xref: /freebsd/sys/netinet/udp_usrreq.c (revision 3642298923e528d795e3a30ec165d2b469e28b40)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipsec.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/domain.h>
40 #include <sys/jail.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mac.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h>
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sx.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 
55 #include <vm/uma.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #ifdef INET6
66 #include <netinet/ip6.h>
67 #endif
68 #include <netinet/ip_icmp.h>
69 #include <netinet/icmp_var.h>
70 #include <netinet/ip_var.h>
71 #ifdef INET6
72 #include <netinet6/ip6_var.h>
73 #endif
74 #include <netinet/udp.h>
75 #include <netinet/udp_var.h>
76 
77 #ifdef FAST_IPSEC
78 #include <netipsec/ipsec.h>
79 #endif /*FAST_IPSEC*/
80 
81 #ifdef IPSEC
82 #include <netinet6/ipsec.h>
83 #endif /*IPSEC*/
84 
85 #include <machine/in_cksum.h>
86 
87 /*
88  * UDP protocol implementation.
89  * Per RFC 768, August, 1980.
90  */
91 #ifndef	COMPAT_42
92 static int	udpcksum = 1;
93 #else
94 static int	udpcksum = 0;		/* XXX */
95 #endif
96 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
97 		&udpcksum, 0, "");
98 
99 int	log_in_vain = 0;
100 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
101     &log_in_vain, 0, "Log all incoming UDP packets");
102 
103 static int	blackhole = 0;
104 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
105 	&blackhole, 0, "Do not send port unreachables for refused connects");
106 
107 static int	strict_mcast_mship = 0;
108 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
109 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
110 
111 struct	inpcbhead udb;		/* from udp_var.h */
112 #define	udb6	udb  /* for KAME src sync over BSD*'s */
113 struct	inpcbinfo udbinfo;
114 
115 #ifndef UDBHASHSIZE
116 #define UDBHASHSIZE 16
117 #endif
118 
119 struct	udpstat udpstat;	/* from udp_var.h */
120 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
121     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
122 
123 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
124 		int off, struct sockaddr_in *udp_in);
125 
126 static int udp_detach(struct socket *so);
127 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
128 		struct mbuf *, struct thread *);
129 
130 void
131 udp_init()
132 {
133 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
134 	LIST_INIT(&udb);
135 	udbinfo.listhead = &udb;
136 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
137 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
138 					&udbinfo.porthashmask);
139 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
140 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
141 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
142 }
143 
144 void
145 udp_input(m, off)
146 	register struct mbuf *m;
147 	int off;
148 {
149 	int iphlen = off;
150 	register struct ip *ip;
151 	register struct udphdr *uh;
152 	register struct inpcb *inp;
153 	struct mbuf *opts = 0;
154 	int len;
155 	struct ip save_ip;
156 	struct sockaddr_in udp_in;
157 
158 	udpstat.udps_ipackets++;
159 
160 	/*
161 	 * Strip IP options, if any; should skip this,
162 	 * make available to user, and use on returned packets,
163 	 * but we don't yet have a way to check the checksum
164 	 * with options still present.
165 	 */
166 	if (iphlen > sizeof (struct ip)) {
167 		ip_stripoptions(m, (struct mbuf *)0);
168 		iphlen = sizeof(struct ip);
169 	}
170 
171 	/*
172 	 * Get IP and UDP header together in first mbuf.
173 	 */
174 	ip = mtod(m, struct ip *);
175 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
176 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
177 			udpstat.udps_hdrops++;
178 			return;
179 		}
180 		ip = mtod(m, struct ip *);
181 	}
182 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
183 
184 	/* destination port of 0 is illegal, based on RFC768. */
185 	if (uh->uh_dport == 0)
186 		goto badunlocked;
187 
188 	/*
189 	 * Construct sockaddr format source address.
190 	 * Stuff source address and datagram in user buffer.
191 	 */
192 	bzero(&udp_in, sizeof(udp_in));
193 	udp_in.sin_len = sizeof(udp_in);
194 	udp_in.sin_family = AF_INET;
195 	udp_in.sin_port = uh->uh_sport;
196 	udp_in.sin_addr = ip->ip_src;
197 
198 	/*
199 	 * Make mbuf data length reflect UDP length.
