xref: /freebsd/sys/netinet/udp_usrreq.c (revision db612abe8df3355d1eb23bb3b50fdd97bc21e979)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 4. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_ipfw.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 
41 #include <sys/param.h>
42 #include <sys/domain.h>
43 #include <sys/eventhandler.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/protosw.h>
52 #include <sys/signalvar.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/systm.h>
59 
60 #include <vm/uma.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #endif
73 #include <netinet/ip_icmp.h>
74 #include <netinet/icmp_var.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/ip_options.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #endif
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
82 
83 #ifdef IPSEC
84 #include <netipsec/ipsec.h>
85 #endif
86 
87 #include <machine/in_cksum.h>
88 
89 #include <security/mac/mac_framework.h>
90 
91 /*
92  * UDP protocol implementation.
93  * Per RFC 768, August, 1980.
94  */
95 
96 /*
97  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
98  * removes the only data integrity mechanism for packets and malformed
99  * packets that would otherwise be discarded due to bad checksums, and may
100  * cause problems (especially for NFS data blocks).
101  */
102 static int	udp_cksum = 1;
103 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
104     0, "");
105 
106 int	udp_log_in_vain = 0;
107 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
108     &udp_log_in_vain, 0, "Log all incoming UDP packets");
109 
110 int	udp_blackhole = 0;
111 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, &udp_blackhole, 0,
112     "Do not send port unreachables for refused connects");
113 
114 u_long	udp_sendspace = 9216;		/* really max datagram size */
115 					/* 40 1K datagrams */
116 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
117     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
118 
119 u_long	udp_recvspace = 40 * (1024 +
120 #ifdef INET6
121 				      sizeof(struct sockaddr_in6)
122 #else
123 				      sizeof(struct sockaddr_in)
124 #endif
125 				      );
126 
127 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
128     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
129 
130 struct inpcbhead	udb;		/* from udp_var.h */
131 struct inpcbinfo	udbinfo;
132 
133 #ifndef UDBHASHSIZE
134 #define	UDBHASHSIZE	16
135 #endif
136 
137 struct udpstat	udpstat;	/* from udp_var.h */
138 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW, &udpstat,
139     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
140 
141 static void	udp_detach(struct socket *so);
142 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
143 		    struct mbuf *, struct thread *);
144 
145 static void
146 udp_zone_change(void *tag)
147 {
148 
149 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
150 }
151 
152 static int
153 udp_inpcb_init(void *mem, int size, int flags)
154 {
155 	struct inpcb *inp;
156 
157 	inp = mem;
158 	INP_LOCK_INIT(inp, "inp", "udpinp");
159 	return (0);
160 }
161 
162 void
163 udp_init(void)
164 {
165 
166 	INP_INFO_LOCK_INIT(&udbinfo, "udp");
167 	LIST_INIT(&udb);
168 	udbinfo.ipi_listhead = &udb;
169 	udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
170 	    &udbinfo.ipi_hashmask);
171 	udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
172 	    &udbinfo.ipi_porthashmask);
173 	udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
174 	    NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
175 	uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
176 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
177 	    EVENTHANDLER_PRI_ANY);
178 }
179 
180 /*
181  * Subroutine of udp_input(), which appends the provided mbuf chain to the
182  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
183  * contains the source address.  If the socket ends up being an IPv6 socket,
184  * udp_append() will convert to a sockaddr_in6 before passing the address
185  * into the socket code.
