xref: /freebsd/sys/netinet/in_pcb.c (revision 6683132d54bd6d589889e43dabdc53d35e38a028)
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 #ifdef INET
90 #include <netinet/in_var.h>
91 #endif
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp_var.h>
94 #ifdef TCPHPTS
95 #include <netinet/tcp_hpts.h>
96 #endif
97 #include <netinet/udp.h>
98 #include <netinet/udp_var.h>
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #endif /* INET6 */
105 #endif
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 in_multi *inm;
1783 	struct in_mfilter *imf;
1784 	struct ip_moptions *imo;
1785 
1786 	INP_INFO_WLOCK(pcbinfo);
1787 	CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1788 		INP_WLOCK(inp);
1789 		imo = inp->inp_moptions;
1790 		if ((inp->inp_vflag & INP_IPV4) &&
1791 		    imo != NULL) {
1792 			/*
1793 			 * Unselect the outgoing interface if it is being
1794 			 * detached.
1795 			 */
1796 			if (imo->imo_multicast_ifp == ifp)
1797 				imo->imo_multicast_ifp = NULL;
1798 
1799 			/*
1800 			 * Drop multicast group membership if we joined
1801 			 * through the interface being detached.
1802 			 *
1803 			 * XXX This can all be deferred to an epoch_call
1804 			 */
1805 restart:
1806 			IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1807 				if ((inm = imf->imf_inm) == NULL)
1808 					continue;
1809 				if (inm->inm_ifp != ifp)
1810 					continue;
1811 				ip_mfilter_remove(&imo->imo_head, imf);
1812 				IN_MULTI_LOCK_ASSERT();
1813 				in_leavegroup_locked(inm, NULL);
1814 				ip_mfilter_free(imf);
1815 				goto restart;
1816 			}
1817 		}
1818 		INP_WUNLOCK(inp);
1819 	}
1820 	INP_INFO_WUNLOCK(pcbinfo);
1821 }
1822 
1823 /*
1824  * Lookup a PCB based on the local address and port.  Caller must hold the
1825  * hash lock.  No inpcb locks or references are acquired.
1826  */
1827 #define INP_LOOKUP_MAPPED_PCB_COST	3
1828 struct inpcb *
1829 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1830     u_short lport, int lookupflags, struct ucred *cred)
1831 {
1832 	struct inpcb *inp;
1833 #ifdef INET6
1834 	int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1835 #else
1836 	int matchwild = 3;
1837 #endif
1838 	int wildcard;
1839 
1840 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1841 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
1842 
1843 	INP_HASH_LOCK_ASSERT(pcbinfo);
1844 
1845 	if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1846 		struct inpcbhead *head;
1847 		/*
1848 		 * Look for an unconnected (wildcard foreign addr) PCB that
1849 		 * matches the local address and port we're looking for.
1850 		 */
1851 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1852 		    0, pcbinfo->ipi_hashmask)];
1853 		CK_LIST_FOREACH(inp, head, inp_hash) {
1854 #ifdef INET6
1855 			/* XXX inp locking */
1856 			if ((inp->inp_vflag & INP_IPV4) == 0)
1857 				continue;
1858 #endif
1859 			if (inp->inp_faddr.s_addr == INADDR_ANY &&
1860 			    inp->inp_laddr.s_addr == laddr.s_addr &&
1861 			    inp->inp_lport == lport) {
1862 				/*
1863 				 * Found?
1864 				 */
1865 				if (cred == NULL ||
1866 				    prison_equal_ip4(cred->cr_prison,
1867 					inp->inp_cred->cr_prison))
1868 					return (inp);
1869 			}
1870 		}
1871 		/*
1872 		 * Not found.
1873 		 */
1874 		return (NULL);
1875 	} else {
1876 		struct inpcbporthead *porthash;
1877 		struct inpcbport *phd;
1878 		struct inpcb *match = NULL;
1879 		/*
1880 		 * Best fit PCB lookup.
1881 		 *
1882 		 * First see if this local port is in use by looking on the
1883 		 * port hash list.
1884 		 */
1885 		porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1886 		    pcbinfo->ipi_porthashmask)];
1887 		CK_LIST_FOREACH(phd, porthash, phd_hash) {
1888 			if (phd->phd_port == lport)
1889 				break;
1890 		}
1891 		if (phd != NULL) {
1892 			/*
1893 			 * Port is in use by one or more PCBs. Look for best
1894 			 * fit.
1895 			 */
1896 			CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1897 				wildcard = 0;
1898 				if (cred != NULL &&
1899 				    !prison_equal_ip4(inp->inp_cred->cr_prison,
1900 					cred->cr_prison))
1901 					continue;
1902 #ifdef INET6
1903 				/* XXX inp locking */
1904 				if ((inp->inp_vflag & INP_IPV4) == 0)
1905 					continue;
1906 				/*
1907 				 * We never select the PCB that has
1908 				 * INP_IPV6 flag and is bound to :: if
1909 				 * we have another PCB which is bound
1910 				 * to 0.0.0.0.  If a PCB has the
1911 				 * INP_IPV6 flag, then we set its cost
1912 				 * higher than IPv4 only PCBs.
1913 				 *
1914 				 * Note that the case only happens
1915 				 * when a socket is bound to ::, under
1916 				 * the condition that the use of the
1917 				 * mapped address is allowed.
1918 				 */
1919 				if ((inp->inp_vflag & INP_IPV6) != 0)
1920 					wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1921 #endif
1922 				if (inp->inp_faddr.s_addr != INADDR_ANY)
1923 					wildcard++;
1924 				if (inp->inp_laddr.s_addr != INADDR_ANY) {
1925 					if (laddr.s_addr == INADDR_ANY)
1926 						wildcard++;
1927 					else if (inp->inp_laddr.s_addr != laddr.s_addr)
1928 						continue;
1929 				} else {
1930 					if (laddr.s_addr != INADDR_ANY)
1931 						wildcard++;
1932 				}
1933 				if (wildcard < matchwild) {
1934 					match = inp;
1935 					matchwild = wildcard;
1936 					if (matchwild == 0)
1937 						break;
1938 				}
1939 			}
1940 		}
1941 		return (match);
1942 	}
1943 }
1944 #undef INP_LOOKUP_MAPPED_PCB_COST
1945 
1946 static struct inpcb *
1947 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1948     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1949     uint16_t fport, int lookupflags)
1950 {
1951 	struct inpcb *local_wild;
1952 	const struct inpcblbgrouphead *hdr;
1953 	struct inpcblbgroup *grp;
1954 	uint32_t idx;
1955 
1956 	INP_HASH_LOCK_ASSERT(pcbinfo);
1957 
1958 	hdr = &pcbinfo->ipi_lbgrouphashbase[
1959 	    INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1960 
1961 	/*
1962 	 * Order of socket selection:
1963 	 * 1. non-wild.