200 	 * If not enough data to reflect UDP length, drop.
201 	 */
202 	len = ntohs((u_short)uh->uh_ulen);
203 	if (ip->ip_len != len) {
204 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
205 			udpstat.udps_badlen++;
206 			goto badunlocked;
207 		}
208 		m_adj(m, len - ip->ip_len);
209 		/* ip->ip_len = len; */
210 	}
211 	/*
212 	 * Save a copy of the IP header in case we want restore it
213 	 * for sending an ICMP error message in response.
214 	 */
215 	if (!blackhole)
216 		save_ip = *ip;
217 
218 	/*
219 	 * Checksum extended UDP header and data.
220 	 */
221 	if (uh->uh_sum) {
222 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
223 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
224 				uh->uh_sum = m->m_pkthdr.csum_data;
225 			else
226 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
227 				    ip->ip_dst.s_addr, htonl((u_short)len +
228 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
229 			uh->uh_sum ^= 0xffff;
230 		} else {
231 			char b[9];
232 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
233 			bzero(((struct ipovly *)ip)->ih_x1, 9);
234 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
235 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
236 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
237 		}
238 		if (uh->uh_sum) {
239 			udpstat.udps_badsum++;
240 			m_freem(m);
241 			return;
242 		}
243 	} else
244 		udpstat.udps_nosum++;
245 
246 	INP_INFO_RLOCK(&udbinfo);
247 
248 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
249 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
250 		struct inpcb *last;
251 		/*
252 		 * Deliver a multicast or broadcast datagram to *all* sockets
253 		 * for which the local and remote addresses and ports match
254 		 * those of the incoming datagram.  This allows more than
255 		 * one process to receive multi/broadcasts on the same port.
256 		 * (This really ought to be done for unicast datagrams as
257 		 * well, but that would cause problems with existing
258 		 * applications that open both address-specific sockets and
259 		 * a wildcard socket listening to the same port -- they would
260 		 * end up receiving duplicates of every unicast datagram.
261 		 * Those applications open the multiple sockets to overcome an
262 		 * inadequacy of the UDP socket interface, but for backwards
263 		 * compatibility we avoid the problem here rather than
264 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
265 		 */
266 
267 		/*
268 		 * Locate pcb(s) for datagram.
269 		 * (Algorithm copied from raw_intr().)
270 		 */
271 		last = NULL;
272 		LIST_FOREACH(inp, &udb, inp_list) {
273 			if (inp->inp_lport != uh->uh_dport)
274 				continue;
275 #ifdef INET6
276 			if ((inp->inp_vflag & INP_IPV4) == 0)
277 				continue;
278 #endif
279 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
280 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
281 					continue;
282 			}
283 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
284 				if (inp->inp_faddr.s_addr !=
285 				    ip->ip_src.s_addr ||
286 				    inp->inp_fport != uh->uh_sport)
287 					continue;
288 			}
289 			INP_LOCK(inp);
290 
291 			/*
292 			 * Check multicast packets to make sure they are only
293 			 * sent to sockets with multicast memberships for the
294 			 * packet's destination address and arrival interface
295 			 */
296 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
297 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
298 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
299 				int mship, foundmship = 0;
300 
301 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
302 					if (MSHIP(inp, mship)->inm_addr.s_addr
303 					    == ip->ip_dst.s_addr &&
304 					    MSHIP(inp, mship)->inm_ifp
305 					    == m->m_pkthdr.rcvif) {
306 						foundmship = 1;
307 						break;
308 					}
309 				}
310 				if (foundmship == 0) {
311 					INP_UNLOCK(inp);
312 					continue;
313 				}
314 			}
315 #undef NMSHIPS
316 #undef MSHIP
317 			if (last != NULL) {
318 				struct mbuf *n;
319 
320 				n = m_copy(m, 0, M_COPYALL);
321 				if (n != NULL)
322 					udp_append(last, ip, n,
323 						   iphlen +
324 						   sizeof(struct udphdr),
325 						   &udp_in);
326 				INP_UNLOCK(last);
327 			}
328 			last = inp;
329 			/*
330 			 * Don't look for additional matches if this one does
331 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
332 			 * socket options set.  This heuristic avoids searching
333 			 * through all pcbs in the common case of a non-shared
334 			 * port.  It * assumes that an application will never
335 			 * clear these options after setting them.
