }
489 
490 /*
491  * Allocate a PCB and associate it with the socket.
492  * On success return with the PCB locked.
493  */
494 int
495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
496 {
497 	struct inpcb *inp;
498 	int error;
499 
500 #ifdef INVARIANTS
501 	if (pcbinfo == &V_tcbinfo) {
502 		INP_INFO_RLOCK_ASSERT(pcbinfo);
503 	} else {
504 		INP_INFO_WLOCK_ASSERT(pcbinfo);
505 	}
506 #endif
507 
508 	error = 0;
509 	inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
510 	if (inp == NULL)
511 		return (ENOBUFS);
512 	bzero(&inp->inp_start_zero, inp_zero_size);
513 	inp->inp_pcbinfo = pcbinfo;
514 	inp->inp_socket = so;
515 	inp->inp_cred = crhold(so->so_cred);
516 	inp->inp_inc.inc_fibnum = so->so_fibnum;
517 #ifdef MAC
518 	error = mac_inpcb_init(inp, M_NOWAIT);
519 	if (error != 0)
520 		goto out;
521 	mac_inpcb_create(so, inp);
522 #endif
523 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
524 	error = ipsec_init_pcbpolicy(inp);
525 	if (error != 0) {
526 #ifdef MAC
527 		mac_inpcb_destroy(inp);
528 #endif
529 		goto out;
530 	}
531 #endif /*IPSEC*/
532 #ifdef INET6
533 	if (INP_SOCKAF(so) == AF_INET6) {
534 		inp->inp_vflag |= INP_IPV6PROTO;
535 		if (V_ip6_v6only)
536 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
537 	}
538 #endif
539 	INP_WLOCK(inp);
540 	INP_LIST_WLOCK(pcbinfo);
541 	CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
542 	pcbinfo->ipi_count++;
543 	so->so_pcb = (caddr_t)inp;
544 #ifdef INET6
545 	if (V_ip6_auto_flowlabel)
546 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
547 #endif
548 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
549 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
550 
551 	/*
552 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
553 	 * to be cleaned up.
554 	 */
555 	inp->inp_route.ro_flags = RT_LLE_CACHE;
556 	INP_LIST_WUNLOCK(pcbinfo);
557 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
558 out:
559 	if (error != 0) {
560 		crfree(inp->inp_cred);
561 		uma_zfree(pcbinfo->ipi_zone, inp);
562 	}
563 #endif
564 	return (error);
565 }
566 
567 #ifdef INET
568 int
569 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
570 {
571 	int anonport, error;
572 
573 	INP_WLOCK_ASSERT(inp);
574 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
575 
576 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
577 		return (EINVAL);
578 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
579 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
580 	    &inp->inp_lport, cred);
581 	if (error)
582 		return (error);
583 	if (in_pcbinshash(inp) != 0) {
584 		inp->inp_laddr.s_addr = INADDR_ANY;
585 		inp->inp_lport = 0;
586 		return (EAGAIN);
587 	}
588 	if (anonport)
589 		inp->inp_flags |= INP_ANONPORT;
590 	return (0);
591 }
592 #endif
593 
594 /*
595  * Select a local port (number) to use.
596  */
597 #if defined(INET) || defined(INET6)
598 int
599 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
600     struct ucred *cred, int lookupflags)
601 {
602 	struct inpcbinfo *pcbinfo;
603 	struct inpcb *tmpinp;
604 	unsigned short *lastport;
605 	int count, dorandom, error;
606 	u_short aux, first, last, lport;
607 #ifdef INET
608 	struct in_addr laddr;
609 #endif
610 
611 	pcbinfo = inp->inp_pcbinfo;
612 
613 	/*
614 	 * Because no actual state changes occur here, a global write lock on
615 	 * the pcbinfo isn't required.
616 	 */
617 	INP_LOCK_ASSERT(inp);
618 	INP_HASH_LOCK_ASSERT(pcbinfo);
619 
620 	if (inp->inp_flags & INP_HIGHPORT) {
621 		first = V_ipport_hifirstauto;	/* sysctl */
622 		last  = V_ipport_hilastauto;
623 		lastport = &pcbinfo->ipi_lasthi;
624 	} else if (inp->inp_flags & INP_LOWPORT) {
625 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
626 		if (error)
627 			return (error);
628 		first = V_ipport_lowfirstauto;	/* 1023 */
629 		last  = V_ipport_lowlastauto;	/* 600 */
630 		lastport = &pcbinfo->ipi_lastlow;
631 	} else {
632 		first = V_ipport_firstauto;	/* sysctl */
633 		last  = V_ipport_lastauto;
634 		lastport = &pcbinfo->ipi_lastport;
635 	}
636 	/*
637 	 * For UDP(-Lite), use random port allocation as long as the user
638 	 * allows it.  For TCP (and as of yet unknown) connections,
639 	 * use random port allocation only if the user allows it AND
640 	 * ipport_tick() allows it.
641 	 */
642 	if (V_ipport_randomized &&
643 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
644 		pcbinfo == &V_ulitecbinfo))
645 		dorandom = 1;
646 	else
647 		dorandom = 0;
648 	/*
649 	 * It makes no sense to do random port allocation if
650 	 * we have the only port available.
651 	 */
652 	if (first == last)
653 		dorandom = 0;
654 	/* Make sure to not include UDP(-Lite) packets in the count. */
655 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
656 		V_ipport_tcpallocs++;
657 	/*
658 	 * Instead of having two loops further down counting up or down
659 	 * make sure that first is always <= last and go with only one
660 	 * code path implementing all logic.
661 	 */
662 	if (first > last) {
663 		aux = first;
664 		first = last;
665 		last = aux;
666 	}
667 
668 #ifdef INET
669 	/* Make the compiler happy. */
670 	laddr.s_addr = 0;
671 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
672 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
673 		    __func__, inp));
674 		laddr = *laddrp;
675 	}
676 #endif
677 	tmpinp = NULL;	/* Make compiler happy. */
678 	lport = *lportp;
679 
680 	if (dorandom)
681 		*lastport = first + (arc4random() % (last - first));
682 
683 	count = last - first;
684 
685 	do {
686 		if (count-- < 0)	/* completely used? */
687 			return (EADDRNOTAVAIL);
688 		++*lastport;
689 		if (*lastport < first || *lastport > last)
690 			*lastport = first;
691 		lport = htons(*lastport);
692 
693 #ifdef INET6
694 		if ((inp->inp_vflag & INP_IPV6) != 0)
695 			tmpinp = in6_pcblookup_local(pcbinfo,
696 			    &inp->in6p_laddr, lport, lookupflags, cred);
697 #endif
698 #if defined(INET) && defined(INET6)
699 		else
700 #endif
701 #ifdef INET
702 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
703 			    lport, lookupflags, cred);
704 #endif
705 	} while (tmpinp != NULL);
706 
707 #ifdef INET
708 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
709 		laddrp->s_addr = laddr.s_addr;
710 #endif
711 	*lportp = lport;
712 
713 	return (0);
714 }
715 
716 /*
717  * Return cached socket options.
718  */
719 int
720 inp_so_options(const struct inpcb *inp)
721 {
722 	int so_options;
723 
724 	so_options = 0;
725 
726 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
727 		so_options |= SO_REUSEPORT_LB;
728 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
729 		so_options |= SO_REUSEPORT;
730 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
731 		so_options |= SO_REUSEADDR;
732 	return (so_options);
733 }
734 #endif /* INET || INET6 */
735 
736 /*
737  * Check if a new BINDMULTI socket is allowed to be created.
738  *
739  * ni points to the new inp.
740  * oi points to the exisitng inp.
741  *
742  * This checks whether the existing inp also has BINDMULTI and
743  * whether the credentials match.
744  */
745 int
746 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
747 {
748 	/* Check permissions match */
749 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
750 	    (ni->inp_cred->cr_uid !=
751 	    oi->inp_cred->cr_uid))
752 		return (0);
753 
754 	/* Check the existing inp has BINDMULTI set */
755 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
756 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
757 		return (0);
758 
759 	/*
760 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
761 	 * it is and it matches the checks.
762 	 */
763 	return (1);
764 }
765 
766 #ifdef INET
767 /*
768  * Set up a bind operation on a PCB, performing port allocation
769  * as required, but do not actually modify the PCB. Callers can
770  * either complete the bind by setting inp_laddr/inp_lport and
771  * calling in_pcbinshash(), or they can just use the resulting
772  * port and address to authorise the sending of a once-off packet.
773  *
774  * On error, the values of *laddrp and *lportp are not changed.
775  */
776 int
777 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
778     u_short *lportp, struct ucred *cred)
779 {
780 	struct socket *so = inp->inp_socket;
781 	struct sockaddr_in *sin;
782 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
783 	struct in_addr laddr;
784 	u_short lport = 0;
785 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
786 	int error;
787 
788 	/*
789 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
790 	 * so that we don't have to add to the (already messy) code below.
791 	 */
792 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
793 
794 	/*
795 	 * No state changes, so read locks are sufficient here.
796 	 */
797 	INP_LOCK_ASSERT(inp);
798 	INP_HASH_LOCK_ASSERT(pcbinfo);
799 
800 	if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
801 		return (EADDRNOTAVAIL);
802 	laddr.s_addr = *laddrp;
803 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
804 		return (EINVAL);
805 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
806 		lookupflags = INPLOOKUP_WILDCARD;
807 	if (nam == NULL) {
808 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
809 			return (error);
810 	} else {
811 		sin = (struct sockaddr_in *)nam;
812 		if (nam->sa_len != sizeof (*sin))
813 			return (EINVAL);
814 #ifdef notdef
815 		/*
816 		 * We should check the family, but old programs
817 		 * incorrectly fail to initialize it.
818 		 */
819 		if (sin->sin_family != AF_INET)
820 			return (EAFNOSUPPORT);
821 #endif
822 		error = prison_local_ip4(cred, &sin->sin_addr);
823 		if (error)
824 			return (error);
825 		if (sin->sin_port != *lportp) {
826 			/* Don't allow the port to change. */
827 			if (*lportp != 0)
828 				return (EINVAL);
829 			lport = sin->sin_port;
830 		}
831 		/* NB: lport is left as 0 if the port isn't being changed. */
832 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
833 			/*
834 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
835 			 * allow complete duplication of binding if
836 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
837 			 * and a multicast address is bound on both
838 			 * new and duplicated sockets.
839 			 */
840 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
841 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
842 			/*
843 			 * XXX: How to deal with SO_REUSEPORT_LB here?
844 			 * Treat same as SO_REUSEPORT for now.
845 			 */
846 			if ((so->so_options &
847 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
848 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
849 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
850 			sin->sin_port = 0;		/* yech... */
851 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
852 			/*
853 			 * Is the address a local IP address?
854 			 * If INP_BINDANY is set, then the socket may be bound
855 			 * to any endpoint address, local or not.
