xref: /freebsd/sys/netinet/udp_usrreq.c (revision 8957464be7644ab4da9559ab479c9a996fbcbd29)
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 #include "opt_mac.h"
41 
42 #include <sys/param.h>
43 #include <sys/domain.h>
44 #include <sys/eventhandler.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/signalvar.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/systm.h>
60 #include <sys/vimage.h>
61 
62 #include <vm/uma.h>
63 
64 #include <net/if.h>
65 #include <net/route.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip.h>
72 #ifdef INET6
73 #include <netinet/ip6.h>
74 #endif
75 #include <netinet/ip_icmp.h>
76 #include <netinet/icmp_var.h>
77 #include <netinet/ip_var.h>
78 #include <netinet/ip_options.h>
79 #ifdef INET6
80 #include <netinet6/ip6_var.h>
81 #endif
82 #include <netinet/udp.h>
83 #include <netinet/udp_var.h>
84 #include <netinet/vinet.h>
85 
86 #ifdef IPSEC
87 #include <netipsec/ipsec.h>
88 #endif
89 
90 #include <machine/in_cksum.h>
91 
92 #include <security/mac/mac_framework.h>
93 
94 /*
95  * UDP protocol implementation.
96  * Per RFC 768, August, 1980.
97  */
98 
99 #ifdef VIMAGE_GLOBALS
100 int	udp_blackhole;
101 #endif
102 
103 /*
104  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
105  * removes the only data integrity mechanism for packets and malformed
106  * packets that would otherwise be discarded due to bad checksums, and may
107  * cause problems (especially for NFS data blocks).
108  */
109 static int	udp_cksum = 1;
110 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
111     0, "compute udp checksum");
112 
113 int	udp_log_in_vain = 0;
114 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
115     &udp_log_in_vain, 0, "Log all incoming UDP packets");
116 
117 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_udp, OID_AUTO, blackhole,
118     CTLFLAG_RW, udp_blackhole, 0,
119     "Do not send port unreachables for refused connects");
120 
121 u_long	udp_sendspace = 9216;		/* really max datagram size */
122 					/* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
125 
126 u_long	udp_recvspace = 40 * (1024 +
127 #ifdef INET6
128 				      sizeof(struct sockaddr_in6)
129 #else
130 				      sizeof(struct sockaddr_in)
131 #endif
132 				      );
133 
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
136 
137 #ifdef VIMAGE_GLOBALS
138 struct inpcbhead	udb;		/* from udp_var.h */
139 struct inpcbinfo	udbinfo;
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 }
162 
163 static int
164 udp_inpcb_init(void *mem, int size, int flags)
165 {
166 	struct inpcb *inp;
167 
168 	inp = mem;
169 	INP_LOCK_INIT(inp, "inp", "udpinp");
170 	return (0);
171 }
172 
173 void
174 udp_init(void)
175 {
176 	INIT_VNET_INET(curvnet);
177 
178 	V_udp_blackhole = 0;
179 
180 	INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
181 	LIST_INIT(&V_udb);
182 	V_udbinfo.ipi_listhead = &V_udb;
183 	V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
184 	    &V_udbinfo.ipi_hashmask);
185 	V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
186 	    &V_udbinfo.ipi_porthashmask);
187 	V_udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
188 	    NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
189 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
190 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
191 	    EVENTHANDLER_PRI_ANY);
192 }
193 
194 /*
195  * Subroutine of udp_input(), which appends the provided mbuf chain to the
196  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
197  * contains the source address.  If the socket ends up being an IPv6 socket,
198  * udp_append() will convert to a sockaddr_in6 before passing the address
199  * into the socket code.
200  */
201 static void
202 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
203     struct sockaddr_in *udp_in)
204 {
205 	struct sockaddr *append_sa;
206 	struct socket *so;
207 	struct mbuf *opts = 0;
208 #ifdef INET6
209 	struct sockaddr_in6 udp_in6;
210 #endif
211 
212 	INP_RLOCK_ASSERT(inp);
213 
214 #ifdef IPSEC
215 	/* Check AH/ESP integrity. */
216 	if (ipsec4_in_reject(n, inp)) {
217 		INIT_VNET_IPSEC(curvnet);
218 		m_freem(n);
219 		V_ipsec4stat.in_polvio++;
220 		return;
221 	}
222 #endif /* IPSEC */
223 #ifdef MAC
224 	if (mac_inpcb_check_deliver(inp, n) != 0) {
225 		m_freem(n);
226 		return;
227 	}
228 #endif
229 	if (inp->inp_flags & INP_CONTROLOPTS ||
230 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
231 #ifdef INET6
232 		if (inp->inp_vflag & INP_IPV6)
233 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
234 		else
235 #endif
236 			ip_savecontrol(inp, &opts, ip, n);
237 	}
238 #ifdef INET6
239 	if (inp->inp_vflag & INP_IPV6) {
240 		bzero(&udp_in6, sizeof(udp_in6));
241 		udp_in6.sin6_len = sizeof(udp_in6);
242 		udp_in6.sin6_family = AF_INET6;
243 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
244 		append_sa = (struct sockaddr *)&udp_in6;
245 	} else
246 #endif
247 		append_sa = (struct sockaddr *)udp_in;
248 	m_adj(n, off);
249 
250 	so = inp->inp_socket;
251 	SOCKBUF_LOCK(&so->so_rcv);
252 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
253 		INIT_VNET_INET(so->so_vnet);
254 		SOCKBUF_UNLOCK(&so->so_rcv);
255 		m_freem(n);
256 		if (opts)
257 			m_freem(opts);
258 		UDPSTAT_INC(udps_fullsock);
259 	} else
260 		sorwakeup_locked(so);
261 }
262 
263 void
264 udp_input(struct mbuf *m, int off)
265 {
266 	INIT_VNET_INET(curvnet);
267 	int iphlen = off;
268 	struct ip *ip;
269 	struct udphdr *uh;
270 	struct ifnet *ifp;
271 	struct inpcb *inp;
272 	int len;
273 	struct ip save_ip;
274 	struct sockaddr_in udp_in;
275 #ifdef IPFIREWALL_FORWARD
276 	struct m_tag *fwd_tag;
277 #endif
278 
279 	ifp = m->m_pkthdr.rcvif;
280 	UDPSTAT_INC(udps_ipackets);
281 
282 	/*
283 	 * Strip IP options, if any; should skip this, make available to
284 	 * user, and use on returned packets, but we don't yet have a way to
285 	 * check the checksum with options still present.
