xref: /freebsd/sys/netinet/udp_usrreq.c (revision e168b357aa7fe7ae2bb9b56373a3aada3ebf56d7)
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 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
388 	INP_UNLOCK(inp);
389 	INP_INFO_RUNLOCK(&udbinfo);
390 	return;
391 
392 badheadlocked:
393 	if (inp)
394 		INP_UNLOCK(inp);
395 	INP_INFO_RUNLOCK(&udbinfo);
396 badunlocked:
397 	m_freem(m);
398 	if (opts)
399 		m_freem(opts);
400 	return;
401 }
402 
403 /*
404  * Subroutine of udp_input(), which appends the provided mbuf chain to the
405  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
406  * contains the source address.  If the socket ends up being an IPv6 socket,
407  * udp_append() will convert to a sockaddr_in6 before passing the address
408  * into the socket code.
409  */
410 static void
411 udp_append(last, ip, n, off, udp_in)
412 	struct inpcb *last;
413 	struct ip *ip;
414 	struct mbuf *n;
415 	int off;
416 	struct sockaddr_in *udp_in;
417 {
418 	struct sockaddr *append_sa;
419 	struct socket *so;
420 	struct mbuf *opts = 0;
421 #ifdef INET6
422 	struct sockaddr_in6 udp_in6;
423 #endif
424 
425 	INP_LOCK_ASSERT(last);
426 
427 #if defined(IPSEC) || defined(FAST_IPSEC)
428 	/* check AH/ESP integrity. */
429 	if (ipsec4_in_reject(n, last)) {
430 #ifdef IPSEC
431 		ipsecstat.in_polvio++;
432 #endif /*IPSEC*/
433 		m_freem(n);
434 		return;
435 	}
436 #endif /*IPSEC || FAST_IPSEC*/
437 #ifdef MAC
438 	if (mac_check_inpcb_deliver(last, n) != 0) {
439 		m_freem(n);
440 		return;
441 	}
442 #endif
443 	if (last->inp_flags & INP_CONTROLOPTS ||
444 	    last->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
445 #ifdef INET6
446 		if (last->inp_vflag & INP_IPV6) {
447 			int savedflags;
448 
449 			savedflags = last->inp_flags;
450 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
451 			ip6_savecontrol(last, n, &opts);
452 			last->inp_flags = savedflags;
453 		} else
454 #endif
455 		ip_savecontrol(last, &opts, ip, n);
456 	}
457 #ifdef INET6
458 	if (last->inp_vflag & INP_IPV6) {
459 		bzero(&udp_in6, sizeof(udp_in6));
460 		udp_in6.sin6_len = sizeof(udp_in6);
461 		udp_in6.sin6_family = AF_INET6;
462 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
463 		append_sa = (struct sockaddr *)&udp_in6;
464 	} else
465 #endif
466 	append_sa = (struct sockaddr *)udp_in;
467 	m_adj(n, off);
468 
469 	so = last->inp_socket;
470 	SOCKBUF_LOCK(&so->so_rcv);
471 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
472 		m_freem(n);
473 		if (opts)
474 			m_freem(opts);
475 		udpstat.udps_fullsock++;
476 		SOCKBUF_UNLOCK(&so->so_rcv);
477 	} else
478 		sorwakeup_locked(so);
479 }
480 
481 /*
482  * Notify a udp user of an asynchronous error;
483  * just wake up so that he can collect error status.
484  */
485 struct inpcb *
486 udp_notify(inp, errno)
487 	register struct inpcb *inp;
488 	int errno;
489 {
490 	inp->inp_socket->so_error = errno;
491 	sorwakeup(inp->inp_socket);
492 	sowwakeup(inp->inp_socket);
493 	return inp;
494 }
495 
496 void
497 udp_ctlinput(cmd, sa, vip)
498 	int cmd;
499 	struct sockaddr *sa;
500 	void *vip;
501 {
502 	struct ip *ip = vip;
503 	struct udphdr *uh;
504 	struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
505 	struct in_addr faddr;
506 	struct inpcb *inp;
507 
508 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
509 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
510 		return;
511 
512 	/*
513 	 * Redirects don't need to be handled up here.