1964 	 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1965 	 *
1966 	 * NOTE:
1967 	 * - Load balanced group does not contain jailed sockets
1968 	 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1969 	 */
1970 	local_wild = NULL;
1971 	CK_LIST_FOREACH(grp, hdr, il_list) {
1972 #ifdef INET6
1973 		if (!(grp->il_vflag & INP_IPV4))
1974 			continue;
1975 #endif
1976 		if (grp->il_lport != lport)
1977 			continue;
1978 
1979 		idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
1980 		    grp->il_inpcnt;
1981 		if (grp->il_laddr.s_addr == laddr->s_addr)
1982 			return (grp->il_inp[idx]);
1983 		if (grp->il_laddr.s_addr == INADDR_ANY &&
1984 		    (lookupflags & INPLOOKUP_WILDCARD) != 0)
1985 			local_wild = grp->il_inp[idx];
1986 	}
1987 	return (local_wild);
1988 }
1989 
1990 #ifdef PCBGROUP
1991 /*
1992  * Lookup PCB in hash list, using pcbgroup tables.
1993  */
1994 static struct inpcb *
1995 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1996     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1997     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1998 {
1999 	struct inpcbhead *head;
2000 	struct inpcb *inp, *tmpinp;
2001 	u_short fport = fport_arg, lport = lport_arg;
2002 	bool locked;
2003 
2004 	/*
2005 	 * First look for an exact match.
2006 	 */
2007 	tmpinp = NULL;
2008 	INP_GROUP_LOCK(pcbgroup);
2009 	head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2010 	    pcbgroup->ipg_hashmask)];
2011 	CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2012 #ifdef INET6
2013 		/* XXX inp locking */
2014 		if ((inp->inp_vflag & INP_IPV4) == 0)
2015 			continue;
2016 #endif
2017 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2018 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2019 		    inp->inp_fport == fport &&
2020 		    inp->inp_lport == lport) {
2021 			/*
2022 			 * XXX We should be able to directly return
2023 			 * the inp here, without any checks.
2024 			 * Well unless both bound with SO_REUSEPORT?
2025 			 */
2026 			if (prison_flag(inp->inp_cred, PR_IP4))
2027 				goto found;
2028 			if (tmpinp == NULL)
2029 				tmpinp = inp;
2030 		}
2031 	}
2032 	if (tmpinp != NULL) {
2033 		inp = tmpinp;
2034 		goto found;
2035 	}
2036 
2037 #ifdef	RSS
2038 	/*
2039 	 * For incoming connections, we may wish to do a wildcard
2040 	 * match for an RSS-local socket.
2041 	 */
2042 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2043 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2044 #ifdef INET6
2045 		struct inpcb *local_wild_mapped = NULL;
2046 #endif
2047 		struct inpcb *jail_wild = NULL;
2048 		struct inpcbhead *head;
2049 		int injail;
2050 
2051 		/*
2052 		 * Order of socket selection - we always prefer jails.
2053 		 *      1. jailed, non-wild.
2054 		 *      2. jailed, wild.
2055 		 *      3. non-jailed, non-wild.
2056 		 *      4. non-jailed, wild.
2057 		 */
2058 
2059 		head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2060 		    lport, 0, pcbgroup->ipg_hashmask)];
2061 		CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2062 #ifdef INET6
2063 			/* XXX inp locking */
2064 			if ((inp->inp_vflag & INP_IPV4) == 0)
2065 				continue;
2066 #endif
2067 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2068 			    inp->inp_lport != lport)
2069 				continue;
2070 
2071 			injail = prison_flag(inp->inp_cred, PR_IP4);
2072 			if (injail) {
2073 				if (prison_check_ip4(inp->inp_cred,
2074 				    &laddr) != 0)
2075 					continue;
2076 			} else {
2077 				if (local_exact != NULL)
2078 					continue;
2079 			}
2080 
2081 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2082 				if (injail)
2083 					goto found;
2084 				else
2085 					local_exact = inp;
2086 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2087 #ifdef INET6
2088 				/* XXX inp locking, NULL check */
2089 				if (inp->inp_vflag & INP_IPV6PROTO)
2090 					local_wild_mapped = inp;
2091 				else
2092 #endif
2093 					if (injail)
2094 						jail_wild = inp;
2095 					else
2096 						local_wild = inp;
2097 			}
2098 		} /* LIST_FOREACH */
2099 
2100 		inp = jail_wild;
2101 		if (inp == NULL)
2102 			inp = local_exact;
2103 		if (inp == NULL)
2104 			inp = local_wild;
2105 #ifdef INET6
2106 		if (inp == NULL)
2107 			inp = local_wild_mapped;
2108 #endif
2109 		if (inp != NULL)
2110 			goto found;
2111 	}
2112 #endif
2113 
2114 	/*
2115 	 * Then look for a wildcard match, if requested.
2116 	 */
2117 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2118 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2119 #ifdef INET6
2120 		struct inpcb *local_wild_mapped = NULL;
2121 #endif
2122 		struct inpcb *jail_wild = NULL;
2123 		struct inpcbhead *head;
2124 		int injail;
2125 
2126 		/*
2127 		 * Order of socket selection - we always prefer jails.
2128 		 *      1. jailed, non-wild.
2129 		 *      2. jailed, wild.
2130 		 *      3. non-jailed, non-wild.
2131 		 *      4. non-jailed, wild.
