1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2008 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: head/sys/netinet/udp_usrreq.c 185435 2008-11-29 14:32:14Z bz $");
36
37#include "opt_ipfw.h"
38#include "opt_inet6.h"
39#include "opt_ipsec.h"
40#include "opt_mac.h"
41
42#include <sys/param.h>
43#include <sys/domain.h>
44#include <sys/eventhandler.h>
45#include <sys/jail.h>
46#include <sys/kernel.h>
47#include <sys/lock.h>
48#include <sys/malloc.h>
49#include <sys/mbuf.h>
50#include <sys/priv.h>
51#include <sys/proc.h>
52#include <sys/protosw.h>
53#include <sys/signalvar.h>
54#include <sys/socket.h>
55#include <sys/socketvar.h>
56#include <sys/sx.h>
57#include <sys/sysctl.h>
58#include <sys/syslog.h>
59#include <sys/systm.h>
60#include <sys/vimage.h>
61
62#include <vm/uma.h>
63
64#include <net/if.h>
65#include <net/route.h>
66
67#include <netinet/in.h>
68#include <netinet/in_pcb.h>
69#include <netinet/in_systm.h>
70#include <netinet/in_var.h>
71#include <netinet/ip.h>
72#ifdef INET6
73#include <netinet/ip6.h>
74#endif
75#include <netinet/ip_icmp.h>
76#include <netinet/icmp_var.h>
77#include <netinet/ip_var.h>
78#include <netinet/ip_options.h>
79#ifdef INET6
80#include <netinet6/ip6_var.h>
81#endif
82#include <netinet/udp.h>
83#include <netinet/udp_var.h>
84
85#ifdef IPSEC
86#include <netipsec/ipsec.h>
87#endif
88
89#include <machine/in_cksum.h>
90
91#include <security/mac/mac_framework.h>
92
93/*
94 * UDP protocol implementation.
95 * Per RFC 768, August, 1980.
96 */
97
98#ifdef VIMAGE_GLOBALS
99int udp_blackhole;
100#endif
101
102/*
103 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
104 * removes the only data integrity mechanism for packets and malformed
105 * packets that would otherwise be discarded due to bad checksums, and may
106 * cause problems (especially for NFS data blocks).
107 */
108static int udp_cksum = 1;
109SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
110 0, "compute udp checksum");
111
112int udp_log_in_vain = 0;
113SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114 &udp_log_in_vain, 0, "Log all incoming UDP packets");
115
116SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_udp, OID_AUTO, blackhole,
117 CTLFLAG_RW, udp_blackhole, 0,
118 "Do not send port unreachables for refused connects");
119
120u_long udp_sendspace = 9216; /* really max datagram size */
121 /* 40 1K datagrams */
122SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
123 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
124
125u_long udp_recvspace = 40 * (1024 +
126#ifdef INET6
127 sizeof(struct sockaddr_in6)
128#else
129 sizeof(struct sockaddr_in)
130#endif
131 );
132
133SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
134 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
135
136#ifdef VIMAGE_GLOBALS
137struct inpcbhead udb; /* from udp_var.h */
138struct inpcbinfo udbinfo;
139struct udpstat udpstat; /* from udp_var.h */
140#endif
141
142#ifndef UDBHASHSIZE
143#define UDBHASHSIZE 128
144#endif
145
146SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_udp, UDPCTL_STATS, stats,
147 CTLFLAG_RW, udpstat, udpstat,
148 "UDP statistics (struct udpstat, netinet/udp_var.h)");
149
150static void udp_detach(struct socket *so);
151static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
152 struct mbuf *, struct thread *);
153
154static void
155udp_zone_change(void *tag)
156{
157
158 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
159}
160
161static int
162udp_inpcb_init(void *mem, int size, int flags)
163{
164 struct inpcb *inp;
165
166 inp = mem;
167 INP_LOCK_INIT(inp, "inp", "udpinp");
168 return (0);
169}
170
171void
172udp_init(void)
173{
174 INIT_VNET_INET(curvnet);
175
176 V_udp_blackhole = 0;
177
178 INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
179 LIST_INIT(&V_udb);
180 V_udbinfo.ipi_listhead = &V_udb;
181 V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
182 &V_udbinfo.ipi_hashmask);
183 V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
184 &V_udbinfo.ipi_porthashmask);
185 V_udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
186 NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
187 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
188 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
189 EVENTHANDLER_PRI_ANY);
190}
191
192/*
193 * Subroutine of udp_input(), which appends the provided mbuf chain to the
194 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
195 * contains the source address. If the socket ends up being an IPv6 socket,
196 * udp_append() will convert to a sockaddr_in6 before passing the address
197 * into the socket code.
198 */
199static void
200udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
201 struct sockaddr_in *udp_in)
202{
203 struct sockaddr *append_sa;
204 struct socket *so;
205 struct mbuf *opts = 0;
206#ifdef INET6
207 struct sockaddr_in6 udp_in6;
208#endif
209
210 INP_RLOCK_ASSERT(inp);
211
212#ifdef IPSEC
213 /* Check AH/ESP integrity. */
214 if (ipsec4_in_reject(n, inp)) {
215 INIT_VNET_IPSEC(curvnet);
216 m_freem(n);
217 V_ipsec4stat.in_polvio++;
218 return;
219 }
220#endif /* IPSEC */
221#ifdef MAC
222 if (mac_inpcb_check_deliver(inp, n) != 0) {
223 m_freem(n);
224 return;
225 }
226#endif
227 if (inp->inp_flags & INP_CONTROLOPTS ||
228 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
229#ifdef INET6
230 if (inp->inp_vflag & INP_IPV6)
231 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
232 else
233#endif
234 ip_savecontrol(inp, &opts, ip, n);
235 }
236#ifdef INET6
237 if (inp->inp_vflag & INP_IPV6) {
238 bzero(&udp_in6, sizeof(udp_in6));
239 udp_in6.sin6_len = sizeof(udp_in6);
240 udp_in6.sin6_family = AF_INET6;
241 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
242 append_sa = (struct sockaddr *)&udp_in6;
243 } else
244#endif
245 append_sa = (struct sockaddr *)udp_in;
246 m_adj(n, off);
247
248 so = inp->inp_socket;
249 SOCKBUF_LOCK(&so->so_rcv);
250 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
251 INIT_VNET_INET(so->so_vnet);
252 SOCKBUF_UNLOCK(&so->so_rcv);
253 m_freem(n);
254 if (opts)
255 m_freem(opts);
256 V_udpstat.udps_fullsock++;
257 } else
258 sorwakeup_locked(so);
259}
260
261void
262udp_input(struct mbuf *m, int off)
263{
264 INIT_VNET_INET(curvnet);
265 int iphlen = off;
266 struct ip *ip;
267 struct udphdr *uh;
268 struct ifnet *ifp;
269 struct inpcb *inp;
270 int len;
271 struct ip save_ip;
272 struct sockaddr_in udp_in;
273#ifdef IPFIREWALL_FORWARD
274 struct m_tag *fwd_tag;
275#endif
276
277 ifp = m->m_pkthdr.rcvif;
278 V_udpstat.udps_ipackets++;
279
280 /*
281 * Strip IP options, if any; should skip this, make available to
282 * user, and use on returned packets, but we don't yet have a way to
283 * check the checksum with options still present.
284 */
285 if (iphlen > sizeof (struct ip)) {
286 ip_stripoptions(m, (struct mbuf *)0);
287 iphlen = sizeof(struct ip);
288 }
289
290 /*
291 * Get IP and UDP header together in first mbuf.
292 */
293 ip = mtod(m, struct ip *);
294 if (m->m_len < iphlen + sizeof(struct udphdr)) {
295 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
296 V_udpstat.udps_hdrops++;
297 return;
298 }
299 ip = mtod(m, struct ip *);
300 }
301 uh = (struct udphdr *)((caddr_t)ip + iphlen);
302
303 /*
304 * Destination port of 0 is illegal, based on RFC768.
305 */
306 if (uh->uh_dport == 0)
307 goto badunlocked;
308
309 /*
310 * Construct sockaddr format source address. Stuff source address
311 * and datagram in user buffer.
312 */
313 bzero(&udp_in, sizeof(udp_in));
314 udp_in.sin_len = sizeof(udp_in);
315 udp_in.sin_family = AF_INET;
316 udp_in.sin_port = uh->uh_sport;
317 udp_in.sin_addr = ip->ip_src;
318
319 /*
320 * Make mbuf data length reflect UDP length. If not enough data to
321 * reflect UDP length, drop.
322 */
323 len = ntohs((u_short)uh->uh_ulen);
324 if (ip->ip_len != len) {
325 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
326 V_udpstat.udps_badlen++;
327 goto badunlocked;
328 }
329 m_adj(m, len - ip->ip_len);
330 /* ip->ip_len = len; */
331 }
332
333 /*
334 * Save a copy of the IP header in case we want restore it for
335 * sending an ICMP error message in response.
336 */
337 if (!V_udp_blackhole)
338 save_ip = *ip;
339 else
340 memset(&save_ip, 0, sizeof(save_ip));
341
342 /*
343 * Checksum extended UDP header and data.
344 */
345 if (uh->uh_sum) {
346 u_short uh_sum;
347
348 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
349 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
350 uh_sum = m->m_pkthdr.csum_data;
351 else
352 uh_sum = in_pseudo(ip->ip_src.s_addr,
353 ip->ip_dst.s_addr, htonl((u_short)len +
354 m->m_pkthdr.csum_data + IPPROTO_UDP));
355 uh_sum ^= 0xffff;
356 } else {
357 char b[9];
358
359 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
360 bzero(((struct ipovly *)ip)->ih_x1, 9);
361 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
362 uh_sum = in_cksum(m, len + sizeof (struct ip));
363 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
364 }
365 if (uh_sum) {
366 V_udpstat.udps_badsum++;
367 m_freem(m);
368 return;
369 }
370 } else
371 V_udpstat.udps_nosum++;
372
373#ifdef IPFIREWALL_FORWARD
374 /*
375 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
376 */
377 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
378 if (fwd_tag != NULL) {
379 struct sockaddr_in *next_hop;
380
381 /*
382 * Do the hack.
