xref: /freebsd/sys/netinet/udp_usrreq.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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_ipfw.h"
34 #include "opt_ipsec.h"
35 #include "opt_inet6.h"
36 #include "opt_mac.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/domain.h>
41 #include <sys/eventhandler.h>
42 #include <sys/jail.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/mac.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/proc.h>
49 #include <sys/protosw.h>
50 #include <sys/signalvar.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sx.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 
57 #include <vm/uma.h>
58 
59 #include <net/if.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/ip_icmp.h>
71 #include <netinet/icmp_var.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/ip_options.h>
74 #ifdef INET6
75 #include <netinet6/ip6_var.h>
76 #endif
77 #include <netinet/udp.h>
78 #include <netinet/udp_var.h>
79 
80 #ifdef FAST_IPSEC
81 #include <netipsec/ipsec.h>
82 #endif /*FAST_IPSEC*/
83 
84 #ifdef IPSEC
85 #include <netinet6/ipsec.h>
86 #endif /*IPSEC*/
87 
88 #include <machine/in_cksum.h>
89 
90 /*
91  * UDP protocol implementation.
92  * Per RFC 768, August, 1980.
93  */
94 #ifndef	COMPAT_42
95 static int	udpcksum = 1;
96 #else
97 static int	udpcksum = 0;		/* XXX */
98 #endif
99 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
100 		&udpcksum, 0, "");
101 
102 int	log_in_vain = 0;
103 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
104     &log_in_vain, 0, "Log all incoming UDP packets");
105 
106 static int	blackhole = 0;
107 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
108 	&blackhole, 0, "Do not send port unreachables for refused connects");
109 
110 static int	strict_mcast_mship = 0;
111 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
112 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
113 
114 struct	inpcbhead udb;		/* from udp_var.h */
115 #define	udb6	udb  /* for KAME src sync over BSD*'s */
116 struct	inpcbinfo udbinfo;
117 
118 #ifndef UDBHASHSIZE
119 #define UDBHASHSIZE 16
120 #endif
121 
122 struct	udpstat udpstat;	/* from udp_var.h */
123 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
124     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
125 
126 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
127 		int off, struct sockaddr_in *udp_in);
128 
129 static void udp_detach(struct socket *so);
130 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
131 		struct mbuf *, struct thread *);
132 
133 static void
134 udp_zone_change(void *tag)
135 {
136 
137 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
138 }
139 
140 void
141 udp_init()
142 {
143 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
144 	LIST_INIT(&udb);
145 	udbinfo.listhead = &udb;
146 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
147 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
148 					&udbinfo.porthashmask);
149 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
150 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
151 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
152 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
153 		EVENTHANDLER_PRI_ANY);
154 }
155 
156 void
157 udp_input(m, off)
158 	register struct mbuf *m;
159 	int off;
160 {
161 	int iphlen = off;
162 	register struct ip *ip;
163 	register struct udphdr *uh;
164 	register struct inpcb *inp;
165 	int len;
166 	struct ip save_ip;
167 	struct sockaddr_in udp_in;
168 #ifdef IPFIREWALL_FORWARD
169 	struct m_tag *fwd_tag;
170 #endif
171 
172 	udpstat.udps_ipackets++;
173 
174 	/*
175 	 * Strip IP options, if any; should skip this,
176 	 * make available to user, and use on returned packets,
177 	 * but we don't yet have a way to check the checksum
178 	 * with options still present.
179 	 */
180 	if (iphlen > sizeof (struct ip)) {
181 		ip_stripoptions(m, (struct mbuf *)0);
182 		iphlen = sizeof(struct ip);
183 	}
184 
185 	/*
186 	 * Get IP and UDP header together in first mbuf.
187 	 */
188 	ip = mtod(m, struct ip *);
189 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
190 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
191 			udpstat.udps_hdrops++;
192 			return;
193 		}
194 		ip = mtod(m, struct ip *);
195 	}
196 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
197 
198 	/* destination port of 0 is illegal, based on RFC768. */
199 	if (uh->uh_dport == 0)
200 		goto badunlocked;
201 
202 	/*
203 	 * Construct sockaddr format source address.
204 	 * Stuff source address and datagram in user buffer.