2132 		 */
2133 		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2134 		    0, pcbinfo->ipi_wildmask)];
2135 		CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2136 #ifdef INET6
2137 			/* XXX inp locking */
2138 			if ((inp->inp_vflag & INP_IPV4) == 0)
2139 				continue;
2140 #endif
2141 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2142 			    inp->inp_lport != lport)
2143 				continue;
2144 
2145 			injail = prison_flag(inp->inp_cred, PR_IP4);
2146 			if (injail) {
2147 				if (prison_check_ip4(inp->inp_cred,
2148 				    &laddr) != 0)
2149 					continue;
2150 			} else {
2151 				if (local_exact != NULL)
2152 					continue;
2153 			}
2154 
2155 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2156 				if (injail)
2157 					goto found;
2158 				else
2159 					local_exact = inp;
2160 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2161 #ifdef INET6
2162 				/* XXX inp locking, NULL check */
2163 				if (inp->inp_vflag & INP_IPV6PROTO)
2164 					local_wild_mapped = inp;
2165 				else
2166 #endif
2167 					if (injail)
2168 						jail_wild = inp;
2169 					else
2170 						local_wild = inp;
2171 			}
2172 		} /* LIST_FOREACH */
2173 		inp = jail_wild;
2174 		if (inp == NULL)
2175 			inp = local_exact;
2176 		if (inp == NULL)
2177 			inp = local_wild;
2178 #ifdef INET6
2179 		if (inp == NULL)
2180 			inp = local_wild_mapped;
2181 #endif
2182 		if (inp != NULL)
2183 			goto found;
2184 	} /* if (lookupflags & INPLOOKUP_WILDCARD) */
2185 	INP_GROUP_UNLOCK(pcbgroup);
2186 	return (NULL);
2187 
2188 found:
2189 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2190 		locked = INP_TRY_WLOCK(inp);
2191 	else if (lookupflags & INPLOOKUP_RLOCKPCB)
2192 		locked = INP_TRY_RLOCK(inp);
2193 	else
2194 		panic("%s: locking bug", __func__);
2195 	if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2196 		if (lookupflags & INPLOOKUP_WLOCKPCB)
2197 			INP_WUNLOCK(inp);
2198 		else
2199 			INP_RUNLOCK(inp);
2200 		return (NULL);
2201 	} else if (!locked)
2202 		in_pcbref(inp);
2203 	INP_GROUP_UNLOCK(pcbgroup);
2204 	if (!locked) {
2205 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2206 			INP_WLOCK(inp);
2207 			if (in_pcbrele_wlocked(inp))
2208 				return (NULL);
2209 		} else {
2210 			INP_RLOCK(inp);
2211 			if (in_pcbrele_rlocked(inp))
2212 				return (NULL);
2213 		}
2214 	}
2215 #ifdef INVARIANTS
2216 	if (lookupflags & INPLOOKUP_WLOCKPCB)
2217 		INP_WLOCK_ASSERT(inp);
2218 	else
2219 		INP_RLOCK_ASSERT(inp);
2220 #endif
2221 	return (inp);
2222 }
2223 #endif /* PCBGROUP */
2224 
2225 /*
2226  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2227  * that the caller has locked the hash list, and will not perform any further
2228  * locking or reference operations on either the hash list or the connection.
2229  */
2230 static struct inpcb *
2231 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2232     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2233     struct ifnet *ifp)
2234 {
2235 	struct inpcbhead *head;
2236 	struct inpcb *inp, *tmpinp;
2237 	u_short fport = fport_arg, lport = lport_arg;
2238 
2239 #ifdef INVARIANTS
2240 	KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2241 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2242 	if (!mtx_owned(&pcbinfo->ipi_hash_lock))
2243 		MPASS(in_epoch_verbose(net_epoch_preempt, 1));
2244 #endif
2245 	/*
2246 	 * First look for an exact match.
2247 	 */
2248 	tmpinp = NULL;
2249 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2250 	    pcbinfo->ipi_hashmask)];
2251 	CK_LIST_FOREACH(inp, head, inp_hash) {
2252 #ifdef INET6
2253 		/* XXX inp locking */
2254 		if ((inp->inp_vflag & INP_IPV4) == 0)
2255 			continue;
2256 #endif
2257 		if (inp->inp_faddr.s_addr == faddr.s_addr &&
2258 		    inp->inp_laddr.s_addr == laddr.s_addr &&
2259 		    inp->inp_fport == fport &&
2260 		    inp->inp_lport == lport) {
2261 			/*
2262 			 * XXX We should be able to directly return
2263 			 * the inp here, without any checks.
2264 			 * Well unless both bound with SO_REUSEPORT?
2265 			 */
2266 			if (prison_flag(inp->inp_cred, PR_IP4))
2267 				return (inp);
2268 			if (tmpinp == NULL)
2269 				tmpinp = inp;
2270 		}
2271 	}
2272 	if (tmpinp != NULL)
2273 		return (tmpinp);
2274 
2275 	/*
2276 	 * Then look in lb group (for wildcard match).
2277 	 */
2278 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2279 		inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2280 		    fport, lookupflags);
2281 		if (inp != NULL)
2282 			return (inp);
2283 	}
2284 
2285 	/*
2286 	 * Then look for a wildcard match, if requested.
2287 	 */
2288 	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2289 		struct inpcb *local_wild = NULL, *local_exact = NULL;
2290 #ifdef INET6
2291 		struct inpcb *local_wild_mapped = NULL;
2292 #endif
2293 		struct inpcb *jail_wild = NULL;
2294 		int injail;
2295 
2296 		/*
2297 		 * Order of socket selection - we always prefer jails.
2298 		 *      1. jailed, non-wild.
2299 		 *      2. jailed, wild.
2300 		 *      3. non-jailed, non-wild.
2301 		 *      4. non-jailed, wild.
2302 		 */
2303 
2304 		head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2305 		    0, pcbinfo->ipi_hashmask)];
2306 		CK_LIST_FOREACH(inp, head, inp_hash) {
2307 #ifdef INET6
2308 			/* XXX inp locking */
2309 			if ((inp->inp_vflag & INP_IPV4) == 0)
2310 				continue;
2311 #endif
2312 			if (inp->inp_faddr.s_addr != INADDR_ANY ||
2313 			    inp->inp_lport != lport)
2314 				continue;
2315 
2316 			injail = prison_flag(inp->inp_cred, PR_IP4);
2317 			if (injail) {
2318 				if (prison_check_ip4(inp->inp_cred,
2319 				    &laddr) != 0)
2320 					continue;
2321 			} else {
2322 				if (local_exact != NULL)
2323 					continue;
2324 			}
2325 
2326 			if (inp->inp_laddr.s_addr == laddr.s_addr) {
2327 				if (injail)
2328 					return (inp);
2329 				else
2330 					local_exact = inp;
2331 			} else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2332 #ifdef INET6
2333 				/* XXX inp locking, NULL check */
2334 				if (inp->inp_vflag & INP_IPV6PROTO)
2335 					local_wild_mapped = inp;
2336 				else
2337 #endif
2338 					if (injail)
2339 						jail_wild = inp;
2340 					else
2341 						local_wild = inp;
2342 			}
2343 		} /* LIST_FOREACH */
2344 		if (jail_wild != NULL)
2345 			return (jail_wild);
2346 		if (local_exact != NULL)
2347 			return (local_exact);
2348 		if (local_wild != NULL)
2349 			return (local_wild);
2350 #ifdef INET6
2351 		if (local_wild_mapped != NULL)
2352 			return (local_wild_mapped);
2353 #endif
2354 	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2355 
2356 	return (NULL);
2357 }
2358 
2359 /*
2360  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2361  * hash list lock, and will return the inpcb locked (i.e., requires
2362  * INPLOOKUP_LOCKPCB).