336 			 */
337 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
338 				break;
339 		}
340 
341 		if (last == NULL) {
342 			/*
343 			 * No matching pcb found; discard datagram.
344 			 * (No need to send an ICMP Port Unreachable
345 			 * for a broadcast or multicast datgram.)
346 			 */
347 			udpstat.udps_noportbcast++;
348 			goto badheadlocked;
349 		}
350 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
351 		    &udp_in);
352 		INP_UNLOCK(last);
353 		INP_INFO_RUNLOCK(&udbinfo);
354 		return;
355 	}
356 	/*
357 	 * Locate pcb for datagram.
358 	 */
359 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
360 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
361 	if (inp == NULL) {
362 		if (log_in_vain) {
363 			char buf[4*sizeof "123"];
364 
365 			strcpy(buf, inet_ntoa(ip->ip_dst));
366 			log(LOG_INFO,
367 			    "Connection attempt to UDP %s:%d from %s:%d\n",
368 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
369 			    ntohs(uh->uh_sport));
370 		}
371 		udpstat.udps_noport++;
372 		if (m->m_flags & (M_BCAST | M_MCAST)) {
373 			udpstat.udps_noportbcast++;
374 			goto badheadlocked;
375 		}
376 		if (blackhole)
377 			goto badheadlocked;
378 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
379 			goto badheadlocked;
380 		*ip = save_ip;
381 		ip->ip_len += iphlen;
382 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
383 		INP_INFO_RUNLOCK(&udbinfo);
384 		return;
385 	}
386 	INP_LOCK(inp);
387 	/* Check the minimum TTL for socket. */
388 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
389 		goto badheadlocked;
390 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
391 	INP_UNLOCK(inp);
392 	INP_INFO_RUNLOCK(&udbinfo);
393 	return;
394 
395 badheadlocked:
396 	if (inp)
397 		INP_UNLOCK(inp);
398 	INP_INFO_RUNLOCK(&udbinfo);
399 badunlocked:
400 	m_freem(m);
401 	if (opts)
402 		m_freem(opts);
403 	return;
404 }
405 
406 /*
407  * Subroutine of udp_input(), which appends the provided mbuf chain to the
408  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
409  * contains the source address.  If the socket ends up being an IPv6 socket,
410  * udp_append() will convert to a sockaddr_in6 before passing the address
411  * into the socket code.
412  */
413 static void
414 udp_append(last, ip, n, off, udp_in)
415 	struct inpcb *last;
416 	struct ip *ip;
417 	struct mbuf *n;
418 	int off;
419 	struct sockaddr_in *udp_in;
420 {
421 	struct sockaddr *append_sa;
422 	struct socket *so;
423 	struct mbuf *opts = 0;
424 #ifdef INET6
425 	struct sockaddr_in6 udp_in6;
426 #endif
427 
428 	INP_LOCK_ASSERT(last);
429 
430 #if defined(IPSEC) || defined(FAST_IPSEC)
431 	/* check AH/ESP integrity. */
432 	if (ipsec4_in_reject(n, last)) {
433 #ifdef IPSEC
434 		ipsecstat.in_polvio++;
435 #endif /*IPSEC*/
436 		m_freem(n);
437 		return;
438 	}
439 #endif /*IPSEC || FAST_IPSEC*/
440 #ifdef MAC
441 	if (mac_check_inpcb_deliver(last, n) != 0) {
442 		m_freem(n);
443 		return;
444 	}
445 #endif
446 	if (last->inp_flags & INP_CONTROLOPTS ||
447 	    last->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
448 #ifdef INET6
449 		if (last->inp_vflag & INP_IPV6) {
450 			int savedflags;
451 
452 			savedflags = last->inp_flags;
453 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
454 			ip6_savecontrol(last, n, &opts);
455 			last->inp_flags = savedflags;
456 		} else
457 #endif
458 		ip_savecontrol(last, &opts, ip, n);
459 	}
460 #ifdef INET6
461 	if (last->inp_vflag & INP_IPV6) {
462 		bzero(&udp_in6, sizeof(udp_in6));
463 		udp_in6.sin6_len = sizeof(udp_in6);
464 		udp_in6.sin6_family = AF_INET6;
465 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
466 		append_sa = (struct sockaddr *)&udp_in6;
467 	} else
468 #endif
469 	append_sa = (struct sockaddr *)udp_in;
470 	m_adj(n, off);
471 
472 	so = last->inp_socket;
473 	SOCKBUF_LOCK(&so->so_rcv);
474 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
475 		m_freem(n);
476 		if (opts)
477 			m_freem(opts);
478 		udpstat.udps_fullsock++;
479 		SOCKBUF_UNLOCK(&so->so_rcv);
480 	} else
481 		sorwakeup_locked(so);
482 }
483 
484 /*
485  * Notify a udp user of an asynchronous error;
486  * just wake up so that he can collect error status.
