xref: /freebsd/sys/netinet/udp_usrreq.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
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/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/priv.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 #include <security/mac/mac_framework.h>
91 
92 /*
93  * UDP protocol implementation.
94  * Per RFC 768, August, 1980.
95  */
96 
97 /*
98  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp
99  * checksums removes the only data integrity mechanism for packets and
100  * malformed packets that would otherwise be discarded by bad checksums
101  * may cause problems (especially for NFS data blocks).
102  */
103 #ifndef	COMPAT_42
104 static int	udpcksum = 1;
105 #else
106 static int	udpcksum = 0;
107 #endif
108 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
109 		&udpcksum, 0, "");
110 
111 int	log_in_vain = 0;
112 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
113     &log_in_vain, 0, "Log all incoming UDP packets");
114 
115 static int	blackhole = 0;
116 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
117 	&blackhole, 0, "Do not send port unreachables for refused connects");
118 
119 static int	strict_mcast_mship = 0;
120 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
121 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
122 
123 struct	inpcbhead udb;		/* from udp_var.h */
124 #define	udb6	udb  /* for KAME src sync over BSD*'s */
125 struct	inpcbinfo udbinfo;
126 
127 #ifndef UDBHASHSIZE
128 #define UDBHASHSIZE 16
129 #endif
130 
131 struct	udpstat udpstat;	/* from udp_var.h */
132 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
133     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
134 
135 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
136 		int off, struct sockaddr_in *udp_in);
137 
138 static void udp_detach(struct socket *so);
139 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
140 		struct mbuf *, struct thread *);
141 
142 static void
143 udp_zone_change(void *tag)
144 {
145 
146 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
147 }
148 
149 static int
150 udp_inpcb_init(void *mem, int size, int flags)
151 {
152 	struct inpcb *inp = mem;
153 
154 	INP_LOCK_INIT(inp, "inp", "udpinp");
155 	return (0);
156 }
157 
158 void
159 udp_init()
160 {
161 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
162 	LIST_INIT(&udb);
163 	udbinfo.listhead = &udb;
164 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
165 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
166 					&udbinfo.porthashmask);
167 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
168 	    NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
169 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
170 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
171 		EVENTHANDLER_PRI_ANY);
172 }
173 
174 void
175 udp_input(m, off)
176 	register struct mbuf *m;
177 	int off;
178 {
179 	int iphlen = off;
180 	register struct ip *ip;
181 	register struct udphdr *uh;
182 	register struct inpcb *inp;
183 	int len;
184 	struct ip save_ip;
185 	struct sockaddr_in udp_in;
186 #ifdef IPFIREWALL_FORWARD
187 	struct m_tag *fwd_tag;
188 #endif
189 
190 	udpstat.udps_ipackets++;
191 
192 	/*
193 	 * Strip IP options, if any; should skip this,
194 	 * make available to user, and use on returned packets,
195 	 * but we don't yet have a way to check the checksum
196 	 * with options still present.
197 	 */
198 	if (iphlen > sizeof (struct ip)) {
199 		ip_stripoptions(m, (struct mbuf *)0);
200 		iphlen = sizeof(struct ip);
201 	}
202 
203 	/*
204 	 * Get IP and UDP header together in first mbuf.
205 	 */
206 	ip = mtod(m, struct ip *);
207 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
208 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
209 			udpstat.udps_hdrops++;
210 			return;
211 		}
212 		ip = mtod(m, struct ip *);
213 	}
214 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
215 
216 	/* destination port of 0 is illegal, based on RFC768. */
217 	if (uh->uh_dport == 0)
218 		goto badunlocked;
219 
220 	/*
221 	 * Construct sockaddr format source address.
222 	 * Stuff source address and datagram in user buffer.
223 	 */
224 	bzero(&udp_in, sizeof(udp_in));
225 	udp_in.sin_len = sizeof(udp_in);
226 	udp_in.sin_family = AF_INET;
227 	udp_in.sin_port = uh->uh_sport;
228 	udp_in.sin_addr = ip->ip_src;
229 
230 	/*
231 	 * Make mbuf data length reflect UDP length.
232 	 * If not enough data to reflect UDP length, drop.