286 	 */
287 	if (iphlen > sizeof (struct ip)) {
288 		ip_stripoptions(m, (struct mbuf *)0);
289 		iphlen = sizeof(struct ip);
290 	}
291 
292 	/*
293 	 * Get IP and UDP header together in first mbuf.
294 	 */
295 	ip = mtod(m, struct ip *);
296 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
297 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
298 			UDPSTAT_INC(udps_hdrops);
299 			return;
300 		}
301 		ip = mtod(m, struct ip *);
302 	}
303 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
304 
305 	/*
306 	 * Destination port of 0 is illegal, based on RFC768.
307 	 */
308 	if (uh->uh_dport == 0)
309 		goto badunlocked;
310 
311 	/*
312 	 * Construct sockaddr format source address.  Stuff source address
313 	 * and datagram in user buffer.
314 	 */
315 	bzero(&udp_in, sizeof(udp_in));
316 	udp_in.sin_len = sizeof(udp_in);
317 	udp_in.sin_family = AF_INET;
318 	udp_in.sin_port = uh->uh_sport;
319 	udp_in.sin_addr = ip->ip_src;
320 
321 	/*
322 	 * Make mbuf data length reflect UDP length.  If not enough data to
323 	 * reflect UDP length, drop.
324 	 */
325 	len = ntohs((u_short)uh->uh_ulen);
326 	if (ip->ip_len != len) {
327 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
328 			UDPSTAT_INC(udps_badlen);
329 			goto badunlocked;
330 		}
331 		m_adj(m, len - ip->ip_len);
332 		/* ip->ip_len = len; */
333 	}
334 
335 	/*
336 	 * Save a copy of the IP header in case we want restore it for
337 	 * sending an ICMP error message in response.
338 	 */
339 	if (!V_udp_blackhole)
340 		save_ip = *ip;
341 	else
342 		memset(&save_ip, 0, sizeof(save_ip));
343 
344 	/*
345 	 * Checksum extended UDP header and data.
346 	 */
347 	if (uh->uh_sum) {
348 		u_short uh_sum;
349 
350 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
351 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
352 				uh_sum = m->m_pkthdr.csum_data;
353 			else
354 				uh_sum = in_pseudo(ip->ip_src.s_addr,
355 				    ip->ip_dst.s_addr, htonl((u_short)len +
356 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
357 			uh_sum ^= 0xffff;
358 		} else {
359 			char b[9];
360 
361 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
362 			bzero(((struct ipovly *)ip)->ih_x1, 9);
363 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
364 			uh_sum = in_cksum(m, len + sizeof (struct ip));
365 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
366 		}
367 		if (uh_sum) {
368 			UDPSTAT_INC(udps_badsum);
369 			m_freem(m);
370 			return;
371 		}
372 	} else
373 		UDPSTAT_INC(udps_nosum);
374 
375 #ifdef IPFIREWALL_FORWARD
376 	/*
377 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
378 	 */
379 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
380 	if (fwd_tag != NULL) {
381 		struct sockaddr_in *next_hop;
382 
383 		/*
384 		 * Do the hack.
385 		 */
386 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
387 		ip->ip_dst = next_hop->sin_addr;
388 		uh->uh_dport = ntohs(next_hop->sin_port);
389 
390 		/*
391 		 * Remove the tag from the packet.  We don't need it anymore.
