xref: /freebsd/sys/netinet/udp_usrreq.c (revision 8d8bc0182ea6f8f85409e76e53053487c7d23a84)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2008 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_ipfw.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 
41 #include <sys/param.h>
42 #include <sys/domain.h>
43 #include <sys/eventhandler.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/protosw.h>
52 #include <sys/signalvar.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/systm.h>
59 #include <sys/vimage.h>
60 
61 #include <vm/uma.h>
62 
63 #include <net/if.h>
64 #include <net/route.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #ifdef INET6
72 #include <netinet/ip6.h>
73 #endif
74 #include <netinet/ip_icmp.h>
75 #include <netinet/icmp_var.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_options.h>
78 #ifdef INET6
79 #include <netinet6/ip6_var.h>
80 #endif
81 #include <netinet/udp.h>
82 #include <netinet/udp_var.h>
83 #include <netinet/vinet.h>
84 
85 #ifdef IPSEC
86 #include <netipsec/ipsec.h>
87 #endif
88 
89 #include <machine/in_cksum.h>
90 
91 #include <security/mac/mac_framework.h>
92 
93 /*
94  * UDP protocol implementation.
95  * Per RFC 768, August, 1980.
96  */
97 
98 #ifdef VIMAGE_GLOBALS
99 int	udp_blackhole;
100 #endif
101 
102 /*
103  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
104  * removes the only data integrity mechanism for packets and malformed
105  * packets that would otherwise be discarded due to bad checksums, and may
106  * cause problems (especially for NFS data blocks).
107  */
108 static int	udp_cksum = 1;
109 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
110     0, "compute udp checksum");
111 
112 int	udp_log_in_vain = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114     &udp_log_in_vain, 0, "Log all incoming UDP packets");
115 
116 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_udp, OID_AUTO, blackhole,
117     CTLFLAG_RW, udp_blackhole, 0,
118     "Do not send port unreachables for refused connects");
119 
120 u_long	udp_sendspace = 9216;		/* really max datagram size */
121 					/* 40 1K datagrams */
122 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
123     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
124 
125 u_long	udp_recvspace = 40 * (1024 +
126 #ifdef INET6
127 				      sizeof(struct sockaddr_in6)
128 #else
129 				      sizeof(struct sockaddr_in)
130 #endif
131 				      );
132 
133 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
134     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
135 
136 #ifdef VIMAGE_GLOBALS
137 struct inpcbhead	udb;		/* from udp_var.h */
138 struct inpcbinfo	udbinfo;
139 static uma_zone_t	udpcb_zone;
140 struct udpstat		udpstat;	/* from udp_var.h */
141 #endif
142 
143 #ifndef UDBHASHSIZE
144 #define	UDBHASHSIZE	128
145 #endif
146 
147 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_udp, UDPCTL_STATS, stats,
148     CTLFLAG_RW, udpstat, udpstat,
149     "UDP statistics (struct udpstat, netinet/udp_var.h)");
150 
151 static void	udp_detach(struct socket *so);
152 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
153 		    struct mbuf *, struct thread *);
154 
155 static void
156 udp_zone_change(void *tag)
157 {
158 	INIT_VNET_INET(curvnet);
159 
160 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
161 	uma_zone_set_max(V_udpcb_zone, maxsockets);
162 }
163 
164 static int
165 udp_inpcb_init(void *mem, int size, int flags)
166 {
167 	struct inpcb *inp;
168 
169 	inp = mem;
170 	INP_LOCK_INIT(inp, "inp", "udpinp");
171 	return (0);
172 }
173 
174 void
175 udp_init(void)
176 {
177 	INIT_VNET_INET(curvnet);
178 
179 	V_udp_blackhole = 0;
180 
181 	INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
182 	LIST_INIT(&V_udb);
183 #ifdef VIMAGE
184 	V_udbinfo.ipi_vnet = curvnet;
185 #endif
186 	V_udbinfo.ipi_listhead = &V_udb;
187 	V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
188 	    &V_udbinfo.ipi_hashmask);
189 	V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
190 	    &V_udbinfo.ipi_porthashmask);
191 	V_udbinfo.ipi_zone = uma_zcreate("udp_inpcb", sizeof(struct inpcb),
192 	    NULL, NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
194 
195 	V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
196 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
197 	uma_zone_set_max(V_udpcb_zone, maxsockets);
198 
199 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
200 	    EVENTHANDLER_PRI_ANY);
201 }
202 
203 int
204 udp_newudpcb(struct inpcb *inp)
205 {
206 	INIT_VNET_INET(curvnet);
207 	struct udpcb *up;
208 
209 	up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
210 	if (up == NULL)
211 		return (ENOBUFS);
212 	inp->inp_ppcb = up;
213 	return (0);
214 }
215 
216 void
217 udp_discardcb(struct udpcb *up)
218 {
219 	INIT_VNET_INET(curvnet);
220 
221 	uma_zfree(V_udpcb_zone, up);
222 }
223 
224 #ifdef VIMAGE
225 void
226 udp_destroy(void)
227 {
228 	INIT_VNET_INET(curvnet);
229 
230 	hashdestroy(V_udbinfo.ipi_hashbase, M_PCB,
231 	    V_udbinfo.ipi_hashmask);
232 	hashdestroy(V_udbinfo.ipi_porthashbase, M_PCB,
233 	    V_udbinfo.ipi_porthashmask);
234 	INP_INFO_LOCK_DESTROY(&V_udbinfo);
235 }
236 #endif
237 
238 /*
239  * Subroutine of udp_input(), which appends the provided mbuf chain to the
240  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
241  * contains the source address.  If the socket ends up being an IPv6 socket,
242  * udp_append() will convert to a sockaddr_in6 before passing the address
243  * into the socket code.