233 	 */
234 	len = ntohs((u_short)uh->uh_ulen);
235 	if (ip->ip_len != len) {
236 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
237 			udpstat.udps_badlen++;
238 			goto badunlocked;
239 		}
240 		m_adj(m, len - ip->ip_len);
241 		/* ip->ip_len = len; */
242 	}
243 	/*
244 	 * Save a copy of the IP header in case we want restore it
245 	 * for sending an ICMP error message in response.
246 	 */
247 	if (!blackhole)
248 		save_ip = *ip;
249 
250 	/*
251 	 * Checksum extended UDP header and data.
252 	 */
253 	if (uh->uh_sum) {
254 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
255 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
256 				uh->uh_sum = m->m_pkthdr.csum_data;
257 			else
258 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
259 				    ip->ip_dst.s_addr, htonl((u_short)len +
260 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
261 			uh->uh_sum ^= 0xffff;
262 		} else {
263 			char b[9];
264 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
265 			bzero(((struct ipovly *)ip)->ih_x1, 9);
266 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
267 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
268 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
269 		}
270 		if (uh->uh_sum) {
271 			udpstat.udps_badsum++;
272 			m_freem(m);
273 			return;
274 		}
275 	} else
276 		udpstat.udps_nosum++;
277 
278 #ifdef IPFIREWALL_FORWARD
279 	/* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
280 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
281 
282 	if (fwd_tag != NULL) {
283 		struct sockaddr_in *next_hop;
284 
285 		/* Do the hack. */
286 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
287 		ip->ip_dst = next_hop->sin_addr;
288 		uh->uh_dport = ntohs(next_hop->sin_port);
289 		/* Remove the tag from the packet.  We don't need it anymore. */
290 		m_tag_delete(m, fwd_tag);
291 	}
292 #endif
293 
294 	INP_INFO_RLOCK(&udbinfo);
295 
296 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
297 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
298 		struct inpcb *last;
299 		/*
300 		 * Deliver a multicast or broadcast datagram to *all* sockets
301 		 * for which the local and remote addresses and ports match
302 		 * those of the incoming datagram.  This allows more than
303 		 * one process to receive multi/broadcasts on the same port.
304 		 * (This really ought to be done for unicast datagrams as
305 		 * well, but that would cause problems with existing
306 		 * applications that open both address-specific sockets and
307 		 * a wildcard socket listening to the same port -- they would
308 		 * end up receiving duplicates of every unicast datagram.
309 		 * Those applications open the multiple sockets to overcome an
310 		 * inadequacy of the UDP socket interface, but for backwards
311 		 * compatibility we avoid the problem here rather than
312 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
313 		 */
314 
315 		/*
316 		 * Locate pcb(s) for datagram.
317 		 * (Algorithm copied from raw_intr().)
318 		 */
319 		last = NULL;
320 		LIST_FOREACH(inp, &udb, inp_list) {
321 			if (inp->inp_lport != uh->uh_dport)
322 				continue;
323 #ifdef INET6
324 			if ((inp->inp_vflag & INP_IPV4) == 0)
325 				continue;
326 #endif
327 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
328 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
329 					continue;
330 			}
331 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
332 				if (inp->inp_faddr.s_addr !=
333 				    ip->ip_src.s_addr ||
334 				    inp->inp_fport != uh->uh_sport)
335 					continue;
336 			}
337 			INP_LOCK(inp);
338 
339 			/*
340 			 * Check multicast packets to make sure they are only
341 			 * sent to sockets with multicast memberships for the
342 			 * packet's destination address and arrival interface
343 			 */
344 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
345 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
346 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
347 				int mship, foundmship = 0;
348 
349 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
350 					if (MSHIP(inp, mship)->inm_addr.s_addr
351 					    == ip->ip_dst.s_addr &&
352 					    MSHIP(inp, mship)->inm_ifp
353 					    == m->m_pkthdr.rcvif) {
354 						foundmship = 1;
355 						break;
356 					}
357 				}
358 				if (foundmship == 0) {
359 					INP_UNLOCK(inp);
360 					continue;
361 				}
362 			}
363 #undef NMSHIPS
364 #undef MSHIP
365 			if (last != NULL) {
366 				struct mbuf *n;
367 
368 				n = m_copy(m, 0, M_COPYALL);
369 				if (n != NULL)
370 					udp_append(last, ip, n,
371 						   iphlen +
372 						   sizeof(struct udphdr),
373 						   &udp_in);
374 				INP_UNLOCK(last);
375 			}
376 			last = inp;
377 			/*
378 			 * Don't look for additional matches if this one does
379 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
380 			 * socket options set.  This heuristic avoids searching
381 			 * through all pcbs in the common case of a non-shared
382 			 * port.  It assumes that an application will never
383 			 * clear these options after setting them.