487  */
488 struct inpcb *
489 udp_notify(inp, errno)
490 	register struct inpcb *inp;
491 	int errno;
492 {
493 	inp->inp_socket->so_error = errno;
494 	sorwakeup(inp->inp_socket);
495 	sowwakeup(inp->inp_socket);
496 	return inp;
497 }
498 
499 void
500 udp_ctlinput(cmd, sa, vip)
501 	int cmd;
502 	struct sockaddr *sa;
503 	void *vip;
504 {
505 	struct ip *ip = vip;
506 	struct udphdr *uh;
507 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
508 	struct in_addr faddr;
509 	struct inpcb *inp;
510 
511 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
512 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
513 		return;
514 
515 	/*
516 	 * Redirects don't need to be handled up here.
517 	 */
518 	if (PRC_IS_REDIRECT(cmd))
519 		return;
520 	/*
521 	 * Hostdead is ugly because it goes linearly through all PCBs.
522 	 * XXX: We never get this from ICMP, otherwise it makes an
523 	 * excellent DoS attack on machines with many connections.
524 	 */
525 	if (cmd == PRC_HOSTDEAD)
526 		ip = 0;
527 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
528 		return;
529 	if (ip) {
530 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
531 		INP_INFO_RLOCK(&udbinfo);
532 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
533 		    ip->ip_src, uh->uh_sport, 0, NULL);
534 		if (inp != NULL) {
535 			INP_LOCK(inp);
536 			if (inp->inp_socket != NULL) {
537 				(*notify)(inp, inetctlerrmap[cmd]);
538 			}
539 			INP_UNLOCK(inp);
540 		}
541 		INP_INFO_RUNLOCK(&udbinfo);
542 	} else
543 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
544 }
545 
546 static int
547 udp_pcblist(SYSCTL_HANDLER_ARGS)
548 {
549 	int error, i, n;
550 	struct inpcb *inp, **inp_list;
551 	inp_gen_t gencnt;
552 	struct xinpgen xig;
553 
554 	/*
555 	 * The process of preparing the TCB list is too time-consuming and
556 	 * resource-intensive to repeat twice on every request.
557 	 */
558 	if (req->oldptr == 0) {
559 		n = udbinfo.ipi_count;
560 		req->oldidx = 2 * (sizeof xig)
561 			+ (n + n/8) * sizeof(struct xinpcb);
562 		return 0;
563 	}
564 
565 	if (req->newptr != 0)
566 		return EPERM;
567 
568 	/*
569 	 * OK, now we're committed to doing something.
