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("$FreeBSD$");
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
61 #include <vm/uma.h>
62
63 #include <net/if.h>
64 #include <net/route.h>
65
66 #include <netinet/in.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #ifdef INET6
72 #include <netinet/ip6.h>
73 #endif
74 #include <netinet/ip_icmp.h>
75 #include <netinet/icmp_var.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_options.h>
78 #ifdef INET6
79 #include <netinet6/ip6_var.h>
80 #endif
81 #include <netinet/udp.h>
82 #include <netinet/udp_var.h>
83 #ifdef INET6
84 #include <netinet6/udp6_var.h>
85 #endif
86
87 #ifdef IPSEC
88 #include <netipsec/ipsec.h>
89 #endif
90
91 #include <machine/in_cksum.h>
92
93 #include <security/mac/mac_framework.h>
94
95 /*
96 * UDP protocol implementation.
97 * Per RFC 768, August, 1980.
98 */
99
100 /*
101 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
102 * removes the only data integrity mechanism for packets and malformed
103 * packets that would otherwise be discarded due to bad checksums, and may
104 * cause problems (especially for NFS data blocks).
105 */
106 static int udp_cksum = 1;
107 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
108 0, "");
109
110 int udp_log_in_vain = 0;
111 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
112 &udp_log_in_vain, 0, "Log all incoming UDP packets");
113
114 int udp_blackhole = 0;
115 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, &udp_blackhole, 0,
116 "Do not send port unreachables for refused connects");
117
118 u_long udp_sendspace = 9216; /* really max datagram size */
119 /* 40 1K datagrams */
120 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
121 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
122
123 u_long udp_recvspace = 40 * (1024 +
124 #ifdef INET6
125 sizeof(struct sockaddr_in6)
126 #else
127 sizeof(struct sockaddr_in)
128 #endif
129 );
130
131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
132 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
133
134 static int udp_soreceive_dgram = 1;
135 SYSCTL_INT(_net_inet_udp, OID_AUTO, soreceive_dgram_enabled,
136 CTLFLAG_RD | CTLFLAG_TUN, &udp_soreceive_dgram, 0,
137 "Use experimental optimized datagram receive");
138
139 struct inpcbhead udb; /* from udp_var.h */
140 struct inpcbinfo udbinfo;
141
142 #ifndef UDBHASHSIZE
143 #define UDBHASHSIZE 128
144 #endif
145
146 struct udpstat udpstat; /* from udp_var.h */
147 SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW, &udpstat,
148 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
149
150 static void udp_detach(struct socket *so);
151 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
152 struct mbuf *, struct thread *);
153
154 static void
155 udp_zone_change(void *tag)
156 {
157
158 uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
159 }
160
161 static int
162 udp_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
171 void
172 udp_init(void)
173 {
174
175 INP_INFO_LOCK_INIT(&udbinfo, "udp");
176 LIST_INIT(&udb);
177 udbinfo.ipi_listhead = &udb;
178 udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
179 &udbinfo.ipi_hashmask);
180 udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
181 &udbinfo.ipi_porthashmask);
182 udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
183 NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
184 uma_zone_set_max(udbinfo.ipi_zone, maxsockets);
185 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
186 EVENTHANDLER_PRI_ANY);
187 TUNABLE_INT_FETCH("net.inet.udp.soreceive_dgram_enabled",
188 &udp_soreceive_dgram);
189 if (udp_soreceive_dgram) {
190 udp_usrreqs.pru_soreceive = soreceive_dgram;
191 #ifdef INET6
192 udp6_usrreqs.pru_soreceive = soreceive_dgram;
193 #endif
194 }
195 }
196
197 /*
198 * Subroutine of udp_input(), which appends the provided mbuf chain to the
199 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
200 * contains the source address. If the socket ends up being an IPv6 socket,
201 * udp_append() will convert to a sockaddr_in6 before passing the address
202 * into the socket code.