383 */
384 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
385 ip->ip_dst = next_hop->sin_addr;
386 uh->uh_dport = ntohs(next_hop->sin_port);
387
388 /*
389 * Remove the tag from the packet. We don't need it anymore.
390 */
391 m_tag_delete(m, fwd_tag);
392 }
393#endif
394
395 INP_INFO_RLOCK(&V_udbinfo);
396 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
397 in_broadcast(ip->ip_dst, ifp)) {
398 struct inpcb *last;
399 struct ip_moptions *imo;
400
401 last = NULL;
402 LIST_FOREACH(inp, &V_udb, inp_list) {
403 if (inp->inp_lport != uh->uh_dport)
404 continue;
405#ifdef INET6
406 if ((inp->inp_vflag & INP_IPV4) == 0)
407 continue;
408#endif
409 if (inp->inp_laddr.s_addr != INADDR_ANY &&
410 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
411 continue;
412 if (inp->inp_faddr.s_addr != INADDR_ANY &&
413 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
414 continue;
415 /*
416 * XXX: Do not check source port of incoming datagram
417 * unless inp_connect() has been called to bind the
418 * fport part of the 4-tuple; the source could be
419 * trying to talk to us with an ephemeral port.
420 */
421 if (inp->inp_fport != 0 &&
422 inp->inp_fport != uh->uh_sport)
423 continue;
424
425 INP_RLOCK(inp);
426
427 /*
428 * Handle socket delivery policy for any-source
429 * and source-specific multicast. [RFC3678]
430 */
431 imo = inp->inp_moptions;
432 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
433 imo != NULL) {
434 struct sockaddr_in sin;
435 struct in_msource *ims;
436 int blocked, mode;
437 size_t idx;
438
439 bzero(&sin, sizeof(struct sockaddr_in));
440 sin.sin_len = sizeof(struct sockaddr_in);
441 sin.sin_family = AF_INET;
442 sin.sin_addr = ip->ip_dst;
443
444 blocked = 0;
445 idx = imo_match_group(imo, ifp,
446 (struct sockaddr *)&sin);
447 if (idx == -1) {
448 /*
449 * No group membership for this socket.
450 * Do not bump udps_noportbcast, as
451 * this will happen further down.
452 */
453 blocked++;
454 } else {
455 /*
456 * Check for a multicast source filter
457 * entry on this socket for this group.
458 * MCAST_EXCLUDE is the default
459 * behaviour. It means default accept;
460 * entries, if present, denote sources
461 * to be excluded from delivery.
462 */
463 ims = imo_match_source(imo, idx,
464 (struct sockaddr *)&udp_in);
465 mode = imo->imo_mfilters[idx].imf_fmode;
466 if ((ims != NULL &&
467 mode == MCAST_EXCLUDE) ||
468 (ims == NULL &&
469 mode == MCAST_INCLUDE)) {
470#ifdef DIAGNOSTIC
471 if (bootverbose) {
472 printf("%s: blocked by"
473 " source filter\n",
474 __func__);
475 }
476#endif
477 V_udpstat.udps_filtermcast++;
478 blocked++;
479 }
480 }
481 if (blocked != 0) {
482 INP_RUNLOCK(inp);
483 continue;
484 }
485 }
486 if (last != NULL) {
487 struct mbuf *n;
488
489 n = m_copy(m, 0, M_COPYALL);
490 if (n != NULL)
491 udp_append(last, ip, n, iphlen +
492 sizeof(struct udphdr), &udp_in);
493 INP_RUNLOCK(last);
494 }
495 last = inp;
496 /*
497 * Don't look for additional matches if this one does
498 * not have either the SO_REUSEPORT or SO_REUSEADDR
499 * socket options set. This heuristic avoids
500 * searching through all pcbs in the common case of a
501 * non-shared port. It assumes that an application
502 * will never clear these options after setting them.
503 */
504 if ((last->inp_socket->so_options &
505 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
506 break;
507 }
508
509 if (last == NULL) {
510 /*
511 * No matching pcb found; discard datagram. (No need
512 * to send an ICMP Port Unreachable for a broadcast
513 * or multicast datgram.)
514 */
515 V_udpstat.udps_noportbcast++;
516 goto badheadlocked;
517 }
518 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
519 &udp_in);
520 INP_RUNLOCK(last);
521 INP_INFO_RUNLOCK(&V_udbinfo);
522 return;
523 }
524
525 /*
526 * Locate pcb for datagram.
527 */
528 inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
529 ip->ip_dst, uh->uh_dport, 1, ifp);
530 if (inp == NULL) {
531 if (udp_log_in_vain) {
532 char buf[4*sizeof "123"];
533
534 strcpy(buf, inet_ntoa(ip->ip_dst));
535 log(LOG_INFO,
536 "Connection attempt to UDP %s:%d from %s:%d\n",
537 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
538 ntohs(uh->uh_sport));
539 }
540 V_udpstat.udps_noport++;
541 if (m->m_flags & (M_BCAST | M_MCAST)) {
542 V_udpstat.udps_noportbcast++;
543 goto badheadlocked;
544 }
545 if (V_udp_blackhole)
546 goto badheadlocked;
547 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
548 goto badheadlocked;
549 *ip = save_ip;
550 ip->ip_len += iphlen;
551 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
552 INP_INFO_RUNLOCK(&V_udbinfo);
553 return;
554 }
555
556 /*
557 * Check the minimum TTL for socket.
558 */
559 INP_RLOCK(inp);
560 INP_INFO_RUNLOCK(&V_udbinfo);
561 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
562 INP_RUNLOCK(inp);
563 goto badunlocked;
564 }
565 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
566 INP_RUNLOCK(inp);
567 return;
568
569badheadlocked:
570 if (inp)
571 INP_RUNLOCK(inp);
572 INP_INFO_RUNLOCK(&V_udbinfo);
573badunlocked:
574 m_freem(m);
575}
576
577/*
578 * Notify a udp user of an asynchronous error; just wake up so that they can
579 * collect error status.
580 */
581struct inpcb *
582udp_notify(struct inpcb *inp, int errno)
583{
584
585 /*
586 * While udp_ctlinput() always calls udp_notify() with a read lock
587 * when invoking it directly, in_pcbnotifyall() currently uses write
588 * locks due to sharing code with TCP. For now, accept either a read
589 * or a write lock, but a read lock is sufficient.
590 */
591 INP_LOCK_ASSERT(inp);
592
593 inp->inp_socket->so_error = errno;
594 sorwakeup(inp->inp_socket);
595 sowwakeup(inp->inp_socket);
596 return (inp);
597}
598
599void
600udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
601{
602 INIT_VNET_INET(curvnet);
603 struct ip *ip = vip;
604 struct udphdr *uh;
605 struct in_addr faddr;
606 struct inpcb *inp;
607
608 faddr = ((struct sockaddr_in *)sa)->sin_addr;
609 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
610 return;
611
612 /*
613 * Redirects don't need to be handled up here.
614 */
615 if (PRC_IS_REDIRECT(cmd))
616 return;
617
618 /*
619 * Hostdead is ugly because it goes linearly through all PCBs.
620 *
621 * XXX: We never get this from ICMP, otherwise it makes an excellent
622 * DoS attack on machines with many connections.
623 */
624 if (cmd == PRC_HOSTDEAD)
625 ip = NULL;
626 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
627 return;
628 if (ip != NULL) {
629 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
630 INP_INFO_RLOCK(&V_udbinfo);
631 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
632 ip->ip_src, uh->uh_sport, 0, NULL);
633 if (inp != NULL) {
634 INP_RLOCK(inp);
635 if (inp->inp_socket != NULL) {
636 udp_notify(inp, inetctlerrmap[cmd]);
637 }
638 INP_RUNLOCK(inp);
639 }
640 INP_INFO_RUNLOCK(&V_udbinfo);
641 } else
642 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
643 udp_notify);
644}
645
646static int
647udp_pcblist(SYSCTL_HANDLER_ARGS)
648{
649 INIT_VNET_INET(curvnet);
650 int error, i, n;
651 struct inpcb *inp, **inp_list;
652 inp_gen_t gencnt;
653 struct xinpgen xig;
654
655 /*
656 * The process of preparing the PCB list is too time-consuming and
657 * resource-intensive to repeat twice on every request.
658 */
659 if (req->oldptr == 0) {
660 n = V_udbinfo.ipi_count;
661 req->oldidx = 2 * (sizeof xig)
662 + (n + n/8) * sizeof(struct xinpcb);
663 return (0);
664 }
665
666 if (req->newptr != 0)
667 return (EPERM);
668
669 /*
670 * OK, now we're committed to doing something.
