xref: /freebsd/sys/netinet/udp_usrreq.c (revision 3d11b6c8f01e1fca5936a11d6996448467851a94)
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/jail.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/mac.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h>
48 #include <sys/protosw.h>
49 #include <sys/signalvar.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sx.h>
53 #include <sys/sysctl.h>
54 #include <sys/syslog.h>
55 
56 #include <vm/uma.h>
57 
58 #include <net/if.h>
59 #include <net/route.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69 #include <netinet/ip_icmp.h>
70 #include <netinet/icmp_var.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_options.h>
73 #ifdef INET6
74 #include <netinet6/ip6_var.h>
75 #endif
76 #include <netinet/udp.h>
77 #include <netinet/udp_var.h>
78 
79 #ifdef FAST_IPSEC
80 #include <netipsec/ipsec.h>
81 #endif /*FAST_IPSEC*/
82 
83 #ifdef IPSEC
84 #include <netinet6/ipsec.h>
85 #endif /*IPSEC*/
86 
87 #include <machine/in_cksum.h>
88 
89 /*
90  * UDP protocol implementation.
91  * Per RFC 768, August, 1980.
92  */
93 #ifndef	COMPAT_42
94 static int	udpcksum = 1;
95 #else
96 static int	udpcksum = 0;		/* XXX */
97 #endif
98 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
99 		&udpcksum, 0, "");
100 
101 int	log_in_vain = 0;
102 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
103     &log_in_vain, 0, "Log all incoming UDP packets");
104 
105 static int	blackhole = 0;
106 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
107 	&blackhole, 0, "Do not send port unreachables for refused connects");
108 
109 static int	strict_mcast_mship = 0;
110 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
111 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
112 
113 struct	inpcbhead udb;		/* from udp_var.h */
114 #define	udb6	udb  /* for KAME src sync over BSD*'s */
115 struct	inpcbinfo udbinfo;
116 
117 #ifndef UDBHASHSIZE
118 #define UDBHASHSIZE 16
119 #endif
120 
121 struct	udpstat udpstat;	/* from udp_var.h */
122 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
123     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
124 
125 static void udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n,
126 		int off, struct sockaddr_in *udp_in);
127 
128 static void udp_detach(struct socket *so);
129 static	int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
130 		struct mbuf *, struct thread *);
131 
132 void
133 udp_init()
134 {
135 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
136 	LIST_INIT(&udb);
137 	udbinfo.listhead = &udb;
138 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
139 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
140 					&udbinfo.porthashmask);
141 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
142 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
143 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
144 }
145 
146 void
147 udp_input(m, off)
148 	register struct mbuf *m;
149 	int off;
150 {
151 	int iphlen = off;
152 	register struct ip *ip;
153 	register struct udphdr *uh;
154 	register struct inpcb *inp;
155 	int len;
156 	struct ip save_ip;
157 	struct sockaddr_in udp_in;
158 #ifdef IPFIREWALL_FORWARD
159 	struct m_tag *fwd_tag;
160 #endif
161 
162 	udpstat.udps_ipackets++;
163 
164 	/*
165 	 * Strip IP options, if any; should skip this,
166 	 * make available to user, and use on returned packets,
167 	 * but we don't yet have a way to check the checksum
168 	 * with options still present.
169 	 */
170 	if (iphlen > sizeof (struct ip)) {
171 		ip_stripoptions(m, (struct mbuf *)0);
172 		iphlen = sizeof(struct ip);
173 	}
174 
175 	/*
176 	 * Get IP and UDP header together in first mbuf.
177 	 */
178 	ip = mtod(m, struct ip *);
179 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
180 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
181 			udpstat.udps_hdrops++;
182 			return;
183 		}
184 		ip = mtod(m, struct ip *);
185 	}
186 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
187 
188 	/* destination port of 0 is illegal, based on RFC768. */
189 	if (uh->uh_dport == 0)
190 		goto badunlocked;
191 
192 	/*
193 	 * Construct sockaddr format source address.
194 	 * Stuff source address and datagram in user buffer.
195 	 */
196 	bzero(&udp_in, sizeof(udp_in));
197 	udp_in.sin_len = sizeof(udp_in);
198 	udp_in.sin_family = AF_INET;
199 	udp_in.sin_port = uh->uh_sport;
200 	udp_in.sin_addr = ip->ip_src;
201 
202 	/*
203 	 * Make mbuf data length reflect UDP length.
