xref: /freebsd/sys/netinet/udp_usrreq.c (revision f6c0136c7fb87ab8277221a306291e386fe944fb)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 4. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
31  * $FreeBSD$
32  */
33 
34 #include "opt_ipfw.h"
35 #include "opt_ipsec.h"
36 #include "opt_inet6.h"
37 #include "opt_mac.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/domain.h>
42 #include <sys/eventhandler.h>
43 #include <sys/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/protosw.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/sx.h>
55 #include <sys/sysctl.h>
56 #include <sys/syslog.h>
57 
58 #include <vm/uma.h>
59 
60 #include <net/if.h>
61 #include <net/route.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/ip_icmp.h>
72 #include <netinet/icmp_var.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_options.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #endif
78 #include <netinet/udp.h>
79 #include <netinet/udp_var.h>
80 
81 #ifdef FAST_IPSEC
82 #include <netipsec/ipsec.h>
83 #endif
84 
85 #ifdef IPSEC
86 #include <netinet6/ipsec.h>
87 #endif
88 
89 #include <machine/in_cksum.h>
90 
91 #include <security/mac/mac_framework.h>
92 
93 /*
94  * UDP protocol implementation.
95  * Per RFC 768, August, 1980.
96  */
97 
98 /*
99  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
100  * removes the only data integrity mechanism for packets and malformed
101  * packets that would otherwise be discarded by bad checksums may cause
102  * problems (especially for NFS data blocks).
103  */
104 static int	udpcksum = 1;
105 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udpcksum,
106     0, "");
107 
108 int	udp_log_in_vain = 0;
109 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
110     &udp_log_in_vain, 0, "Log all incoming UDP packets");
111 
112 static int	blackhole = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, &blackhole, 0,
114     "Do not send port unreachables for refused connects");
115 
116 static int	strict_mcast_mship = 0;
117 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
118     &strict_mcast_mship, 0, "Only send multicast to member sockets");
119 
120 struct	inpcbhead udb;		/* from udp_var.h */
121 struct	inpcbinfo udbinfo;
122 
123 #ifndef UDBHASHSIZE
124 #define UDBHASHSIZE 16
125 #endif
126 
127 struct	udpstat udpstat;	/* from udp_var.h */
128 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW, &udpstat,
129     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
130 
131 static void	udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
132 		    int off, struct sockaddr_in *udp_in);
133 
134 static void	udp_detach(struct socket *so);
135 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
136 		    struct mbuf *, struct thread *);
137 
138 static void
139 udp_zone_change(void *tag)
140 {
141 
142 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
143 }
144 
145 static int
146 udp_inpcb_init(void *mem, int size, int flags)
147 {
148 	struct inpcb *inp = mem;
149 
150 	INP_LOCK_INIT(inp, "inp", "udpinp");
151 	return (0);
152 }
153 
154 void
155 udp_init()
156 {
157 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
158 	LIST_INIT(&udb);
159 	udbinfo.listhead = &udb;
160 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
161 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
162 	    &udbinfo.porthashmask);
163 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
164 	    NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
165 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
166 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
167 		EVENTHANDLER_PRI_ANY);
168 }
169 
170 void
171 udp_input(struct mbuf *m, int off)
172 {
173 	int iphlen = off;
174 	struct ip *ip;
175 	struct udphdr *uh;
176 	struct inpcb *inp;
177 	int len;
178 	struct ip save_ip;
179 	struct sockaddr_in udp_in;
180 #ifdef IPFIREWALL_FORWARD
181 	struct m_tag *fwd_tag;
182 #endif
183 
184 	udpstat.udps_ipackets++;
185 
186 	/*
187 	 * Strip IP options, if any; should skip this, make available to
188 	 * user, and use on returned packets, but we don't yet have a way to
189 	 * check the checksum with options still present.
190 	 */
191 	if (iphlen > sizeof (struct ip)) {
192 		ip_stripoptions(m, (struct mbuf *)0);
193 		iphlen = sizeof(struct ip);
194 	}
195 
196 	/*
197 	 * Get IP and UDP header together in first mbuf.
