xref: /freebsd/sys/netinet/in_pcb.c (revision e430d1ed78d021db4e9760d9800393b627156745)
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 #ifdef NUMA
514 	inp->inp_numa_domain = M_NODOM;
515 #endif
516 	inp->inp_pcbinfo = pcbinfo;
517 	inp->inp_socket = so;
518 	inp->inp_cred = crhold(so->so_cred);
519 	inp->inp_inc.inc_fibnum = so->so_fibnum;
520 #ifdef MAC
521 	error = mac_inpcb_init(inp, M_NOWAIT);
522 	if (error != 0)
523 		goto out;
524 	mac_inpcb_create(so, inp);
525 #endif
526 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
527 	error = ipsec_init_pcbpolicy(inp);
528 	if (error != 0) {
529 #ifdef MAC
530 		mac_inpcb_destroy(inp);
531 #endif
532 		goto out;
533 	}
534 #endif /*IPSEC*/
535 #ifdef INET6
536 	if (INP_SOCKAF(so) == AF_INET6) {
537 		inp->inp_vflag |= INP_IPV6PROTO;
538 		if (V_ip6_v6only)
539 			inp->inp_flags |= IN6P_IPV6_V6ONLY;
540 	}
541 #endif
542 	INP_WLOCK(inp);
543 	INP_LIST_WLOCK(pcbinfo);
544 	CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
545 	pcbinfo->ipi_count++;
546 	so->so_pcb = (caddr_t)inp;
547 #ifdef INET6
548 	if (V_ip6_auto_flowlabel)
549 		inp->inp_flags |= IN6P_AUTOFLOWLABEL;
550 #endif
551 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
552 	refcount_init(&inp->inp_refcount, 1);	/* Reference from inpcbinfo */
553 
554 	/*
555 	 * Routes in inpcb's can cache L2 as well; they are guaranteed
556 	 * to be cleaned up.
557 	 */
558 	inp->inp_route.ro_flags = RT_LLE_CACHE;
559 	INP_LIST_WUNLOCK(pcbinfo);
560 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
561 out:
562 	if (error != 0) {
563 		crfree(inp->inp_cred);
564 		uma_zfree(pcbinfo->ipi_zone, inp);
565 	}
566 #endif
567 	return (error);
568 }
569 
570 #ifdef INET
571 int
572 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
573 {
574 	int anonport, error;
575 
576 	INP_WLOCK_ASSERT(inp);
577 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
578 
579 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
580 		return (EINVAL);
581 	anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
582 	error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
583 	    &inp->inp_lport, cred);
584 	if (error)
585 		return (error);
586 	if (in_pcbinshash(inp) != 0) {
587 		inp->inp_laddr.s_addr = INADDR_ANY;
588 		inp->inp_lport = 0;
589 		return (EAGAIN);
590 	}
591 	if (anonport)
592 		inp->inp_flags |= INP_ANONPORT;
593 	return (0);
594 }
595 #endif
596 
597 /*
598  * Select a local port (number) to use.
599  */
600 #if defined(INET) || defined(INET6)
601 int
602 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
603     struct ucred *cred, int lookupflags)
604 {
605 	struct inpcbinfo *pcbinfo;
606 	struct inpcb *tmpinp;
607 	unsigned short *lastport;
608 	int count, dorandom, error;
609 	u_short aux, first, last, lport;
610 #ifdef INET
611 	struct in_addr laddr;
612 #endif
613 
614 	pcbinfo = inp->inp_pcbinfo;
615 
616 	/*
617 	 * Because no actual state changes occur here, a global write lock on
618 	 * the pcbinfo isn't required.
619 	 */
620 	INP_LOCK_ASSERT(inp);
621 	INP_HASH_LOCK_ASSERT(pcbinfo);
622 
623 	if (inp->inp_flags & INP_HIGHPORT) {
624 		first = V_ipport_hifirstauto;	/* sysctl */
625 		last  = V_ipport_hilastauto;
626 		lastport = &pcbinfo->ipi_lasthi;
627 	} else if (inp->inp_flags & INP_LOWPORT) {
628 		error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
629 		if (error)
630 			return (error);
631 		first = V_ipport_lowfirstauto;	/* 1023 */
632 		last  = V_ipport_lowlastauto;	/* 600 */
633 		lastport = &pcbinfo->ipi_lastlow;
634 	} else {
635 		first = V_ipport_firstauto;	/* sysctl */
636 		last  = V_ipport_lastauto;
637 		lastport = &pcbinfo->ipi_lastport;
638 	}
639 	/*
640 	 * For UDP(-Lite), use random port allocation as long as the user
641 	 * allows it.  For TCP (and as of yet unknown) connections,
642 	 * use random port allocation only if the user allows it AND
643 	 * ipport_tick() allows it.
644 	 */
645 	if (V_ipport_randomized &&
646 		(!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
647 		pcbinfo == &V_ulitecbinfo))
648 		dorandom = 1;
649 	else
650 		dorandom = 0;
651 	/*
652 	 * It makes no sense to do random port allocation if
653 	 * we have the only port available.
654 	 */
655 	if (first == last)
656 		dorandom = 0;
657 	/* Make sure to not include UDP(-Lite) packets in the count. */
658 	if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
659 		V_ipport_tcpallocs++;
660 	/*
661 	 * Instead of having two loops further down counting up or down
662 	 * make sure that first is always <= last and go with only one
663 	 * code path implementing all logic.
664 	 */
665 	if (first > last) {
666 		aux = first;
667 		first = last;
668 		last = aux;
669 	}
670 
671 #ifdef INET
672 	/* Make the compiler happy. */
673 	laddr.s_addr = 0;
674 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
675 		KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
676 		    __func__, inp));
677 		laddr = *laddrp;
678 	}
679 #endif
680 	tmpinp = NULL;	/* Make compiler happy. */
681 	lport = *lportp;
682 
683 	if (dorandom)
684 		*lastport = first + (arc4random() % (last - first));
685 
686 	count = last - first;
687 
688 	do {
689 		if (count-- < 0)	/* completely used? */
690 			return (EADDRNOTAVAIL);
691 		++*lastport;
692 		if (*lastport < first || *lastport > last)
693 			*lastport = first;
694 		lport = htons(*lastport);
695 
696 #ifdef INET6
697 		if ((inp->inp_vflag & INP_IPV6) != 0)
698 			tmpinp = in6_pcblookup_local(pcbinfo,
699 			    &inp->in6p_laddr, lport, lookupflags, cred);
700 #endif
701 #if defined(INET) && defined(INET6)
702 		else
703 #endif
704 #ifdef INET
705 			tmpinp = in_pcblookup_local(pcbinfo, laddr,
706 			    lport, lookupflags, cred);
707 #endif
708 	} while (tmpinp != NULL);
709 
710 #ifdef INET
711 	if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
712 		laddrp->s_addr = laddr.s_addr;
713 #endif
714 	*lportp = lport;
715 
716 	return (0);
717 }
718 
719 /*
720  * Return cached socket options.
721  */
722 int
723 inp_so_options(const struct inpcb *inp)
724 {
725 	int so_options;
726 
727 	so_options = 0;
728 
729 	if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
730 		so_options |= SO_REUSEPORT_LB;
731 	if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
732 		so_options |= SO_REUSEPORT;
733 	if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
734 		so_options |= SO_REUSEADDR;
735 	return (so_options);
736 }
737 #endif /* INET || INET6 */
738 
739 /*
740  * Check if a new BINDMULTI socket is allowed to be created.
741  *
742  * ni points to the new inp.
743  * oi points to the exisitng inp.
744  *
745  * This checks whether the existing inp also has BINDMULTI and
746  * whether the credentials match.
747  */
748 int
749 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
750 {
751 	/* Check permissions match */
752 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
753 	    (ni->inp_cred->cr_uid !=
754 	    oi->inp_cred->cr_uid))
755 		return (0);
756 
757 	/* Check the existing inp has BINDMULTI set */
758 	if ((ni->inp_flags2 & INP_BINDMULTI) &&
759 	    ((oi->inp_flags2 & INP_BINDMULTI) == 0))
760 		return (0);
761 
762 	/*
763 	 * We're okay - either INP_BINDMULTI isn't set on ni, or
764 	 * it is and it matches the checks.
765 	 */
766 	return (1);
767 }
768 
769 #ifdef INET
770 /*
771  * Set up a bind operation on a PCB, performing port allocation
772  * as required, but do not actually modify the PCB. Callers can
773  * either complete the bind by setting inp_laddr/inp_lport and
774  * calling in_pcbinshash(), or they can just use the resulting
775  * port and address to authorise the sending of a once-off packet.
776  *
777  * On error, the values of *laddrp and *lportp are not changed.
778  */
779 int
780 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
781     u_short *lportp, struct ucred *cred)
782 {
783 	struct socket *so = inp->inp_socket;
784 	struct sockaddr_in *sin;
785 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
786 	struct in_addr laddr;
787 	u_short lport = 0;
788 	int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
789 	int error;
790 
791 	/*
792 	 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
793 	 * so that we don't have to add to the (already messy) code below.
794 	 */
795 	int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
796 
797 	/*
798 	 * No state changes, so read locks are sufficient here.
799 	 */
800 	INP_LOCK_ASSERT(inp);
801 	INP_HASH_LOCK_ASSERT(pcbinfo);
802 
803 	if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
804 		return (EADDRNOTAVAIL);
805 	laddr.s_addr = *laddrp;
806 	if (nam != NULL && laddr.s_addr != INADDR_ANY)
807 		return (EINVAL);
808 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
809 		lookupflags = INPLOOKUP_WILDCARD;
810 	if (nam == NULL) {
811 		if ((error = prison_local_ip4(cred, &laddr)) != 0)
812 			return (error);
813 	} else {
814 		sin = (struct sockaddr_in *)nam;
815 		if (nam->sa_len != sizeof (*sin))
816 			return (EINVAL);
817 #ifdef notdef
818 		/*
819 		 * We should check the family, but old programs
820 		 * incorrectly fail to initialize it.
821 		 */
822 		if (sin->sin_family != AF_INET)
823 			return (EAFNOSUPPORT);
824 #endif
825 		error = prison_local_ip4(cred, &sin->sin_addr);
826 		if (error)
827 			return (error);
828 		if (sin->sin_port != *lportp) {
829 			/* Don't allow the port to change. */
830 			if (*lportp != 0)
831 				return (EINVAL);
832 			lport = sin->sin_port;
833 		}
834 		/* NB: lport is left as 0 if the port isn't being changed. */
835 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
836 			/*
837 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
838 			 * allow complete duplication of binding if
839 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
840 			 * and a multicast address is bound on both
841 			 * new and duplicated sockets.
842 			 */
843 			if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
844 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
845 			/*
846 			 * XXX: How to deal with SO_REUSEPORT_LB here?
847 			 * Treat same as SO_REUSEPORT for now.
848 			 */
849 			if ((so->so_options &
850 			    (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
851 				reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
852 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
853 			sin->sin_port = 0;		/* yech... */
854 			bzero(&sin->sin_zero, sizeof(sin->sin_zero));
855 			/*
856 			 * Is the address a local IP address?
857 			 * If INP_BINDANY is set, then the socket may be bound
858 			 * to any endpoint address, local or not.