856 			 */
857 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
858 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
859 				return (EADDRNOTAVAIL);
860 		}
861 		laddr = sin->sin_addr;
862 		if (lport) {
863 			struct inpcb *t;
864 			struct tcptw *tw;
865 
866 			/* GROSS */
867 			if (ntohs(lport) <= V_ipport_reservedhigh &&
868 			    ntohs(lport) >= V_ipport_reservedlow &&
869 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
870 			    0))
871 				return (EACCES);
872 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
873 			    priv_check_cred(inp->inp_cred,
874 			    PRIV_NETINET_REUSEPORT, 0) != 0) {
875 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
876 				    lport, INPLOOKUP_WILDCARD, cred);
877 	/*
878 	 * XXX
879 	 * This entire block sorely needs a rewrite.
880 	 */
881 				if (t &&
882 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
883 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
884 				    (so->so_type != SOCK_STREAM ||
885 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
886 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
887 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
888 				     (t->inp_flags2 & INP_REUSEPORT) ||
889 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
890 				    (inp->inp_cred->cr_uid !=
891 				     t->inp_cred->cr_uid))
892 					return (EADDRINUSE);
893 
894 				/*
895 				 * If the socket is a BINDMULTI socket, then
896 				 * the credentials need to match and the
897 				 * original socket also has to have been bound
898 				 * with BINDMULTI.
899 				 */
900 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
901 					return (EADDRINUSE);
902 			}
903 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
904 			    lport, lookupflags, cred);
905 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
906 				/*
907 				 * XXXRW: If an incpb has had its timewait
908 				 * state recycled, we treat the address as
909 				 * being in use (for now).  This is better
910 				 * than a panic, but not desirable.
911 				 */
912 				tw = intotw(t);
913 				if (tw == NULL ||
914 				    ((reuseport & tw->tw_so_options) == 0 &&
915 					(reuseport_lb &
916 				            tw->tw_so_options) == 0)) {
917 					return (EADDRINUSE);
918 				}
919 			} else if (t &&
920 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
921 				   (reuseport & inp_so_options(t)) == 0 &&
922 				   (reuseport_lb & inp_so_options(t)) == 0) {
923 #ifdef INET6
924 				if (ntohl(sin->sin_addr.s_addr) !=
925 				    INADDR_ANY ||
926 				    ntohl(t->inp_laddr.s_addr) !=
927 				    INADDR_ANY ||
928 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
929 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
930 #endif
931 						return (EADDRINUSE);
932 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
933 					return (EADDRINUSE);
934 			}
935 		}
936 	}
937 	if (*lportp != 0)
938 		lport = *lportp;
939 	if (lport == 0) {
940 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
941 		if (error != 0)
942 			return (error);
943 
944 	}
945 	*laddrp = laddr.s_addr;
946 	*lportp = lport;
947 	return (0);
948 }
949 
950 /*
951  * Connect from a socket to a specified address.
952  * Both address and port must be specified in argument sin.
953  * If don't have a local address for this socket yet,
954  * then pick one.
955  */
956 int
957 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
958     struct ucred *cred, struct mbuf *m)
959 {
960 	u_short lport, fport;
961 	in_addr_t laddr, faddr;
962 	int anonport, error;
963 
964 	INP_WLOCK_ASSERT(inp);
965 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
966 
967 	lport = inp->inp_lport;
968 	laddr = inp->inp_laddr.s_addr;
969 	anonport = (lport == 0);
970 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
971 	    NULL, cred);
972 	if (error)
973 		return (error);
974 
975 	/* Do the initial binding of the local address if required. */
976 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
977 		inp->inp_lport = lport;
978 		inp->inp_laddr.s_addr = laddr;
979 		if (in_pcbinshash(inp) != 0) {
980 			inp->inp_laddr.s_addr = INADDR_ANY;
981 			inp->inp_lport = 0;
982 			return (EAGAIN);
983 		}
984 	}
985 
986 	/* Commit the remaining changes. */
987 	inp->inp_lport = lport;
988 	inp->inp_laddr.s_addr = laddr;
989 	inp->inp_faddr.s_addr = faddr;
990 	inp->inp_fport = fport;
991 	in_pcbrehash_mbuf(inp, m);
992 
993 	if (anonport)
994 		inp->inp_flags |= INP_ANONPORT;
995 	return (0);
996 }
997 
998 int
999 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1000 {
1001 
1002 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1003 }
1004 
1005 /*
1006  * Do proper source address selection on an unbound socket in case
1007  * of connect. Take jails into account as well.
1008  */
1009 int
1010 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1011     struct ucred *cred)
1012 {
1013 	struct ifaddr *ifa;
1014 	struct sockaddr *sa;
1015 	struct sockaddr_in *sin;
1016 	struct route sro;
1017 	int error;
1018 
1019 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1020 	/*
1021 	 * Bypass source address selection and use the primary jail IP
1022 	 * if requested.
1023 	 */
1024 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1025 		return (0);
1026 
1027 	error = 0;
1028 	bzero(&sro, sizeof(sro));
1029 
1030 	sin = (struct sockaddr_in *)&sro.ro_dst;
1031 	sin->sin_family = AF_INET;
1032 	sin->sin_len = sizeof(struct sockaddr_in);
1033 	sin->sin_addr.s_addr = faddr->s_addr;
1034 
1035 	/*
1036 	 * If route is known our src addr is taken from the i/f,
1037 	 * else punt.
1038 	 *
1039 	 * Find out route to destination.
1040 	 */
1041 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1042 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1043 
1044 	/*
1045 	 * If we found a route, use the address corresponding to
1046 	 * the outgoing interface.
1047 	 *
1048 	 * Otherwise assume faddr is reachable on a directly connected
1049 	 * network and try to find a corresponding interface to take
1050 	 * the source address from.
1051 	 */
1052 	NET_EPOCH_ENTER();
1053 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1054 		struct in_ifaddr *ia;
1055 		struct ifnet *ifp;
1056 
1057 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1058 					inp->inp_socket->so_fibnum));
1059 		if (ia == NULL) {
1060 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1061 						inp->inp_socket->so_fibnum));
1062 
1063 		}
1064 		if (ia == NULL) {
1065 			error = ENETUNREACH;
1066 			goto done;
1067 		}
1068 
1069 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1070 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1071 			goto done;
1072 		}
1073 
1074 		ifp = ia->ia_ifp;
1075 		ia = NULL;
1076 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1077 
1078 			sa = ifa->ifa_addr;
1079 			if (sa->sa_family != AF_INET)
1080 				continue;
1081 			sin = (struct sockaddr_in *)sa;
1082 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1083 				ia = (struct in_ifaddr *)ifa;
1084 				break;
1085 			}
1086 		}
1087 		if (ia != NULL) {
1088 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1089 			goto done;
1090 		}
1091 
1092 		/* 3. As a last resort return the 'default' jail address. */
1093 		error = prison_get_ip4(cred, laddr);
1094 		goto done;
1095 	}
1096 
1097 	/*
1098 	 * If the outgoing interface on the route found is not
1099 	 * a loopback interface, use the address from that interface.
1100 	 * In case of jails do those three steps:
1101 	 * 1. check if the interface address belongs to the jail. If so use it.
1102 	 * 2. check if we have any address on the outgoing interface
1103 	 *    belonging to this jail. If so use it.
1104 	 * 3. as a last resort return the 'default' jail address.
1105 	 */
1106 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1107 		struct in_ifaddr *ia;
1108 		struct ifnet *ifp;
1109 
1110 		/* If not jailed, use the default returned. */
1111 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1112 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1113 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1114 			goto done;
1115 		}
1116 
1117 		/* Jailed. */
1118 		/* 1. Check if the iface address belongs to the jail. */
1119 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1120 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1121 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1122 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1123 			goto done;
1124 		}
1125 
1126 		/*
1127 		 * 2. Check if we have any address on the outgoing interface
1128 		 *    belonging to this jail.
1129 		 */
1130 		ia = NULL;
1131 		ifp = sro.ro_rt->rt_ifp;
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 	 * The outgoing interface is marked with 'loopback net', so a route
1154 	 * to ourselves is here.
1155 	 * Try to find the interface of the destination address and then
1156 	 * take the address from there. That interface is not necessarily
1157 	 * a loopback interface.
1158 	 * In case of jails, check that it is an address of the jail
1159 	 * and if we cannot find, fall back to the 'default' jail address.
1160 	 */
1161 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1162 		struct sockaddr_in sain;
1163 		struct in_ifaddr *ia;
1164 
1165 		bzero(&sain, sizeof(struct sockaddr_in));
1166 		sain.sin_family = AF_INET;
1167 		sain.sin_len = sizeof(struct sockaddr_in);
1168 		sain.sin_addr.s_addr = faddr->s_addr;
1169 
1170 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1171 					inp->inp_socket->so_fibnum));
1172 		if (ia == NULL)
1173 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1174 						inp->inp_socket->so_fibnum));
1175 		if (ia == NULL)
1176 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1177 
1178 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1179 			if (ia == NULL) {
1180 				error = ENETUNREACH;
1181 				goto done;
1182 			}
1183 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1184 			goto done;
1185 		}
1186 
1187 		/* Jailed. */
1188 		if (ia != NULL) {
1189 			struct ifnet *ifp;
1190 
1191 			ifp = ia->ia_ifp;
1192 			ia = NULL;
1193 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1194 				sa = ifa->ifa_addr;
1195 				if (sa->sa_family != AF_INET)
1196 					continue;
1197 				sin = (struct sockaddr_in *)sa;
1198 				if (prison_check_ip4(cred,
1199 				    &sin->sin_addr) == 0) {
1200 					ia = (struct in_ifaddr *)ifa;
1201 					break;
1202 				}
1203 			}
1204 			if (ia != NULL) {
1205 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1206 				goto done;
1207 			}
1208 		}
1209 
1210 		/* 3. As a last resort return the 'default' jail address. */
1211 		error = prison_get_ip4(cred, laddr);
1212 		goto done;
1213 	}
1214 
1215 done:
1216 	NET_EPOCH_EXIT();
1217 	if (sro.ro_rt != NULL)
1218 		RTFREE(sro.ro_rt);
1219 	return (error);
1220 }
1221 
1222 /*
1223  * Set up for a connect from a socket to the specified address.
1224  * On entry, *laddrp and *lportp should contain the current local
1225  * address and port for the PCB; these are updated to the values
1226  * that should be placed in inp_laddr and inp_lport to complete
1227  * the connect.
1228  *
1229  * On success, *faddrp and *fportp will be set to the remote address
1230  * and port. These are not updated in the error case.
1231  *
1232  * If the operation fails because the connection already exists,
1233  * *oinpp will be set to the PCB of that connection so that the
1234  * caller can decide to override it. In all other cases, *oinpp
1235  * is set to NULL.
1236  */
1237 int
1238 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1239     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1240     struct inpcb **oinpp, struct ucred *cred)
1241 {
1242 	struct rm_priotracker in_ifa_tracker;
1243 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1244 	struct in_ifaddr *ia;
1245 	struct inpcb *oinp;
1246 	struct in_addr laddr, faddr;
1247 	u_short lport, fport;
1248 	int error;
1249 
1250 	/*
1251 	 * Because a global state change doesn't actually occur here, a read
1252 	 * lock is sufficient.