244  */
245 static void
246 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
247     struct sockaddr_in *udp_in)
248 {
249 	struct sockaddr *append_sa;
250 	struct socket *so;
251 	struct mbuf *opts = 0;
252 #ifdef INET6
253 	struct sockaddr_in6 udp_in6;
254 #endif
255 
256 	INP_RLOCK_ASSERT(inp);
257 
258 #ifdef IPSEC
259 	/* Check AH/ESP integrity. */
260 	if (ipsec4_in_reject(n, inp)) {
261 		INIT_VNET_IPSEC(curvnet);
262 		m_freem(n);
263 		V_ipsec4stat.in_polvio++;
264 		return;
265 	}
266 #endif /* IPSEC */
267 #ifdef MAC
268 	if (mac_inpcb_check_deliver(inp, n) != 0) {
269 		m_freem(n);
270 		return;
271 	}
272 #endif
273 	if (inp->inp_flags & INP_CONTROLOPTS ||
274 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
275 #ifdef INET6
276 		if (inp->inp_vflag & INP_IPV6)
277 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
278 		else
279 #endif
280 			ip_savecontrol(inp, &opts, ip, n);
281 	}
282 #ifdef INET6
283 	if (inp->inp_vflag & INP_IPV6) {
284 		bzero(&udp_in6, sizeof(udp_in6));
285 		udp_in6.sin6_len = sizeof(udp_in6);
286 		udp_in6.sin6_family = AF_INET6;
287 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
288 		append_sa = (struct sockaddr *)&udp_in6;
289 	} else
290 #endif
291 		append_sa = (struct sockaddr *)udp_in;
292 	m_adj(n, off);
293 
294 	so = inp->inp_socket;
295 	SOCKBUF_LOCK(&so->so_rcv);
296 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
297 		INIT_VNET_INET(so->so_vnet);
298 		SOCKBUF_UNLOCK(&so->so_rcv);
299 		m_freem(n);
300 		if (opts)
301 			m_freem(opts);
302 		UDPSTAT_INC(udps_fullsock);
303 	} else
304 		sorwakeup_locked(so);
305 }
306 
307 void
308 udp_input(struct mbuf *m, int off)
309 {
310 	INIT_VNET_INET(curvnet);
311 	int iphlen = off;
312 	struct ip *ip;
313 	struct udphdr *uh;
314 	struct ifnet *ifp;
315 	struct inpcb *inp;
316 	struct udpcb *up;
317 	int len;
318 	struct ip save_ip;
319 	struct sockaddr_in udp_in;
320 #ifdef IPFIREWALL_FORWARD
321 	struct m_tag *fwd_tag;
322 #endif
323 
324 	ifp = m->m_pkthdr.rcvif;
325 	UDPSTAT_INC(udps_ipackets);
326 
327 	/*
328 	 * Strip IP options, if any; should skip this, make available to
329 	 * user, and use on returned packets, but we don't yet have a way to
330 	 * check the checksum with options still present.
331 	 */
332 	if (iphlen > sizeof (struct ip)) {
333 		ip_stripoptions(m, (struct mbuf *)0);
334 		iphlen = sizeof(struct ip);
335 	}
336 
337 	/*
338 	 * Get IP and UDP header together in first mbuf.
339 	 */
340 	ip = mtod(m, struct ip *);
341 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
342 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
343 			UDPSTAT_INC(udps_hdrops);
344 			return;
345 		}
346 		ip = mtod(m, struct ip *);
347 	}
348 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
349 
350 	/*
351 	 * Destination port of 0 is illegal, based on RFC768.
352 	 */
353 	if (uh->uh_dport == 0)
354 		goto badunlocked;
355 
356 	/*
357 	 * Construct sockaddr format source address.  Stuff source address
358 	 * and datagram in user buffer.
359 	 */
360 	bzero(&udp_in, sizeof(udp_in));
361 	udp_in.sin_len = sizeof(udp_in);
362 	udp_in.sin_family = AF_INET;
363 	udp_in.sin_port = uh->uh_sport;
364 	udp_in.sin_addr = ip->ip_src;
365 
366 	/*
367 	 * Make mbuf data length reflect UDP length.  If not enough data to
368 	 * reflect UDP length, drop.
369 	 */
370 	len = ntohs((u_short)uh->uh_ulen);
371 	if (ip->ip_len != len) {
372 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
373 			UDPSTAT_INC(udps_badlen);
374 			goto badunlocked;
375 		}
376 		m_adj(m, len - ip->ip_len);
377 		/* ip->ip_len = len; */
378 	}
379 
380 	/*
381 	 * Save a copy of the IP header in case we want restore it for
382 	 * sending an ICMP error message in response.
383 	 */
384 	if (!V_udp_blackhole)
385 		save_ip = *ip;
386 	else
387 		memset(&save_ip, 0, sizeof(save_ip));
388 
389 	/*
390 	 * Checksum extended UDP header and data.