859 			 */
860 			if ((inp->inp_flags & INP_BINDANY) == 0 &&
861 			    ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
862 				return (EADDRNOTAVAIL);
863 		}
864 		laddr = sin->sin_addr;
865 		if (lport) {
866 			struct inpcb *t;
867 			struct tcptw *tw;
868 
869 			/* GROSS */
870 			if (ntohs(lport) <= V_ipport_reservedhigh &&
871 			    ntohs(lport) >= V_ipport_reservedlow &&
872 			    priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
873 				return (EACCES);
874 			if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
875 			    priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
876 				t = in_pcblookup_local(pcbinfo, sin->sin_addr,
877 				    lport, INPLOOKUP_WILDCARD, cred);
878 	/*
879 	 * XXX
880 	 * This entire block sorely needs a rewrite.
881 	 */
882 				if (t &&
883 				    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
884 				    ((t->inp_flags & INP_TIMEWAIT) == 0) &&
885 				    (so->so_type != SOCK_STREAM ||
886 				     ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
887 				    (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
888 				     ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
889 				     (t->inp_flags2 & INP_REUSEPORT) ||
890 				     (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
891 				    (inp->inp_cred->cr_uid !=
892 				     t->inp_cred->cr_uid))
893 					return (EADDRINUSE);
894 
895 				/*
896 				 * If the socket is a BINDMULTI socket, then
897 				 * the credentials need to match and the
898 				 * original socket also has to have been bound
899 				 * with BINDMULTI.
900 				 */
901 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
902 					return (EADDRINUSE);
903 			}
904 			t = in_pcblookup_local(pcbinfo, sin->sin_addr,
905 			    lport, lookupflags, cred);
906 			if (t && (t->inp_flags & INP_TIMEWAIT)) {
907 				/*
908 				 * XXXRW: If an incpb has had its timewait
909 				 * state recycled, we treat the address as
910 				 * being in use (for now).  This is better
911 				 * than a panic, but not desirable.
912 				 */
913 				tw = intotw(t);
914 				if (tw == NULL ||
915 				    ((reuseport & tw->tw_so_options) == 0 &&
916 					(reuseport_lb &
917 				            tw->tw_so_options) == 0)) {
918 					return (EADDRINUSE);
919 				}
920 			} else if (t &&
921 				   ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
922 				   (reuseport & inp_so_options(t)) == 0 &&
923 				   (reuseport_lb & inp_so_options(t)) == 0) {
924 #ifdef INET6
925 				if (ntohl(sin->sin_addr.s_addr) !=
926 				    INADDR_ANY ||
927 				    ntohl(t->inp_laddr.s_addr) !=
928 				    INADDR_ANY ||
929 				    (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
930 				    (t->inp_vflag & INP_IPV6PROTO) == 0)
931 #endif
932 						return (EADDRINUSE);
933 				if (t && (! in_pcbbind_check_bindmulti(inp, t)))
934 					return (EADDRINUSE);
935 			}
936 		}
937 	}
938 	if (*lportp != 0)
939 		lport = *lportp;
940 	if (lport == 0) {
941 		error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
942 		if (error != 0)
943 			return (error);
944 
945 	}
946 	*laddrp = laddr.s_addr;
947 	*lportp = lport;
948 	return (0);
949 }
950 
951 /*
952  * Connect from a socket to a specified address.
953  * Both address and port must be specified in argument sin.
954  * If don't have a local address for this socket yet,
955  * then pick one.
956  */
957 int
958 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
959     struct ucred *cred, struct mbuf *m)
960 {
961 	u_short lport, fport;
962 	in_addr_t laddr, faddr;
963 	int anonport, error;
964 
965 	INP_WLOCK_ASSERT(inp);
966 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
967 
968 	lport = inp->inp_lport;
969 	laddr = inp->inp_laddr.s_addr;
970 	anonport = (lport == 0);
971 	error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
972 	    NULL, cred);
973 	if (error)
974 		return (error);
975 
976 	/* Do the initial binding of the local address if required. */
977 	if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
978 		inp->inp_lport = lport;
979 		inp->inp_laddr.s_addr = laddr;
980 		if (in_pcbinshash(inp) != 0) {
981 			inp->inp_laddr.s_addr = INADDR_ANY;
982 			inp->inp_lport = 0;
983 			return (EAGAIN);
984 		}
985 	}
986 
987 	/* Commit the remaining changes. */
988 	inp->inp_lport = lport;
989 	inp->inp_laddr.s_addr = laddr;
990 	inp->inp_faddr.s_addr = faddr;
991 	inp->inp_fport = fport;
992 	in_pcbrehash_mbuf(inp, m);
993 
994 	if (anonport)
995 		inp->inp_flags |= INP_ANONPORT;
996 	return (0);
997 }
998 
999 int
1000 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1001 {
1002 
1003 	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1004 }
1005 
1006 /*
1007  * Do proper source address selection on an unbound socket in case
1008  * of connect. Take jails into account as well.
1009  */
1010 int
1011 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1012     struct ucred *cred)
1013 {
1014 	struct ifaddr *ifa;
1015 	struct sockaddr *sa;
1016 	struct sockaddr_in *sin;
1017 	struct route sro;
1018 	struct epoch_tracker et;
1019 	int error;
1020 
1021 	KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1022 	/*
1023 	 * Bypass source address selection and use the primary jail IP
1024 	 * if requested.
1025 	 */
1026 	if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1027 		return (0);
1028 
1029 	error = 0;
1030 	bzero(&sro, sizeof(sro));
1031 
1032 	sin = (struct sockaddr_in *)&sro.ro_dst;
1033 	sin->sin_family = AF_INET;
1034 	sin->sin_len = sizeof(struct sockaddr_in);
1035 	sin->sin_addr.s_addr = faddr->s_addr;
1036 
1037 	/*
1038 	 * If route is known our src addr is taken from the i/f,
1039 	 * else punt.
1040 	 *
1041 	 * Find out route to destination.
1042 	 */
1043 	if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1044 		in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1045 
1046 	/*
1047 	 * If we found a route, use the address corresponding to
1048 	 * the outgoing interface.
1049 	 *
1050 	 * Otherwise assume faddr is reachable on a directly connected
1051 	 * network and try to find a corresponding interface to take
1052 	 * the source address from.
1053 	 */
1054 	NET_EPOCH_ENTER(et);
1055 	if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1056 		struct in_ifaddr *ia;
1057 		struct ifnet *ifp;
1058 
1059 		ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1060 					inp->inp_socket->so_fibnum));
1061 		if (ia == NULL) {
1062 			ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1063 						inp->inp_socket->so_fibnum));
1064 
1065 		}
1066 		if (ia == NULL) {
1067 			error = ENETUNREACH;
1068 			goto done;
1069 		}
1070 
1071 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1072 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1073 			goto done;
1074 		}
1075 
1076 		ifp = ia->ia_ifp;
1077 		ia = NULL;
1078 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1079 
1080 			sa = ifa->ifa_addr;
1081 			if (sa->sa_family != AF_INET)
1082 				continue;
1083 			sin = (struct sockaddr_in *)sa;
1084 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1085 				ia = (struct in_ifaddr *)ifa;
1086 				break;
1087 			}
1088 		}
1089 		if (ia != NULL) {
1090 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1091 			goto done;
1092 		}
1093 
1094 		/* 3. As a last resort return the 'default' jail address. */
1095 		error = prison_get_ip4(cred, laddr);
1096 		goto done;
1097 	}
1098 
1099 	/*
1100 	 * If the outgoing interface on the route found is not
1101 	 * a loopback interface, use the address from that interface.
1102 	 * In case of jails do those three steps:
1103 	 * 1. check if the interface address belongs to the jail. If so use it.
1104 	 * 2. check if we have any address on the outgoing interface
1105 	 *    belonging to this jail. If so use it.
1106 	 * 3. as a last resort return the 'default' jail address.
1107 	 */
1108 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1109 		struct in_ifaddr *ia;
1110 		struct ifnet *ifp;
1111 
1112 		/* If not jailed, use the default returned. */
1113 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1114 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1115 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1116 			goto done;
1117 		}
1118 
1119 		/* Jailed. */
1120 		/* 1. Check if the iface address belongs to the jail. */
1121 		sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1122 		if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1123 			ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1124 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1125 			goto done;
1126 		}
1127 
1128 		/*
1129 		 * 2. Check if we have any address on the outgoing interface
1130 		 *    belonging to this jail.
1131 		 */
1132 		ia = NULL;
1133 		ifp = sro.ro_rt->rt_ifp;
1134 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1135 			sa = ifa->ifa_addr;
1136 			if (sa->sa_family != AF_INET)
1137 				continue;
1138 			sin = (struct sockaddr_in *)sa;
1139 			if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1140 				ia = (struct in_ifaddr *)ifa;
1141 				break;
1142 			}
1143 		}
1144 		if (ia != NULL) {
1145 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1146 			goto done;
1147 		}
1148 
1149 		/* 3. As a last resort return the 'default' jail address. */
1150 		error = prison_get_ip4(cred, laddr);
1151 		goto done;
1152 	}
1153 
1154 	/*
1155 	 * The outgoing interface is marked with 'loopback net', so a route
1156 	 * to ourselves is here.
1157 	 * Try to find the interface of the destination address and then
1158 	 * take the address from there. That interface is not necessarily
1159 	 * a loopback interface.
1160 	 * In case of jails, check that it is an address of the jail
1161 	 * and if we cannot find, fall back to the 'default' jail address.
1162 	 */
1163 	if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1164 		struct sockaddr_in sain;
1165 		struct in_ifaddr *ia;
1166 
1167 		bzero(&sain, sizeof(struct sockaddr_in));
1168 		sain.sin_family = AF_INET;
1169 		sain.sin_len = sizeof(struct sockaddr_in);
1170 		sain.sin_addr.s_addr = faddr->s_addr;
1171 
1172 		ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1173 					inp->inp_socket->so_fibnum));
1174 		if (ia == NULL)
1175 			ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1176 						inp->inp_socket->so_fibnum));
1177 		if (ia == NULL)
1178 			ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1179 
1180 		if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1181 			if (ia == NULL) {
1182 				error = ENETUNREACH;
1183 				goto done;
1184 			}
1185 			laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1186 			goto done;
1187 		}
1188 
1189 		/* Jailed. */
1190 		if (ia != NULL) {
1191 			struct ifnet *ifp;
1192 
1193 			ifp = ia->ia_ifp;
1194 			ia = NULL;
1195 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1196 				sa = ifa->ifa_addr;
1197 				if (sa->sa_family != AF_INET)
1198 					continue;
1199 				sin = (struct sockaddr_in *)sa;
1200 				if (prison_check_ip4(cred,
1201 				    &sin->sin_addr) == 0) {
1202 					ia = (struct in_ifaddr *)ifa;
1203 					break;
1204 				}
1205 			}
1206 			if (ia != NULL) {
1207 				laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1208 				goto done;
1209 			}
1210 		}
1211 
1212 		/* 3. As a last resort return the 'default' jail address. */
1213 		error = prison_get_ip4(cred, laddr);
1214 		goto done;
1215 	}
1216 
1217 done:
1218 	NET_EPOCH_EXIT(et);
1219 	if (sro.ro_rt != NULL)
1220 		RTFREE(sro.ro_rt);
1221 	return (error);
1222 }
1223 
1224 /*
1225  * Set up for a connect from a socket to the specified address.