205 	 */
206 	bzero(&udp_in, sizeof(udp_in));
207 	udp_in.sin_len = sizeof(udp_in);
208 	udp_in.sin_family = AF_INET;
209 	udp_in.sin_port = uh->uh_sport;
210 	udp_in.sin_addr = ip->ip_src;
211 
212 	/*
213 	 * Make mbuf data length reflect UDP length.
214 	 * If not enough data to reflect UDP length, drop.
215 	 */
216 	len = ntohs((u_short)uh->uh_ulen);
217 	if (ip->ip_len != len) {
218 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
219 			udpstat.udps_badlen++;
220 			goto badunlocked;
221 		}
222 		m_adj(m, len - ip->ip_len);
223 		/* ip->ip_len = len; */
224 	}
225 	/*
226 	 * Save a copy of the IP header in case we want restore it
227 	 * for sending an ICMP error message in response.
228 	 */
229 	if (!blackhole)
230 		save_ip = *ip;
231 
232 	/*
233 	 * Checksum extended UDP header and data.
234 	 */
235 	if (uh->uh_sum) {
236 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
237 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
238 				uh->uh_sum = m->m_pkthdr.csum_data;
239 			else
240 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
241 				    ip->ip_dst.s_addr, htonl((u_short)len +
242 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
243 			uh->uh_sum ^= 0xffff;
244 		} else {
245 			char b[9];
246 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
247 			bzero(((struct ipovly *)ip)->ih_x1, 9);
248 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
249 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
250 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
251 		}
252 		if (uh->uh_sum) {
253 			udpstat.udps_badsum++;
254 			m_freem(m);
255 			return;
256 		}
257 	} else
258 		udpstat.udps_nosum++;
259 
260 #ifdef IPFIREWALL_FORWARD
261 	/* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
262 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
263 
264 	if (fwd_tag != NULL) {
265 		struct sockaddr_in *next_hop;
266 
267 		/* Do the hack. */
268 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
269 		ip->ip_dst = next_hop->sin_addr;
270 		uh->uh_dport = ntohs(next_hop->sin_port);
271 		/* Remove the tag from the packet.  We don't need it anymore. */
272 		m_tag_delete(m, fwd_tag);
273 	}
274 #endif
275 
276 	INP_INFO_RLOCK(&udbinfo);
277 
278 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
279 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
280 		struct inpcb *last;
281 		/*
282 		 * Deliver a multicast or broadcast datagram to *all* sockets
283 		 * for which the local and remote addresses and ports match
284 		 * those of the incoming datagram.  This allows more than
285 		 * one process to receive multi/broadcasts on the same port.
286 		 * (This really ought to be done for unicast datagrams as
287 		 * well, but that would cause problems with existing
288 		 * applications that open both address-specific sockets and
289 		 * a wildcard socket listening to the same port -- they would
290 		 * end up receiving duplicates of every unicast datagram.
291 		 * Those applications open the multiple sockets to overcome an
292 		 * inadequacy of the UDP socket interface, but for backwards
293 		 * compatibility we avoid the problem here rather than
294 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
295 		 */
296 
297 		/*
298 		 * Locate pcb(s) for datagram.
299 		 * (Algorithm copied from raw_intr().)
300 		 */
301 		last = NULL;
302 		LIST_FOREACH(inp, &udb, inp_list) {
303 			if (inp->inp_lport != uh->uh_dport)
304 				continue;
305 #ifdef INET6
306 			if ((inp->inp_vflag & INP_IPV4) == 0)
307 				continue;
308 #endif
309 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
310 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
311 					continue;
312 			}
313 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
314 				if (inp->inp_faddr.s_addr !=
315 				    ip->ip_src.s_addr ||
316 				    inp->inp_fport != uh->uh_sport)
317 					continue;
318 			}
319 			INP_LOCK(inp);
320 
321 			/*
322 			 * Check multicast packets to make sure they are only
323 			 * sent to sockets with multicast memberships for the
324 			 * packet's destination address and arrival interface
325 			 */
326 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
327 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
328 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
329 				int mship, foundmship = 0;
330 
331 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
332 					if (MSHIP(inp, mship)->inm_addr.s_addr
333 					    == ip->ip_dst.s_addr &&
334 					    MSHIP(inp, mship)->inm_ifp
335 					    == m->m_pkthdr.rcvif) {
336 						foundmship = 1;
337 						break;
338 					}
339 				}
340 				if (foundmship == 0) {
341 					INP_UNLOCK(inp);
342 					continue;
343 				}
344 			}
345 #undef NMSHIPS
346 #undef MSHIP
347 			if (last != NULL) {
348 				struct mbuf *n;
349 
350 				n = m_copy(m, 0, M_COPYALL);
351 				if (n != NULL)
352 					udp_append(last, ip, n,
353 						   iphlen +
354 						   sizeof(struct udphdr),
355 						   &udp_in);
356 				INP_UNLOCK(last);
357 			}
358 			last = inp;
359 			/*
360 			 * Don't look for additional matches if this one does
361 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
362 			 * socket options set.  This heuristic avoids searching
363 			 * through all pcbs in the common case of a non-shared
364 			 * port.  It * assumes that an application will never
365 			 * clear these options after setting them.