391 	 */
392 	if (uh->uh_sum) {
393 		u_short uh_sum;
394 
395 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
396 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
397 				uh_sum = m->m_pkthdr.csum_data;
398 			else
399 				uh_sum = in_pseudo(ip->ip_src.s_addr,
400 				    ip->ip_dst.s_addr, htonl((u_short)len +
401 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
402 			uh_sum ^= 0xffff;
403 		} else {
404 			char b[9];
405 
406 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
407 			bzero(((struct ipovly *)ip)->ih_x1, 9);
408 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
409 			uh_sum = in_cksum(m, len + sizeof (struct ip));
410 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
411 		}
412 		if (uh_sum) {
413 			UDPSTAT_INC(udps_badsum);
414 			m_freem(m);
415 			return;
416 		}
417 	} else
418 		UDPSTAT_INC(udps_nosum);
419 
420 #ifdef IPFIREWALL_FORWARD
421 	/*
422 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
423 	 */
424 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
425 	if (fwd_tag != NULL) {
426 		struct sockaddr_in *next_hop;
427 
428 		/*
429 		 * Do the hack.
430 		 */
431 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
432 		ip->ip_dst = next_hop->sin_addr;
433 		uh->uh_dport = ntohs(next_hop->sin_port);
434 
435 		/*
436 		 * Remove the tag from the packet.  We don't need it anymore.
437 		 */
438 		m_tag_delete(m, fwd_tag);
439 	}
440 #endif
441 
442 	INP_INFO_RLOCK(&V_udbinfo);
443 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
444 	    in_broadcast(ip->ip_dst, ifp)) {
445 		struct inpcb *last;
446 		struct ip_moptions *imo;
447 
448 		last = NULL;
449 		LIST_FOREACH(inp, &V_udb, inp_list) {
450 			if (inp->inp_lport != uh->uh_dport)
451 				continue;
452 #ifdef INET6
453 			if ((inp->inp_vflag & INP_IPV4) == 0)
454 				continue;
455 #endif
456 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
457 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
458 				continue;
459 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
460 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
461 				continue;
462 			if (inp->inp_fport != 0 &&
463 			    inp->inp_fport != uh->uh_sport)
464 				continue;
465 
466 			INP_RLOCK(inp);
467 
468 			/*
469 			 * Handle socket delivery policy for any-source
470 			 * and source-specific multicast. [RFC3678]
471 			 */
472 			imo = inp->inp_moptions;
473 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
474 			    imo != NULL) {
475 				struct sockaddr_in	 group;
476 				int			 blocked;
477 
478 				bzero(&group, sizeof(struct sockaddr_in));
479 				group.sin_len = sizeof(struct sockaddr_in);
480 				group.sin_family = AF_INET;
481 				group.sin_addr = ip->ip_dst;
482 
483 				blocked = imo_multi_filter(imo, ifp,
484 					(struct sockaddr *)&group,
485 					(struct sockaddr *)&udp_in);
486 				if (blocked != MCAST_PASS) {
487 					if (blocked == MCAST_NOTGMEMBER)
488 						IPSTAT_INC(ips_notmember);
489 					if (blocked == MCAST_NOTSMEMBER ||
490 					    blocked == MCAST_MUTED)
491 						UDPSTAT_INC(udps_filtermcast);
492 					INP_RUNLOCK(inp);
493 					continue;
494 				}
495 			}
496 			if (last != NULL) {
497 				struct mbuf *n;
498 
499 				n = m_copy(m, 0, M_COPYALL);
500 				up = intoudpcb(last);
501 				if (up->u_tun_func == NULL) {
502 					if (n != NULL)
503 						udp_append(last,
504 						    ip, n,
505 						    iphlen +
506 						    sizeof(struct udphdr),
507 						    &udp_in);
508 				} else {
509 					/*
510 					 * Engage the tunneling protocol we
511 					 * will have to leave the info_lock
512 					 * up, since we are hunting through
513 					 * multiple UDP's.
514 					 */
515 
516 					(*up->u_tun_func)(n, iphlen, last);
517 				}
518 				INP_RUNLOCK(last);
519 			}
520 			last = inp;
521 			/*
522 			 * Don't look for additional matches if this one does
523 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
524 			 * socket options set.  This heuristic avoids
525 			 * searching through all pcbs in the common case of a
526 			 * non-shared port.  It assumes that an application
527 			 * will never clear these options after setting them.
528 			 */
529 			if ((last->inp_socket->so_options &
530 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
531 				break;
532 		}
533 
534 		if (last == NULL) {
535 			/*
536 			 * No matching pcb found; discard datagram.  (No need
537 			 * to send an ICMP Port Unreachable for a broadcast
538 			 * or multicast datgram.)
539 			 */
540 			UDPSTAT_INC(udps_noportbcast);
541 			goto badheadlocked;
542 		}
543 		up = intoudpcb(last);
544 		if (up->u_tun_func == NULL) {
545 			udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
546 			    &udp_in);
547 		} else {
548 			/*
549 			 * Engage the tunneling protocol.
550 			 */
551 			(*up->u_tun_func)(m, iphlen, last);
552 		}
553 		INP_RUNLOCK(last);
554 		INP_INFO_RUNLOCK(&V_udbinfo);
555 		return;
556 	}
557 
558 	/*
559 	 * Locate pcb for datagram.
560 	 */
561 	inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
562 	    ip->ip_dst, uh->uh_dport, 1, ifp);
563 	if (inp == NULL) {
564 		if (udp_log_in_vain) {
565 			char buf[4*sizeof "123"];
566 
567 			strcpy(buf, inet_ntoa(ip->ip_dst));
568 			log(LOG_INFO,
569 			    "Connection attempt to UDP %s:%d from %s:%d\n",
570 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
571 			    ntohs(uh->uh_sport));
572 		}
573 		UDPSTAT_INC(udps_noport);
574 		if (m->m_flags & (M_BCAST | M_MCAST)) {
575 			UDPSTAT_INC(udps_noportbcast);
576 			goto badheadlocked;
577 		}
578 		if (V_udp_blackhole)
579 			goto badheadlocked;
580 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
581 			goto badheadlocked;
582 		*ip = save_ip;
583 		ip->ip_len += iphlen;
584 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
585 		INP_INFO_RUNLOCK(&V_udbinfo);
586 		return;
587 	}
588 
589 	/*
590 	 * Check the minimum TTL for socket.