514 	 */
515 	if (PRC_IS_REDIRECT(cmd))
516 		return;
517 	/*
518 	 * Hostdead is ugly because it goes linearly through all PCBs.
519 	 * XXX: We never get this from ICMP, otherwise it makes an
520 	 * excellent DoS attack on machines with many connections.
521 	 */
522 	if (cmd == PRC_HOSTDEAD)
523 		ip = 0;
524 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
525 		return;
526 	if (ip) {
527 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
528 		INP_INFO_RLOCK(&udbinfo);
529 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
530 		    ip->ip_src, uh->uh_sport, 0, NULL);
531 		if (inp != NULL) {
532 			INP_LOCK(inp);
533 			if (inp->inp_socket != NULL) {
534 				(*notify)(inp, inetctlerrmap[cmd]);
535 			}
536 			INP_UNLOCK(inp);
537 		}
538 		INP_INFO_RUNLOCK(&udbinfo);
539 	} else
540 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify);
541 }
542 
543 static int
544 udp_pcblist(SYSCTL_HANDLER_ARGS)
545 {
546 	int error, i, n;
547 	struct inpcb *inp, **inp_list;
548 	inp_gen_t gencnt;
549 	struct xinpgen xig;
550 
551 	/*
552 	 * The process of preparing the TCB list is too time-consuming and
553 	 * resource-intensive to repeat twice on every request.
554 	 */
555 	if (req->oldptr == 0) {
556 		n = udbinfo.ipi_count;
557 		req->oldidx = 2 * (sizeof xig)
558 			+ (n + n/8) * sizeof(struct xinpcb);
559 		return 0;
560 	}
561 
562 	if (req->newptr != 0)
563 		return EPERM;
564 
565 	/*
566 	 * OK, now we're committed to doing something.
567 	 */
568 	INP_INFO_RLOCK(&udbinfo);
569 	gencnt = udbinfo.ipi_gencnt;
570 	n = udbinfo.ipi_count;
571 	INP_INFO_RUNLOCK(&udbinfo);
572 
573 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
574 		+ n * sizeof(struct xinpcb));
575 	if (error != 0)
576 		return (error);
577 
578 	xig.xig_len = sizeof xig;
579 	xig.xig_count = n;
580 	xig.xig_gen = gencnt;
581 	xig.xig_sogen = so_gencnt;
582 	error = SYSCTL_OUT(req, &xig, sizeof xig);
583 	if (error)
584 		return error;
585 
586 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
587 	if (inp_list == 0)
588 		return ENOMEM;
589 
590 	INP_INFO_RLOCK(&udbinfo);
591 	for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n;
592 	     inp = LIST_NEXT(inp, inp_list)) {
593 		INP_LOCK(inp);
594 		if (inp->inp_gencnt <= gencnt &&
595 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
596 			inp_list[i++] = inp;
597 		INP_UNLOCK(inp);
598 	}
599 	INP_INFO_RUNLOCK(&udbinfo);
600 	n = i;
601 
602 	error = 0;
603 	for (i = 0; i < n; i++) {
604 		inp = inp_list[i];
605 		if (inp->inp_gencnt <= gencnt) {
606 			struct xinpcb xi;
607 			bzero(&xi, sizeof(xi));
608 			xi.xi_len = sizeof xi;
609 			/* XXX should avoid extra copy */
610 			bcopy(inp, &xi.xi_inp, sizeof *inp);
611 			if (inp->inp_socket)
612 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
613 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
614 			error = SYSCTL_OUT(req, &xi, sizeof xi);
615 		}
616 	}
617 	if (!error) {
618 		/*
619 		 * Give the user an updated idea of our state.
620 		 * If the generation differs from what we told
621 		 * her before, she knows that something happened
622 		 * while we were processing this request, and it
623 		 * might be necessary to retry.