671 */
672 INP_INFO_RLOCK(&V_udbinfo);
673 gencnt = V_udbinfo.ipi_gencnt;
674 n = V_udbinfo.ipi_count;
675 INP_INFO_RUNLOCK(&V_udbinfo);
676
677 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
678 + n * sizeof(struct xinpcb));
679 if (error != 0)
680 return (error);
681
682 xig.xig_len = sizeof xig;
683 xig.xig_count = n;
684 xig.xig_gen = gencnt;
685 xig.xig_sogen = so_gencnt;
686 error = SYSCTL_OUT(req, &xig, sizeof xig);
687 if (error)
688 return (error);
689
690 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
691 if (inp_list == 0)
692 return (ENOMEM);
693
694 INP_INFO_RLOCK(&V_udbinfo);
695 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
696 inp = LIST_NEXT(inp, inp_list)) {
697 INP_RLOCK(inp);
698 if (inp->inp_gencnt <= gencnt &&
699 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
700 inp_list[i++] = inp;
701 INP_RUNLOCK(inp);
702 }
703 INP_INFO_RUNLOCK(&V_udbinfo);
704 n = i;
705
706 error = 0;
707 for (i = 0; i < n; i++) {
708 inp = inp_list[i];
709 INP_RLOCK(inp);
710 if (inp->inp_gencnt <= gencnt) {
711 struct xinpcb xi;
712 bzero(&xi, sizeof(xi));
713 xi.xi_len = sizeof xi;
714 /* XXX should avoid extra copy */
715 bcopy(inp, &xi.xi_inp, sizeof *inp);
716 if (inp->inp_socket)
717 sotoxsocket(inp->inp_socket, &xi.xi_socket);
718 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
719 INP_RUNLOCK(inp);
720 error = SYSCTL_OUT(req, &xi, sizeof xi);
721 } else
722 INP_RUNLOCK(inp);
723 }
724 if (!error) {
725 /*
726 * Give the user an updated idea of our state. If the
727 * generation differs from what we told her before, she knows
728 * that something happened while we were processing this
729 * request, and it might be necessary to retry.
730 */
731 INP_INFO_RLOCK(&V_udbinfo);
732 xig.xig_gen = V_udbinfo.ipi_gencnt;
733 xig.xig_sogen = so_gencnt;
734 xig.xig_count = V_udbinfo.ipi_count;
735 INP_INFO_RUNLOCK(&V_udbinfo);
736 error = SYSCTL_OUT(req, &xig, sizeof xig);
737 }
738 free(inp_list, M_TEMP);
739 return (error);
740}
741
742SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
743 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
744
745static int
746udp_getcred(SYSCTL_HANDLER_ARGS)
747{
748 INIT_VNET_INET(curvnet);
749 struct xucred xuc;
750 struct sockaddr_in addrs[2];
751 struct inpcb *inp;
752 int error;
753
754 error = priv_check(req->td, PRIV_NETINET_GETCRED);
755 if (error)
756 return (error);
757 error = SYSCTL_IN(req, addrs, sizeof(addrs));
758 if (error)
759 return (error);
760 INP_INFO_RLOCK(&V_udbinfo);
761 inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
762 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
763 if (inp != NULL) {
764 INP_RLOCK(inp);
765 INP_INFO_RUNLOCK(&V_udbinfo);
766 if (inp->inp_socket == NULL)
767 error = ENOENT;
768 if (error == 0)
769 error = cr_canseeinpcb(req->td->td_ucred, inp);
770 if (error == 0)
771 cru2x(inp->inp_cred, &xuc);
772 INP_RUNLOCK(inp);
773 } else {
774 INP_INFO_RUNLOCK(&V_udbinfo);
775 error = ENOENT;
776 }
777 if (error == 0)
778 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
779 return (error);
780}
781
782SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
783 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
784 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
785
786static int
787udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
788 struct mbuf *control, struct thread *td)
789{
790 INIT_VNET_INET(inp->inp_vnet);
791 struct udpiphdr *ui;
792 int len = m->m_pkthdr.len;
793 struct in_addr faddr, laddr;
794 struct cmsghdr *cm;
795 struct sockaddr_in *sin, src;
796 int error = 0;
797 int ipflags;
798 u_short fport, lport;
799 int unlock_udbinfo;
800
801 /*
802 * udp_output() may need to temporarily bind or connect the current
803 * inpcb. As such, we don't know up front whether we will need the
804 * pcbinfo lock or not. Do any work to decide what is needed up
805 * front before acquiring any locks.
806 */
807 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
808 if (control)
809 m_freem(control);
810 m_freem(m);
811 return (EMSGSIZE);
812 }
813
814 src.sin_family = 0;
815 if (control != NULL) {
816 /*
817 * XXX: Currently, we assume all the optional information is
818 * stored in a single mbuf.
819 */
820 if (control->m_next) {
821 m_freem(control);
822 m_freem(m);
823 return (EINVAL);
824 }
825 for (; control->m_len > 0;
826 control->m_data += CMSG_ALIGN(cm->cmsg_len),
827 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
828 cm = mtod(control, struct cmsghdr *);
829 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
830 || cm->cmsg_len > control->m_len) {
831 error = EINVAL;
832 break;
833 }
834 if (cm->cmsg_level != IPPROTO_IP)
835 continue;
836
837 switch (cm->cmsg_type) {
838 case IP_SENDSRCADDR:
839 if (cm->cmsg_len !=
840 CMSG_LEN(sizeof(struct in_addr))) {
841 error = EINVAL;
842 break;
843 }
844 bzero(&src, sizeof(src));
845 src.sin_family = AF_INET;
846 src.sin_len = sizeof(src);
847 src.sin_port = inp->inp_lport;
848 src.sin_addr =
849 *(struct in_addr *)CMSG_DATA(cm);
850 break;
851
852 default:
853 error = ENOPROTOOPT;
854 break;
855 }
856 if (error)
857 break;
858 }
859 m_freem(control);
860 }
861 if (error) {
862 m_freem(m);
863 return (error);
864 }
865
866 /*
867 * Depending on whether or not the application has bound or connected
868 * the socket, we may have to do varying levels of work. The optimal
869 * case is for a connected UDP socket, as a global lock isn't
870 * required at all.
871 *
872 * In order to decide which we need, we require stability of the
873 * inpcb binding, which we ensure by acquiring a read lock on the
874 * inpcb. This doesn't strictly follow the lock order, so we play
875 * the trylock and retry game; note that we may end up with more
876 * conservative locks than required the second time around, so later
877 * assertions have to accept that. Further analysis of the number of
878 * misses under contention is required.
879 */
880 sin = (struct sockaddr_in *)addr;
881 INP_RLOCK(inp);
882 if (sin != NULL &&
883 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
884 INP_RUNLOCK(inp);
885 INP_INFO_WLOCK(&V_udbinfo);
886 INP_WLOCK(inp);
887 unlock_udbinfo = 2;
888 } else if ((sin != NULL && (
889 (sin->sin_addr.s_addr == INADDR_ANY) ||
890 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
891 (inp->inp_laddr.s_addr == INADDR_ANY) ||
892 (inp->inp_lport == 0))) ||
893 (src.sin_family == AF_INET)) {
894 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
895 INP_RUNLOCK(inp);
896 INP_INFO_RLOCK(&V_udbinfo);
897 INP_RLOCK(inp);
898 }
899 unlock_udbinfo = 1;
900 } else
901 unlock_udbinfo = 0;
902
903 /*
904 * If the IP_SENDSRCADDR control message was specified, override the
905 * source address for this datagram. Its use is invalidated if the
906 * address thus specified is incomplete or clobbers other inpcbs.
907 */
908 laddr = inp->inp_laddr;
909 lport = inp->inp_lport;
910 if (src.sin_family == AF_INET) {
911 INP_INFO_LOCK_ASSERT(&V_udbinfo);
912 if ((lport == 0) ||
913 (laddr.s_addr == INADDR_ANY &&
914 src.sin_addr.s_addr == INADDR_ANY)) {
915 error = EINVAL;
916 goto release;
917 }
918 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
919 &laddr.s_addr, &lport, td->td_ucred);
920 if (error)
921 goto release;
922 }
923
924 /*
925 * If a UDP socket has been connected, then a local address/port will
926 * have been selected and bound.
927 *
928 * If a UDP socket has not been connected to, then an explicit
929 * destination address must be used, in which case a local
930 * address/port may not have been selected and bound.
931 */
932 if (sin != NULL) {
933 INP_LOCK_ASSERT(inp);
934 if (inp->inp_faddr.s_addr != INADDR_ANY) {
935 error = EISCONN;
936 goto release;
937 }
938
939 /*
940 * Jail may rewrite the destination address, so let it do
941 * that before we use it.
942 */
943 if (prison_remote_ip4(td->td_ucred, &sin->sin_addr) != 0) {
944 error = EINVAL;
945 goto release;
946 }
947
948 /*
949 * If a local address or port hasn't yet been selected, or if
950 * the destination address needs to be rewritten due to using
951 * a special INADDR_ constant, invoke in_pcbconnect_setup()
952 * to do the heavy lifting. Once a port is selected, we
953 * commit the binding back to the socket; we also commit the
954 * binding of the address if in jail.
955 *
956 * If we already have a valid binding and we're not
957 * requesting a destination address rewrite, use a fast path.
958 */
959 if (inp->inp_laddr.s_addr == INADDR_ANY ||
960 inp->inp_lport == 0 ||
961 sin->sin_addr.s_addr == INADDR_ANY ||
962 sin->sin_addr.s_addr == INADDR_BROADCAST) {
963 INP_INFO_LOCK_ASSERT(&V_udbinfo);
964 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
965 &lport, &faddr.s_addr, &fport, NULL,
966 td->td_ucred);
967 if (error)
968 goto release;
969
970 /*
971 * XXXRW: Why not commit the port if the address is
972 * !INADDR_ANY?
973 */
974 /* Commit the local port if newly assigned. */
975 if (inp->inp_laddr.s_addr == INADDR_ANY &&
976 inp->inp_lport == 0) {
977 INP_INFO_WLOCK_ASSERT(&V_udbinfo);
978 INP_WLOCK_ASSERT(inp);
979 /*
980 * Remember addr if jailed, to prevent
981 * rebinding.
982 */
983 if (jailed(td->td_ucred))
984 inp->inp_laddr = laddr;
985 inp->inp_lport = lport;
986 if (in_pcbinshash(inp) != 0) {
987 inp->inp_lport = 0;
988 error = EAGAIN;
989 goto release;
990 }
991 inp->inp_flags |= INP_ANONPORT;
992 }
993 } else {
994 faddr = sin->sin_addr;
995 fport = sin->sin_port;
996 }
997 } else {
998 INP_LOCK_ASSERT(inp);
999 faddr = inp->inp_faddr;
1000 fport = inp->inp_fport;
1001 if (faddr.s_addr == INADDR_ANY) {
1002 error = ENOTCONN;
1003 goto release;
1004 }
1005 }
1006
1007 /*
1008 * Calculate data length and get a mbuf for UDP, IP, and possible
1009 * link-layer headers. Immediate slide the data pointer back forward
1010 * since we won't use that space at this layer.