570 	 */
571 	INP_INFO_RLOCK(&udbinfo);
572 	gencnt = udbinfo.ipi_gencnt;
573 	n = udbinfo.ipi_count;
574 	INP_INFO_RUNLOCK(&udbinfo);
575 
576 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
577 		+ n * sizeof(struct xinpcb));
578 	if (error != 0)
579 		return (error);
580 
581 	xig.xig_len = sizeof xig;
582 	xig.xig_count = n;
583 	xig.xig_gen = gencnt;
584 	xig.xig_sogen = so_gencnt;
585 	error = SYSCTL_OUT(req, &xig, sizeof xig);
586 	if (error)
587 		return error;
588 
589 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
590 	if (inp_list == 0)
591 		return ENOMEM;
592 
593 	INP_INFO_RLOCK(&udbinfo);
594 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
595 	     inp = LIST_NEXT(inp, inp_list)) {
596 		INP_LOCK(inp);
597 		if (inp->inp_gencnt <= gencnt &&
598 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
599 			inp_list[i++] = inp;
600 		INP_UNLOCK(inp);
601 	}
602 	INP_INFO_RUNLOCK(&udbinfo);
603 	n = i;
604 
605 	error = 0;
606 	for (i = 0; i < n; i++) {
607 		inp = inp_list[i];
608 		if (inp->inp_gencnt <= gencnt) {
609 			struct xinpcb xi;
610 			bzero(&xi, sizeof(xi));
611 			xi.xi_len = sizeof xi;
612 			/* XXX should avoid extra copy */
613 			bcopy(inp, &xi.xi_inp, sizeof *inp);
614 			if (inp->inp_socket)
615 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
616 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
617 			error = SYSCTL_OUT(req, &xi, sizeof xi);
618 		}
619 	}
620 	if (!error) {
621 		/*
622 		 * Give the user an updated idea of our state.
623 		 * If the generation differs from what we told
624 		 * her before, she knows that something happened
625 		 * while we were processing this request, and it
626 		 * might be necessary to retry.
627 		 */
628 		INP_INFO_RLOCK(&udbinfo);
629 		xig.xig_gen = udbinfo.ipi_gencnt;
630 		xig.xig_sogen = so_gencnt;
631 		xig.xig_count = udbinfo.ipi_count;
632 		INP_INFO_RUNLOCK(&udbinfo);
633 		error = SYSCTL_OUT(req, &xig, sizeof xig);
634 	}
635 	free(inp_list, M_TEMP);
636 	return error;
637 }
638 
639 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
640 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
641 
642 static int
643 udp_getcred(SYSCTL_HANDLER_ARGS)
644 {
645 	struct xucred xuc;
646 	struct sockaddr_in addrs[2];
647 	struct inpcb *inp;
648 	int error;
649 
650 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
651 	if (error)
652 		return (error);
653 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
654 	if (error)
655 		return (error);
656 	INP_INFO_RLOCK(&udbinfo);
657 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
658 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
659 	if (inp == NULL || inp->inp_socket == NULL) {
660 		error = ENOENT;
661 		goto out;
662 	}
663 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
664 	if (error)
665 		goto out;
666 	cru2x(inp->inp_socket->so_cred, &xuc);
667 out:
668 	INP_INFO_RUNLOCK(&udbinfo);
669 	if (error == 0)
670 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
671 	return (error);
672 }
673 
674 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
675     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
676     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
677 
678 static int
679 udp_output(inp, m, addr, control, td)
680 	register struct inpcb *inp;
681 	struct mbuf *m;
682 	struct sockaddr *addr;
683 	struct mbuf *control;
684 	struct thread *td;
685 {
686 	register struct udpiphdr *ui;
687 	register int len = m->m_pkthdr.len;
688 	struct in_addr faddr, laddr;
689 	struct cmsghdr *cm;
690 	struct sockaddr_in *sin, src;
691 	int error = 0;
692 	int ipflags;
693 	u_short fport, lport;
694 	int unlock_udbinfo;
695 
696 	/*
697 	 * udp_output() may need to temporarily bind or connect the current
698 	 * inpcb.  As such, we don't know up front what inpcb locks we will
699 	 * need.  Do any work to decide what is needed up front before
700 	 * acquiring locks.
701 	 */
702 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
703 		if (control)
704 			m_freem(control);
705 		m_freem(m);
706 		return EMSGSIZE;
707 	}
708 
709 	src.sin_addr.s_addr = INADDR_ANY;
710 	if (control != NULL) {
711 		/*
712 		 * XXX: Currently, we assume all the optional information
713 		 * is stored in a single mbuf.