384 			 */
385 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
386 				break;
387 		}
388 
389 		if (last == NULL) {
390 			/*
391 			 * No matching pcb found; discard datagram.
392 			 * (No need to send an ICMP Port Unreachable
393 			 * for a broadcast or multicast datgram.)
394 			 */
395 			udpstat.udps_noportbcast++;
396 			goto badheadlocked;
397 		}
398 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
399 		    &udp_in);
400 		INP_UNLOCK(last);
401 		INP_INFO_RUNLOCK(&udbinfo);
402 		return;
403 	}
404 	/*
405 	 * Locate pcb for datagram.
406 	 */
407 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
408 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
409 	if (inp == NULL) {
410 		if (log_in_vain) {
411 			char buf[4*sizeof "123"];
412 
413 			strcpy(buf, inet_ntoa(ip->ip_dst));
414 			log(LOG_INFO,
415 			    "Connection attempt to UDP %s:%d from %s:%d\n",
416 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
417 			    ntohs(uh->uh_sport));
418 		}
419 		udpstat.udps_noport++;
420 		if (m->m_flags & (M_BCAST | M_MCAST)) {
421 			udpstat.udps_noportbcast++;
422 			goto badheadlocked;
423 		}
424 		if (blackhole)
425 			goto badheadlocked;
426 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
427 			goto badheadlocked;
428 		*ip = save_ip;
429 		ip->ip_len += iphlen;
430 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
431 		INP_INFO_RUNLOCK(&udbinfo);
432 		return;
433 	}
434 	INP_LOCK(inp);
435 	/* Check the minimum TTL for socket. */
436 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
437 		goto badheadlocked;
438 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
439 	INP_UNLOCK(inp);
440 	INP_INFO_RUNLOCK(&udbinfo);
441 	return;
442 
443 badheadlocked:
444 	if (inp)
445 		INP_UNLOCK(inp);
446 	INP_INFO_RUNLOCK(&udbinfo);
447 badunlocked:
448 	m_freem(m);
449 	return;
450 }
451 
452 /*
453  * Subroutine of udp_input(), which appends the provided mbuf chain to the
454  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
455  * contains the source address.  If the socket ends up being an IPv6 socket,
456  * udp_append() will convert to a sockaddr_in6 before passing the address
457  * into the socket code.
458  */
459 static void
460 udp_append(inp, ip, n, off, udp_in)
461 	struct inpcb *inp;
462 	struct ip *ip;
463 	struct mbuf *n;
464 	int off;
465 	struct sockaddr_in *udp_in;
466 {
467 	struct sockaddr *append_sa;
468 	struct socket *so;
469 	struct mbuf *opts = 0;
470 #ifdef INET6
471 	struct sockaddr_in6 udp_in6;
472 #endif
473 
474 	INP_LOCK_ASSERT(inp);
475 
476 #if defined(IPSEC) || defined(FAST_IPSEC)
477 	/* check AH/ESP integrity. */
478 	if (ipsec4_in_reject(n, inp)) {
479 #ifdef IPSEC
480 		ipsecstat.in_polvio++;
481 #endif /*IPSEC*/
482 		m_freem(n);
483 		return;
484 	}
485 #endif /*IPSEC || FAST_IPSEC*/
486 #ifdef MAC
487 	if (mac_check_inpcb_deliver(inp, n) != 0) {
488 		m_freem(n);
489 		return;
490 	}
491 #endif
492 	if (inp->inp_flags & INP_CONTROLOPTS ||
493 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
494 #ifdef INET6
495 		if (inp->inp_vflag & INP_IPV6) {
496 			int savedflags;
497 
498 			savedflags = inp->inp_flags;
499 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
500 			ip6_savecontrol(inp, n, &opts);
501 			inp->inp_flags = savedflags;
502 		} else
503 #endif
504 		ip_savecontrol(inp, &opts, ip, n);
505 	}
506 #ifdef INET6
507 	if (inp->inp_vflag & INP_IPV6) {
508 		bzero(&udp_in6, sizeof(udp_in6));
509 		udp_in6.sin6_len = sizeof(udp_in6);
510 		udp_in6.sin6_family = AF_INET6;
511 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
512 		append_sa = (struct sockaddr *)&udp_in6;
513 	} else
514 #endif
515 	append_sa = (struct sockaddr *)udp_in;
516 	m_adj(n, off);
517 
518 	so = inp->inp_socket;
519 	SOCKBUF_LOCK(&so->so_rcv);
520 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
521 		m_freem(n);
522 		if (opts)
523 			m_freem(opts);
524 		udpstat.udps_fullsock++;
525 		SOCKBUF_UNLOCK(&so->so_rcv);
526 	} else
527 		sorwakeup_locked(so);
528 }
529 
530 /*
531  * Notify a udp user of an asynchronous error;
532  * just wake up so that he can collect error status.