198 	 */
199 	ip = mtod(m, struct ip *);
200 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
201 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
202 			udpstat.udps_hdrops++;
203 			return;
204 		}
205 		ip = mtod(m, struct ip *);
206 	}
207 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
208 
209 	/*
210 	 * Destination port of 0 is illegal, based on RFC768.
211 	 */
212 	if (uh->uh_dport == 0)
213 		goto badunlocked;
214 
215 	/*
216 	 * Construct sockaddr format source address.  Stuff source address
217 	 * and datagram in user buffer.
218 	 */
219 	bzero(&udp_in, sizeof(udp_in));
220 	udp_in.sin_len = sizeof(udp_in);
221 	udp_in.sin_family = AF_INET;
222 	udp_in.sin_port = uh->uh_sport;
223 	udp_in.sin_addr = ip->ip_src;
224 
225 	/*
226 	 * Make mbuf data length reflect UDP length.
227 	 * If not enough data to reflect UDP length, drop.
228 	 */
229 	len = ntohs((u_short)uh->uh_ulen);
230 	if (ip->ip_len != len) {
231 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
232 			udpstat.udps_badlen++;
233 			goto badunlocked;
234 		}
235 		m_adj(m, len - ip->ip_len);
236 		/* ip->ip_len = len; */
237 	}
238 
239 	/*
240 	 * Save a copy of the IP header in case we want restore it for
241 	 * sending an ICMP error message in response.
242 	 */
243 	if (!blackhole)
244 		save_ip = *ip;
245 
246 	/*
247 	 * Checksum extended UDP header and data.
248 	 */
249 	if (uh->uh_sum) {
250 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
251 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
252 				uh->uh_sum = m->m_pkthdr.csum_data;
253 			else
254 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
255 				    ip->ip_dst.s_addr, htonl((u_short)len +
256 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
257 			uh->uh_sum ^= 0xffff;
258 		} else {
259 			char b[9];
260 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
261 			bzero(((struct ipovly *)ip)->ih_x1, 9);
262 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
263 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
264 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
265 		}
266 		if (uh->uh_sum) {
267 			udpstat.udps_badsum++;
268 			m_freem(m);
269 			return;
270 		}
271 	} else
272 		udpstat.udps_nosum++;
273 
274 #ifdef IPFIREWALL_FORWARD
275 	/*
276 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
277 	 */
278 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
279 	if (fwd_tag != NULL) {
280 		struct sockaddr_in *next_hop;
281 
282 		/*
283 		 * Do the hack.
284 		 */
285 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
286 		ip->ip_dst = next_hop->sin_addr;
287 		uh->uh_dport = ntohs(next_hop->sin_port);
288 
289 		/*
290 		 * Remove the tag from the packet.  We don't need it anymore.
291 		 */
292 		m_tag_delete(m, fwd_tag);
293 	}
294 #endif
295 
296 	INP_INFO_RLOCK(&udbinfo);
297 
298 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
299 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
300 		struct inpcb *last;
301 
302 		/*
303 		 * Deliver a multicast or broadcast datagram to *all* sockets
304 		 * for which the local and remote addresses and ports match
305 		 * those of the incoming datagram.  This allows more than one
306 		 * process to receive multi/broadcasts on the same port.
307 		 * (This really ought to be done for unicast datagrams as
308 		 * well, but that would cause problems with existing
309 		 * applications that open both address-specific sockets and a
310 		 * wildcard socket listening to the same port -- they would
311 		 * end up receiving duplicates of every unicast datagram.
312 		 * Those applications open the multiple sockets to overcome
313 		 * an inadequacy of the UDP socket interface, but for
314 		 * backwards compatibility we avoid the problem here rather
315 		 * than fixing the interface.  Maybe 4.5BSD will remedy
316 		 * this?)