624 		 */
625 		INP_INFO_RLOCK(&udbinfo);
626 		xig.xig_gen = udbinfo.ipi_gencnt;
627 		xig.xig_sogen = so_gencnt;
628 		xig.xig_count = udbinfo.ipi_count;
629 		INP_INFO_RUNLOCK(&udbinfo);
630 		error = SYSCTL_OUT(req, &xig, sizeof xig);
631 	}
632 	free(inp_list, M_TEMP);
633 	return error;
634 }
635 
636 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
637 	    udp_pcblist, "S,xinpcb", "List of active UDP sockets");
638 
639 static int
640 udp_getcred(SYSCTL_HANDLER_ARGS)
641 {
642 	struct xucred xuc;
643 	struct sockaddr_in addrs[2];
644 	struct inpcb *inp;
645 	int error;
646 
647 	error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
648 	if (error)
649 		return (error);
650 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
651 	if (error)
652 		return (error);
653 	INP_INFO_RLOCK(&udbinfo);
654 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
655 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
656 	if (inp == NULL || inp->inp_socket == NULL) {
657 		error = ENOENT;
658 		goto out;
659 	}
660 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
661 	if (error)
662 		goto out;
663 	cru2x(inp->inp_socket->so_cred, &xuc);
664 out:
665 	INP_INFO_RUNLOCK(&udbinfo);
666 	if (error == 0)
667 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
668 	return (error);
669 }
670 
671 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
672     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
673     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
674 
675 static int
676 udp_output(inp, m, addr, control, td)
677 	register struct inpcb *inp;
678 	struct mbuf *m;
679 	struct sockaddr *addr;
680 	struct mbuf *control;
681 	struct thread *td;
682 {
683 	register struct udpiphdr *ui;
684 	register int len = m->m_pkthdr.len;
685 	struct in_addr faddr, laddr;
686 	struct cmsghdr *cm;
687 	struct sockaddr_in *sin, src;
688 	int error = 0;
689 	int ipflags;
690 	u_short fport, lport;
691 	int unlock_udbinfo;
692 
693 	/*
694 	 * udp_output() may need to temporarily bind or connect the current
695 	 * inpcb.  As such, we don't know up front what inpcb locks we will
696 	 * need.  Do any work to decide what is needed up front before
697 	 * acquiring locks.
698 	 */
699 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
700 		if (control)
701 			m_freem(control);
702 		m_freem(m);
703 		return EMSGSIZE;
704 	}
705 
706 	src.sin_addr.s_addr = INADDR_ANY;
707 	if (control != NULL) {
708 		/*
709 		 * XXX: Currently, we assume all the optional information
710 		 * is stored in a single mbuf.
711 		 */
712 		if (control->m_next) {
713 			m_freem(control);
714 			m_freem(m);
715 			return EINVAL;
716 		}
717 		for (; control->m_len > 0;
718 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
719 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
720 			cm = mtod(control, struct cmsghdr *);
721 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 ||
722 			    cm->cmsg_len > control->m_len) {
723 				error = EINVAL;
724 				break;
725 			}
726 			if (cm->cmsg_level != IPPROTO_IP)
727 				continue;
728 
729 			switch (cm->cmsg_type) {
730 			case IP_SENDSRCADDR:
731 				if (cm->cmsg_len !=
732 				    CMSG_LEN(sizeof(struct in_addr))) {
733 					error = EINVAL;
734 					break;
735 				}