533  */
534 struct inpcb *
535 udp_notify(inp, errno)
536 	register struct inpcb *inp;
537 	int errno;
538 {
539 	inp->inp_socket->so_error = errno;
540 	sorwakeup(inp->inp_socket);
541 	sowwakeup(inp->inp_socket);
542 	return inp;
543 }
544 
545 void
546 udp_ctlinput(cmd, sa, vip)
547 	int cmd;
548 	struct sockaddr *sa;
549 	void *vip;
550 {
551 	struct ip *ip = vip;
552 	struct udphdr *uh;
553 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
554 	struct in_addr faddr;
555 	struct inpcb *inp;
556 
557 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
558 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
559 		return;
560 
561 	/*
562 	 * Redirects don't need to be handled up here.
563 	 */
564 	if (PRC_IS_REDIRECT(cmd))
565 		return;
566 	/*
567 	 * Hostdead is ugly because it goes linearly through all PCBs.
568 	 * XXX: We never get this from ICMP, otherwise it makes an
569 	 * excellent DoS attack on machines with many connections.
570 	 */
571 	if (cmd == PRC_HOSTDEAD)
572 		ip = 0;
573 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
574 		return;
575 	if (ip) {
576 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
577 		INP_INFO_RLOCK(&udbinfo);
578 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
579 		    ip->ip_src, uh->uh_sport, 0, NULL);
580 		if (inp != NULL) {
581 			INP_LOCK(inp);
582 			if (inp->inp_socket != NULL) {
583 				(*notify)(inp, inetctlerrmap[cmd]);
584 			}
585 			INP_UNLOCK(inp);
586 		}
587 		INP_INFO_RUNLOCK(&udbinfo);
588 	} else
589 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
590 }
591 
592 static int
593 udp_pcblist(SYSCTL_HANDLER_ARGS)
594 {
595 	int error, i, n;
596 	struct inpcb *inp, **inp_list;
597 	inp_gen_t gencnt;
598 	struct xinpgen xig;
599 
600 	/*
601 	 * The process of preparing the TCB list is too time-consuming and
602 	 * resource-intensive to repeat twice on every request.
603 	 */
604 	if (req->oldptr == 0) {
605 		n = udbinfo.ipi_count;
606 		req->oldidx = 2 * (sizeof xig)
607 			+ (n + n/8) * sizeof(struct xinpcb);
608 		return 0;
609 	}
610 
611 	if (req->newptr != 0)
612 		return EPERM;
613 
614 	/*
615 	 * OK, now we're committed to doing something.