1011 */
1012 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1013 if (m == NULL) {
1014 error = ENOBUFS;
1015 goto release;
1016 }
1017 m->m_data += max_linkhdr;
1018 m->m_len -= max_linkhdr;
1019 m->m_pkthdr.len -= max_linkhdr;
1020
1021 /*
1022 * Fill in mbuf with extended UDP header and addresses and length put
1023 * into network format.
1024 */
1025 ui = mtod(m, struct udpiphdr *);
1026 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1027 ui->ui_pr = IPPROTO_UDP;
1028 ui->ui_src = laddr;
1029 ui->ui_dst = faddr;
1030 ui->ui_sport = lport;
1031 ui->ui_dport = fport;
1032 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1033
1034 /*
1035 * Set the Don't Fragment bit in the IP header.
1036 */
1037 if (inp->inp_flags & INP_DONTFRAG) {
1038 struct ip *ip;
1039
1040 ip = (struct ip *)&ui->ui_i;
1041 ip->ip_off |= IP_DF;
1042 }
1043
1044 ipflags = 0;
1045 if (inp->inp_socket->so_options & SO_DONTROUTE)
1046 ipflags |= IP_ROUTETOIF;
1047 if (inp->inp_socket->so_options & SO_BROADCAST)
1048 ipflags |= IP_ALLOWBROADCAST;
1049 if (inp->inp_flags & INP_ONESBCAST)
1050 ipflags |= IP_SENDONES;
1051
1052#ifdef MAC
1053 mac_inpcb_create_mbuf(inp, m);
1054#endif
1055
1056 /*
1057 * Set up checksum and output datagram.
1058 */
1059 if (udp_cksum) {
1060 if (inp->inp_flags & INP_ONESBCAST)
1061 faddr.s_addr = INADDR_BROADCAST;
1062 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1063 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1064 m->m_pkthdr.csum_flags = CSUM_UDP;
1065 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1066 } else
1067 ui->ui_sum = 0;
1068 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1069 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1070 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */
1071 V_udpstat.udps_opackets++;
1072
1073 if (unlock_udbinfo == 2)
1074 INP_INFO_WUNLOCK(&V_udbinfo);
1075 else if (unlock_udbinfo == 1)
1076 INP_INFO_RUNLOCK(&V_udbinfo);
1077 error = ip_output(m, inp->inp_options, NULL, ipflags,
1078 inp->inp_moptions, inp);
1079 if (unlock_udbinfo == 2)
1080 INP_WUNLOCK(inp);
1081 else
1082 INP_RUNLOCK(inp);
1083 return (error);
1084
1085release:
1086 if (unlock_udbinfo == 2) {
1087 INP_WUNLOCK(inp);
1088 INP_INFO_WUNLOCK(&V_udbinfo);
1089 } else if (unlock_udbinfo == 1) {
1090 INP_RUNLOCK(inp);
1091 INP_INFO_RUNLOCK(&V_udbinfo);
1092 } else
1093 INP_RUNLOCK(inp);
1094 m_freem(m);
1095 return (error);
1096}
1097
1098static void
1099udp_abort(struct socket *so)
1100{
1101 INIT_VNET_INET(so->so_vnet);
1102 struct inpcb *inp;
1103
1104 inp = sotoinpcb(so);
1105 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1106 INP_INFO_WLOCK(&V_udbinfo);
1107 INP_WLOCK(inp);
1108 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1109 in_pcbdisconnect(inp);
1110 inp->inp_laddr.s_addr = INADDR_ANY;
1111 soisdisconnected(so);
1112 }
1113 INP_WUNLOCK(inp);
1114 INP_INFO_WUNLOCK(&V_udbinfo);
1115}
1116
1117static int
1118udp_attach(struct socket *so, int proto, struct thread *td)
1119{
1120 INIT_VNET_INET(so->so_vnet);
1121 struct inpcb *inp;
1122 int error;
1123
1124 inp = sotoinpcb(so);
1125 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1126 error = soreserve(so, udp_sendspace, udp_recvspace);
1127 if (error)
1128 return (error);
1129 INP_INFO_WLOCK(&V_udbinfo);
1130 error = in_pcballoc(so, &V_udbinfo);
1131 if (error) {
1132 INP_INFO_WUNLOCK(&V_udbinfo);
1133 return (error);
1134 }
1135
1136 inp = (struct inpcb *)so->so_pcb;
1137 INP_INFO_WUNLOCK(&V_udbinfo);
1138 inp->inp_vflag |= INP_IPV4;
1139 inp->inp_ip_ttl = V_ip_defttl;
1140 INP_WUNLOCK(inp);
1141 return (0);
1142}
1143
1144static int
1145udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1146{
1147 INIT_VNET_INET(so->so_vnet);
1148 struct inpcb *inp;
1149 int error;
1150
1151 inp = sotoinpcb(so);
1152 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1153 INP_INFO_WLOCK(&V_udbinfo);
1154 INP_WLOCK(inp);
1155 error = in_pcbbind(inp, nam, td->td_ucred);
1156 INP_WUNLOCK(inp);
1157 INP_INFO_WUNLOCK(&V_udbinfo);
1158 return (error);
1159}
1160
1161static void
1162udp_close(struct socket *so)
1163{
1164 INIT_VNET_INET(so->so_vnet);
1165 struct inpcb *inp;
1166
1167 inp = sotoinpcb(so);
1168 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1169 INP_INFO_WLOCK(&V_udbinfo);
1170 INP_WLOCK(inp);
1171 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1172 in_pcbdisconnect(inp);
1173 inp->inp_laddr.s_addr = INADDR_ANY;
1174 soisdisconnected(so);
1175 }
1176 INP_WUNLOCK(inp);
1177 INP_INFO_WUNLOCK(&V_udbinfo);
1178}
1179
1180static int
1181udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1182{
1183 INIT_VNET_INET(so->so_vnet);
1184 struct inpcb *inp;
1185 int error;
1186 struct sockaddr_in *sin;
1187
1188 inp = sotoinpcb(so);
1189 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1190 INP_INFO_WLOCK(&V_udbinfo);
1191 INP_WLOCK(inp);
1192 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1193 INP_WUNLOCK(inp);
1194 INP_INFO_WUNLOCK(&V_udbinfo);
1195 return (EISCONN);
1196 }
1197 sin = (struct sockaddr_in *)nam;
1198 if (prison_remote_ip4(td->td_ucred, &sin->sin_addr) != 0) {
1199 INP_WUNLOCK(inp);
1200 INP_INFO_WUNLOCK(&udbinfo);
1201 return (EAFNOSUPPORT);
1202 }
1203 error = in_pcbconnect(inp, nam, td->td_ucred);
1204 if (error == 0)
1205 soisconnected(so);
1206 INP_WUNLOCK(inp);
1207 INP_INFO_WUNLOCK(&V_udbinfo);
1208 return (error);
1209}
1210
1211static void
1212udp_detach(struct socket *so)
1213{
1214 INIT_VNET_INET(so->so_vnet);
1215 struct inpcb *inp;
1216
1217 inp = sotoinpcb(so);
1218 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1219 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1220 ("udp_detach: not disconnected"));
1221 INP_INFO_WLOCK(&V_udbinfo);
1222 INP_WLOCK(inp);
1223 in_pcbdetach(inp);
1224 in_pcbfree(inp);
1225 INP_INFO_WUNLOCK(&V_udbinfo);
1226}
1227
1228static int
1229udp_disconnect(struct socket *so)
1230{
1231 INIT_VNET_INET(so->so_vnet);
1232 struct inpcb *inp;
1233
1234 inp = sotoinpcb(so);
1235 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1236 INP_INFO_WLOCK(&V_udbinfo);
1237 INP_WLOCK(inp);
1238 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1239 INP_WUNLOCK(inp);
1240 INP_INFO_WUNLOCK(&V_udbinfo);
1241 return (ENOTCONN);
1242 }
1243
1244 in_pcbdisconnect(inp);
1245 inp->inp_laddr.s_addr = INADDR_ANY;
1246 SOCK_LOCK(so);
1247 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1248 SOCK_UNLOCK(so);
1249 INP_WUNLOCK(inp);
1250 INP_INFO_WUNLOCK(&V_udbinfo);
1251 return (0);
1252}
1253
1254static int
1255udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1256 struct mbuf *control, struct thread *td)
1257{
1258 struct inpcb *inp;
1259
1260 inp = sotoinpcb(so);
1261 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1262 return (udp_output(inp, m, addr, control, td));
1263}
1264
1265int
1266udp_shutdown(struct socket *so)
1267{
1268 struct inpcb *inp;
1269
1270 inp = sotoinpcb(so);
1271 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1272 INP_WLOCK(inp);
1273 socantsendmore(so);
1274 INP_WUNLOCK(inp);
1275 return (0);
1276}
1277
1278struct pr_usrreqs udp_usrreqs = {
1279 .pru_abort = udp_abort,
1280 .pru_attach = udp_attach,
1281 .pru_bind = udp_bind,
1282 .pru_connect = udp_connect,
1283 .pru_control = in_control,
1284 .pru_detach = udp_detach,
1285 .pru_disconnect = udp_disconnect,
1286 .pru_peeraddr = in_getpeeraddr,
1287 .pru_send = udp_send,
1288 .pru_soreceive = soreceive_dgram,
1289 .pru_sosend = sosend_dgram,
1290 .pru_shutdown = udp_shutdown,
1291 .pru_sockaddr = in_getsockaddr,
1292 .pru_sosetlabel = in_pcbsosetlabel,
1293 .pru_close = udp_close,
1294};
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2008 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: head/sys/netinet/udp_usrreq.c 185435 2008-11-29 14:32:14Z bz $");
36
37#include "opt_ipfw.h"
38#include "opt_inet6.h"
39#include "opt_ipsec.h"
40#include "opt_mac.h"
41
42#include <sys/param.h>
43#include <sys/domain.h>
44#include <sys/eventhandler.h>
45#include <sys/jail.h>
46#include <sys/kernel.h>
47#include <sys/lock.h>
48#include <sys/malloc.h>
49#include <sys/mbuf.h>
50#include <sys/priv.h>
51#include <sys/proc.h>
52#include <sys/protosw.h>
53#include <sys/signalvar.h>
54#include <sys/socket.h>
55#include <sys/socketvar.h>
56#include <sys/sx.h>
57#include <sys/sysctl.h>
58#include <sys/syslog.h>
59#include <sys/systm.h>
60#include <sys/vimage.h>
61
62#include <vm/uma.h>
63
64#include <net/if.h>
65#include <net/route.h>
66
67#include <netinet/in.h>
68#include <netinet/in_pcb.h>
69#include <netinet/in_systm.h>
70#include <netinet/in_var.h>
71#include <netinet/ip.h>
72#ifdef INET6
73#include <netinet/ip6.h>
74#endif
75#include <netinet/ip_icmp.h>
76#include <netinet/icmp_var.h>
77#include <netinet/ip_var.h>
78#include <netinet/ip_options.h>
79#ifdef INET6
80#include <netinet6/ip6_var.h>
81#endif
82#include <netinet/udp.h>
83#include <netinet/udp_var.h>
84
85#ifdef IPSEC
86#include <netipsec/ipsec.h>
87#endif
88
89#include <machine/in_cksum.h>
90
91#include <security/mac/mac_framework.h>
92
93/*
94 * UDP protocol implementation.
