xref: /freebsd/sys/netinet/udp_usrreq.c (revision 6b3455a7665208c366849f0b2b3bc916fb97516e)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipsec.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/domain.h>
40 #include <sys/jail.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mac.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h>
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sx.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 
55 #include <vm/uma.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #endif
68 #include <netinet/ip_icmp.h>
69 #include <netinet/icmp_var.h>
70 #include <netinet/ip_var.h>
71 #ifdef INET6
72 #include <netinet6/ip6_var.h>
73 #endif
74 #include <netinet/udp.h>
75 #include <netinet/udp_var.h>
76 
77 #ifdef FAST_IPSEC
78 #include <netipsec/ipsec.h>
79 #endif /*FAST_IPSEC*/
80 
81 #ifdef IPSEC
82 #include <netinet6/ipsec.h>
83 #endif /*IPSEC*/
84 
85 #include <machine/in_cksum.h>
86 
87 /*
88  * UDP protocol implementation.
89  * Per RFC 768, August, 1980.
90  */
91 #ifndef	COMPAT_42
92 static int	udpcksum = 1;
93 #else
94 static int	udpcksum = 0;		/* XXX */
95 #endif
96 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
97 		&udpcksum, 0, "");
98 
99 int	log_in_vain = 0;
100 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
101     &log_in_vain, 0, "Log all incoming UDP packets");
102 
103 static int	blackhole = 0;
104 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
105 	&blackhole, 0, "Do not send port unreachables for refused connects");
106 
107 static int	strict_mcast_mship = 0;
108 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
109 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
110 
111 struct	inpcbhead udb;		/* from udp_var.h */
112 #define	udb6	udb  /* for KAME src sync over BSD*'s */
113 struct	inpcbinfo udbinfo;
114 
115 #ifndef UDBHASHSIZE
116 #define UDBHASHSIZE 16
117 #endif
118 
119 struct	udpstat udpstat;	/* from udp_var.h */
120 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
121     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
122 
123 static struct	sockaddr_in udp_in = { sizeof(udp_in), AF_INET };
124 #ifdef INET6
125 struct udp_in6 {
126 	struct sockaddr_in6	uin6_sin;
127 	u_char			uin6_init_done : 1;
128 } udp_in6 = {
129 	{ sizeof(udp_in6.uin6_sin), AF_INET6 },
130 	0
131 };
132 struct udp_ip6 {
133 	struct ip6_hdr		uip6_ip6;
134 	u_char			uip6_init_done : 1;
135 } udp_ip6;
136 #endif /* INET6 */
137 
138 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
139 		int off);
140 #ifdef INET6
141 static void ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip);
142 #endif
143 
144 static int udp_detach(struct socket *so);
145 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
146 		struct mbuf *, struct thread *);
147 
148 void
149 udp_init()
150 {
151 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
152 	LIST_INIT(&udb);
153 	udbinfo.listhead = &udb;
154 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
155 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
156 					&udbinfo.porthashmask);
157 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
158 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
159 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
160 }
161 
162 void
163 udp_input(m, off)
164 	register struct mbuf *m;
165 	int off;
166 {
167 	int iphlen = off;
168 	register struct ip *ip;
169 	register struct udphdr *uh;
170 	register struct inpcb *inp;
171 	struct mbuf *opts = 0;
172 	int len;
173 	struct ip save_ip;
174 
175 	udpstat.udps_ipackets++;
176 
177 	/*
178 	 * Strip IP options, if any; should skip this,
179 	 * make available to user, and use on returned packets,
180 	 * but we don't yet have a way to check the checksum
181 	 * with options still present.
182 	 */
183 	if (iphlen > sizeof (struct ip)) {
184 		ip_stripoptions(m, (struct mbuf *)0);
185 		iphlen = sizeof(struct ip);
186 	}
187 
188 	/*
189 	 * Get IP and UDP header together in first mbuf.
190 	 */
191 	ip = mtod(m, struct ip *);
192 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
193 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
194 			udpstat.udps_hdrops++;
195 			return;
196 		}
197 		ip = mtod(m, struct ip *);
198 	}
199 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
200 
201 	/* destination port of 0 is illegal, based on RFC768. */
202 	if (uh->uh_dport == 0)
203 		goto badunlocked;
204 
205 	/*
206 	 * Construct sockaddr format source address.