186  */
187 static void
188 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
189     struct sockaddr_in *udp_in)
190 {
191 	struct sockaddr *append_sa;
192 	struct socket *so;
193 	struct mbuf *opts = 0;
194 #ifdef INET6
195 	struct sockaddr_in6 udp_in6;
196 #endif
197 
198 	INP_WLOCK_ASSERT(inp);
199 
200 #ifdef IPSEC
201 	/* Check AH/ESP integrity. */
202 	if (ipsec4_in_reject(n, inp)) {
203 		m_freem(n);
204 		ipsec4stat.in_polvio++;
205 		return;
206 	}
207 #endif /* IPSEC */
208 #ifdef MAC
209 	if (mac_inpcb_check_deliver(inp, n) != 0) {
210 		m_freem(n);
211 		return;
212 	}
213 #endif
214 	if (inp->inp_flags & INP_CONTROLOPTS ||
215 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
216 #ifdef INET6
217 		if (inp->inp_vflag & INP_IPV6) {
218 			int savedflags;
219 
220 			savedflags = inp->inp_flags;
221 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
222 			ip6_savecontrol(inp, n, &opts);
223 			inp->inp_flags = savedflags;
224 		} else
225 #endif
226 			ip_savecontrol(inp, &opts, ip, n);
227 	}
228 #ifdef INET6
229 	if (inp->inp_vflag & INP_IPV6) {
230 		bzero(&udp_in6, sizeof(udp_in6));
231 		udp_in6.sin6_len = sizeof(udp_in6);
232 		udp_in6.sin6_family = AF_INET6;
233 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
234 		append_sa = (struct sockaddr *)&udp_in6;
235 	} else
236 #endif
237 		append_sa = (struct sockaddr *)udp_in;
238 	m_adj(n, off);
239 
240 	so = inp->inp_socket;
241 	SOCKBUF_LOCK(&so->so_rcv);
242 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
243 		SOCKBUF_UNLOCK(&so->so_rcv);
244 		m_freem(n);
245 		if (opts)
246 			m_freem(opts);
247 		udpstat.udps_fullsock++;
248 	} else
249 		sorwakeup_locked(so);
250 }
251 
252 void
253 udp_input(struct mbuf *m, int off)
254 {
255 	int iphlen = off;
256 	struct ip *ip;
257 	struct udphdr *uh;
258 	struct ifnet *ifp;
259 	struct inpcb *inp;
260 	int len;
261 	struct ip save_ip;
262 	struct sockaddr_in udp_in;
263 #ifdef IPFIREWALL_FORWARD
264 	struct m_tag *fwd_tag;
265 #endif
266 
267 	ifp = m->m_pkthdr.rcvif;
268 	udpstat.udps_ipackets++;
269 
270 	/*
271 	 * Strip IP options, if any; should skip this, make available to
272 	 * user, and use on returned packets, but we don't yet have a way to
273 	 * check the checksum with options still present.
274 	 */
275 	if (iphlen > sizeof (struct ip)) {
276 		ip_stripoptions(m, (struct mbuf *)0);
277 		iphlen = sizeof(struct ip);
278 	}
279 
280 	/*
281 	 * Get IP and UDP header together in first mbuf.
282 	 */
283 	ip = mtod(m, struct ip *);
284 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
285 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
286 			udpstat.udps_hdrops++;
287 			return;
288 		}
289 		ip = mtod(m, struct ip *);
290 	}
291 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
292 
293 	/*
294 	 * Destination port of 0 is illegal, based on RFC768.
295 	 */
296 	if (uh->uh_dport == 0)
297 		goto badunlocked;
298 
299 	/*
300 	 * Construct sockaddr format source address.  Stuff source address
301 	 * and datagram in user buffer.
302 	 */
303 	bzero(&udp_in, sizeof(udp_in));
304 	udp_in.sin_len = sizeof(udp_in);
305 	udp_in.sin_family = AF_INET;
306 	udp_in.sin_port = uh->uh_sport;
307 	udp_in.sin_addr = ip->ip_src;
308 
309 	/*
310 	 * Make mbuf data length reflect UDP length.  If not enough data to
311 	 * reflect UDP length, drop.
312 	 */
313 	len = ntohs((u_short)uh->uh_ulen);
314 	if (ip->ip_len != len) {
315 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
316 			udpstat.udps_badlen++;
317 			goto badunlocked;
318 		}
319 		m_adj(m, len - ip->ip_len);
320 		/* ip->ip_len = len; */
321 	}
322 
323 	/*
324 	 * Save a copy of the IP header in case we want restore it for
325 	 * sending an ICMP error message in response.
326 	 */
327 	if (!udp_blackhole)
328 		save_ip = *ip;
329 	else
330 		memset(&save_ip, 0, sizeof(save_ip));
331 
332 	/*
333 	 * Checksum extended UDP header and data.
334 	 */
335 	if (uh->uh_sum) {
336 		u_short uh_sum;
337 
338 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
339 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
340 				uh_sum = m->m_pkthdr.csum_data;
341 			else
342 				uh_sum = in_pseudo(ip->ip_src.s_addr,
343 				    ip->ip_dst.s_addr, htonl((u_short)len +
344 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
345 			uh_sum ^= 0xffff;
346 		} else {
347 			char b[9];
348 
349 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
350 			bzero(((struct ipovly *)ip)->ih_x1, 9);
351 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
352 			uh_sum = in_cksum(m, len + sizeof (struct ip));
353 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
354 		}
355 		if (uh_sum) {
356 			udpstat.udps_badsum++;
357 			m_freem(m);
358 			return;
359 		}
360 	} else
361 		udpstat.udps_nosum++;
362 
363 #ifdef IPFIREWALL_FORWARD
364 	/*
365 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
366 	 */
367 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
368 	if (fwd_tag != NULL) {
369 		struct sockaddr_in *next_hop;
370 
371 		/*
372 		 * Do the hack.
