xref: /dflybsd-src/sys/netinet/udp_usrreq.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
67  * $FreeBSD: src/sys/netinet/udp_usrreq.c,v 1.64.2.18 2003/01/24 05:11:34 sam Exp $
68  * $DragonFly: src/sys/netinet/udp_usrreq.c,v 1.47 2008/11/11 10:46:58 sephe Exp $
69  */
70 
71 #include "opt_ipsec.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/kernel.h>
77 #include <sys/malloc.h>
78 #include <sys/mbuf.h>
79 #include <sys/domain.h>
80 #include <sys/proc.h>
81 #include <sys/priv.h>
82 #include <sys/protosw.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/sysctl.h>
86 #include <sys/syslog.h>
87 #include <sys/thread2.h>
88 #include <sys/in_cksum.h>
89 
90 #include <machine/stdarg.h>
91 
92 #include <vm/vm_zone.h>
93 
94 #include <net/if.h>
95 #include <net/route.h>
96 #include <net/netmsg2.h>
97 
98 #include <netinet/in.h>
99 #include <netinet/in_systm.h>
100 #include <netinet/ip.h>
101 #ifdef INET6
102 #include <netinet/ip6.h>
103 #endif
104 #include <netinet/in_pcb.h>
105 #include <netinet/in_var.h>
106 #include <netinet/ip_var.h>
107 #ifdef INET6
108 #include <netinet6/ip6_var.h>
109 #endif
110 #include <netinet/ip_icmp.h>
111 #include <netinet/icmp_var.h>
112 #include <netinet/udp.h>
113 #include <netinet/udp_var.h>
114 
115 #ifdef FAST_IPSEC
116 #include <netproto/ipsec/ipsec.h>
117 #endif
118 
119 #ifdef IPSEC
120 #include <netinet6/ipsec.h>
121 #endif
122 
123 int	udp_mpsafe_proto = 0;
124 TUNABLE_INT("net.inet.udp.mpsafe_proto", &udp_mpsafe_proto);
125 
126 int	udp_mpsafe_thread = NETMSG_SERVICE_ADAPTIVE;
127 TUNABLE_INT("net.inet.udp.mpsafe_thread", &udp_mpsafe_thread);
128 SYSCTL_INT(_net_inet_udp, OID_AUTO, mpsafe_thread, CTLFLAG_RW,
129 	   &udp_mpsafe_thread, 0,
130 	   "0:BGL, 1:Adaptive BGL, 2:No BGL(experimental)");
131 
132 /*
133  * UDP protocol implementation.
134  * Per RFC 768, August, 1980.
135  */
136 #ifndef	COMPAT_42
137 static int	udpcksum = 1;
138 #else
139 static int	udpcksum = 0;		/* XXX */
140 #endif
141 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
142 		&udpcksum, 0, "");
143 
144 int	log_in_vain = 0;
145 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
146     &log_in_vain, 0, "Log all incoming UDP packets");
147 
148 static int	blackhole = 0;
149 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
150 	&blackhole, 0, "Do not send port unreachables for refused connects");
151 
152 static int	strict_mcast_mship = 1;
153 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW,
154 	&strict_mcast_mship, 0, "Only send multicast to member sockets");
155 
156 struct	inpcbinfo udbinfo;
157 
158 #ifndef UDBHASHSIZE
159 #define UDBHASHSIZE 16
160 #endif
161 
162 struct	udpstat udpstat;	/* from udp_var.h */
163 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
164     &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
165 
166 static struct	sockaddr_in udp_in = { sizeof udp_in, AF_INET };
167 #ifdef INET6
168 struct udp_in6 {
169 	struct sockaddr_in6	uin6_sin;
170 	u_char			uin6_init_done : 1;
171 } udp_in6 = {
172 	{ sizeof udp_in6.uin6_sin, AF_INET6 },
173 	0
174 };
175 struct udp_ip6 {
176 	struct ip6_hdr		uip6_ip6;
177 	u_char			uip6_init_done : 1;
178 } udp_ip6;
179 #endif /* INET6 */
180 
181 static void udp_append (struct inpcb *last, struct ip *ip,
182 			    struct mbuf *n, int off);
183 #ifdef INET6
184 static void ip_2_ip6_hdr (struct ip6_hdr *ip6, struct ip *ip);
185 #endif
186 
187 static int udp_connect_oncpu(struct socket *so, struct thread *td,
188 			struct sockaddr_in *sin, struct sockaddr_in *if_sin);
189 static int udp_detach (struct socket *so);
190 static	int udp_output (struct inpcb *, struct mbuf *, struct sockaddr *,
191 			    struct mbuf *, struct thread *);
192 
193 void
194 udp_init(void)
195 {
196 	in_pcbinfo_init(&udbinfo);
197 	udbinfo.hashbase = hashinit(UDBHASHSIZE, M_PCB, &udbinfo.hashmask);
198 	udbinfo.porthashbase = hashinit(UDBHASHSIZE, M_PCB,
199 					&udbinfo.porthashmask);
200 	udbinfo.wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB,
201 					    &udbinfo.wildcardhashmask);
202 	udbinfo.ipi_zone = zinit("udpcb", sizeof(struct inpcb), maxsockets,
203 				 ZONE_INTERRUPT, 0);
204 	udp_thread_init();
205 }
206 
207 /*
208  * Check multicast packets to make sure they are only sent to sockets with
209  * multicast memberships for the packet's destination address and arrival
210  * interface.  Multicast packets to multicast-unaware sockets are also
211  * disallowed.