207 	 * Stuff source address and datagram in user buffer.
208 	 */
209 	udp_in.sin_port = uh->uh_sport;
210 	udp_in.sin_addr = ip->ip_src;
211 #ifdef INET6
212 	udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0;
213 #endif
214 
215 	/*
216 	 * Make mbuf data length reflect UDP length.
217 	 * If not enough data to reflect UDP length, drop.
218 	 */
219 	len = ntohs((u_short)uh->uh_ulen);
220 	if (ip->ip_len != len) {
221 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
222 			udpstat.udps_badlen++;
223 			goto badunlocked;
224 		}
225 		m_adj(m, len - ip->ip_len);
226 		/* ip->ip_len = len; */
227 	}
228 	/*
229 	 * Save a copy of the IP header in case we want restore it
230 	 * for sending an ICMP error message in response.
231 	 */
232 	if (!blackhole)
233 		save_ip = *ip;
234 
235 	/*
236 	 * Checksum extended UDP header and data.
237 	 */
238 	if (uh->uh_sum) {
239 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
240 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
241 				uh->uh_sum = m->m_pkthdr.csum_data;
242 			else
243 	                	uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
244 				    ip->ip_dst.s_addr, htonl((u_short)len +
245 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
246 			uh->uh_sum ^= 0xffff;
247 		} else {
248 			char b[9];
249 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
250 			bzero(((struct ipovly *)ip)->ih_x1, 9);
251 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
252 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
253 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
254 		}
255 		if (uh->uh_sum) {
256 			udpstat.udps_badsum++;
257 			m_freem(m);
258 			return;
259 		}
260 	} else
261 		udpstat.udps_nosum++;
262 
263 	INP_INFO_RLOCK(&udbinfo);
264 
265 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
266 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
267 		struct inpcb *last;
268 		/*
269 		 * Deliver a multicast or broadcast datagram to *all* sockets
270 		 * for which the local and remote addresses and ports match
271 		 * those of the incoming datagram.  This allows more than
272 		 * one process to receive multi/broadcasts on the same port.
273 		 * (This really ought to be done for unicast datagrams as
274 		 * well, but that would cause problems with existing
275 		 * applications that open both address-specific sockets and
276 		 * a wildcard socket listening to the same port -- they would
277 		 * end up receiving duplicates of every unicast datagram.
278 		 * Those applications open the multiple sockets to overcome an
279 		 * inadequacy of the UDP socket interface, but for backwards
280 		 * compatibility we avoid the problem here rather than
281 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
282 		 */
283 
284 		/*
285 		 * Locate pcb(s) for datagram.
286 		 * (Algorithm copied from raw_intr().)
287 		 */
288 		last = NULL;
289 		LIST_FOREACH(inp, &udb, inp_list) {
290 			INP_LOCK(inp);
291 			if (inp->inp_lport != uh->uh_dport) {
292 		docontinue:
293 				INP_UNLOCK(inp);
294 				continue;
295 			}
296 #ifdef INET6
297 			if ((inp->inp_vflag & INP_IPV4) == 0)
298 				goto docontinue;
299 #endif
300 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
301 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
302 					goto docontinue;
303 			}
304 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
305 				if (inp->inp_faddr.s_addr !=
306 				    ip->ip_src.s_addr ||
307 				    inp->inp_fport != uh->uh_sport)
308 					goto docontinue;
309 			}
310 
311 			/*
312 			 * Check multicast packets to make sure they are only
313 			 * sent to sockets with multicast memberships for the
314 			 * packet's destination address and arrival interface
315 			 */
316 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
317 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
318 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
319 				int mship, foundmship = 0;
320 
321 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
322 					if (MSHIP(inp, mship)->inm_addr.s_addr
323 					    == ip->ip_dst.s_addr &&
324 					    MSHIP(inp, mship)->inm_ifp
325 					    == m->m_pkthdr.rcvif) {
326 						foundmship = 1;
327 						break;
328 					}
329 				}
330 				if (foundmship == 0)
331 					goto docontinue;
332 			}
333 #undef NMSHIPS
334 #undef MSHIP
335 			if (last != NULL) {
336 				struct mbuf *n;
337 
338 				n = m_copy(m, 0, M_COPYALL);
339 				if (n != NULL)
340 					udp_append(last, ip, n,
341 						   iphlen +
342 						   sizeof(struct udphdr));
343 				INP_UNLOCK(last);
344 			}
345 			last = inp;
346 			/*
347 			 * Don't look for additional matches if this one does
348 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
349 			 * socket options set.  This heuristic avoids searching
350 			 * through all pcbs in the common case of a non-shared
351 			 * port.  It * assumes that an application will never
352 			 * clear these options after setting them.