203 */
204 static void
205 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
206 struct sockaddr_in *udp_in)
207 {
208 struct sockaddr *append_sa;
209 struct socket *so;
210 struct mbuf *opts = 0;
211 #ifdef INET6
212 struct sockaddr_in6 udp_in6;
213 #endif
214
215 INP_RLOCK_ASSERT(inp);
216
217 #ifdef IPSEC
218 /* Check AH/ESP integrity. */
219 if (ipsec4_in_reject(n, inp)) {
220 m_freem(n);
221 ipsec4stat.in_polvio++;
222 return;
223 }
224 #endif /* IPSEC */
225 #ifdef MAC
226 if (mac_check_inpcb_deliver(inp, n) != 0) {
227 m_freem(n);
228 return;
229 }
230 #endif
231 if (inp->inp_flags & INP_CONTROLOPTS ||
232 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
233 #ifdef INET6
234 if (inp->inp_vflag & INP_IPV6)
235 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
236 else
237 #endif
238 ip_savecontrol(inp, &opts, ip, n);
239 }
240 #ifdef INET6
241 if (inp->inp_vflag & INP_IPV6) {
242 bzero(&udp_in6, sizeof(udp_in6));
243 udp_in6.sin6_len = sizeof(udp_in6);
244 udp_in6.sin6_family = AF_INET6;
245 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
246 append_sa = (struct sockaddr *)&udp_in6;
247 } else
248 #endif
249 append_sa = (struct sockaddr *)udp_in;
250 m_adj(n, off);
251
252 so = inp->inp_socket;
253 SOCKBUF_LOCK(&so->so_rcv);
254 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
255 SOCKBUF_UNLOCK(&so->so_rcv);
256 m_freem(n);
257 if (opts)
258 m_freem(opts);
259 udpstat.udps_fullsock++;
260 } else
261 sorwakeup_locked(so);
262 }
263
264 void
265 udp_input(struct mbuf *m, int off)
266 {
267 int iphlen = off;
268 struct ip *ip;
269 struct udphdr *uh;
270 struct ifnet *ifp;
271 struct inpcb *inp;
272 int len;
273 struct ip save_ip;
274 struct sockaddr_in udp_in;
275 #ifdef IPFIREWALL_FORWARD
276 struct m_tag *fwd_tag;
277 #endif
278
279 ifp = m->m_pkthdr.rcvif;
280 udpstat.udps_ipackets++;
281
282 /*
283 * Strip IP options, if any; should skip this, make available to
284 * user, and use on returned packets, but we don't yet have a way to
285 * check the checksum with options still present.
286 */
287 if (iphlen > sizeof (struct ip)) {
288 ip_stripoptions(m, (struct mbuf *)0);
289 iphlen = sizeof(struct ip);
290 }
291
292 /*
293 * Get IP and UDP header together in first mbuf.
294 */
295 ip = mtod(m, struct ip *);
296 if (m->m_len < iphlen + sizeof(struct udphdr)) {
297 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
298 udpstat.udps_hdrops++;
299 return;
300 }
301 ip = mtod(m, struct ip *);
302 }
303 uh = (struct udphdr *)((caddr_t)ip + iphlen);
304
305 /*
306 * Destination port of 0 is illegal, based on RFC768.
307 */
308 if (uh->uh_dport == 0)
309 goto badunlocked;
310
311 /*
312 * Construct sockaddr format source address. Stuff source address
313 * and datagram in user buffer.
314 */
315 bzero(&udp_in, sizeof(udp_in));
316 udp_in.sin_len = sizeof(udp_in);
317 udp_in.sin_family = AF_INET;
318 udp_in.sin_port = uh->uh_sport;
319 udp_in.sin_addr = ip->ip_src;
320
321 /*
322 * Make mbuf data length reflect UDP length. If not enough data to
323 * reflect UDP length, drop.
324 */
325 len = ntohs((u_short)uh->uh_ulen);
326 if (ip->ip_len != len) {
327 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
328 udpstat.udps_badlen++;
329 goto badunlocked;
330 }
331 m_adj(m, len - ip->ip_len);
332 /* ip->ip_len = len; */
333 }
334
335 /*
336 * Save a copy of the IP header in case we want restore it for
337 * sending an ICMP error message in response.