1226  * On entry, *laddrp and *lportp should contain the current local
1227  * address and port for the PCB; these are updated to the values
1228  * that should be placed in inp_laddr and inp_lport to complete
1229  * the connect.
1230  *
1231  * On success, *faddrp and *fportp will be set to the remote address
1232  * and port. These are not updated in the error case.
1233  *
1234  * If the operation fails because the connection already exists,
1235  * *oinpp will be set to the PCB of that connection so that the
1236  * caller can decide to override it. In all other cases, *oinpp
1237  * is set to NULL.
1238  */
1239 int
1240 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1241     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1242     struct inpcb **oinpp, struct ucred *cred)
1243 {
1244 	struct rm_priotracker in_ifa_tracker;
1245 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1246 	struct in_ifaddr *ia;
1247 	struct inpcb *oinp;
1248 	struct in_addr laddr, faddr;
1249 	u_short lport, fport;
1250 	int error;
1251 
1252 	/*
1253 	 * Because a global state change doesn't actually occur here, a read
1254 	 * lock is sufficient.
1255 	 */
1256 	INP_LOCK_ASSERT(inp);
1257 	INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1258 
1259 	if (oinpp != NULL)
1260 		*oinpp = NULL;
1261 	if (nam->sa_len != sizeof (*sin))
1262 		return (EINVAL);
1263 	if (sin->sin_family != AF_INET)
1264 		return (EAFNOSUPPORT);
1265 	if (sin->sin_port == 0)
1266 		return (EADDRNOTAVAIL);
1267 	laddr.s_addr = *laddrp;
1268 	lport = *lportp;
1269 	faddr = sin->sin_addr;
1270 	fport = sin->sin_port;
1271 
1272 	if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1273 		/*
1274 		 * If the destination address is INADDR_ANY,
1275 		 * use the primary local address.
1276 		 * If the supplied address is INADDR_BROADCAST,
1277 		 * and the primary interface supports broadcast,
1278 		 * choose the broadcast address for that interface.
1279 		 */
1280 		if (faddr.s_addr == INADDR_ANY) {
1281 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1282 			faddr =
1283 			    IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1284 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1285 			if (cred != NULL &&
1286 			    (error = prison_get_ip4(cred, &faddr)) != 0)
1287 				return (error);
1288 		} else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1289 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1290 			if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1291 			    IFF_BROADCAST)
1292 				faddr = satosin(&CK_STAILQ_FIRST(
1293 				    &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1294 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1295 		}
1296 	}
1297 	if (laddr.s_addr == INADDR_ANY) {
1298 		error = in_pcbladdr(inp, &faddr, &laddr, cred);
1299 		/*
1300 		 * If the destination address is multicast and an outgoing
1301 		 * interface has been set as a multicast option, prefer the
1302 		 * address of that interface as our source address.
1303 		 */
1304 		if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1305 		    inp->inp_moptions != NULL) {
1306 			struct ip_moptions *imo;
1307 			struct ifnet *ifp;
1308 
1309 			imo = inp->inp_moptions;
1310 			if (imo->imo_multicast_ifp != NULL) {
1311 				ifp = imo->imo_multicast_ifp;
1312 				IN_IFADDR_RLOCK(&in_ifa_tracker);
1313 				CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1314 					if ((ia->ia_ifp == ifp) &&
1315 					    (cred == NULL ||
1316 					    prison_check_ip4(cred,
1317 					    &ia->ia_addr.sin_addr) == 0))
1318 						break;
1319 				}
1320 				if (ia == NULL)
1321 					error = EADDRNOTAVAIL;
1322 				else {
1323 					laddr = ia->ia_addr.sin_addr;
1324 					error = 0;
1325 				}
1326 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1327 			}
1328 		}
1329 		if (error)
1330 			return (error);
1331 	}
1332 	oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1333 	    laddr, lport, 0, NULL);
1334 	if (oinp != NULL) {
1335 		if (oinpp != NULL)
1336 			*oinpp = oinp;
1337 		return (EADDRINUSE);
1338 	}
1339 	if (lport == 0) {
1340 		error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1341 		    cred);
1342 		if (error)
1343 			return (error);
1344 	}
1345 	*laddrp = laddr.s_addr;
1346 	*lportp = lport;
1347 	*faddrp = faddr.s_addr;
1348 	*fportp = fport;
1349 	return (0);
1350 }
1351 
1352 void
1353 in_pcbdisconnect(struct inpcb *inp)
1354 {
1355 
1356 	INP_WLOCK_ASSERT(inp);
1357 	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1358 
1359 	inp->inp_faddr.s_addr = INADDR_ANY;
1360 	inp->inp_fport = 0;
1361 	in_pcbrehash(inp);
1362 }
1363 #endif /* INET */
1364 
1365 /*
1366  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1367  * For most protocols, this will be invoked immediately prior to calling
1368  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1369  * socket, in which case in_pcbfree() is deferred.
1370  */
1371 void
1372 in_pcbdetach(struct inpcb *inp)
1373 {
1374 
1375 	KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1376 
1377 #ifdef RATELIMIT
1378 	if (inp->inp_snd_tag != NULL)
1379 		in_pcbdetach_txrtlmt(inp);
1380 #endif
1381 	inp->inp_socket->so_pcb = NULL;
1382 	inp->inp_socket = NULL;
1383 }
1384 
1385 /*
1386  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1387  * stability of an inpcb pointer despite the inpcb lock being released.  This
1388  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1389  * but where the inpcb lock may already held, or when acquiring a reference
1390  * via a pcbgroup.
1391  *
1392  * in_pcbref() should be used only to provide brief memory stability, and
1393  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1394  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1395  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1396  * lock and rele are the *only* safe operations that may be performed on the
1397  * inpcb.
1398  *
1399  * While the inpcb will not be freed, releasing the inpcb lock means that the
1400  * connection's state may change, so the caller should be careful to
1401  * revalidate any cached state on reacquiring the lock.  Drop the reference
1402  * using in_pcbrele().
1403  */
1404 void
1405 in_pcbref(struct inpcb *inp)
1406 {
1407 
1408 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1409 
1410 	refcount_acquire(&inp->inp_refcount);
1411 }
1412 
1413 /*
1414  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1415  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1416  * return a flag indicating whether or not the inpcb remains valid.  If it is
1417  * valid, we return with the inpcb lock held.
1418  *
1419  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1420  * reference on an inpcb.  Historically more work was done here (actually, in
1421  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1422  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1423  * about memory stability (and continued use of the write lock).
1424  */
1425 int
1426 in_pcbrele_rlocked(struct inpcb *inp)
1427 {
1428 	struct inpcbinfo *pcbinfo;
1429 
1430 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1431 
1432 	INP_RLOCK_ASSERT(inp);
1433 
1434 	if (refcount_release(&inp->inp_refcount) == 0) {
1435 		/*
1436 		 * If the inpcb has been freed, let the caller know, even if
1437 		 * this isn't the last reference.
1438 		 */
1439 		if (inp->inp_flags2 & INP_FREED) {
1440 			INP_RUNLOCK(inp);
1441 			return (1);
1442 		}
1443 		return (0);
1444 	}
1445 
1446 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1447 #ifdef TCPHPTS
1448 	if (inp->inp_in_hpts || inp->inp_in_input) {
1449 		struct tcp_hpts_entry *hpts;
1450 		/*
1451 		 * We should not be on the hpts at
1452 		 * this point in any form. we must
1453 		 * get the lock to be sure.
1454 		 */
1455 		hpts = tcp_hpts_lock(inp);
1456 		if (inp->inp_in_hpts)
1457 			panic("Hpts:%p inp:%p at free still on hpts",
1458 			      hpts, inp);
1459 		mtx_unlock(&hpts->p_mtx);
1460 		hpts = tcp_input_lock(inp);
1461 		if (inp->inp_in_input)
1462 			panic("Hpts:%p inp:%p at free still on input hpts",
1463 			      hpts, inp);
1464 		mtx_unlock(&hpts->p_mtx);
1465 	}
1466 #endif
1467 	INP_RUNLOCK(inp);
1468 	pcbinfo = inp->inp_pcbinfo;
1469 	uma_zfree(pcbinfo->ipi_zone, inp);
1470 	return (1);
1471 }
1472 
1473 int
1474 in_pcbrele_wlocked(struct inpcb *inp)
1475 {
1476 	struct inpcbinfo *pcbinfo;
1477 
1478 	KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1479 
1480 	INP_WLOCK_ASSERT(inp);
1481 
1482 	if (refcount_release(&inp->inp_refcount) == 0) {
1483 		/*
1484 		 * If the inpcb has been freed, let the caller know, even if
1485 		 * this isn't the last reference.
1486 		 */
1487 		if (inp->inp_flags2 & INP_FREED) {
1488 			INP_WUNLOCK(inp);
1489 			return (1);
1490 		}
1491 		return (0);
1492 	}
1493 
1494 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1495 #ifdef TCPHPTS
1496 	if (inp->inp_in_hpts || inp->inp_in_input) {
1497 		struct tcp_hpts_entry *hpts;
1498 		/*
1499 		 * We should not be on the hpts at
1500 		 * this point in any form. we must
1501 		 * get the lock to be sure.
1502 		 */
1503 		hpts = tcp_hpts_lock(inp);
1504 		if (inp->inp_in_hpts)
1505 			panic("Hpts:%p inp:%p at free still on hpts",
1506 			      hpts, inp);
1507 		mtx_unlock(&hpts->p_mtx);
1508 		hpts = tcp_input_lock(inp);
1509 		if (inp->inp_in_input)
1510 			panic("Hpts:%p inp:%p at free still on input hpts",
1511 			      hpts, inp);
1512 		mtx_unlock(&hpts->p_mtx);
1513 	}
1514 #endif
1515 	INP_WUNLOCK(inp);
1516 	pcbinfo = inp->inp_pcbinfo;
1517 	uma_zfree(pcbinfo->ipi_zone, inp);
1518 	return (1);
1519 }
1520 
1521 /*
1522  * Temporary wrapper.