1253 	 */
1254 	INP_LOCK_ASSERT(inp);
1255 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1256 
1257 	if (oinpp != NULL)
1258 		*oinpp = NULL;
1259 	if (nam->sa_len != sizeof (*sin))
1260 		return (EINVAL);
1261 	if (sin->sin_family != AF_INET)
1262 		return (EAFNOSUPPORT);
1263 	if (sin->sin_port == 0)
1264 		return (EADDRNOTAVAIL);
1265 	laddr.s_addr = *laddrp;
1266 	lport = *lportp;
1267 	faddr = sin->sin_addr;
1268 	fport = sin->sin_port;
1269 
1270 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1271 		/*
1272 		 * If the destination address is INADDR_ANY,
1273 		 * use the primary local address.
1274 		 * If the supplied address is INADDR_BROADCAST,
1275 		 * and the primary interface supports broadcast,
1276 		 * choose the broadcast address for that interface.
1277 		 */
1278 		if (faddr.s_addr == INADDR_ANY) {
1279 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1280 			faddr =
1281 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1282 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1283 			if (cred != NULL &&
1284 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1285 				return (error);
1286 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1287 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1288 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1289 			    IFF_BROADCAST)
1290 				faddr = satosin(&CK_STAILQ_FIRST(
1291 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1292 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1293 		}
1294 	}
1295 	if (laddr.s_addr == INADDR_ANY) {
1296 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1297 		/*
1298 		 * If the destination address is multicast and an outgoing
1299 		 * interface has been set as a multicast option, prefer the
1300 		 * address of that interface as our source address.
1301 		 */
1302 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1303 		    inp->inp_moptions != NULL) {
1304 			struct ip_moptions *imo;
1305 			struct ifnet *ifp;
1306 
1307 			imo = inp->inp_moptions;
1308 			if (imo->imo_multicast_ifp != NULL) {
1309 				ifp = imo->imo_multicast_ifp;
1310 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1311 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1312 					if ((ia->ia_ifp == ifp) &&
1313 					    (cred == NULL ||
1314 					    prison_check_ip4(cred,
1315 					    &ia->ia_addr.sin_addr) == 0))
1316 						break;
1317 				}
1318 				if (ia == NULL)
1319 					error = EADDRNOTAVAIL;
1320 				else {
1321 					laddr = ia->ia_addr.sin_addr;
1322 					error = 0;
1323 				}
1324 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1325 			}
1326 		}
1327 		if (error)
1328 			return (error);
1329 	}
1330 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1331 	    laddr, lport, 0, NULL);
1332 	if (oinp != NULL) {
1333 		if (oinpp != NULL)
1334 			*oinpp = oinp;
1335 		return (EADDRINUSE);
1336 	}
1337 	if (lport == 0) {
1338 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1339 		    cred);
1340 		if (error)
1341 			return (error);
1342 	}
1343 	*laddrp = laddr.s_addr;
1344 	*lportp = lport;
1345 	*faddrp = faddr.s_addr;
1346 	*fportp = fport;
1347 	return (0);
1348 }
1349 
1350 void
1351 in_pcbdisconnect(struct inpcb *inp)
1352 {
1353 
1354 	INP_WLOCK_ASSERT(inp);
1355 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1356 
1357 	inp->inp_faddr.s_addr = INADDR_ANY;
1358 	inp->inp_fport = 0;
1359 	in_pcbrehash(inp);
1360 }
1361 #endif /* INET */
1362 
1363 /*
1364  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1365  * For most protocols, this will be invoked immediately prior to calling
1366  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1367  * socket, in which case in_pcbfree() is deferred.
1368  */
1369 void
1370 in_pcbdetach(struct inpcb *inp)
1371 {
1372 
1373 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1374 
1375 #ifdef RATELIMIT
1376 	if (inp->inp_snd_tag != NULL)
1377 		in_pcbdetach_txrtlmt(inp);
1378 #endif
1379 	inp->inp_socket->so_pcb = NULL;
1380 	inp->inp_socket = NULL;
1381 }
1382 
1383 /*
1384  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1385  * stability of an inpcb pointer despite the inpcb lock being released.  This
1386  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1387  * but where the inpcb lock may already held, or when acquiring a reference
1388  * via a pcbgroup.
1389  *
1390  * in_pcbref() should be used only to provide brief memory stability, and
1391  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1392  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1393  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1394  * lock and rele are the *only* safe operations that may be performed on the
1395  * inpcb.
1396  *
1397  * While the inpcb will not be freed, releasing the inpcb lock means that the
1398  * connection's state may change, so the caller should be careful to
1399  * revalidate any cached state on reacquiring the lock.  Drop the reference
1400  * using in_pcbrele().
1401  */
1402 void
1403 in_pcbref(struct inpcb *inp)
1404 {
1405 
1406 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1407 
1408 	refcount_acquire(&inp->inp_refcount);
1409 }
1410 
1411 /*
1412  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1413  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1414  * return a flag indicating whether or not the inpcb remains valid.  If it is
1415  * valid, we return with the inpcb lock held.
1416  *
1417  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1418  * reference on an inpcb.  Historically more work was done here (actually, in
1419  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1420  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1421  * about memory stability (and continued use of the write lock).
1422  */
1423 int
1424 in_pcbrele_rlocked(struct inpcb *inp)
1425 {
1426 	struct inpcbinfo *pcbinfo;
1427 
1428 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1429 
1430 	INP_RLOCK_ASSERT(inp);
1431 
1432 	if (refcount_release(&inp->inp_refcount) == 0) {
1433 		/*
1434 		 * If the inpcb has been freed, let the caller know, even if
1435 		 * this isn't the last reference.
1436 		 */
1437 		if (inp->inp_flags2 & INP_FREED) {
1438 			INP_RUNLOCK(inp);
1439 			return (1);
1440 		}
1441 		return (0);
1442 	}
1443 
1444 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1445 #ifdef TCPHPTS
1446 	if (inp->inp_in_hpts || inp->inp_in_input) {
1447 		struct tcp_hpts_entry *hpts;
1448 		/*
1449 		 * We should not be on the hpts at
1450 		 * this point in any form. we must
1451 		 * get the lock to be sure.
1452 		 */
1453 		hpts = tcp_hpts_lock(inp);
1454 		if (inp->inp_in_hpts)
1455 			panic("Hpts:%p inp:%p at free still on hpts",
1456 			      hpts, inp);
1457 		mtx_unlock(&hpts->p_mtx);
1458 		hpts = tcp_input_lock(inp);
1459 		if (inp->inp_in_input)
1460 			panic("Hpts:%p inp:%p at free still on input hpts",
1461 			      hpts, inp);
1462 		mtx_unlock(&hpts->p_mtx);
1463 	}
1464 #endif
1465 	INP_RUNLOCK(inp);
1466 	pcbinfo = inp->inp_pcbinfo;
1467 	uma_zfree(pcbinfo->ipi_zone, inp);
1468 	return (1);
1469 }
1470 
1471 int
1472 in_pcbrele_wlocked(struct inpcb *inp)
1473 {
1474 	struct inpcbinfo *pcbinfo;
1475 
1476 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1477 
1478 	INP_WLOCK_ASSERT(inp);
1479 
1480 	if (refcount_release(&inp->inp_refcount) == 0) {
1481 		/*
1482 		 * If the inpcb has been freed, let the caller know, even if
1483 		 * this isn't the last reference.
1484 		 */
1485 		if (inp->inp_flags2 & INP_FREED) {
1486 			INP_WUNLOCK(inp);
1487 			return (1);
1488 		}
1489 		return (0);
1490 	}
1491 
1492 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1493 #ifdef TCPHPTS
1494 	if (inp->inp_in_hpts || inp->inp_in_input) {
1495 		struct tcp_hpts_entry *hpts;
1496 		/*
1497 		 * We should not be on the hpts at
1498 		 * this point in any form. we must
1499 		 * get the lock to be sure.
1500 		 */
1501 		hpts = tcp_hpts_lock(inp);
1502 		if (inp->inp_in_hpts)
1503 			panic("Hpts:%p inp:%p at free still on hpts",
1504 			      hpts, inp);
1505 		mtx_unlock(&hpts->p_mtx);
1506 		hpts = tcp_input_lock(inp);
1507 		if (inp->inp_in_input)
1508 			panic("Hpts:%p inp:%p at free still on input hpts",
1509 			      hpts, inp);
1510 		mtx_unlock(&hpts->p_mtx);
1511 	}
1512 #endif
1513 	INP_WUNLOCK(inp);
1514 	pcbinfo = inp->inp_pcbinfo;
1515 	uma_zfree(pcbinfo->ipi_zone, inp);
1516 	return (1);
1517 }
1518 
1519 /*
1520  * Temporary wrapper.
1521  */
1522 int
1523 in_pcbrele(struct inpcb *inp)
1524 {
1525 
1526 	return (in_pcbrele_wlocked(inp));
1527 }
1528 
1529 void
1530 in_pcblist_rele_rlocked(epoch_context_t ctx)
1531 {
1532 	struct in_pcblist *il;
1533 	struct inpcb *inp;
1534 	struct inpcbinfo *pcbinfo;
1535 	int i, n;
1536 
1537 	il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1538 	pcbinfo = il->il_pcbinfo;
1539 	n = il->il_count;
1540 	INP_INFO_WLOCK(pcbinfo);
1541 	for (i = 0; i < n; i++) {
1542 		inp = il->il_inp_list[i];
1543 		INP_RLOCK(inp);
1544 		if (!in_pcbrele_rlocked(inp))
1545 			INP_RUNLOCK(inp);
1546 	}
1547 	INP_INFO_WUNLOCK(pcbinfo);
1548 	free(il, M_TEMP);
1549 }
1550 
1551 static void
1552 inpcbport_free(epoch_context_t ctx)
1553 {
1554 	struct inpcbport *phd;
1555 
1556 	phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1557 	free(phd, M_PCB);
1558 }
1559 
1560 static void
1561 in_pcbfree_deferred(epoch_context_t ctx)
1562 {
1563 	struct inpcb *inp;
1564 	int released __unused;
1565 
1566 	inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1567 
1568 	INP_WLOCK(inp);
1569 #ifdef INET
1570 	struct ip_moptions *imo = inp->inp_moptions;
1571 	inp->inp_moptions = NULL;
1572 #endif
1573 	/* XXXRW: Do as much as possible here. */
1574 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1575 	if (inp->inp_sp != NULL)
1576 		ipsec_delete_pcbpolicy(inp);
1577 #endif
1578 #ifdef INET6
1579 	struct ip6_moptions *im6o = NULL;
1580 	if (inp->inp_vflag & INP_IPV6PROTO) {
1581 		ip6_freepcbopts(inp->in6p_outputopts);
1582 		im6o = inp->in6p_moptions;
1583 		inp->in6p_moptions = NULL;
1584 	}
1585 #endif
1586 	if (inp->inp_options)
1587 		(void)m_free(inp->inp_options);
1588 	inp->inp_vflag = 0;
1589 	crfree(inp->inp_cred);
1590 #ifdef MAC
1591 	mac_inpcb_destroy(inp);
1592 #endif
1593 	released = in_pcbrele_wlocked(inp);
1594 	MPASS(released);
1595 #ifdef INET6
1596 	ip6_freemoptions(im6o);
1597 #endif
1598 #ifdef INET
1599 	inp_freemoptions(imo);
1600 #endif
1601 }
1602 
1603 /*
1604  * Unconditionally schedule an inpcb to be freed by decrementing its
1605  * reference count, which should occur only after the inpcb has been detached
1606  * from its socket.  If another thread holds a temporary reference (acquired
1607  * using in_pcbref()) then the free is deferred until that reference is
1608  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1609  * work, including removal from global lists, is done in this context, where
1610  * the pcbinfo lock is held.