212  *
213  * Returns 0 if the packet is acceptable, -1 if it is not.
214  */
215 static __inline int
216 check_multicast_membership(struct ip *ip, struct inpcb *inp, struct mbuf *m)
217 {
218 	int mshipno;
219 	struct ip_moptions *mopt;
220 
221 	if (strict_mcast_mship == 0 ||
222 	    !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
223 		return (0);
224 	}
225 	mopt = inp->inp_moptions;
226 	if (mopt == NULL)
227 		return (-1);
228 	for (mshipno = 0; mshipno <= mopt->imo_num_memberships; ++mshipno) {
229 		struct in_multi *maddr = mopt->imo_membership[mshipno];
230 
231 		if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr &&
232 		    m->m_pkthdr.rcvif == maddr->inm_ifp) {
233 			return (0);
234 		}
235 	}
236 	return (-1);
237 }
238 
239 void
240 udp_input(struct mbuf *m, ...)
241 {
242 	int iphlen;
243 	struct ip *ip;
244 	struct udphdr *uh;
245 	struct inpcb *inp;
246 	struct mbuf *opts = NULL;
247 	int len, off, proto;
248 	struct ip save_ip;
249 	struct sockaddr *append_sa;
250 	__va_list ap;
251 
252 	__va_start(ap, m);
253 	off = __va_arg(ap, int);
254 	proto = __va_arg(ap, int);
255 	__va_end(ap);
256 
257 	iphlen = off;
258 	udpstat.udps_ipackets++;
259 
260 	/*
261 	 * Strip IP options, if any; should skip this,
262 	 * make available to user, and use on returned packets,
263 	 * but we don't yet have a way to check the checksum
264 	 * with options still present.
265 	 */
266 	if (iphlen > sizeof(struct ip)) {
267 		ip_stripoptions(m);
268 		iphlen = sizeof(struct ip);
269 	}
270 
271 	/*
272 	 * IP and UDP headers are together in first mbuf.
273 	 * Already checked and pulled up in ip_demux().
274 	 */
275 	KASSERT(m->m_len >= iphlen + sizeof(struct udphdr),
276 	    ("UDP header not in one mbuf"));
277 
278 	ip = mtod(m, struct ip *);
279 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
280 
281 	/* destination port of 0 is illegal, based on RFC768. */
282 	if (uh->uh_dport == 0)
283 		goto bad;
284 
285 	/*
286 	 * Make mbuf data length reflect UDP length.
287 	 * If not enough data to reflect UDP length, drop.
288 	 */
289 	len = ntohs((u_short)uh->uh_ulen);
290 	if (ip->ip_len != len) {
291 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
292 			udpstat.udps_badlen++;
293 			goto bad;
294 		}
295 		m_adj(m, len - ip->ip_len);
296 		/* ip->ip_len = len; */
297 	}
298 	/*
299 	 * Save a copy of the IP header in case we want restore it
300 	 * for sending an ICMP error message in response.
301 	 */
302 	save_ip = *ip;
303 
304 	/*
305 	 * Checksum extended UDP header and data.
306 	 */
307 	if (uh->uh_sum) {
308 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
309 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
310 				uh->uh_sum = m->m_pkthdr.csum_data;
311 			else
312 				uh->uh_sum = in_pseudo(ip->ip_src.s_addr,
313 				    ip->ip_dst.s_addr, htonl((u_short)len +
314 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
315 			uh->uh_sum ^= 0xffff;
316 		} else {
317 			char b[9];
318 
319 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
320 			bzero(((struct ipovly *)ip)->ih_x1, 9);
321 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
322 			uh->uh_sum = in_cksum(m, len + sizeof(struct ip));
323 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
324 		}
325 		if (uh->uh_sum) {
326 			udpstat.udps_badsum++;
327 			m_freem(m);
328 			return;
329 		}
330 	} else
331 		udpstat.udps_nosum++;
332 
333 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
334 	    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
335 		struct inpcb *last;
336 
337 		/*
338 		 * Deliver a multicast or broadcast datagram to *all* sockets
339 		 * for which the local and remote addresses and ports match
340 		 * those of the incoming datagram.  This allows more than
341 		 * one process to receive multi/broadcasts on the same port.