1523  */
1524 int
1525 in_pcbrele(struct inpcb *inp)
1526 {
1527 
1528 	return (in_pcbrele_wlocked(inp));
1529 }
1530 
1531 void
1532 in_pcblist_rele_rlocked(epoch_context_t ctx)
1533 {
1534 	struct in_pcblist *il;
1535 	struct inpcb *inp;
1536 	struct inpcbinfo *pcbinfo;
1537 	int i, n;
1538 
1539 	il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1540 	pcbinfo = il->il_pcbinfo;
1541 	n = il->il_count;
1542 	INP_INFO_WLOCK(pcbinfo);
1543 	for (i = 0; i < n; i++) {
1544 		inp = il->il_inp_list[i];
1545 		INP_RLOCK(inp);
1546 		if (!in_pcbrele_rlocked(inp))
1547 			INP_RUNLOCK(inp);
1548 	}
1549 	INP_INFO_WUNLOCK(pcbinfo);
1550 	free(il, M_TEMP);
1551 }
1552 
1553 static void
1554 inpcbport_free(epoch_context_t ctx)
1555 {
1556 	struct inpcbport *phd;
1557 
1558 	phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1559 	free(phd, M_PCB);
1560 }
1561 
1562 static void
1563 in_pcbfree_deferred(epoch_context_t ctx)
1564 {
1565 	struct inpcb *inp;
1566 	int released __unused;
1567 
1568 	inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1569 
1570 	INP_WLOCK(inp);
1571 	CURVNET_SET(inp->inp_vnet);
1572 #ifdef INET
1573 	struct ip_moptions *imo = inp->inp_moptions;
1574 	inp->inp_moptions = NULL;
1575 #endif
1576 	/* XXXRW: Do as much as possible here. */
1577 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1578 	if (inp->inp_sp != NULL)
1579 		ipsec_delete_pcbpolicy(inp);
1580 #endif
1581 #ifdef INET6
1582 	struct ip6_moptions *im6o = NULL;
1583 	if (inp->inp_vflag & INP_IPV6PROTO) {
1584 		ip6_freepcbopts(inp->in6p_outputopts);
1585 		im6o = inp->in6p_moptions;
1586 		inp->in6p_moptions = NULL;
1587 	}
1588 #endif
1589 	if (inp->inp_options)
1590 		(void)m_free(inp->inp_options);
1591 	inp->inp_vflag = 0;
1592 	crfree(inp->inp_cred);
1593 #ifdef MAC
1594 	mac_inpcb_destroy(inp);
1595 #endif
1596 	released = in_pcbrele_wlocked(inp);
1597 	MPASS(released);
1598 #ifdef INET6
1599 	ip6_freemoptions(im6o);
1600 #endif
1601 #ifdef INET
1602 	inp_freemoptions(imo);
1603 #endif
1604 	CURVNET_RESTORE();
1605 }
1606 
1607 /*
1608  * Unconditionally schedule an inpcb to be freed by decrementing its
1609  * reference count, which should occur only after the inpcb has been detached
1610  * from its socket.  If another thread holds a temporary reference (acquired
1611  * using in_pcbref()) then the free is deferred until that reference is
1612  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1613  * work, including removal from global lists, is done in this context, where
1614  * the pcbinfo lock is held.
1615  */
1616 void
1617 in_pcbfree(struct inpcb *inp)
1618 {
1619 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1620 
1621 	KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1622 	KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1623 	    ("%s: called twice for pcb %p", __func__, inp));
1624 	if (inp->inp_flags2 & INP_FREED) {
1625 		INP_WUNLOCK(inp);
1626 		return;
1627 	}
1628 
1629 #ifdef INVARIANTS
1630 	if (pcbinfo == &V_tcbinfo) {
1631 		INP_INFO_LOCK_ASSERT(pcbinfo);
1632 	} else {
1633 		INP_INFO_WLOCK_ASSERT(pcbinfo);
1634 	}
1635 #endif
1636 	INP_WLOCK_ASSERT(inp);
1637 	INP_LIST_WLOCK(pcbinfo);
1638 	in_pcbremlists(inp);
1639 	INP_LIST_WUNLOCK(pcbinfo);
1640 	RO_INVALIDATE_CACHE(&inp->inp_route);
1641 	/* mark as destruction in progress */
1642 	inp->inp_flags2 |= INP_FREED;
1643 	INP_WUNLOCK(inp);
1644 	epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred);
1645 }
1646 
1647 /*
1648  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1649  * port reservation, and preventing it from being returned by inpcb lookups.
1650  *
1651  * It is used by TCP to mark an inpcb as unused and avoid future packet
1652  * delivery or event notification when a socket remains open but TCP has
1653  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1654  * or a RST on the wire, and allows the port binding to be reused while still
1655  * maintaining the invariant that so_pcb always points to a valid inpcb until
1656  * in_pcbdetach().
1657  *
1658  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1659  * in_pcbnotifyall() and in_pcbpurgeif0()?
1660  */
1661 void
1662 in_pcbdrop(struct inpcb *inp)
1663 {
1664 
1665 	INP_WLOCK_ASSERT(inp);
1666 #ifdef INVARIANTS
1667 	if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1668 		MPASS(inp->inp_refcount > 1);
1669 #endif
1670 
1671 	/*
1672 	 * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1673 	 * the hash lock...?
1674 	 */
1675 	inp->inp_flags |= INP_DROPPED;
1676 	if (inp->inp_flags & INP_INHASHLIST) {
1677 		struct inpcbport *phd = inp->inp_phd;
1678 
1679 		INP_HASH_WLOCK(inp->inp_pcbinfo);
1680 		in_pcbremlbgrouphash(inp);
1681 		CK_LIST_REMOVE(inp, inp_hash);
1682 		CK_LIST_REMOVE(inp, inp_portlist);
1683 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1684 			CK_LIST_REMOVE(phd, phd_hash);
1685 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
1686 		}
1687 		INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1688 		inp->inp_flags &= ~INP_INHASHLIST;
1689 #ifdef PCBGROUP
1690 		in_pcbgroup_remove(inp);
1691 #endif
1692 	}
1693 }
1694 
1695 #ifdef INET
1696 /*
1697  * Common routines to return the socket addresses associated with inpcbs.
1698  */
1699 struct sockaddr *
1700 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1701 {
1702 	struct sockaddr_in *sin;
1703 
1704 	sin = malloc(sizeof *sin, M_SONAME,
1705 		M_WAITOK | M_ZERO);
1706 	sin->sin_family = AF_INET;
1707 	sin->sin_len = sizeof(*sin);
1708 	sin->sin_addr = *addr_p;
1709 	sin->sin_port = port;
1710 
1711 	return (struct sockaddr *)sin;
1712 }
1713 
1714 int
1715 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1716 {
1717 	struct inpcb *inp;
1718 	struct in_addr addr;
1719 	in_port_t port;
1720 
1721 	inp = sotoinpcb(so);
1722 	KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1723 
1724 	INP_RLOCK(inp);
1725 	port = inp->inp_lport;
1726 	addr = inp->inp_laddr;
1727 	INP_RUNLOCK(inp);
1728 
1729 	*nam = in_sockaddr(port, &addr);
1730 	return 0;
1731 }
1732 
1733 int
1734 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1735 {
1736 	struct inpcb *inp;
1737 	struct in_addr addr;
1738 	in_port_t port;
1739 
1740 	inp = sotoinpcb(so);
1741 	KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1742 
1743 	INP_RLOCK(inp);
1744 	port = inp->inp_fport;
1745 	addr = inp->inp_faddr;
1746 	INP_RUNLOCK(inp);
1747 
1748 	*nam = in_sockaddr(port, &addr);
1749 	return 0;
1750 }
1751 
1752 void
1753 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1754     struct inpcb *(*notify)(struct inpcb *, int))
1755 {
1756 	struct inpcb *inp, *inp_temp;
1757 
1758 	INP_INFO_WLOCK(pcbinfo);
1759 	CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1760 		INP_WLOCK(inp);
1761 #ifdef INET6
1762 		if ((inp->inp_vflag & INP_IPV4) == 0) {
1763 			INP_WUNLOCK(inp);
1764 			continue;
1765 		}
1766 #endif
1767 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
1768 		    inp->inp_socket == NULL) {
1769 			INP_WUNLOCK(inp);
1770 			continue;
1771 		}
1772 		if ((*notify)(inp, errno))
1773 			INP_WUNLOCK(inp);
1774 	}
1775 	INP_INFO_WUNLOCK(pcbinfo);
1776 }
1777 
1778 void
1779 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1780 {
1781 	struct inpcb *inp;
1782 	struct ip_moptions *imo;
1783 	int i, gap;
1784 
1785 	INP_INFO_WLOCK(pcbinfo);
1786 	CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1787 		INP_WLOCK(inp);
1788 		imo = inp->inp_moptions;
1789 		if ((inp->inp_vflag & INP_IPV4) &&
1790 		    imo != NULL) {
1791 			/*
1792 			 * Unselect the outgoing interface if it is being
1793 			 * detached.
1794 			 */
1795 			if (imo->imo_multicast_ifp == ifp)
1796 				imo->imo_multicast_ifp = NULL;
1797 
1798 			/*
1799 			 * Drop multicast group membership if we joined
1800 			 * through the interface being detached.
1801 			 *
1802 			 * XXX This can all be deferred to an epoch_call
1803 			 */
1804 			for (i = 0, gap = 0; i < imo->imo_num_memberships;
1805 			    i++) {
1806 				if (imo->imo_membership[i]->inm_ifp == ifp) {
1807 					IN_MULTI_LOCK_ASSERT();
1808 					in_leavegroup_locked(imo->imo_membership[i], NULL);
1809 					gap++;
1810 				} else if (gap != 0)
1811 					imo->imo_membership[i - gap] =
1812 					    imo->imo_membership[i];
1813 			}
1814 			imo->imo_num_memberships -= gap;
1815 		}
1816 		INP_WUNLOCK(inp);
1817 	}
1818 	INP_INFO_WUNLOCK(pcbinfo);
1819 }
1820 
1821 /*
1822  * Lookup a PCB based on the local address and port.  Caller must hold the
1823  * hash lock.  No inpcb locks or references are acquired.
1824  */
1825 #define INP_LOOKUP_MAPPED_PCB_COST	3
1826 struct inpcb *
1827 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1828     u_short lport, int lookupflags, struct ucred *cred)
1829 {
1830 	struct inpcb *inp;
1831 #ifdef INET6
1832 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1833 #else
1834 	int matchwild = 3;
1835 #endif
1836 	int wildcard;
1837 
1838 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1839 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1840 
1841 	INP_HASH_LOCK_ASSERT(pcbinfo);
1842 
1843 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1844 		struct inpcbhead *head;
1845 		/*
1846 		 * Look for an unconnected (wildcard foreign addr) PCB that
1847 		 * matches the local address and port we're looking for.
1848 		 */
1849 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1850 		    0, pcbinfo->ipi_hashmask)];
1851 		CK_LIST_FOREACH(inp, head, inp_hash) {
1852 #ifdef INET6
1853 			/* XXX inp locking */
1854 			if ((inp->inp_vflag & INP_IPV4) == 0)
1855 				continue;
1856 #endif
1857 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1858 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1859 			    inp->inp_lport == lport) {
1860 				/*
1861 				 * Found?
1862 				 */
1863 				if (cred == NULL ||
1864 				    prison_equal_ip4(cred->cr_prison,
1865 					inp->inp_cred->cr_prison))
1866 					return (inp);
1867 			}
1868 		}
1869 		/*
1870 		 * Not found.
1871 		 */
1872 		return (NULL);
1873 	} else {
1874 		struct inpcbporthead *porthash;
1875 		struct inpcbport *phd;
1876 		struct inpcb *match = NULL;
1877 		/*
1878 		 * Best fit PCB lookup.
1879 		 *
1880 		 * First see if this local port is in use by looking on the
1881 		 * port hash list.
1882 		 */
1883 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1884 		    pcbinfo->ipi_porthashmask)];
1885 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
1886 			if (phd->phd_port == lport)
1887 				break;
1888 		}
1889 		if (phd != NULL) {
1890 			/*
1891 			 * Port is in use by one or more PCBs. Look for best
1892 			 * fit.