373 		 */
374 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
375 		ip->ip_dst = next_hop->sin_addr;
376 		uh->uh_dport = ntohs(next_hop->sin_port);
377 
378 		/*
379 		 * Remove the tag from the packet.  We don't need it anymore.
380 		 */
381 		m_tag_delete(m, fwd_tag);
382 	}
383 #endif
384 
385 	INP_INFO_RLOCK(&udbinfo);
386 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
387 	    in_broadcast(ip->ip_dst, ifp)) {
388 		struct inpcb *last;
389 		struct ip_moptions *imo;
390 
391 		last = NULL;
392 		LIST_FOREACH(inp, &udb, inp_list) {
393 			if (inp->inp_lport != uh->uh_dport)
394 				continue;
395 #ifdef INET6
396 			if ((inp->inp_vflag & INP_IPV4) == 0)
397 				continue;
398 #endif
399 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
400 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
401 				continue;
402 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
403 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
404 				continue;
405 			/*
406 			 * XXX: Do not check source port of incoming datagram
407 			 * unless inp_connect() has been called to bind the
408 			 * fport part of the 4-tuple; the source could be
409 			 * trying to talk to us with an ephemeral port.
410 			 */
411 			if (inp->inp_fport != 0 &&
412 			    inp->inp_fport != uh->uh_sport)
413 				continue;
414 
415 			INP_WLOCK(inp);
416 
417 			/*
418 			 * Handle socket delivery policy for any-source
419 			 * and source-specific multicast. [RFC3678]
420 			 */
421 			imo = inp->inp_moptions;
422 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
423 			    imo != NULL) {
424 				struct sockaddr_in	 sin;
425 				struct in_msource	*ims;
426 				int			 blocked, mode;
427 				size_t			 idx;
428 
429 				bzero(&sin, sizeof(struct sockaddr_in));
430 				sin.sin_len = sizeof(struct sockaddr_in);
431 				sin.sin_family = AF_INET;
432 				sin.sin_addr = ip->ip_dst;
433 
434 				blocked = 0;
435 				idx = imo_match_group(imo, ifp,
436 				    (struct sockaddr *)&sin);
437 				if (idx == -1) {
438 					/*
439 					 * No group membership for this socket.
440 					 * Do not bump udps_noportbcast, as
441 					 * this will happen further down.
442 					 */
443 					blocked++;
444 				} else {
445 					/*
446 					 * Check for a multicast source filter
447 					 * entry on this socket for this group.
448 					 * MCAST_EXCLUDE is the default
449 					 * behaviour.  It means default accept;
450 					 * entries, if present, denote sources
451 					 * to be excluded from delivery.
452 					 */
453 					ims = imo_match_source(imo, idx,
454 					    (struct sockaddr *)&udp_in);
455 					mode = imo->imo_mfilters[idx].imf_fmode;
456 					if ((ims != NULL &&
457 					     mode == MCAST_EXCLUDE) ||
458 					    (ims == NULL &&
459 					     mode == MCAST_INCLUDE)) {
460 #ifdef DIAGNOSTIC
461 						if (bootverbose) {
462 							printf("%s: blocked by"
463 							    " source filter\n",
464 							    __func__);
465 						}
466 #endif
467 						udpstat.udps_filtermcast++;
468 						blocked++;
469 					}
470 				}
471 				if (blocked != 0) {
472 					INP_WUNLOCK(inp);
473 					continue;
474 				}
475 			}
476 			if (last != NULL) {
477 				struct mbuf *n;
478 
479 				n = m_copy(m, 0, M_COPYALL);
480 				if (n != NULL)
481 					udp_append(last, ip, n, iphlen +
482 					    sizeof(struct udphdr), &udp_in);
483 				INP_WUNLOCK(last);
484 			}
485 			last = inp;
486 			/*
487 			 * Don't look for additional matches if this one does
488 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
489 			 * socket options set.  This heuristic avoids
490 			 * searching through all pcbs in the common case of a
491 			 * non-shared port.  It assumes that an application
492 			 * will never clear these options after setting them.