392 		 */
393 		m_tag_delete(m, fwd_tag);
394 	}
395 #endif
396 
397 	INP_INFO_RLOCK(&V_udbinfo);
398 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
399 	    in_broadcast(ip->ip_dst, ifp)) {
400 		struct inpcb *last;
401 		struct ip_moptions *imo;
402 
403 		last = NULL;
404 		LIST_FOREACH(inp, &V_udb, inp_list) {
405 			if (inp->inp_lport != uh->uh_dport)
406 				continue;
407 #ifdef INET6
408 			if ((inp->inp_vflag & INP_IPV4) == 0)
409 				continue;
410 #endif
411 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
412 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
413 				continue;
414 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
415 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
416 				continue;
417 			if (inp->inp_fport != 0 &&
418 			    inp->inp_fport != uh->uh_sport)
419 				continue;
420 
421 			INP_RLOCK(inp);
422 
423 			/*
424 			 * Handle socket delivery policy for any-source
425 			 * and source-specific multicast. [RFC3678]
426 			 */
427 			imo = inp->inp_moptions;
428 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
429 			    imo != NULL) {
430 				struct sockaddr_in	 group;
431 				int			 blocked;
432 
433 				bzero(&group, sizeof(struct sockaddr_in));
434 				group.sin_len = sizeof(struct sockaddr_in);
435 				group.sin_family = AF_INET;
436 				group.sin_addr = ip->ip_dst;
437 
438 				blocked = imo_multi_filter(imo, ifp,
439 					(struct sockaddr *)&group,
440 					(struct sockaddr *)&udp_in);
441 				if (blocked != MCAST_PASS) {
442 					if (blocked == MCAST_NOTGMEMBER)
443 						IPSTAT_INC(ips_notmember);
444 					if (blocked == MCAST_NOTSMEMBER ||
445 					    blocked == MCAST_MUTED)
446 						UDPSTAT_INC(udps_filtermcast);
447 					INP_RUNLOCK(inp);
448 					continue;
449 				}
450 			}
451 			if (last != NULL) {
452 				struct mbuf *n;
453 
454 				n = m_copy(m, 0, M_COPYALL);
455 				if (last->inp_ppcb == NULL) {
456 					if (n != NULL)
457 						udp_append(last,
458 						    ip, n,
459 						    iphlen +
460 						    sizeof(struct udphdr),
461 						    &udp_in);
462 					INP_RUNLOCK(last);
463 				} else {
464 					/*
465 					 * Engage the tunneling protocol we
466 					 * will have to leave the info_lock
467 					 * up, since we are hunting through
468 					 * multiple UDP's.
469 					 *
470 					 */
471 					udp_tun_func_t tunnel_func;
472 
473 					tunnel_func = (udp_tun_func_t)last->inp_ppcb;
474 					tunnel_func(n, iphlen, last);
475 					INP_RUNLOCK(last);
476 				}
477 			}
478 			last = inp;
479 			/*
480 			 * Don't look for additional matches if this one does
481 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
482 			 * socket options set.  This heuristic avoids
483 			 * searching through all pcbs in the common case of a
484 			 * non-shared port.  It assumes that an application
485 			 * will never clear these options after setting them.
486 			 */
487 			if ((last->inp_socket->so_options &
488 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
489 				break;
490 		}
491 
492 		if (last == NULL) {
493 			/*
494 			 * No matching pcb found; discard datagram.  (No need
495 			 * to send an ICMP Port Unreachable for a broadcast
496 			 * or multicast datgram.)
497 			 */
498 			UDPSTAT_INC(udps_noportbcast);
499 			goto badheadlocked;
500 		}
501 		if (last->inp_ppcb == NULL) {
502 			udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
503 			    &udp_in);
504 			INP_RUNLOCK(last);
505 			INP_INFO_RUNLOCK(&V_udbinfo);
506 		} else {
507 			/*
508 			 * Engage the tunneling protocol.
509 			 */
510 			udp_tun_func_t tunnel_func;
511 
512 			tunnel_func = (udp_tun_func_t)last->inp_ppcb;
513 			tunnel_func(m, iphlen, last);
514 			INP_RUNLOCK(last);
515 			INP_INFO_RUNLOCK(&V_udbinfo);
516 		}
517 		return;
518 	}
519 
520 	/*
521 	 * Locate pcb for datagram.
522 	 */
523 	inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
524 	    ip->ip_dst, uh->uh_dport, 1, ifp);
525 	if (inp == NULL) {
526 		if (udp_log_in_vain) {
527 			char buf[4*sizeof "123"];
528 
529 			strcpy(buf, inet_ntoa(ip->ip_dst));
530 			log(LOG_INFO,
531 			    "Connection attempt to UDP %s:%d from %s:%d\n",
532 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
533 			    ntohs(uh->uh_sport));
534 		}
535 		UDPSTAT_INC(udps_noport);
536 		if (m->m_flags & (M_BCAST | M_MCAST)) {
537 			UDPSTAT_INC(udps_noportbcast);
538 			goto badheadlocked;
539 		}
540 		if (V_udp_blackhole)
541 			goto badheadlocked;
542 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
543 			goto badheadlocked;
544 		*ip = save_ip;
545 		ip->ip_len += iphlen;
546 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
547 		INP_INFO_RUNLOCK(&V_udbinfo);
548 		return;
549 	}
550 
551 	/*
552 	 * Check the minimum TTL for socket.
553 	 */
554 	INP_RLOCK(inp);
555 	INP_INFO_RUNLOCK(&V_udbinfo);
556 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
557 		INP_RUNLOCK(inp);
558 		goto badunlocked;
559 	}
560 	if (inp->inp_ppcb != NULL) {
561 		/*
562 		 * Engage the tunneling protocol.