338 */
339 if (!udp_blackhole)
340 save_ip = *ip;
341 else
342 memset(&save_ip, 0, sizeof(save_ip));
343
344 /*
345 * Checksum extended UDP header and data.
346 */
347 if (uh->uh_sum) {
348 u_short uh_sum;
349
350 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
351 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
352 uh_sum = m->m_pkthdr.csum_data;
353 else
354 uh_sum = in_pseudo(ip->ip_src.s_addr,
355 ip->ip_dst.s_addr, htonl((u_short)len +
356 m->m_pkthdr.csum_data + IPPROTO_UDP));
357 uh_sum ^= 0xffff;
358 } else {
359 char b[9];
360
361 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
362 bzero(((struct ipovly *)ip)->ih_x1, 9);
363 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
364 uh_sum = in_cksum(m, len + sizeof (struct ip));
365 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
366 }
367 if (uh_sum) {
368 udpstat.udps_badsum++;
369 m_freem(m);
370 return;
371 }
372 } else
373 udpstat.udps_nosum++;
374
375 #ifdef IPFIREWALL_FORWARD
376 /*
377 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
378 */
379 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
380 if (fwd_tag != NULL) {
381 struct sockaddr_in *next_hop;
382
383 /*
384 * Do the hack.
385 */
386 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
387 ip->ip_dst = next_hop->sin_addr;
388 uh->uh_dport = ntohs(next_hop->sin_port);
389
390 /*
391 * Remove the tag from the packet. We don't need it anymore.
392 */
393 m_tag_delete(m, fwd_tag);
394 }
395 #endif
396
397 INP_INFO_RLOCK(&udbinfo);
398 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
399 in_broadcast(ip->ip_dst, ifp)) {
400 struct inpcb *last;
401 struct ip_moptions *imo;
402
403 last = NULL;
404 LIST_FOREACH(inp, &udb, inp_list) {
405 if (inp->inp_lport != uh->uh_dport)
406 continue;
407 #ifdef INET6
408 if ((inp->inp_vflag & INP_IPV4) == 0)
409 continue;
410 #endif
411 if (inp->inp_laddr.s_addr != INADDR_ANY &&
412 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
413 continue;
414 if (inp->inp_faddr.s_addr != INADDR_ANY &&
415 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
416 continue;
417 /*
418 * XXX: Do not check source port of incoming datagram
419 * unless inp_connect() has been called to bind the
420 * fport part of the 4-tuple; the source could be
421 * trying to talk to us with an ephemeral port.
422 */
423 if (inp->inp_fport != 0 &&
424 inp->inp_fport != uh->uh_sport)
425 continue;
426
427 INP_RLOCK(inp);
428
429 /*
430 * Handle socket delivery policy for any-source
431 * and source-specific multicast. [RFC3678]
432 */
433 imo = inp->inp_moptions;
434 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
435 imo != NULL) {
436 struct sockaddr_in sin;
437 struct in_msource *ims;
438 int blocked, mode;
439 size_t idx;
440
441 bzero(&sin, sizeof(struct sockaddr_in));
442 sin.sin_len = sizeof(struct sockaddr_in);
443 sin.sin_family = AF_INET;
444 sin.sin_addr = ip->ip_dst;
445
446 blocked = 0;
447 idx = imo_match_group(imo, ifp,
448 (struct sockaddr *)&sin);
449 if (idx == -1) {
450 /*
451 * No group membership for this socket.
452 * Do not bump udps_noportbcast, as
453 * this will happen further down.
454 */
455 blocked++;
456 } else {
457 /*
458 * Check for a multicast source filter
459 * entry on this socket for this group.
460 * MCAST_EXCLUDE is the default
461 * behaviour. It means default accept;
462 * entries, if present, denote sources
463 * to be excluded from delivery.