366 			 */
367 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
368 				break;
369 		}
370 
371 		if (last == NULL) {
372 			/*
373 			 * No matching pcb found; discard datagram.
374 			 * (No need to send an ICMP Port Unreachable
375 			 * for a broadcast or multicast datgram.)
376 			 */
377 			udpstat.udps_noportbcast++;
378 			goto badheadlocked;
379 		}
380 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
381 		    &udp_in);
382 		INP_UNLOCK(last);
383 		INP_INFO_RUNLOCK(&udbinfo);
384 		return;
385 	}
386 	/*
387 	 * Locate pcb for datagram.
388 	 */
389 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
390 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
391 	if (inp == NULL) {
392 		if (log_in_vain) {
393 			char buf[4*sizeof "123"];
394 
395 			strcpy(buf, inet_ntoa(ip->ip_dst));
396 			log(LOG_INFO,
397 			    "Connection attempt to UDP %s:%d from %s:%d\n",
398 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
399 			    ntohs(uh->uh_sport));
400 		}
401 		udpstat.udps_noport++;
402 		if (m->m_flags & (M_BCAST | M_MCAST)) {
403 			udpstat.udps_noportbcast++;
404 			goto badheadlocked;
405 		}
406 		if (blackhole)
407 			goto badheadlocked;
408 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
409 			goto badheadlocked;
410 		*ip = save_ip;
411 		ip->ip_len += iphlen;
412 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
413 		INP_INFO_RUNLOCK(&udbinfo);
414 		return;
415 	}
416 	INP_LOCK(inp);
417 	/* Check the minimum TTL for socket. */
418 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
419 		goto badheadlocked;
420 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
421 	INP_UNLOCK(inp);
422 	INP_INFO_RUNLOCK(&udbinfo);
423 	return;
424 
425 badheadlocked:
426 	if (inp)
427 		INP_UNLOCK(inp);
428 	INP_INFO_RUNLOCK(&udbinfo);
429 badunlocked:
430 	m_freem(m);
431 	return;
432 }
433 
434 /*
435  * Subroutine of udp_input(), which appends the provided mbuf chain to the
436  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
437  * contains the source address.  If the socket ends up being an IPv6 socket,
438  * udp_append() will convert to a sockaddr_in6 before passing the address
439  * into the socket code.