563 		 */
564 		udp_tun_func_t tunnel_func;
565 
566 		tunnel_func = (udp_tun_func_t)inp->inp_ppcb;
567 		tunnel_func(m, iphlen, inp);
568 		INP_RUNLOCK(inp);
569 		return;
570 	}
571 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
572 	INP_RUNLOCK(inp);
573 	return;
574 
575 badheadlocked:
576 	if (inp)
577 		INP_RUNLOCK(inp);
578 	INP_INFO_RUNLOCK(&V_udbinfo);
579 badunlocked:
580 	m_freem(m);
581 }
582 
583 /*
584  * Notify a udp user of an asynchronous error; just wake up so that they can
585  * collect error status.
586  */
587 struct inpcb *
588 udp_notify(struct inpcb *inp, int errno)
589 {
590 
591 	/*
592 	 * While udp_ctlinput() always calls udp_notify() with a read lock
593 	 * when invoking it directly, in_pcbnotifyall() currently uses write
594 	 * locks due to sharing code with TCP.  For now, accept either a read
595 	 * or a write lock, but a read lock is sufficient.
596 	 */
597 	INP_LOCK_ASSERT(inp);
598 
599 	inp->inp_socket->so_error = errno;
600 	sorwakeup(inp->inp_socket);
601 	sowwakeup(inp->inp_socket);
602 	return (inp);
603 }
604 
605 void
606 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
607 {
608 	INIT_VNET_INET(curvnet);
609 	struct ip *ip = vip;
610 	struct udphdr *uh;
611 	struct in_addr faddr;
612 	struct inpcb *inp;
613 
614 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
615 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
616 		return;
617 
618 	/*
619 	 * Redirects don't need to be handled up here.
620 	 */
621 	if (PRC_IS_REDIRECT(cmd))
622 		return;
623 
624 	/*
625 	 * Hostdead is ugly because it goes linearly through all PCBs.
626 	 *
627 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
628 	 * DoS attack on machines with many connections.
629 	 */
630 	if (cmd == PRC_HOSTDEAD)
631 		ip = NULL;
632 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
633 		return;
634 	if (ip != NULL) {
635 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
636 		INP_INFO_RLOCK(&V_udbinfo);
637 		inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
638 		    ip->ip_src, uh->uh_sport, 0, NULL);
639 		if (inp != NULL) {
640 			INP_RLOCK(inp);
641 			if (inp->inp_socket != NULL) {
642 				udp_notify(inp, inetctlerrmap[cmd]);
643 			}
644 			INP_RUNLOCK(inp);
645 		}
646 		INP_INFO_RUNLOCK(&V_udbinfo);
647 	} else
648 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
649 		    udp_notify);
650 }
651 
652 static int
653 udp_pcblist(SYSCTL_HANDLER_ARGS)
654 {
655 	INIT_VNET_INET(curvnet);
656 	int error, i, n;
657 	struct inpcb *inp, **inp_list;
658 	inp_gen_t gencnt;
659 	struct xinpgen xig;
660 
661 	/*
662 	 * The process of preparing the PCB list is too time-consuming and
663 	 * resource-intensive to repeat twice on every request.
664 	 */
665 	if (req->oldptr == 0) {
666 		n = V_udbinfo.ipi_count;
667 		req->oldidx = 2 * (sizeof xig)
668 			+ (n + n/8) * sizeof(struct xinpcb);
669 		return (0);
670 	}
671 
672 	if (req->newptr != 0)
673 		return (EPERM);
674 
675 	/*
676 	 * OK, now we're committed to doing something.
677 	 */
678 	INP_INFO_RLOCK(&V_udbinfo);
679 	gencnt = V_udbinfo.ipi_gencnt;
680 	n = V_udbinfo.ipi_count;
681 	INP_INFO_RUNLOCK(&V_udbinfo);
682 
683 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
684 		+ n * sizeof(struct xinpcb));
685 	if (error != 0)
686 		return (error);
687 
688 	xig.xig_len = sizeof xig;
689 	xig.xig_count = n;
690 	xig.xig_gen = gencnt;
691 	xig.xig_sogen = so_gencnt;
692 	error = SYSCTL_OUT(req, &xig, sizeof xig);
693 	if (error)
694 		return (error);
695 
696 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
697 	if (inp_list == 0)
698 		return (ENOMEM);
699 
700 	INP_INFO_RLOCK(&V_udbinfo);
701 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
702 	     inp = LIST_NEXT(inp, inp_list)) {
703 		INP_RLOCK(inp);
704 		if (inp->inp_gencnt <= gencnt &&
705 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
706 			inp_list[i++] = inp;
707 		INP_RUNLOCK(inp);
708 	}
709 	INP_INFO_RUNLOCK(&V_udbinfo);
710 	n = i;
711 
712 	error = 0;
713 	for (i = 0; i < n; i++) {
714 		inp = inp_list[i];
715 		INP_RLOCK(inp);
716 		if (inp->inp_gencnt <= gencnt) {
717 			struct xinpcb xi;
718 			bzero(&xi, sizeof(xi));
719 			xi.xi_len = sizeof xi;
720 			/* XXX should avoid extra copy */
721 			bcopy(inp, &xi.xi_inp, sizeof *inp);
722 			if (inp->inp_socket)
723 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
724 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
725 			INP_RUNLOCK(inp);
726 			error = SYSCTL_OUT(req, &xi, sizeof xi);
727 		} else
728 			INP_RUNLOCK(inp);
729 	}
730 	if (!error) {
731 		/*
732 		 * Give the user an updated idea of our state.  If the
733 		 * generation differs from what we told her before, she knows
734 		 * that something happened while we were processing this
735 		 * request, and it might be necessary to retry.