464 */
465 ims = imo_match_source(imo, idx,
466 (struct sockaddr *)&udp_in);
467 mode = imo->imo_mfilters[idx].imf_fmode;
468 if ((ims != NULL &&
469 mode == MCAST_EXCLUDE) ||
470 (ims == NULL &&
471 mode == MCAST_INCLUDE)) {
472 #ifdef DIAGNOSTIC
473 if (bootverbose) {
474 printf("%s: blocked by"
475 " source filter\n",
476 __func__);
477 }
478 #endif
479 udpstat.udps_filtermcast++;
480 blocked++;
481 }
482 }
483 if (blocked != 0) {
484 INP_RUNLOCK(inp);
485 continue;
486 }
487 }
488 if (last != NULL) {
489 struct mbuf *n;
490
491 n = m_copy(m, 0, M_COPYALL);
492 if (n != NULL)
493 udp_append(last, ip, n, iphlen +
494 sizeof(struct udphdr), &udp_in);
495 INP_RUNLOCK(last);
496 }
497 last = inp;
498 /*
499 * Don't look for additional matches if this one does
500 * not have either the SO_REUSEPORT or SO_REUSEADDR
501 * socket options set. This heuristic avoids
502 * searching through all pcbs in the common case of a
503 * non-shared port. It assumes that an application
504 * will never clear these options after setting them.
505 */
506 if ((last->inp_socket->so_options &
507 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
508 break;
509 }
510
511 if (last == NULL) {
512 /*
513 * No matching pcb found; discard datagram. (No need
514 * to send an ICMP Port Unreachable for a broadcast
515 * or multicast datgram.)
516 */
517 udpstat.udps_noportbcast++;
518 goto badheadlocked;
519 }
520 udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
521 &udp_in);
522 INP_RUNLOCK(last);
523 INP_INFO_RUNLOCK(&udbinfo);
524 return;
525 }
526
527 /*
528 * Locate pcb for datagram.
529 */
530 inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport,
531 ip->ip_dst, uh->uh_dport, 1, ifp);
532 if (inp == NULL) {
533 if (udp_log_in_vain) {
534 char buf[4*sizeof "123"];
535
536 strcpy(buf, inet_ntoa(ip->ip_dst));
537 log(LOG_INFO,
538 "Connection attempt to UDP %s:%d from %s:%d\n",
539 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
540 ntohs(uh->uh_sport));
541 }
542 udpstat.udps_noport++;
543 if (m->m_flags & (M_BCAST | M_MCAST)) {
544 udpstat.udps_noportbcast++;
545 goto badheadlocked;
546 }
547 if (udp_blackhole)
548 goto badheadlocked;
549 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
550 goto badheadlocked;
551 *ip = save_ip;
552 ip->ip_len += iphlen;
553 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
554 INP_INFO_RUNLOCK(&udbinfo);
555 return;
556 }
557
558 /*
559 * Check the minimum TTL for socket.
560 */
561 INP_RLOCK(inp);
562 INP_INFO_RUNLOCK(&udbinfo);
563 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
564 INP_RUNLOCK(inp);
565 goto badunlocked;
566 }
567 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
568 INP_RUNLOCK(inp);
569 return;
570
571 badheadlocked:
572 if (inp)
573 INP_RUNLOCK(inp);
574 INP_INFO_RUNLOCK(&udbinfo);
575 badunlocked:
576 m_freem(m);
577 }
578
579 /*
580 * Notify a udp user of an asynchronous error; just wake up so that they can
581 * collect error status.
582 */
583 struct inpcb *
584 udp_notify(struct inpcb *inp, int errno)
585 {
586
587 /*
588 * While udp_ctlinput() always calls udp_notify() with a read lock
589 * when invoking it directly, in_pcbnotifyall() currently uses write
590 * locks due to sharing code with TCP. For now, accept either a read
591 * or a write lock, but a read lock is sufficient.
592 */
593 INP_LOCK_ASSERT(inp);
594
595 inp->inp_socket->so_error = errno;
596 sorwakeup(inp->inp_socket);
597 sowwakeup(inp->inp_socket);
598 return (inp);
599 }
600
601 void
602 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
603 {
604 struct ip *ip = vip;
605 struct udphdr *uh;
606 struct in_addr faddr;
607 struct inpcb *inp;
608
609 faddr = ((struct sockaddr_in *)sa)->sin_addr;
610 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
611 return;
612
613 /*
614 * Redirects don't need to be handled up here.