342 		 * (This really ought to be done for unicast datagrams as
343 		 * well, but that would cause problems with existing
344 		 * applications that open both address-specific sockets and
345 		 * a wildcard socket listening to the same port -- they would
346 		 * end up receiving duplicates of every unicast datagram.
347 		 * Those applications open the multiple sockets to overcome an
348 		 * inadequacy of the UDP socket interface, but for backwards
349 		 * compatibility we avoid the problem here rather than
350 		 * fixing the interface.  Maybe 4.5BSD will remedy this?)
351 		 */
352 
353 		/*
354 		 * Construct sockaddr format source address.
355 		 */
356 		udp_in.sin_port = uh->uh_sport;
357 		udp_in.sin_addr = ip->ip_src;
358 		/*
359 		 * Locate pcb(s) for datagram.
360 		 * (Algorithm copied from raw_intr().)
361 		 */
362 		last = NULL;
363 #ifdef INET6
364 		udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0;
365 #endif
366 		LIST_FOREACH(inp, &udbinfo.pcblisthead, inp_list) {
367 			if (inp->inp_flags & INP_PLACEMARKER)
368 				continue;
369 #ifdef INET6
370 			if (!(inp->inp_vflag & INP_IPV4))
371 				continue;
372 #endif
373 			if (inp->inp_lport != uh->uh_dport)
374 				continue;
375 			if (inp->inp_laddr.s_addr != INADDR_ANY) {
376 				if (inp->inp_laddr.s_addr !=
377 				    ip->ip_dst.s_addr)
378 					continue;
379 			}
380 			if (inp->inp_faddr.s_addr != INADDR_ANY) {
381 				if (inp->inp_faddr.s_addr !=
382 				    ip->ip_src.s_addr ||
383 				    inp->inp_fport != uh->uh_sport)
384 					continue;
385 			}
386 
387 			if (check_multicast_membership(ip, inp, m) < 0)
388 				continue;
389 
390 			if (last != NULL) {
391 				struct mbuf *n;
392 
393 #ifdef IPSEC
394 				/* check AH/ESP integrity. */
395 				if (ipsec4_in_reject_so(m, last->inp_socket))
396 					ipsecstat.in_polvio++;
397 					/* do not inject data to pcb */
398 				else
399 #endif /*IPSEC*/
400 #ifdef FAST_IPSEC
401 				/* check AH/ESP integrity. */
402 				if (ipsec4_in_reject(m, last))
403 					;
404 				else
405 #endif /*FAST_IPSEC*/
406 				if ((n = m_copypacket(m, MB_DONTWAIT)) != NULL)
407 					udp_append(last, ip, n,
408 						   iphlen +
409 						   sizeof(struct udphdr));
410 			}
411 			last = inp;
412 			/*
413 			 * Don't look for additional matches if this one does
414 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
415 			 * socket options set.  This heuristic avoids searching
416 			 * through all pcbs in the common case of a non-shared
417 			 * port.  It * assumes that an application will never
418 			 * clear these options after setting them.
419 			 */
420 			if (!(last->inp_socket->so_options &
421 			    (SO_REUSEPORT | SO_REUSEADDR)))
422 				break;
423 		}
424 
425 		if (last == NULL) {
426 			/*
427 			 * No matching pcb found; discard datagram.
428 			 * (No need to send an ICMP Port Unreachable
429 			 * for a broadcast or multicast datgram.)
430 			 */
431 			udpstat.udps_noportbcast++;
432 			goto bad;
433 		}
434 #ifdef IPSEC
435 		/* check AH/ESP integrity. */
436 		if (ipsec4_in_reject_so(m, last->inp_socket)) {
437 			ipsecstat.in_polvio++;
438 			goto bad;
439 		}
440 #endif /*IPSEC*/
441 #ifdef FAST_IPSEC
442 		/* check AH/ESP integrity. */
443 		if (ipsec4_in_reject(m, last))
444 			goto bad;
445 #endif /*FAST_IPSEC*/
446 		udp_append(last, ip, m, iphlen + sizeof(struct udphdr));
447 		return;
448 	}
449 	/*
450 	 * Locate pcb for datagram.