591 	 */
592 	INP_RLOCK(inp);
593 	INP_INFO_RUNLOCK(&V_udbinfo);
594 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
595 		INP_RUNLOCK(inp);
596 		goto badunlocked;
597 	}
598 	up = intoudpcb(inp);
599 	if (up->u_tun_func == NULL) {
600 		udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
601 	} else {
602 		/*
603 		 * Engage the tunneling protocol.
604 		 */
605 
606 		(*up->u_tun_func)(m, iphlen, inp);
607 	}
608 	INP_RUNLOCK(inp);
609 	return;
610 
611 badheadlocked:
612 	if (inp)
613 		INP_RUNLOCK(inp);
614 	INP_INFO_RUNLOCK(&V_udbinfo);
615 badunlocked:
616 	m_freem(m);
617 }
618 
619 /*
620  * Notify a udp user of an asynchronous error; just wake up so that they can
621  * collect error status.
622  */
623 struct inpcb *
624 udp_notify(struct inpcb *inp, int errno)
625 {
626 
627 	/*
628 	 * While udp_ctlinput() always calls udp_notify() with a read lock
629 	 * when invoking it directly, in_pcbnotifyall() currently uses write
630 	 * locks due to sharing code with TCP.  For now, accept either a read
631 	 * or a write lock, but a read lock is sufficient.
632 	 */
633 	INP_LOCK_ASSERT(inp);
634 
635 	inp->inp_socket->so_error = errno;
636 	sorwakeup(inp->inp_socket);
637 	sowwakeup(inp->inp_socket);
638 	return (inp);
639 }
640 
641 void
642 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
643 {
644 	INIT_VNET_INET(curvnet);
645 	struct ip *ip = vip;
646 	struct udphdr *uh;
647 	struct in_addr faddr;
648 	struct inpcb *inp;
649 
650 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
651 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
652 		return;
653 
654 	/*
655 	 * Redirects don't need to be handled up here.
656 	 */
657 	if (PRC_IS_REDIRECT(cmd))
658 		return;
659 
660 	/*
661 	 * Hostdead is ugly because it goes linearly through all PCBs.
662 	 *
663 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
664 	 * DoS attack on machines with many connections.
665 	 */
666 	if (cmd == PRC_HOSTDEAD)
667 		ip = NULL;
668 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
669 		return;
670 	if (ip != NULL) {
671 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
672 		INP_INFO_RLOCK(&V_udbinfo);
673 		inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
674 		    ip->ip_src, uh->uh_sport, 0, NULL);
675 		if (inp != NULL) {
676 			INP_RLOCK(inp);
677 			if (inp->inp_socket != NULL) {
678 				udp_notify(inp, inetctlerrmap[cmd]);
679 			}
680 			INP_RUNLOCK(inp);
681 		}
682 		INP_INFO_RUNLOCK(&V_udbinfo);
683 	} else
684 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
685 		    udp_notify);
686 }
687 
688 static int
689 udp_pcblist(SYSCTL_HANDLER_ARGS)
690 {
691 	INIT_VNET_INET(curvnet);
692 	int error, i, n;
693 	struct inpcb *inp, **inp_list;
694 	inp_gen_t gencnt;
695 	struct xinpgen xig;
696 
697 	/*
698 	 * The process of preparing the PCB list is too time-consuming and
699 	 * resource-intensive to repeat twice on every request.
700 	 */
701 	if (req->oldptr == 0) {
702 		n = V_udbinfo.ipi_count;
703 		req->oldidx = 2 * (sizeof xig)
704 			+ (n + n/8) * sizeof(struct xinpcb);
705 		return (0);
706 	}
707 
708 	if (req->newptr != 0)
709 		return (EPERM);
710 
711 	/*
712 	 * OK, now we're committed to doing something.
713 	 */
714 	INP_INFO_RLOCK(&V_udbinfo);
715 	gencnt = V_udbinfo.ipi_gencnt;
716 	n = V_udbinfo.ipi_count;
717 	INP_INFO_RUNLOCK(&V_udbinfo);
718 
719 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
720 		+ n * sizeof(struct xinpcb));
721 	if (error != 0)
722 		return (error);
723 
724 	xig.xig_len = sizeof xig;
725 	xig.xig_count = n;
726 	xig.xig_gen = gencnt;
727 	xig.xig_sogen = so_gencnt;
728 	error = SYSCTL_OUT(req, &xig, sizeof xig);
729 	if (error)
730 		return (error);
731 
732 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
733 	if (inp_list == 0)
734 		return (ENOMEM);
735 
736 	INP_INFO_RLOCK(&V_udbinfo);
737 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
738 	     inp = LIST_NEXT(inp, inp_list)) {
739 		INP_RLOCK(inp);
740 		if (inp->inp_gencnt <= gencnt &&
741 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
742 			inp_list[i++] = inp;
743 		INP_RUNLOCK(inp);
744 	}
745 	INP_INFO_RUNLOCK(&V_udbinfo);
746 	n = i;
747 
748 	error = 0;
749 	for (i = 0; i < n; i++) {
750 		inp = inp_list[i];
751 		INP_RLOCK(inp);
752 		if (inp->inp_gencnt <= gencnt) {
753 			struct xinpcb xi;
754 			bzero(&xi, sizeof(xi));
755 			xi.xi_len = sizeof xi;
756 			/* XXX should avoid extra copy */
757 			bcopy(inp, &xi.xi_inp, sizeof *inp);
758 			if (inp->inp_socket)
759 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
760 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
761 			INP_RUNLOCK(inp);
762 			error = SYSCTL_OUT(req, &xi, sizeof xi);
763 		} else
764 			INP_RUNLOCK(inp);
765 	}
766 	if (!error) {
767 		/*
768 		 * Give the user an updated idea of our state.  If the
769 		 * generation differs from what we told her before, she knows
770 		 * that something happened while we were processing this
771 		 * request, and it might be necessary to retry.