1611  */
1612 void
1613 in_pcbfree(struct inpcb *inp)
1614 {
1615 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1616 
1617 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1618 	KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1619 	    ("%s: called twice for pcb %p", __func__, inp));
1620 	if (inp->inp_flags2 & INP_FREED) {
1621 		INP_WUNLOCK(inp);
1622 		return;
1623 	}
1624 
1625 #ifdef INVARIANTS
1626 	if (pcbinfo == &V_tcbinfo) {
1627 		INP_INFO_LOCK_ASSERT(pcbinfo);
1628 	} else {
1629 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1630 	}
1631 #endif
1632 	INP_WLOCK_ASSERT(inp);
1633 	INP_LIST_WLOCK(pcbinfo);
1634 	in_pcbremlists(inp);
1635 	INP_LIST_WUNLOCK(pcbinfo);
1636 	RO_INVALIDATE_CACHE(&inp->inp_route);
1637 	/* mark as destruction in progress */
1638 	inp->inp_flags2 |= INP_FREED;
1639 	INP_WUNLOCK(inp);
1640 	epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1641 }
1642 
1643 /*
1644  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1645  * port reservation, and preventing it from being returned by inpcb lookups.
1646  *
1647  * It is used by TCP to mark an inpcb as unused and avoid future packet
1648  * delivery or event notification when a socket remains open but TCP has
1649  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1650  * or a RST on the wire, and allows the port binding to be reused while still
1651  * maintaining the invariant that so_pcb always points to a valid inpcb until
1652  * in_pcbdetach().
1653  *
1654  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1655  * in_pcbnotifyall() and in_pcbpurgeif0()?
1656  */
1657 void
1658 in_pcbdrop(struct inpcb *inp)
1659 {
1660 
1661 	INP_WLOCK_ASSERT(inp);
1662 #ifdef INVARIANTS
1663 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1664 		MPASS(inp->inp_refcount > 1);
1665 #endif
1666 
1667 	/*
1668 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1669 	 * the hash lock...?
1670 	 */
1671 	inp->inp_flags |= INP_DROPPED;
1672 	if (inp->inp_flags & INP_INHASHLIST) {
1673 		struct inpcbport *phd = inp->inp_phd;
1674 
1675 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1676 		in_pcbremlbgrouphash(inp);
1677 		CK_LIST_REMOVE(inp, inp_hash);
1678 		CK_LIST_REMOVE(inp, inp_portlist);
1679 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1680 			CK_LIST_REMOVE(phd, phd_hash);
1681 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1682 		}
1683 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1684 		inp->inp_flags &= ~INP_INHASHLIST;
1685 #ifdef PCBGROUP
1686 		in_pcbgroup_remove(inp);
1687 #endif
1688 	}
1689 }
1690 
1691 #ifdef INET
1692 /*
1693  * Common routines to return the socket addresses associated with inpcbs.
1694  */
1695 struct sockaddr *
1696 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1697 {
1698 	struct sockaddr_in *sin;
1699 
1700 	sin = malloc(sizeof *sin, M_SONAME,
1701 		M_WAITOK | M_ZERO);
1702 	sin->sin_family = AF_INET;
1703 	sin->sin_len = sizeof(*sin);
1704 	sin->sin_addr = *addr_p;
1705 	sin->sin_port = port;
1706 
1707 	return (struct sockaddr *)sin;
1708 }
1709 
1710 int
1711 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1712 {
1713 	struct inpcb *inp;
1714 	struct in_addr addr;
1715 	in_port_t port;
1716 
1717 	inp = sotoinpcb(so);
1718 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1719 
1720 	INP_RLOCK(inp);
1721 	port = inp->inp_lport;
1722 	addr = inp->inp_laddr;
1723 	INP_RUNLOCK(inp);
1724 
1725 	*nam = in_sockaddr(port, &addr);
1726 	return 0;
1727 }
1728 
1729 int
1730 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1731 {
1732 	struct inpcb *inp;
1733 	struct in_addr addr;
1734 	in_port_t port;
1735 
1736 	inp = sotoinpcb(so);
1737 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1738 
1739 	INP_RLOCK(inp);
1740 	port = inp->inp_fport;
1741 	addr = inp->inp_faddr;
1742 	INP_RUNLOCK(inp);
1743 
1744 	*nam = in_sockaddr(port, &addr);
1745 	return 0;
1746 }
1747 
1748 void
1749 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1750     struct inpcb *(*notify)(struct inpcb *, int))
1751 {
1752 	struct inpcb *inp, *inp_temp;
1753 
1754 	INP_INFO_WLOCK(pcbinfo);
1755 	CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1756 		INP_WLOCK(inp);
1757 #ifdef INET6
1758 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1759 			INP_WUNLOCK(inp);
1760 			continue;
1761 		}
1762 #endif
1763 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1764 		    inp->inp_socket == NULL) {
1765 			INP_WUNLOCK(inp);
1766 			continue;
1767 		}
1768 		if ((*notify)(inp, errno))
1769 			INP_WUNLOCK(inp);
1770 	}
1771 	INP_INFO_WUNLOCK(pcbinfo);
1772 }
1773 
1774 void
1775 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1776 {
1777 	struct inpcb *inp;
1778 	struct ip_moptions *imo;
1779 	int i, gap;
1780 
1781 	INP_INFO_WLOCK(pcbinfo);
1782 	CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1783 		INP_WLOCK(inp);
1784 		imo = inp->inp_moptions;
1785 		if ((inp->inp_vflag & INP_IPV4) &&
1786 		    imo != NULL) {
1787 			/*
1788 			 * Unselect the outgoing interface if it is being
1789 			 * detached.
1790 			 */
1791 			if (imo->imo_multicast_ifp == ifp)
1792 				imo->imo_multicast_ifp = NULL;
1793 
1794 			/*
1795 			 * Drop multicast group membership if we joined
1796 			 * through the interface being detached.
1797 			 *
1798 			 * XXX This can all be deferred to an epoch_call
1799 			 */
1800 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1801 			    i++) {
1802 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1803 					IN_MULTI_LOCK_ASSERT();
1804 					in_leavegroup_locked(imo->imo_membership[i], NULL);
1805 					gap++;
1806 				} else if (gap != 0)
1807 					imo->imo_membership[i - gap] =
1808 					    imo->imo_membership[i];
1809 			}
1810 			imo->imo_num_memberships -= gap;
1811 		}
1812 		INP_WUNLOCK(inp);
1813 	}
1814 	INP_INFO_WUNLOCK(pcbinfo);
1815 }
1816 
1817 /*
1818  * Lookup a PCB based on the local address and port.  Caller must hold the
1819  * hash lock.  No inpcb locks or references are acquired.
1820  */
1821 #define INP_LOOKUP_MAPPED_PCB_COST	3
1822 struct inpcb *
1823 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1824     u_short lport, int lookupflags, struct ucred *cred)
1825 {
1826 	struct inpcb *inp;
1827 #ifdef INET6
1828 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1829 #else
1830 	int matchwild = 3;
1831 #endif
1832 	int wildcard;
1833 
1834 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1835 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1836 
1837 	INP_HASH_LOCK_ASSERT(pcbinfo);
1838 
1839 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1840 		struct inpcbhead *head;
1841 		/*
1842 		 * Look for an unconnected (wildcard foreign addr) PCB that
1843 		 * matches the local address and port we're looking for.
1844 		 */
1845 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1846 		    0, pcbinfo->ipi_hashmask)];
1847 		CK_LIST_FOREACH(inp, head, inp_hash) {
1848 #ifdef INET6
1849 			/* XXX inp locking */
1850 			if ((inp->inp_vflag & INP_IPV4) == 0)
1851 				continue;
1852 #endif
1853 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1854 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1855 			    inp->inp_lport == lport) {
1856 				/*
1857 				 * Found?
1858 				 */
1859 				if (cred == NULL ||
1860 				    prison_equal_ip4(cred->cr_prison,
1861 					inp->inp_cred->cr_prison))
1862 					return (inp);
1863 			}
1864 		}
1865 		/*
1866 		 * Not found.
1867 		 */
1868 		return (NULL);
1869 	} else {
1870 		struct inpcbporthead *porthash;
1871 		struct inpcbport *phd;
1872 		struct inpcb *match = NULL;
1873 		/*
1874 		 * Best fit PCB lookup.
1875 		 *
1876 		 * First see if this local port is in use by looking on the
1877 		 * port hash list.
1878 		 */
1879 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1880 		    pcbinfo->ipi_porthashmask)];
1881 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
1882 			if (phd->phd_port == lport)
1883 				break;
1884 		}
1885 		if (phd != NULL) {
1886 			/*
1887 			 * Port is in use by one or more PCBs. Look for best
1888 			 * fit.
1889 			 */
1890 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1891 				wildcard = 0;
1892 				if (cred != NULL &&
1893 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1894 					cred->cr_prison))
1895 					continue;
1896 #ifdef INET6
1897 				/* XXX inp locking */
1898 				if ((inp->inp_vflag & INP_IPV4) == 0)
1899 					continue;
1900 				/*
1901 				 * We never select the PCB that has
1902 				 * INP_IPV6 flag and is bound to :: if
1903 				 * we have another PCB which is bound
1904 				 * to 0.0.0.0.  If a PCB has the
1905 				 * INP_IPV6 flag, then we set its cost
1906 				 * higher than IPv4 only PCBs.
1907 				 *
1908 				 * Note that the case only happens
1909 				 * when a socket is bound to ::, under
1910 				 * the condition that the use of the
1911 				 * mapped address is allowed.