615 */
616 if (PRC_IS_REDIRECT(cmd))
617 return;
618
619 /*
620 * Hostdead is ugly because it goes linearly through all PCBs.
621 *
622 * XXX: We never get this from ICMP, otherwise it makes an excellent
623 * DoS attack on machines with many connections.
624 */
625 if (cmd == PRC_HOSTDEAD)
626 ip = NULL;
627 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
628 return;
629 if (ip != NULL) {
630 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
631 INP_INFO_RLOCK(&udbinfo);
632 inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport,
633 ip->ip_src, uh->uh_sport, 0, NULL);
634 if (inp != NULL) {
635 INP_RLOCK(inp);
636 if (inp->inp_socket != NULL) {
637 udp_notify(inp, inetctlerrmap[cmd]);
638 }
639 INP_RUNLOCK(inp);
640 }
641 INP_INFO_RUNLOCK(&udbinfo);
642 } else
643 in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd],
644 udp_notify);
645 }
646
647 static int
648 udp_pcblist(SYSCTL_HANDLER_ARGS)
649 {
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 = 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(&udbinfo);
673 gencnt = udbinfo.ipi_gencnt;
674 n = udbinfo.ipi_count;
675 INP_INFO_RUNLOCK(&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(&udbinfo);
695 for (inp = LIST_FIRST(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(&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(&udbinfo);
732 xig.xig_gen = udbinfo.ipi_gencnt;
733 xig.xig_sogen = so_gencnt;
734 xig.xig_count = udbinfo.ipi_count;
735 INP_INFO_RUNLOCK(&udbinfo);
736 error = SYSCTL_OUT(req, &xig, sizeof xig);
737 }
738 free(inp_list, M_TEMP);
739 return (error);
740 }
741
742 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
743 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
744
745 static int
746 udp_getcred(SYSCTL_HANDLER_ARGS)
747 {
748 struct xucred xuc;
749 struct sockaddr_in addrs[2];
750 struct inpcb *inp;
751 int error;
752
753 error = priv_check(req->td, PRIV_NETINET_GETCRED);
754 if (error)
755 return (error);
756 error = SYSCTL_IN(req, addrs, sizeof(addrs));
757 if (error)
758 return (error);
759 INP_INFO_RLOCK(&udbinfo);
760 inp = in_pcblookup_hash(&udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
761 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
762 if (inp != NULL) {
763 INP_RLOCK(inp);
764 INP_INFO_RUNLOCK(&udbinfo);
765 if (inp->inp_socket == NULL)
766 error = ENOENT;
767 if (error == 0)
768 error = cr_canseeinpcb(req->td->td_ucred, inp);
769 if (error == 0)
770 cru2x(inp->inp_cred, &xuc);
771 INP_RUNLOCK(inp);
772 } else {
773 INP_INFO_RUNLOCK(&udbinfo);
774 error = ENOENT;
775 }
776 if (error == 0)
777 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
778 return (error);
779 }
780
781 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
782 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
783 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
784
785 static int
786 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
787 struct mbuf *control, struct thread *td)
788 {
789 struct udpiphdr *ui;
790 int len = m->m_pkthdr.len;
791 struct in_addr faddr, laddr;
792 struct cmsghdr *cm;
793 struct sockaddr_in *sin, src;
794 int error = 0;
795 int ipflags;
796 u_short fport, lport;
797 int unlock_udbinfo;
798
799 /*
800 * udp_output() may need to temporarily bind or connect the current
801 * inpcb. As such, we don't know up front whether we will need the
802 * pcbinfo lock or not. Do any work to decide what is needed up
803 * front before acquiring any locks.
804 */
805 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
806 if (control)
807 m_freem(control);
808 m_freem(m);
809 return (EMSGSIZE);
810 }
811
812 src.sin_family = 0;
813 if (control != NULL) {
814 /*
815 * XXX: Currently, we assume all the optional information is
816 * stored in a single mbuf.