616 	 */
617 	INP_INFO_RLOCK(&udbinfo);
618 	gencnt = udbinfo.ipi_gencnt;
619 	n = udbinfo.ipi_count;
620 	INP_INFO_RUNLOCK(&udbinfo);
621 
622 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
623 		+ n * sizeof(struct xinpcb));
624 	if (error != 0)
625 		return (error);
626 
627 	xig.xig_len = sizeof xig;
628 	xig.xig_count = n;
629 	xig.xig_gen = gencnt;
630 	xig.xig_sogen = so_gencnt;
631 	error = SYSCTL_OUT(req, &xig, sizeof xig);
632 	if (error)
633 		return error;
634 
635 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
636 	if (inp_list == 0)
637 		return ENOMEM;
638 
639 	INP_INFO_RLOCK(&udbinfo);
640 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
641 	     inp = LIST_NEXT(inp, inp_list)) {
642 		INP_LOCK(inp);
643 		if (inp->inp_gencnt <= gencnt &&
644 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
645 			inp_list[i++] = inp;
646 		INP_UNLOCK(inp);
647 	}
648 	INP_INFO_RUNLOCK(&udbinfo);
649 	n = i;
650 
651 	error = 0;
652 	for (i = 0; i < n; i++) {
653 		inp = inp_list[i];
654 		INP_LOCK(inp);
655 		if (inp->inp_gencnt <= gencnt) {
656 			struct xinpcb xi;
657 			bzero(&xi, sizeof(xi));
658 			xi.xi_len = sizeof xi;
659 			/* XXX should avoid extra copy */
660 			bcopy(inp, &xi.xi_inp, sizeof *inp);
661 			if (inp->inp_socket)
662 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
663 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
664 			INP_UNLOCK(inp);
665 			error = SYSCTL_OUT(req, &xi, sizeof xi);
666 		} else
667 			INP_UNLOCK(inp);
668 	}
669 	if (!error) {
670 		/*
671 		 * Give the user an updated idea of our state.
672 		 * If the generation differs from what we told
673 		 * her before, she knows that something happened
674 		 * while we were processing this request, and it
675 		 * might be necessary to retry.
676 		 */
677 		INP_INFO_RLOCK(&udbinfo);
678 		xig.xig_gen = udbinfo.ipi_gencnt;
679 		xig.xig_sogen = so_gencnt;
680 		xig.xig_count = udbinfo.ipi_count;
681 		INP_INFO_RUNLOCK(&udbinfo);
682 		error = SYSCTL_OUT(req, &xig, sizeof xig);
683 	}
684 	free(inp_list, M_TEMP);
685 	return error;
686 }
687 
688 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
689 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
690 
691 static int
692 udp_getcred(SYSCTL_HANDLER_ARGS)
693 {
694 	struct xucred xuc;
695 	struct sockaddr_in addrs[2];
696 	struct inpcb *inp;
697 	int error;
698 
699 	error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED,
700 	    SUSER_ALLOWJAIL);
701 	if (error)
702 		return (error);
703 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
704 	if (error)
705 		return (error);
706 	INP_INFO_RLOCK(&udbinfo);
707 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
708 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
709 	if (inp == NULL || inp->inp_socket == NULL) {
710 		error = ENOENT;
711 		goto out;
712 	}
713 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
714 	if (error)
715 		goto out;
716 	cru2x(inp->inp_socket->so_cred, &xuc);
717 out:
718 	INP_INFO_RUNLOCK(&udbinfo);
719 	if (error == 0)
720 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
721 	return (error);
722 }
723 
724 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
725     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
726     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
727 
728 static int
729 udp_output(inp, m, addr, control, td)
730 	register struct inpcb *inp;
731 	struct mbuf *m;
732 	struct sockaddr *addr;
733 	struct mbuf *control;
734 	struct thread *td;
735 {
736 	register struct udpiphdr *ui;
737 	register int len = m->m_pkthdr.len;
738 	struct in_addr faddr, laddr;
739 	struct cmsghdr *cm;
740 	struct sockaddr_in *sin, src;
741 	int error = 0;
742 	int ipflags;
743 	u_short fport, lport;
744 	int unlock_udbinfo;
745 
746 	/*
747 	 * udp_output() may need to temporarily bind or connect the current
748 	 * inpcb.  As such, we don't know up front what inpcb locks we will
749 	 * need.  Do any work to decide what is needed up front before
750 	 * acquiring locks.
751 	 */
752 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
753 		if (control)
754 			m_freem(control);
755 		m_freem(m);
756 		return EMSGSIZE;
757 	}
758 
759 	src.sin_addr.s_addr = INADDR_ANY;
760 	if (control != NULL) {
761 		/*
762 		 * XXX: Currently, we assume all the optional information
763 		 * is stored in a single mbuf.