1912 				 */
1913 				if ((inp->inp_vflag & INP_IPV6) != 0)
1914 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1915 #endif
1916 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1917 					wildcard++;
1918 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1919 					if (laddr.s_addr == INADDR_ANY)
1920 						wildcard++;
1921 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1922 						continue;
1923 				} else {
1924 					if (laddr.s_addr != INADDR_ANY)
1925 						wildcard++;
1926 				}
1927 				if (wildcard < matchwild) {
1928 					match = inp;
1929 					matchwild = wildcard;
1930 					if (matchwild == 0)
1931 						break;
1932 				}
1933 			}
1934 		}
1935 		return (match);
1936 	}
1937 }
1938 #undef INP_LOOKUP_MAPPED_PCB_COST
1939 
1940 static struct inpcb *
1941 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1942     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1943     uint16_t fport, int lookupflags)
1944 {
1945 	struct inpcb *local_wild;
1946 	const struct inpcblbgrouphead *hdr;
1947 	struct inpcblbgroup *grp;
1948 	uint32_t idx;
1949 
1950 	INP_HASH_LOCK_ASSERT(pcbinfo);
1951 
1952 	hdr = &pcbinfo->ipi_lbgrouphashbase[
1953 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1954 
1955 	/*
1956 	 * Order of socket selection:
1957 	 * 1. non-wild.
1958 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1959 	 *
1960 	 * NOTE:
1961 	 * - Load balanced group does not contain jailed sockets
1962 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1963 	 */
1964 	local_wild = NULL;
1965 	CK_LIST_FOREACH(grp, hdr, il_list) {
1966 #ifdef INET6
1967 		if (!(grp->il_vflag & INP_IPV4))
1968 			continue;
1969 #endif
1970 		if (grp->il_lport != lport)
1971 			continue;
1972 
1973 		idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1974 		    grp->il_inpcnt;
1975 		if (grp->il_laddr.s_addr == laddr->s_addr)
1976 			return (grp->il_inp[idx]);
1977 		if (grp->il_laddr.s_addr == INADDR_ANY &&
1978 		    (lookupflags & INPLOOKUP_WILDCARD) != 0)
1979 			local_wild = grp->il_inp[idx];
1980 	}
1981 	return (local_wild);
1982 }
1983 
1984 #ifdef PCBGROUP
1985 /*
1986  * Lookup PCB in hash list, using pcbgroup tables.
1987  */
1988 static struct inpcb *
1989 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1990     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1991     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1992 {
1993 	struct inpcbhead *head;
1994 	struct inpcb *inp, *tmpinp;
1995 	u_short fport = fport_arg, lport = lport_arg;
1996 	bool locked;
1997 
1998 	/*
1999 	 * First look for an exact match.
2000 	 */
2001 	tmpinp = NULL;
2002 	INP_GROUP_LOCK(pcbgroup);
2003 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2004 	    pcbgroup->ipg_hashmask)];
2005 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2006 #ifdef INET6
2007 		/* XXX inp locking */
2008 		if ((inp->inp_vflag & INP_IPV4) == 0)
2009 			continue;
2010 #endif
2011 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2012 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2013 		    inp->inp_fport == fport &&
2014 		    inp->inp_lport == lport) {
2015 			/*
2016 			 * XXX We should be able to directly return
2017 			 * the inp here, without any checks.
2018 			 * Well unless both bound with SO_REUSEPORT?
2019 			 */
2020 			if (prison_flag(inp->inp_cred, PR_IP4))
2021 				goto found;
2022 			if (tmpinp == NULL)
2023 				tmpinp = inp;
2024 		}
2025 	}
2026 	if (tmpinp != NULL) {
2027 		inp = tmpinp;
2028 		goto found;
2029 	}
2030 
2031 #ifdef	RSS
2032 	/*
2033 	 * For incoming connections, we may wish to do a wildcard
2034 	 * match for an RSS-local socket.
2035 	 */
2036 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2037 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2038 #ifdef INET6
2039 		struct inpcb *local_wild_mapped = NULL;
2040 #endif
2041 		struct inpcb *jail_wild = NULL;
2042 		struct inpcbhead *head;
2043 		int injail;
2044 
2045 		/*
2046 		 * Order of socket selection - we always prefer jails.
2047 		 *      1. jailed, non-wild.
2048 		 *      2. jailed, wild.
2049 		 *      3. non-jailed, non-wild.
2050 		 *      4. non-jailed, wild.
2051 		 */
2052 
2053 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2054 		    lport, 0, pcbgroup->ipg_hashmask)];
2055 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2056 #ifdef INET6
2057 			/* XXX inp locking */
2058 			if ((inp->inp_vflag & INP_IPV4) == 0)
2059 				continue;
2060 #endif
2061 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2062 			    inp->inp_lport != lport)
2063 				continue;
2064 
2065 			injail = prison_flag(inp->inp_cred, PR_IP4);
2066 			if (injail) {
2067 				if (prison_check_ip4(inp->inp_cred,
2068 				    &laddr) != 0)
2069 					continue;
2070 			} else {
2071 				if (local_exact != NULL)
2072 					continue;
2073 			}
2074 
2075 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2076 				if (injail)
2077 					goto found;
2078 				else
2079 					local_exact = inp;
2080 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2081 #ifdef INET6
2082 				/* XXX inp locking, NULL check */
2083 				if (inp->inp_vflag & INP_IPV6PROTO)
2084 					local_wild_mapped = inp;
2085 				else
2086 #endif
2087 					if (injail)
2088 						jail_wild = inp;
2089 					else
2090 						local_wild = inp;
2091 			}
2092 		} /* LIST_FOREACH */
2093 
2094 		inp = jail_wild;
2095 		if (inp == NULL)
2096 			inp = local_exact;
2097 		if (inp == NULL)
2098 			inp = local_wild;
2099 #ifdef INET6
2100 		if (inp == NULL)
2101 			inp = local_wild_mapped;
2102 #endif
2103 		if (inp != NULL)
2104 			goto found;
2105 	}
2106 #endif
2107 
2108 	/*
2109 	 * Then look for a wildcard match, if requested.
2110 	 */
2111 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2112 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2113 #ifdef INET6
2114 		struct inpcb *local_wild_mapped = NULL;
2115 #endif
2116 		struct inpcb *jail_wild = NULL;
2117 		struct inpcbhead *head;
2118 		int injail;
2119 
2120 		/*
2121 		 * Order of socket selection - we always prefer jails.
2122 		 *      1. jailed, non-wild.
2123 		 *      2. jailed, wild.
2124 		 *      3. non-jailed, non-wild.
2125 		 *      4. non-jailed, wild.
2126 		 */
2127 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2128 		    0, pcbinfo->ipi_wildmask)];
2129 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2130 #ifdef INET6
2131 			/* XXX inp locking */
2132 			if ((inp->inp_vflag & INP_IPV4) == 0)
2133 				continue;
2134 #endif
2135 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2136 			    inp->inp_lport != lport)
2137 				continue;
2138 
2139 			injail = prison_flag(inp->inp_cred, PR_IP4);
2140 			if (injail) {
2141 				if (prison_check_ip4(inp->inp_cred,
2142 				    &laddr) != 0)
2143 					continue;
2144 			} else {
2145 				if (local_exact != NULL)
2146 					continue;
2147 			}
2148 
2149 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2150 				if (injail)
2151 					goto found;
2152 				else
2153 					local_exact = inp;
2154 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2155 #ifdef INET6
2156 				/* XXX inp locking, NULL check */
2157 				if (inp->inp_vflag & INP_IPV6PROTO)
2158 					local_wild_mapped = inp;
2159 				else
2160 #endif
2161 					if (injail)
2162 						jail_wild = inp;
2163 					else
2164 						local_wild = inp;
2165 			}
2166 		} /* LIST_FOREACH */
2167 		inp = jail_wild;
2168 		if (inp == NULL)
2169 			inp = local_exact;
2170 		if (inp == NULL)
2171 			inp = local_wild;
2172 #ifdef INET6
2173 		if (inp == NULL)
2174 			inp = local_wild_mapped;
2175 #endif
2176 		if (inp != NULL)
2177 			goto found;
2178 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2179 	INP_GROUP_UNLOCK(pcbgroup);
2180 	return (NULL);
2181 
2182 found:
2183 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2184 		locked = INP_TRY_WLOCK(inp);
2185 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2186 		locked = INP_TRY_RLOCK(inp);
2187 	else
2188 		panic("%s: locking bug", __func__);
2189 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2190 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2191 			INP_WUNLOCK(inp);
2192 		else
2193 			INP_RUNLOCK(inp);
2194 		return (NULL);
2195 	} else if (!locked)
2196 		in_pcbref(inp);
2197 	INP_GROUP_UNLOCK(pcbgroup);
2198 	if (!locked) {
2199 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2200 			INP_WLOCK(inp);
2201 			if (in_pcbrele_wlocked(inp))
2202 				return (NULL);
2203 		} else {
2204 			INP_RLOCK(inp);
2205 			if (in_pcbrele_rlocked(inp))
2206 				return (NULL);
2207 		}
2208 	}
2209 #ifdef INVARIANTS
2210 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2211 		INP_WLOCK_ASSERT(inp);
2212 	else
2213 		INP_RLOCK_ASSERT(inp);
2214 #endif
2215 	return (inp);
2216 }
2217 #endif /* PCBGROUP */
2218 
2219 /*
2220  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2221  * that the caller has locked the hash list, and will not perform any further
2222  * locking or reference operations on either the hash list or the connection.
2223  */
2224 static struct inpcb *
2225 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2226     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2227     struct ifnet *ifp)
2228 {
2229 	struct inpcbhead *head;
2230 	struct inpcb *inp, *tmpinp;
2231 	u_short fport = fport_arg, lport = lport_arg;
2232 
2233 #ifdef INVARIANTS
2234 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2235 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2236 	if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2237 		MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2238 #endif
2239 	/*
2240 	 * First look for an exact match.
2241 	 */
2242 	tmpinp = NULL;
2243 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2244 	    pcbinfo->ipi_hashmask)];
2245 	CK_LIST_FOREACH(inp, head, inp_hash) {
2246 #ifdef INET6
2247 		/* XXX inp locking */
2248 		if ((inp->inp_vflag & INP_IPV4) == 0)
2249 			continue;
2250 #endif
2251 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2252 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2253 		    inp->inp_fport == fport &&
2254 		    inp->inp_lport == lport) {
2255 			/*
2256 			 * XXX We should be able to directly return
2257 			 * the inp here, without any checks.
2258 			 * Well unless both bound with SO_REUSEPORT?
2259 			 */
2260 			if (prison_flag(inp->inp_cred, PR_IP4))
2261 				return (inp);
2262 			if (tmpinp == NULL)
2263 				tmpinp = inp;
2264 		}
2265 	}
2266 	if (tmpinp != NULL)
2267 		return (tmpinp);
2268 
2269 	/*
2270 	 * Then look in lb group (for wildcard match).
2271 	 */
2272 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2273 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2274 		    fport, lookupflags);
2275 		if (inp != NULL)
2276 			return (inp);
2277 	}
2278 
2279 	/*
2280 	 * Then look for a wildcard match, if requested.
2281 	 */
2282 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2283 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2284 #ifdef INET6
2285 		struct inpcb *local_wild_mapped = NULL;
2286 #endif
2287 		struct inpcb *jail_wild = NULL;
2288 		int injail;
2289 
2290 		/*
2291 		 * Order of socket selection - we always prefer jails.