736 		 */
737 		INP_INFO_RLOCK(&V_udbinfo);
738 		xig.xig_gen = V_udbinfo.ipi_gencnt;
739 		xig.xig_sogen = so_gencnt;
740 		xig.xig_count = V_udbinfo.ipi_count;
741 		INP_INFO_RUNLOCK(&V_udbinfo);
742 		error = SYSCTL_OUT(req, &xig, sizeof xig);
743 	}
744 	free(inp_list, M_TEMP);
745 	return (error);
746 }
747 
748 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
749     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
750 
751 static int
752 udp_getcred(SYSCTL_HANDLER_ARGS)
753 {
754 	INIT_VNET_INET(curvnet);
755 	struct xucred xuc;
756 	struct sockaddr_in addrs[2];
757 	struct inpcb *inp;
758 	int error;
759 
760 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
761 	if (error)
762 		return (error);
763 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
764 	if (error)
765 		return (error);
766 	INP_INFO_RLOCK(&V_udbinfo);
767 	inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
768 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
769 	if (inp != NULL) {
770 		INP_RLOCK(inp);
771 		INP_INFO_RUNLOCK(&V_udbinfo);
772 		if (inp->inp_socket == NULL)
773 			error = ENOENT;
774 		if (error == 0)
775 			error = cr_canseeinpcb(req->td->td_ucred, inp);
776 		if (error == 0)
777 			cru2x(inp->inp_cred, &xuc);
778 		INP_RUNLOCK(inp);
779 	} else {
780 		INP_INFO_RUNLOCK(&V_udbinfo);
781 		error = ENOENT;
782 	}
783 	if (error == 0)
784 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
785 	return (error);
786 }
787 
788 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
789     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
790     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
791 
792 static int
793 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
794     struct mbuf *control, struct thread *td)
795 {
796 	INIT_VNET_INET(inp->inp_vnet);
797 	struct udpiphdr *ui;
798 	int len = m->m_pkthdr.len;
799 	struct in_addr faddr, laddr;
800 	struct cmsghdr *cm;
801 	struct sockaddr_in *sin, src;
802 	int error = 0;
803 	int ipflags;
804 	u_short fport, lport;
805 	int unlock_udbinfo;
806 
807 	/*
808 	 * udp_output() may need to temporarily bind or connect the current
809 	 * inpcb.  As such, we don't know up front whether we will need the
810 	 * pcbinfo lock or not.  Do any work to decide what is needed up
811 	 * front before acquiring any locks.
812 	 */
813 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
814 		if (control)
815 			m_freem(control);
816 		m_freem(m);
817 		return (EMSGSIZE);
818 	}
819 
820 	src.sin_family = 0;
821 	if (control != NULL) {
822 		/*
823 		 * XXX: Currently, we assume all the optional information is
824 		 * stored in a single mbuf.
825 		 */
826 		if (control->m_next) {
827 			m_freem(control);
828 			m_freem(m);
829 			return (EINVAL);
830 		}
831 		for (; control->m_len > 0;
832 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
833 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
834 			cm = mtod(control, struct cmsghdr *);
835 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
836 			    || cm->cmsg_len > control->m_len) {
837 				error = EINVAL;
838 				break;
839 			}
840 			if (cm->cmsg_level != IPPROTO_IP)
841 				continue;
842 
843 			switch (cm->cmsg_type) {
844 			case IP_SENDSRCADDR:
845 				if (cm->cmsg_len !=
846 				    CMSG_LEN(sizeof(struct in_addr))) {
847 					error = EINVAL;
848 					break;
849 				}
850 				bzero(&src, sizeof(src));
851 				src.sin_family = AF_INET;
852 				src.sin_len = sizeof(src);
853 				src.sin_port = inp->inp_lport;
854 				src.sin_addr =
855 				    *(struct in_addr *)CMSG_DATA(cm);
856 				break;
857 
858 			default:
859 				error = ENOPROTOOPT;
860 				break;
861 			}
862 			if (error)
863 				break;
864 		}
865 		m_freem(control);
866 	}
867 	if (error) {
868 		m_freem(m);
869 		return (error);
870 	}
871 
872 	/*
873 	 * Depending on whether or not the application has bound or connected
874 	 * the socket, we may have to do varying levels of work.  The optimal
875 	 * case is for a connected UDP socket, as a global lock isn't
876 	 * required at all.
877 	 *
878 	 * In order to decide which we need, we require stability of the
879 	 * inpcb binding, which we ensure by acquiring a read lock on the
880 	 * inpcb.  This doesn't strictly follow the lock order, so we play
881 	 * the trylock and retry game; note that we may end up with more
882 	 * conservative locks than required the second time around, so later
883 	 * assertions have to accept that.  Further analysis of the number of
884 	 * misses under contention is required.