493 			 */
494 			if ((last->inp_socket->so_options &
495 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
496 				break;
497 		}
498 
499 		if (last == NULL) {
500 			/*
501 			 * No matching pcb found; discard datagram.  (No need
502 			 * to send an ICMP Port Unreachable for a broadcast
503 			 * or multicast datgram.)
504 			 */
505 			udpstat.udps_noportbcast++;
506 			goto badheadlocked;
507 		}
508 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
509 		    &udp_in);
510 		INP_WUNLOCK(last);
511 		INP_INFO_RUNLOCK(&udbinfo);
512 		return;
513 	}
514 
515 	/*
516 	 * Locate pcb for datagram.
517 	 */
518 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
519 	    ip->ip_dst, uh->uh_dport, 1, ifp);
520 	if (inp == NULL) {
521 		if (udp_log_in_vain) {
522 			char buf[4*sizeof "123"];
523 
524 			strcpy(buf, inet_ntoa(ip->ip_dst));
525 			log(LOG_INFO,
526 			    "Connection attempt to UDP %s:%d from %s:%d\n",
527 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
528 			    ntohs(uh->uh_sport));
529 		}
530 		udpstat.udps_noport++;
531 		if (m->m_flags & (M_BCAST | M_MCAST)) {
532 			udpstat.udps_noportbcast++;
533 			goto badheadlocked;
534 		}
535 		if (udp_blackhole)
536 			goto badheadlocked;
537 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
538 			goto badheadlocked;
539 		*ip = save_ip;
540 		ip->ip_len += iphlen;
541 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
542 		INP_INFO_RUNLOCK(&udbinfo);
543 		return;
544 	}
545 
546 	/*
547 	 * Check the minimum TTL for socket.
548 	 */
549 	INP_WLOCK(inp);
550 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
551 		goto badheadlocked;
552 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
553 	INP_WUNLOCK(inp);
554 	INP_INFO_RUNLOCK(&udbinfo);
555 	return;
556 
557 badheadlocked:
558 	if (inp)
559 		INP_WUNLOCK(inp);
560 	INP_INFO_RUNLOCK(&udbinfo);
561 badunlocked:
562 	m_freem(m);
563 }
564 
565 /*
566  * Notify a udp user of an asynchronous error; just wake up so that they can
567  * collect error status.
568  */
569 struct inpcb *
570 udp_notify(struct inpcb *inp, int errno)
571 {
572 
573 	INP_WLOCK_ASSERT(inp);
574 
575 	inp->inp_socket->so_error = errno;
576 	sorwakeup(inp->inp_socket);
577 	sowwakeup(inp->inp_socket);
578 	return (inp);
579 }
580 
581 void
582 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
583 {
584 	struct ip *ip = vip;
585 	struct udphdr *uh;
586 	struct in_addr faddr;
587 	struct inpcb *inp;
588 
589 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
590 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
591 		return;
592 
593 	/*
594 	 * Redirects don't need to be handled up here.
595 	 */
596 	if (PRC_IS_REDIRECT(cmd))
597 		return;
598 
599 	/*
600 	 * Hostdead is ugly because it goes linearly through all PCBs.
601 	 *
602 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
603 	 * DoS attack on machines with many connections.
604 	 */
605 	if (cmd == PRC_HOSTDEAD)
606 		ip = NULL;
607 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
608 		return;
609 	if (ip != NULL) {
610 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
611 		INP_INFO_RLOCK(&udbinfo);
612 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
613 		    ip->ip_src, uh->uh_sport, 0, NULL);
614 		if (inp != NULL) {
615 			INP_WLOCK(inp);
616 			if (inp->inp_socket != NULL) {
617 				udp_notify(inp, inetctlerrmap[cmd]);
618 			}
619 			INP_WUNLOCK(inp);
620 		}
621 		INP_INFO_RUNLOCK(&udbinfo);
622 	} else
623 		in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd],
624 		    udp_notify);
625 }
626 
627 static int
628 udp_pcblist(SYSCTL_HANDLER_ARGS)
629 {
630 	int error, i, n;
631 	struct inpcb *inp, **inp_list;
632 	inp_gen_t gencnt;
633 	struct xinpgen xig;
634 
635 	/*
636 	 * The process of preparing the PCB list is too time-consuming and
637 	 * resource-intensive to repeat twice on every request.