440  */
441 static void
442 udp_append(inp, ip, n, off, udp_in)
443 	struct inpcb *inp;
444 	struct ip *ip;
445 	struct mbuf *n;
446 	int off;
447 	struct sockaddr_in *udp_in;
448 {
449 	struct sockaddr *append_sa;
450 	struct socket *so;
451 	struct mbuf *opts = 0;
452 #ifdef INET6
453 	struct sockaddr_in6 udp_in6;
454 #endif
455 
456 	INP_LOCK_ASSERT(inp);
457 
458 #if defined(IPSEC) || defined(FAST_IPSEC)
459 	/* check AH/ESP integrity. */
460 	if (ipsec4_in_reject(n, inp)) {
461 #ifdef IPSEC
462 		ipsecstat.in_polvio++;
463 #endif /*IPSEC*/
464 		m_freem(n);
465 		return;
466 	}
467 #endif /*IPSEC || FAST_IPSEC*/
468 #ifdef MAC
469 	if (mac_check_inpcb_deliver(inp, n) != 0) {
470 		m_freem(n);
471 		return;
472 	}
473 #endif
474 	if (inp->inp_flags & INP_CONTROLOPTS ||
475 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
476 #ifdef INET6
477 		if (inp->inp_vflag & INP_IPV6) {
478 			int savedflags;
479 
480 			savedflags = inp->inp_flags;
481 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
482 			ip6_savecontrol(inp, n, &opts);
483 			inp->inp_flags = savedflags;
484 		} else
485 #endif
486 		ip_savecontrol(inp, &opts, ip, n);
487 	}
488 #ifdef INET6
489 	if (inp->inp_vflag & INP_IPV6) {
490 		bzero(&udp_in6, sizeof(udp_in6));
491 		udp_in6.sin6_len = sizeof(udp_in6);
492 		udp_in6.sin6_family = AF_INET6;
493 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
494 		append_sa = (struct sockaddr *)&udp_in6;
495 	} else
496 #endif
497 	append_sa = (struct sockaddr *)udp_in;
498 	m_adj(n, off);
499 
500 	so = inp->inp_socket;
501 	SOCKBUF_LOCK(&so->so_rcv);
502 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
503 		m_freem(n);
504 		if (opts)
505 			m_freem(opts);
506 		udpstat.udps_fullsock++;
507 		SOCKBUF_UNLOCK(&so->so_rcv);
508 	} else
509 		sorwakeup_locked(so);
510 }
511 
512 /*
513  * Notify a udp user of an asynchronous error;
514  * just wake up so that he can collect error status.
515  */
516 struct inpcb *
517 udp_notify(inp, errno)
518 	register struct inpcb *inp;
519 	int errno;
520 {
521 	inp->inp_socket->so_error = errno;
522 	sorwakeup(inp->inp_socket);
523 	sowwakeup(inp->inp_socket);
524 	return inp;
525 }
526 
527 void
528 udp_ctlinput(cmd, sa, vip)
529 	int cmd;
530 	struct sockaddr *sa;
531 	void *vip;
532 {
533 	struct ip *ip = vip;
534 	struct udphdr *uh;
535 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
536 	struct in_addr faddr;
537 	struct inpcb *inp;
538 
539 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
540 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
541 		return;
542 
543 	/*
544 	 * Redirects don't need to be handled up here.
545 	 */
546 	if (PRC_IS_REDIRECT(cmd))
547 		return;
548 	/*
549 	 * Hostdead is ugly because it goes linearly through all PCBs.
550 	 * XXX: We never get this from ICMP, otherwise it makes an
551 	 * excellent DoS attack on machines with many connections.
552 	 */
553 	if (cmd == PRC_HOSTDEAD)
554 		ip = 0;
555 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
556 		return;
557 	if (ip) {
558 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
559 		INP_INFO_RLOCK(&udbinfo);
560 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
561 		    ip->ip_src, uh->uh_sport, 0, NULL);
562 		if (inp != NULL) {
563 			INP_LOCK(inp);
564 			if (inp->inp_socket != NULL) {
565 				(*notify)(inp, inetctlerrmap[cmd]);
566 			}
567 			INP_UNLOCK(inp);
568 		}
569 		INP_INFO_RUNLOCK(&udbinfo);
570 	} else
571 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
572 }
573 
574 static int
575 udp_pcblist(SYSCTL_HANDLER_ARGS)
576 {
577 	int error, i, n;
578 	struct inpcb *inp, **inp_list;
579 	inp_gen_t gencnt;
580 	struct xinpgen xig;
581 
582 	/*
583 	 * The process of preparing the TCB list is too time-consuming and
584 	 * resource-intensive to repeat twice on every request.
585 	 */
586 	if (req->oldptr == 0) {
587 		n = udbinfo.ipi_count;
588 		req->oldidx = 2 * (sizeof xig)
589 			+ (n + n/8) * sizeof(struct xinpcb);
590 		return 0;
591 	}
592 
593 	if (req->newptr != 0)
594 		return EPERM;
595 
596 	/*
597 	 * OK, now we're committed to doing something.