451 	 */
452 	inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
453 	    ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif);
454 	if (inp == NULL) {
455 		if (log_in_vain) {
456 			char buf[sizeof "aaa.bbb.ccc.ddd"];
457 
458 			strcpy(buf, inet_ntoa(ip->ip_dst));
459 			log(LOG_INFO,
460 			    "Connection attempt to UDP %s:%d from %s:%d\n",
461 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
462 			    ntohs(uh->uh_sport));
463 		}
464 		udpstat.udps_noport++;
465 		if (m->m_flags & (M_BCAST | M_MCAST)) {
466 			udpstat.udps_noportbcast++;
467 			goto bad;
468 		}
469 		if (blackhole)
470 			goto bad;
471 #ifdef ICMP_BANDLIM
472 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
473 			goto bad;
474 #endif
475 		*ip = save_ip;
476 		ip->ip_len += iphlen;
477 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
478 		return;
479 	}
480 #ifdef IPSEC
481 	if (ipsec4_in_reject_so(m, inp->inp_socket)) {
482 		ipsecstat.in_polvio++;
483 		goto bad;
484 	}
485 #endif /*IPSEC*/
486 #ifdef FAST_IPSEC
487 	if (ipsec4_in_reject(m, inp))
488 		goto bad;
489 #endif /*FAST_IPSEC*/
490 	/*
491 	 * Check the minimum TTL for socket.
492 	 */
493 	if (ip->ip_ttl < inp->inp_ip_minttl)
494 		goto bad;
495 
496 	/*
497 	 * Construct sockaddr format source address.
498 	 * Stuff source address and datagram in user buffer.
499 	 */
500 	udp_in.sin_port = uh->uh_sport;
501 	udp_in.sin_addr = ip->ip_src;
502 	if ((inp->inp_flags & INP_CONTROLOPTS) ||
503 	    (inp->inp_socket->so_options & SO_TIMESTAMP)) {
504 #ifdef INET6
505 		if (inp->inp_vflag & INP_IPV6) {
506 			int savedflags;
507 
508 			ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
509 			savedflags = inp->inp_flags;
510 			inp->inp_flags &= ~INP_UNMAPPABLEOPTS;
511 			ip6_savecontrol(inp, &opts, &udp_ip6.uip6_ip6, m);
512 			inp->inp_flags = savedflags;
513 		} else
514 #endif
515 		ip_savecontrol(inp, &opts, ip, m);
516 	}
517 	m_adj(m, iphlen + sizeof(struct udphdr));
518 #ifdef INET6
519 	if (inp->inp_vflag & INP_IPV6) {
520 		in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
521 		append_sa = (struct sockaddr *)&udp_in6;
522 	} else
523 #endif
524 		append_sa = (struct sockaddr *)&udp_in;
525 	if (ssb_appendaddr(&inp->inp_socket->so_rcv, append_sa, m, opts) == 0) {
526 		udpstat.udps_fullsock++;
527 		goto bad;
528 	}
529 	sorwakeup(inp->inp_socket);
530 	return;
531 bad:
532 	m_freem(m);
533 	if (opts)
534 		m_freem(opts);
535 	return;
536 }
537 
538 #ifdef INET6
539 static void
540 ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip)
541 {
542 	bzero(ip6, sizeof *ip6);
543 
544 	ip6->ip6_vfc = IPV6_VERSION;
545 	ip6->ip6_plen = ip->ip_len;
546 	ip6->ip6_nxt = ip->ip_p;
547 	ip6->ip6_hlim = ip->ip_ttl;
548 	ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] =
549 		IPV6_ADDR_INT32_SMP;
550 	ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr;
551 	ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr;
552 }
553 #endif
554 
555 /*
556  * subroutine of udp_input(), mainly for source code readability.
557  * caller must properly init udp_ip6 and udp_in6 beforehand.
558  */
559 static void
560 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off)
561 {
562 	struct sockaddr *append_sa;
563 	struct mbuf *opts = NULL;
564 
565 	if (last->inp_flags & INP_CONTROLOPTS ||
566 	    last->inp_socket->so_options & SO_TIMESTAMP) {
567 #ifdef INET6
568 		if (last->inp_vflag & INP_IPV6) {
569 			int savedflags;
570 
571 			if (udp_ip6.uip6_init_done == 0) {
572 				ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip);
573 				udp_ip6.uip6_init_done = 1;
574 			}
575 			savedflags = last->inp_flags;
576 			last->inp_flags &= ~INP_UNMAPPABLEOPTS;
577 			ip6_savecontrol(last, &opts, &udp_ip6.uip6_ip6, n);
578 			last->inp_flags = savedflags;
579 		} else
580 #endif
581 		ip_savecontrol(last, &opts, ip, n);
582 	}
583 #ifdef INET6
584 	if (last->inp_vflag & INP_IPV6) {
585 		if (udp_in6.uin6_init_done == 0) {
586 			in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin);
587 			udp_in6.uin6_init_done = 1;
588 		}
589 		append_sa = (struct sockaddr *)&udp_in6.uin6_sin;
590 	} else
591 #endif
592 		append_sa = (struct sockaddr *)&udp_in;
593 	m_adj(n, off);
594 	if (ssb_appendaddr(&last->inp_socket->so_rcv, append_sa, n, opts) == 0) {
595 		m_freem(n);
596 		if (opts)
597 			m_freem(opts);
598 		udpstat.udps_fullsock++;
599 	} else
600 		sorwakeup(last->inp_socket);
601 }
602 
603 /*
604  * Notify a udp user of an asynchronous error;
605  * just wake up so that he can collect error status.