317 		 */
318 		last = NULL;
319 		LIST_FOREACH(inp, &udb, inp_list) {
320 			if (inp->inp_lport != uh->uh_dport)
321 				continue;
322 #ifdef INET6
323 			if ((inp->inp_vflag & INP_IPV4) == 0)
324 				continue;
325 #endif
326 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
327 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
328 					continue;
329 			}
330 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
331 				if (inp->inp_faddr.s_addr !=
332 				    ip->ip_src.s_addr ||
333 				    inp->inp_fport != uh->uh_sport)
334 					continue;
335 			}
336 
337 			/*
338 			 * Check multicast packets to make sure they are only
339 			 * sent to sockets with multicast memberships for the
340 			 * packet's destination address and arrival interface
341 			 */
342 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
343 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
344 			INP_LOCK(inp);
345 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
346 				int mship, foundmship = 0;
347 
348 				for (mship = 0; mship < NMSHIPS(inp);
349 				    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, iphlen +
371 					    sizeof(struct udphdr), &udp_in);
372 				INP_UNLOCK(last);
373 			}
374 			last = inp;
375 			/*
376 			 * Don't look for additional matches if this one does
377 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
378 			 * socket options set.  This heuristic avoids
379 			 * searching through all pcbs in the common case of a
380 			 * non-shared port.  It assumes that an application
381 			 * will never clear these options after setting them.
382 			 */
383 			if ((last->inp_socket->so_options &
384 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
385 				break;
386 		}
387 
388 		if (last == NULL) {
389 			/*
390 			 * No matching pcb found; discard datagram.  (No need
391 			 * to send an ICMP Port Unreachable for a broadcast
392 			 * or multicast datgram.)
393 			 */
394 			udpstat.udps_noportbcast++;
395 			goto badheadlocked;
396 		}
397 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
398 		    &udp_in);
399 		INP_UNLOCK(last);
400 		INP_INFO_RUNLOCK(&udbinfo);
401 		return;
402 	}
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 (udp_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 
435 	/*
436 	 * Check the minimum TTL for socket.
437 	 */
438 	INP_LOCK(inp);
439 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
440 		goto badheadlocked;
441 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
442 	INP_UNLOCK(inp);
443 	INP_INFO_RUNLOCK(&udbinfo);
444 	return;
445 
446 badheadlocked:
447 	if (inp)
448 		INP_UNLOCK(inp);
449 	INP_INFO_RUNLOCK(&udbinfo);
450 badunlocked:
451 	m_freem(m);
452 }
453 
454 /*
455  * Subroutine of udp_input(), which appends the provided mbuf chain to the
456  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
457  * contains the source address.  If the socket ends up being an IPv6 socket,
458  * udp_append() will convert to a sockaddr_in6 before passing the address
459  * into the socket code.
460  */
461 static void
462 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
463     struct sockaddr_in *udp_in)
464 {
465 	struct sockaddr *append_sa;
466 	struct socket *so;
467 	struct mbuf *opts = 0;
468 #ifdef INET6
469 	struct sockaddr_in6 udp_in6;
470 #endif
471 
472 	INP_LOCK_ASSERT(inp);
473 
474 #if defined(IPSEC) || defined(FAST_IPSEC)
475 	/* check AH/ESP integrity. */
476 	if (ipsec4_in_reject(n, inp)) {
477 #ifdef IPSEC
478 		ipsecstat.in_polvio++;
479 #endif
480 		m_freem(n);
481 		return;
482 	}
483 #endif /*IPSEC || FAST_IPSEC*/
484 #ifdef MAC
485 	if (mac_check_inpcb_deliver(inp, n) != 0) {
486 		m_freem(n);
487 		return;
488 	}
489 #endif
490 	if (inp->inp_flags & INP_CONTROLOPTS ||
491 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
492 #ifdef INET6
493 		if (inp->inp_vflag & INP_IPV6) {
494 			int savedflags;
495 
496 			savedflags = inp->inp_flags;
497 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
498 			ip6_savecontrol(inp, n, &opts);
499 			inp->inp_flags = savedflags;
500 		} else
501 #endif
502 			ip_savecontrol(inp, &opts, ip, n);
503 	}
504 #ifdef INET6
505 	if (inp->inp_vflag & INP_IPV6) {
506 		bzero(&udp_in6, sizeof(udp_in6));
507 		udp_in6.sin6_len = sizeof(udp_in6);
508 		udp_in6.sin6_family = AF_INET6;
509 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
510 		append_sa = (struct sockaddr *)&udp_in6;
511 	} else
512 #endif
513 		append_sa = (struct sockaddr *)udp_in;
514 	m_adj(n, off);
515 
516 	so = inp->inp_socket;
517 	SOCKBUF_LOCK(&so->so_rcv);
518 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
519 		m_freem(n);
520 		if (opts)
521 			m_freem(opts);
522 		udpstat.udps_fullsock++;
523 		SOCKBUF_UNLOCK(&so->so_rcv);
524 	} else
525 		sorwakeup_locked(so);
526 }
527 
528 /*
529  * Notify a udp user of an asynchronous error; just wake up so that they can
530  * collect error status.