638 	 */
639 	if (req->oldptr == 0) {
640 		n = udbinfo.ipi_count;
641 		req->oldidx = 2 * (sizeof xig)
642 			+ (n + n/8) * sizeof(struct xinpcb);
643 		return (0);
644 	}
645 
646 	if (req->newptr != 0)
647 		return (EPERM);
648 
649 	/*
650 	 * OK, now we're committed to doing something.
651 	 */
652 	INP_INFO_RLOCK(&udbinfo);
653 	gencnt = udbinfo.ipi_gencnt;
654 	n = udbinfo.ipi_count;
655 	INP_INFO_RUNLOCK(&udbinfo);
656 
657 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
658 		+ n * sizeof(struct xinpcb));
659 	if (error != 0)
660 		return (error);
661 
662 	xig.xig_len = sizeof xig;
663 	xig.xig_count = n;
664 	xig.xig_gen = gencnt;
665 	xig.xig_sogen = so_gencnt;
666 	error = SYSCTL_OUT(req, &xig, sizeof xig);
667 	if (error)
668 		return (error);
669 
670 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
671 	if (inp_list == 0)
672 		return (ENOMEM);
673 
674 	INP_INFO_RLOCK(&udbinfo);
675 	for (inp = LIST_FIRST(udbinfo.ipi_listhead), i = 0; inp && i < n;
676 	     inp = LIST_NEXT(inp, inp_list)) {
677 		INP_WLOCK(inp);
678 		if (inp->inp_gencnt <= gencnt &&
679 		    cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
680 			inp_list[i++] = inp;
681 		INP_WUNLOCK(inp);
682 	}
683 	INP_INFO_RUNLOCK(&udbinfo);
684 	n = i;
685 
686 	error = 0;
687 	for (i = 0; i < n; i++) {
688 		inp = inp_list[i];
689 		INP_WLOCK(inp);
690 		if (inp->inp_gencnt <= gencnt) {
691 			struct xinpcb xi;
692 			bzero(&xi, sizeof(xi));
693 			xi.xi_len = sizeof xi;
694 			/* XXX should avoid extra copy */
695 			bcopy(inp, &xi.xi_inp, sizeof *inp);
696 			if (inp->inp_socket)
697 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
698 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
699 			INP_WUNLOCK(inp);
700 			error = SYSCTL_OUT(req, &xi, sizeof xi);
701 		} else
702 			INP_WUNLOCK(inp);
703 	}
704 	if (!error) {
705 		/*
706 		 * Give the user an updated idea of our state.  If the
707 		 * generation differs from what we told her before, she knows
708 		 * that something happened while we were processing this
709 		 * request, and it might be necessary to retry.
710 		 */
711 		INP_INFO_RLOCK(&udbinfo);
712 		xig.xig_gen = udbinfo.ipi_gencnt;
713 		xig.xig_sogen = so_gencnt;
714 		xig.xig_count = udbinfo.ipi_count;
715 		INP_INFO_RUNLOCK(&udbinfo);
716 		error = SYSCTL_OUT(req, &xig, sizeof xig);
717 	}
718 	free(inp_list, M_TEMP);
719 	return (error);
720 }
721 
722 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
723     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
724 
725 static int
726 udp_getcred(SYSCTL_HANDLER_ARGS)
727 {
728 	struct xucred xuc;
729 	struct sockaddr_in addrs[2];
730 	struct inpcb *inp;
731 	int error;
732 
733 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
734 	if (error)
735 		return (error);
736 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
737 	if (error)
738 		return (error);
739 	INP_INFO_RLOCK(&udbinfo);
740 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
741 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
742 	if (inp == NULL || inp->inp_socket == NULL) {
743 		error = ENOENT;
744 		goto out;
745 	}
746 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
747 	if (error)
748 		goto out;
749 	cru2x(inp->inp_socket->so_cred, &xuc);
750 out:
751 	INP_INFO_RUNLOCK(&udbinfo);
752 	if (error == 0)
753 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
754 	return (error);
755 }
756 
757 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
758     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
759     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
760 
761 static int
762 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
763     struct mbuf *control, struct thread *td)
764 {
765 	struct udpiphdr *ui;
766 	int len = m->m_pkthdr.len;
767 	struct in_addr faddr, laddr;
768 	struct cmsghdr *cm;
769 	struct sockaddr_in *sin, src;
770 	int error = 0;
771 	int ipflags;
772 	u_short fport, lport;
773 	int unlock_udbinfo;
774 
775 	/*
776 	 * udp_output() may need to temporarily bind or connect the current
777 	 * inpcb.  As such, we don't know up front whether we will need the
778 	 * pcbinfo lock or not.  Do any work to decide what is needed up
779 	 * front before acquiring any locks.