95 * Per RFC 768, August, 1980.
96 */
97
98#ifdef VIMAGE_GLOBALS
99int udp_blackhole;
100#endif
101
102/*
103 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
104 * removes the only data integrity mechanism for packets and malformed
105 * packets that would otherwise be discarded due to bad checksums, and may
106 * cause problems (especially for NFS data blocks).
107 */
108static int udp_cksum = 1;
109SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
110 0, "compute udp checksum");
111
112int udp_log_in_vain = 0;
113SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114 &udp_log_in_vain, 0, "Log all incoming UDP packets");
115
116SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_udp, OID_AUTO, blackhole,
117 CTLFLAG_RW, udp_blackhole, 0,
118 "Do not send port unreachables for refused connects");
119
120u_long udp_sendspace = 9216; /* really max datagram size */
121 /* 40 1K datagrams */
122SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
123 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
124
125u_long udp_recvspace = 40 * (1024 +
126#ifdef INET6
127 sizeof(struct sockaddr_in6)
128#else
129 sizeof(struct sockaddr_in)
130#endif
131 );
132
133SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
134 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
135
136#ifdef VIMAGE_GLOBALS
137struct inpcbhead udb; /* from udp_var.h */
138struct inpcbinfo udbinfo;
139struct udpstat udpstat; /* from udp_var.h */
140#endif
141
142#ifndef UDBHASHSIZE
143#define UDBHASHSIZE 128
144#endif
145
146SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_udp, UDPCTL_STATS, stats,
147 CTLFLAG_RW, udpstat, udpstat,
148 "UDP statistics (struct udpstat, netinet/udp_var.h)");
149
150static void udp_detach(struct socket *so);
151static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
152 struct mbuf *, struct thread *);
153
154static void
155udp_zone_change(void *tag)
156{
157
158 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
159}
160
161static int
162udp_inpcb_init(void *mem, int size, int flags)
163{
164 struct inpcb *inp;
165
166 inp = mem;
167 INP_LOCK_INIT(inp, "inp", "udpinp");
168 return (0);
169}
170
171void
172udp_init(void)
173{
174 INIT_VNET_INET(curvnet);
175
176 V_udp_blackhole = 0;
177
178 INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
179 LIST_INIT(&V_udb);
180 V_udbinfo.ipi_listhead = &V_udb;
181 V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
182 &V_udbinfo.ipi_hashmask);
183 V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
184 &V_udbinfo.ipi_porthashmask);
185 V_udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
186 NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
187 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
188 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
189 EVENTHANDLER_PRI_ANY);
190}
191
192/*
193 * Subroutine of udp_input(), which appends the provided mbuf chain to the
194 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
195 * contains the source address. If the socket ends up being an IPv6 socket,
196 * udp_append() will convert to a sockaddr_in6 before passing the address
197 * into the socket code.
198 */
199static void
200udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
201 struct sockaddr_in *udp_in)
202{
203 struct sockaddr *append_sa;
204 struct socket *so;
205 struct mbuf *opts = 0;
206#ifdef INET6
207 struct sockaddr_in6 udp_in6;
208#endif
209
210 INP_RLOCK_ASSERT(inp);
211
212#ifdef IPSEC
213 /* Check AH/ESP integrity. */
214 if (ipsec4_in_reject(n, inp)) {
215 INIT_VNET_IPSEC(curvnet);
216 m_freem(n);
217 V_ipsec4stat.in_polvio++;
218 return;
219 }
220#endif /* IPSEC */
221#ifdef MAC
222 if (mac_inpcb_check_deliver(inp, n) != 0) {
223 m_freem(n);
224 return;
225 }
226#endif
227 if (inp->inp_flags & INP_CONTROLOPTS ||
228 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
229#ifdef INET6
230 if (inp->inp_vflag & INP_IPV6)
231 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
232 else
233#endif
234 ip_savecontrol(inp, &opts, ip, n);
235 }
236#ifdef INET6
237 if (inp->inp_vflag & INP_IPV6) {
238 bzero(&udp_in6, sizeof(udp_in6));
239 udp_in6.sin6_len = sizeof(udp_in6);
240 udp_in6.sin6_family = AF_INET6;
241 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
242 append_sa = (struct sockaddr *)&udp_in6;
243 } else
244#endif
245 append_sa = (struct sockaddr *)udp_in;
246 m_adj(n, off);
247
248 so = inp->inp_socket;
249 SOCKBUF_LOCK(&so->so_rcv);
250 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
251 INIT_VNET_INET(so->so_vnet);
252 SOCKBUF_UNLOCK(&so->so_rcv);
253 m_freem(n);
254 if (opts)
255 m_freem(opts);
256 V_udpstat.udps_fullsock++;
257 } else
258 sorwakeup_locked(so);
259}
260
261void
262udp_input(struct mbuf *m, int off)
263{
264 INIT_VNET_INET(curvnet);
265 int iphlen = off;
266 struct ip *ip;
267 struct udphdr *uh;
268 struct ifnet *ifp;
269 struct inpcb *inp;
270 int len;
271 struct ip save_ip;
272 struct sockaddr_in udp_in;
273#ifdef IPFIREWALL_FORWARD
274 struct m_tag *fwd_tag;
275#endif
276
277 ifp = m->m_pkthdr.rcvif;
278 V_udpstat.udps_ipackets++;
279
280 /*
281 * Strip IP options, if any; should skip this, make available to
282 * user, and use on returned packets, but we don't yet have a way to
283 * check the checksum with options still present.
284 */
285 if (iphlen > sizeof (struct ip)) {
286 ip_stripoptions(m, (struct mbuf *)0);
287 iphlen = sizeof(struct ip);
288 }
289
290 /*
291 * Get IP and UDP header together in first mbuf.
292 */
293 ip = mtod(m, struct ip *);
294 if (m->m_len < iphlen + sizeof(struct udphdr)) {
295 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
296 V_udpstat.udps_hdrops++;
297 return;
298 }
299 ip = mtod(m, struct ip *);
300 }
301 uh = (struct udphdr *)((caddr_t)ip + iphlen);
302
303 /*
304 * Destination port of 0 is illegal, based on RFC768.
305 */
306 if (uh->uh_dport == 0)
307 goto badunlocked;
308
309 /*
310 * Construct sockaddr format source address. Stuff source address
311 * and datagram in user buffer.
312 */
313 bzero(&udp_in, sizeof(udp_in));
314 udp_in.sin_len = sizeof(udp_in);
315 udp_in.sin_family = AF_INET;
316 udp_in.sin_port = uh->uh_sport;
317 udp_in.sin_addr = ip->ip_src;
318
319 /*
320 * Make mbuf data length reflect UDP length. If not enough data to
321 * reflect UDP length, drop.
322 */
323 len = ntohs((u_short)uh->uh_ulen);
324 if (ip->ip_len != len) {
325 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
326 V_udpstat.udps_badlen++;
327 goto badunlocked;
328 }
329 m_adj(m, len - ip->ip_len);
330 /* ip->ip_len = len; */
331 }
332
333 /*
334 * Save a copy of the IP header in case we want restore it for
335 * sending an ICMP error message in response.
336 */
337 if (!V_udp_blackhole)
338 save_ip = *ip;
339 else
340 memset(&save_ip, 0, sizeof(save_ip));
341
342 /*
343 * Checksum extended UDP header and data.
344 */
345 if (uh->uh_sum) {
346 u_short uh_sum;
347
348 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
349 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
350 uh_sum = m->m_pkthdr.csum_data;
351 else
352 uh_sum = in_pseudo(ip->ip_src.s_addr,
353 ip->ip_dst.s_addr, htonl((u_short)len +
354 m->m_pkthdr.csum_data + IPPROTO_UDP));
355 uh_sum ^= 0xffff;
356 } else {
357 char b[9];
358
359 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
360 bzero(((struct ipovly *)ip)->ih_x1, 9);
361 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
362 uh_sum = in_cksum(m, len + sizeof (struct ip));
363 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
364 }
365 if (uh_sum) {
366 V_udpstat.udps_badsum++;
367 m_freem(m);
368 return;
369 }
370 } else
371 V_udpstat.udps_nosum++;
372
373#ifdef IPFIREWALL_FORWARD
374 /*
375 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
376 */
377 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
378 if (fwd_tag != NULL) {
379 struct sockaddr_in *next_hop;
380
381 /*
382 * Do the hack.