353 			 */
354 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
355 				break;
356 		}
357 
358 		if (last == NULL) {
359 			/*
360 			 * No matching pcb found; discard datagram.
361 			 * (No need to send an ICMP Port Unreachable
362 			 * for a broadcast or multicast datgram.)
363 			 */
364 			udpstat.udps_noportbcast++;
365 			goto badheadlocked;
366 		}
367 		INP_INFO_RUNLOCK(&udbinfo);
368 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr));
369 		INP_UNLOCK(last);
370 		return;
371 	}
372 	/*
373 	 * Locate pcb for datagram.
374 	 */
375 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
376 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
377 	if (inp == NULL) {
378 		if (log_in_vain) {
379 			char buf[4*sizeof "123"];
380 
381 			strcpy(buf, inet_ntoa(ip->ip_dst));
382 			log(LOG_INFO,
383 			    "Connection attempt to UDP %s:%d from %s:%d\n",
384 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
385 			    ntohs(uh->uh_sport));
386 		}
387 		udpstat.udps_noport++;
388 		if (m->m_flags & (M_BCAST | M_MCAST)) {
389 			udpstat.udps_noportbcast++;
390 			goto badheadlocked;
391 		}
392 		if (blackhole)
393 			goto badheadlocked;
394 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
395 			goto badheadlocked;
396 		*ip = save_ip;
397 		ip->ip_len += iphlen;
398 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
399 		INP_INFO_RUNLOCK(&udbinfo);
400 		return;
401 	}
402 	INP_LOCK(inp);
403 	INP_INFO_RUNLOCK(&udbinfo);
404 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr));
405 	INP_UNLOCK(inp);
406 	return;
407 
408 badheadlocked:
409 	INP_INFO_RUNLOCK(&udbinfo);
410 	if (inp)
411 		INP_UNLOCK(inp);
412 badunlocked:
413 	m_freem(m);
414 	if (opts)
415 		m_freem(opts);
416 	return;
417 }
418 
419 #ifdef INET6
420 static void
421 ip_2_ip6_hdr(ip6, ip)
422 	struct ip6_hdr *ip6;
423 	struct ip *ip;
424 {
425 	bzero(ip6, sizeof(*ip6));
426 
427 	ip6->ip6_vfc = IPV6_VERSION;
428 	ip6->ip6_plen = ip->ip_len;
429 	ip6->ip6_nxt = ip->ip_p;
430 	ip6->ip6_hlim = ip->ip_ttl;
431 	ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] =
432 		IPV6_ADDR_INT32_SMP;
433 	ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr;
434 	ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr;
435 }
436 #endif
437 
438 /*
439  * subroutine of udp_input(), mainly for source code readability.
440  * caller must properly init udp_ip6 and udp_in6 beforehand.