764 		 */
765 		if (control->m_next) {
766 			m_freem(control);
767 			m_freem(m);
768 			return EINVAL;
769 		}
770 		for (; control->m_len > 0;
771 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
772 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
773 			cm = mtod(control, struct cmsghdr *);
774 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
775 			    cm->cmsg_len > control->m_len) {
776 				error = EINVAL;
777 				break;
778 			}
779 			if (cm->cmsg_level != IPPROTO_IP)
780 				continue;
781 
782 			switch (cm->cmsg_type) {
783 			case IP_SENDSRCADDR:
784 				if (cm->cmsg_len !=
785 				    CMSG_LEN(sizeof(struct in_addr))) {
786 					error = EINVAL;
787 					break;
788 				}
789 				bzero(&src, sizeof(src));
790 				src.sin_family = AF_INET;
791 				src.sin_len = sizeof(src);
792 				src.sin_port = inp->inp_lport;
793 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
794 				break;
795 			default:
796 				error = ENOPROTOOPT;
797 				break;
798 			}
799 			if (error)
800 				break;
801 		}
802 		m_freem(control);
803 	}
804 	if (error) {
805 		m_freem(m);
806 		return error;
807 	}
808 
809 	if (src.sin_addr.s_addr != INADDR_ANY ||
810 	    addr != NULL) {
811 		INP_INFO_WLOCK(&udbinfo);
812 		unlock_udbinfo = 1;
813 	} else
814 		unlock_udbinfo = 0;
815 	INP_LOCK(inp);
816 
817 #ifdef MAC
818 	mac_create_mbuf_from_inpcb(inp, m);
819 #endif
820 
821 	laddr = inp->inp_laddr;
822 	lport = inp->inp_lport;
823 	if (src.sin_addr.s_addr != INADDR_ANY) {
824 		if (lport == 0) {
825 			error = EINVAL;
826 			goto release;
827 		}
828 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
829 		    &laddr.s_addr, &lport, td->td_ucred);
830 		if (error)
831 			goto release;
832 	}
833 
834 	if (addr) {
835 		sin = (struct sockaddr_in *)addr;
836 		if (jailed(td->td_ucred))
837 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
838 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
839 			error = EISCONN;
840 			goto release;
841 		}
842 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
843 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
844 		if (error)
845 			goto release;
846 
847 		/* Commit the local port if newly assigned. */
848 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
849 		    inp->inp_lport == 0) {
850 			/*
851 			 * Remember addr if jailed, to prevent rebinding.
852 			 */
853 			if (jailed(td->td_ucred))
854 				inp->inp_laddr = laddr;
855 			inp->inp_lport = lport;
856 			if (in_pcbinshash(inp) != 0) {
857 				inp->inp_lport = 0;
858 				error = EAGAIN;
859 				goto release;
860 			}
861 			inp->inp_flags |= INP_ANONPORT;
862 		}
863 	} else {
864 		faddr = inp->inp_faddr;
865 		fport = inp->inp_fport;
866 		if (faddr.s_addr == INADDR_ANY) {
867 			error = ENOTCONN;
868 			goto release;
869 		}
870 	}
871 
872 	/*
873 	 * Calculate data length and get a mbuf for UDP, IP, and possible
874 	 * link-layer headers.  Immediate slide the data pointer back forward
875 	 * since we won't use that space at this layer.
876 	 */
877 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
878 	if (m == NULL) {
879 		error = ENOBUFS;
880 		goto release;
881 	}
882 	m->m_data += max_linkhdr;
883 	m->m_len -= max_linkhdr;
884 	m->m_pkthdr.len -= max_linkhdr;
885 
886 	/*
887 	 * Fill in mbuf with extended UDP header
888 	 * and addresses and length put into network format.
889 	 */
890 	ui = mtod(m, struct udpiphdr *);
891 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
892 	ui->ui_pr = IPPROTO_UDP;
893 	ui->ui_src = laddr;
894 	ui->ui_dst = faddr;
895 	ui->ui_sport = lport;
896 	ui->ui_dport = fport;
897 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
898 
899 	/*
900 	 * Set the Don't Fragment bit in the IP header.
901 	 */
902 	if (inp->inp_flags & INP_DONTFRAG) {
903 		struct ip *ip;
904 		ip = (struct ip *)&ui->ui_i;
905 		ip->ip_off |= IP_DF;
906 	}
907 
908 	ipflags = 0;
909 	if (inp->inp_socket->so_options & SO_DONTROUTE)
910 		ipflags |= IP_ROUTETOIF;
911 	if (inp->inp_socket->so_options & SO_BROADCAST)
912 		ipflags |= IP_ALLOWBROADCAST;
913 	if (inp->inp_flags & INP_ONESBCAST)
914 		ipflags |= IP_SENDONES;
915 
916 	/*
917 	 * Set up checksum and output datagram.