383 */
384 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
385 ip->ip_dst = next_hop->sin_addr;
386 uh->uh_dport = ntohs(next_hop->sin_port);
387
388 /*
389 * Remove the tag from the packet. We don't need it anymore.
390 */
391 m_tag_delete(m, fwd_tag);
392 }
393#endif
394
395 INP_INFO_RLOCK(&V_udbinfo);
396 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
397 in_broadcast(ip->ip_dst, ifp)) {
398 struct inpcb *last;
399 struct ip_moptions *imo;
400
401 last = NULL;
402 LIST_FOREACH(inp, &V_udb, inp_list) {
403 if (inp->inp_lport != uh->uh_dport)
404 continue;
405#ifdef INET6
406 if ((inp->inp_vflag & INP_IPV4) == 0)
407 continue;
408#endif
409 if (inp->inp_laddr.s_addr != INADDR_ANY &&
410 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
411 continue;
412 if (inp->inp_faddr.s_addr != INADDR_ANY &&
413 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
414 continue;
415 /*
416 * XXX: Do not check source port of incoming datagram
417 * unless inp_connect() has been called to bind the
418 * fport part of the 4-tuple; the source could be
419 * trying to talk to us with an ephemeral port.
420 */
421 if (inp->inp_fport != 0 &&
422 inp->inp_fport != uh->uh_sport)
423 continue;
424
425 INP_RLOCK(inp);
426
427 /*
428 * Handle socket delivery policy for any-source
429 * and source-specific multicast. [RFC3678]
430 */
431 imo = inp->inp_moptions;
432 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
433 imo != NULL) {
434 struct sockaddr_in sin;
435 struct in_msource *ims;
436 int blocked, mode;
437 size_t idx;
438
439 bzero(&sin, sizeof(struct sockaddr_in));
440 sin.sin_len = sizeof(struct sockaddr_in);
441 sin.sin_family = AF_INET;
442 sin.sin_addr = ip->ip_dst;
443
444 blocked = 0;
445 idx = imo_match_group(imo, ifp,
446 (struct sockaddr *)&sin);
447 if (idx == -1) {
448 /*
449 * No group membership for this socket.
450 * Do not bump udps_noportbcast, as
451 * this will happen further down.
452 */
453 blocked++;
454 } else {
455 /*
456 * Check for a multicast source filter
457 * entry on this socket for this group.
458 * MCAST_EXCLUDE is the default
459 * behaviour. It means default accept;
460 * entries, if present, denote sources
461 * to be excluded from delivery.
462 */
463 ims = imo_match_source(imo, idx,
464 (struct sockaddr *)&udp_in);
465 mode = imo->imo_mfilters[idx].imf_fmode;
466 if ((ims != NULL &&
467 mode == MCAST_EXCLUDE) ||
468 (ims == NULL &&
469 mode == MCAST_INCLUDE)) {
470#ifdef DIAGNOSTIC
471 if (bootverbose) {
472 printf("%s: blocked by"
473 " source filter\n",
474 __func__);
475 }
476#endif
477 V_udpstat.udps_filtermcast++;
478 blocked++;
479 }
480 }
481 if (blocked != 0) {
482 INP_RUNLOCK(inp);
483 continue;
484 }
485 }
486 if (last != NULL) {
487 struct mbuf *n;
488
489 n = m_copy(m, 0, M_COPYALL);
490 if (n != NULL)
491 udp_append(last, ip, n, iphlen +
492 sizeof(struct udphdr), &udp_in);
493 INP_RUNLOCK(last);
494 }
495 last = inp;
496 /*
497 * Don't look for additional matches if this one does
498 * not have either the SO_REUSEPORT or SO_REUSEADDR
499 * socket options set. This heuristic avoids
500 * searching through all pcbs in the common case of a
501 * non-shared port. It assumes that an application
502 * will never clear these options after setting them.
503 */
504 if ((last->inp_socket->so_options &
505 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
506 break;
507 }
508
509 if (last == NULL) {
510 /*
511 * No matching pcb found; discard datagram. (No need
512 * to send an ICMP Port Unreachable for a broadcast
513 * or multicast datgram.)
514 */
515 V_udpstat.udps_noportbcast++;
516 goto badheadlocked;
517 }
518 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
519 &udp_in);
520 INP_RUNLOCK(last);
521 INP_INFO_RUNLOCK(&V_udbinfo);
522 return;
523 }
524
525 /*
526 * Locate pcb for datagram.
527 */
528 inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
529 ip->ip_dst, uh->uh_dport, 1, ifp);
530 if (inp == NULL) {
531 if (udp_log_in_vain) {
532 char buf[4*sizeof "123"];
533
534 strcpy(buf, inet_ntoa(ip->ip_dst));
535 log(LOG_INFO,
536 "Connection attempt to UDP %s:%d from %s:%d\n",
537 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
538 ntohs(uh->uh_sport));
539 }
540 V_udpstat.udps_noport++;
541 if (m->m_flags & (M_BCAST | M_MCAST)) {
542 V_udpstat.udps_noportbcast++;
543 goto badheadlocked;
544 }
545 if (V_udp_blackhole)
546 goto badheadlocked;
547 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
548 goto badheadlocked;
549 *ip = save_ip;
550 ip->ip_len += iphlen;
551 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
552 INP_INFO_RUNLOCK(&V_udbinfo);
553 return;
554 }
555
556 /*
557 * Check the minimum TTL for socket.
558 */
559 INP_RLOCK(inp);
560 INP_INFO_RUNLOCK(&V_udbinfo);
561 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
562 INP_RUNLOCK(inp);
563 goto badunlocked;
564 }
565 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
566 INP_RUNLOCK(inp);
567 return;
568
569badheadlocked:
570 if (inp)
571 INP_RUNLOCK(inp);
572 INP_INFO_RUNLOCK(&V_udbinfo);
573badunlocked:
574 m_freem(m);
575}
576
577/*
578 * Notify a udp user of an asynchronous error; just wake up so that they can
579 * collect error status.
580 */
581struct inpcb *
582udp_notify(struct inpcb *inp, int errno)
583{
584
585 /*
586 * While udp_ctlinput() always calls udp_notify() with a read lock
587 * when invoking it directly, in_pcbnotifyall() currently uses write
588 * locks due to sharing code with TCP. For now, accept either a read
589 * or a write lock, but a read lock is sufficient.
590 */
591 INP_LOCK_ASSERT(inp);
592
593 inp->inp_socket->so_error = errno;
594 sorwakeup(inp->inp_socket);
595 sowwakeup(inp->inp_socket);
596 return (inp);
597}
598
599void
600udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
601{
602 INIT_VNET_INET(curvnet);
603 struct ip *ip = vip;
604 struct udphdr *uh;
605 struct in_addr faddr;
606 struct inpcb *inp;
607
608 faddr = ((struct sockaddr_in *)sa)->sin_addr;
609 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
610 return;
611
612 /*
613 * Redirects don't need to be handled up here.
614 */
615 if (PRC_IS_REDIRECT(cmd))
616 return;
617
618 /*
619 * Hostdead is ugly because it goes linearly through all PCBs.
620 *
621 * XXX: We never get this from ICMP, otherwise it makes an excellent
622 * DoS attack on machines with many connections.
623 */
624 if (cmd == PRC_HOSTDEAD)
625 ip = NULL;
626 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
627 return;
628 if (ip != NULL) {
629 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
630 INP_INFO_RLOCK(&V_udbinfo);
631 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
632 ip->ip_src, uh->uh_sport, 0, NULL);
633 if (inp != NULL) {
634 INP_RLOCK(inp);
635 if (inp->inp_socket != NULL) {
636 udp_notify(inp, inetctlerrmap[cmd]);
637 }
638 INP_RUNLOCK(inp);
639 }
640 INP_INFO_RUNLOCK(&V_udbinfo);
641 } else
642 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
643 udp_notify);
644}
645
646static int
647udp_pcblist(SYSCTL_HANDLER_ARGS)
648{
649 INIT_VNET_INET(curvnet);
650 int error, i, n;
651 struct inpcb *inp, **inp_list;
652 inp_gen_t gencnt;
653 struct xinpgen xig;
654
655 /*
656 * The process of preparing the PCB list is too time-consuming and
657 * resource-intensive to repeat twice on every request.
658 */
659 if (req->oldptr == 0) {
660 n = V_udbinfo.ipi_count;
661 req->oldidx = 2 * (sizeof xig)
662 + (n + n/8) * sizeof(struct xinpcb);
663 return (0);
664 }
665
666 if (req->newptr != 0)
667 return (EPERM);
668
669 /*
670 * OK, now we're committed to doing something.