885 	 */
886 	sin = (struct sockaddr_in *)addr;
887 	INP_RLOCK(inp);
888 	if (sin != NULL &&
889 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
890 		INP_RUNLOCK(inp);
891 		INP_INFO_WLOCK(&V_udbinfo);
892 		INP_WLOCK(inp);
893 		unlock_udbinfo = 2;
894 	} else if ((sin != NULL && (
895 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
896 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
897 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
898 	    (inp->inp_lport == 0))) ||
899 	    (src.sin_family == AF_INET)) {
900 		if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
901 			INP_RUNLOCK(inp);
902 			INP_INFO_RLOCK(&V_udbinfo);
903 			INP_RLOCK(inp);
904 		}
905 		unlock_udbinfo = 1;
906 	} else
907 		unlock_udbinfo = 0;
908 
909 	/*
910 	 * If the IP_SENDSRCADDR control message was specified, override the
911 	 * source address for this datagram.  Its use is invalidated if the
912 	 * address thus specified is incomplete or clobbers other inpcbs.
913 	 */
914 	laddr = inp->inp_laddr;
915 	lport = inp->inp_lport;
916 	if (src.sin_family == AF_INET) {
917 		INP_INFO_LOCK_ASSERT(&V_udbinfo);
918 		if ((lport == 0) ||
919 		    (laddr.s_addr == INADDR_ANY &&
920 		     src.sin_addr.s_addr == INADDR_ANY)) {
921 			error = EINVAL;
922 			goto release;
923 		}
924 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
925 		    &laddr.s_addr, &lport, td->td_ucred);
926 		if (error)
927 			goto release;
928 	}
929 
930 	/*
931 	 * If a UDP socket has been connected, then a local address/port will
932 	 * have been selected and bound.
933 	 *
934 	 * If a UDP socket has not been connected to, then an explicit
935 	 * destination address must be used, in which case a local
936 	 * address/port may not have been selected and bound.
937 	 */
938 	if (sin != NULL) {
939 		INP_LOCK_ASSERT(inp);
940 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
941 			error = EISCONN;
942 			goto release;
943 		}
944 
945 		/*
946 		 * Jail may rewrite the destination address, so let it do
947 		 * that before we use it.
948 		 */
949 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
950 		if (error)
951 			goto release;
952 
953 		/*
954 		 * If a local address or port hasn't yet been selected, or if
955 		 * the destination address needs to be rewritten due to using
956 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
957 		 * to do the heavy lifting.  Once a port is selected, we
958 		 * commit the binding back to the socket; we also commit the
959 		 * binding of the address if in jail.
960 		 *
961 		 * If we already have a valid binding and we're not
962 		 * requesting a destination address rewrite, use a fast path.
963 		 */
964 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
965 		    inp->inp_lport == 0 ||
966 		    sin->sin_addr.s_addr == INADDR_ANY ||
967 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
968 			INP_INFO_LOCK_ASSERT(&V_udbinfo);
969 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
970 			    &lport, &faddr.s_addr, &fport, NULL,
971 			    td->td_ucred);
972 			if (error)
973 				goto release;
974 
975 			/*
976 			 * XXXRW: Why not commit the port if the address is
977 			 * !INADDR_ANY?
978 			 */
979 			/* Commit the local port if newly assigned. */
980 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
981 			    inp->inp_lport == 0) {
982 				INP_INFO_WLOCK_ASSERT(&V_udbinfo);
983 				INP_WLOCK_ASSERT(inp);
984 				/*
985 				 * Remember addr if jailed, to prevent
986 				 * rebinding.
987 				 */
988 				if (jailed(td->td_ucred))
989 					inp->inp_laddr = laddr;
990 				inp->inp_lport = lport;
991 				if (in_pcbinshash(inp) != 0) {
992 					inp->inp_lport = 0;
993 					error = EAGAIN;
994 					goto release;
995 				}
996 				inp->inp_flags |= INP_ANONPORT;
997 			}
998 		} else {
999 			faddr = sin->sin_addr;
1000 			fport = sin->sin_port;
1001 		}
1002 	} else {
1003 		INP_LOCK_ASSERT(inp);
1004 		faddr = inp->inp_faddr;
1005 		fport = inp->inp_fport;
1006 		if (faddr.s_addr == INADDR_ANY) {
1007 			error = ENOTCONN;
1008 			goto release;
1009 		}
1010 	}
1011 
1012 	/*
1013 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1014 	 * link-layer headers.  Immediate slide the data pointer back forward
1015 	 * since we won't use that space at this layer.
1016 	 */
1017 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1018 	if (m == NULL) {
1019 		error = ENOBUFS;
1020 		goto release;
1021 	}
1022 	m->m_data += max_linkhdr;
1023 	m->m_len -= max_linkhdr;
1024 	m->m_pkthdr.len -= max_linkhdr;
1025 
1026 	/*
1027 	 * Fill in mbuf with extended UDP header and addresses and length put
1028 	 * into network format.
1029 	 */
1030 	ui = mtod(m, struct udpiphdr *);
1031 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1032 	ui->ui_pr = IPPROTO_UDP;
1033 	ui->ui_src = laddr;
1034 	ui->ui_dst = faddr;
1035 	ui->ui_sport = lport;
1036 	ui->ui_dport = fport;
1037 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1038 
1039 	/*
1040 	 * Set the Don't Fragment bit in the IP header.