2363  */
2364 static struct inpcb *
2365 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2366     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2367     struct ifnet *ifp)
2368 {
2369 	struct inpcb *inp;
2370 
2371 	INP_HASH_RLOCK(pcbinfo);
2372 	inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2373 	    (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2374 	if (inp != NULL) {
2375 		if (lookupflags & INPLOOKUP_WLOCKPCB) {
2376 			INP_WLOCK(inp);
2377 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2378 				INP_WUNLOCK(inp);
2379 				inp = NULL;
2380 			}
2381 		} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2382 			INP_RLOCK(inp);
2383 			if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2384 				INP_RUNLOCK(inp);
2385 				inp = NULL;
2386 			}
2387 		} else
2388 			panic("%s: locking bug", __func__);
2389 #ifdef INVARIANTS
2390 		if (inp != NULL) {
2391 			if (lookupflags & INPLOOKUP_WLOCKPCB)
2392 				INP_WLOCK_ASSERT(inp);
2393 			else
2394 				INP_RLOCK_ASSERT(inp);
2395 		}
2396 #endif
2397 	}
2398 	INP_HASH_RUNLOCK(pcbinfo);
2399 	return (inp);
2400 }
2401 
2402 /*
2403  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2404  * from which a pre-calculated hash value may be extracted.
2405  *
2406  * Possibly more of this logic should be in in_pcbgroup.c.
2407  */
2408 struct inpcb *
2409 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2410     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2411 {
2412 #if defined(PCBGROUP) && !defined(RSS)
2413 	struct inpcbgroup *pcbgroup;
2414 #endif
2415 
2416 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2417 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2418 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2419 	    ("%s: LOCKPCB not set", __func__));
2420 
2421 	/*
2422 	 * When not using RSS, use connection groups in preference to the
2423 	 * reservation table when looking up 4-tuples.  When using RSS, just
2424 	 * use the reservation table, due to the cost of the Toeplitz hash
2425 	 * in software.
2426 	 *
2427 	 * XXXRW: This policy belongs in the pcbgroup code, as in principle
2428 	 * we could be doing RSS with a non-Toeplitz hash that is affordable
2429 	 * in software.
2430 	 */
2431 #if defined(PCBGROUP) && !defined(RSS)
2432 	if (in_pcbgroup_enabled(pcbinfo)) {
2433 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2434 		    fport);
2435 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2436 		    laddr, lport, lookupflags, ifp));
2437 	}
2438 #endif
2439 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2440 	    lookupflags, ifp));
2441 }
2442 
2443 struct inpcb *
2444 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2445     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2446     struct ifnet *ifp, struct mbuf *m)
2447 {
2448 #ifdef PCBGROUP
2449 	struct inpcbgroup *pcbgroup;
2450 #endif
2451 
2452 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2453 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
2454 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2455 	    ("%s: LOCKPCB not set", __func__));
2456 
2457 #ifdef PCBGROUP
2458 	/*
2459 	 * If we can use a hardware-generated hash to look up the connection
2460 	 * group, use that connection group to find the inpcb.  Otherwise
2461 	 * fall back on a software hash -- or the reservation table if we're
2462 	 * using RSS.
2463 	 *
2464 	 * XXXRW: As above, that policy belongs in the pcbgroup code.
2465 	 */
2466 	if (in_pcbgroup_enabled(pcbinfo) &&
2467 	    !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2468 		pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2469 		    m->m_pkthdr.flowid);
2470 		if (pcbgroup != NULL)
2471 			return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2472 			    fport, laddr, lport, lookupflags, ifp));
2473 #ifndef RSS
2474 		pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2475 		    fport);
2476 		return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2477 		    laddr, lport, lookupflags, ifp));
2478 #endif
2479 	}
2480 #endif
2481 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2482 	    lookupflags, ifp));
2483 }
2484 #endif /* INET */
2485 
2486 /*
2487  * Insert PCB onto various hash lists.
2488  */
2489 static int
2490 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2491 {
2492 	struct inpcbhead *pcbhash;
2493 	struct inpcbporthead *pcbporthash;
2494 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2495 	struct inpcbport *phd;
2496 	u_int32_t hashkey_faddr;
2497 	int so_options;
2498 
2499 	INP_WLOCK_ASSERT(inp);
2500 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2501 
2502 	KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2503 	    ("in_pcbinshash: INP_INHASHLIST"));
2504 
2505 #ifdef INET6
2506 	if (inp->inp_vflag & INP_IPV6)
2507 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2508 	else
2509 #endif
2510 	hashkey_faddr = inp->inp_faddr.s_addr;
2511 
2512 	pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2513 		 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2514 
2515 	pcbporthash = &pcbinfo->ipi_porthashbase[
2516 	    INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2517 
2518 	/*
2519 	 * Add entry to load balance group.
2520 	 * Only do this if SO_REUSEPORT_LB is set.
2521 	 */
2522 	so_options = inp_so_options(inp);
2523 	if (so_options & SO_REUSEPORT_LB) {
2524 		int ret = in_pcbinslbgrouphash(inp);
2525 		if (ret) {
2526 			/* pcb lb group malloc fail (ret=ENOBUFS). */
2527 			return (ret);
2528 		}
2529 	}
2530 
2531 	/*
2532 	 * Go through port list and look for a head for this lport.
2533 	 */
2534 	CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2535 		if (phd->phd_port == inp->inp_lport)
2536 			break;
2537 	}
2538 	/*
2539 	 * If none exists, malloc one and tack it on.
2540 	 */
2541 	if (phd == NULL) {
2542 		phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2543 		if (phd == NULL) {
2544 			return (ENOBUFS); /* XXX */
2545 		}
2546 		bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2547 		phd->phd_port = inp->inp_lport;
2548 		CK_LIST_INIT(&phd->phd_pcblist);
2549 		CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2550 	}
2551 	inp->inp_phd = phd;
2552 	CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2553 	CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2554 	inp->inp_flags |= INP_INHASHLIST;
2555 #ifdef PCBGROUP
2556 	if (do_pcbgroup_update)
2557 		in_pcbgroup_update(inp);
2558 #endif
2559 	return (0);
2560 }
2561 
2562 /*
2563  * For now, there are two public interfaces to insert an inpcb into the hash
2564  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2565  * is used only in the TCP syncache, where in_pcbinshash is called before the
2566  * full 4-tuple is set for the inpcb, and we don't want to install in the
2567  * pcbgroup until later.
2568  *
2569  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2570  * connection groups, and partially initialised inpcbs should not be exposed
2571  * to either reservation hash tables or pcbgroups.
2572  */
2573 int
2574 in_pcbinshash(struct inpcb *inp)
2575 {
2576 
2577 	return (in_pcbinshash_internal(inp, 1));
2578 }
2579 
2580 int
2581 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2582 {
2583 
2584 	return (in_pcbinshash_internal(inp, 0));
2585 }
2586 
2587 /*
2588  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2589  * changed. NOTE: This does not handle the case of the lport changing (the
2590  * hashed port list would have to be updated as well), so the lport must
2591  * not change after in_pcbinshash() has been called.