1893 			 */
1894 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1895 				wildcard = 0;
1896 				if (cred != NULL &&
1897 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1898 					cred->cr_prison))
1899 					continue;
1900 #ifdef INET6
1901 				/* XXX inp locking */
1902 				if ((inp->inp_vflag & INP_IPV4) == 0)
1903 					continue;
1904 				/*
1905 				 * We never select the PCB that has
1906 				 * INP_IPV6 flag and is bound to :: if
1907 				 * we have another PCB which is bound
1908 				 * to 0.0.0.0.  If a PCB has the
1909 				 * INP_IPV6 flag, then we set its cost
1910 				 * higher than IPv4 only PCBs.
1911 				 *
1912 				 * Note that the case only happens
1913 				 * when a socket is bound to ::, under
1914 				 * the condition that the use of the
1915 				 * mapped address is allowed.
1916 				 */
1917 				if ((inp->inp_vflag & INP_IPV6) != 0)
1918 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1919 #endif
1920 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1921 					wildcard++;
1922 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1923 					if (laddr.s_addr == INADDR_ANY)
1924 						wildcard++;
1925 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1926 						continue;
1927 				} else {
1928 					if (laddr.s_addr != INADDR_ANY)
1929 						wildcard++;
1930 				}
1931 				if (wildcard < matchwild) {
1932 					match = inp;
1933 					matchwild = wildcard;
1934 					if (matchwild == 0)
1935 						break;
1936 				}
1937 			}
1938 		}
1939 		return (match);
1940 	}
1941 }
1942 #undef INP_LOOKUP_MAPPED_PCB_COST
1943 
1944 static struct inpcb *
1945 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1946     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1947     uint16_t fport, int lookupflags)
1948 {
1949 	struct inpcb *local_wild;
1950 	const struct inpcblbgrouphead *hdr;
1951 	struct inpcblbgroup *grp;
1952 	uint32_t idx;
1953 
1954 	INP_HASH_LOCK_ASSERT(pcbinfo);
1955 
1956 	hdr = &pcbinfo->ipi_lbgrouphashbase[
1957 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1958 
1959 	/*
1960 	 * Order of socket selection:
1961 	 * 1. non-wild.
1962 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1963 	 *
1964 	 * NOTE:
1965 	 * - Load balanced group does not contain jailed sockets
1966 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1967 	 */
1968 	local_wild = NULL;
1969 	CK_LIST_FOREACH(grp, hdr, il_list) {
1970 #ifdef INET6
1971 		if (!(grp->il_vflag & INP_IPV4))
1972 			continue;
1973 #endif
1974 		if (grp->il_lport != lport)
1975 			continue;
1976 
1977 		idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1978 		    grp->il_inpcnt;
1979 		if (grp->il_laddr.s_addr == laddr->s_addr)
1980 			return (grp->il_inp[idx]);
1981 		if (grp->il_laddr.s_addr == INADDR_ANY &&
1982 		    (lookupflags & INPLOOKUP_WILDCARD) != 0)
1983 			local_wild = grp->il_inp[idx];
1984 	}
1985 	return (local_wild);
1986 }
1987 
1988 #ifdef PCBGROUP
1989 /*
1990  * Lookup PCB in hash list, using pcbgroup tables.
1991  */
1992 static struct inpcb *
1993 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1994     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1995     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1996 {
1997 	struct inpcbhead *head;
1998 	struct inpcb *inp, *tmpinp;
1999 	u_short fport = fport_arg, lport = lport_arg;
2000 	bool locked;
2001 
2002 	/*
2003 	 * First look for an exact match.
2004 	 */
2005 	tmpinp = NULL;
2006 	INP_GROUP_LOCK(pcbgroup);
2007 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2008 	    pcbgroup->ipg_hashmask)];
2009 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2010 #ifdef INET6
2011 		/* XXX inp locking */
2012 		if ((inp->inp_vflag & INP_IPV4) == 0)
2013 			continue;
2014 #endif
2015 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2016 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2017 		    inp->inp_fport == fport &&
2018 		    inp->inp_lport == lport) {
2019 			/*
2020 			 * XXX We should be able to directly return
2021 			 * the inp here, without any checks.
2022 			 * Well unless both bound with SO_REUSEPORT?
2023 			 */
2024 			if (prison_flag(inp->inp_cred, PR_IP4))
2025 				goto found;
2026 			if (tmpinp == NULL)
2027 				tmpinp = inp;
2028 		}
2029 	}
2030 	if (tmpinp != NULL) {
2031 		inp = tmpinp;
2032 		goto found;
2033 	}
2034 
2035 #ifdef	RSS
2036 	/*
2037 	 * For incoming connections, we may wish to do a wildcard
2038 	 * match for an RSS-local socket.
2039 	 */
2040 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2041 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2042 #ifdef INET6
2043 		struct inpcb *local_wild_mapped = NULL;
2044 #endif
2045 		struct inpcb *jail_wild = NULL;
2046 		struct inpcbhead *head;
2047 		int injail;
2048 
2049 		/*
2050 		 * Order of socket selection - we always prefer jails.
2051 		 *      1. jailed, non-wild.
2052 		 *      2. jailed, wild.
2053 		 *      3. non-jailed, non-wild.
2054 		 *      4. non-jailed, wild.
2055 		 */
2056 
2057 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2058 		    lport, 0, pcbgroup->ipg_hashmask)];
2059 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2060 #ifdef INET6
2061 			/* XXX inp locking */
2062 			if ((inp->inp_vflag & INP_IPV4) == 0)
2063 				continue;
2064 #endif
2065 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2066 			    inp->inp_lport != lport)
2067 				continue;
2068 
2069 			injail = prison_flag(inp->inp_cred, PR_IP4);
2070 			if (injail) {
2071 				if (prison_check_ip4(inp->inp_cred,
2072 				    &laddr) != 0)
2073 					continue;
2074 			} else {
2075 				if (local_exact != NULL)
2076 					continue;
2077 			}
2078 
2079 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2080 				if (injail)
2081 					goto found;
2082 				else
2083 					local_exact = inp;
2084 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2085 #ifdef INET6
2086 				/* XXX inp locking, NULL check */
2087 				if (inp->inp_vflag & INP_IPV6PROTO)
2088 					local_wild_mapped = inp;
2089 				else
2090 #endif
2091 					if (injail)
2092 						jail_wild = inp;
2093 					else
2094 						local_wild = inp;
2095 			}
2096 		} /* LIST_FOREACH */
2097 
2098 		inp = jail_wild;
2099 		if (inp == NULL)
2100 			inp = local_exact;
2101 		if (inp == NULL)
2102 			inp = local_wild;
2103 #ifdef INET6
2104 		if (inp == NULL)
2105 			inp = local_wild_mapped;
2106 #endif
2107 		if (inp != NULL)
2108 			goto found;
2109 	}
2110 #endif
2111 
2112 	/*
2113 	 * Then look for a wildcard match, if requested.
2114 	 */
2115 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2116 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2117 #ifdef INET6
2118 		struct inpcb *local_wild_mapped = NULL;
2119 #endif
2120 		struct inpcb *jail_wild = NULL;
2121 		struct inpcbhead *head;
2122 		int injail;
2123 
2124 		/*
2125 		 * Order of socket selection - we always prefer jails.
2126 		 *      1. jailed, non-wild.
2127 		 *      2. jailed, wild.
2128 		 *      3. non-jailed, non-wild.
2129 		 *      4. non-jailed, wild.
2130 		 */
2131 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2132 		    0, pcbinfo->ipi_wildmask)];
2133 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2134 #ifdef INET6
2135 			/* XXX inp locking */
2136 			if ((inp->inp_vflag & INP_IPV4) == 0)
2137 				continue;
2138 #endif
2139 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2140 			    inp->inp_lport != lport)
2141 				continue;
2142 
2143 			injail = prison_flag(inp->inp_cred, PR_IP4);
2144 			if (injail) {
2145 				if (prison_check_ip4(inp->inp_cred,
2146 				    &laddr) != 0)
2147 					continue;
2148 			} else {
2149 				if (local_exact != NULL)
2150 					continue;
2151 			}
2152 
2153 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2154 				if (injail)
2155 					goto found;
2156 				else
2157 					local_exact = inp;
2158 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2159 #ifdef INET6
2160 				/* XXX inp locking, NULL check */
2161 				if (inp->inp_vflag & INP_IPV6PROTO)
2162 					local_wild_mapped = inp;
2163 				else
2164 #endif
2165 					if (injail)
2166 						jail_wild = inp;
2167 					else
2168 						local_wild = inp;
2169 			}
2170 		} /* LIST_FOREACH */
2171 		inp = jail_wild;
2172 		if (inp == NULL)
2173 			inp = local_exact;
2174 		if (inp == NULL)
2175 			inp = local_wild;
2176 #ifdef INET6
2177 		if (inp == NULL)
2178 			inp = local_wild_mapped;
2179 #endif
2180 		if (inp != NULL)
2181 			goto found;
2182 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2183 	INP_GROUP_UNLOCK(pcbgroup);
2184 	return (NULL);
2185 
2186 found:
2187 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2188 		locked = INP_TRY_WLOCK(inp);
2189 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2190 		locked = INP_TRY_RLOCK(inp);
2191 	else
2192 		panic("%s: locking bug", __func__);
2193 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2194 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2195 			INP_WUNLOCK(inp);
2196 		else
2197 			INP_RUNLOCK(inp);
2198 		return (NULL);
2199 	} else if (!locked)
2200 		in_pcbref(inp);
2201 	INP_GROUP_UNLOCK(pcbgroup);
2202 	if (!locked) {
2203 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2204 			INP_WLOCK(inp);
2205 			if (in_pcbrele_wlocked(inp))
2206 				return (NULL);
2207 		} else {
2208 			INP_RLOCK(inp);
2209 			if (in_pcbrele_rlocked(inp))
2210 				return (NULL);
2211 		}
2212 	}
2213 #ifdef INVARIANTS
2214 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2215 		INP_WLOCK_ASSERT(inp);
2216 	else
2217 		INP_RLOCK_ASSERT(inp);
2218 #endif
2219 	return (inp);
2220 }
2221 #endif /* PCBGROUP */
2222 
2223 /*
2224  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2225  * that the caller has locked the hash list, and will not perform any further
2226  * locking or reference operations on either the hash list or the connection.
2227  */
2228 static struct inpcb *
2229 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2230     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2231     struct ifnet *ifp)
2232 {
2233 	struct inpcbhead *head;
2234 	struct inpcb *inp, *tmpinp;
2235 	u_short fport = fport_arg, lport = lport_arg;
2236 
2237 #ifdef INVARIANTS
2238 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2239 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2240 	if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2241 		MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2242 #endif
2243 	/*
2244 	 * First look for an exact match.
2245 	 */
2246 	tmpinp = NULL;
2247 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2248 	    pcbinfo->ipi_hashmask)];
2249 	CK_LIST_FOREACH(inp, head, inp_hash) {
2250 #ifdef INET6
2251 		/* XXX inp locking */
2252 		if ((inp->inp_vflag & INP_IPV4) == 0)
2253 			continue;
2254 #endif
2255 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2256 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2257 		    inp->inp_fport == fport &&
2258 		    inp->inp_lport == lport) {
2259 			/*
2260 			 * XXX We should be able to directly return
2261 			 * the inp here, without any checks.