736 				bzero(&src, sizeof(src));
737 				src.sin_family = AF_INET;
738 				src.sin_len = sizeof(src);
739 				src.sin_port = inp->inp_lport;
740 				src.sin_addr = *(struct in_addr *)CMSG_DATA(cm);
741 				break;
742 			default:
743 				error = ENOPROTOOPT;
744 				break;
745 			}
746 			if (error)
747 				break;
748 		}
749 		m_freem(control);
750 	}
751 	if (error) {
752 		m_freem(m);
753 		return error;
754 	}
755 
756 	if (src.sin_addr.s_addr != INADDR_ANY ||
757 	    addr != NULL) {
758 		INP_INFO_WLOCK(&udbinfo);
759 		unlock_udbinfo = 1;
760 	} else
761 		unlock_udbinfo = 0;
762 	INP_LOCK(inp);
763 
764 #ifdef MAC
765 	mac_create_mbuf_from_inpcb(inp, m);
766 #endif
767 
768 	laddr = inp->inp_laddr;
769 	lport = inp->inp_lport;
770 	if (src.sin_addr.s_addr != INADDR_ANY) {
771 		if (lport == 0) {
772 			error = EINVAL;
773 			goto release;
774 		}
775 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
776 		    &laddr.s_addr, &lport, td->td_ucred);
777 		if (error)
778 			goto release;
779 	}
780 
781 	if (addr) {
782 		sin = (struct sockaddr_in *)addr;
783 		if (jailed(td->td_ucred))
784 			prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
785 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
786 			error = EISCONN;
787 			goto release;
788 		}
789 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
790 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
791 		if (error)
792 			goto release;
793 
794 		/* Commit the local port if newly assigned. */
795 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
796 		    inp->inp_lport == 0) {
797 			/*
798 			 * Remember addr if jailed, to prevent rebinding.
799 			 */
800 			if (jailed(td->td_ucred))
801 				inp->inp_laddr = laddr;
802 			inp->inp_lport = lport;
803 			if (in_pcbinshash(inp) != 0) {
804 				inp->inp_lport = 0;
805 				error = EAGAIN;
806 				goto release;
807 			}
808 			inp->inp_flags |= INP_ANONPORT;
809 		}
810 	} else {
811 		faddr = inp->inp_faddr;
812 		fport = inp->inp_fport;
813 		if (faddr.s_addr == INADDR_ANY) {
814 			error = ENOTCONN;
815 			goto release;
816 		}
817 	}
818 
819 	/*
820 	 * Calculate data length and get a mbuf for UDP, IP, and possible
821 	 * link-layer headers.  Immediate slide the data pointer back forward
822 	 * since we won't use that space at this layer.
823 	 */
824 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
825 	if (m == NULL) {
826 		error = ENOBUFS;
827 		goto release;
828 	}
829 	m->m_data += max_linkhdr;
830 	m->m_len -= max_linkhdr;
831 	m->m_pkthdr.len -= max_linkhdr;
832 
833 	/*
834 	 * Fill in mbuf with extended UDP header
835 	 * and addresses and length put into network format.
836 	 */
837 	ui = mtod(m, struct udpiphdr *);
838 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
839 	ui->ui_pr = IPPROTO_UDP;
840 	ui->ui_src = laddr;
841 	ui->ui_dst = faddr;
842 	ui->ui_sport = lport;
843 	ui->ui_dport = fport;
844 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
845 
846 	ipflags = 0;
847 	if (inp->inp_socket->so_options & SO_DONTROUTE)
848 		ipflags |= IP_ROUTETOIF;
849 	if (inp->inp_socket->so_options & SO_BROADCAST)
850 		ipflags |= IP_ALLOWBROADCAST;
851 	if (inp->inp_flags & INP_ONESBCAST)
852 		ipflags |= IP_SENDONES;
853 
854 	/*
855 	 * Set up checksum and output datagram.