441  */
442 static void
443 udp_append(last, ip, n, off)
444 	struct inpcb *last;
445 	struct ip *ip;
446 	struct mbuf *n;
447 	int off;
448 {
449 	struct sockaddr *append_sa;
450 	struct socket *so;
451 	struct mbuf *opts = 0;
452 
453 	INP_LOCK_ASSERT(last);
454 
455 #if defined(IPSEC) || defined(FAST_IPSEC)
456 	/* check AH/ESP integrity. */
457 	if (ipsec4_in_reject(n, last)) {
458 #ifdef IPSEC
459 		ipsecstat.in_polvio++;
460 #endif /*IPSEC*/
461 		m_freem(n);
462 		return;
463 	}
464 #endif /*IPSEC || FAST_IPSEC*/
465 #ifdef MAC
466 	if (mac_check_inpcb_deliver(last, n) != 0) {
467 		m_freem(n);
468 		return;
469 	}
470 #endif
471 	if (last->inp_flags & INP_CONTROLOPTS ||
472 	    last->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
473 #ifdef INET6
474 		if (last->inp_vflag & INP_IPV6) {
475 			int savedflags;
476 
477 			if (udp_ip6.uip6_init_done == 0) {
478 				ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
479 				udp_ip6.uip6_init_done = 1;
480 			}
481 			savedflags = last->inp_flags;
482 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
483 			ip6_savecontrol(last, n, &opts);
484 			last->inp_flags = savedflags;
485 		} else
486 #endif
487 		ip_savecontrol(last, &opts, ip, n);
488 	}
489 #ifdef INET6
490 	if (last->inp_vflag & INP_IPV6) {
491 		if (udp_in6.uin6_init_done == 0) {
492 			in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
493 			udp_in6.uin6_init_done = 1;
494 		}
495 		append_sa = (struct sockaddr *)&udp_in6.uin6_sin;
496 	} else
497 #endif
498 	append_sa = (struct sockaddr *)&udp_in;
499 	m_adj(n, off);
500 
501 	so = last->inp_socket;
502 	SOCKBUF_LOCK(&so->so_rcv);
503 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
504 		m_freem(n);
505 		if (opts)
506 			m_freem(opts);
507 		udpstat.udps_fullsock++;
508 		SOCKBUF_UNLOCK(&so->so_rcv);
509 	} else
510 		sorwakeup_locked(so);
511 }
512 
513 /*
514  * Notify a udp user of an asynchronous error;
515  * just wake up so that he can collect error status.
516  */
517 struct inpcb *
518 udp_notify(inp, errno)
519 	register struct inpcb *inp;
520 	int errno;
521 {
522 	inp->inp_socket->so_error = errno;
523 	sorwakeup(inp->inp_socket);
524 	sowwakeup(inp->inp_socket);
525 	return inp;
526 }
527 
528 void
529 udp_ctlinput(cmd, sa, vip)
530 	int cmd;
531 	struct sockaddr *sa;
532 	void *vip;
533 {
534 	struct ip *ip = vip;
535 	struct udphdr *uh;
536 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
537         struct in_addr faddr;
538 	struct inpcb *inp;
539 	int s;
540 
541 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
542 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
543         	return;
544 
545 	/*
546 	 * Redirects don't need to be handled up here.
547 	 */
548 	if (PRC_IS_REDIRECT(cmd))
549 		return;
550 	/*
551 	 * Hostdead is ugly because it goes linearly through all PCBs.
552 	 * XXX: We never get this from ICMP, otherwise it makes an
553 	 * excellent DoS attack on machines with many connections.
554 	 */
555 	if (cmd == PRC_HOSTDEAD)
556 		ip = 0;
557 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
558 		return;
559 	if (ip) {
560 		s = splnet();
561 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
562 		INP_INFO_RLOCK(&udbinfo);
563 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
564                     ip->ip_src, uh->uh_sport, 0, NULL);
565 		if (inp != NULL) {
566 			INP_LOCK(inp);
567 			if (inp->inp_socket != NULL) {
568 				(*notify)(inp, inetctlerrmap[cmd]);
569 			}
570 			INP_UNLOCK(inp);
571 		}
572 		INP_INFO_RUNLOCK(&udbinfo);
573 		splx(s);
574 	} else
575 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
576 }
577 
578 static int
579 udp_pcblist(SYSCTL_HANDLER_ARGS)
580 {
581 	int error, i, n, s;
582 	struct inpcb *inp, **inp_list;
583 	inp_gen_t gencnt;
584 	struct xinpgen xig;
585 
586 	/*
587 	 * The process of preparing the TCB list is too time-consuming and
588 	 * resource-intensive to repeat twice on every request.
589 	 */
590 	if (req->oldptr == 0) {
591 		n = udbinfo.ipi_count;
592 		req->oldidx = 2 * (sizeof xig)
593 			+ (n + n/8) * sizeof(struct xinpcb);
594 		return 0;
595 	}
596 
597 	if (req->newptr != 0)
598 		return EPERM;
599 
600 	/*
601 	 * OK, now we're committed to doing something.