204 	 * If not enough data to reflect UDP length, drop.
205 	 */
206 	len = ntohs((u_short)uh->uh_ulen);
207 	if (ip->ip_len != len) {
208 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
209 			udpstat.udps_badlen++;
210 			goto badunlocked;
211 		}
212 		m_adj(m, len - ip->ip_len);
213 		/* ip->ip_len = len; */
214 	}
215 	/*
216 	 * Save a copy of the IP header in case we want restore it
217 	 * for sending an ICMP error message in response.
218 	 */
219 	if (!blackhole)
220 		save_ip = *ip;
221 
222 	/*
223 	 * Checksum extended UDP header and data.
224 	 */
225 	if (uh->uh_sum) {
226 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
227 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
228 				uh->uh_sum = m->m_pkthdr.csum_data;
229 			else
230 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
231 				    ip->ip_dst.s_addr, htonl((u_short)len +
232 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
233 			uh->uh_sum ^= 0xffff;
234 		} else {
235 			char b[9];
236 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
237 			bzero(((struct ipovly *)ip)->ih_x1, 9);
238 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
239 			uh->uh_sum = in_cksum(m, len + sizeof (struct ip));
240 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
241 		}
242 		if (uh->uh_sum) {
243 			udpstat.udps_badsum++;
244 			m_freem(m);
245 			return;
246 		}
247 	} else
248 		udpstat.udps_nosum++;
249 
250 #ifdef IPFIREWALL_FORWARD
251 	/* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
252 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
253 
254 	if (fwd_tag != NULL) {
255 		struct sockaddr_in *next_hop;
256 
257 		/* Do the hack. */
258 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
259 		ip->ip_dst = next_hop->sin_addr;
260 		uh->uh_dport = ntohs(next_hop->sin_port);
261 		/* Remove the tag from the packet.  We don't need it anymore. */
262 		m_tag_delete(m, fwd_tag);
263 	}
264 #endif
265 
266 	INP_INFO_RLOCK(&udbinfo);
267 
268 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
269 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
270 		struct inpcb *last;
271 		/*
272 		 * Deliver a multicast or broadcast datagram to *all* sockets
273 		 * for which the local and remote addresses and ports match
274 		 * those of the incoming datagram.  This allows more than
275 		 * one process to receive multi/broadcasts on the same port.
276 		 * (This really ought to be done for unicast datagrams as
277 		 * well, but that would cause problems with existing
278 		 * applications that open both address-specific sockets and
279 		 * a wildcard socket listening to the same port -- they would
280 		 * end up receiving duplicates of every unicast datagram.
281 		 * Those applications open the multiple sockets to overcome an
282 		 * inadequacy of the UDP socket interface, but for backwards
283 		 * compatibility we avoid the problem here rather than
284 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
285 		 */
286 
287 		/*
288 		 * Locate pcb(s) for datagram.
289 		 * (Algorithm copied from raw_intr().)