918 	 */
919 	if (udpcksum) {
920 		if (inp->inp_flags & INP_ONESBCAST)
921 			faddr.s_addr = INADDR_BROADCAST;
922 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
923 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
924 		m->m_pkthdr.csum_flags = CSUM_UDP;
925 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
926 	} else {
927 		ui->ui_sum = 0;
928 	}
929 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
930 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
931 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
932 	udpstat.udps_opackets++;
933 
934 	if (unlock_udbinfo)
935 		INP_INFO_WUNLOCK(&udbinfo);
936 	error = ip_output(m, inp->inp_options, NULL, ipflags,
937 	    inp->inp_moptions, inp);
938 	INP_UNLOCK(inp);
939 	return (error);
940 
941 release:
942 	INP_UNLOCK(inp);
943 	if (unlock_udbinfo)
944 		INP_INFO_WUNLOCK(&udbinfo);
945 	m_freem(m);
946 	return (error);
947 }
948 
949 u_long	udp_sendspace = 9216;		/* really max datagram size */
950 					/* 40 1K datagrams */
951 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
952     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
953 
954 u_long	udp_recvspace = 40 * (1024 +
955 #ifdef INET6
956 				      sizeof(struct sockaddr_in6)
957 #else
958 				      sizeof(struct sockaddr_in)
959 #endif
960 				      );
961 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
962     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
963 
964 static void
965 udp_abort(struct socket *so)
966 {
967 	struct inpcb *inp;
968 
969 	inp = sotoinpcb(so);
970 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
971 	INP_INFO_WLOCK(&udbinfo);
972 	INP_LOCK(inp);
973 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
974 		in_pcbdisconnect(inp);
975 		inp->inp_laddr.s_addr = INADDR_ANY;
976 		soisdisconnected(so);
977 	}
978 	INP_UNLOCK(inp);
979 	INP_INFO_WUNLOCK(&udbinfo);
980 }
981 
982 static int
983 udp_attach(struct socket *so, int proto, struct thread *td)
984 {
985 	struct inpcb *inp;
986 	int error;
987 
988 	inp = sotoinpcb(so);
989 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
990 	error = soreserve(so, udp_sendspace, udp_recvspace);
991 	if (error)
992 		return error;
993 	INP_INFO_WLOCK(&udbinfo);
994 	error = in_pcballoc(so, &udbinfo);
995 	if (error) {
996 		INP_INFO_WUNLOCK(&udbinfo);
997 		return error;
998 	}
999 
1000 	inp = (struct inpcb *)so->so_pcb;
1001 	INP_INFO_WUNLOCK(&udbinfo);
1002 	inp->inp_vflag |= INP_IPV4;
1003 	inp->inp_ip_ttl = ip_defttl;
1004 	INP_UNLOCK(inp);
1005 	return 0;
1006 }
1007 
1008 static int
1009 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1010 {
1011 	struct inpcb *inp;
1012 	int error;
1013 
1014 	inp = sotoinpcb(so);
1015 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1016 	INP_INFO_WLOCK(&udbinfo);
1017 	INP_LOCK(inp);
1018 	error = in_pcbbind(inp, nam, td->td_ucred);
1019 	INP_UNLOCK(inp);
1020 	INP_INFO_WUNLOCK(&udbinfo);
1021 	return error;
1022 }
1023 
1024 static void
1025 udp_close(struct socket *so)
1026 {
1027 	struct inpcb *inp;
1028 
1029 	inp = sotoinpcb(so);
1030 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1031 	INP_INFO_WLOCK(&udbinfo);
1032 	INP_LOCK(inp);
1033 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1034 		in_pcbdisconnect(inp);
1035 		inp->inp_laddr.s_addr = INADDR_ANY;
1036 		soisdisconnected(so);
1037 	}
1038 	INP_UNLOCK(inp);
1039 	INP_INFO_WUNLOCK(&udbinfo);
1040 }
1041 
1042 static int
1043 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1044 {