817 */
818 if (control->m_next) {
819 m_freem(control);
820 m_freem(m);
821 return (EINVAL);
822 }
823 for (; control->m_len > 0;
824 control->m_data += CMSG_ALIGN(cm->cmsg_len),
825 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
826 cm = mtod(control, struct cmsghdr *);
827 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
828 || cm->cmsg_len > control->m_len) {
829 error = EINVAL;
830 break;
831 }
832 if (cm->cmsg_level != IPPROTO_IP)
833 continue;
834
835 switch (cm->cmsg_type) {
836 case IP_SENDSRCADDR:
837 if (cm->cmsg_len !=
838 CMSG_LEN(sizeof(struct in_addr))) {
839 error = EINVAL;
840 break;
841 }
842 bzero(&src, sizeof(src));
843 src.sin_family = AF_INET;
844 src.sin_len = sizeof(src);
845 src.sin_port = inp->inp_lport;
846 src.sin_addr =
847 *(struct in_addr *)CMSG_DATA(cm);
848 break;
849
850 default:
851 error = ENOPROTOOPT;
852 break;
853 }
854 if (error)
855 break;
856 }
857 m_freem(control);
858 }
859 if (error) {
860 m_freem(m);
861 return (error);
862 }
863
864 /*
865 * Depending on whether or not the application has bound or connected
866 * the socket, we may have to do varying levels of work. The optimal
867 * case is for a connected UDP socket, as a global lock isn't
868 * required at all.
869 *
870 * In order to decide which we need, we require stability of the
871 * inpcb binding, which we ensure by acquiring a read lock on the
872 * inpcb. This doesn't strictly follow the lock order, so we play
873 * the trylock and retry game; note that we may end up with more
874 * conservative locks than required the second time around, so later
875 * assertions have to accept that. Further analysis of the number of
876 * misses under contention is required.
877 */
878 sin = (struct sockaddr_in *)addr;
879 INP_RLOCK(inp);
880 if (sin != NULL &&
881 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
882 INP_RUNLOCK(inp);
883 INP_INFO_WLOCK(&udbinfo);
884 INP_WLOCK(inp);
885 unlock_udbinfo = 2;
886 } else if ((sin != NULL && (
887 (sin->sin_addr.s_addr == INADDR_ANY) ||
888 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
889 (inp->inp_laddr.s_addr == INADDR_ANY) ||
890 (inp->inp_lport == 0))) ||
891 (src.sin_family == AF_INET)) {
892 if (!INP_INFO_TRY_RLOCK(&udbinfo)) {
893 INP_RUNLOCK(inp);
894 INP_INFO_RLOCK(&udbinfo);
895 INP_RLOCK(inp);
896 }
897 unlock_udbinfo = 1;
898 } else
899 unlock_udbinfo = 0;
900
901 /*
902 * If the IP_SENDSRCADDR control message was specified, override the
903 * source address for this datagram. Its use is invalidated if the
904 * address thus specified is incomplete or clobbers other inpcbs.
905 */
906 laddr = inp->inp_laddr;
907 lport = inp->inp_lport;
908 if (src.sin_family == AF_INET) {
909 INP_INFO_LOCK_ASSERT(&udbinfo);
910 if ((lport == 0) ||
911 (laddr.s_addr == INADDR_ANY &&
912 src.sin_addr.s_addr == INADDR_ANY)) {
913 error = EINVAL;
914 goto release;
915 }
916 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
917 &laddr.s_addr, &lport, td->td_ucred);
918 if (error)
919 goto release;
920 }
921
922 /*
923 * If a UDP socket has been connected, then a local address/port will
924 * have been selected and bound.
925 *
926 * If a UDP socket has not been connected to, then an explicit
927 * destination address must be used, in which case a local
928 * address/port may not have been selected and bound.
929 */
930 if (sin != NULL) {
931 INP_LOCK_ASSERT(inp);
932 if (inp->inp_faddr.s_addr != INADDR_ANY) {
933 error = EISCONN;
934 goto release;
935 }
936
937 /*
938 * Jail may rewrite the destination address, so let it do
939 * that before we use it.
940 */
941 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
942 if (error)
943 goto release;
944
945 /*
946 * If a local address or port hasn't yet been selected, or if
947 * the destination address needs to be rewritten due to using
948 * a special INADDR_ constant, invoke in_pcbconnect_setup()
949 * to do the heavy lifting. Once a port is selected, we
950 * commit the binding back to the socket; we also commit the
951 * binding of the address if in jail.