290 		 */
291 		last = NULL;
292 		LIST_FOREACH(inp, &udb, inp_list) {
293 			if (inp->inp_lport != uh->uh_dport)
294 				continue;
295 #ifdef INET6
296 			if ((inp->inp_vflag & INP_IPV4) == 0)
297 				continue;
298 #endif
299 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
300 				if (inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
301 					continue;
302 			}
303 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
304 				if (inp->inp_faddr.s_addr !=
305 				    ip->ip_src.s_addr ||
306 				    inp->inp_fport != uh->uh_sport)
307 					continue;
308 			}
309 			INP_LOCK(inp);
310 
311 			/*
312 			 * Check multicast packets to make sure they are only
313 			 * sent to sockets with multicast memberships for the
314 			 * packet's destination address and arrival interface
315 			 */
316 #define MSHIP(_inp, n) ((_inp)->inp_moptions->imo_membership[(n)])
317 #define NMSHIPS(_inp) ((_inp)->inp_moptions->imo_num_memberships)
318 			if (strict_mcast_mship && inp->inp_moptions != NULL) {
319 				int mship, foundmship = 0;
320 
321 				for (mship = 0; mship < NMSHIPS(inp); mship++) {
322 					if (MSHIP(inp, mship)->inm_addr.s_addr
323 					    == ip->ip_dst.s_addr &&
324 					    MSHIP(inp, mship)->inm_ifp
325 					    == m->m_pkthdr.rcvif) {
326 						foundmship = 1;
327 						break;
328 					}
329 				}
330 				if (foundmship == 0) {
331 					INP_UNLOCK(inp);
332 					continue;
333 				}
334 			}
335 #undef NMSHIPS
336 #undef MSHIP
337 			if (last != NULL) {
338 				struct mbuf *n;
339 
340 				n = m_copy(m, 0, M_COPYALL);
341 				if (n != NULL)
342 					udp_append(last, ip, n,
343 						   iphlen +
344 						   sizeof(struct udphdr),
345 						   &udp_in);
346 				INP_UNLOCK(last);
347 			}
348 			last = inp;
349 			/*
350 			 * Don't look for additional matches if this one does
351 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
352 			 * socket options set.  This heuristic avoids searching
353 			 * through all pcbs in the common case of a non-shared
354 			 * port.  It * assumes that an application will never
355 			 * clear these options after setting them.
356 			 */
357 			if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0)
358 				break;
359 		}
360 
361 		if (last == NULL) {
362 			/*
363 			 * No matching pcb found; discard datagram.
364 			 * (No need to send an ICMP Port Unreachable
365 			 * for a broadcast or multicast datgram.)
366 			 */
367 			udpstat.udps_noportbcast++;
368 			goto badheadlocked;
369 		}
370 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
371 		    &udp_in);
372 		INP_UNLOCK(last);
373 		INP_INFO_RUNLOCK(&udbinfo);
374 		return;
375 	}
376 	/*
377 	 * Locate pcb for datagram.
378 	 */
379 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
380 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
381 	if (inp == NULL) {
382 		if (log_in_vain) {
383 			char buf[4*sizeof "123"];
384 
385 			strcpy(buf, inet_ntoa(ip->ip_dst));
386 			log(LOG_INFO,
387 			    "Connection attempt to UDP %s:%d from %s:%d\n",
388 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
389 			    ntohs(uh->uh_sport));
390 		}
391 		udpstat.udps_noport++;
392 		if (m->m_flags & (M_BCAST | M_MCAST)) {
393 			udpstat.udps_noportbcast++;
394 			goto badheadlocked;
395 		}
396 		if (blackhole)
397 			goto badheadlocked;
398 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
399 			goto badheadlocked;
400 		*ip = save_ip;
401 		ip->ip_len += iphlen;
402 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
403 		INP_INFO_RUNLOCK(&udbinfo);
404 		return;
405 	}
406 	INP_LOCK(inp);
407 	/* Check the minimum TTL for socket. */
408 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
409 		goto badheadlocked;
410 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
411 	INP_UNLOCK(inp);
412 	INP_INFO_RUNLOCK(&udbinfo);
413 	return;
414 
415 badheadlocked:
416 	if (inp)
417 		INP_UNLOCK(inp);
418 	INP_INFO_RUNLOCK(&udbinfo);
419 badunlocked:
420 	m_freem(m);
421 	return;
422 }
423 
424 /*
425  * Subroutine of udp_input(), which appends the provided mbuf chain to the
426  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
427  * contains the source address.  If the socket ends up being an IPv6 socket,
428  * udp_append() will convert to a sockaddr_in6 before passing the address
429  * into the socket code.