2592  */
2593 void
2594 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2595 {
2596 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2597 	struct inpcbhead *head;
2598 	u_int32_t hashkey_faddr;
2599 
2600 	INP_WLOCK_ASSERT(inp);
2601 	INP_HASH_WLOCK_ASSERT(pcbinfo);
2602 
2603 	KASSERT(inp->inp_flags & INP_INHASHLIST,
2604 	    ("in_pcbrehash: !INP_INHASHLIST"));
2605 
2606 #ifdef INET6
2607 	if (inp->inp_vflag & INP_IPV6)
2608 		hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2609 	else
2610 #endif
2611 	hashkey_faddr = inp->inp_faddr.s_addr;
2612 
2613 	head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2614 		inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2615 
2616 	CK_LIST_REMOVE(inp, inp_hash);
2617 	CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2618 
2619 #ifdef PCBGROUP
2620 	if (m != NULL)
2621 		in_pcbgroup_update_mbuf(inp, m);
2622 	else
2623 		in_pcbgroup_update(inp);
2624 #endif
2625 }
2626 
2627 void
2628 in_pcbrehash(struct inpcb *inp)
2629 {
2630 
2631 	in_pcbrehash_mbuf(inp, NULL);
2632 }
2633 
2634 /*
2635  * Remove PCB from various lists.
2636  */
2637 static void
2638 in_pcbremlists(struct inpcb *inp)
2639 {
2640 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2641 
2642 #ifdef INVARIANTS
2643 	if (pcbinfo == &V_tcbinfo) {
2644 		INP_INFO_RLOCK_ASSERT(pcbinfo);
2645 	} else {
2646 		INP_INFO_WLOCK_ASSERT(pcbinfo);
2647 	}
2648 #endif
2649 
2650 	INP_WLOCK_ASSERT(inp);
2651 	INP_LIST_WLOCK_ASSERT(pcbinfo);
2652 
2653 	inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2654 	if (inp->inp_flags & INP_INHASHLIST) {
2655 		struct inpcbport *phd = inp->inp_phd;
2656 
2657 		INP_HASH_WLOCK(pcbinfo);
2658 
2659 		/* XXX: Only do if SO_REUSEPORT_LB set? */
2660 		in_pcbremlbgrouphash(inp);
2661 
2662 		CK_LIST_REMOVE(inp, inp_hash);
2663 		CK_LIST_REMOVE(inp, inp_portlist);
2664 		if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2665 			CK_LIST_REMOVE(phd, phd_hash);
2666 			epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free);
2667 		}
2668 		INP_HASH_WUNLOCK(pcbinfo);
2669 		inp->inp_flags &= ~INP_INHASHLIST;
2670 	}
2671 	CK_LIST_REMOVE(inp, inp_list);
2672 	pcbinfo->ipi_count--;
2673 #ifdef PCBGROUP
2674 	in_pcbgroup_remove(inp);
2675 #endif
2676 }
2677 
2678 /*
2679  * Check for alternatives when higher level complains
2680  * about service problems.  For now, invalidate cached
2681  * routing information.  If the route was created dynamically
2682  * (by a redirect), time to try a default gateway again.
2683  */
2684 void
2685 in_losing(struct inpcb *inp)
2686 {
2687 
2688 	RO_INVALIDATE_CACHE(&inp->inp_route);
2689 	return;
2690 }
2691 
2692 /*
2693  * A set label operation has occurred at the socket layer, propagate the
2694  * label change into the in_pcb for the socket.
2695  */
2696 void
2697 in_pcbsosetlabel(struct socket *so)
2698 {
2699 #ifdef MAC
2700 	struct inpcb *inp;
2701 
2702 	inp = sotoinpcb(so);
2703 	KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2704 
2705 	INP_WLOCK(inp);
2706 	SOCK_LOCK(so);
2707 	mac_inpcb_sosetlabel(so, inp);
2708 	SOCK_UNLOCK(so);
2709 	INP_WUNLOCK(inp);
2710 #endif
2711 }
2712 
2713 /*
2714  * ipport_tick runs once per second, determining if random port allocation
2715  * should be continued.  If more than ipport_randomcps ports have been
2716  * allocated in the last second, then we return to sequential port
2717  * allocation. We return to random allocation only once we drop below
2718  * ipport_randomcps for at least ipport_randomtime seconds.
2719  */
2720 static void
2721 ipport_tick(void *xtp)
2722 {
2723 	VNET_ITERATOR_DECL(vnet_iter);
2724 
2725 	VNET_LIST_RLOCK_NOSLEEP();
2726 	VNET_FOREACH(vnet_iter) {
2727 		CURVNET_SET(vnet_iter);	/* XXX appease INVARIANTS here */
2728 		if (V_ipport_tcpallocs <=
2729 		    V_ipport_tcplastcount + V_ipport_randomcps) {
2730 			if (V_ipport_stoprandom > 0)
2731 				V_ipport_stoprandom--;
2732 		} else
2733 			V_ipport_stoprandom = V_ipport_randomtime;
2734 		V_ipport_tcplastcount = V_ipport_tcpallocs;
2735 		CURVNET_RESTORE();
2736 	}
2737 	VNET_LIST_RUNLOCK_NOSLEEP();
2738 	callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2739 }
2740 
2741 static void
2742 ip_fini(void *xtp)
2743 {
2744 
2745 	callout_stop(&ipport_tick_callout);
2746 }
2747 
2748 /*
2749  * The ipport_callout should start running at about the time we attach the
2750  * inet or inet6 domains.