671 */
672 INP_INFO_RLOCK(&V_udbinfo);
673 gencnt = V_udbinfo.ipi_gencnt;
674 n = V_udbinfo.ipi_count;
675 INP_INFO_RUNLOCK(&V_udbinfo);
676
677 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
678 + n * sizeof(struct xinpcb));
679 if (error != 0)
680 return (error);
681
682 xig.xig_len = sizeof xig;
683 xig.xig_count = n;
684 xig.xig_gen = gencnt;
685 xig.xig_sogen = so_gencnt;
686 error = SYSCTL_OUT(req, &xig, sizeof xig);
687 if (error)
688 return (error);
689
690 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
691 if (inp_list == 0)
692 return (ENOMEM);
693
694 INP_INFO_RLOCK(&V_udbinfo);
695 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
696 inp = LIST_NEXT(inp, inp_list)) {
697 INP_RLOCK(inp);
698 if (inp->inp_gencnt <= gencnt &&
699 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
700 inp_list[i++] = inp;
701 INP_RUNLOCK(inp);
702 }
703 INP_INFO_RUNLOCK(&V_udbinfo);
704 n = i;
705
706 error = 0;
707 for (i = 0; i < n; i++) {
708 inp = inp_list[i];
709 INP_RLOCK(inp);
710 if (inp->inp_gencnt <= gencnt) {
711 struct xinpcb xi;
712 bzero(&xi, sizeof(xi));
713 xi.xi_len = sizeof xi;
714 /* XXX should avoid extra copy */
715 bcopy(inp, &xi.xi_inp, sizeof *inp);
716 if (inp->inp_socket)
717 sotoxsocket(inp->inp_socket, &xi.xi_socket);
718 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
719 INP_RUNLOCK(inp);
720 error = SYSCTL_OUT(req, &xi, sizeof xi);
721 } else
722 INP_RUNLOCK(inp);
723 }
724 if (!error) {
725 /*
726 * Give the user an updated idea of our state. If the
727 * generation differs from what we told her before, she knows
728 * that something happened while we were processing this
729 * request, and it might be necessary to retry.
730 */
731 INP_INFO_RLOCK(&V_udbinfo);
732 xig.xig_gen = V_udbinfo.ipi_gencnt;
733 xig.xig_sogen = so_gencnt;
734 xig.xig_count = V_udbinfo.ipi_count;
735 INP_INFO_RUNLOCK(&V_udbinfo);
736 error = SYSCTL_OUT(req, &xig, sizeof xig);
737 }
738 free(inp_list, M_TEMP);
739 return (error);
740}
741
742SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
743 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
744
745static int
746udp_getcred(SYSCTL_HANDLER_ARGS)
747{
748 INIT_VNET_INET(curvnet);
749 struct xucred xuc;
750 struct sockaddr_in addrs[2];
751 struct inpcb *inp;
752 int error;
753
754 error = priv_check(req->td, PRIV_NETINET_GETCRED);
755 if (error)
756 return (error);
757 error = SYSCTL_IN(req, addrs, sizeof(addrs));
758 if (error)
759 return (error);
760 INP_INFO_RLOCK(&V_udbinfo);
761 inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
762 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
763 if (inp != NULL) {
764 INP_RLOCK(inp);
765 INP_INFO_RUNLOCK(&V_udbinfo);
766 if (inp->inp_socket == NULL)
767 error = ENOENT;
768 if (error == 0)
769 error = cr_canseeinpcb(req->td->td_ucred, inp);
770 if (error == 0)
771 cru2x(inp->inp_cred, &xuc);
772 INP_RUNLOCK(inp);
773 } else {
774 INP_INFO_RUNLOCK(&V_udbinfo);
775 error = ENOENT;
776 }
777 if (error == 0)
778 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
779 return (error);
780}
781
782SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
783 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
784 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
785
786static int
787udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
788 struct mbuf *control, struct thread *td)
789{
790 INIT_VNET_INET(inp->inp_vnet);
791 struct udpiphdr *ui;
792 int len = m->m_pkthdr.len;
793 struct in_addr faddr, laddr;
794 struct cmsghdr *cm;
795 struct sockaddr_in *sin, src;
796 int error = 0;
797 int ipflags;
798 u_short fport, lport;
799 int unlock_udbinfo;
800
801 /*
802 * udp_output() may need to temporarily bind or connect the current
803 * inpcb. As such, we don't know up front whether we will need the
804 * pcbinfo lock or not. Do any work to decide what is needed up
805 * front before acquiring any locks.
806 */
807 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
808 if (control)
809 m_freem(control);
810 m_freem(m);
811 return (EMSGSIZE);
812 }
813
814 src.sin_family = 0;
815 if (control != NULL) {
816 /*
817 * XXX: Currently, we assume all the optional information is
818 * stored in a single mbuf.
819 */
820 if (control->m_next) {
821 m_freem(control);
822 m_freem(m);
823 return (EINVAL);
824 }
825 for (; control->m_len > 0;
826 control->m_data += CMSG_ALIGN(cm->cmsg_len),
827 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
828 cm = mtod(control, struct cmsghdr *);
829 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
830 || cm->cmsg_len > control->m_len) {
831 error = EINVAL;
832 break;
833 }
834 if (cm->cmsg_level != IPPROTO_IP)
835 continue;
836
837 switch (cm->cmsg_type) {
838 case IP_SENDSRCADDR:
839 if (cm->cmsg_len !=
840 CMSG_LEN(sizeof(struct in_addr))) {
841 error = EINVAL;
842 break;
843 }
844 bzero(&src, sizeof(src));
845 src.sin_family = AF_INET;
846 src.sin_len = sizeof(src);
847 src.sin_port = inp->inp_lport;
848 src.sin_addr =
849 *(struct in_addr *)CMSG_DATA(cm);
850 break;
851
852 default:
853 error = ENOPROTOOPT;
854 break;
855 }
856 if (error)
857 break;
858 }
859 m_freem(control);
860 }
861 if (error) {
862 m_freem(m);
863 return (error);
864 }
865
866 /*
867 * Depending on whether or not the application has bound or connected
868 * the socket, we may have to do varying levels of work. The optimal
869 * case is for a connected UDP socket, as a global lock isn't
870 * required at all.
871 *
872 * In order to decide which we need, we require stability of the
873 * inpcb binding, which we ensure by acquiring a read lock on the
874 * inpcb. This doesn't strictly follow the lock order, so we play
875 * the trylock and retry game; note that we may end up with more
876 * conservative locks than required the second time around, so later
877 * assertions have to accept that. Further analysis of the number of
878 * misses under contention is required.
879 */
880 sin = (struct sockaddr_in *)addr;
881 INP_RLOCK(inp);
882 if (sin != NULL &&
883 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
884 INP_RUNLOCK(inp);
885 INP_INFO_WLOCK(&V_udbinfo);
886 INP_WLOCK(inp);
887 unlock_udbinfo = 2;
888 } else if ((sin != NULL && (
889 (sin->sin_addr.s_addr == INADDR_ANY) ||
890 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
891 (inp->inp_laddr.s_addr == INADDR_ANY) ||
892 (inp->inp_lport == 0))) ||
893 (src.sin_family == AF_INET)) {
894 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
895 INP_RUNLOCK(inp);
896 INP_INFO_RLOCK(&V_udbinfo);
897 INP_RLOCK(inp);
898 }
899 unlock_udbinfo = 1;
900 } else
901 unlock_udbinfo = 0;
902
903 /*
904 * If the IP_SENDSRCADDR control message was specified, override the
905 * source address for this datagram. Its use is invalidated if the
906 * address thus specified is incomplete or clobbers other inpcbs.
907 */
908 laddr = inp->inp_laddr;
909 lport = inp->inp_lport;
910 if (src.sin_family == AF_INET) {
911 INP_INFO_LOCK_ASSERT(&V_udbinfo);
912 if ((lport == 0) ||
913 (laddr.s_addr == INADDR_ANY &&
914 src.sin_addr.s_addr == INADDR_ANY)) {
915 error = EINVAL;
916 goto release;
917 }
918 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
919 &laddr.s_addr, &lport, td->td_ucred);
920 if (error)
921 goto release;
922 }
923
924 /*
925 * If a UDP socket has been connected, then a local address/port will
926 * have been selected and bound.
927 *
928 * If a UDP socket has not been connected to, then an explicit
929 * destination address must be used, in which case a local
930 * address/port may not have been selected and bound.
931 */
932 if (sin != NULL) {
933 INP_LOCK_ASSERT(inp);
934 if (inp->inp_faddr.s_addr != INADDR_ANY) {
935 error = EISCONN;
936 goto release;
937 }
938
939 /*
940 * Jail may rewrite the destination address, so let it do
941 * that before we use it.
942 */
943 if (prison_remote_ip4(td->td_ucred, &sin->sin_addr) != 0) {
944 error = EINVAL;
945 goto release;
946 }
947
948 /*
949 * If a local address or port hasn't yet been selected, or if
950 * the destination address needs to be rewritten due to using
951 * a special INADDR_ constant, invoke in_pcbconnect_setup()
952 * to do the heavy lifting. Once a port is selected, we
953 * commit the binding back to the socket; we also commit the
954 * binding of the address if in jail.
955 *
956 * If we already have a valid binding and we're not
957 * requesting a destination address rewrite, use a fast path.
958 */
959 if (inp->inp_laddr.s_addr == INADDR_ANY ||
960 inp->inp_lport == 0 ||
961 sin->sin_addr.s_addr == INADDR_ANY ||
962 sin->sin_addr.s_addr == INADDR_BROADCAST) {
963 INP_INFO_LOCK_ASSERT(&V_udbinfo);
964 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
965 &lport, &faddr.s_addr, &fport, NULL,
966 td->td_ucred);
967 if (error)
968 goto release;
969
970 /*
971 * XXXRW: Why not commit the port if the address is
972 * !INADDR_ANY?
973 */
974 /* Commit the local port if newly assigned. */
975 if (inp->inp_laddr.s_addr == INADDR_ANY &&
976 inp->inp_lport == 0) {
977 INP_INFO_WLOCK_ASSERT(&V_udbinfo);
978 INP_WLOCK_ASSERT(inp);
979 /*
980 * Remember addr if jailed, to prevent
981 * rebinding.
982 */
983 if (jailed(td->td_ucred))
984 inp->inp_laddr = laddr;
985 inp->inp_lport = lport;
986 if (in_pcbinshash(inp) != 0) {
987 inp->inp_lport = 0;
988 error = EAGAIN;
989 goto release;
990 }
991 inp->inp_flags |= INP_ANONPORT;
992 }
993 } else {
994 faddr = sin->sin_addr;
995 fport = sin->sin_port;
996 }
997 } else {
998 INP_LOCK_ASSERT(inp);
999 faddr = inp->inp_faddr;
1000 fport = inp->inp_fport;
1001 if (faddr.s_addr == INADDR_ANY) {
1002 error = ENOTCONN;
1003 goto release;
1004 }
1005 }
1006
1007 /*
1008 * Calculate data length and get a mbuf for UDP, IP, and possible
1009 * link-layer headers. Immediate slide the data pointer back forward
1010 * since we won't use that space at this layer.