430  */
431 static void
432 udp_append(last, ip, n, off, udp_in)
433 	struct inpcb *last;
434 	struct ip *ip;
435 	struct mbuf *n;
436 	int off;
437 	struct sockaddr_in *udp_in;
438 {
439 	struct sockaddr *append_sa;
440 	struct socket *so;
441 	struct mbuf *opts = 0;
442 #ifdef INET6
443 	struct sockaddr_in6 udp_in6;
444 #endif
445 
446 	INP_LOCK_ASSERT(last);
447 
448 #if defined(IPSEC) || defined(FAST_IPSEC)
449 	/* check AH/ESP integrity. */
450 	if (ipsec4_in_reject(n, last)) {
451 #ifdef IPSEC
452 		ipsecstat.in_polvio++;
453 #endif /*IPSEC*/
454 		m_freem(n);
455 		return;
456 	}
457 #endif /*IPSEC || FAST_IPSEC*/
458 #ifdef MAC
459 	if (mac_check_inpcb_deliver(last, n) != 0) {
460 		m_freem(n);
461 		return;
462 	}
463 #endif
464 	if (last->inp_flags & INP_CONTROLOPTS ||
465 	    last->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
466 #ifdef INET6
467 		if (last->inp_vflag & INP_IPV6) {
468 			int savedflags;
469 
470 			savedflags = last->inp_flags;
471 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
472 			ip6_savecontrol(last, n, &opts);
473 			last->inp_flags = savedflags;
474 		} else
475 #endif
476 		ip_savecontrol(last, &opts, ip, n);
477 	}
478 #ifdef INET6
479 	if (last->inp_vflag & INP_IPV6) {
480 		bzero(&udp_in6, sizeof(udp_in6));
481 		udp_in6.sin6_len = sizeof(udp_in6);
482 		udp_in6.sin6_family = AF_INET6;
483 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
484 		append_sa = (struct sockaddr *)&udp_in6;
485 	} else
486 #endif
487 	append_sa = (struct sockaddr *)udp_in;
488 	m_adj(n, off);
489 
490 	so = last->inp_socket;
491 	SOCKBUF_LOCK(&so->so_rcv);
492 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
493 		m_freem(n);
494 		if (opts)
495 			m_freem(opts);
496 		udpstat.udps_fullsock++;
497 		SOCKBUF_UNLOCK(&so->so_rcv);
498 	} else
499 		sorwakeup_locked(so);
500 }
501 
502 /*
503  * Notify a udp user of an asynchronous error;
504  * just wake up so that he can collect error status.
505  */
506 struct inpcb *
507 udp_notify(inp, errno)
508 	register struct inpcb *inp;
509 	int errno;
510 {
511 	inp->inp_socket->so_error = errno;
512 	sorwakeup(inp->inp_socket);
513 	sowwakeup(inp->inp_socket);
514 	return inp;
515 }
516 
517 void
518 udp_ctlinput(cmd, sa, vip)
519 	int cmd;
520 	struct sockaddr *sa;
521 	void *vip;
522 {
523 	struct ip *ip = vip;
524 	struct udphdr *uh;
525 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
526 	struct in_addr faddr;
527 	struct inpcb *inp;
528 
529 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
530 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
531 		return;
532 
533 	/*
534 	 * Redirects don't need to be handled up here.
535 	 */
536 	if (PRC_IS_REDIRECT(cmd))
537 		return;
538 	/*
539 	 * Hostdead is ugly because it goes linearly through all PCBs.
540 	 * XXX: We never get this from ICMP, otherwise it makes an
541 	 * excellent DoS attack on machines with many connections.
542 	 */
543 	if (cmd == PRC_HOSTDEAD)
544 		ip = 0;
545 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
546 		return;
547 	if (ip) {
548 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
549 		INP_INFO_RLOCK(&udbinfo);
550 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
551 		    ip->ip_src, uh->uh_sport, 0, NULL);
552 		if (inp != NULL) {
553 			INP_LOCK(inp);
554 			if (inp->inp_socket != NULL) {
555 				(*notify)(inp, inetctlerrmap[cmd]);
556 			}
557 			INP_UNLOCK(inp);
558 		}
559 		INP_INFO_RUNLOCK(&udbinfo);
560 	} else
561 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
562 }
563 
564 static int
565 udp_pcblist(SYSCTL_HANDLER_ARGS)
566 {
567 	int error, i, n;
568 	struct inpcb *inp, **inp_list;
569 	inp_gen_t gencnt;
570 	struct xinpgen xig;
571 
572 	/*
573 	 * The process of preparing the TCB list is too time-consuming and
574 	 * resource-intensive to repeat twice on every request.
575 	 */
576 	if (req->oldptr == 0) {
577 		n = udbinfo.ipi_count;
578 		req->oldidx = 2 * (sizeof xig)
579 			+ (n + n/8) * sizeof(struct xinpcb);
580 		return 0;
581 	}
582 
583 	if (req->newptr != 0)
584 		return EPERM;
585 
586 	/*
587 	 * OK, now we're committed to doing something.