1041 	 */
1042 	if (inp->inp_flags & INP_DONTFRAG) {
1043 		struct ip *ip;
1044 
1045 		ip = (struct ip *)&ui->ui_i;
1046 		ip->ip_off |= IP_DF;
1047 	}
1048 
1049 	ipflags = 0;
1050 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1051 		ipflags |= IP_ROUTETOIF;
1052 	if (inp->inp_socket->so_options & SO_BROADCAST)
1053 		ipflags |= IP_ALLOWBROADCAST;
1054 	if (inp->inp_flags & INP_ONESBCAST)
1055 		ipflags |= IP_SENDONES;
1056 
1057 #ifdef MAC
1058 	mac_inpcb_create_mbuf(inp, m);
1059 #endif
1060 
1061 	/*
1062 	 * Set up checksum and output datagram.
1063 	 */
1064 	if (udp_cksum) {
1065 		if (inp->inp_flags & INP_ONESBCAST)
1066 			faddr.s_addr = INADDR_BROADCAST;
1067 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1068 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1069 		m->m_pkthdr.csum_flags = CSUM_UDP;
1070 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1071 	} else
1072 		ui->ui_sum = 0;
1073 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1074 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1075 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
1076 	UDPSTAT_INC(udps_opackets);
1077 
1078 	if (unlock_udbinfo == 2)
1079 		INP_INFO_WUNLOCK(&V_udbinfo);
1080 	else if (unlock_udbinfo == 1)
1081 		INP_INFO_RUNLOCK(&V_udbinfo);
1082 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1083 	    inp->inp_moptions, inp);
1084 	if (unlock_udbinfo == 2)
1085 		INP_WUNLOCK(inp);
1086 	else
1087 		INP_RUNLOCK(inp);
1088 	return (error);
1089 
1090 release:
1091 	if (unlock_udbinfo == 2) {
1092 		INP_WUNLOCK(inp);
1093 		INP_INFO_WUNLOCK(&V_udbinfo);
1094 	} else if (unlock_udbinfo == 1) {
1095 		INP_RUNLOCK(inp);
1096 		INP_INFO_RUNLOCK(&V_udbinfo);
1097 	} else
1098 		INP_RUNLOCK(inp);
1099 	m_freem(m);
1100 	return (error);
1101 }
1102 
1103 static void
1104 udp_abort(struct socket *so)
1105 {
1106 	INIT_VNET_INET(so->so_vnet);
1107 	struct inpcb *inp;
1108 
1109 	inp = sotoinpcb(so);
1110 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1111 	INP_INFO_WLOCK(&V_udbinfo);
1112 	INP_WLOCK(inp);
1113 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1114 		in_pcbdisconnect(inp);
1115 		inp->inp_laddr.s_addr = INADDR_ANY;
1116 		soisdisconnected(so);
1117 	}
1118 	INP_WUNLOCK(inp);
1119 	INP_INFO_WUNLOCK(&V_udbinfo);
1120 }
1121 
1122 static int
1123 udp_attach(struct socket *so, int proto, struct thread *td)
1124 {
1125 	INIT_VNET_INET(so->so_vnet);
1126 	struct inpcb *inp;
1127 	int error;
1128 
1129 	inp = sotoinpcb(so);
1130 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1131 	error = soreserve(so, udp_sendspace, udp_recvspace);
1132 	if (error)
1133 		return (error);
1134 	INP_INFO_WLOCK(&V_udbinfo);
1135 	error = in_pcballoc(so, &V_udbinfo);
1136 	if (error) {
1137 		INP_INFO_WUNLOCK(&V_udbinfo);
1138 		return (error);
1139 	}
1140 
1141 	inp = (struct inpcb *)so->so_pcb;
1142 	INP_INFO_WUNLOCK(&V_udbinfo);
1143 	inp->inp_vflag |= INP_IPV4;
1144 	inp->inp_ip_ttl = V_ip_defttl;
1145 	/*
1146 	 * UDP does not have a per-protocol pcb (inp->inp_ppcb).
1147 	 * We use this pointer for kernel tunneling pointer.
1148 	 * If we ever need to have a protocol block we will
1149 	 * need to move this function pointer there. Null
1150 	 * in this pointer means "do the normal thing".