714 		 */
715 		if (control->m_next) {
716 			m_freem(control);
717 			m_freem(m);
718 			return EINVAL;
719 		}
720 		for (; control->m_len > 0;
721 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
722 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
723 			cm = mtod(control, struct cmsghdr *);
724 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
725 			    cm->cmsg_len > control->m_len) {
726 				error = EINVAL;
727 				break;
728 			}
729 			if (cm->cmsg_level != IPPROTO_IP)
730 				continue;
731 
732 			switch (cm->cmsg_type) {
733 			case IP_SENDSRCADDR:
734 				if (cm->cmsg_len !=
735 				    CMSG_LEN(sizeof(struct in_addr))) {
736 					error = EINVAL;
737 					break;
738 				}
739 				bzero(&src, sizeof(src));
740 				src.sin_family = AF_INET;
741 				src.sin_len = sizeof(src);
742 				src.sin_port = inp->inp_lport;
743 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
744 				break;
745 			default:
746 				error = ENOPROTOOPT;
747 				break;
748 			}
749 			if (error)
750 				break;
751 		}
752 		m_freem(control);
753 	}
754 	if (error) {
755 		m_freem(m);
756 		return error;
757 	}
758 
759 	if (src.sin_addr.s_addr != INADDR_ANY ||
760 	    addr != NULL) {
761 		INP_INFO_WLOCK(&udbinfo);
762 		unlock_udbinfo = 1;
763 	} else
764 		unlock_udbinfo = 0;
765 	INP_LOCK(inp);
766 
767 #ifdef MAC
768 	mac_create_mbuf_from_inpcb(inp, m);
769 #endif
770 
771 	laddr = inp->inp_laddr;
772 	lport = inp->inp_lport;
773 	if (src.sin_addr.s_addr != INADDR_ANY) {
774 		if (lport == 0) {
775 			error = EINVAL;
776 			goto release;
777 		}
778 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
779 		    &laddr.s_addr, &lport, td->td_ucred);
780 		if (error)
781 			goto release;
782 	}
783 
784 	if (addr) {
785 		sin = (struct sockaddr_in *)addr;
786 		if (jailed(td->td_ucred))
787 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
788 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
789 			error = EISCONN;
790 			goto release;
791 		}
792 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
793 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
794 		if (error)
795 			goto release;
796 
797 		/* Commit the local port if newly assigned. */
798 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
799 		    inp->inp_lport == 0) {
800 			/*
801 			 * Remember addr if jailed, to prevent rebinding.
802 			 */
803 			if (jailed(td->td_ucred))
804 				inp->inp_laddr = laddr;
805 			inp->inp_lport = lport;
806 			if (in_pcbinshash(inp) != 0) {
807 				inp->inp_lport = 0;
808 				error = EAGAIN;
809 				goto release;
810 			}
811 			inp->inp_flags |= INP_ANONPORT;
812 		}
813 	} else {
814 		faddr = inp->inp_faddr;
815 		fport = inp->inp_fport;
816 		if (faddr.s_addr == INADDR_ANY) {
817 			error = ENOTCONN;
818 			goto release;
819 		}
820 	}
821 
822 	/*
823 	 * Calculate data length and get a mbuf for UDP, IP, and possible
824 	 * link-layer headers.  Immediate slide the data pointer back forward
825 	 * since we won't use that space at this layer.
826 	 */
827 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
828 	if (m == NULL) {
829 		error = ENOBUFS;
830 		goto release;
831 	}
832 	m->m_data += max_linkhdr;
833 	m->m_len -= max_linkhdr;
834 	m->m_pkthdr.len -= max_linkhdr;
835 
836 	/*
837 	 * Fill in mbuf with extended UDP header
838 	 * and addresses and length put into network format.
839 	 */
840 	ui = mtod(m, struct udpiphdr *);
841 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
842 	ui->ui_pr = IPPROTO_UDP;
843 	ui->ui_src = laddr;
844 	ui->ui_dst = faddr;
845 	ui->ui_sport = lport;
846 	ui->ui_dport = fport;
847 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
848 
849 	ipflags = 0;
850 	if (inp->inp_socket->so_options & SO_DONTROUTE)
851 		ipflags |= IP_ROUTETOIF;
852 	if (inp->inp_socket->so_options & SO_BROADCAST)
853 		ipflags |= IP_ALLOWBROADCAST;
854 	if (inp->inp_flags & INP_ONESBCAST)
855 		ipflags |= IP_SENDONES;
856 
857 	/*
858 	 * Set up checksum and output datagram.