588 	 */
589 	INP_INFO_RLOCK(&udbinfo);
590 	gencnt = udbinfo.ipi_gencnt;
591 	n = udbinfo.ipi_count;
592 	INP_INFO_RUNLOCK(&udbinfo);
593 
594 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
595 		+ n * sizeof(struct xinpcb));
596 	if (error != 0)
597 		return (error);
598 
599 	xig.xig_len = sizeof xig;
600 	xig.xig_count = n;
601 	xig.xig_gen = gencnt;
602 	xig.xig_sogen = so_gencnt;
603 	error = SYSCTL_OUT(req, &xig, sizeof xig);
604 	if (error)
605 		return error;
606 
607 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
608 	if (inp_list == 0)
609 		return ENOMEM;
610 
611 	INP_INFO_RLOCK(&udbinfo);
612 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
613 	     inp = LIST_NEXT(inp, inp_list)) {
614 		INP_LOCK(inp);
615 		if (inp->inp_gencnt <= gencnt &&
616 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
617 			inp_list[i++] = inp;
618 		INP_UNLOCK(inp);
619 	}
620 	INP_INFO_RUNLOCK(&udbinfo);
621 	n = i;
622 
623 	error = 0;
624 	for (i = 0; i < n; i++) {
625 		inp = inp_list[i];
626 		if (inp->inp_gencnt <= gencnt) {
627 			struct xinpcb xi;
628 			bzero(&xi, sizeof(xi));
629 			xi.xi_len = sizeof xi;
630 			/* XXX should avoid extra copy */
631 			bcopy(inp, &xi.xi_inp, sizeof *inp);
632 			if (inp->inp_socket)
633 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
634 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
635 			error = SYSCTL_OUT(req, &xi, sizeof xi);
636 		}
637 	}
638 	if (!error) {
639 		/*
640 		 * Give the user an updated idea of our state.
641 		 * If the generation differs from what we told
642 		 * her before, she knows that something happened
643 		 * while we were processing this request, and it
644 		 * might be necessary to retry.
645 		 */
646 		INP_INFO_RLOCK(&udbinfo);
647 		xig.xig_gen = udbinfo.ipi_gencnt;
648 		xig.xig_sogen = so_gencnt;
649 		xig.xig_count = udbinfo.ipi_count;
650 		INP_INFO_RUNLOCK(&udbinfo);
651 		error = SYSCTL_OUT(req, &xig, sizeof xig);
652 	}
653 	free(inp_list, M_TEMP);
654 	return error;
655 }
656 
657 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
658 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
659 
660 static int
661 udp_getcred(SYSCTL_HANDLER_ARGS)
662 {
663 	struct xucred xuc;
664 	struct sockaddr_in addrs[2];
665 	struct inpcb *inp;
666 	int error;
667 
668 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
669 	if (error)
670 		return (error);
671 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
672 	if (error)
673 		return (error);
674 	INP_INFO_RLOCK(&udbinfo);
675 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
676 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
677 	if (inp == NULL || inp->inp_socket == NULL) {
678 		error = ENOENT;
679 		goto out;
680 	}
681 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
682 	if (error)
683 		goto out;
684 	cru2x(inp->inp_socket->so_cred, &xuc);
685 out:
686 	INP_INFO_RUNLOCK(&udbinfo);
687 	if (error == 0)
688 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
689 	return (error);
690 }
691 
692 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
693     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
694     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
695 
696 static int
697 udp_output(inp, m, addr, control, td)
698 	register struct inpcb *inp;
699 	struct mbuf *m;
700 	struct sockaddr *addr;
701 	struct mbuf *control;
702 	struct thread *td;
703 {
704 	register struct udpiphdr *ui;
705 	register int len = m->m_pkthdr.len;
706 	struct in_addr faddr, laddr;
707 	struct cmsghdr *cm;
708 	struct sockaddr_in *sin, src;
709 	int error = 0;
710 	int ipflags;
711 	u_short fport, lport;
712 	int unlock_udbinfo;
713 
714 	/*
715 	 * udp_output() may need to temporarily bind or connect the current
716 	 * inpcb.  As such, we don't know up front what inpcb locks we will
717 	 * need.  Do any work to decide what is needed up front before
718 	 * acquiring locks.