598 	 */
599 	INP_INFO_RLOCK(&udbinfo);
600 	gencnt = udbinfo.ipi_gencnt;
601 	n = udbinfo.ipi_count;
602 	INP_INFO_RUNLOCK(&udbinfo);
603 
604 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
605 		+ n * sizeof(struct xinpcb));
606 	if (error != 0)
607 		return (error);
608 
609 	xig.xig_len = sizeof xig;
610 	xig.xig_count = n;
611 	xig.xig_gen = gencnt;
612 	xig.xig_sogen = so_gencnt;
613 	error = SYSCTL_OUT(req, &xig, sizeof xig);
614 	if (error)
615 		return error;
616 
617 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
618 	if (inp_list == 0)
619 		return ENOMEM;
620 
621 	INP_INFO_RLOCK(&udbinfo);
622 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
623 	     inp = LIST_NEXT(inp, inp_list)) {
624 		INP_LOCK(inp);
625 		if (inp->inp_gencnt <= gencnt &&
626 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
627 			inp_list[i++] = inp;
628 		INP_UNLOCK(inp);
629 	}
630 	INP_INFO_RUNLOCK(&udbinfo);
631 	n = i;
632 
633 	error = 0;
634 	for (i = 0; i < n; i++) {
635 		inp = inp_list[i];
636 		if (inp->inp_gencnt <= gencnt) {
637 			struct xinpcb xi;
638 			bzero(&xi, sizeof(xi));
639 			xi.xi_len = sizeof xi;
640 			/* XXX should avoid extra copy */
641 			bcopy(inp, &xi.xi_inp, sizeof *inp);
642 			if (inp->inp_socket)
643 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
644 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
645 			error = SYSCTL_OUT(req, &xi, sizeof xi);
646 		}
647 	}
648 	if (!error) {
649 		/*
650 		 * Give the user an updated idea of our state.
651 		 * If the generation differs from what we told
652 		 * her before, she knows that something happened
653 		 * while we were processing this request, and it
654 		 * might be necessary to retry.
655 		 */
656 		INP_INFO_RLOCK(&udbinfo);
657 		xig.xig_gen = udbinfo.ipi_gencnt;
658 		xig.xig_sogen = so_gencnt;
659 		xig.xig_count = udbinfo.ipi_count;
660 		INP_INFO_RUNLOCK(&udbinfo);
661 		error = SYSCTL_OUT(req, &xig, sizeof xig);
662 	}
663 	free(inp_list, M_TEMP);
664 	return error;
665 }
666 
667 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
668 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
669 
670 static int
671 udp_getcred(SYSCTL_HANDLER_ARGS)
672 {
673 	struct xucred xuc;
674 	struct sockaddr_in addrs[2];
675 	struct inpcb *inp;
676 	int error;
677 
678 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
679 	if (error)
680 		return (error);
681 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
682 	if (error)
683 		return (error);
684 	INP_INFO_RLOCK(&udbinfo);
685 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
686 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
687 	if (inp == NULL || inp->inp_socket == NULL) {
688 		error = ENOENT;
689 		goto out;
690 	}
691 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
692 	if (error)
693 		goto out;
694 	cru2x(inp->inp_socket->so_cred, &xuc);
695 out:
696 	INP_INFO_RUNLOCK(&udbinfo);
697 	if (error == 0)
698 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
699 	return (error);
700 }
701 
702 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
703     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
704     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
705 
706 static int
707 udp_output(inp, m, addr, control, td)
708 	register struct inpcb *inp;
709 	struct mbuf *m;
710 	struct sockaddr *addr;
711 	struct mbuf *control;
712 	struct thread *td;
713 {
714 	register struct udpiphdr *ui;
715 	register int len = m->m_pkthdr.len;
716 	struct in_addr faddr, laddr;
717 	struct cmsghdr *cm;
718 	struct sockaddr_in *sin, src;
719 	int error = 0;
720 	int ipflags;
721 	u_short fport, lport;
722 	int unlock_udbinfo;
723 
724 	/*
725 	 * udp_output() may need to temporarily bind or connect the current
726 	 * inpcb.  As such, we don't know up front what inpcb locks we will
727 	 * need.  Do any work to decide what is needed up front before
728 	 * acquiring locks.