772 		 */
773 		INP_INFO_RLOCK(&V_udbinfo);
774 		xig.xig_gen = V_udbinfo.ipi_gencnt;
775 		xig.xig_sogen = so_gencnt;
776 		xig.xig_count = V_udbinfo.ipi_count;
777 		INP_INFO_RUNLOCK(&V_udbinfo);
778 		error = SYSCTL_OUT(req, &xig, sizeof xig);
779 	}
780 	free(inp_list, M_TEMP);
781 	return (error);
782 }
783 
784 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
785     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
786 
787 static int
788 udp_getcred(SYSCTL_HANDLER_ARGS)
789 {
790 	INIT_VNET_INET(curvnet);
791 	struct xucred xuc;
792 	struct sockaddr_in addrs[2];
793 	struct inpcb *inp;
794 	int error;
795 
796 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
797 	if (error)
798 		return (error);
799 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
800 	if (error)
801 		return (error);
802 	INP_INFO_RLOCK(&V_udbinfo);
803 	inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
804 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
805 	if (inp != NULL) {
806 		INP_RLOCK(inp);
807 		INP_INFO_RUNLOCK(&V_udbinfo);
808 		if (inp->inp_socket == NULL)
809 			error = ENOENT;
810 		if (error == 0)
811 			error = cr_canseeinpcb(req->td->td_ucred, inp);
812 		if (error == 0)
813 			cru2x(inp->inp_cred, &xuc);
814 		INP_RUNLOCK(inp);
815 	} else {
816 		INP_INFO_RUNLOCK(&V_udbinfo);
817 		error = ENOENT;
818 	}
819 	if (error == 0)
820 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
821 	return (error);
822 }
823 
824 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
825     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
826     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
827 
828 static int
829 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
830     struct mbuf *control, struct thread *td)
831 {
832 	INIT_VNET_INET(inp->inp_vnet);
833 	struct udpiphdr *ui;
834 	int len = m->m_pkthdr.len;
835 	struct in_addr faddr, laddr;
836 	struct cmsghdr *cm;
837 	struct sockaddr_in *sin, src;
838 	int error = 0;
839 	int ipflags;
840 	u_short fport, lport;
841 	int unlock_udbinfo;
842 
843 	/*
844 	 * udp_output() may need to temporarily bind or connect the current
845 	 * inpcb.  As such, we don't know up front whether we will need the
846 	 * pcbinfo lock or not.  Do any work to decide what is needed up
847 	 * front before acquiring any locks.
848 	 */
849 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
850 		if (control)
851 			m_freem(control);
852 		m_freem(m);
853 		return (EMSGSIZE);
854 	}
855 
856 	src.sin_family = 0;
857 	if (control != NULL) {
858 		/*
859 		 * XXX: Currently, we assume all the optional information is
860 		 * stored in a single mbuf.
861 		 */
862 		if (control->m_next) {
863 			m_freem(control);
864 			m_freem(m);
865 			return (EINVAL);
866 		}
867 		for (; control->m_len > 0;
868 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
869 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
870 			cm = mtod(control, struct cmsghdr *);
871 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
872 			    || cm->cmsg_len > control->m_len) {
873 				error = EINVAL;
874 				break;
875 			}
876 			if (cm->cmsg_level != IPPROTO_IP)
877 				continue;
878 
879 			switch (cm->cmsg_type) {
880 			case IP_SENDSRCADDR:
881 				if (cm->cmsg_len !=
882 				    CMSG_LEN(sizeof(struct in_addr))) {
883 					error = EINVAL;
884 					break;
885 				}
886 				bzero(&src, sizeof(src));
887 				src.sin_family = AF_INET;
888 				src.sin_len = sizeof(src);
889 				src.sin_port = inp->inp_lport;
890 				src.sin_addr =
891 				    *(struct in_addr *)CMSG_DATA(cm);
892 				break;
893 
894 			default:
895 				error = ENOPROTOOPT;
896 				break;
897 			}
898 			if (error)
899 				break;
900 		}
901 		m_freem(control);
902 	}
903 	if (error) {
904 		m_freem(m);
905 		return (error);
906 	}
907 
908 	/*
909 	 * Depending on whether or not the application has bound or connected
910 	 * the socket, we may have to do varying levels of work.  The optimal
911 	 * case is for a connected UDP socket, as a global lock isn't
912 	 * required at all.
913 	 *
914 	 * In order to decide which we need, we require stability of the
915 	 * inpcb binding, which we ensure by acquiring a read lock on the
916 	 * inpcb.  This doesn't strictly follow the lock order, so we play
917 	 * the trylock and retry game; note that we may end up with more
918 	 * conservative locks than required the second time around, so later
919 	 * assertions have to accept that.  Further analysis of the number of
920 	 * misses under contention is required.