719 	 */
720 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
721 		if (control)
722 			m_freem(control);
723 		m_freem(m);
724 		return EMSGSIZE;
725 	}
726 
727 	src.sin_addr.s_addr = INADDR_ANY;
728 	if (control != NULL) {
729 		/*
730 		 * XXX: Currently, we assume all the optional information
731 		 * is stored in a single mbuf.
732 		 */
733 		if (control->m_next) {
734 			m_freem(control);
735 			m_freem(m);
736 			return EINVAL;
737 		}
738 		for (; control->m_len > 0;
739 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
740 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
741 			cm = mtod(control, struct cmsghdr *);
742 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
743 			    cm->cmsg_len > control->m_len) {
744 				error = EINVAL;
745 				break;
746 			}
747 			if (cm->cmsg_level != IPPROTO_IP)
748 				continue;
749 
750 			switch (cm->cmsg_type) {
751 			case IP_SENDSRCADDR:
752 				if (cm->cmsg_len !=
753 				    CMSG_LEN(sizeof(struct in_addr))) {
754 					error = EINVAL;
755 					break;
756 				}
757 				bzero(&src, sizeof(src));
758 				src.sin_family = AF_INET;
759 				src.sin_len = sizeof(src);
760 				src.sin_port = inp->inp_lport;
761 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
762 				break;
763 			default:
764 				error = ENOPROTOOPT;
765 				break;
766 			}
767 			if (error)
768 				break;
769 		}
770 		m_freem(control);
771 	}
772 	if (error) {
773 		m_freem(m);
774 		return error;
775 	}
776 
777 	if (src.sin_addr.s_addr != INADDR_ANY ||
778 	    addr != NULL) {
779 		INP_INFO_WLOCK(&udbinfo);
780 		unlock_udbinfo = 1;
781 	} else
782 		unlock_udbinfo = 0;
783 	INP_LOCK(inp);
784 
785 #ifdef MAC
786 	mac_create_mbuf_from_inpcb(inp, m);
787 #endif
788 
789 	laddr = inp->inp_laddr;
790 	lport = inp->inp_lport;
791 	if (src.sin_addr.s_addr != INADDR_ANY) {
792 		if (lport == 0) {
793 			error = EINVAL;
794 			goto release;
795 		}
796 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
797 		    &laddr.s_addr, &lport, td->td_ucred);
798 		if (error)
799 			goto release;
800 	}
801 
802 	if (addr) {
803 		sin = (struct sockaddr_in *)addr;
804 		if (jailed(td->td_ucred))
805 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
806 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
807 			error = EISCONN;
808 			goto release;
809 		}
810 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
811 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
812 		if (error)
813 			goto release;
814 
815 		/* Commit the local port if newly assigned. */
816 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
817 		    inp->inp_lport == 0) {
818 			/*
819 			 * Remember addr if jailed, to prevent rebinding.
820 			 */
821 			if (jailed(td->td_ucred))
822 				inp->inp_laddr = laddr;
823 			inp->inp_lport = lport;
824 			if (in_pcbinshash(inp) != 0) {
825 				inp->inp_lport = 0;
826 				error = EAGAIN;
827 				goto release;
828 			}
829 			inp->inp_flags |= INP_ANONPORT;
830 		}
831 	} else {
832 		faddr = inp->inp_faddr;
833 		fport = inp->inp_fport;
834 		if (faddr.s_addr == INADDR_ANY) {
835 			error = ENOTCONN;
836 			goto release;
837 		}
838 	}
839 
840 	/*
841 	 * Calculate data length and get a mbuf for UDP, IP, and possible
842 	 * link-layer headers.  Immediate slide the data pointer back forward
843 	 * since we won't use that space at this layer.
844 	 */
845 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
846 	if (m == NULL) {
847 		error = ENOBUFS;
848 		goto release;
849 	}
850 	m->m_data += max_linkhdr;
851 	m->m_len -= max_linkhdr;
852 	m->m_pkthdr.len -= max_linkhdr;
853 
854 	/*
855 	 * Fill in mbuf with extended UDP header
856 	 * and addresses and length put into network format.