729 	 */
730 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
731 		if (control)
732 			m_freem(control);
733 		m_freem(m);
734 		return EMSGSIZE;
735 	}
736 
737 	src.sin_addr.s_addr = INADDR_ANY;
738 	if (control != NULL) {
739 		/*
740 		 * XXX: Currently, we assume all the optional information
741 		 * is stored in a single mbuf.
742 		 */
743 		if (control->m_next) {
744 			m_freem(control);
745 			m_freem(m);
746 			return EINVAL;
747 		}
748 		for (; control->m_len > 0;
749 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
750 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
751 			cm = mtod(control, struct cmsghdr *);
752 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
753 			    cm->cmsg_len > control->m_len) {
754 				error = EINVAL;
755 				break;
756 			}
757 			if (cm->cmsg_level != IPPROTO_IP)
758 				continue;
759 
760 			switch (cm->cmsg_type) {
761 			case IP_SENDSRCADDR:
762 				if (cm->cmsg_len !=
763 				    CMSG_LEN(sizeof(struct in_addr))) {
764 					error = EINVAL;
765 					break;
766 				}
767 				bzero(&src, sizeof(src));
768 				src.sin_family = AF_INET;
769 				src.sin_len = sizeof(src);
770 				src.sin_port = inp->inp_lport;
771 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
772 				break;
773 			default:
774 				error = ENOPROTOOPT;
775 				break;
776 			}
777 			if (error)
778 				break;
779 		}
780 		m_freem(control);
781 	}
782 	if (error) {
783 		m_freem(m);
784 		return error;
785 	}
786 
787 	if (src.sin_addr.s_addr != INADDR_ANY ||
788 	    addr != NULL) {
789 		INP_INFO_WLOCK(&udbinfo);
790 		unlock_udbinfo = 1;
791 	} else
792 		unlock_udbinfo = 0;
793 	INP_LOCK(inp);
794 
795 #ifdef MAC
796 	mac_create_mbuf_from_inpcb(inp, m);
797 #endif
798 
799 	laddr = inp->inp_laddr;
800 	lport = inp->inp_lport;
801 	if (src.sin_addr.s_addr != INADDR_ANY) {
802 		if (lport == 0) {
803 			error = EINVAL;
804 			goto release;
805 		}
806 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
807 		    &laddr.s_addr, &lport, td->td_ucred);
808 		if (error)
809 			goto release;
810 	}
811 
812 	if (addr) {
813 		sin = (struct sockaddr_in *)addr;
814 		if (jailed(td->td_ucred))
815 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
816 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
817 			error = EISCONN;
818 			goto release;
819 		}
820 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
821 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
822 		if (error)
823 			goto release;
824 
825 		/* Commit the local port if newly assigned. */
826 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
827 		    inp->inp_lport == 0) {
828 			/*
829 			 * Remember addr if jailed, to prevent rebinding.
830 			 */
831 			if (jailed(td->td_ucred))
832 				inp->inp_laddr = laddr;
833 			inp->inp_lport = lport;
834 			if (in_pcbinshash(inp) != 0) {
835 				inp->inp_lport = 0;
836 				error = EAGAIN;
837 				goto release;
838 			}
839 			inp->inp_flags |= INP_ANONPORT;
840 		}
841 	} else {
842 		faddr = inp->inp_faddr;
843 		fport = inp->inp_fport;
844 		if (faddr.s_addr == INADDR_ANY) {
845 			error = ENOTCONN;
846 			goto release;
847 		}
848 	}
849 
850 	/*
851 	 * Calculate data length and get a mbuf for UDP, IP, and possible
852 	 * link-layer headers.  Immediate slide the data pointer back forward
853 	 * since we won't use that space at this layer.
854 	 */
855 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
856 	if (m == NULL) {
857 		error = ENOBUFS;
858 		goto release;
859 	}
860 	m->m_data += max_linkhdr;
861 	m->m_len -= max_linkhdr;
862 	m->m_pkthdr.len -= max_linkhdr;
863 
864 	/*
865 	 * Fill in mbuf with extended UDP header
866 	 * and addresses and length put into network format.