1151 	 */
1152 	inp->inp_ppcb = NULL;
1153 	INP_WUNLOCK(inp);
1154 	return (0);
1155 }
1156 
1157 int
1158 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1159 {
1160 	struct inpcb *inp;
1161 
1162 	inp = (struct inpcb *)so->so_pcb;
1163 	KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram"));
1164 	KASSERT(so->so_pcb != NULL, ("udp_set_kernel_tunneling: NULL inp"));
1165 	if (so->so_type != SOCK_DGRAM) {
1166 		/* Not UDP socket... sorry! */
1167 		return (ENOTSUP);
1168 	}
1169 	if (inp == NULL) {
1170 		/* NULL INP? */
1171 		return (EINVAL);
1172 	}
1173 	INP_WLOCK(inp);
1174 	if (inp->inp_ppcb != NULL) {
1175 		INP_WUNLOCK(inp);
1176 		return (EBUSY);
1177 	}
1178 	inp->inp_ppcb = f;
1179 	INP_WUNLOCK(inp);
1180 	return (0);
1181 }
1182 
1183 static int
1184 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1185 {
1186 	INIT_VNET_INET(so->so_vnet);
1187 	struct inpcb *inp;
1188 	int error;
1189 
1190 	inp = sotoinpcb(so);
1191 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1192 	INP_INFO_WLOCK(&V_udbinfo);
1193 	INP_WLOCK(inp);
1194 	error = in_pcbbind(inp, nam, td->td_ucred);
1195 	INP_WUNLOCK(inp);
1196 	INP_INFO_WUNLOCK(&V_udbinfo);
1197 	return (error);
1198 }
1199 
1200 static void
1201 udp_close(struct socket *so)
1202 {
1203 	INIT_VNET_INET(so->so_vnet);
1204 	struct inpcb *inp;
1205 
1206 	inp = sotoinpcb(so);
1207 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1208 	INP_INFO_WLOCK(&V_udbinfo);
1209 	INP_WLOCK(inp);
1210 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1211 		in_pcbdisconnect(inp);
1212 		inp->inp_laddr.s_addr = INADDR_ANY;
1213 		soisdisconnected(so);
1214 	}
1215 	INP_WUNLOCK(inp);
1216 	INP_INFO_WUNLOCK(&V_udbinfo);
1217 }
1218 
1219 static int
1220 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1221 {
1222 	INIT_VNET_INET(so->so_vnet);
1223 	struct inpcb *inp;
1224 	int error;
1225 	struct sockaddr_in *sin;
1226 
1227 	inp = sotoinpcb(so);
1228 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1229 	INP_INFO_WLOCK(&V_udbinfo);
1230 	INP_WLOCK(inp);
1231 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1232 		INP_WUNLOCK(inp);
1233 		INP_INFO_WUNLOCK(&V_udbinfo);
1234 		return (EISCONN);
1235 	}
1236 	sin = (struct sockaddr_in *)nam;
1237 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1238 	if (error != 0) {
1239 		INP_WUNLOCK(inp);
1240 		INP_INFO_WUNLOCK(&V_udbinfo);
1241 		return (error);
1242 	}
1243 	error = in_pcbconnect(inp, nam, td->td_ucred);
1244 	if (error == 0)
1245 		soisconnected(so);
1246 	INP_WUNLOCK(inp);
1247 	INP_INFO_WUNLOCK(&V_udbinfo);
1248 	return (error);
1249 }
1250 
1251 static void
1252 udp_detach(struct socket *so)
1253 {
1254 	INIT_VNET_INET(so->so_vnet);
1255 	struct inpcb *inp;
1256 
1257 	inp = sotoinpcb(so);
1258 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1259 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1260 	    ("udp_detach: not disconnected"));
1261 	INP_INFO_WLOCK(&V_udbinfo);
1262 	INP_WLOCK(inp);
1263 	in_pcbdetach(inp);
1264 	in_pcbfree(inp);
1265 	INP_INFO_WUNLOCK(&V_udbinfo);
1266 }
1267 
1268 static int
1269 udp_disconnect(struct socket *so)
1270 {
1271 	INIT_VNET_INET(so->so_vnet);
1272 	struct inpcb *inp;
1273 
1274 	inp = sotoinpcb(so);
1275 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1276 	INP_INFO_WLOCK(&V_udbinfo);
1277 	INP_WLOCK(inp);
1278 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1279 		INP_WUNLOCK(inp);
1280 		INP_INFO_WUNLOCK(&V_udbinfo);
1281 		return (ENOTCONN);
1282 	}
1283 
1284 	in_pcbdisconnect(inp);
1285 	inp->inp_laddr.s_addr = INADDR_ANY;
1286 	SOCK_LOCK(so);
1287 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1288 	SOCK_UNLOCK(so);
1289 	INP_WUNLOCK(inp);
1290 	INP_INFO_WUNLOCK(&V_udbinfo);
1291 	return (0);
1292 }
1293 
1294 static int
1295 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1296     struct mbuf *control, struct thread *td)
1297 {
1298 	struct inpcb *inp;
1299 
1300 	inp = sotoinpcb(so);
1301 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1302 	return (udp_output(inp, m, addr, control, td));
1303 }
1304 
1305 int
1306 udp_shutdown(struct socket *so)
1307 {
1308 	struct inpcb *inp;
1309 
1310 	inp = sotoinpcb(so);
1311 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1312 	INP_WLOCK(inp);
1313 	socantsendmore(so);
1314 	INP_WUNLOCK(inp);
1315 	return (0);
1316 }
1317 
1318 struct pr_usrreqs udp_usrreqs = {
1319 	.pru_abort =		udp_abort,
1320 	.pru_attach =		udp_attach,
1321 	.pru_bind =		udp_bind,
1322 	.pru_connect =		udp_connect,
1323 	.pru_control =		in_control,
1324 	.pru_detach =		udp_detach,
1325 	.pru_disconnect =	udp_disconnect,
1326 	.pru_peeraddr =		in_getpeeraddr,
1327 	.pru_send =		udp_send,
1328 	.pru_soreceive =	soreceive_dgram,
1329 	.pru_sosend =		sosend_dgram,
1330 	.pru_shutdown =		udp_shutdown,
1331 	.pru_sockaddr =		in_getsockaddr,
1332 	.pru_sosetlabel =	in_pcbsosetlabel,
1333 	.pru_close =		udp_close,
1334 };
1335