952 *
953 * If we already have a valid binding and we're not
954 * requesting a destination address rewrite, use a fast path.
955 */
956 if (inp->inp_laddr.s_addr == INADDR_ANY ||
957 inp->inp_lport == 0 ||
958 sin->sin_addr.s_addr == INADDR_ANY ||
959 sin->sin_addr.s_addr == INADDR_BROADCAST) {
960 INP_INFO_LOCK_ASSERT(&udbinfo);
961 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
962 &lport, &faddr.s_addr, &fport, NULL,
963 td->td_ucred);
964 if (error)
965 goto release;
966
967 /*
968 * XXXRW: Why not commit the port if the address is
969 * !INADDR_ANY?
970 */
971 /* Commit the local port if newly assigned. */
972 if (inp->inp_laddr.s_addr == INADDR_ANY &&
973 inp->inp_lport == 0) {
974 INP_INFO_WLOCK_ASSERT(&udbinfo);
975 INP_WLOCK_ASSERT(inp);
976 /*
977 * Remember addr if jailed, to prevent
978 * rebinding.
979 */
980 if (jailed(td->td_ucred))
981 inp->inp_laddr = laddr;
982 inp->inp_lport = lport;
983 if (in_pcbinshash(inp) != 0) {
984 inp->inp_lport = 0;
985 error = EAGAIN;
986 goto release;
987 }
988 inp->inp_flags |= INP_ANONPORT;
989 }
990 } else {
991 faddr = sin->sin_addr;
992 fport = sin->sin_port;
993 }
994 } else {
995 INP_LOCK_ASSERT(inp);
996 faddr = inp->inp_faddr;
997 fport = inp->inp_fport;
998 if (faddr.s_addr == INADDR_ANY) {
999 error = ENOTCONN;
1000 goto release;
1001 }
1002 }
1003
1004 /*
1005 * Calculate data length and get a mbuf for UDP, IP, and possible
1006 * link-layer headers. Immediate slide the data pointer back forward
1007 * since we won't use that space at this layer.
1008 */
1009 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1010 if (m == NULL) {
1011 error = ENOBUFS;
1012 goto release;
1013 }
1014 m->m_data += max_linkhdr;
1015 m->m_len -= max_linkhdr;
1016 m->m_pkthdr.len -= max_linkhdr;
1017
1018 /*
1019 * Fill in mbuf with extended UDP header and addresses and length put
1020 * into network format.
1021 */
1022 ui = mtod(m, struct udpiphdr *);
1023 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1024 ui->ui_pr = IPPROTO_UDP;
1025 ui->ui_src = laddr;
1026 ui->ui_dst = faddr;
1027 ui->ui_sport = lport;
1028 ui->ui_dport = fport;
1029 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1030
1031 /*
1032 * Set the Don't Fragment bit in the IP header.
1033 */
1034 if (inp->inp_flags & INP_DONTFRAG) {
1035 struct ip *ip;
1036
1037 ip = (struct ip *)&ui->ui_i;
1038 ip->ip_off |= IP_DF;
1039 }
1040
1041 ipflags = 0;
1042 if (inp->inp_socket->so_options & SO_DONTROUTE)
1043 ipflags |= IP_ROUTETOIF;
1044 if (inp->inp_socket->so_options & SO_BROADCAST)
1045 ipflags |= IP_ALLOWBROADCAST;
1046 if (inp->inp_flags & INP_ONESBCAST)
1047 ipflags |= IP_SENDONES;
1048
1049 #ifdef MAC
1050 mac_create_mbuf_from_inpcb(inp, m);
1051 #endif
1052
1053 /*
1054 * Set up checksum and output datagram.