1045 	struct inpcb *inp;
1046 	int error;
1047 	struct sockaddr_in *sin;
1048 
1049 	inp = sotoinpcb(so);
1050 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1051 	INP_INFO_WLOCK(&udbinfo);
1052 	INP_LOCK(inp);
1053 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1054 		INP_UNLOCK(inp);
1055 		INP_INFO_WUNLOCK(&udbinfo);
1056 		return EISCONN;
1057 	}
1058 	sin = (struct sockaddr_in *)nam;
1059 	if (jailed(td->td_ucred))
1060 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1061 	error = in_pcbconnect(inp, nam, td->td_ucred);
1062 	if (error == 0)
1063 		soisconnected(so);
1064 	INP_UNLOCK(inp);
1065 	INP_INFO_WUNLOCK(&udbinfo);
1066 	return error;
1067 }
1068 
1069 static void
1070 udp_detach(struct socket *so)
1071 {
1072 	struct inpcb *inp;
1073 
1074 	inp = sotoinpcb(so);
1075 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1076 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1077 	    ("udp_detach: not disconnected"));
1078 	INP_INFO_WLOCK(&udbinfo);
1079 	INP_LOCK(inp);
1080 	in_pcbdetach(inp);
1081 	in_pcbfree(inp);
1082 	INP_INFO_WUNLOCK(&udbinfo);
1083 }
1084 
1085 static int
1086 udp_disconnect(struct socket *so)
1087 {
1088 	struct inpcb *inp;
1089 
1090 	inp = sotoinpcb(so);
1091 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1092 	INP_INFO_WLOCK(&udbinfo);
1093 	INP_LOCK(inp);
1094 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1095 		INP_INFO_WUNLOCK(&udbinfo);
1096 		INP_UNLOCK(inp);
1097 		return ENOTCONN;
1098 	}
1099 
1100 	in_pcbdisconnect(inp);
1101 	inp->inp_laddr.s_addr = INADDR_ANY;
1102 	SOCK_LOCK(so);
1103 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1104 	SOCK_UNLOCK(so);
1105 	INP_UNLOCK(inp);
1106 	INP_INFO_WUNLOCK(&udbinfo);
1107 	return 0;
1108 }
1109 
1110 static int
1111 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1112 	    struct mbuf *control, struct thread *td)
1113 {
1114 	struct inpcb *inp;
1115 
1116 	inp = sotoinpcb(so);
1117 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1118 	return udp_output(inp, m, addr, control, td);
1119 }
1120 
1121 int
1122 udp_shutdown(struct socket *so)
1123 {
1124 	struct inpcb *inp;
1125 
1126 	inp = sotoinpcb(so);
1127 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1128 	INP_LOCK(inp);
1129 	socantsendmore(so);
1130 	INP_UNLOCK(inp);
1131 	return 0;
1132 }
1133 
1134 /*
1135  * This is the wrapper function for in_setsockaddr.  We just pass down
1136  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1137  * here because in_setsockaddr will call malloc and might block.
1138  */
1139 static int
1140 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1141 {
1142 	return (in_setsockaddr(so, nam, &udbinfo));
1143 }
1144 
1145 /*
1146  * This is the wrapper function for in_setpeeraddr.  We just pass down
1147  * the pcbinfo for in_setpeeraddr to lock.
1148  */
1149 static int
1150 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1151 {
1152 	return (in_setpeeraddr(so, nam, &udbinfo));
1153 }
1154 
1155 struct pr_usrreqs udp_usrreqs = {
1156 	.pru_abort =		udp_abort,
1157 	.pru_attach =		udp_attach,
1158 	.pru_bind =		udp_bind,
1159 	.pru_connect =		udp_connect,
1160 	.pru_control =		in_control,
1161 	.pru_detach =		udp_detach,
1162 	.pru_disconnect =	udp_disconnect,
1163 	.pru_peeraddr =		udp_peeraddr,
1164 	.pru_send =		udp_send,
1165 	.pru_sosend =		sosend_dgram,
1166 	.pru_shutdown =		udp_shutdown,
1167 	.pru_sockaddr =		udp_sockaddr,
1168 	.pru_sosetlabel =	in_pcbsosetlabel,
1169 	.pru_close =		udp_close,
1170 };
1171