531  */
532 struct inpcb *
533 udp_notify(struct inpcb *inp, int errno)
534 {
535 
536 	inp->inp_socket->so_error = errno;
537 	sorwakeup(inp->inp_socket);
538 	sowwakeup(inp->inp_socket);
539 	return (inp);
540 }
541 
542 void
543 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
544 {
545 	struct ip *ip = vip;
546 	struct udphdr *uh;
547 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
548 	struct in_addr faddr;
549 	struct inpcb *inp;
550 
551 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
552 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
553 		return;
554 
555 	/*
556 	 * Redirects don't need to be handled up here.
557 	 */
558 	if (PRC_IS_REDIRECT(cmd))
559 		return;
560 
561 	/*
562 	 * Hostdead is ugly because it goes linearly through all PCBs.
563 	 *
564 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
565 	 * DoS attack on machines with many connections.
566 	 */
567 	if (cmd == PRC_HOSTDEAD)
568 		ip = 0;
569 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
570 		return;
571 	if (ip) {
572 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
573 		INP_INFO_RLOCK(&udbinfo);
574 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
575 		    ip->ip_src, uh->uh_sport, 0, NULL);
576 		if (inp != NULL) {
577 			INP_LOCK(inp);
578 			if (inp->inp_socket != NULL) {
579 				(*notify)(inp, inetctlerrmap[cmd]);
580 			}
581 			INP_UNLOCK(inp);
582 		}
583 		INP_INFO_RUNLOCK(&udbinfo);
584 	} else
585 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
586 }
587 
588 static int
589 udp_pcblist(SYSCTL_HANDLER_ARGS)
590 {
591 	int error, i, n;
592 	struct inpcb *inp, **inp_list;
593 	inp_gen_t gencnt;
594 	struct xinpgen xig;
595 
596 	/*
597 	 * The process of preparing the TCB list is too time-consuming and
598 	 * resource-intensive to repeat twice on every request.
599 	 */
600 	if (req->oldptr == 0) {
601 		n = udbinfo.ipi_count;
602 		req->oldidx = 2 * (sizeof xig)
603 			+ (n + n/8) * sizeof(struct xinpcb);
604 		return (0);
605 	}
606 
607 	if (req->newptr != 0)
608 		return (EPERM);
609 
610 	/*
611 	 * OK, now we're committed to doing something.