856 	 */
857 	if (udpcksum) {
858 		if (inp->inp_flags & INP_ONESBCAST)
859 			faddr.s_addr = INADDR_BROADCAST;
860 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
861 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
862 		m->m_pkthdr.csum_flags = CSUM_UDP;
863 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
864 	} else {
865 		ui->ui_sum = 0;
866 	}
867 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
868 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
869 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
870 	udpstat.udps_opackets++;
871 
872 	if (unlock_udbinfo)
873 		INP_INFO_WUNLOCK(&udbinfo);
874 	error = ip_output(m, inp->inp_options, NULL, ipflags,
875 	    inp->inp_moptions, inp);
876 	INP_UNLOCK(inp);
877 	return (error);
878 
879 release:
880 	INP_UNLOCK(inp);
881 	if (unlock_udbinfo)
882 		INP_INFO_WUNLOCK(&udbinfo);
883 	m_freem(m);
884 	return (error);
885 }
886 
887 u_long	udp_sendspace = 9216;		/* really max datagram size */
888 					/* 40 1K datagrams */
889 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
890     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
891 
892 u_long	udp_recvspace = 40 * (1024 +
893 #ifdef INET6
894 				      sizeof(struct sockaddr_in6)
895 #else
896 				      sizeof(struct sockaddr_in)
897 #endif
898 				      );
899 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
900     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
901 
902 static int
903 udp_abort(struct socket *so)
904 {
905 	struct inpcb *inp;
906 
907 	INP_INFO_WLOCK(&udbinfo);
908 	inp = sotoinpcb(so);
909 	if (inp == 0) {
910 		INP_INFO_WUNLOCK(&udbinfo);
911 		return EINVAL;	/* ??? possible? panic instead? */
912 	}
913 	INP_LOCK(inp);
914 	soisdisconnected(so);
915 	in_pcbdetach(inp);
916 	INP_INFO_WUNLOCK(&udbinfo);
917 	return 0;
918 }
919 
920 static int
921 udp_attach(struct socket *so, int proto, struct thread *td)
922 {
923 	struct inpcb *inp;
924 	int error;
925 
926 	INP_INFO_WLOCK(&udbinfo);
927 	inp = sotoinpcb(so);
928 	if (inp != 0) {
929 		INP_INFO_WUNLOCK(&udbinfo);
930 		return EINVAL;
931 	}
932 	error = soreserve(so, udp_sendspace, udp_recvspace);
933 	if (error) {
934 		INP_INFO_WUNLOCK(&udbinfo);
935 		return error;
936 	}
937 	error = in_pcballoc(so, &udbinfo, "udpinp");
938 	if (error) {
939 		INP_INFO_WUNLOCK(&udbinfo);
940 		return error;
941 	}
942 
943 	inp = (struct inpcb *)so->so_pcb;
944 	INP_LOCK(inp);
945 	INP_INFO_WUNLOCK(&udbinfo);
946 	inp->inp_vflag |= INP_IPV4;
947 	inp->inp_ip_ttl = ip_defttl;
948 	INP_UNLOCK(inp);
949 	return 0;
950 }
951 
952 static int
953 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
954 {
955 	struct inpcb *inp;
956 	int error;
957 
958 	INP_INFO_WLOCK(&udbinfo);
959 	inp = sotoinpcb(so);
960 	if (inp == 0) {
961 		INP_INFO_WUNLOCK(&udbinfo);
962 		return EINVAL;
963 	}
964 	INP_LOCK(inp);
965 	error = in_pcbbind(inp, nam, td->td_ucred);
966 	INP_UNLOCK(inp);
967 	INP_INFO_WUNLOCK(&udbinfo);
968 	return error;
969 }
970 
971 static int
972 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
973 {
974 	struct inpcb *inp;
975 	int error;
976 	struct sockaddr_in *sin;
977 
978 	INP_INFO_WLOCK(&udbinfo);
979 	inp = sotoinpcb(so);
980 	if (inp == 0) {
981 		INP_INFO_WUNLOCK(&udbinfo);