2262 			 * Well unless both bound with SO_REUSEPORT?
2263 			 */
2264 			if (prison_flag(inp->inp_cred, PR_IP4))
2265 				return (inp);
2266 			if (tmpinp == NULL)
2267 				tmpinp = inp;
2268 		}
2269 	}
2270 	if (tmpinp != NULL)
2271 		return (tmpinp);
2272 
2273 	/*
2274 	 * Then look in lb group (for wildcard match).
2275 	 */
2276 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2277 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2278 		    fport, lookupflags);
2279 		if (inp != NULL)
2280 			return (inp);
2281 	}
2282 
2283 	/*
2284 	 * Then look for a wildcard match, if requested.
2285 	 */
2286 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2287 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2288 #ifdef INET6
2289 		struct inpcb *local_wild_mapped = NULL;
2290 #endif
2291 		struct inpcb *jail_wild = NULL;
2292 		int injail;
2293 
2294 		/*
2295 		 * Order of socket selection - we always prefer jails.
2296 		 *      1. jailed, non-wild.
2297 		 *      2. jailed, wild.
2298 		 *      3. non-jailed, non-wild.
2299 		 *      4. non-jailed, wild.
2300 		 */
2301 
2302 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2303 		    0, pcbinfo->ipi_hashmask)];
2304 		CK_LIST_FOREACH(inp, head, inp_hash) {
2305 #ifdef INET6
2306 			/* XXX inp locking */
2307 			if ((inp->inp_vflag & INP_IPV4) == 0)
2308 				continue;
2309 #endif
2310 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2311 			    inp->inp_lport != lport)
2312 				continue;
2313 
2314 			injail = prison_flag(inp->inp_cred, PR_IP4);
2315 			if (injail) {
2316 				if (prison_check_ip4(inp->inp_cred,
2317 				    &laddr) != 0)
2318 					continue;
2319 			} else {
2320 				if (local_exact != NULL)
2321 					continue;
2322 			}
2323 
2324 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2325 				if (injail)
2326 					return (inp);
2327 				else
2328 					local_exact = inp;
2329 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2330 #ifdef INET6
2331 				/* XXX inp locking, NULL check */
2332 				if (inp->inp_vflag & INP_IPV6PROTO)
2333 					local_wild_mapped = inp;
2334 				else
2335 #endif
2336 					if (injail)
2337 						jail_wild = inp;
2338 					else
2339 						local_wild = inp;
2340 			}
2341 		} /* LIST_FOREACH */
2342 		if (jail_wild != NULL)
2343 			return (jail_wild);
2344 		if (local_exact != NULL)
2345 			return (local_exact);
2346 		if (local_wild != NULL)
2347 			return (local_wild);
2348 #ifdef INET6
2349 		if (local_wild_mapped != NULL)
2350 			return (local_wild_mapped);
2351 #endif
2352 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2353 
2354 	return (NULL);
2355 }
2356 
2357 /*
2358  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2359  * hash list lock, and will return the inpcb locked (i.e., requires
2360  * INPLOOKUP_LOCKPCB).
2361  */
2362 static struct inpcb *
2363 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2364     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2365     struct ifnet *ifp)
2366 {
2367 	struct inpcb *inp;
2368 
2369 	INP_HASH_RLOCK(pcbinfo);
2370 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2371 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2372 	if (inp != NULL) {
2373 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2374 			INP_WLOCK(inp);
2375 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2376 				INP_WUNLOCK(inp);
2377 				inp = NULL;
2378 			}
2379 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2380 			INP_RLOCK(inp);
2381 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2382 				INP_RUNLOCK(inp);
2383 				inp = NULL;
2384 			}
2385 		} else
2386 			panic("%s: locking bug", __func__);
2387 #ifdef INVARIANTS
2388 		if (inp != NULL) {
2389 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2390 				INP_WLOCK_ASSERT(inp);
2391 			else
2392 				INP_RLOCK_ASSERT(inp);
2393 		}
2394 #endif
2395 	}
2396 	INP_HASH_RUNLOCK(pcbinfo);
2397 	return (inp);
2398 }
2399 
2400 /*
2401  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2402  * from which a pre-calculated hash value may be extracted.
2403  *
2404  * Possibly more of this logic should be in in_pcbgroup.c.
2405  */
2406 struct inpcb *
2407 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2408     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2409 {
2410 #if defined(PCBGROUP) && !defined(RSS)
2411 	struct inpcbgroup *pcbgroup;
2412 #endif
2413 
2414 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2415 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2416 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2417 	    ("%s: LOCKPCB not set", __func__));
2418 
2419 	/*
2420 	 * When not using RSS, use connection groups in preference to the
2421 	 * reservation table when looking up 4-tuples.  When using RSS, just
2422 	 * use the reservation table, due to the cost of the Toeplitz hash
2423 	 * in software.
2424 	 *
2425 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2426 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2427 	 * in software.
2428 	 */
2429 #if defined(PCBGROUP) && !defined(RSS)
2430 	if (in_pcbgroup_enabled(pcbinfo)) {
2431 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2432 		    fport);
2433 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2434 		    laddr, lport, lookupflags, ifp));
2435 	}
2436 #endif
2437 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2438 	    lookupflags, ifp));
2439 }
2440 
2441 struct inpcb *
2442 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2443     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2444     struct ifnet *ifp, struct mbuf *m)
2445 {
2446 #ifdef PCBGROUP
2447 	struct inpcbgroup *pcbgroup;
2448 #endif
2449 
2450 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2451 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2452 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2453 	    ("%s: LOCKPCB not set", __func__));
2454 
2455 #ifdef PCBGROUP
2456 	/*
2457 	 * If we can use a hardware-generated hash to look up the connection
2458 	 * group, use that connection group to find the inpcb.  Otherwise
2459 	 * fall back on a software hash -- or the reservation table if we're
2460 	 * using RSS.
2461 	 *
2462 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2463 	 */
2464 	if (in_pcbgroup_enabled(pcbinfo) &&
2465 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2466 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2467 		    m->m_pkthdr.flowid);
2468 		if (pcbgroup != NULL)
2469 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2470 			    fport, laddr, lport, lookupflags, ifp));
2471 #ifndef RSS
2472 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2473 		    fport);
2474 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2475 		    laddr, lport, lookupflags, ifp));
2476 #endif
2477 	}
2478 #endif
2479 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2480 	    lookupflags, ifp));
2481 }
2482 #endif /* INET */
2483 
2484 /*
2485  * Insert PCB onto various hash lists.
2486  */
2487 static int
2488 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2489 {
2490 	struct inpcbhead *pcbhash;
2491 	struct inpcbporthead *pcbporthash;
2492 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2493 	struct inpcbport *phd;
2494 	u_int32_t hashkey_faddr;
2495 	int so_options;
2496 
2497 	INP_WLOCK_ASSERT(inp);
2498 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2499 
2500 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2501 	    ("in_pcbinshash: INP_INHASHLIST"));
2502 
2503 #ifdef INET6
2504 	if (inp->inp_vflag & INP_IPV6)
2505 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2506 	else
2507 #endif
2508 	hashkey_faddr = inp->inp_faddr.s_addr;
2509 
2510 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2511 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2512 
2513 	pcbporthash = &pcbinfo->ipi_porthashbase[
2514 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2515 
2516 	/*
2517 	 * Add entry to load balance group.
2518 	 * Only do this if SO_REUSEPORT_LB is set.
2519 	 */
2520 	so_options = inp_so_options(inp);
2521 	if (so_options & SO_REUSEPORT_LB) {
2522 		int ret = in_pcbinslbgrouphash(inp);
2523 		if (ret) {
2524 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2525 			return (ret);
2526 		}
2527 	}
2528 
2529 	/*
2530 	 * Go through port list and look for a head for this lport.
2531 	 */
2532 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2533 		if (phd->phd_port == inp->inp_lport)
2534 			break;
2535 	}
2536 	/*
2537 	 * If none exists, malloc one and tack it on.
2538 	 */
2539 	if (phd == NULL) {
2540 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2541 		if (phd == NULL) {
2542 			return (ENOBUFS); /* XXX */
2543 		}
2544 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2545 		phd->phd_port = inp->inp_lport;
2546 		CK_LIST_INIT(&phd->phd_pcblist);
2547 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2548 	}
2549 	inp->inp_phd = phd;
2550 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2551 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2552 	inp->inp_flags |= INP_INHASHLIST;
2553 #ifdef PCBGROUP
2554 	if (do_pcbgroup_update)
2555 		in_pcbgroup_update(inp);
2556 #endif
2557 	return (0);
2558 }
2559 
2560 /*
2561  * For now, there are two public interfaces to insert an inpcb into the hash
2562  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2563  * is used only in the TCP syncache, where in_pcbinshash is called before the
2564  * full 4-tuple is set for the inpcb, and we don't want to install in the
2565  * pcbgroup until later.
2566  *
2567  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2568  * connection groups, and partially initialised inpcbs should not be exposed
2569  * to either reservation hash tables or pcbgroups.
2570  */
2571 int
2572 in_pcbinshash(struct inpcb *inp)
2573 {
2574 
2575 	return (in_pcbinshash_internal(inp, 1));
2576 }
2577 
2578 int
2579 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2580 {
2581 
2582 	return (in_pcbinshash_internal(inp, 0));
2583 }
2584 
2585 /*
2586  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2587  * changed. NOTE: This does not handle the case of the lport changing (the
2588  * hashed port list would have to be updated as well), so the lport must
2589  * not change after in_pcbinshash() has been called.
2590  */
2591 void
2592 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2593 {
2594 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2595 	struct inpcbhead *head;
2596 	u_int32_t hashkey_faddr;
2597 
2598 	INP_WLOCK_ASSERT(inp);
2599 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2600 
2601 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2602 	    ("in_pcbrehash: !INP_INHASHLIST"));
2603 
2604 #ifdef INET6
2605 	if (inp->inp_vflag & INP_IPV6)
2606 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2607 	else
2608 #endif
2609 	hashkey_faddr = inp->inp_faddr.s_addr;
2610 
2611 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2612 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2613 
2614 	CK_LIST_REMOVE(inp, inp_hash);
2615 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2616 
2617 #ifdef PCBGROUP
2618 	if (m != NULL)
2619 		in_pcbgroup_update_mbuf(inp, m);
2620 	else
2621 		in_pcbgroup_update(inp);
2622 #endif
2623 }
2624 
2625 void
2626 in_pcbrehash(struct inpcb *inp)
2627 {
2628 
2629 	in_pcbrehash_mbuf(inp, NULL);
2630 }
2631 
2632 /*
2633  * Remove PCB from various lists.