921 	 */
922 	sin = (struct sockaddr_in *)addr;
923 	INP_RLOCK(inp);
924 	if (sin != NULL &&
925 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
926 		INP_RUNLOCK(inp);
927 		INP_INFO_WLOCK(&V_udbinfo);
928 		INP_WLOCK(inp);
929 		unlock_udbinfo = 2;
930 	} else if ((sin != NULL && (
931 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
932 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
933 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
934 	    (inp->inp_lport == 0))) ||
935 	    (src.sin_family == AF_INET)) {
936 		if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
937 			INP_RUNLOCK(inp);
938 			INP_INFO_RLOCK(&V_udbinfo);
939 			INP_RLOCK(inp);
940 		}
941 		unlock_udbinfo = 1;
942 	} else
943 		unlock_udbinfo = 0;
944 
945 	/*
946 	 * If the IP_SENDSRCADDR control message was specified, override the
947 	 * source address for this datagram.  Its use is invalidated if the
948 	 * address thus specified is incomplete or clobbers other inpcbs.
949 	 */
950 	laddr = inp->inp_laddr;
951 	lport = inp->inp_lport;
952 	if (src.sin_family == AF_INET) {
953 		INP_INFO_LOCK_ASSERT(&V_udbinfo);
954 		if ((lport == 0) ||
955 		    (laddr.s_addr == INADDR_ANY &&
956 		     src.sin_addr.s_addr == INADDR_ANY)) {
957 			error = EINVAL;
958 			goto release;
959 		}
960 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
961 		    &laddr.s_addr, &lport, td->td_ucred);
962 		if (error)
963 			goto release;
964 	}
965 
966 	/*
967 	 * If a UDP socket has been connected, then a local address/port will
968 	 * have been selected and bound.
969 	 *
970 	 * If a UDP socket has not been connected to, then an explicit
971 	 * destination address must be used, in which case a local
972 	 * address/port may not have been selected and bound.
973 	 */
974 	if (sin != NULL) {
975 		INP_LOCK_ASSERT(inp);
976 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
977 			error = EISCONN;
978 			goto release;
979 		}
980 
981 		/*
982 		 * Jail may rewrite the destination address, so let it do
983 		 * that before we use it.
984 		 */
985 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
986 		if (error)
987 			goto release;
988 
989 		/*
990 		 * If a local address or port hasn't yet been selected, or if
991 		 * the destination address needs to be rewritten due to using
992 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
993 		 * to do the heavy lifting.  Once a port is selected, we
994 		 * commit the binding back to the socket; we also commit the
995 		 * binding of the address if in jail.
996 		 *
997 		 * If we already have a valid binding and we're not
998 		 * requesting a destination address rewrite, use a fast path.
999 		 */
1000 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
1001 		    inp->inp_lport == 0 ||
1002 		    sin->sin_addr.s_addr == INADDR_ANY ||
1003 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
1004 			INP_INFO_LOCK_ASSERT(&V_udbinfo);
1005 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1006 			    &lport, &faddr.s_addr, &fport, NULL,
1007 			    td->td_ucred);
1008 			if (error)
1009 				goto release;
1010 
1011 			/*
1012 			 * XXXRW: Why not commit the port if the address is
1013 			 * !INADDR_ANY?
1014 			 */
1015 			/* Commit the local port if newly assigned. */
1016 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
1017 			    inp->inp_lport == 0) {
1018 				INP_INFO_WLOCK_ASSERT(&V_udbinfo);
1019 				INP_WLOCK_ASSERT(inp);
1020 				/*
1021 				 * Remember addr if jailed, to prevent
1022 				 * rebinding.
1023 				 */
1024 				if (prison_flag(td->td_ucred, PR_IP4))
1025 					inp->inp_laddr = laddr;
1026 				inp->inp_lport = lport;
1027 				if (in_pcbinshash(inp) != 0) {
1028 					inp->inp_lport = 0;
1029 					error = EAGAIN;
1030 					goto release;
1031 				}
1032 				inp->inp_flags |= INP_ANONPORT;
1033 			}
1034 		} else {
1035 			faddr = sin->sin_addr;
1036 			fport = sin->sin_port;
1037 		}
1038 	} else {
1039 		INP_LOCK_ASSERT(inp);
1040 		faddr = inp->inp_faddr;
1041 		fport = inp->inp_fport;
1042 		if (faddr.s_addr == INADDR_ANY) {
1043 			error = ENOTCONN;
1044 			goto release;
1045 		}
1046 	}
1047 
1048 	/*
1049 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1050 	 * link-layer headers.  Immediate slide the data pointer back forward
1051 	 * since we won't use that space at this layer.
1052 	 */
1053 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1054 	if (m == NULL) {
1055 		error = ENOBUFS;
1056 		goto release;
1057 	}
1058 	m->m_data += max_linkhdr;
1059 	m->m_len -= max_linkhdr;
1060 	m->m_pkthdr.len -= max_linkhdr;
1061 
1062 	/*
1063 	 * Fill in mbuf with extended UDP header and addresses and length put
1064 	 * into network format.
1065 	 */
1066 	ui = mtod(m, struct udpiphdr *);
1067 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1068 	ui->ui_pr = IPPROTO_UDP;
1069 	ui->ui_src = laddr;
1070 	ui->ui_dst = faddr;
1071 	ui->ui_sport = lport;
1072 	ui->ui_dport = fport;
1073 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1074 
1075 	/*
1076 	 * Set the Don't Fragment bit in the IP header.