612 	 */
613 	INP_INFO_RLOCK(&udbinfo);
614 	gencnt = udbinfo.ipi_gencnt;
615 	n = udbinfo.ipi_count;
616 	INP_INFO_RUNLOCK(&udbinfo);
617 
618 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
619 		+ n * sizeof(struct xinpcb));
620 	if (error != 0)
621 		return (error);
622 
623 	xig.xig_len = sizeof xig;
624 	xig.xig_count = n;
625 	xig.xig_gen = gencnt;
626 	xig.xig_sogen = so_gencnt;
627 	error = SYSCTL_OUT(req, &xig, sizeof xig);
628 	if (error)
629 		return (error);
630 
631 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
632 	if (inp_list == 0)
633 		return (ENOMEM);
634 
635 	INP_INFO_RLOCK(&udbinfo);
636 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
637 	     inp = LIST_NEXT(inp, inp_list)) {
638 		INP_LOCK(inp);
639 		if (inp->inp_gencnt <= gencnt &&
640 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
641 			inp_list[i++] = inp;
642 		INP_UNLOCK(inp);
643 	}
644 	INP_INFO_RUNLOCK(&udbinfo);
645 	n = i;
646 
647 	error = 0;
648 	for (i = 0; i < n; i++) {
649 		inp = inp_list[i];
650 		INP_LOCK(inp);
651 		if (inp->inp_gencnt <= gencnt) {
652 			struct xinpcb xi;
653 			bzero(&xi, sizeof(xi));
654 			xi.xi_len = sizeof xi;
655 			/* XXX should avoid extra copy */
656 			bcopy(inp, &xi.xi_inp, sizeof *inp);
657 			if (inp->inp_socket)
658 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
659 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
660 			INP_UNLOCK(inp);
661 			error = SYSCTL_OUT(req, &xi, sizeof xi);
662 		} else
663 			INP_UNLOCK(inp);
664 	}
665 	if (!error) {
666 		/*
667 		 * Give the user an updated idea of our state.  If the
668 		 * generation differs from what we told her before, she knows
669 		 * that something happened while we were processing this
670 		 * request, and it might be necessary to retry.
671 		 */
672 		INP_INFO_RLOCK(&udbinfo);
673 		xig.xig_gen = udbinfo.ipi_gencnt;
674 		xig.xig_sogen = so_gencnt;
675 		xig.xig_count = udbinfo.ipi_count;
676 		INP_INFO_RUNLOCK(&udbinfo);
677 		error = SYSCTL_OUT(req, &xig, sizeof xig);
678 	}
679 	free(inp_list, M_TEMP);
680 	return (error);
681 }
682 
683 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
684     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
685 
686 static int
687 udp_getcred(SYSCTL_HANDLER_ARGS)
688 {
689 	struct xucred xuc;
690 	struct sockaddr_in addrs[2];
691 	struct inpcb *inp;
692 	int error;
693 
694 	error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED,
695 	    SUSER_ALLOWJAIL);
696 	if (error)
697 		return (error);
698 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
699 	if (error)
700 		return (error);
701 	INP_INFO_RLOCK(&udbinfo);
702 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
703 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
704 	if (inp == NULL || inp->inp_socket == NULL) {
705 		error = ENOENT;
706 		goto out;
707 	}
708 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
709 	if (error)
710 		goto out;
711 	cru2x(inp->inp_socket->so_cred, &xuc);
712 out:
713 	INP_INFO_RUNLOCK(&udbinfo);
714 	if (error == 0)
715 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
716 	return (error);
717 }
718 
719 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
720     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
721     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
722 
723 static int
724 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
725     struct mbuf *control, struct thread *td)
726 {
727 	struct udpiphdr *ui;
728 	int len = m->m_pkthdr.len;
729 	struct in_addr faddr, laddr;
730 	struct cmsghdr *cm;
731 	struct sockaddr_in *sin, src;
732 	int error = 0;
733 	int ipflags;
734 	u_short fport, lport;
735 	int unlock_udbinfo;
736 
737 	/*
738 	 * udp_output() may need to temporarily bind or connect the current
739 	 * inpcb.  As such, we don't know up front what inpcb locks we will
740 	 * need.  Do any work to decide what is needed up front before
741 	 * acquiring locks.
742 	 */
743 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
744 		if (control)
745 			m_freem(control);
746 		m_freem(m);
747 		return (EMSGSIZE);
748 	}
749 
750 	src.sin_family = 0;
751 	if (control != NULL) {
752 		/*
753 		 * XXX: Currently, we assume all the optional information is
754 		 * stored in a single mbuf.