602 	 */
603 	s = splnet();
604 	INP_INFO_RLOCK(&udbinfo);
605 	gencnt = udbinfo.ipi_gencnt;
606 	n = udbinfo.ipi_count;
607 	INP_INFO_RUNLOCK(&udbinfo);
608 	splx(s);
609 
610 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
611 		+ n * sizeof(struct xinpcb));
612 	if (error != 0)
613 		return (error);
614 
615 	xig.xig_len = sizeof xig;
616 	xig.xig_count = n;
617 	xig.xig_gen = gencnt;
618 	xig.xig_sogen = so_gencnt;
619 	error = SYSCTL_OUT(req, &xig, sizeof xig);
620 	if (error)
621 		return error;
622 
623 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
624 	if (inp_list == 0)
625 		return ENOMEM;
626 
627 	s = splnet();
628 	INP_INFO_RLOCK(&udbinfo);
629 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
630 	     inp = LIST_NEXT(inp, inp_list)) {
631 		INP_LOCK(inp);
632 		if (inp->inp_gencnt <= gencnt &&
633 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
634 			inp_list[i++] = inp;
635 		INP_UNLOCK(inp);
636 	}
637 	INP_INFO_RUNLOCK(&udbinfo);
638 	splx(s);
639 	n = i;
640 
641 	error = 0;
642 	for (i = 0; i < n; i++) {
643 		inp = inp_list[i];
644 		if (inp->inp_gencnt <= gencnt) {
645 			struct xinpcb xi;
646 			xi.xi_len = sizeof xi;
647 			/* XXX should avoid extra copy */
648 			bcopy(inp, &xi.xi_inp, sizeof *inp);
649 			if (inp->inp_socket)
650 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
651 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
652 			error = SYSCTL_OUT(req, &xi, sizeof xi);
653 		}
654 	}
655 	if (!error) {
656 		/*
657 		 * Give the user an updated idea of our state.
658 		 * If the generation differs from what we told
659 		 * her before, she knows that something happened
660 		 * while we were processing this request, and it
661 		 * might be necessary to retry.
662 		 */
663 		s = splnet();
664 		INP_INFO_RLOCK(&udbinfo);
665 		xig.xig_gen = udbinfo.ipi_gencnt;
666 		xig.xig_sogen = so_gencnt;
667 		xig.xig_count = udbinfo.ipi_count;
668 		INP_INFO_RUNLOCK(&udbinfo);
669 		splx(s);
670 		error = SYSCTL_OUT(req, &xig, sizeof xig);
671 	}
672 	free(inp_list, M_TEMP);
673 	return error;
674 }
675 
676 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
677 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
678 
679 static int
680 udp_getcred(SYSCTL_HANDLER_ARGS)
681 {
682 	struct xucred xuc;
683 	struct sockaddr_in addrs[2];
684 	struct inpcb *inp;
685 	int error, s;
686 
687 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
688 	if (error)
689 		return (error);
690 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
691 	if (error)
692 		return (error);
693 	s = splnet();
694 	INP_INFO_RLOCK(&udbinfo);
695 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
696 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
697 	if (inp == NULL || inp->inp_socket == NULL) {
698 		error = ENOENT;
699 		goto out;
700 	}
701 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
702 	if (error)
703 		goto out;
704 	cru2x(inp->inp_socket->so_cred, &xuc);
705 out:
706 	INP_INFO_RUNLOCK(&udbinfo);
707 	splx(s);
708 	if (error == 0)
709 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
710 	return (error);
711 }
712 
713 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
714     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
715     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
716 
717 static int
718 udp_output(inp, m, addr, control, td)
719 	register struct inpcb *inp;
720 	struct mbuf *m;
721 	struct sockaddr *addr;
722 	struct mbuf *control;
723 	struct thread *td;
724 {
725 	register struct udpiphdr *ui;
726 	register int len = m->m_pkthdr.len;
727 	struct in_addr faddr, laddr;
728 	struct cmsghdr *cm;
729 	struct sockaddr_in *sin, src;
730 	int error = 0;
731 	int ipflags;
732 	u_short fport, lport;
733 
734 	INP_LOCK_ASSERT(inp);
735 #ifdef MAC
736 	mac_create_mbuf_from_inpcb(inp, m);
737 #endif
738 
739 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
740 		error = EMSGSIZE;
741 		if (control)
742 			m_freem(control);
743 		goto release;
744 	}
745 
746 	src.sin_addr.s_addr = INADDR_ANY;
747 	if (control != NULL) {
748 		/*
749 		 * XXX: Currently, we assume all the optional information
750 		 * is stored in a single mbuf.