780 	 */
781 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
782 		if (control)
783 			m_freem(control);
784 		m_freem(m);
785 		return (EMSGSIZE);
786 	}
787 
788 	src.sin_family = 0;
789 	if (control != NULL) {
790 		/*
791 		 * XXX: Currently, we assume all the optional information is
792 		 * stored in a single mbuf.
793 		 */
794 		if (control->m_next) {
795 			m_freem(control);
796 			m_freem(m);
797 			return (EINVAL);
798 		}
799 		for (; control->m_len > 0;
800 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
801 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
802 			cm = mtod(control, struct cmsghdr *);
803 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
804 			    || cm->cmsg_len > control->m_len) {
805 				error = EINVAL;
806 				break;
807 			}
808 			if (cm->cmsg_level != IPPROTO_IP)
809 				continue;
810 
811 			switch (cm->cmsg_type) {
812 			case IP_SENDSRCADDR:
813 				if (cm->cmsg_len !=
814 				    CMSG_LEN(sizeof(struct in_addr))) {
815 					error = EINVAL;
816 					break;
817 				}
818 				bzero(&src, sizeof(src));
819 				src.sin_family = AF_INET;
820 				src.sin_len = sizeof(src);
821 				src.sin_port = inp->inp_lport;
822 				src.sin_addr =
823 				    *(struct in_addr *)CMSG_DATA(cm);
824 				break;
825 
826 			default:
827 				error = ENOPROTOOPT;
828 				break;
829 			}
830 			if (error)
831 				break;
832 		}
833 		m_freem(control);
834 	}
835 	if (error) {
836 		m_freem(m);
837 		return (error);
838 	}
839 
840 	if (src.sin_family == AF_INET || addr != NULL) {
841 		INP_INFO_WLOCK(&udbinfo);
842 		unlock_udbinfo = 1;
843 	} else
844 		unlock_udbinfo = 0;
845 	INP_WLOCK(inp);
846 
847 #ifdef MAC
848 	mac_inpcb_create_mbuf(inp, m);
849 #endif
850 
851 	/*
852 	 * If the IP_SENDSRCADDR control message was specified, override the
853 	 * source address for this datagram.  Its use is invalidated if the
854 	 * address thus specified is incomplete or clobbers other inpcbs.
855 	 */
856 	laddr = inp->inp_laddr;
857 	lport = inp->inp_lport;
858 	if (src.sin_family == AF_INET) {
859 		if ((lport == 0) ||
860 		    (laddr.s_addr == INADDR_ANY &&
861 		     src.sin_addr.s_addr == INADDR_ANY)) {
862 			error = EINVAL;
863 			goto release;
864 		}
865 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
866 		    &laddr.s_addr, &lport, td->td_ucred);
867 		if (error)
868 			goto release;
869 	}
870 
871 	if (addr) {
872 		sin = (struct sockaddr_in *)addr;
873 		if (jailed(td->td_ucred))
874 			prison_remote_ip(td->td_ucred, 0,
875 			    &sin->sin_addr.s_addr);
876 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
877 			error = EISCONN;
878 			goto release;
879 		}
880 		error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport,
881 		    &faddr.s_addr, &fport, NULL, td->td_ucred);
882 		if (error)
883 			goto release;
884 
885 		/* Commit the local port if newly assigned. */
886 		if (inp->inp_laddr.s_addr == INADDR_ANY &&
887 		    inp->inp_lport == 0) {
888 			/*
889 			 * Remember addr if jailed, to prevent rebinding.
890 			 */
891 			if (jailed(td->td_ucred))
892 				inp->inp_laddr = laddr;
893 			inp->inp_lport = lport;
894 			if (in_pcbinshash(inp) != 0) {
895 				inp->inp_lport = 0;
896 				error = EAGAIN;
897 				goto release;
898 			}
899 			inp->inp_flags |= INP_ANONPORT;
900 		}
901 	} else {
902 		faddr = inp->inp_faddr;
903 		fport = inp->inp_fport;
904 		if (faddr.s_addr == INADDR_ANY) {
905 			error = ENOTCONN;
906 			goto release;
907 		}
908 	}
909 
910 	/*
911 	 * Calculate data length and get a mbuf for UDP, IP, and possible
912 	 * link-layer headers.  Immediate slide the data pointer back forward
913 	 * since we won't use that space at this layer.