1077 	 */
1078 	if (inp->inp_flags & INP_DONTFRAG) {
1079 		struct ip *ip;
1080 
1081 		ip = (struct ip *)&ui->ui_i;
1082 		ip->ip_off |= IP_DF;
1083 	}
1084 
1085 	ipflags = 0;
1086 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1087 		ipflags |= IP_ROUTETOIF;
1088 	if (inp->inp_socket->so_options & SO_BROADCAST)
1089 		ipflags |= IP_ALLOWBROADCAST;
1090 	if (inp->inp_flags & INP_ONESBCAST)
1091 		ipflags |= IP_SENDONES;
1092 
1093 #ifdef MAC
1094 	mac_inpcb_create_mbuf(inp, m);
1095 #endif
1096 
1097 	/*
1098 	 * Set up checksum and output datagram.
1099 	 */
1100 	if (udp_cksum) {
1101 		if (inp->inp_flags & INP_ONESBCAST)
1102 			faddr.s_addr = INADDR_BROADCAST;
1103 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1104 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1105 		m->m_pkthdr.csum_flags = CSUM_UDP;
1106 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1107 	} else
1108 		ui->ui_sum = 0;
1109 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1110 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1111 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
1112 	UDPSTAT_INC(udps_opackets);
1113 
1114 	if (unlock_udbinfo == 2)
1115 		INP_INFO_WUNLOCK(&V_udbinfo);
1116 	else if (unlock_udbinfo == 1)
1117 		INP_INFO_RUNLOCK(&V_udbinfo);
1118 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1119 	    inp->inp_moptions, inp);
1120 	if (unlock_udbinfo == 2)
1121 		INP_WUNLOCK(inp);
1122 	else
1123 		INP_RUNLOCK(inp);
1124 	return (error);
1125 
1126 release:
1127 	if (unlock_udbinfo == 2) {
1128 		INP_WUNLOCK(inp);
1129 		INP_INFO_WUNLOCK(&V_udbinfo);
1130 	} else if (unlock_udbinfo == 1) {
1131 		INP_RUNLOCK(inp);
1132 		INP_INFO_RUNLOCK(&V_udbinfo);
1133 	} else
1134 		INP_RUNLOCK(inp);
1135 	m_freem(m);
1136 	return (error);
1137 }
1138 
1139 static void
1140 udp_abort(struct socket *so)
1141 {
1142 	INIT_VNET_INET(so->so_vnet);
1143 	struct inpcb *inp;
1144 
1145 	inp = sotoinpcb(so);
1146 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1147 	INP_INFO_WLOCK(&V_udbinfo);
1148 	INP_WLOCK(inp);
1149 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1150 		in_pcbdisconnect(inp);
1151 		inp->inp_laddr.s_addr = INADDR_ANY;
1152 		soisdisconnected(so);
1153 	}
1154 	INP_WUNLOCK(inp);
1155 	INP_INFO_WUNLOCK(&V_udbinfo);
1156 }
1157 
1158 static int
1159 udp_attach(struct socket *so, int proto, struct thread *td)
1160 {
1161 	INIT_VNET_INET(so->so_vnet);
1162 	struct inpcb *inp;
1163 	int error;
1164 
1165 	inp = sotoinpcb(so);
1166 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1167 	error = soreserve(so, udp_sendspace, udp_recvspace);
1168 	if (error)
1169 		return (error);
1170 	INP_INFO_WLOCK(&V_udbinfo);
1171 	error = in_pcballoc(so, &V_udbinfo);
1172 	if (error) {
1173 		INP_INFO_WUNLOCK(&V_udbinfo);
1174 		return (error);
1175 	}
1176 
1177 	inp = (struct inpcb *)so->so_pcb;
1178 	inp->inp_vflag |= INP_IPV4;
1179 	inp->inp_ip_ttl = V_ip_defttl;
1180 
1181 	error = udp_newudpcb(inp);
1182 	if (error) {
1183 		in_pcbdetach(inp);
1184 		in_pcbfree(inp);
1185 		INP_INFO_WUNLOCK(&V_udbinfo);
1186 		return (error);
1187 	}
1188 
1189 	INP_WUNLOCK(inp);
1190 	INP_INFO_WUNLOCK(&V_udbinfo);
1191 	return (0);
1192 }
1193 
1194 int
1195 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1196 {
1197 	struct inpcb *inp;
1198 	struct udpcb *up;
1199 
1200 	KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram"));
1201 	KASSERT(so->so_pcb != NULL, ("udp_set_kernel_tunneling: NULL inp"));
1202 	if (so->so_type != SOCK_DGRAM) {
1203 		/* Not UDP socket... sorry! */
1204 		return (ENOTSUP);
1205 	}
1206 	inp = (struct inpcb *)so->so_pcb;
1207 	if (inp == NULL) {
1208 		/* NULL INP? */
1209 		return (EINVAL);
1210 	}
1211 	INP_WLOCK(inp);
1212 	up = intoudpcb(inp);
1213 	if (up->u_tun_func != NULL) {
1214 		INP_WUNLOCK(inp);
1215 		return (EBUSY);
1216 	}
1217 	up->u_tun_func = f;
1218 	INP_WUNLOCK(inp);
1219 	return (0);
1220 }
1221 
1222 static int
1223 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1224 {
1225 	INIT_VNET_INET(so->so_vnet);
1226 	struct inpcb *inp;
1227 	int error;
1228 
1229 	inp = sotoinpcb(so);