2751  */
2752 static void
2753 ipport_tick_init(const void *unused __unused)
2754 {
2755 
2756 	/* Start ipport_tick. */
2757 	callout_init(&ipport_tick_callout, 1);
2758 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2759 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2760 		SHUTDOWN_PRI_DEFAULT);
2761 }
2762 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2763     ipport_tick_init, NULL);
2764 
2765 void
2766 inp_wlock(struct inpcb *inp)
2767 {
2768 
2769 	INP_WLOCK(inp);
2770 }
2771 
2772 void
2773 inp_wunlock(struct inpcb *inp)
2774 {
2775 
2776 	INP_WUNLOCK(inp);
2777 }
2778 
2779 void
2780 inp_rlock(struct inpcb *inp)
2781 {
2782 
2783 	INP_RLOCK(inp);
2784 }
2785 
2786 void
2787 inp_runlock(struct inpcb *inp)
2788 {
2789 
2790 	INP_RUNLOCK(inp);
2791 }
2792 
2793 #ifdef INVARIANT_SUPPORT
2794 void
2795 inp_lock_assert(struct inpcb *inp)
2796 {
2797 
2798 	INP_WLOCK_ASSERT(inp);
2799 }
2800 
2801 void
2802 inp_unlock_assert(struct inpcb *inp)
2803 {
2804 
2805 	INP_UNLOCK_ASSERT(inp);
2806 }
2807 #endif
2808 
2809 void
2810 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2811 {
2812 	struct inpcb *inp;
2813 
2814 	INP_INFO_WLOCK(&V_tcbinfo);
2815 	CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2816 		INP_WLOCK(inp);
2817 		func(inp, arg);
2818 		INP_WUNLOCK(inp);
2819 	}
2820 	INP_INFO_WUNLOCK(&V_tcbinfo);
2821 }
2822 
2823 struct socket *
2824 inp_inpcbtosocket(struct inpcb *inp)
2825 {
2826 
2827 	INP_WLOCK_ASSERT(inp);
2828 	return (inp->inp_socket);
2829 }
2830 
2831 struct tcpcb *
2832 inp_inpcbtotcpcb(struct inpcb *inp)
2833 {
2834 
2835 	INP_WLOCK_ASSERT(inp);
2836 	return ((struct tcpcb *)inp->inp_ppcb);
2837 }
2838 
2839 int
2840 inp_ip_tos_get(const struct inpcb *inp)
2841 {
2842 
2843 	return (inp->inp_ip_tos);
2844 }
2845 
2846 void
2847 inp_ip_tos_set(struct inpcb *inp, int val)
2848 {
2849 
2850 	inp->inp_ip_tos = val;
2851 }
2852 
2853 void
2854 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2855     uint32_t *faddr, uint16_t *fp)
2856 {
2857 
2858 	INP_LOCK_ASSERT(inp);
2859 	*laddr = inp->inp_laddr.s_addr;
2860 	*faddr = inp->inp_faddr.s_addr;
2861 	*lp = inp->inp_lport;
2862 	*fp = inp->inp_fport;
2863 }
2864 
2865 struct inpcb *
2866 so_sotoinpcb(struct socket *so)
2867 {
2868 
2869 	return (sotoinpcb(so));
2870 }
2871 
2872 struct tcpcb *
2873 so_sototcpcb(struct socket *so)
2874 {
2875 
2876 	return (sototcpcb(so));
2877 }
2878 
2879 /*
2880  * Create an external-format (``xinpcb'') structure using the information in
2881  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2882  * reduce the spew of irrelevant information over this interface, to isolate
2883  * user code from changes in the kernel structure, and potentially to provide
2884  * information-hiding if we decide that some of this information should be
2885  * hidden from users.
2886  */
2887 void
2888 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2889 {
2890 
2891 	bzero(xi, sizeof(*xi));
2892 	xi->xi_len = sizeof(struct xinpcb);
2893 	if (inp->inp_socket)
2894 		sotoxsocket(inp->inp_socket, &xi->xi_socket);
2895 	bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2896 	xi->inp_gencnt = inp->inp_gencnt;
2897 	xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2898 	xi->inp_flow = inp->inp_flow;
2899 	xi->inp_flowid = inp->inp_flowid;
2900 	xi->inp_flowtype = inp->inp_flowtype;
2901 	xi->inp_flags = inp->inp_flags;
2902 	xi->inp_flags2 = inp->inp_flags2;
2903 	xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2904 	xi->in6p_cksum = inp->in6p_cksum;
2905 	xi->in6p_hops = inp->in6p_hops;
2906 	xi->inp_ip_tos = inp->inp_ip_tos;
2907 	xi->inp_vflag = inp->inp_vflag;
2908 	xi->inp_ip_ttl = inp->inp_ip_ttl;
2909 	xi->inp_ip_p = inp->inp_ip_p;
2910 	xi->inp_ip_minttl = inp->inp_ip_minttl;
2911 }
2912 
2913 #ifdef DDB
2914 static void
2915 db_print_indent(int indent)
2916 {
2917 	int i;
2918 
2919 	for (i = 0; i < indent; i++)
2920 		db_printf(" ");
2921 }
2922 
2923 static void
2924 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2925 {
2926 	char faddr_str[48], laddr_str[48];
2927 
2928 	db_print_indent(indent);
2929 	db_printf("%s at %p\n", name, inc);
2930 
2931 	indent += 2;
2932 
2933 #ifdef INET6
2934 	if (inc->inc_flags & INC_ISIPV6) {
2935 		/* IPv6. */
2936 		ip6_sprintf(laddr_str, &inc->inc6_laddr);
2937 		ip6_sprintf(faddr_str, &inc->inc6_faddr);
2938 	} else
2939 #endif
2940 	{
2941 		/* IPv4. */
2942 		inet_ntoa_r(inc->inc_laddr, laddr_str);
2943 		inet_ntoa_r(inc->inc_faddr, faddr_str);
2944 	}
2945 	db_print_indent(indent);
2946 	db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2947 	    ntohs(inc->inc_lport));
2948 	db_print_indent(indent);
2949 	db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2950 	    ntohs(inc->inc_fport));
2951 }
2952 
2953 static void
2954 db_print_inpflags(int inp_flags)
2955 {
2956 	int comma;
2957 
2958 	comma = 0;
2959 	if (inp_flags & INP_RECVOPTS) {
2960 		db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2961 		comma = 1;
2962 	}
2963 	if (inp_flags & INP_RECVRETOPTS) {
2964 		db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2965 		comma = 1;
2966 	}
2967 	if (inp_flags & INP_RECVDSTADDR) {
2968 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2969 		comma = 1;
2970 	}
2971 	if (inp_flags & INP_ORIGDSTADDR) {
2972 		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2973 		comma = 1;
2974 	}
2975 	if (inp_flags & INP_HDRINCL) {
2976 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
2977 		comma = 1;
2978 	}
2979 	if (inp_flags & INP_HIGHPORT) {
2980 		db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2981 		comma = 1;
2982 	}
2983 	if (inp_flags & INP_LOWPORT) {
2984 		db_printf("%sINP_LOWPORT", comma ? ", " : "");
2985 		comma = 1;
2986 	}
2987 	if (inp_flags & INP_ANONPORT) {
2988 		db_printf("%sINP_ANONPORT", comma ? ", " : "");
2989 		comma = 1;
2990 	}
2991 	if (inp_flags & INP_RECVIF) {
2992 		db_printf("%sINP_RECVIF", comma ? ", " : "");
2993 		comma = 1;
2994 	}
2995 	if (inp_flags & INP_MTUDISC) {
2996 		db_printf("%sINP_MTUDISC", comma ? ", " : "");
2997 		comma = 1;
2998 	}
2999 	if (inp_flags & INP_RECVTTL) {
3000 		db_printf("%sINP_RECVTTL", comma ? ", " : "");
3001 		comma = 1;
3002 	}
3003 	if (inp_flags & INP_DONTFRAG) {