914 	 */
915 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
916 	if (m == NULL) {
917 		error = ENOBUFS;
918 		goto release;
919 	}
920 	m->m_data += max_linkhdr;
921 	m->m_len -= max_linkhdr;
922 	m->m_pkthdr.len -= max_linkhdr;
923 
924 	/*
925 	 * Fill in mbuf with extended UDP header and addresses and length put
926 	 * into network format.
927 	 */
928 	ui = mtod(m, struct udpiphdr *);
929 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
930 	ui->ui_pr = IPPROTO_UDP;
931 	ui->ui_src = laddr;
932 	ui->ui_dst = faddr;
933 	ui->ui_sport = lport;
934 	ui->ui_dport = fport;
935 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
936 
937 	/*
938 	 * Set the Don't Fragment bit in the IP header.
939 	 */
940 	if (inp->inp_flags & INP_DONTFRAG) {
941 		struct ip *ip;
942 
943 		ip = (struct ip *)&ui->ui_i;
944 		ip->ip_off |= IP_DF;
945 	}
946 
947 	ipflags = 0;
948 	if (inp->inp_socket->so_options & SO_DONTROUTE)
949 		ipflags |= IP_ROUTETOIF;
950 	if (inp->inp_socket->so_options & SO_BROADCAST)
951 		ipflags |= IP_ALLOWBROADCAST;
952 	if (inp->inp_flags & INP_ONESBCAST)
953 		ipflags |= IP_SENDONES;
954 
955 	/*
956 	 * Set up checksum and output datagram.
957 	 */
958 	if (udp_cksum) {
959 		if (inp->inp_flags & INP_ONESBCAST)
960 			faddr.s_addr = INADDR_BROADCAST;
961 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
962 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
963 		m->m_pkthdr.csum_flags = CSUM_UDP;
964 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
965 	} else
966 		ui->ui_sum = 0;
967 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
968 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
969 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
970 	udpstat.udps_opackets++;
971 
972 	if (unlock_udbinfo)
973 		INP_INFO_WUNLOCK(&udbinfo);
974 	error = ip_output(m, inp->inp_options, NULL, ipflags,
975 	    inp->inp_moptions, inp);
976 	INP_WUNLOCK(inp);
977 	return (error);
978 
979 release:
980 	INP_WUNLOCK(inp);
981 	if (unlock_udbinfo)
982 		INP_INFO_WUNLOCK(&udbinfo);
983 	m_freem(m);
984 	return (error);
985 }
986 
987 static void
988 udp_abort(struct socket *so)
989 {
990 	struct inpcb *inp;
991 
992 	inp = sotoinpcb(so);
993 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
994 	INP_INFO_WLOCK(&udbinfo);
995 	INP_WLOCK(inp);
996 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
997 		in_pcbdisconnect(inp);
998 		inp->inp_laddr.s_addr = INADDR_ANY;
999 		soisdisconnected(so);
1000 	}
1001 	INP_WUNLOCK(inp);
1002 	INP_INFO_WUNLOCK(&udbinfo);
1003 }
1004 
1005 static int
1006 udp_attach(struct socket *so, int proto, struct thread *td)
1007 {
1008 	struct inpcb *inp;
1009 	int error;
1010 
1011 	inp = sotoinpcb(so);
1012 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1013 	error = soreserve(so, udp_sendspace, udp_recvspace);
1014 	if (error)
1015 		return (error);
1016 	INP_INFO_WLOCK(&udbinfo);
1017 	error = in_pcballoc(so, &udbinfo);
1018 	if (error) {
1019 		INP_INFO_WUNLOCK(&udbinfo);
1020 		return (error);
1021 	}
1022 
1023 	inp = (struct inpcb *)so->so_pcb;
1024 	INP_INFO_WUNLOCK(&udbinfo);
1025 	inp->inp_vflag |= INP_IPV4;
1026 	inp->inp_ip_ttl = ip_defttl;
1027 	INP_WUNLOCK(inp);
1028 	return (0);
1029 }
1030 
1031 static int
1032 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1033 {
1034 	struct inpcb *inp;
1035 	int error;
1036 
1037 	inp = sotoinpcb(so);
1038 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1039 	INP_INFO_WLOCK(&udbinfo);
1040 	INP_WLOCK(inp);
1041 	error = in_pcbbind(inp, nam, td->td_ucred);