859 	 */
860 	if (udpcksum) {
861 		if (inp->inp_flags & INP_ONESBCAST)
862 			faddr.s_addr = INADDR_BROADCAST;
863 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
864 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
865 		m->m_pkthdr.csum_flags = CSUM_UDP;
866 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
867 	} else {
868 		ui->ui_sum = 0;
869 	}
870 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
871 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
872 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
873 	udpstat.udps_opackets++;
874 
875 	if (unlock_udbinfo)
876 		INP_INFO_WUNLOCK(&udbinfo);
877 	error = ip_output(m, inp->inp_options, NULL, ipflags,
878 	    inp->inp_moptions, inp);
879 	INP_UNLOCK(inp);
880 	return (error);
881 
882 release:
883 	INP_UNLOCK(inp);
884 	if (unlock_udbinfo)
885 		INP_INFO_WUNLOCK(&udbinfo);
886 	m_freem(m);
887 	return (error);
888 }
889 
890 u_long	udp_sendspace = 9216;		/* really max datagram size */
891 					/* 40 1K datagrams */
892 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
893     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
894 
895 u_long	udp_recvspace = 40 * (1024 +
896 #ifdef INET6
897 				      sizeof(struct sockaddr_in6)
898 #else
899 				      sizeof(struct sockaddr_in)
900 #endif
901 				      );
902 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
903     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
904 
905 static int
906 udp_abort(struct socket *so)
907 {
908 	struct inpcb *inp;
909 
910 	INP_INFO_WLOCK(&udbinfo);
911 	inp = sotoinpcb(so);
912 	if (inp == 0) {
913 		INP_INFO_WUNLOCK(&udbinfo);
914 		return EINVAL;	/* ??? possible? panic instead? */
915 	}
916 	INP_LOCK(inp);
917 	soisdisconnected(so);
918 	in_pcbdetach(inp);
919 	INP_INFO_WUNLOCK(&udbinfo);
920 	return 0;
921 }
922 
923 static int
924 udp_attach(struct socket *so, int proto, struct thread *td)
925 {
926 	struct inpcb *inp;
927 	int error;
928 
929 	INP_INFO_WLOCK(&udbinfo);
930 	inp = sotoinpcb(so);
931 	if (inp != 0) {
932 		INP_INFO_WUNLOCK(&udbinfo);
933 		return EINVAL;
934 	}
935 	error = soreserve(so, udp_sendspace, udp_recvspace);
936 	if (error) {
937 		INP_INFO_WUNLOCK(&udbinfo);
938 		return error;
939 	}
940 	error = in_pcballoc(so, &udbinfo, "udpinp");
941 	if (error) {
942 		INP_INFO_WUNLOCK(&udbinfo);
943 		return error;
944 	}
945 
946 	inp = (struct inpcb *)so->so_pcb;
947 	INP_LOCK(inp);
948 	INP_INFO_WUNLOCK(&udbinfo);
949 	inp->inp_vflag |= INP_IPV4;
950 	inp->inp_ip_ttl = ip_defttl;
951 	INP_UNLOCK(inp);
952 	return 0;
953 }
954 
955 static int
956 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
957 {
958 	struct inpcb *inp;
959 	int error;
960 
961 	INP_INFO_WLOCK(&udbinfo);
962 	inp = sotoinpcb(so);
963 	if (inp == 0) {
964 		INP_INFO_WUNLOCK(&udbinfo);
965 		return EINVAL;
966 	}
967 	INP_LOCK(inp);
968 	error = in_pcbbind(inp, nam, td->td_ucred);
969 	INP_UNLOCK(inp);
970 	INP_INFO_WUNLOCK(&udbinfo);
971 	return error;
972 }
973 
974 static int
975 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
976 {
977 	struct inpcb *inp;
978 	int error;
979 	struct sockaddr_in *sin;
980 
981 	INP_INFO_WLOCK(&udbinfo);
982 	inp = sotoinpcb(so);
983 	if (inp == 0) {