3004 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3005 		comma = 1;
3006 	}
3007 	if (inp_flags & INP_RECVTOS) {
3008 		db_printf("%sINP_RECVTOS", comma ? ", " : "");
3009 		comma = 1;
3010 	}
3011 	if (inp_flags & IN6P_IPV6_V6ONLY) {
3012 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3013 		comma = 1;
3014 	}
3015 	if (inp_flags & IN6P_PKTINFO) {
3016 		db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3017 		comma = 1;
3018 	}
3019 	if (inp_flags & IN6P_HOPLIMIT) {
3020 		db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3021 		comma = 1;
3022 	}
3023 	if (inp_flags & IN6P_HOPOPTS) {
3024 		db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3025 		comma = 1;
3026 	}
3027 	if (inp_flags & IN6P_DSTOPTS) {
3028 		db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3029 		comma = 1;
3030 	}
3031 	if (inp_flags & IN6P_RTHDR) {
3032 		db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3033 		comma = 1;
3034 	}
3035 	if (inp_flags & IN6P_RTHDRDSTOPTS) {
3036 		db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3037 		comma = 1;
3038 	}
3039 	if (inp_flags & IN6P_TCLASS) {
3040 		db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3041 		comma = 1;
3042 	}
3043 	if (inp_flags & IN6P_AUTOFLOWLABEL) {
3044 		db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3045 		comma = 1;
3046 	}
3047 	if (inp_flags & INP_TIMEWAIT) {
3048 		db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3049 		comma  = 1;
3050 	}
3051 	if (inp_flags & INP_ONESBCAST) {
3052 		db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3053 		comma  = 1;
3054 	}
3055 	if (inp_flags & INP_DROPPED) {
3056 		db_printf("%sINP_DROPPED", comma ? ", " : "");
3057 		comma  = 1;
3058 	}
3059 	if (inp_flags & INP_SOCKREF) {
3060 		db_printf("%sINP_SOCKREF", comma ? ", " : "");
3061 		comma  = 1;
3062 	}
3063 	if (inp_flags & IN6P_RFC2292) {
3064 		db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3065 		comma = 1;
3066 	}
3067 	if (inp_flags & IN6P_MTU) {
3068 		db_printf("IN6P_MTU%s", comma ? ", " : "");
3069 		comma = 1;
3070 	}
3071 }
3072 
3073 static void
3074 db_print_inpvflag(u_char inp_vflag)
3075 {
3076 	int comma;
3077 
3078 	comma = 0;
3079 	if (inp_vflag & INP_IPV4) {
3080 		db_printf("%sINP_IPV4", comma ? ", " : "");
3081 		comma  = 1;
3082 	}
3083 	if (inp_vflag & INP_IPV6) {
3084 		db_printf("%sINP_IPV6", comma ? ", " : "");
3085 		comma  = 1;
3086 	}
3087 	if (inp_vflag & INP_IPV6PROTO) {
3088 		db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3089 		comma  = 1;
3090 	}
3091 }
3092 
3093 static void
3094 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3095 {
3096 
3097 	db_print_indent(indent);
3098 	db_printf("%s at %p\n", name, inp);
3099 
3100 	indent += 2;
3101 
3102 	db_print_indent(indent);
3103 	db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3104 
3105 	db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3106 
3107 	db_print_indent(indent);
3108 	db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3109 	    inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3110 
3111 	db_print_indent(indent);
3112 	db_printf("inp_label: %p   inp_flags: 0x%x (",
3113 	   inp->inp_label, inp->inp_flags);
3114 	db_print_inpflags(inp->inp_flags);
3115 	db_printf(")\n");
3116 
3117 	db_print_indent(indent);
3118 	db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3119 	    inp->inp_vflag);
3120 	db_print_inpvflag(inp->inp_vflag);
3121 	db_printf(")\n");
3122 
3123 	db_print_indent(indent);
3124 	db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3125 	    inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3126 
3127 	db_print_indent(indent);
3128 #ifdef INET6
3129 	if (inp->inp_vflag & INP_IPV6) {
3130 		db_printf("in6p_options: %p   in6p_outputopts: %p   "
3131 		    "in6p_moptions: %p\n", inp->in6p_options,
3132 		    inp->in6p_outputopts, inp->in6p_moptions);
3133 		db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3134 		    "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3135 		    inp->in6p_hops);
3136 	} else
3137 #endif
3138 	{
3139 		db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3140 		    "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3141 		    inp->inp_options, inp->inp_moptions);
3142 	}
3143 
3144 	db_print_indent(indent);
3145 	db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3146 	    (uintmax_t)inp->inp_gencnt);
3147 }
3148 
3149 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3150 {
3151 	struct inpcb *inp;
3152 
3153 	if (!have_addr) {
3154 		db_printf("usage: show inpcb <addr>\n");
3155 		return;
3156 	}
3157 	inp = (struct inpcb *)addr;
3158 
3159 	db_print_inpcb(inp, "inpcb", 0);
3160 }
3161 #endif /* DDB */
3162 
3163 #ifdef RATELIMIT
3164 /*
3165  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3166  * if any.
3167  */
3168 int
3169 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3170 {
3171 	union if_snd_tag_modify_params params = {
3172 		.rate_limit.max_rate = max_pacing_rate,
3173 	};
3174 	struct m_snd_tag *mst;
3175 	struct ifnet *ifp;
3176 	int error;
3177 
3178 	mst = inp->inp_snd_tag;
3179 	if (mst == NULL)
3180 		return (EINVAL);
3181 
3182 	ifp = mst->ifp;
3183 	if (ifp == NULL)
3184 		return (EINVAL);
3185 
3186 	if (ifp->if_snd_tag_modify == NULL) {
3187 		error = EOPNOTSUPP;
3188 	} else {
3189 		error = ifp->if_snd_tag_modify(mst, &params);
3190 	}
3191 	return (error);
3192 }
3193 
3194 /*
3195  * Query existing TX rate limit based on the existing
3196  * "inp->inp_snd_tag", if any.
3197  */
3198 int
3199 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3200 {
3201 	union if_snd_tag_query_params params = { };
3202 	struct m_snd_tag *mst;
3203 	struct ifnet *ifp;
3204 	int error;
3205 
3206 	mst = inp->inp_snd_tag;
3207 	if (mst == NULL)
3208 		return (EINVAL);
3209 
3210 	ifp = mst->ifp;
3211 	if (ifp == NULL)
3212 		return (EINVAL);
3213 
3214 	if (ifp->if_snd_tag_query == NULL) {
3215 		error = EOPNOTSUPP;
3216 	} else {
3217 		error = ifp->if_snd_tag_query(mst, &params);
3218 		if (error == 0 &&  p_max_pacing_rate != NULL)
3219 			*p_max_pacing_rate = params.rate_limit.max_rate;
3220 	}
3221 	return (error);
3222 }
3223 
3224 /*
3225  * Query existing TX queue level based on the existing
3226  * "inp->inp_snd_tag", if any.