1042 	INP_WUNLOCK(inp);
1043 	INP_INFO_WUNLOCK(&udbinfo);
1044 	return (error);
1045 }
1046 
1047 static void
1048 udp_close(struct socket *so)
1049 {
1050 	struct inpcb *inp;
1051 
1052 	inp = sotoinpcb(so);
1053 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1054 	INP_INFO_WLOCK(&udbinfo);
1055 	INP_WLOCK(inp);
1056 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1057 		in_pcbdisconnect(inp);
1058 		inp->inp_laddr.s_addr = INADDR_ANY;
1059 		soisdisconnected(so);
1060 	}
1061 	INP_WUNLOCK(inp);
1062 	INP_INFO_WUNLOCK(&udbinfo);
1063 }
1064 
1065 static int
1066 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1067 {
1068 	struct inpcb *inp;
1069 	int error;
1070 	struct sockaddr_in *sin;
1071 
1072 	inp = sotoinpcb(so);
1073 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1074 	INP_INFO_WLOCK(&udbinfo);
1075 	INP_WLOCK(inp);
1076 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1077 		INP_WUNLOCK(inp);
1078 		INP_INFO_WUNLOCK(&udbinfo);
1079 		return (EISCONN);
1080 	}
1081 	sin = (struct sockaddr_in *)nam;
1082 	if (jailed(td->td_ucred))
1083 		prison_remote_ip(td->td_ucred, 0, &sin->sin_addr.s_addr);
1084 	error = in_pcbconnect(inp, nam, td->td_ucred);
1085 	if (error == 0)
1086 		soisconnected(so);
1087 	INP_WUNLOCK(inp);
1088 	INP_INFO_WUNLOCK(&udbinfo);
1089 	return (error);
1090 }
1091 
1092 static void
1093 udp_detach(struct socket *so)
1094 {
1095 	struct inpcb *inp;
1096 
1097 	inp = sotoinpcb(so);
1098 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1099 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1100 	    ("udp_detach: not disconnected"));
1101 	INP_INFO_WLOCK(&udbinfo);
1102 	INP_WLOCK(inp);
1103 	in_pcbdetach(inp);
1104 	in_pcbfree(inp);
1105 	INP_INFO_WUNLOCK(&udbinfo);
1106 }
1107 
1108 static int
1109 udp_disconnect(struct socket *so)
1110 {
1111 	struct inpcb *inp;
1112 
1113 	inp = sotoinpcb(so);
1114 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1115 	INP_INFO_WLOCK(&udbinfo);
1116 	INP_WLOCK(inp);
1117 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1118 		INP_WUNLOCK(inp);
1119 		INP_INFO_WUNLOCK(&udbinfo);
1120 		return (ENOTCONN);
1121 	}
1122 
1123 	in_pcbdisconnect(inp);
1124 	inp->inp_laddr.s_addr = INADDR_ANY;
1125 	SOCK_LOCK(so);
1126 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1127 	SOCK_UNLOCK(so);
1128 	INP_WUNLOCK(inp);
1129 	INP_INFO_WUNLOCK(&udbinfo);
1130 	return (0);
1131 }
1132 
1133 static int
1134 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1135     struct mbuf *control, struct thread *td)
1136 {
1137 	struct inpcb *inp;
1138 
1139 	inp = sotoinpcb(so);
1140 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1141 	return (udp_output(inp, m, addr, control, td));
1142 }
1143 
1144 int
1145 udp_shutdown(struct socket *so)
1146 {
1147 	struct inpcb *inp;
1148 
1149 	inp = sotoinpcb(so);
1150 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1151 	INP_WLOCK(inp);
1152 	socantsendmore(so);
1153 	INP_WUNLOCK(inp);
1154 	return (0);
1155 }
1156 
1157 struct pr_usrreqs udp_usrreqs = {
1158 	.pru_abort =		udp_abort,
1159 	.pru_attach =		udp_attach,
1160 	.pru_bind =		udp_bind,
1161 	.pru_connect =		udp_connect,
1162 	.pru_control =		in_control,
1163 	.pru_detach =		udp_detach,
1164 	.pru_disconnect =	udp_disconnect,
1165 	.pru_peeraddr =		in_getpeeraddr,
1166 	.pru_send =		udp_send,
1167 	.pru_sosend =		sosend_dgram,
1168 	.pru_shutdown =		udp_shutdown,
1169 	.pru_sockaddr =		in_getsockaddr,
1170 	.pru_sosetlabel =	in_pcbsosetlabel,
1171 	.pru_close =		udp_close,
1172 };
1173