857 	 */
858 	ui = mtod(m, struct udpiphdr *);
859 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
860 	ui->ui_pr = IPPROTO_UDP;
861 	ui->ui_src = laddr;
862 	ui->ui_dst = faddr;
863 	ui->ui_sport = lport;
864 	ui->ui_dport = fport;
865 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
866 
867 	/*
868 	 * Set the Don't Fragment bit in the IP header.
869 	 */
870 	if (inp->inp_flags & INP_DONTFRAG) {
871 		struct ip *ip;
872 		ip = (struct ip *)&ui->ui_i;
873 		ip->ip_off |= IP_DF;
874 	}
875 
876 	ipflags = 0;
877 	if (inp->inp_socket->so_options & SO_DONTROUTE)
878 		ipflags |= IP_ROUTETOIF;
879 	if (inp->inp_socket->so_options & SO_BROADCAST)
880 		ipflags |= IP_ALLOWBROADCAST;
881 	if (inp->inp_vflag & INP_ONESBCAST)
882 		ipflags |= IP_SENDONES;
883 
884 	/*
885 	 * Set up checksum and output datagram.
886 	 */
887 	if (udpcksum) {
888 		if (inp->inp_vflag & INP_ONESBCAST)
889 			faddr.s_addr = INADDR_BROADCAST;
890 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
891 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
892 		m->m_pkthdr.csum_flags = CSUM_UDP;
893 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
894 	} else {
895 		ui->ui_sum = 0;
896 	}
897 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
898 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
899 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
900 	udpstat.udps_opackets++;
901 
902 	if (unlock_udbinfo)
903 		INP_INFO_WUNLOCK(&udbinfo);
904 	error = ip_output(m, inp->inp_options, NULL, ipflags,
905 	    inp->inp_moptions, inp);
906 	INP_UNLOCK(inp);
907 	return (error);
908 
909 release:
910 	INP_UNLOCK(inp);
911 	if (unlock_udbinfo)
912 		INP_INFO_WUNLOCK(&udbinfo);
913 	m_freem(m);
914 	return (error);
915 }
916 
917 u_long	udp_sendspace = 9216;		/* really max datagram size */
918 					/* 40 1K datagrams */
919 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
920     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
921 
922 u_long	udp_recvspace = 40 * (1024 +
923 #ifdef INET6
924 				      sizeof(struct sockaddr_in6)
925 #else
926 				      sizeof(struct sockaddr_in)
927 #endif
928 				      );
929 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
930     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
931 
932 static void
933 udp_abort(struct socket *so)
934 {
935 	struct inpcb *inp;
936 
937 	inp = sotoinpcb(so);
938 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
939 	INP_INFO_WLOCK(&udbinfo);
940 	INP_LOCK(inp);
941 	soisdisconnected(so);
942 	in_pcbdetach(inp);
943 	in_pcbfree(inp);
944 	INP_INFO_WUNLOCK(&udbinfo);
945 }
946 
947 static int
948 udp_attach(struct socket *so, int proto, struct thread *td)
949 {
950 	struct inpcb *inp;
951 	int error;
952 
953 	inp = sotoinpcb(so);
954 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
955 	INP_INFO_WLOCK(&udbinfo);
956 	error = soreserve(so, udp_sendspace, udp_recvspace);
957 	if (error) {
958 		INP_INFO_WUNLOCK(&udbinfo);
959 		return error;
960 	}
961 	error = in_pcballoc(so, &udbinfo, "udpinp");
962 	if (error) {
963 		INP_INFO_WUNLOCK(&udbinfo);
964 		return error;
965 	}
966 
967 	inp = (struct inpcb *)so->so_pcb;
968 	INP_LOCK(inp);
969 	INP_INFO_WUNLOCK(&udbinfo);
970 	inp->inp_vflag |= INP_IPV4;
971 	inp->inp_ip_ttl = ip_defttl;
972 	INP_UNLOCK(inp);
973 	return 0;
974 }
975 
976 static int
977 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
978 {
979 	struct inpcb *inp;
980 	int error;
981 
982 	inp = sotoinpcb(so);
983 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