755 		 */
756 		if (control->m_next) {
757 			m_freem(control);
758 			m_freem(m);
759 			return (EINVAL);
760 		}
761 		for (; control->m_len > 0;
762 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
763 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
764 			cm = mtod(control, struct cmsghdr *);
765 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
766 			    cm->cmsg_len > control->m_len) {
767 				error = EINVAL;
768 				break;
769 			}
770 			if (cm->cmsg_level != IPPROTO_IP)
771 				continue;
772 
773 			switch (cm->cmsg_type) {
774 			case IP_SENDSRCADDR:
775 				if (cm->cmsg_len !=
776 				    CMSG_LEN(sizeof(struct in_addr))) {
777 					error = EINVAL;
778 					break;
779 				}
780 				bzero(&src, sizeof(src));
781 				src.sin_family = AF_INET;
782 				src.sin_len = sizeof(src);
783 				src.sin_port = inp->inp_lport;
784 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
785 				break;
786 			default:
787 				error = ENOPROTOOPT;
788 				break;
789 			}
790 			if (error)
791 				break;
792 		}
793 		m_freem(control);
794 	}
795 	if (error) {
796 		m_freem(m);
797 		return (error);
798 	}
799 
800 	if (src.sin_family == AF_INET || addr != NULL) {
801 		INP_INFO_WLOCK(&udbinfo);
802 		unlock_udbinfo = 1;
803 	} else
804 		unlock_udbinfo = 0;
805 	INP_LOCK(inp);
806 
807 #ifdef MAC
808 	mac_create_mbuf_from_inpcb(inp, m);
809 #endif
810 
811 	/*
812 	 * If the IP_SENDSRCADDR control message was specified, override the
813 	 * source address for this datagram. Its use is invalidated if the
814 	 * address thus specified is incomplete or clobbers other inpcbs.
815 	 */
816 	laddr = inp->inp_laddr;
817 	lport = inp->inp_lport;
818 	if (src.sin_family == AF_INET) {
819 		if ((lport == 0) ||
820 		    (laddr.s_addr == INADDR_ANY &&
821 		     src.sin_addr.s_addr == INADDR_ANY)) {
822 			error = EINVAL;
823 			goto release;
824 		}
825 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
826 		    &laddr.s_addr, &lport, td->td_ucred);
827 		if (error)
828 			goto release;
829 	}
830 
831 	if (addr) {
832 		sin = (struct sockaddr_in *)addr;
833 		if (jailed(td->td_ucred))
834 			prison_remote_ip(td->td_ucred, 0,
835 			    &sin->sin_addr.s_addr);
836 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
837 			error = EISCONN;
838 			goto release;
839 		}
840 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
841 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
842 		if (error)
843 			goto release;
844 
845 		/* Commit the local port if newly assigned. */
846 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
847 		    inp->inp_lport == 0) {
848 			/*
849 			 * Remember addr if jailed, to prevent rebinding.
850 			 */
851 			if (jailed(td->td_ucred))
852 				inp->inp_laddr = laddr;
853 			inp->inp_lport = lport;
854 			if (in_pcbinshash(inp) != 0) {
855 				inp->inp_lport = 0;
856 				error = EAGAIN;
857 				goto release;
858 			}
859 			inp->inp_flags |= INP_ANONPORT;
860 		}
861 	} else {
862 		faddr = inp->inp_faddr;
863 		fport = inp->inp_fport;
864 		if (faddr.s_addr == INADDR_ANY) {
865 			error = ENOTCONN;
866 			goto release;
867 		}
868 	}
869 
870 	/*
871 	 * Calculate data length and get a mbuf for UDP, IP, and possible
872 	 * link-layer headers.  Immediate slide the data pointer back forward
873 	 * since we won't use that space at this layer.
874 	 */
875 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
876 	if (m == NULL) {
877 		error = ENOBUFS;
878 		goto release;
879 	}
880 	m->m_data += max_linkhdr;
881 	m->m_len -= max_linkhdr;
882 	m->m_pkthdr.len -= max_linkhdr;
883 
884 	/*
885 	 * Fill in mbuf with extended UDP header and addresses and length put
886 	 * into network format.
887 	 */
888 	ui = mtod(m, struct udpiphdr *);
889 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
890 	ui->ui_pr = IPPROTO_UDP;
891 	ui->ui_src = laddr;
892 	ui->ui_dst = faddr;
893 	ui->ui_sport = lport;
894 	ui->ui_dport = fport;
895 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
896 
897 	/*
898 	 * Set the Don't Fragment bit in the IP header.