606  */
607 void
608 udp_notify(struct inpcb *inp, int error)
609 {
610 	inp->inp_socket->so_error = error;
611 	sorwakeup(inp->inp_socket);
612 	sowwakeup(inp->inp_socket);
613 }
614 
615 struct netmsg_udp_notify {
616 	struct netmsg	nm_nmsg;
617 	void		(*nm_notify)(struct inpcb *, int);
618 	struct in_addr	nm_faddr;
619 	int		nm_arg;
620 };
621 
622 static void
623 udp_notifyall_oncpu(struct netmsg *netmsg)
624 {
625 	struct netmsg_udp_notify *nmsg = (struct netmsg_udp_notify *)netmsg;
626 	int nextcpu;
627 
628 	in_pcbnotifyall(&udbinfo.pcblisthead, nmsg->nm_faddr, nmsg->nm_arg,
629 			nmsg->nm_notify);
630 
631 	nextcpu = mycpuid + 1;
632 	if (nextcpu < ncpus2)
633 		lwkt_forwardmsg(udp_cport(nextcpu), &netmsg->nm_lmsg);
634 	else
635 		lwkt_replymsg(&netmsg->nm_lmsg, 0);
636 }
637 
638 static void
639 udp_rtchange(struct inpcb *inp, int err)
640 {
641 #ifdef SMP
642 	/* XXX Nuke this, once UDP inpcbs are CPU localized */
643 	if (inp->inp_route.ro_rt && inp->inp_route.ro_rt->rt_cpuid == mycpuid) {
644 		rtfree(inp->inp_route.ro_rt);
645 		inp->inp_route.ro_rt = NULL;
646 		/*
647 		 * A new route can be allocated the next time
648 		 * output is attempted.
649 		 */
650 	}
651 #else
652 	in_rtchange(inp, err);
653 #endif
654 }
655 
656 void
657 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
658 {
659 	struct ip *ip = vip;
660 	struct udphdr *uh;
661 	void (*notify) (struct inpcb *, int) = udp_notify;
662 	struct in_addr faddr;
663 	struct inpcb *inp;
664 
665 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
666 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
667 		return;
668 
669 	if (PRC_IS_REDIRECT(cmd)) {
670 		ip = NULL;
671 		notify = udp_rtchange;
672 	} else if (cmd == PRC_HOSTDEAD)
673 		ip = NULL;
674 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
675 		return;
676 	if (ip) {
677 		crit_enter();
678 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
679 		inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
680 					ip->ip_src, uh->uh_sport, 0, NULL);
681 		if (inp != NULL && inp->inp_socket != NULL)
682 			(*notify)(inp, inetctlerrmap[cmd]);
683 		crit_exit();
684 	} else if (PRC_IS_REDIRECT(cmd)) {
685 		struct netmsg_udp_notify nmsg;
686 
687 		KKASSERT(&curthread->td_msgport == cpu_portfn(0));
688 		netmsg_init(&nmsg.nm_nmsg, NULL, &curthread->td_msgport,
689 			    0, udp_notifyall_oncpu);
690 		nmsg.nm_faddr = faddr;
691 		nmsg.nm_arg = inetctlerrmap[cmd];
692 		nmsg.nm_notify = notify;
693 
694 		lwkt_domsg(udp_cport(0), &nmsg.nm_nmsg.nm_lmsg, 0);
695 	} else {
696 		/*
697 		 * XXX We should forward msg upon PRC_HOSTHEAD and ip == NULL,
698 		 * once UDP inpcbs are CPU localized
699 		 */
700 		KKASSERT(&curthread->td_msgport == udp_cport(0));
701 		in_pcbnotifyall(&udbinfo.pcblisthead, faddr, inetctlerrmap[cmd],
702 				notify);
703 	}
704 }
705 
706 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, &udbinfo, 0,
707 	    in_pcblist_global, "S,xinpcb", "List of active UDP sockets");
708 
709 static int
710 udp_getcred(SYSCTL_HANDLER_ARGS)
711 {
712 	struct sockaddr_in addrs[2];
713 	struct inpcb *inp;
714 	int error;
715 
716 	error = priv_check(req->td, PRIV_ROOT);
717 	if (error)
718 		return (error);
719 	error = SYSCTL_IN(req, addrs, sizeof addrs);
720 	if (error)
721 		return (error);
722 	crit_enter();
723 	inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
724 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
725 	if (inp == NULL || inp->inp_socket == NULL) {
726 		error = ENOENT;
727 		goto out;
728 	}
729 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
730 out:
731 	crit_exit();
732 	return (error);
733 }
734 
735 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
736     0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection");
737 
738 static int
739 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *dstaddr,
740 	   struct mbuf *control, struct thread *td)
741 {
742 	struct udpiphdr *ui;
743 	int len = m->m_pkthdr.len;
744 	struct sockaddr_in *sin;	/* really is initialized before use */
745 	int error = 0, lport_any = 0;
746 
747 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
748 		error = EMSGSIZE;
749 		goto release;
750 	}
751 
752 	if (inp->inp_lport == 0) {	/* unbound socket */
753 		error = in_pcbbind(inp, NULL, td);
754 		if (error)
755 			goto release;
756 		in_pcbinswildcardhash(inp);
757 		lport_any = 1;
758 	}
759 
760 	if (dstaddr != NULL) {		/* destination address specified */
761 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
762 			/* already connected */
763 			error = EISCONN;
764 			goto release;
765 		}
766 		sin = (struct sockaddr_in *)dstaddr;
767 		if (!prison_remote_ip(td, (struct sockaddr *)&sin)) {
768 			error = EAFNOSUPPORT; /* IPv6 only jail */
769 			goto release;
770 		}
771 	} else {
772 		if (inp->inp_faddr.s_addr == INADDR_ANY) {
773 			/* no destination specified and not already connected */
774 			error = ENOTCONN;
775 			goto release;
776 		}
777 		sin = NULL;
778 	}
779 
780 	/*
781 	 * Calculate data length and get a mbuf
782 	 * for UDP and IP headers.