984 	INP_INFO_WLOCK(&udbinfo);
985 	INP_LOCK(inp);
986 	error = in_pcbbind(inp, nam, td->td_ucred);
987 	INP_UNLOCK(inp);
988 	INP_INFO_WUNLOCK(&udbinfo);
989 	return error;
990 }
991 
992 static int
993 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
994 {
995 	struct inpcb *inp;
996 	int error;
997 	struct sockaddr_in *sin;
998 
999 	inp = sotoinpcb(so);
1000 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1001 	INP_INFO_WLOCK(&udbinfo);
1002 	INP_LOCK(inp);
1003 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1004 		INP_UNLOCK(inp);
1005 		INP_INFO_WUNLOCK(&udbinfo);
1006 		return EISCONN;
1007 	}
1008 	sin = (struct sockaddr_in *)nam;
1009 	if (jailed(td->td_ucred))
1010 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1011 	error = in_pcbconnect(inp, nam, td->td_ucred);
1012 	if (error == 0)
1013 		soisconnected(so);
1014 	INP_UNLOCK(inp);
1015 	INP_INFO_WUNLOCK(&udbinfo);
1016 	return error;
1017 }
1018 
1019 static void
1020 udp_detach(struct socket *so)
1021 {
1022 	struct inpcb *inp;
1023 
1024 	inp = sotoinpcb(so);
1025 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1026 	INP_INFO_WLOCK(&udbinfo);
1027 	INP_LOCK(inp);
1028 	in_pcbdetach(inp);
1029 	in_pcbfree(inp);
1030 	INP_INFO_WUNLOCK(&udbinfo);
1031 }
1032 
1033 static int
1034 udp_disconnect(struct socket *so)
1035 {
1036 	struct inpcb *inp;
1037 
1038 	inp = sotoinpcb(so);
1039 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1040 	INP_INFO_WLOCK(&udbinfo);
1041 	INP_LOCK(inp);
1042 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1043 		INP_INFO_WUNLOCK(&udbinfo);
1044 		INP_UNLOCK(inp);
1045 		return ENOTCONN;
1046 	}
1047 
1048 	in_pcbdisconnect(inp);
1049 	inp->inp_laddr.s_addr = INADDR_ANY;
1050 	INP_UNLOCK(inp);
1051 	INP_INFO_WUNLOCK(&udbinfo);
1052 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1053 	return 0;
1054 }
1055 
1056 static int
1057 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1058 	    struct mbuf *control, struct thread *td)
1059 {
1060 	struct inpcb *inp;
1061 
1062 	inp = sotoinpcb(so);
1063 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1064 	return udp_output(inp, m, addr, control, td);
1065 }
1066 
1067 int
1068 udp_shutdown(struct socket *so)
1069 {
1070 	struct inpcb *inp;
1071 
1072 	inp = sotoinpcb(so);
1073 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1074 	INP_LOCK(inp);
1075 	socantsendmore(so);
1076 	INP_UNLOCK(inp);
1077 	return 0;
1078 }
1079 
1080 /*
1081  * This is the wrapper function for in_setsockaddr.  We just pass down
1082  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1083  * here because in_setsockaddr will call malloc and might block.
1084  */
1085 static int
1086 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1087 {
1088 	return (in_setsockaddr(so, nam, &udbinfo));
1089 }
1090 
1091 /*
1092  * This is the wrapper function for in_setpeeraddr.  We just pass down
1093  * the pcbinfo for in_setpeeraddr to lock.
1094  */
1095 static int
1096 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1097 {
1098 	return (in_setpeeraddr(so, nam, &udbinfo));
1099 }
1100 
1101 struct pr_usrreqs udp_usrreqs = {
1102 	.pru_abort =		udp_abort,
1103 	.pru_attach =		udp_attach,
1104 	.pru_bind =		udp_bind,
1105 	.pru_connect =		udp_connect,
1106 	.pru_control =		in_control,
1107 	.pru_detach =		udp_detach,
1108 	.pru_disconnect =	udp_disconnect,
1109 	.pru_peeraddr =		udp_peeraddr,
1110 	.pru_send =		udp_send,
1111 	.pru_shutdown =		udp_shutdown,
1112 	.pru_sockaddr =		udp_sockaddr,
1113 	.pru_sosetlabel =	in_pcbsosetlabel
1114 };
1115