867 	 */
868 	ui = mtod(m, struct udpiphdr *);
869 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
870 	ui->ui_pr = IPPROTO_UDP;
871 	ui->ui_src = laddr;
872 	ui->ui_dst = faddr;
873 	ui->ui_sport = lport;
874 	ui->ui_dport = fport;
875 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
876 
877 	/*
878 	 * Set the Don't Fragment bit in the IP header.
879 	 */
880 	if (inp->inp_flags & INP_DONTFRAG) {
881 		struct ip *ip;
882 		ip = (struct ip *)&ui->ui_i;
883 		ip->ip_off |= IP_DF;
884 	}
885 
886 	ipflags = 0;
887 	if (inp->inp_socket->so_options & SO_DONTROUTE)
888 		ipflags |= IP_ROUTETOIF;
889 	if (inp->inp_socket->so_options & SO_BROADCAST)
890 		ipflags |= IP_ALLOWBROADCAST;
891 	if (inp->inp_vflag & INP_ONESBCAST)
892 		ipflags |= IP_SENDONES;
893 
894 	/*
895 	 * Set up checksum and output datagram.
896 	 */
897 	if (udpcksum) {
898 		if (inp->inp_vflag & INP_ONESBCAST)
899 			faddr.s_addr = INADDR_BROADCAST;
900 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
901 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
902 		m->m_pkthdr.csum_flags = CSUM_UDP;
903 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
904 	} else {
905 		ui->ui_sum = 0;
906 	}
907 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
908 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
909 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
910 	udpstat.udps_opackets++;
911 
912 	if (unlock_udbinfo)
913 		INP_INFO_WUNLOCK(&udbinfo);
914 	error = ip_output(m, inp->inp_options, NULL, ipflags,
915 	    inp->inp_moptions, inp);
916 	INP_UNLOCK(inp);
917 	return (error);
918 
919 release:
920 	INP_UNLOCK(inp);
921 	if (unlock_udbinfo)
922 		INP_INFO_WUNLOCK(&udbinfo);
923 	m_freem(m);
924 	return (error);
925 }
926 
927 u_long	udp_sendspace = 9216;		/* really max datagram size */
928 					/* 40 1K datagrams */
929 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
930     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
931 
932 u_long	udp_recvspace = 40 * (1024 +
933 #ifdef INET6
934 				      sizeof(struct sockaddr_in6)
935 #else
936 				      sizeof(struct sockaddr_in)
937 #endif
938 				      );
939 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
940     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
941 
942 static void
943 udp_abort(struct socket *so)
944 {
945 	struct inpcb *inp;
946 
947 	inp = sotoinpcb(so);
948 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
949 	INP_INFO_WLOCK(&udbinfo);
950 	INP_LOCK(inp);
951 	soisdisconnected(so);
952 	in_pcbdetach(inp);
953 	in_pcbfree(inp);
954 	INP_INFO_WUNLOCK(&udbinfo);
955 }
956 
957 static int
958 udp_attach(struct socket *so, int proto, struct thread *td)
959 {
960 	struct inpcb *inp;
961 	int error;
962 
963 	inp = sotoinpcb(so);
964 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
965 	error = soreserve(so, udp_sendspace, udp_recvspace);
966 	if (error)
967 		return error;
968 	INP_INFO_WLOCK(&udbinfo);
969 	error = in_pcballoc(so, &udbinfo, "udpinp");
970 	if (error) {
971 		INP_INFO_WUNLOCK(&udbinfo);
972 		return error;
973 	}
974 
975 	inp = (struct inpcb *)so->so_pcb;
976 	INP_LOCK(inp);
977 	INP_INFO_WUNLOCK(&udbinfo);
978 	inp->inp_vflag |= INP_IPV4;
979 	inp->inp_ip_ttl = ip_defttl;
980 	INP_UNLOCK(inp);
981 	return 0;
982 }
983 
984 static int
985 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
986 {
987 	struct inpcb *inp;
988 	int error;
989 
990 	inp = sotoinpcb(so);
991 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
992 	INP_INFO_WLOCK(&udbinfo);
993 	INP_LOCK(inp);
994 	error = in_pcbbind(inp, nam, td->td_ucred);