783 	 */
784 	M_PREPEND(m, sizeof(struct udpiphdr), MB_DONTWAIT);
785 	if (m == NULL) {
786 		error = ENOBUFS;
787 		goto release;
788 	}
789 
790 	/*
791 	 * Fill in mbuf with extended UDP header
792 	 * and addresses and length put into network format.
793 	 */
794 	ui = mtod(m, struct udpiphdr *);
795 	bzero(ui->ui_x1, sizeof ui->ui_x1);	/* XXX still needed? */
796 	ui->ui_pr = IPPROTO_UDP;
797 
798 	/*
799 	 * Set destination address.
800 	 */
801 	if (dstaddr != NULL) {			/* use specified destination */
802 		ui->ui_dst = sin->sin_addr;
803 		ui->ui_dport = sin->sin_port;
804 	} else {				/* use connected destination */
805 		ui->ui_dst = inp->inp_faddr;
806 		ui->ui_dport = inp->inp_fport;
807 	}
808 
809 	/*
810 	 * Set source address.
811 	 */
812 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
813 		struct sockaddr_in *if_sin;
814 
815 		if (dstaddr == NULL) {
816 			/*
817 			 * connect() had (or should have) failed because
818 			 * the interface had no IP address, but the
819 			 * application proceeded to call send() anyways.
820 			 */
821 			error = ENOTCONN;
822 			goto release;
823 		}
824 
825 		/* Look up outgoing interface. */
826 		if ((error = in_pcbladdr(inp, dstaddr, &if_sin, td)))
827 			goto release;
828 		ui->ui_src = if_sin->sin_addr;	/* use address of interface */
829 	} else {
830 		ui->ui_src = inp->inp_laddr;	/* use non-null bound address */
831 	}
832 	ui->ui_sport = inp->inp_lport;
833 	KASSERT(inp->inp_lport != 0, ("inp lport should have been bound"));
834 
835 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
836 
837 	/*
838 	 * Set up checksum and output datagram.
839 	 */
840 	if (udpcksum) {
841 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr,
842 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
843 		m->m_pkthdr.csum_flags = CSUM_UDP;
844 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
845 	} else {
846 		ui->ui_sum = 0;
847 	}
848 	((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len;
849 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
850 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
851 	udpstat.udps_opackets++;
852 
853 	error = ip_output(m, inp->inp_options, &inp->inp_route,
854 	    (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)) |
855 	    IP_DEBUGROUTE,
856 	    inp->inp_moptions, inp);
857 
858 	/*
859 	 * If this is the first data gram sent on an unbound and unconnected
860 	 * UDP socket, lport will be changed in this function.  If target
861 	 * CPU after this lport changing is no longer the current CPU, then
862 	 * free the route entry allocated on the current CPU.