751 		 */
752 		if (control->m_next) {
753 			error = EINVAL;
754 			m_freem(control);
755 			goto release;
756 		}
757 		for (; control->m_len > 0;
758 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
759 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
760 			cm = mtod(control, struct cmsghdr *);
761 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
762 			    cm->cmsg_len > control->m_len) {
763 				error = EINVAL;
764 				break;
765 			}
766 			if (cm->cmsg_level != IPPROTO_IP)
767 				continue;
768 
769 			switch (cm->cmsg_type) {
770 			case IP_SENDSRCADDR:
771 				if (cm->cmsg_len !=
772 				    CMSG_LEN(sizeof(struct in_addr))) {
773 					error = EINVAL;
774 					break;
775 				}
776 				bzero(&src, sizeof(src));
777 				src.sin_family = AF_INET;
778 				src.sin_len = sizeof(src);
779 				src.sin_port = inp->inp_lport;
780 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
781 				break;
782 			default:
783 				error = ENOPROTOOPT;
784 				break;
785 			}
786 			if (error)
787 				break;
788 		}
789 		m_freem(control);
790 	}
791 	if (error)
792 		goto release;
793 	laddr = inp->inp_laddr;
794 	lport = inp->inp_lport;
795 	if (src.sin_addr.s_addr != INADDR_ANY) {
796 		if (lport == 0) {
797 			error = EINVAL;
798 			goto release;
799 		}
800 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
801 		    &laddr.s_addr, &lport, td->td_ucred);
802 		if (error)
803 			goto release;
804 	}
805 
806 	if (addr) {
807 		sin = (struct sockaddr_in *)addr;
808 		if (td && jailed(td->td_ucred))
809 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
810 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
811 			error = EISCONN;
812 			goto release;
813 		}
814 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
815 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
816 		if (error)
817 			goto release;
818 
819 		/* Commit the local port if newly assigned. */
820 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
821 		    inp->inp_lport == 0) {
822 			inp->inp_lport = lport;
823 			if (in_pcbinshash(inp) != 0) {
824 				inp->inp_lport = 0;
825 				error = EAGAIN;
826 				goto release;
827 			}
828 			inp->inp_flags |= INP_ANONPORT;
829 		}
830 	} else {
831 		faddr = inp->inp_faddr;
832 		fport = inp->inp_fport;
833 		if (faddr.s_addr == INADDR_ANY) {
834 			error = ENOTCONN;
835 			goto release;
836 		}
837 	}
838 	/*
839 	 * Calculate data length and get a mbuf
840 	 * for UDP and IP headers.
841 	 */
842 	M_PREPEND(m, sizeof(struct udpiphdr), M_DONTWAIT);
843 	if (m == 0) {
844 		error = ENOBUFS;
845 		goto release;
846 	}
847 
848 	/*
849 	 * Fill in mbuf with extended UDP header
850 	 * and addresses and length put into network format.
851 	 */
852 	ui = mtod(m, struct udpiphdr *);
853 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
854 	ui->ui_pr = IPPROTO_UDP;
855 	ui->ui_src = laddr;
856 	ui->ui_dst = faddr;
857 	ui->ui_sport = lport;
858 	ui->ui_dport = fport;
859 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
860 
861 	ipflags = inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST);
862 	if (inp->inp_flags & INP_ONESBCAST)
863 		ipflags |= IP_SENDONES;
864 
865 	/*
866 	 * Set up checksum and output datagram.