2292 		 *      1. jailed, non-wild.
2293 		 *      2. jailed, wild.
2294 		 *      3. non-jailed, non-wild.
2295 		 *      4. non-jailed, wild.
2296 		 */
2297 
2298 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2299 		    0, pcbinfo->ipi_hashmask)];
2300 		CK_LIST_FOREACH(inp, head, inp_hash) {
2301 #ifdef INET6
2302 			/* XXX inp locking */
2303 			if ((inp->inp_vflag & INP_IPV4) == 0)
2304 				continue;
2305 #endif
2306 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2307 			    inp->inp_lport != lport)
2308 				continue;
2309 
2310 			injail = prison_flag(inp->inp_cred, PR_IP4);
2311 			if (injail) {
2312 				if (prison_check_ip4(inp->inp_cred,
2313 				    &laddr) != 0)
2314 					continue;
2315 			} else {
2316 				if (local_exact != NULL)
2317 					continue;
2318 			}
2319 
2320 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2321 				if (injail)
2322 					return (inp);
2323 				else
2324 					local_exact = inp;
2325 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2326 #ifdef INET6
2327 				/* XXX inp locking, NULL check */
2328 				if (inp->inp_vflag & INP_IPV6PROTO)
2329 					local_wild_mapped = inp;
2330 				else
2331 #endif
2332 					if (injail)
2333 						jail_wild = inp;
2334 					else
2335 						local_wild = inp;
2336 			}
2337 		} /* LIST_FOREACH */
2338 		if (jail_wild != NULL)
2339 			return (jail_wild);
2340 		if (local_exact != NULL)
2341 			return (local_exact);
2342 		if (local_wild != NULL)
2343 			return (local_wild);
2344 #ifdef INET6
2345 		if (local_wild_mapped != NULL)
2346 			return (local_wild_mapped);
2347 #endif
2348 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2349 
2350 	return (NULL);
2351 }
2352 
2353 /*
2354  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2355  * hash list lock, and will return the inpcb locked (i.e., requires
2356  * INPLOOKUP_LOCKPCB).
2357  */
2358 static struct inpcb *
2359 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2360     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2361     struct ifnet *ifp)
2362 {
2363 	struct inpcb *inp;
2364 
2365 	INP_HASH_RLOCK(pcbinfo);
2366 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2367 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2368 	if (inp != NULL) {
2369 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2370 			INP_WLOCK(inp);
2371 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2372 				INP_WUNLOCK(inp);
2373 				inp = NULL;
2374 			}
2375 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2376 			INP_RLOCK(inp);
2377 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2378 				INP_RUNLOCK(inp);
2379 				inp = NULL;
2380 			}
2381 		} else
2382 			panic("%s: locking bug", __func__);
2383 #ifdef INVARIANTS
2384 		if (inp != NULL) {
2385 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2386 				INP_WLOCK_ASSERT(inp);
2387 			else
2388 				INP_RLOCK_ASSERT(inp);
2389 		}
2390 #endif
2391 	}
2392 	INP_HASH_RUNLOCK(pcbinfo);
2393 	return (inp);
2394 }
2395 
2396 /*
2397  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2398  * from which a pre-calculated hash value may be extracted.
2399  *
2400  * Possibly more of this logic should be in in_pcbgroup.c.
2401  */
2402 struct inpcb *
2403 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2404     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2405 {
2406 #if defined(PCBGROUP) && !defined(RSS)
2407 	struct inpcbgroup *pcbgroup;
2408 #endif
2409 
2410 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2411 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2412 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2413 	    ("%s: LOCKPCB not set", __func__));
2414 
2415 	/*
2416 	 * When not using RSS, use connection groups in preference to the
2417 	 * reservation table when looking up 4-tuples.  When using RSS, just
2418 	 * use the reservation table, due to the cost of the Toeplitz hash
2419 	 * in software.
2420 	 *
2421 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2422 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2423 	 * in software.
2424 	 */
2425 #if defined(PCBGROUP) && !defined(RSS)
2426 	if (in_pcbgroup_enabled(pcbinfo)) {
2427 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2428 		    fport);
2429 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2430 		    laddr, lport, lookupflags, ifp));
2431 	}
2432 #endif
2433 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2434 	    lookupflags, ifp));
2435 }
2436 
2437 struct inpcb *
2438 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2439     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2440     struct ifnet *ifp, struct mbuf *m)
2441 {
2442 #ifdef PCBGROUP
2443 	struct inpcbgroup *pcbgroup;
2444 #endif
2445 
2446 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2447 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2448 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2449 	    ("%s: LOCKPCB not set", __func__));
2450 
2451 #ifdef PCBGROUP
2452 	/*
2453 	 * If we can use a hardware-generated hash to look up the connection
2454 	 * group, use that connection group to find the inpcb.  Otherwise
2455 	 * fall back on a software hash -- or the reservation table if we're
2456 	 * using RSS.
2457 	 *
2458 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2459 	 */
2460 	if (in_pcbgroup_enabled(pcbinfo) &&
2461 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2462 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2463 		    m->m_pkthdr.flowid);
2464 		if (pcbgroup != NULL)
2465 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2466 			    fport, laddr, lport, lookupflags, ifp));
2467 #ifndef RSS
2468 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2469 		    fport);
2470 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2471 		    laddr, lport, lookupflags, ifp));
2472 #endif
2473 	}
2474 #endif
2475 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2476 	    lookupflags, ifp));
2477 }
2478 #endif /* INET */
2479 
2480 /*
2481  * Insert PCB onto various hash lists.
2482  */
2483 static int
2484 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2485 {
2486 	struct inpcbhead *pcbhash;
2487 	struct inpcbporthead *pcbporthash;
2488 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2489 	struct inpcbport *phd;
2490 	u_int32_t hashkey_faddr;
2491 	int so_options;
2492 
2493 	INP_WLOCK_ASSERT(inp);
2494 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2495 
2496 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2497 	    ("in_pcbinshash: INP_INHASHLIST"));
2498 
2499 #ifdef INET6
2500 	if (inp->inp_vflag & INP_IPV6)
2501 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2502 	else
2503 #endif
2504 	hashkey_faddr = inp->inp_faddr.s_addr;
2505 
2506 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2507 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2508 
2509 	pcbporthash = &pcbinfo->ipi_porthashbase[
2510 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2511 
2512 	/*
2513 	 * Add entry to load balance group.
2514 	 * Only do this if SO_REUSEPORT_LB is set.
2515 	 */
2516 	so_options = inp_so_options(inp);
2517 	if (so_options & SO_REUSEPORT_LB) {
2518 		int ret = in_pcbinslbgrouphash(inp);
2519 		if (ret) {
2520 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2521 			return (ret);
2522 		}
2523 	}
2524 
2525 	/*
2526 	 * Go through port list and look for a head for this lport.
2527 	 */
2528 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2529 		if (phd->phd_port == inp->inp_lport)
2530 			break;
2531 	}
2532 	/*
2533 	 * If none exists, malloc one and tack it on.
2534 	 */
2535 	if (phd == NULL) {
2536 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2537 		if (phd == NULL) {
2538 			return (ENOBUFS); /* XXX */
2539 		}
2540 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2541 		phd->phd_port = inp->inp_lport;
2542 		CK_LIST_INIT(&phd->phd_pcblist);
2543 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2544 	}
2545 	inp->inp_phd = phd;
2546 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2547 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2548 	inp->inp_flags |= INP_INHASHLIST;
2549 #ifdef PCBGROUP
2550 	if (do_pcbgroup_update)
2551 		in_pcbgroup_update(inp);
2552 #endif
2553 	return (0);
2554 }
2555 
2556 /*
2557  * For now, there are two public interfaces to insert an inpcb into the hash
2558  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2559  * is used only in the TCP syncache, where in_pcbinshash is called before the
2560  * full 4-tuple is set for the inpcb, and we don't want to install in the
2561  * pcbgroup until later.
2562  *
2563  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2564  * connection groups, and partially initialised inpcbs should not be exposed
2565  * to either reservation hash tables or pcbgroups.
2566  */
2567 int
2568 in_pcbinshash(struct inpcb *inp)
2569 {
2570 
2571 	return (in_pcbinshash_internal(inp, 1));
2572 }
2573 
2574 int
2575 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2576 {
2577 
2578 	return (in_pcbinshash_internal(inp, 0));
2579 }
2580 
2581 /*
2582  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2583  * changed. NOTE: This does not handle the case of the lport changing (the
2584  * hashed port list would have to be updated as well), so the lport must
2585  * not change after in_pcbinshash() has been called.
2586  */
2587 void
2588 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2589 {
2590 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2591 	struct inpcbhead *head;
2592 	u_int32_t hashkey_faddr;
2593 
2594 	INP_WLOCK_ASSERT(inp);
2595 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2596 
2597 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2598 	    ("in_pcbrehash: !INP_INHASHLIST"));
2599 
2600 #ifdef INET6
2601 	if (inp->inp_vflag & INP_IPV6)
2602 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2603 	else
2604 #endif
2605 	hashkey_faddr = inp->inp_faddr.s_addr;
2606 
2607 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2608 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2609 
2610 	CK_LIST_REMOVE(inp, inp_hash);
2611 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2612 
2613 #ifdef PCBGROUP
2614 	if (m != NULL)
2615 		in_pcbgroup_update_mbuf(inp, m);
2616 	else
2617 		in_pcbgroup_update(inp);
2618 #endif
2619 }
2620 
2621 void
2622 in_pcbrehash(struct inpcb *inp)
2623 {
2624 
2625 	in_pcbrehash_mbuf(inp, NULL);
2626 }
2627 
2628 /*
2629  * Remove PCB from various lists.
2630  */
2631 static void
2632 in_pcbremlists(struct inpcb *inp)
2633 {
2634 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2635 
2636 #ifdef INVARIANTS
2637 	if (pcbinfo == &V_tcbinfo) {
2638 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2639 	} else {
2640 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2641 	}
2642 #endif
2643 
2644 	INP_WLOCK_ASSERT(inp);
2645 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2646 
2647 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2648 	if (inp->inp_flags & INP_INHASHLIST) {
2649 		struct inpcbport *phd = inp->inp_phd;
2650 
2651 		INP_HASH_WLOCK(pcbinfo);
2652 
2653 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2654 		in_pcbremlbgrouphash(inp);
2655 
2656 		CK_LIST_REMOVE(inp, inp_hash);
2657 		CK_LIST_REMOVE(inp, inp_portlist);
2658 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2659 			CK_LIST_REMOVE(phd, phd_hash);
2660 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2661 		}
2662 		INP_HASH_WUNLOCK(pcbinfo);
2663 		inp->inp_flags &= ~INP_INHASHLIST;
2664 	}
2665 	CK_LIST_REMOVE(inp, inp_list);
2666 	pcbinfo->ipi_count--;
2667 #ifdef PCBGROUP
2668 	in_pcbgroup_remove(inp);
2669 #endif
2670 }
2671 
2672 /*
2673  * Check for alternatives when higher level complains
2674  * about service problems.  For now, invalidate cached
2675  * routing information.  If the route was created dynamically
2676  * (by a redirect), time to try a default gateway again.