3227  */
3228 int
3229 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3230 {
3231 	union if_snd_tag_query_params params = { };
3232 	struct m_snd_tag *mst;
3233 	struct ifnet *ifp;
3234 	int error;
3235 
3236 	mst = inp->inp_snd_tag;
3237 	if (mst == NULL)
3238 		return (EINVAL);
3239 
3240 	ifp = mst->ifp;
3241 	if (ifp == NULL)
3242 		return (EINVAL);
3243 
3244 	if (ifp->if_snd_tag_query == NULL)
3245 		return (EOPNOTSUPP);
3246 
3247 	error = ifp->if_snd_tag_query(mst, &params);
3248 	if (error == 0 &&  p_txqueue_level != NULL)
3249 		*p_txqueue_level = params.rate_limit.queue_level;
3250 	return (error);
3251 }
3252 
3253 /*
3254  * Allocate a new TX rate limit send tag from the network interface
3255  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3256  */
3257 int
3258 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3259     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3260 {
3261 	union if_snd_tag_alloc_params params = {
3262 		.rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3263 		    IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3264 		.rate_limit.hdr.flowid = flowid,
3265 		.rate_limit.hdr.flowtype = flowtype,
3266 		.rate_limit.max_rate = max_pacing_rate,
3267 	};
3268 	int error;
3269 
3270 	INP_WLOCK_ASSERT(inp);
3271 
3272 	if (inp->inp_snd_tag != NULL)
3273 		return (EINVAL);
3274 
3275 	if (ifp->if_snd_tag_alloc == NULL) {
3276 		error = EOPNOTSUPP;
3277 	} else {
3278 		error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3279 	}
3280 	return (error);
3281 }
3282 
3283 /*
3284  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3285  * if any:
3286  */
3287 void
3288 in_pcbdetach_txrtlmt(struct inpcb *inp)
3289 {
3290 	struct m_snd_tag *mst;
3291 
3292 	INP_WLOCK_ASSERT(inp);
3293 
3294 	mst = inp->inp_snd_tag;
3295 	inp->inp_snd_tag = NULL;
3296 
3297 	if (mst == NULL)
3298 		return;
3299 
3300 	m_snd_tag_rele(mst);
3301 }
3302 
3303 /*
3304  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3305  * is set in the fast path and will attach/detach/modify the TX rate
3306  * limit send tag based on the socket's so_max_pacing_rate value.
3307  */
3308 void
3309 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3310 {
3311 	struct socket *socket;
3312 	uint32_t max_pacing_rate;
3313 	bool did_upgrade;
3314 	int error;
3315 
3316 	if (inp == NULL)
3317 		return;
3318 
3319 	socket = inp->inp_socket;
3320 	if (socket == NULL)
3321 		return;
3322 
3323 	if (!INP_WLOCKED(inp)) {
3324 		/*
3325 		 * NOTE: If the write locking fails, we need to bail
3326 		 * out and use the non-ratelimited ring for the
3327 		 * transmit until there is a new chance to get the
3328 		 * write lock.
3329 		 */
3330 		if (!INP_TRY_UPGRADE(inp))
3331 			return;
3332 		did_upgrade = 1;
3333 	} else {
3334 		did_upgrade = 0;
3335 	}
3336 
3337 	/*
3338 	 * NOTE: The so_max_pacing_rate value is read unlocked,
3339 	 * because atomic updates are not required since the variable
3340 	 * is checked at every mbuf we send. It is assumed that the
3341 	 * variable read itself will be atomic.
3342 	 */
3343 	max_pacing_rate = socket->so_max_pacing_rate;
3344 
3345 	/*
3346 	 * If the existing send tag is for the wrong interface due to
3347 	 * a route change, first drop the existing tag.  Set the
3348 	 * CHANGED flag so that we will keep trying to allocate a new
3349 	 * tag if we fail to allocate one this time.
3350 	 */
3351 	if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3352 		in_pcbdetach_txrtlmt(inp);
3353 		inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3354 	}
3355 
3356 	/*
3357 	 * NOTE: When attaching to a network interface a reference is
3358 	 * made to ensure the network interface doesn't go away until
3359 	 * all ratelimit connections are gone. The network interface
3360 	 * pointers compared below represent valid network interfaces,
3361 	 * except when comparing towards NULL.
3362 	 */
3363 	if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3364 		error = 0;
3365 	} else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3366 		if (inp->inp_snd_tag != NULL)
3367 			in_pcbdetach_txrtlmt(inp);
3368 		error = 0;
3369 	} else if (inp->inp_snd_tag == NULL) {
3370 		/*
3371 		 * In order to utilize packet pacing with RSS, we need
3372 		 * to wait until there is a valid RSS hash before we
3373 		 * can proceed:
3374 		 */
3375 		if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3376 			error = EAGAIN;
3377 		} else {
3378 			error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3379 			    mb->m_pkthdr.flowid, max_pacing_rate);
3380 		}
3381 	} else {
3382 		error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3383 	}
3384 	if (error == 0 || error == EOPNOTSUPP)
3385 		inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3386 	if (did_upgrade)
3387 		INP_DOWNGRADE(inp);
3388 }
3389 
3390 /*
3391  * Track route changes for TX rate limiting.
3392  */
3393 void
3394 in_pcboutput_eagain(struct inpcb *inp)
3395 {
3396 	bool did_upgrade;
3397 
3398 	if (inp == NULL)
3399 		return;
3400 
3401 	if (inp->inp_snd_tag == NULL)
3402 		return;
3403 
3404 	if (!INP_WLOCKED(inp)) {
3405 		/*
3406 		 * NOTE: If the write locking fails, we need to bail
3407 		 * out and use the non-ratelimited ring for the
3408 		 * transmit until there is a new chance to get the
3409 		 * write lock.
3410 		 */
3411 		if (!INP_TRY_UPGRADE(inp))
3412 			return;
3413 		did_upgrade = 1;
3414 	} else {
3415 		did_upgrade = 0;
3416 	}
3417 
3418 	/* detach rate limiting */
3419 	in_pcbdetach_txrtlmt(inp);
3420 
3421 	/* make sure new mbuf send tag allocation is made */
3422 	inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3423 
3424 	if (did_upgrade)
3425 		INP_DOWNGRADE(inp);
3426 }
3427 #endif /* RATELIMIT */
3428