982 		return EINVAL;
983 	}
984 	INP_LOCK(inp);
985 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
986 		INP_UNLOCK(inp);
987 		INP_INFO_WUNLOCK(&udbinfo);
988 		return EISCONN;
989 	}
990 	sin = (struct sockaddr_in *)nam;
991 	if (jailed(td->td_ucred))
992 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
993 	error = in_pcbconnect(inp, nam, td->td_ucred);
994 	if (error == 0)
995 		soisconnected(so);
996 	INP_UNLOCK(inp);
997 	INP_INFO_WUNLOCK(&udbinfo);
998 	return error;
999 }
1000 
1001 static int
1002 udp_detach(struct socket *so)
1003 {
1004 	struct inpcb *inp;
1005 
1006 	INP_INFO_WLOCK(&udbinfo);
1007 	inp = sotoinpcb(so);
1008 	if (inp == 0) {
1009 		INP_INFO_WUNLOCK(&udbinfo);
1010 		return EINVAL;
1011 	}
1012 	INP_LOCK(inp);
1013 	in_pcbdetach(inp);
1014 	INP_INFO_WUNLOCK(&udbinfo);
1015 	return 0;
1016 }
1017 
1018 static int
1019 udp_disconnect(struct socket *so)
1020 {
1021 	struct inpcb *inp;
1022 
1023 	INP_INFO_WLOCK(&udbinfo);
1024 	inp = sotoinpcb(so);
1025 	if (inp == 0) {
1026 		INP_INFO_WUNLOCK(&udbinfo);
1027 		return EINVAL;
1028 	}
1029 	INP_LOCK(inp);
1030 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1031 		INP_INFO_WUNLOCK(&udbinfo);
1032 		INP_UNLOCK(inp);
1033 		return ENOTCONN;
1034 	}
1035 
1036 	in_pcbdisconnect(inp);
1037 	inp->inp_laddr.s_addr = INADDR_ANY;
1038 	INP_UNLOCK(inp);
1039 	INP_INFO_WUNLOCK(&udbinfo);
1040 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1041 	return 0;
1042 }
1043 
1044 static int
1045 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1046 	    struct mbuf *control, struct thread *td)
1047 {
1048 	struct inpcb *inp;
1049 
1050 	inp = sotoinpcb(so);
1051 	return udp_output(inp, m, addr, control, td);
1052 }
1053 
1054 int
1055 udp_shutdown(struct socket *so)
1056 {
1057 	struct inpcb *inp;
1058 
1059 	INP_INFO_RLOCK(&udbinfo);
1060 	inp = sotoinpcb(so);
1061 	if (inp == 0) {
1062 		INP_INFO_RUNLOCK(&udbinfo);
1063 		return EINVAL;
1064 	}
1065 	INP_LOCK(inp);
1066 	INP_INFO_RUNLOCK(&udbinfo);
1067 	socantsendmore(so);
1068 	INP_UNLOCK(inp);
1069 	return 0;
1070 }
1071 
1072 /*
1073  * This is the wrapper function for in_setsockaddr.  We just pass down
1074  * the pcbinfo for in_setsockaddr to lock.  We don't want to do the locking
1075  * here because in_setsockaddr will call malloc and might block.
1076  */
1077 static int
1078 udp_sockaddr(struct socket *so, struct sockaddr **nam)
1079 {
1080 	return (in_setsockaddr(so, nam, &udbinfo));
1081 }
1082 
1083 /*
1084  * This is the wrapper function for in_setpeeraddr.  We just pass down
1085  * the pcbinfo for in_setpeeraddr to lock.
1086  */
1087 static int
1088 udp_peeraddr(struct socket *so, struct sockaddr **nam)
1089 {
1090 	return (in_setpeeraddr(so, nam, &udbinfo));
1091 }
1092 
1093 struct pr_usrreqs udp_usrreqs = {
1094 	.pru_abort =		udp_abort,
1095 	.pru_attach =		udp_attach,
1096 	.pru_bind =		udp_bind,
1097 	.pru_connect =		udp_connect,
1098 	.pru_control =		in_control,
1099 	.pru_detach =		udp_detach,
1100 	.pru_disconnect =	udp_disconnect,
1101 	.pru_peeraddr =		udp_peeraddr,
1102 	.pru_send =		udp_send,
1103 	.pru_shutdown =		udp_shutdown,
1104 	.pru_sockaddr =		udp_sockaddr,
1105 	.pru_sosetlabel =	in_pcbsosetlabel
1106 };
1107