1011 */
1012 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1013 if (m == NULL) {
1014 error = ENOBUFS;
1015 goto release;
1016 }
1017 m->m_data += max_linkhdr;
1018 m->m_len -= max_linkhdr;
1019 m->m_pkthdr.len -= max_linkhdr;
1020
1021 /*
1022 * Fill in mbuf with extended UDP header and addresses and length put
1023 * into network format.
1024 */
1025 ui = mtod(m, struct udpiphdr *);
1026 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1027 ui->ui_pr = IPPROTO_UDP;
1028 ui->ui_src = laddr;
1029 ui->ui_dst = faddr;
1030 ui->ui_sport = lport;
1031 ui->ui_dport = fport;
1032 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1033
1034 /*
1035 * Set the Don't Fragment bit in the IP header.
1036 */
1037 if (inp->inp_flags & INP_DONTFRAG) {
1038 struct ip *ip;
1039
1040 ip = (struct ip *)&ui->ui_i;
1041 ip->ip_off |= IP_DF;
1042 }
1043
1044 ipflags = 0;
1045 if (inp->inp_socket->so_options & SO_DONTROUTE)
1046 ipflags |= IP_ROUTETOIF;
1047 if (inp->inp_socket->so_options & SO_BROADCAST)
1048 ipflags |= IP_ALLOWBROADCAST;
1049 if (inp->inp_flags & INP_ONESBCAST)
1050 ipflags |= IP_SENDONES;
1051
1052#ifdef MAC
1053 mac_inpcb_create_mbuf(inp, m);
1054#endif
1055
1056 /*
1057 * Set up checksum and output datagram.
1058 */
1059 if (udp_cksum) {
1060 if (inp->inp_flags & INP_ONESBCAST)
1061 faddr.s_addr = INADDR_BROADCAST;
1062 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1063 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1064 m->m_pkthdr.csum_flags = CSUM_UDP;
1065 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1066 } else
1067 ui->ui_sum = 0;
1068 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1069 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1070 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */
1071 V_udpstat.udps_opackets++;
1072
1073 if (unlock_udbinfo == 2)
1074 INP_INFO_WUNLOCK(&V_udbinfo);
1075 else if (unlock_udbinfo == 1)
1076 INP_INFO_RUNLOCK(&V_udbinfo);
1077 error = ip_output(m, inp->inp_options, NULL, ipflags,
1078 inp->inp_moptions, inp);
1079 if (unlock_udbinfo == 2)
1080 INP_WUNLOCK(inp);
1081 else
1082 INP_RUNLOCK(inp);
1083 return (error);
1084
1085release:
1086 if (unlock_udbinfo == 2) {
1087 INP_WUNLOCK(inp);
1088 INP_INFO_WUNLOCK(&V_udbinfo);
1089 } else if (unlock_udbinfo == 1) {
1090 INP_RUNLOCK(inp);
1091 INP_INFO_RUNLOCK(&V_udbinfo);
1092 } else
1093 INP_RUNLOCK(inp);
1094 m_freem(m);
1095 return (error);
1096}
1097
1098static void
1099udp_abort(struct socket *so)
1100{
1101 INIT_VNET_INET(so->so_vnet);
1102 struct inpcb *inp;
1103
1104 inp = sotoinpcb(so);
1105 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1106 INP_INFO_WLOCK(&V_udbinfo);
1107 INP_WLOCK(inp);
1108 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1109 in_pcbdisconnect(inp);
1110 inp->inp_laddr.s_addr = INADDR_ANY;
1111 soisdisconnected(so);
1112 }
1113 INP_WUNLOCK(inp);
1114 INP_INFO_WUNLOCK(&V_udbinfo);
1115}
1116
1117static int
1118udp_attach(struct socket *so, int proto, struct thread *td)
1119{
1120 INIT_VNET_INET(so->so_vnet);
1121 struct inpcb *inp;
1122 int error;
1123
1124 inp = sotoinpcb(so);
1125 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1126 error = soreserve(so, udp_sendspace, udp_recvspace);
1127 if (error)
1128 return (error);
1129 INP_INFO_WLOCK(&V_udbinfo);
1130 error = in_pcballoc(so, &V_udbinfo);
1131 if (error) {
1132 INP_INFO_WUNLOCK(&V_udbinfo);
1133 return (error);
1134 }
1135
1136 inp = (struct inpcb *)so->so_pcb;
1137 INP_INFO_WUNLOCK(&V_udbinfo);
1138 inp->inp_vflag |= INP_IPV4;
1139 inp->inp_ip_ttl = V_ip_defttl;
1140 INP_WUNLOCK(inp);
1141 return (0);
1142}
1143
1144static int
1145udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1146{
1147 INIT_VNET_INET(so->so_vnet);
1148 struct inpcb *inp;
1149 int error;
1150
1151 inp = sotoinpcb(so);
1152 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1153 INP_INFO_WLOCK(&V_udbinfo);
1154 INP_WLOCK(inp);
1155 error = in_pcbbind(inp, nam, td->td_ucred);
1156 INP_WUNLOCK(inp);
1157 INP_INFO_WUNLOCK(&V_udbinfo);
1158 return (error);
1159}
1160
1161static void
1162udp_close(struct socket *so)
1163{
1164 INIT_VNET_INET(so->so_vnet);
1165 struct inpcb *inp;
1166
1167 inp = sotoinpcb(so);
1168 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1169 INP_INFO_WLOCK(&V_udbinfo);
1170 INP_WLOCK(inp);
1171 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1172 in_pcbdisconnect(inp);
1173 inp->inp_laddr.s_addr = INADDR_ANY;
1174 soisdisconnected(so);
1175 }
1176 INP_WUNLOCK(inp);
1177 INP_INFO_WUNLOCK(&V_udbinfo);
1178}
1179
1180static int
1181udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1182{
1183 INIT_VNET_INET(so->so_vnet);
1184 struct inpcb *inp;
1185 int error;
1186 struct sockaddr_in *sin;
1187
1188 inp = sotoinpcb(so);
1189 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1190 INP_INFO_WLOCK(&V_udbinfo);
1191 INP_WLOCK(inp);
1192 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1193 INP_WUNLOCK(inp);
1194 INP_INFO_WUNLOCK(&V_udbinfo);
1195 return (EISCONN);
1196 }
1197 sin = (struct sockaddr_in *)nam;
1198 if (prison_remote_ip4(td->td_ucred, &sin->sin_addr) != 0) {
1199 INP_WUNLOCK(inp);
1200 INP_INFO_WUNLOCK(&udbinfo);
1201 return (EAFNOSUPPORT);
1202 }
1203 error = in_pcbconnect(inp, nam, td->td_ucred);
1204 if (error == 0)
1205 soisconnected(so);
1206 INP_WUNLOCK(inp);
1207 INP_INFO_WUNLOCK(&V_udbinfo);
1208 return (error);
1209}
1210
1211static void
1212udp_detach(struct socket *so)
1213{
1214 INIT_VNET_INET(so->so_vnet);
1215 struct inpcb *inp;
1216
1217 inp = sotoinpcb(so);
1218 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1219 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1220 ("udp_detach: not disconnected"));
1221 INP_INFO_WLOCK(&V_udbinfo);
1222 INP_WLOCK(inp);
1223 in_pcbdetach(inp);
1224 in_pcbfree(inp);
1225 INP_INFO_WUNLOCK(&V_udbinfo);
1226}
1227
1228static int
1229udp_disconnect(struct socket *so)
1230{
1231 INIT_VNET_INET(so->so_vnet);
1232 struct inpcb *inp;
1233
1234 inp = sotoinpcb(so);
1235 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1236 INP_INFO_WLOCK(&V_udbinfo);
1237 INP_WLOCK(inp);
1238 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1239 INP_WUNLOCK(inp);
1240 INP_INFO_WUNLOCK(&V_udbinfo);
1241 return (ENOTCONN);
1242 }
1243
1244 in_pcbdisconnect(inp);
1245 inp->inp_laddr.s_addr = INADDR_ANY;
1246 SOCK_LOCK(so);
1247 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1248 SOCK_UNLOCK(so);
1249 INP_WUNLOCK(inp);
1250 INP_INFO_WUNLOCK(&V_udbinfo);
1251 return (0);
1252}
1253
1254static int
1255udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1256 struct mbuf *control, struct thread *td)
1257{
1258 struct inpcb *inp;
1259
1260 inp = sotoinpcb(so);
1261 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1262 return (udp_output(inp, m, addr, control, td));
1263}
1264
1265int
1266udp_shutdown(struct socket *so)
1267{
1268 struct inpcb *inp;
1269
1270 inp = sotoinpcb(so);
1271 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1272 INP_WLOCK(inp);
1273 socantsendmore(so);
1274 INP_WUNLOCK(inp);
1275 return (0);
1276}
1277
1278struct pr_usrreqs udp_usrreqs = {
1279 .pru_abort = udp_abort,
1280 .pru_attach = udp_attach,
1281 .pru_bind = udp_bind,
1282 .pru_connect = udp_connect,
1283 .pru_control = in_control,
1284 .pru_detach = udp_detach,
1285 .pru_disconnect = udp_disconnect,
1286 .pru_peeraddr = in_getpeeraddr,
1287 .pru_send = udp_send,
1288 .pru_soreceive = soreceive_dgram,
1289 .pru_sosend = sosend_dgram,
1290 .pru_shutdown = udp_shutdown,
1291 .pru_sockaddr = in_getsockaddr,
1292 .pru_sosetlabel = in_pcbsosetlabel,
1293 .pru_close = udp_close,
1294};