863 	 */
864 	if (lport_any) {
865 		if (udp_addrcpu(inp->inp_faddr.s_addr, inp->inp_fport,
866 		    inp->inp_laddr.s_addr, inp->inp_lport) != mycpuid) {
867 			struct route *ro = &inp->inp_route;
868 
869 			if (ro->ro_rt != NULL)
870 				RTFREE(ro->ro_rt);
871 			bzero(ro, sizeof(*ro));
872 		}
873 	}
874 	return (error);
875 
876 release:
877 	m_freem(m);
878 	return (error);
879 }
880 
881 u_long	udp_sendspace = 9216;		/* really max datagram size */
882 					/* 40 1K datagrams */
883 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
884     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
885 
886 u_long	udp_recvspace = 40 * (1024 +
887 #ifdef INET6
888 				      sizeof(struct sockaddr_in6)
889 #else
890 				      sizeof(struct sockaddr_in)
891 #endif
892 				      );
893 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
894     &udp_recvspace, 0, "Maximum incoming UDP datagram size");
895 
896 static int
897 udp_abort(struct socket *so)
898 {
899 	struct inpcb *inp;
900 
901 	inp = so->so_pcb;
902 	if (inp == NULL)
903 		return EINVAL;	/* ??? possible? panic instead? */
904 	soisdisconnected(so);
905 	crit_enter();
906 	in_pcbdetach(inp);
907 	crit_exit();
908 	return 0;
909 }
910 
911 static int
912 udp_attach(struct socket *so, int proto, struct pru_attach_info *ai)
913 {
914 	struct inpcb *inp;
915 	int error;
916 
917 	inp = so->so_pcb;
918 	if (inp != NULL)
919 		return EINVAL;
920 
921 	error = soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit);
922 	if (error)
923 		return error;
924 	crit_enter();
925 	error = in_pcballoc(so, &udbinfo);
926 	crit_exit();
927 	if (error)
928 		return error;
929 	so->so_port = udp_soport_attach(so);
930 
931 	inp = (struct inpcb *)so->so_pcb;
932 	inp->inp_vflag |= INP_IPV4;
933 	inp->inp_ip_ttl = ip_defttl;
934 	return 0;
935 }
936 
937 static int
938 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
939 {
940 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
941 	struct inpcb *inp;
942 	int error;
943 
944 	inp = so->so_pcb;
945 	if (inp == NULL)
946 		return EINVAL;
947 	crit_enter();
948 	error = in_pcbbind(inp, nam, td);
949 	crit_exit();
950 	if (error == 0) {
951 		if (sin->sin_addr.s_addr != INADDR_ANY)
952 			inp->inp_flags |= INP_WASBOUND_NOTANY;
953 		in_pcbinswildcardhash(inp);
954 	}
955 	return error;
956 }
957 
958 #ifdef SMP
959 
960 struct netmsg_udp_connect {
961 	struct netmsg		nm_netmsg;
962 	struct socket		*nm_so;
963 	struct sockaddr_in	*nm_sin;
964 	struct sockaddr_in	*nm_ifsin;
965 	struct thread		*nm_td;
966 };
967 
968 static void
969 udp_connect_handler(netmsg_t netmsg)
970 {
971 	struct netmsg_udp_connect *msg = (void *)netmsg;
972 	int error;
973 
974 	error = udp_connect_oncpu(msg->nm_so, msg->nm_td,
975 				  msg->nm_sin, msg->nm_ifsin);
976 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
977 }
978 
979 #endif
980 
981 static int
982 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
983 {
984 	struct inpcb *inp;
985 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
986 	struct sockaddr_in *if_sin;
987 	lwkt_port_t port;
988 	int error;
989 
990 	inp = so->so_pcb;
991 	if (inp == NULL)
992 		return EINVAL;
993 	if (inp->inp_faddr.s_addr != INADDR_ANY)
994 		return EISCONN;
995 	error = 0;
996 
997 	/*
998 	 * Bind if we have to
999 	 */
1000 	if (td->td_proc && td->td_proc->p_ucred->cr_prison != NULL &&
1001 	    inp->inp_laddr.s_addr == INADDR_ANY) {
1002 		error = in_pcbbind(inp, NULL, td);
1003 		if (error)
1004 			return (error);
1005 	}
1006 
1007 	/*
1008 	 * Calculate the correct protocol processing thread.  The connect
1009 	 * operation must run there.
1010 	 */
1011 	error = in_pcbladdr(inp, nam, &if_sin, td);
1012 	if (error)
1013 		return(error);
1014 	if (!prison_remote_ip(td, nam))
1015 		return(EAFNOSUPPORT); /* IPv6 only jail */
1016 
1017 	port = udp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1018 			    inp->inp_laddr.s_addr, inp->inp_lport);
1019 #ifdef SMP
1020 	if (port != &curthread->td_msgport) {
1021 		struct netmsg_udp_connect msg;
1022 		struct route *ro = &inp->inp_route;
1023 
1024 		/*
1025 		 * in_pcbladdr() may have allocated a route entry for us
1026 		 * on the current CPU, but we need a route entry on the
1027 		 * inpcb's owner CPU, so free it here.
1028 		 */
1029 		if (ro->ro_rt != NULL)
1030 			RTFREE(ro->ro_rt);
1031 		bzero(ro, sizeof(*ro));
1032 
1033 		/*
1034 		 * NOTE: We haven't set so->so_port yet do not pass so
1035 		 *       to netmsg_init() or it will be improperly forwarded.