984 		INP_INFO_WUNLOCK(&udbinfo);
985 		return EINVAL;
986 	}
987 	INP_LOCK(inp);
988 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
989 		INP_UNLOCK(inp);
990 		INP_INFO_WUNLOCK(&udbinfo);
991 		return EISCONN;
992 	}
993 	sin = (struct sockaddr_in *)nam;
994 	if (jailed(td->td_ucred))
995 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
996 	error = in_pcbconnect(inp, nam, td->td_ucred);
997 	if (error == 0)
998 		soisconnected(so);
999 	INP_UNLOCK(inp);
1000 	INP_INFO_WUNLOCK(&udbinfo);
1001 	return error;
1002 }
1003 
1004 static int
1005 udp_detach(struct socket *so)
1006 {
1007 	struct inpcb *inp;
1008 
1009 	INP_INFO_WLOCK(&udbinfo);
1010 	inp = sotoinpcb(so);
1011 	if (inp == 0) {
1012 		INP_INFO_WUNLOCK(&udbinfo);
1013 		return EINVAL;
1014 	}
1015 	INP_LOCK(inp);
1016 	in_pcbdetach(inp);
1017 	INP_INFO_WUNLOCK(&udbinfo);
1018 	return 0;
1019 }
1020 
1021 static int
1022 udp_disconnect(struct socket *so)
1023 {
1024 	struct inpcb *inp;
1025 
1026 	INP_INFO_WLOCK(&udbinfo);
1027 	inp = sotoinpcb(so);
1028 	if (inp == 0) {
1029 		INP_INFO_WUNLOCK(&udbinfo);
1030 		return EINVAL;
1031 	}
1032 	INP_LOCK(inp);
1033 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1034 		INP_INFO_WUNLOCK(&udbinfo);
1035 		INP_UNLOCK(inp);
1036 		return ENOTCONN;
1037 	}
1038 
1039 	in_pcbdisconnect(inp);
1040 	inp->inp_laddr.s_addr = INADDR_ANY;
1041 	INP_UNLOCK(inp);
1042 	INP_INFO_WUNLOCK(&udbinfo);
1043 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1044 	return 0;
1045 }
1046 
1047 static int
1048 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1049 	    struct mbuf *control, struct thread *td)
1050 {
1051 	struct inpcb *inp;
1052 
1053 	inp = sotoinpcb(so);
1054 	return udp_output(inp, m, addr, control, td);
1055 }
1056 
1057 int
1058 udp_shutdown(struct socket *so)
1059 {
1060 	struct inpcb *inp;
1061 
1062 	INP_INFO_RLOCK(&udbinfo);
1063 	inp = sotoinpcb(so);
1064 	if (inp == 0) {
1065 		INP_INFO_RUNLOCK(&udbinfo);
1066 		return EINVAL;
1067 	}
1068 	INP_LOCK(inp);
1069 	INP_INFO_RUNLOCK(&udbinfo);
1070 	socantsendmore(so);
1071 	INP_UNLOCK(inp);
1072 	return 0;
1073 }
1074 
1075 /*
1076  * This is the wrapper function for in_setsockaddr.  We just pass down
1077  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1078  * here because in_setsockaddr will call malloc and might block.
1079  */
1080 static int
1081 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1082 {
1083 	return (in_setsockaddr(so, nam, &udbinfo));
1084 }
1085 
1086 /*
1087  * This is the wrapper function for in_setpeeraddr.  We just pass down
1088  * the pcbinfo for in_setpeeraddr to lock.
1089  */
1090 static int
1091 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1092 {
1093 	return (in_setpeeraddr(so, nam, &udbinfo));
1094 }
1095 
1096 struct pr_usrreqs udp_usrreqs = {
1097 	.pru_abort =		udp_abort,
1098 	.pru_attach =		udp_attach,
1099 	.pru_bind =		udp_bind,
1100 	.pru_connect =		udp_connect,
1101 	.pru_control =		in_control,
1102 	.pru_detach =		udp_detach,
1103 	.pru_disconnect =	udp_disconnect,
1104 	.pru_peeraddr =		udp_peeraddr,
1105 	.pru_send =		udp_send,
1106 	.pru_shutdown =		udp_shutdown,
1107 	.pru_sockaddr =		udp_sockaddr,
1108 	.pru_sosetlabel =	in_pcbsosetlabel
1109 };
1110