867 	 */
868 	if (udpcksum) {
869 		if (inp->inp_flags & INP_ONESBCAST)
870 			faddr.s_addr = INADDR_BROADCAST;
871 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
872 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
873 		m->m_pkthdr.csum_flags = CSUM_UDP;
874 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
875 	} else {
876 		ui->ui_sum = 0;
877 	}
878 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
879 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
880 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
881 	udpstat.udps_opackets++;
882 
883 	error = ip_output(m, inp->inp_options, NULL, ipflags,
884 	    inp->inp_moptions, inp);
885 	return (error);
886 
887 release:
888 	m_freem(m);
889 	return (error);
890 }
891 
892 u_long	udp_sendspace = 9216;		/* really max datagram size */
893 					/* 40 1K datagrams */
894 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
895     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
896 
897 u_long	udp_recvspace = 40 * (1024 +
898 #ifdef INET6
899 				      sizeof(struct sockaddr_in6)
900 #else
901 				      sizeof(struct sockaddr_in)
902 #endif
903 				      );
904 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
905     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
906 
907 static int
908 udp_abort(struct socket *so)
909 {
910 	struct inpcb *inp;
911 	int s;
912 
913 	INP_INFO_WLOCK(&udbinfo);
914 	inp = sotoinpcb(so);
915 	if (inp == 0) {
916 		INP_INFO_WUNLOCK(&udbinfo);
917 		return EINVAL;	/* ??? possible? panic instead? */
918 	}
919 	INP_LOCK(inp);
920 	soisdisconnected(so);
921 	s = splnet();
922 	in_pcbdetach(inp);
923 	INP_INFO_WUNLOCK(&udbinfo);
924 	splx(s);
925 	return 0;
926 }
927 
928 static int
929 udp_attach(struct socket *so, int proto, struct thread *td)
930 {
931 	struct inpcb *inp;
932 	int s, error;
933 
934 	INP_INFO_WLOCK(&udbinfo);
935 	inp = sotoinpcb(so);
936 	if (inp != 0) {
937 		INP_INFO_WUNLOCK(&udbinfo);
938 		return EINVAL;
939 	}
940 	error = soreserve(so, udp_sendspace, udp_recvspace);
941 	if (error) {
942 		INP_INFO_WUNLOCK(&udbinfo);
943 		return error;
944 	}
945 	s = splnet();
946 	error = in_pcballoc(so, &udbinfo, "udpinp");
947 	splx(s);
948 	if (error) {
949 		INP_INFO_WUNLOCK(&udbinfo);
950 		return error;
951 	}
952 
953 	inp = (struct inpcb *)so->so_pcb;
954 	INP_LOCK(inp);
955 	INP_INFO_WUNLOCK(&udbinfo);
956 	inp->inp_vflag |= INP_IPV4;
957 	inp->inp_ip_ttl = ip_defttl;
958 	INP_UNLOCK(inp);
959 	return 0;
960 }
961 
962 static int
963 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
964 {
965 	struct inpcb *inp;
966 	int s, error;
967 
968 	INP_INFO_WLOCK(&udbinfo);
969 	inp = sotoinpcb(so);
970 	if (inp == 0) {
971 		INP_INFO_WUNLOCK(&udbinfo);
972 		return EINVAL;
973 	}
974 	INP_LOCK(inp);
975 	s = splnet();
976 	error = in_pcbbind(inp, nam, td->td_ucred);
977 	splx(s);
978 	INP_UNLOCK(inp);
979 	INP_INFO_WUNLOCK(&udbinfo);
980 	return error;
981 }
982 
983 static int
984 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
985 {
986 	struct inpcb *inp;
987 	int s, error;
988 	struct sockaddr_in *sin;
989 
990 	INP_INFO_WLOCK(&udbinfo);
991 	inp = sotoinpcb(so);
992 	if (inp == 0) {
993 		INP_INFO_WUNLOCK(&udbinfo);
994 		return EINVAL;
995 	}
996 	INP_LOCK(inp);
997 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
998 		INP_UNLOCK(inp);
999 		INP_INFO_WUNLOCK(&udbinfo);