1230 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1231 	INP_INFO_WLOCK(&V_udbinfo);
1232 	INP_WLOCK(inp);
1233 	error = in_pcbbind(inp, nam, td->td_ucred);
1234 	INP_WUNLOCK(inp);
1235 	INP_INFO_WUNLOCK(&V_udbinfo);
1236 	return (error);
1237 }
1238 
1239 static void
1240 udp_close(struct socket *so)
1241 {
1242 	INIT_VNET_INET(so->so_vnet);
1243 	struct inpcb *inp;
1244 
1245 	inp = sotoinpcb(so);
1246 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1247 	INP_INFO_WLOCK(&V_udbinfo);
1248 	INP_WLOCK(inp);
1249 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1250 		in_pcbdisconnect(inp);
1251 		inp->inp_laddr.s_addr = INADDR_ANY;
1252 		soisdisconnected(so);
1253 	}
1254 	INP_WUNLOCK(inp);
1255 	INP_INFO_WUNLOCK(&V_udbinfo);
1256 }
1257 
1258 static int
1259 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1260 {
1261 	INIT_VNET_INET(so->so_vnet);
1262 	struct inpcb *inp;
1263 	int error;
1264 	struct sockaddr_in *sin;
1265 
1266 	inp = sotoinpcb(so);
1267 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1268 	INP_INFO_WLOCK(&V_udbinfo);
1269 	INP_WLOCK(inp);
1270 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1271 		INP_WUNLOCK(inp);
1272 		INP_INFO_WUNLOCK(&V_udbinfo);
1273 		return (EISCONN);
1274 	}
1275 	sin = (struct sockaddr_in *)nam;
1276 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1277 	if (error != 0) {
1278 		INP_WUNLOCK(inp);
1279 		INP_INFO_WUNLOCK(&V_udbinfo);
1280 		return (error);
1281 	}
1282 	error = in_pcbconnect(inp, nam, td->td_ucred);
1283 	if (error == 0)
1284 		soisconnected(so);
1285 	INP_WUNLOCK(inp);
1286 	INP_INFO_WUNLOCK(&V_udbinfo);
1287 	return (error);
1288 }
1289 
1290 static void
1291 udp_detach(struct socket *so)
1292 {
1293 	INIT_VNET_INET(so->so_vnet);
1294 	struct inpcb *inp;
1295 	struct udpcb *up;
1296 
1297 	inp = sotoinpcb(so);
1298 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1299 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1300 	    ("udp_detach: not disconnected"));
1301 	INP_INFO_WLOCK(&V_udbinfo);
1302 	INP_WLOCK(inp);
1303 	up = intoudpcb(inp);
1304 	KASSERT(up != NULL, ("%s: up == NULL", __func__));
1305 	inp->inp_ppcb = NULL;
1306 	in_pcbdetach(inp);
1307 	in_pcbfree(inp);
1308 	INP_INFO_WUNLOCK(&V_udbinfo);
1309 	udp_discardcb(up);
1310 }
1311 
1312 static int
1313 udp_disconnect(struct socket *so)
1314 {
1315 	INIT_VNET_INET(so->so_vnet);
1316 	struct inpcb *inp;
1317 
1318 	inp = sotoinpcb(so);
1319 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1320 	INP_INFO_WLOCK(&V_udbinfo);
1321 	INP_WLOCK(inp);
1322 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1323 		INP_WUNLOCK(inp);
1324 		INP_INFO_WUNLOCK(&V_udbinfo);
1325 		return (ENOTCONN);
1326 	}
1327 
1328 	in_pcbdisconnect(inp);
1329 	inp->inp_laddr.s_addr = INADDR_ANY;
1330 	SOCK_LOCK(so);
1331 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1332 	SOCK_UNLOCK(so);
1333 	INP_WUNLOCK(inp);
1334 	INP_INFO_WUNLOCK(&V_udbinfo);
1335 	return (0);
1336 }
1337 
1338 static int
1339 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1340     struct mbuf *control, struct thread *td)
1341 {
1342 	struct inpcb *inp;
1343 
1344 	inp = sotoinpcb(so);
1345 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1346 	return (udp_output(inp, m, addr, control, td));
1347 }
1348 
1349 int
1350 udp_shutdown(struct socket *so)
1351 {
1352 	struct inpcb *inp;
1353 
1354 	inp = sotoinpcb(so);
1355 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1356 	INP_WLOCK(inp);
1357 	socantsendmore(so);
1358 	INP_WUNLOCK(inp);
1359 	return (0);
1360 }
1361 
1362 struct pr_usrreqs udp_usrreqs = {
1363 	.pru_abort =		udp_abort,
1364 	.pru_attach =		udp_attach,
1365 	.pru_bind =		udp_bind,
1366 	.pru_connect =		udp_connect,
1367 	.pru_control =		in_control,
1368 	.pru_detach =		udp_detach,
1369 	.pru_disconnect =	udp_disconnect,
1370 	.pru_peeraddr =		in_getpeeraddr,
1371 	.pru_send =		udp_send,
1372 	.pru_soreceive =	soreceive_dgram,
1373 	.pru_sosend =		sosend_dgram,
1374 	.pru_shutdown =		udp_shutdown,
1375 	.pru_sockaddr =		in_getsockaddr,
1376 	.pru_sosetlabel =	in_pcbsosetlabel,
1377 	.pru_close =		udp_close,
1378 };
1379