899 	 */
900 	if (inp->inp_flags & INP_DONTFRAG) {
901 		struct ip *ip;
902 
903 		ip = (struct ip *)&ui->ui_i;
904 		ip->ip_off |= IP_DF;
905 	}
906 
907 	ipflags = 0;
908 	if (inp->inp_socket->so_options & SO_DONTROUTE)
909 		ipflags |= IP_ROUTETOIF;
910 	if (inp->inp_socket->so_options & SO_BROADCAST)
911 		ipflags |= IP_ALLOWBROADCAST;
912 	if (inp->inp_flags & INP_ONESBCAST)
913 		ipflags |= IP_SENDONES;
914 
915 	/*
916 	 * Set up checksum and output datagram.
917 	 */
918 	if (udpcksum) {
919 		if (inp->inp_flags & INP_ONESBCAST)
920 			faddr.s_addr = INADDR_BROADCAST;
921 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
922 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
923 		m->m_pkthdr.csum_flags = CSUM_UDP;
924 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
925 	} else
926 		ui->ui_sum = 0;
927 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
928 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
929 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
930 	udpstat.udps_opackets++;
931 
932 	if (unlock_udbinfo)
933 		INP_INFO_WUNLOCK(&udbinfo);
934 	error = ip_output(m, inp->inp_options, NULL, ipflags,
935 	    inp->inp_moptions, inp);
936 	INP_UNLOCK(inp);
937 	return (error);
938 
939 release:
940 	INP_UNLOCK(inp);
941 	if (unlock_udbinfo)
942 		INP_INFO_WUNLOCK(&udbinfo);
943 	m_freem(m);
944 	return (error);
945 }
946 
947 u_long	udp_sendspace = 9216;		/* really max datagram size */
948 					/* 40 1K datagrams */
949 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
950     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
951 
952 u_long	udp_recvspace = 40 * (1024 +
953 #ifdef INET6
954 				      sizeof(struct sockaddr_in6)
955 #else
956 				      sizeof(struct sockaddr_in)
957 #endif
958 				      );
959 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
960     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
961 
962 static void
963 udp_abort(struct socket *so)
964 {
965 	struct inpcb *inp;
966 
967 	inp = sotoinpcb(so);
968 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
969 	INP_INFO_WLOCK(&udbinfo);
970 	INP_LOCK(inp);
971 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
972 		in_pcbdisconnect(inp);
973 		inp->inp_laddr.s_addr = INADDR_ANY;
974 		soisdisconnected(so);
975 	}
976 	INP_UNLOCK(inp);
977 	INP_INFO_WUNLOCK(&udbinfo);
978 }
979 
980 static int
981 udp_attach(struct socket *so, int proto, struct thread *td)
982 {
983 	struct inpcb *inp;
984 	int error;
985 
986 	inp = sotoinpcb(so);
987 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
988 	error = soreserve(so, udp_sendspace, udp_recvspace);
989 	if (error)
990 		return (error);
991 	INP_INFO_WLOCK(&udbinfo);
992 	error = in_pcballoc(so, &udbinfo);
993 	if (error) {
994 		INP_INFO_WUNLOCK(&udbinfo);
995 		return (error);
996 	}
997 
998 	inp = (struct inpcb *)so->so_pcb;
999 	INP_INFO_WUNLOCK(&udbinfo);
1000 	inp->inp_vflag |= INP_IPV4;
1001 	inp->inp_ip_ttl = ip_defttl;
1002 	INP_UNLOCK(inp);
1003 	return (0);
1004 }
1005 
1006 static int
1007 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1008 {
1009 	struct inpcb *inp;
1010 	int error;
1011 
1012 	inp = sotoinpcb(so);
1013 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1014 	INP_INFO_WLOCK(&udbinfo);
1015 	INP_LOCK(inp);
1016 	error = in_pcbbind(inp, nam, td->td_ucred);
1017 	INP_UNLOCK(inp);
1018 	INP_INFO_WUNLOCK(&udbinfo);
1019 	return (error);
1020 }
1021 
1022 static void
1023 udp_close(struct socket *so)
1024 {
1025 	struct inpcb *inp;