1000 		return EISCONN;
1001 	}
1002 	s = splnet();
1003 	sin = (struct sockaddr_in *)nam;
1004 	if (td && jailed(td->td_ucred))
1005 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1006 	error = in_pcbconnect(inp, nam, td->td_ucred);
1007 	splx(s);
1008 	if (error == 0)
1009 		soisconnected(so);
1010 	INP_UNLOCK(inp);
1011 	INP_INFO_WUNLOCK(&udbinfo);
1012 	return error;
1013 }
1014 
1015 static int
1016 udp_detach(struct socket *so)
1017 {
1018 	struct inpcb *inp;
1019 	int s;
1020 
1021 	INP_INFO_WLOCK(&udbinfo);
1022 	inp = sotoinpcb(so);
1023 	if (inp == 0) {
1024 		INP_INFO_WUNLOCK(&udbinfo);
1025 		return EINVAL;
1026 	}
1027 	INP_LOCK(inp);
1028 	s = splnet();
1029 	in_pcbdetach(inp);
1030 	INP_INFO_WUNLOCK(&udbinfo);
1031 	splx(s);
1032 	return 0;
1033 }
1034 
1035 static int
1036 udp_disconnect(struct socket *so)
1037 {
1038 	struct inpcb *inp;
1039 	int s;
1040 
1041 	INP_INFO_WLOCK(&udbinfo);
1042 	inp = sotoinpcb(so);
1043 	if (inp == 0) {
1044 		INP_INFO_WUNLOCK(&udbinfo);
1045 		return EINVAL;
1046 	}
1047 	INP_LOCK(inp);
1048 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1049 		INP_INFO_WUNLOCK(&udbinfo);
1050 		INP_UNLOCK(inp);
1051 		return ENOTCONN;
1052 	}
1053 
1054 	s = splnet();
1055 	in_pcbdisconnect(inp);
1056 	inp->inp_laddr.s_addr = INADDR_ANY;
1057 	INP_UNLOCK(inp);
1058 	INP_INFO_WUNLOCK(&udbinfo);
1059 	splx(s);
1060 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1061 	return 0;
1062 }
1063 
1064 static int
1065 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1066 	    struct mbuf *control, struct thread *td)
1067 {
1068 	struct inpcb *inp;
1069 	int ret;
1070 
1071 	INP_INFO_WLOCK(&udbinfo);
1072 	inp = sotoinpcb(so);
1073 	if (inp == 0) {
1074 		INP_INFO_WUNLOCK(&udbinfo);
1075 		m_freem(m);
1076 		return EINVAL;
1077 	}
1078 	INP_LOCK(inp);
1079 	ret = udp_output(inp, m, addr, control, td);
1080 	INP_UNLOCK(inp);
1081 	INP_INFO_WUNLOCK(&udbinfo);
1082 	return ret;
1083 }
1084 
1085 int
1086 udp_shutdown(struct socket *so)
1087 {
1088 	struct inpcb *inp;
1089 
1090 	INP_INFO_RLOCK(&udbinfo);
1091 	inp = sotoinpcb(so);
1092 	if (inp == 0) {
1093 		INP_INFO_RUNLOCK(&udbinfo);
1094 		return EINVAL;
1095 	}
1096 	INP_LOCK(inp);
1097 	INP_INFO_RUNLOCK(&udbinfo);
1098 	socantsendmore(so);
1099 	INP_UNLOCK(inp);
1100 	return 0;
1101 }
1102 
1103 /*
1104  * This is the wrapper function for in_setsockaddr.  We just pass down
1105  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1106  * here because in_setsockaddr will call malloc and might block.
1107  */
1108 static int
1109 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1110 {
1111 	return (in_setsockaddr(so, nam, &udbinfo));
1112 }
1113 
1114 /*
1115  * This is the wrapper function for in_setpeeraddr.  We just pass down
1116  * the pcbinfo for in_setpeeraddr to lock.
1117  */
1118 static int
1119 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1120 {
1121 	return (in_setpeeraddr(so, nam, &udbinfo));
1122 }
1123 
1124 struct pr_usrreqs udp_usrreqs = {
1125 	udp_abort, pru_accept_notsupp, udp_attach, udp_bind, udp_connect,
1126 	pru_connect2_notsupp, in_control, udp_detach, udp_disconnect,
1127 	pru_listen_notsupp, udp_peeraddr, pru_rcvd_notsupp,
1128 	pru_rcvoob_notsupp, udp_send, pru_sense_null, udp_shutdown,
1129 	udp_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel
1130 };
1131