1036 		 */
1037 		netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1038 			    0, udp_connect_handler);
1039 		msg.nm_so = so;
1040 		msg.nm_sin = sin;
1041 		msg.nm_ifsin = if_sin;
1042 		msg.nm_td = td;
1043 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1044 	} else {
1045 		error = udp_connect_oncpu(so, td, sin, if_sin);
1046 	}
1047 #else
1048 	error = udp_connect_oncpu(so, td, sin, if_sin);
1049 #endif
1050 	return (error);
1051 }
1052 
1053 static int
1054 udp_connect_oncpu(struct socket *so, struct thread *td,
1055 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
1056 {
1057 	struct inpcb *inp;
1058 	int error;
1059 
1060 	inp = so->so_pcb;
1061 	if (inp->inp_flags & INP_WILDCARD)
1062 		in_pcbremwildcardhash(inp);
1063 	error = in_pcbconnect(inp, (struct sockaddr *)sin, td);
1064 
1065 	if (error == 0) {
1066 		/*
1067 		 * No more errors can occur, finish adjusting the socket
1068 		 * and change the processing port to reflect the connected
1069 		 * socket.  Once set we can no longer safely mess with the
1070 		 * socket.
1071 		 */
1072 		soisconnected(so);
1073 		sosetport(so, &curthread->td_msgport);
1074 	} else if (error == EAFNOSUPPORT) {	/* connection dissolved */
1075 		/*
1076 		 * Follow traditional BSD behavior and retain
1077 		 * the local port binding.  But, fix the old misbehavior
1078 		 * of overwriting any previously bound local address.
1079 		 */
1080 		if (!(inp->inp_flags & INP_WASBOUND_NOTANY))
1081 			inp->inp_laddr.s_addr = INADDR_ANY;
1082 		in_pcbinswildcardhash(inp);
1083 	}
1084 	return error;
1085 }
1086 
1087 static int
1088 udp_detach(struct socket *so)
1089 {
1090 	struct inpcb *inp;
1091 
1092 	inp = so->so_pcb;
1093 	if (inp == NULL)
1094 		return EINVAL;
1095 	crit_enter();
1096 	in_pcbdetach(inp);
1097 	crit_exit();
1098 	return 0;
1099 }
1100 
1101 static int
1102 udp_disconnect(struct socket *so)
1103 {
1104 	struct route *ro;
1105 	struct inpcb *inp;
1106 
1107 	inp = so->so_pcb;
1108 	if (inp == NULL)
1109 		return EINVAL;
1110 	if (inp->inp_faddr.s_addr == INADDR_ANY)
1111 		return ENOTCONN;
1112 
1113 	crit_enter();
1114 	in_pcbdisconnect(inp);
1115 	crit_exit();
1116 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1117 
1118 	ro = &inp->inp_route;
1119 	if (ro->ro_rt != NULL)
1120 		RTFREE(ro->ro_rt);
1121 	bzero(ro, sizeof(*ro));
1122 
1123 	return 0;
1124 }
1125 
1126 static int
1127 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1128 	    struct mbuf *control, struct thread *td)
1129 {
1130 	struct inpcb *inp;
1131 
1132 	inp = so->so_pcb;
1133 	if (inp == NULL) {
1134 		m_freem(m);
1135 		return EINVAL;
1136 	}
1137 	return udp_output(inp, m, addr, control, td);
1138 }
1139 
1140 int
1141 udp_shutdown(struct socket *so)
1142 {
1143 	struct inpcb *inp;
1144 
1145 	inp = so->so_pcb;
1146 	if (inp == NULL)
1147 		return EINVAL;
1148 	socantsendmore(so);
1149 	return 0;
1150 }
1151 
1152 struct pr_usrreqs udp_usrreqs = {
1153 	.pru_abort = udp_abort,
1154 	.pru_accept = pru_accept_notsupp,
1155 	.pru_attach = udp_attach,
1156 	.pru_bind = udp_bind,
1157 	.pru_connect = udp_connect,
1158 	.pru_connect2 = pru_connect2_notsupp,
1159 	.pru_control = in_control,
1160 	.pru_detach = udp_detach,
1161 	.pru_disconnect = udp_disconnect,
1162 	.pru_listen = pru_listen_notsupp,
1163 	.pru_peeraddr = in_setpeeraddr,
1164 	.pru_rcvd = pru_rcvd_notsupp,
1165 	.pru_rcvoob = pru_rcvoob_notsupp,
1166 	.pru_send = udp_send,
1167 	.pru_sense = pru_sense_null,
1168 	.pru_shutdown = udp_shutdown,
1169 	.pru_sockaddr = in_setsockaddr,
1170 	.pru_sosend = sosendudp,
1171 	.pru_soreceive = soreceive,
1172 	.pru_sopoll = sopoll
1173 };
1174 
1175