2634  */
2635 static void
2636 in_pcbremlists(struct inpcb *inp)
2637 {
2638 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2639 
2640 #ifdef INVARIANTS
2641 	if (pcbinfo == &V_tcbinfo) {
2642 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2643 	} else {
2644 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2645 	}
2646 #endif
2647 
2648 	INP_WLOCK_ASSERT(inp);
2649 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2650 
2651 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2652 	if (inp->inp_flags & INP_INHASHLIST) {
2653 		struct inpcbport *phd = inp->inp_phd;
2654 
2655 		INP_HASH_WLOCK(pcbinfo);
2656 
2657 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2658 		in_pcbremlbgrouphash(inp);
2659 
2660 		CK_LIST_REMOVE(inp, inp_hash);
2661 		CK_LIST_REMOVE(inp, inp_portlist);
2662 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2663 			CK_LIST_REMOVE(phd, phd_hash);
2664 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2665 		}
2666 		INP_HASH_WUNLOCK(pcbinfo);
2667 		inp->inp_flags &= ~INP_INHASHLIST;
2668 	}
2669 	CK_LIST_REMOVE(inp, inp_list);
2670 	pcbinfo->ipi_count--;
2671 #ifdef PCBGROUP
2672 	in_pcbgroup_remove(inp);
2673 #endif
2674 }
2675 
2676 /*
2677  * Check for alternatives when higher level complains
2678  * about service problems.  For now, invalidate cached
2679  * routing information.  If the route was created dynamically
2680  * (by a redirect), time to try a default gateway again.
2681  */
2682 void
2683 in_losing(struct inpcb *inp)
2684 {
2685 
2686 	RO_INVALIDATE_CACHE(&inp->inp_route);
2687 	return;
2688 }
2689 
2690 /*
2691  * A set label operation has occurred at the socket layer, propagate the
2692  * label change into the in_pcb for the socket.
2693  */
2694 void
2695 in_pcbsosetlabel(struct socket *so)
2696 {
2697 #ifdef MAC
2698 	struct inpcb *inp;
2699 
2700 	inp = sotoinpcb(so);
2701 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2702 
2703 	INP_WLOCK(inp);
2704 	SOCK_LOCK(so);
2705 	mac_inpcb_sosetlabel(so, inp);
2706 	SOCK_UNLOCK(so);
2707 	INP_WUNLOCK(inp);
2708 #endif
2709 }
2710 
2711 /*
2712  * ipport_tick runs once per second, determining if random port allocation
2713  * should be continued.  If more than ipport_randomcps ports have been
2714  * allocated in the last second, then we return to sequential port
2715  * allocation. We return to random allocation only once we drop below
2716  * ipport_randomcps for at least ipport_randomtime seconds.
2717  */
2718 static void
2719 ipport_tick(void *xtp)
2720 {
2721 	VNET_ITERATOR_DECL(vnet_iter);
2722 
2723 	VNET_LIST_RLOCK_NOSLEEP();
2724 	VNET_FOREACH(vnet_iter) {
2725 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2726 		if (V_ipport_tcpallocs <=
2727 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2728 			if (V_ipport_stoprandom > 0)
2729 				V_ipport_stoprandom--;
2730 		} else
2731 			V_ipport_stoprandom = V_ipport_randomtime;
2732 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2733 		CURVNET_RESTORE();
2734 	}
2735 	VNET_LIST_RUNLOCK_NOSLEEP();
2736 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2737 }
2738 
2739 static void
2740 ip_fini(void *xtp)
2741 {
2742 
2743 	callout_stop(&ipport_tick_callout);
2744 }
2745 
2746 /*
2747  * The ipport_callout should start running at about the time we attach the
2748  * inet or inet6 domains.
2749  */
2750 static void
2751 ipport_tick_init(const void *unused __unused)
2752 {
2753 
2754 	/* Start ipport_tick. */
2755 	callout_init(&ipport_tick_callout, 1);
2756 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2757 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2758 		SHUTDOWN_PRI_DEFAULT);
2759 }
2760 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2761     ipport_tick_init, NULL);
2762 
2763 void
2764 inp_wlock(struct inpcb *inp)
2765 {
2766 
2767 	INP_WLOCK(inp);
2768 }
2769 
2770 void
2771 inp_wunlock(struct inpcb *inp)
2772 {
2773 
2774 	INP_WUNLOCK(inp);
2775 }
2776 
2777 void
2778 inp_rlock(struct inpcb *inp)
2779 {
2780 
2781 	INP_RLOCK(inp);
2782 }
2783 
2784 void
2785 inp_runlock(struct inpcb *inp)
2786 {
2787 
2788 	INP_RUNLOCK(inp);
2789 }
2790 
2791 #ifdef INVARIANT_SUPPORT
2792 void
2793 inp_lock_assert(struct inpcb *inp)
2794 {
2795 
2796 	INP_WLOCK_ASSERT(inp);
2797 }
2798 
2799 void
2800 inp_unlock_assert(struct inpcb *inp)
2801 {
2802 
2803 	INP_UNLOCK_ASSERT(inp);
2804 }
2805 #endif
2806 
2807 void
2808 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2809 {
2810 	struct inpcb *inp;
2811 
2812 	INP_INFO_WLOCK(&V_tcbinfo);
2813 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2814 		INP_WLOCK(inp);
2815 		func(inp, arg);
2816 		INP_WUNLOCK(inp);
2817 	}
2818 	INP_INFO_WUNLOCK(&V_tcbinfo);
2819 }
2820 
2821 struct socket *
2822 inp_inpcbtosocket(struct inpcb *inp)
2823 {
2824 
2825 	INP_WLOCK_ASSERT(inp);
2826 	return (inp->inp_socket);
2827 }
2828 
2829 struct tcpcb *
2830 inp_inpcbtotcpcb(struct inpcb *inp)
2831 {
2832 
2833 	INP_WLOCK_ASSERT(inp);
2834 	return ((struct tcpcb *)inp->inp_ppcb);
2835 }
2836 
2837 int
2838 inp_ip_tos_get(const struct inpcb *inp)
2839 {
2840 
2841 	return (inp->inp_ip_tos);
2842 }
2843 
2844 void
2845 inp_ip_tos_set(struct inpcb *inp, int val)
2846 {
2847 
2848 	inp->inp_ip_tos = val;
2849 }
2850 
2851 void
2852 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2853     uint32_t *faddr, uint16_t *fp)
2854 {
2855 
2856 	INP_LOCK_ASSERT(inp);
2857 	*laddr = inp->inp_laddr.s_addr;
2858 	*faddr = inp->inp_faddr.s_addr;
2859 	*lp = inp->inp_lport;
2860 	*fp = inp->inp_fport;
2861 }
2862 
2863 struct inpcb *
2864 so_sotoinpcb(struct socket *so)
2865 {
2866 
2867 	return (sotoinpcb(so));
2868 }
2869 
2870 struct tcpcb *
2871 so_sototcpcb(struct socket *so)
2872 {
2873 
2874 	return (sototcpcb(so));
2875 }
2876 
2877 /*
2878  * Create an external-format (``xinpcb'') structure using the information in
2879  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2880  * reduce the spew of irrelevant information over this interface, to isolate
2881  * user code from changes in the kernel structure, and potentially to provide
2882  * information-hiding if we decide that some of this information should be
2883  * hidden from users.
2884  */
2885 void
2886 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2887 {
2888 
2889 	bzero(xi, sizeof(*xi));
2890 	xi->xi_len = sizeof(struct xinpcb);
2891 	if (inp->inp_socket)
2892 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2893 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2894 	xi->inp_gencnt = inp->inp_gencnt;
2895 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2896 	xi->inp_flow = inp->inp_flow;
2897 	xi->inp_flowid = inp->inp_flowid;
2898 	xi->inp_flowtype = inp->inp_flowtype;
2899 	xi->inp_flags = inp->inp_flags;
2900 	xi->inp_flags2 = inp->inp_flags2;
2901 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2902 	xi->in6p_cksum = inp->in6p_cksum;
2903 	xi->in6p_hops = inp->in6p_hops;
2904 	xi->inp_ip_tos = inp->inp_ip_tos;
2905 	xi->inp_vflag = inp->inp_vflag;
2906 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2907 	xi->inp_ip_p = inp->inp_ip_p;
2908 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2909 }
2910 
2911 #ifdef DDB
2912 static void
2913 db_print_indent(int indent)
2914 {
2915 	int i;
2916 
2917 	for (i = 0; i < indent; i++)
2918 		db_printf(" ");
2919 }
2920 
2921 static void
2922 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2923 {
2924 	char faddr_str[48], laddr_str[48];
2925 
2926 	db_print_indent(indent);
2927 	db_printf("%s at %p\n", name, inc);
2928 
2929 	indent += 2;
2930 
2931 #ifdef INET6
2932 	if (inc->inc_flags & INC_ISIPV6) {
2933 		/* IPv6. */
2934 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2935 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2936 	} else
2937 #endif
2938 	{
2939 		/* IPv4. */
2940 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2941 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2942 	}
2943 	db_print_indent(indent);
2944 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2945 	    ntohs(inc->inc_lport));
2946 	db_print_indent(indent);
2947 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2948 	    ntohs(inc->inc_fport));
2949 }
2950 
2951 static void
2952 db_print_inpflags(int inp_flags)
2953 {
2954 	int comma;
2955 
2956 	comma = 0;
2957 	if (inp_flags & INP_RECVOPTS) {
2958 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2959 		comma = 1;
2960 	}
2961 	if (inp_flags & INP_RECVRETOPTS) {
2962 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2963 		comma = 1;
2964 	}
2965 	if (inp_flags & INP_RECVDSTADDR) {
2966 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2967 		comma = 1;
2968 	}
2969 	if (inp_flags & INP_ORIGDSTADDR) {
2970 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2971 		comma = 1;
2972 	}
2973 	if (inp_flags & INP_HDRINCL) {
2974 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2975 		comma = 1;
2976 	}
2977 	if (inp_flags & INP_HIGHPORT) {
2978 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2979 		comma = 1;
2980 	}
2981 	if (inp_flags & INP_LOWPORT) {
2982 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2983 		comma = 1;
2984 	}
2985 	if (inp_flags & INP_ANONPORT) {
2986 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2987 		comma = 1;
2988 	}
2989 	if (inp_flags & INP_RECVIF) {
2990 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2991 		comma = 1;
2992 	}
2993 	if (inp_flags & INP_MTUDISC) {
2994 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2995 		comma = 1;
2996 	}
2997 	if (inp_flags & INP_RECVTTL) {
2998 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
2999 		comma = 1;
3000 	}
3001 	if (inp_flags & INP_DONTFRAG) {
3002 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3003 		comma = 1;
3004 	}
3005 	if (inp_flags & INP_RECVTOS) {
3006 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3007 		comma = 1;
3008 	}
3009 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3010 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3011 		comma = 1;
3012 	}
3013 	if (inp_flags & IN6P_PKTINFO) {
3014 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3015 		comma = 1;
3016 	}
3017 	if (inp_flags & IN6P_HOPLIMIT) {
3018 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3019 		comma = 1;
3020 	}
3021 	if (inp_flags & IN6P_HOPOPTS) {
3022 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3023 		comma = 1;
3024 	}
3025 	if (inp_flags & IN6P_DSTOPTS) {
3026 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3027 		comma = 1;
3028 	}
3029 	if (inp_flags & IN6P_RTHDR) {
3030 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3031 		comma = 1;
3032 	}
3033 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3034 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3035 		comma = 1;
3036 	}
3037 	if (inp_flags & IN6P_TCLASS) {
3038 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3039 		comma = 1;
3040 	}
3041 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3042 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3043 		comma = 1;
3044 	}
3045 	if (inp_flags & INP_TIMEWAIT) {
3046 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3047 		comma  = 1;
3048 	}
3049 	if (inp_flags & INP_ONESBCAST) {
3050 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3051 		comma  = 1;
3052 	}
3053 	if (inp_flags & INP_DROPPED) {
3054 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3055 		comma  = 1;
3056 	}
3057 	if (inp_flags & INP_SOCKREF) {
3058 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3059 		comma  = 1;
3060 	}
3061 	if (inp_flags & IN6P_RFC2292) {
3062 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3063 		comma = 1;
3064 	}
3065 	if (inp_flags & IN6P_MTU) {
3066 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3067 		comma = 1;
3068 	}
3069 }
3070 
3071 static void
3072 db_print_inpvflag(u_char inp_vflag)
3073 {
3074 	int comma;
3075 
3076 	comma = 0;
3077 	if (inp_vflag & INP_IPV4) {
3078 		db_printf("%sINP_IPV4", comma ? ", " : "");
3079 		comma  = 1;
3080 	}
3081 	if (inp_vflag & INP_IPV6) {
3082 		db_printf("%sINP_IPV6", comma ? ", " : "");
3083 		comma  = 1;
3084 	}
3085 	if (inp_vflag & INP_IPV6PROTO) {
3086 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3087 		comma  = 1;
3088 	}
3089 }
3090 
3091 static void
3092 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3093 {
3094 
3095 	db_print_indent(indent);
3096 	db_printf("%s at %p\n", name, inp);
3097 
3098 	indent += 2;
3099 
3100 	db_print_indent(indent);
3101 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3102 
3103 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3104 
3105 	db_print_indent(indent);
3106 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3107 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3108 
3109 	db_print_indent(indent);
3110 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3111 	   inp->inp_label, inp->inp_flags);
3112 	db_print_inpflags(inp->inp_flags);
3113 	db_printf(")\n");
3114 
3115 	db_print_indent(indent);
3116 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3117 	    inp->inp_vflag);
3118 	db_print_inpvflag(inp->inp_vflag);
3119 	db_printf(")\n");
3120 
3121 	db_print_indent(indent);
3122 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3123 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3124 
3125 	db_print_indent(indent);
3126 #ifdef INET6
3127 	if (inp->inp_vflag & INP_IPV6) {
3128 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3129 		    "in6p_moptions: %p\n", inp->in6p_options,
3130 		    inp->in6p_outputopts, inp->in6p_moptions);
3131 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3132 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3133 		    inp->in6p_hops);
3134 	} else
3135 #endif
3136 	{
3137 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3138 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3139 		    inp->inp_options, inp->inp_moptions);
3140 	}
3141 
3142 	db_print_indent(indent);
3143 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3144 	    (uintmax_t)inp->inp_gencnt);
3145 }
3146 
3147 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3148 {
3149 	struct inpcb *inp;
3150 
3151 	if (!have_addr) {
3152 		db_printf("usage: show inpcb <addr>\n");
3153 		return;
3154 	}
3155 	inp = (struct inpcb *)addr;
3156 
3157 	db_print_inpcb(inp, "inpcb", 0);
3158 }
3159 #endif /* DDB */
3160 
3161 #ifdef RATELIMIT
3162 /*
3163  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3164  * if any.