995 	INP_UNLOCK(inp);
996 	INP_INFO_WUNLOCK(&udbinfo);
997 	return error;
998 }
999 
1000 static int
1001 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1002 {
1003 	struct inpcb *inp;
1004 	int error;
1005 	struct sockaddr_in *sin;
1006 
1007 	inp = sotoinpcb(so);
1008 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1009 	INP_INFO_WLOCK(&udbinfo);
1010 	INP_LOCK(inp);
1011 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1012 		INP_UNLOCK(inp);
1013 		INP_INFO_WUNLOCK(&udbinfo);
1014 		return EISCONN;
1015 	}
1016 	sin = (struct sockaddr_in *)nam;
1017 	if (jailed(td->td_ucred))
1018 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1019 	error = in_pcbconnect(inp, nam, td->td_ucred);
1020 	if (error == 0)
1021 		soisconnected(so);
1022 	INP_UNLOCK(inp);
1023 	INP_INFO_WUNLOCK(&udbinfo);
1024 	return error;
1025 }
1026 
1027 static void
1028 udp_detach(struct socket *so)
1029 {
1030 	struct inpcb *inp;
1031 
1032 	inp = sotoinpcb(so);
1033 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1034 	INP_INFO_WLOCK(&udbinfo);
1035 	INP_LOCK(inp);
1036 	in_pcbdetach(inp);
1037 	in_pcbfree(inp);
1038 	INP_INFO_WUNLOCK(&udbinfo);
1039 }
1040 
1041 static int
1042 udp_disconnect(struct socket *so)
1043 {
1044 	struct inpcb *inp;
1045 
1046 	inp = sotoinpcb(so);
1047 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1048 	INP_INFO_WLOCK(&udbinfo);
1049 	INP_LOCK(inp);
1050 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1051 		INP_INFO_WUNLOCK(&udbinfo);
1052 		INP_UNLOCK(inp);
1053 		return ENOTCONN;
1054 	}
1055 
1056 	in_pcbdisconnect(inp);
1057 	inp->inp_laddr.s_addr = INADDR_ANY;
1058 	SOCK_LOCK(so);
1059 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1060 	SOCK_UNLOCK(so);
1061 	INP_UNLOCK(inp);
1062 	INP_INFO_WUNLOCK(&udbinfo);
1063 	return 0;
1064 }
1065 
1066 static int
1067 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1068 	    struct mbuf *control, struct thread *td)
1069 {
1070 	struct inpcb *inp;
1071 
1072 	inp = sotoinpcb(so);
1073 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1074 	return udp_output(inp, m, addr, control, td);
1075 }
1076 
1077 int
1078 udp_shutdown(struct socket *so)
1079 {
1080 	struct inpcb *inp;
1081 
1082 	inp = sotoinpcb(so);
1083 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1084 	INP_LOCK(inp);
1085 	socantsendmore(so);
1086 	INP_UNLOCK(inp);
1087 	return 0;
1088 }
1089 
1090 /*
1091  * This is the wrapper function for in_setsockaddr.  We just pass down
1092  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1093  * here because in_setsockaddr will call malloc and might block.
1094  */
1095 static int
1096 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1097 {
1098 	return (in_setsockaddr(so, nam, &udbinfo));
1099 }
1100 
1101 /*
1102  * This is the wrapper function for in_setpeeraddr.  We just pass down
1103  * the pcbinfo for in_setpeeraddr to lock.
1104  */
1105 static int
1106 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1107 {
1108 	return (in_setpeeraddr(so, nam, &udbinfo));
1109 }
1110 
1111 struct pr_usrreqs udp_usrreqs = {
1112 	.pru_abort =		udp_abort,
1113 	.pru_attach =		udp_attach,
1114 	.pru_bind =		udp_bind,
1115 	.pru_connect =		udp_connect,
1116 	.pru_control =		in_control,
1117 	.pru_detach =		udp_detach,
1118 	.pru_disconnect =	udp_disconnect,
1119 	.pru_peeraddr =		udp_peeraddr,
1120 	.pru_send =		udp_send,
1121 	.pru_sosend =		sosend_dgram,
1122 	.pru_shutdown =		udp_shutdown,
1123 	.pru_sockaddr =		udp_sockaddr,
1124 	.pru_sosetlabel =	in_pcbsosetlabel
1125 };
1126