2677  */
2678 void
2679 in_losing(struct inpcb *inp)
2680 {
2681 
2682 	RO_INVALIDATE_CACHE(&inp->inp_route);
2683 	return;
2684 }
2685 
2686 /*
2687  * A set label operation has occurred at the socket layer, propagate the
2688  * label change into the in_pcb for the socket.
2689  */
2690 void
2691 in_pcbsosetlabel(struct socket *so)
2692 {
2693 #ifdef MAC
2694 	struct inpcb *inp;
2695 
2696 	inp = sotoinpcb(so);
2697 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2698 
2699 	INP_WLOCK(inp);
2700 	SOCK_LOCK(so);
2701 	mac_inpcb_sosetlabel(so, inp);
2702 	SOCK_UNLOCK(so);
2703 	INP_WUNLOCK(inp);
2704 #endif
2705 }
2706 
2707 /*
2708  * ipport_tick runs once per second, determining if random port allocation
2709  * should be continued.  If more than ipport_randomcps ports have been
2710  * allocated in the last second, then we return to sequential port
2711  * allocation. We return to random allocation only once we drop below
2712  * ipport_randomcps for at least ipport_randomtime seconds.
2713  */
2714 static void
2715 ipport_tick(void *xtp)
2716 {
2717 	VNET_ITERATOR_DECL(vnet_iter);
2718 
2719 	VNET_LIST_RLOCK_NOSLEEP();
2720 	VNET_FOREACH(vnet_iter) {
2721 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2722 		if (V_ipport_tcpallocs <=
2723 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2724 			if (V_ipport_stoprandom > 0)
2725 				V_ipport_stoprandom--;
2726 		} else
2727 			V_ipport_stoprandom = V_ipport_randomtime;
2728 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2729 		CURVNET_RESTORE();
2730 	}
2731 	VNET_LIST_RUNLOCK_NOSLEEP();
2732 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2733 }
2734 
2735 static void
2736 ip_fini(void *xtp)
2737 {
2738 
2739 	callout_stop(&ipport_tick_callout);
2740 }
2741 
2742 /*
2743  * The ipport_callout should start running at about the time we attach the
2744  * inet or inet6 domains.
2745  */
2746 static void
2747 ipport_tick_init(const void *unused __unused)
2748 {
2749 
2750 	/* Start ipport_tick. */
2751 	callout_init(&ipport_tick_callout, 1);
2752 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2753 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2754 		SHUTDOWN_PRI_DEFAULT);
2755 }
2756 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2757     ipport_tick_init, NULL);
2758 
2759 void
2760 inp_wlock(struct inpcb *inp)
2761 {
2762 
2763 	INP_WLOCK(inp);
2764 }
2765 
2766 void
2767 inp_wunlock(struct inpcb *inp)
2768 {
2769 
2770 	INP_WUNLOCK(inp);
2771 }
2772 
2773 void
2774 inp_rlock(struct inpcb *inp)
2775 {
2776 
2777 	INP_RLOCK(inp);
2778 }
2779 
2780 void
2781 inp_runlock(struct inpcb *inp)
2782 {
2783 
2784 	INP_RUNLOCK(inp);
2785 }
2786 
2787 #ifdef INVARIANT_SUPPORT
2788 void
2789 inp_lock_assert(struct inpcb *inp)
2790 {
2791 
2792 	INP_WLOCK_ASSERT(inp);
2793 }
2794 
2795 void
2796 inp_unlock_assert(struct inpcb *inp)
2797 {
2798 
2799 	INP_UNLOCK_ASSERT(inp);
2800 }
2801 #endif
2802 
2803 void
2804 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2805 {
2806 	struct inpcb *inp;
2807 
2808 	INP_INFO_WLOCK(&V_tcbinfo);
2809 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2810 		INP_WLOCK(inp);
2811 		func(inp, arg);
2812 		INP_WUNLOCK(inp);
2813 	}
2814 	INP_INFO_WUNLOCK(&V_tcbinfo);
2815 }
2816 
2817 struct socket *
2818 inp_inpcbtosocket(struct inpcb *inp)
2819 {
2820 
2821 	INP_WLOCK_ASSERT(inp);
2822 	return (inp->inp_socket);
2823 }
2824 
2825 struct tcpcb *
2826 inp_inpcbtotcpcb(struct inpcb *inp)
2827 {
2828 
2829 	INP_WLOCK_ASSERT(inp);
2830 	return ((struct tcpcb *)inp->inp_ppcb);
2831 }
2832 
2833 int
2834 inp_ip_tos_get(const struct inpcb *inp)
2835 {
2836 
2837 	return (inp->inp_ip_tos);
2838 }
2839 
2840 void
2841 inp_ip_tos_set(struct inpcb *inp, int val)
2842 {
2843 
2844 	inp->inp_ip_tos = val;
2845 }
2846 
2847 void
2848 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2849     uint32_t *faddr, uint16_t *fp)
2850 {
2851 
2852 	INP_LOCK_ASSERT(inp);
2853 	*laddr = inp->inp_laddr.s_addr;
2854 	*faddr = inp->inp_faddr.s_addr;
2855 	*lp = inp->inp_lport;
2856 	*fp = inp->inp_fport;
2857 }
2858 
2859 struct inpcb *
2860 so_sotoinpcb(struct socket *so)
2861 {
2862 
2863 	return (sotoinpcb(so));
2864 }
2865 
2866 struct tcpcb *
2867 so_sototcpcb(struct socket *so)
2868 {
2869 
2870 	return (sototcpcb(so));
2871 }
2872 
2873 /*
2874  * Create an external-format (``xinpcb'') structure using the information in
2875  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2876  * reduce the spew of irrelevant information over this interface, to isolate
2877  * user code from changes in the kernel structure, and potentially to provide
2878  * information-hiding if we decide that some of this information should be
2879  * hidden from users.
2880  */
2881 void
2882 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2883 {
2884 
2885 	bzero(xi, sizeof(*xi));
2886 	xi->xi_len = sizeof(struct xinpcb);
2887 	if (inp->inp_socket)
2888 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2889 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2890 	xi->inp_gencnt = inp->inp_gencnt;
2891 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2892 	xi->inp_flow = inp->inp_flow;
2893 	xi->inp_flowid = inp->inp_flowid;
2894 	xi->inp_flowtype = inp->inp_flowtype;
2895 	xi->inp_flags = inp->inp_flags;
2896 	xi->inp_flags2 = inp->inp_flags2;
2897 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2898 	xi->in6p_cksum = inp->in6p_cksum;
2899 	xi->in6p_hops = inp->in6p_hops;
2900 	xi->inp_ip_tos = inp->inp_ip_tos;
2901 	xi->inp_vflag = inp->inp_vflag;
2902 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2903 	xi->inp_ip_p = inp->inp_ip_p;
2904 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2905 }
2906 
2907 #ifdef DDB
2908 static void
2909 db_print_indent(int indent)
2910 {
2911 	int i;
2912 
2913 	for (i = 0; i < indent; i++)
2914 		db_printf(" ");
2915 }
2916 
2917 static void
2918 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2919 {
2920 	char faddr_str[48], laddr_str[48];
2921 
2922 	db_print_indent(indent);
2923 	db_printf("%s at %p\n", name, inc);
2924 
2925 	indent += 2;
2926 
2927 #ifdef INET6
2928 	if (inc->inc_flags & INC_ISIPV6) {
2929 		/* IPv6. */
2930 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2931 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2932 	} else
2933 #endif
2934 	{
2935 		/* IPv4. */
2936 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2937 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2938 	}
2939 	db_print_indent(indent);
2940 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2941 	    ntohs(inc->inc_lport));
2942 	db_print_indent(indent);
2943 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2944 	    ntohs(inc->inc_fport));
2945 }
2946 
2947 static void
2948 db_print_inpflags(int inp_flags)
2949 {
2950 	int comma;
2951 
2952 	comma = 0;
2953 	if (inp_flags & INP_RECVOPTS) {
2954 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2955 		comma = 1;
2956 	}
2957 	if (inp_flags & INP_RECVRETOPTS) {
2958 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2959 		comma = 1;
2960 	}
2961 	if (inp_flags & INP_RECVDSTADDR) {
2962 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2963 		comma = 1;
2964 	}
2965 	if (inp_flags & INP_ORIGDSTADDR) {
2966 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2967 		comma = 1;
2968 	}
2969 	if (inp_flags & INP_HDRINCL) {
2970 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2971 		comma = 1;
2972 	}
2973 	if (inp_flags & INP_HIGHPORT) {
2974 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2975 		comma = 1;
2976 	}
2977 	if (inp_flags & INP_LOWPORT) {
2978 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2979 		comma = 1;
2980 	}
2981 	if (inp_flags & INP_ANONPORT) {
2982 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2983 		comma = 1;
2984 	}
2985 	if (inp_flags & INP_RECVIF) {
2986 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2987 		comma = 1;
2988 	}
2989 	if (inp_flags & INP_MTUDISC) {
2990 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2991 		comma = 1;
2992 	}
2993 	if (inp_flags & INP_RECVTTL) {
2994 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2995 		comma = 1;
2996 	}
2997 	if (inp_flags & INP_DONTFRAG) {
2998 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2999 		comma = 1;
3000 	}
3001 	if (inp_flags & INP_RECVTOS) {
3002 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3003 		comma = 1;
3004 	}
3005 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3006 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3007 		comma = 1;
3008 	}
3009 	if (inp_flags & IN6P_PKTINFO) {
3010 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3011 		comma = 1;
3012 	}
3013 	if (inp_flags & IN6P_HOPLIMIT) {
3014 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3015 		comma = 1;
3016 	}
3017 	if (inp_flags & IN6P_HOPOPTS) {
3018 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3019 		comma = 1;
3020 	}
3021 	if (inp_flags & IN6P_DSTOPTS) {
3022 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3023 		comma = 1;
3024 	}
3025 	if (inp_flags & IN6P_RTHDR) {
3026 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3027 		comma = 1;
3028 	}
3029 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3030 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3031 		comma = 1;
3032 	}
3033 	if (inp_flags & IN6P_TCLASS) {
3034 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3035 		comma = 1;
3036 	}
3037 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3038 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3039 		comma = 1;
3040 	}
3041 	if (inp_flags & INP_TIMEWAIT) {
3042 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3043 		comma  = 1;
3044 	}
3045 	if (inp_flags & INP_ONESBCAST) {
3046 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3047 		comma  = 1;
3048 	}
3049 	if (inp_flags & INP_DROPPED) {
3050 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3051 		comma  = 1;
3052 	}
3053 	if (inp_flags & INP_SOCKREF) {
3054 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3055 		comma  = 1;
3056 	}
3057 	if (inp_flags & IN6P_RFC2292) {
3058 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3059 		comma = 1;
3060 	}
3061 	if (inp_flags & IN6P_MTU) {
3062 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3063 		comma = 1;
3064 	}
3065 }
3066 
3067 static void
3068 db_print_inpvflag(u_char inp_vflag)
3069 {
3070 	int comma;
3071 
3072 	comma = 0;
3073 	if (inp_vflag & INP_IPV4) {
3074 		db_printf("%sINP_IPV4", comma ? ", " : "");
3075 		comma  = 1;
3076 	}
3077 	if (inp_vflag & INP_IPV6) {
3078 		db_printf("%sINP_IPV6", comma ? ", " : "");
3079 		comma  = 1;
3080 	}
3081 	if (inp_vflag & INP_IPV6PROTO) {
3082 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3083 		comma  = 1;
3084 	}
3085 }
3086 
3087 static void
3088 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3089 {
3090 
3091 	db_print_indent(indent);
3092 	db_printf("%s at %p\n", name, inp);
3093 
3094 	indent += 2;
3095 
3096 	db_print_indent(indent);
3097 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3098 
3099 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3100 
3101 	db_print_indent(indent);
3102 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3103 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3104 
3105 	db_print_indent(indent);
3106 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3107 	   inp->inp_label, inp->inp_flags);
3108 	db_print_inpflags(inp->inp_flags);
3109 	db_printf(")\n");
3110 
3111 	db_print_indent(indent);
3112 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3113 	    inp->inp_vflag);
3114 	db_print_inpvflag(inp->inp_vflag);
3115 	db_printf(")\n");
3116 
3117 	db_print_indent(indent);
3118 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3119 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3120 
3121 	db_print_indent(indent);
3122 #ifdef INET6
3123 	if (inp->inp_vflag & INP_IPV6) {
3124 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3125 		    "in6p_moptions: %p\n", inp->in6p_options,
3126 		    inp->in6p_outputopts, inp->in6p_moptions);
3127 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3128 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3129 		    inp->in6p_hops);
3130 	} else
3131 #endif
3132 	{
3133 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3134 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3135 		    inp->inp_options, inp->inp_moptions);
3136 	}
3137 
3138 	db_print_indent(indent);
3139 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3140 	    (uintmax_t)inp->inp_gencnt);
3141 }
3142 
3143 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3144 {
3145 	struct inpcb *inp;
3146 
3147 	if (!have_addr) {
3148 		db_printf("usage: show inpcb <addr>\n");
3149 		return;
3150 	}
3151 	inp = (struct inpcb *)addr;
3152 
3153 	db_print_inpcb(inp, "inpcb", 0);
3154 }
3155 #endif /* DDB */
3156 
3157 #ifdef RATELIMIT
3158 /*
3159  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3160  * if any.