1026 
1027 	inp = sotoinpcb(so);
1028 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1029 	INP_INFO_WLOCK(&udbinfo);
1030 	INP_LOCK(inp);
1031 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1032 		in_pcbdisconnect(inp);
1033 		inp->inp_laddr.s_addr = INADDR_ANY;
1034 		soisdisconnected(so);
1035 	}
1036 	INP_UNLOCK(inp);
1037 	INP_INFO_WUNLOCK(&udbinfo);
1038 }
1039 
1040 static int
1041 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1042 {
1043 	struct inpcb *inp;
1044 	int error;
1045 	struct sockaddr_in *sin;
1046 
1047 	inp = sotoinpcb(so);
1048 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1049 	INP_INFO_WLOCK(&udbinfo);
1050 	INP_LOCK(inp);
1051 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1052 		INP_UNLOCK(inp);
1053 		INP_INFO_WUNLOCK(&udbinfo);
1054 		return (EISCONN);
1055 	}
1056 	sin = (struct sockaddr_in *)nam;
1057 	if (jailed(td->td_ucred))
1058 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1059 	error = in_pcbconnect(inp, nam, td->td_ucred);
1060 	if (error == 0)
1061 		soisconnected(so);
1062 	INP_UNLOCK(inp);
1063 	INP_INFO_WUNLOCK(&udbinfo);
1064 	return (error);
1065 }
1066 
1067 static void
1068 udp_detach(struct socket *so)
1069 {
1070 	struct inpcb *inp;
1071 
1072 	inp = sotoinpcb(so);
1073 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1074 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1075 	    ("udp_detach: not disconnected"));
1076 	INP_INFO_WLOCK(&udbinfo);
1077 	INP_LOCK(inp);
1078 	in_pcbdetach(inp);
1079 	in_pcbfree(inp);
1080 	INP_INFO_WUNLOCK(&udbinfo);
1081 }
1082 
1083 static int
1084 udp_disconnect(struct socket *so)
1085 {
1086 	struct inpcb *inp;
1087 
1088 	inp = sotoinpcb(so);
1089 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1090 	INP_INFO_WLOCK(&udbinfo);
1091 	INP_LOCK(inp);
1092 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1093 		INP_INFO_WUNLOCK(&udbinfo);
1094 		INP_UNLOCK(inp);
1095 		return (ENOTCONN);
1096 	}
1097 
1098 	in_pcbdisconnect(inp);
1099 	inp->inp_laddr.s_addr = INADDR_ANY;
1100 	SOCK_LOCK(so);
1101 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1102 	SOCK_UNLOCK(so);
1103 	INP_UNLOCK(inp);
1104 	INP_INFO_WUNLOCK(&udbinfo);
1105 	return (0);
1106 }
1107 
1108 static int
1109 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1110     struct mbuf *control, struct thread *td)
1111 {
1112 	struct inpcb *inp;
1113 
1114 	inp = sotoinpcb(so);
1115 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1116 	return (udp_output(inp, m, addr, control, td));
1117 }
1118 
1119 int
1120 udp_shutdown(struct socket *so)
1121 {
1122 	struct inpcb *inp;
1123 
1124 	inp = sotoinpcb(so);
1125 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1126 	INP_LOCK(inp);
1127 	socantsendmore(so);
1128 	INP_UNLOCK(inp);
1129 	return (0);
1130 }
1131 
1132 /*
1133  * This is the wrapper function for in_setsockaddr.  We just pass down the
1134  * pcbinfo for in_setsockaddr() to lock.  We don't want to do the locking
1135  * here because in_setsockaddr() will call malloc and might block.
1136  */
1137 static int
1138 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1139 {
1140 
1141 	return (in_setsockaddr(so, nam, &udbinfo));
1142 }
1143 
1144 /*
1145  * This is the wrapper function for in_setpeeraddr().  We just pass down the
1146  * pcbinfo for in_setpeeraddr() to lock.
1147  */
1148 static int
1149 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1150 {
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