3165  */
3166 int
3167 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3168 {
3169 	union if_snd_tag_modify_params params = {
3170 		.rate_limit.max_rate = max_pacing_rate,
3171 	};
3172 	struct m_snd_tag *mst;
3173 	struct ifnet *ifp;
3174 	int error;
3175 
3176 	mst = inp->inp_snd_tag;
3177 	if (mst == NULL)
3178 		return (EINVAL);
3179 
3180 	ifp = mst->ifp;
3181 	if (ifp == NULL)
3182 		return (EINVAL);
3183 
3184 	if (ifp->if_snd_tag_modify == NULL) {
3185 		error = EOPNOTSUPP;
3186 	} else {
3187 		error = ifp->if_snd_tag_modify(mst, &params);
3188 	}
3189 	return (error);
3190 }
3191 
3192 /*
3193  * Query existing TX rate limit based on the existing
3194  * "inp->inp_snd_tag", if any.
3195  */
3196 int
3197 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3198 {
3199 	union if_snd_tag_query_params params = { };
3200 	struct m_snd_tag *mst;
3201 	struct ifnet *ifp;
3202 	int error;
3203 
3204 	mst = inp->inp_snd_tag;
3205 	if (mst == NULL)
3206 		return (EINVAL);
3207 
3208 	ifp = mst->ifp;
3209 	if (ifp == NULL)
3210 		return (EINVAL);
3211 
3212 	if (ifp->if_snd_tag_query == NULL) {
3213 		error = EOPNOTSUPP;
3214 	} else {
3215 		error = ifp->if_snd_tag_query(mst, &params);
3216 		if (error == 0 &&  p_max_pacing_rate != NULL)
3217 			*p_max_pacing_rate = params.rate_limit.max_rate;
3218 	}
3219 	return (error);
3220 }
3221 
3222 /*
3223  * Query existing TX queue level based on the existing
3224  * "inp->inp_snd_tag", if any.
3225  */
3226 int
3227 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3228 {
3229 	union if_snd_tag_query_params params = { };
3230 	struct m_snd_tag *mst;
3231 	struct ifnet *ifp;
3232 	int error;
3233 
3234 	mst = inp->inp_snd_tag;
3235 	if (mst == NULL)
3236 		return (EINVAL);
3237 
3238 	ifp = mst->ifp;
3239 	if (ifp == NULL)
3240 		return (EINVAL);
3241 
3242 	if (ifp->if_snd_tag_query == NULL)
3243 		return (EOPNOTSUPP);
3244 
3245 	error = ifp->if_snd_tag_query(mst, &params);
3246 	if (error == 0 &&  p_txqueue_level != NULL)
3247 		*p_txqueue_level = params.rate_limit.queue_level;
3248 	return (error);
3249 }
3250 
3251 /*
3252  * Allocate a new TX rate limit send tag from the network interface
3253  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3254  */
3255 int
3256 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3257     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3258 {
3259 	union if_snd_tag_alloc_params params = {
3260 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3261 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3262 		.rate_limit.hdr.flowid = flowid,
3263 		.rate_limit.hdr.flowtype = flowtype,
3264 		.rate_limit.max_rate = max_pacing_rate,
3265 	};
3266 	int error;
3267 
3268 	INP_WLOCK_ASSERT(inp);
3269 
3270 	if (inp->inp_snd_tag != NULL)
3271 		return (EINVAL);
3272 
3273 	if (ifp->if_snd_tag_alloc == NULL) {
3274 		error = EOPNOTSUPP;
3275 	} else {
3276 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3277 
3278 		/*
3279 		 * At success increment the refcount on
3280 		 * the send tag's network interface:
3281 		 */
3282 		if (error == 0)
3283 			if_ref(inp->inp_snd_tag->ifp);
3284 	}
3285 	return (error);
3286 }
3287 
3288 /*
3289  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3290  * if any:
3291  */
3292 void
3293 in_pcbdetach_txrtlmt(struct inpcb *inp)
3294 {
3295 	struct m_snd_tag *mst;
3296 	struct ifnet *ifp;
3297 
3298 	INP_WLOCK_ASSERT(inp);
3299 
3300 	mst = inp->inp_snd_tag;
3301 	inp->inp_snd_tag = NULL;
3302 
3303 	if (mst == NULL)
3304 		return;
3305 
3306 	ifp = mst->ifp;
3307 	if (ifp == NULL)
3308 		return;
3309 
3310 	/*
3311 	 * If the device was detached while we still had reference(s)
3312 	 * on the ifp, we assume if_snd_tag_free() was replaced with
3313 	 * stubs.
3314 	 */
3315 	ifp->if_snd_tag_free(mst);
3316 
3317 	/* release reference count on network interface */
3318 	if_rele(ifp);
3319 }
3320 
3321 /*
3322  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3323  * is set in the fast path and will attach/detach/modify the TX rate
3324  * limit send tag based on the socket's so_max_pacing_rate value.
3325  */
3326 void
3327 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3328 {
3329 	struct socket *socket;
3330 	uint32_t max_pacing_rate;
3331 	bool did_upgrade;
3332 	int error;
3333 
3334 	if (inp == NULL)
3335 		return;
3336 
3337 	socket = inp->inp_socket;
3338 	if (socket == NULL)
3339 		return;
3340 
3341 	if (!INP_WLOCKED(inp)) {
3342 		/*
3343 		 * NOTE: If the write locking fails, we need to bail
3344 		 * out and use the non-ratelimited ring for the
3345 		 * transmit until there is a new chance to get the
3346 		 * write lock.
3347 		 */
3348 		if (!INP_TRY_UPGRADE(inp))
3349 			return;
3350 		did_upgrade = 1;
3351 	} else {
3352 		did_upgrade = 0;
3353 	}
3354 
3355 	/*
3356 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3357 	 * because atomic updates are not required since the variable
3358 	 * is checked at every mbuf we send. It is assumed that the
3359 	 * variable read itself will be atomic.
3360 	 */
3361 	max_pacing_rate = socket->so_max_pacing_rate;
3362 
3363 	/*
3364 	 * NOTE: When attaching to a network interface a reference is
3365 	 * made to ensure the network interface doesn't go away until
3366 	 * all ratelimit connections are gone. The network interface
3367 	 * pointers compared below represent valid network interfaces,
3368 	 * except when comparing towards NULL.
3369 	 */
3370 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3371 		error = 0;
3372 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3373 		if (inp->inp_snd_tag != NULL)
3374 			in_pcbdetach_txrtlmt(inp);
3375 		error = 0;
3376 	} else if (inp->inp_snd_tag == NULL) {
3377 		/*
3378 		 * In order to utilize packet pacing with RSS, we need
3379 		 * to wait until there is a valid RSS hash before we
3380 		 * can proceed:
3381 		 */
3382 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3383 			error = EAGAIN;
3384 		} else {
3385 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3386 			    mb->m_pkthdr.flowid, max_pacing_rate);
3387 		}
3388 	} else {
3389 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3390 	}
3391 	if (error == 0 || error == EOPNOTSUPP)
3392 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3393 	if (did_upgrade)
3394 		INP_DOWNGRADE(inp);
3395 }
3396 
3397 /*
3398  * Track route changes for TX rate limiting.
3399  */
3400 void
3401 in_pcboutput_eagain(struct inpcb *inp)
3402 {
3403 	bool did_upgrade;
3404 
3405 	if (inp == NULL)
3406 		return;
3407 
3408 	if (inp->inp_snd_tag == NULL)
3409 		return;
3410 
3411 	if (!INP_WLOCKED(inp)) {
3412 		/*
3413 		 * NOTE: If the write locking fails, we need to bail
3414 		 * out and use the non-ratelimited ring for the
3415 		 * transmit until there is a new chance to get the
3416 		 * write lock.
3417 		 */
3418 		if (!INP_TRY_UPGRADE(inp))
3419 			return;
3420 		did_upgrade = 1;
3421 	} else {
3422 		did_upgrade = 0;
3423 	}
3424 
3425 	/* detach rate limiting */
3426 	in_pcbdetach_txrtlmt(inp);
3427 
3428 	/* make sure new mbuf send tag allocation is made */
3429 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3430 
3431 	if (did_upgrade)
3432 		INP_DOWNGRADE(inp);
3433 }
3434 #endif /* RATELIMIT */
3435