3161  */
3162 int
3163 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3164 {
3165 	union if_snd_tag_modify_params params = {
3166 		.rate_limit.max_rate = max_pacing_rate,
3167 	};
3168 	struct m_snd_tag *mst;
3169 	struct ifnet *ifp;
3170 	int error;
3171 
3172 	mst = inp->inp_snd_tag;
3173 	if (mst == NULL)
3174 		return (EINVAL);
3175 
3176 	ifp = mst->ifp;
3177 	if (ifp == NULL)
3178 		return (EINVAL);
3179 
3180 	if (ifp->if_snd_tag_modify == NULL) {
3181 		error = EOPNOTSUPP;
3182 	} else {
3183 		error = ifp->if_snd_tag_modify(mst, &params);
3184 	}
3185 	return (error);
3186 }
3187 
3188 /*
3189  * Query existing TX rate limit based on the existing
3190  * "inp->inp_snd_tag", if any.
3191  */
3192 int
3193 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3194 {
3195 	union if_snd_tag_query_params params = { };
3196 	struct m_snd_tag *mst;
3197 	struct ifnet *ifp;
3198 	int error;
3199 
3200 	mst = inp->inp_snd_tag;
3201 	if (mst == NULL)
3202 		return (EINVAL);
3203 
3204 	ifp = mst->ifp;
3205 	if (ifp == NULL)
3206 		return (EINVAL);
3207 
3208 	if (ifp->if_snd_tag_query == NULL) {
3209 		error = EOPNOTSUPP;
3210 	} else {
3211 		error = ifp->if_snd_tag_query(mst, &params);
3212 		if (error == 0 &&  p_max_pacing_rate != NULL)
3213 			*p_max_pacing_rate = params.rate_limit.max_rate;
3214 	}
3215 	return (error);
3216 }
3217 
3218 /*
3219  * Query existing TX queue level based on the existing
3220  * "inp->inp_snd_tag", if any.
3221  */
3222 int
3223 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3224 {
3225 	union if_snd_tag_query_params params = { };
3226 	struct m_snd_tag *mst;
3227 	struct ifnet *ifp;
3228 	int error;
3229 
3230 	mst = inp->inp_snd_tag;
3231 	if (mst == NULL)
3232 		return (EINVAL);
3233 
3234 	ifp = mst->ifp;
3235 	if (ifp == NULL)
3236 		return (EINVAL);
3237 
3238 	if (ifp->if_snd_tag_query == NULL)
3239 		return (EOPNOTSUPP);
3240 
3241 	error = ifp->if_snd_tag_query(mst, &params);
3242 	if (error == 0 &&  p_txqueue_level != NULL)
3243 		*p_txqueue_level = params.rate_limit.queue_level;
3244 	return (error);
3245 }
3246 
3247 /*
3248  * Allocate a new TX rate limit send tag from the network interface
3249  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3250  */
3251 int
3252 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3253     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3254 {
3255 	union if_snd_tag_alloc_params params = {
3256 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3257 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3258 		.rate_limit.hdr.flowid = flowid,
3259 		.rate_limit.hdr.flowtype = flowtype,
3260 		.rate_limit.max_rate = max_pacing_rate,
3261 	};
3262 	int error;
3263 
3264 	INP_WLOCK_ASSERT(inp);
3265 
3266 	if (inp->inp_snd_tag != NULL)
3267 		return (EINVAL);
3268 
3269 	if (ifp->if_snd_tag_alloc == NULL) {
3270 		error = EOPNOTSUPP;
3271 	} else {
3272 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3273 
3274 		/*
3275 		 * At success increment the refcount on
3276 		 * the send tag's network interface:
3277 		 */
3278 		if (error == 0)
3279 			if_ref(inp->inp_snd_tag->ifp);
3280 	}
3281 	return (error);
3282 }
3283 
3284 /*
3285  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3286  * if any:
3287  */
3288 void
3289 in_pcbdetach_txrtlmt(struct inpcb *inp)
3290 {
3291 	struct m_snd_tag *mst;
3292 	struct ifnet *ifp;
3293 
3294 	INP_WLOCK_ASSERT(inp);
3295 
3296 	mst = inp->inp_snd_tag;
3297 	inp->inp_snd_tag = NULL;
3298 
3299 	if (mst == NULL)
3300 		return;
3301 
3302 	ifp = mst->ifp;
3303 	if (ifp == NULL)
3304 		return;
3305 
3306 	/*
3307 	 * If the device was detached while we still had reference(s)
3308 	 * on the ifp, we assume if_snd_tag_free() was replaced with
3309 	 * stubs.
3310 	 */
3311 	ifp->if_snd_tag_free(mst);
3312 
3313 	/* release reference count on network interface */
3314 	if_rele(ifp);
3315 }
3316 
3317 /*
3318  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3319  * is set in the fast path and will attach/detach/modify the TX rate
3320  * limit send tag based on the socket's so_max_pacing_rate value.
3321  */
3322 void
3323 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3324 {
3325 	struct socket *socket;
3326 	uint32_t max_pacing_rate;
3327 	bool did_upgrade;
3328 	int error;
3329 
3330 	if (inp == NULL)
3331 		return;
3332 
3333 	socket = inp->inp_socket;
3334 	if (socket == NULL)
3335 		return;
3336 
3337 	if (!INP_WLOCKED(inp)) {
3338 		/*
3339 		 * NOTE: If the write locking fails, we need to bail
3340 		 * out and use the non-ratelimited ring for the
3341 		 * transmit until there is a new chance to get the
3342 		 * write lock.
3343 		 */
3344 		if (!INP_TRY_UPGRADE(inp))
3345 			return;
3346 		did_upgrade = 1;
3347 	} else {
3348 		did_upgrade = 0;
3349 	}
3350 
3351 	/*
3352 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3353 	 * because atomic updates are not required since the variable
3354 	 * is checked at every mbuf we send. It is assumed that the
3355 	 * variable read itself will be atomic.
3356 	 */
3357 	max_pacing_rate = socket->so_max_pacing_rate;
3358 
3359 	/*
3360 	 * NOTE: When attaching to a network interface a reference is
3361 	 * made to ensure the network interface doesn't go away until
3362 	 * all ratelimit connections are gone. The network interface
3363 	 * pointers compared below represent valid network interfaces,
3364 	 * except when comparing towards NULL.
3365 	 */
3366 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3367 		error = 0;
3368 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3369 		if (inp->inp_snd_tag != NULL)
3370 			in_pcbdetach_txrtlmt(inp);
3371 		error = 0;
3372 	} else if (inp->inp_snd_tag == NULL) {
3373 		/*
3374 		 * In order to utilize packet pacing with RSS, we need
3375 		 * to wait until there is a valid RSS hash before we
3376 		 * can proceed:
3377 		 */
3378 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3379 			error = EAGAIN;
3380 		} else {
3381 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3382 			    mb->m_pkthdr.flowid, max_pacing_rate);
3383 		}
3384 	} else {
3385 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3386 	}
3387 	if (error == 0 || error == EOPNOTSUPP)
3388 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3389 	if (did_upgrade)
3390 		INP_DOWNGRADE(inp);
3391 }
3392 
3393 /*
3394  * Track route changes for TX rate limiting.
3395  */
3396 void
3397 in_pcboutput_eagain(struct inpcb *inp)
3398 {
3399 	struct socket *socket;
3400 	bool did_upgrade;
3401 
3402 	if (inp == NULL)
3403 		return;
3404 
3405 	socket = inp->inp_socket;
3406 	if (socket == NULL)
3407 		return;
3408 
3409 	if (inp->inp_snd_tag == NULL)
3410 		return;
3411 
3412 	if (!INP_WLOCKED(inp)) {
3413 		/*
3414 		 * NOTE: If the write locking fails, we need to bail
3415 		 * out and use the non-ratelimited ring for the
3416 		 * transmit until there is a new chance to get the
3417 		 * write lock.
3418 		 */
3419 		if (!INP_TRY_UPGRADE(inp))
3420 			return;
3421 		did_upgrade = 1;
3422 	} else {
3423 		did_upgrade = 0;
3424 	}
3425 
3426 	/* detach rate limiting */
3427 	in_pcbdetach_txrtlmt(inp);
3428 
3429 	/* make sure new mbuf send tag allocation is made */
3430 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3431 
3432 	if (did_upgrade)
3433 		INP_DOWNGRADE(inp);
3434 }
3435 #endif /* RATELIMIT */
3436