1055 */
1056 if (udp_cksum) {
1057 if (inp->inp_flags & INP_ONESBCAST)
1058 faddr.s_addr = INADDR_BROADCAST;
1059 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1060 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1061 m->m_pkthdr.csum_flags = CSUM_UDP;
1062 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1063 } else
1064 ui->ui_sum = 0;
1065 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1066 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1067 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */
1068 udpstat.udps_opackets++;
1069
1070 if (unlock_udbinfo == 2)
1071 INP_INFO_WUNLOCK(&udbinfo);
1072 else if (unlock_udbinfo == 1)
1073 INP_INFO_RUNLOCK(&udbinfo);
1074 error = ip_output(m, inp->inp_options, NULL, ipflags,
1075 inp->inp_moptions, inp);
1076 if (unlock_udbinfo == 2)
1077 INP_WUNLOCK(inp);
1078 else
1079 INP_RUNLOCK(inp);
1080 return (error);
1081
1082 release:
1083 if (unlock_udbinfo == 2) {
1084 INP_WUNLOCK(inp);
1085 INP_INFO_WUNLOCK(&udbinfo);
1086 } else if (unlock_udbinfo == 1) {
1087 INP_RUNLOCK(inp);
1088 INP_INFO_RUNLOCK(&udbinfo);
1089 } else
1090 INP_RUNLOCK(inp);
1091 m_freem(m);
1092 return (error);
1093 }
1094
1095 static void
1096 udp_abort(struct socket *so)
1097 {
1098 struct inpcb *inp;
1099
1100 inp = sotoinpcb(so);
1101 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1102 INP_INFO_WLOCK(&udbinfo);
1103 INP_WLOCK(inp);
1104 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1105 in_pcbdisconnect(inp);
1106 inp->inp_laddr.s_addr = INADDR_ANY;
1107 soisdisconnected(so);
1108 }
1109 INP_WUNLOCK(inp);
1110 INP_INFO_WUNLOCK(&udbinfo);
1111 }
1112
1113 static int
1114 udp_attach(struct socket *so, int proto, struct thread *td)
1115 {
1116 struct inpcb *inp;
1117 int error;
1118
1119 inp = sotoinpcb(so);
1120 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1121 error = soreserve(so, udp_sendspace, udp_recvspace);
1122 if (error)
1123 return (error);
1124 INP_INFO_WLOCK(&udbinfo);
1125 error = in_pcballoc(so, &udbinfo);
1126 if (error) {
1127 INP_INFO_WUNLOCK(&udbinfo);
1128 return (error);
1129 }
1130
1131 inp = (struct inpcb *)so->so_pcb;
1132 INP_INFO_WUNLOCK(&udbinfo);
1133 inp->inp_vflag |= INP_IPV4;
1134 inp->inp_ip_ttl = ip_defttl;
1135 INP_WUNLOCK(inp);
1136 return (0);
1137 }
1138
1139 static int
1140 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1141 {
1142 struct inpcb *inp;
1143 int error;
1144
1145 inp = sotoinpcb(so);
1146 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1147 INP_INFO_WLOCK(&udbinfo);
1148 INP_WLOCK(inp);
1149 error = in_pcbbind(inp, nam, td->td_ucred);
1150 INP_WUNLOCK(inp);
1151 INP_INFO_WUNLOCK(&udbinfo);
1152 return (error);
1153 }
1154
1155 static void
1156 udp_close(struct socket *so)
1157 {
1158 struct inpcb *inp;
1159
1160 inp = sotoinpcb(so);
1161 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1162 INP_INFO_WLOCK(&udbinfo);
1163 INP_WLOCK(inp);
1164 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1165 in_pcbdisconnect(inp);
1166 inp->inp_laddr.s_addr = INADDR_ANY;
1167 soisdisconnected(so);
1168 }
1169 INP_WUNLOCK(inp);
1170 INP_INFO_WUNLOCK(&udbinfo);
1171 }
1172
1173 static int
1174 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1175 {
1176 struct inpcb *inp;
1177 int error;
1178 struct sockaddr_in *sin;
1179
1180 inp = sotoinpcb(so);
1181 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1182 INP_INFO_WLOCK(&udbinfo);
1183 INP_WLOCK(inp);
1184 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1185 INP_WUNLOCK(inp);
1186 INP_INFO_WUNLOCK(&udbinfo);
1187 return (EISCONN);
1188 }
1189 sin = (struct sockaddr_in *)nam;
1190 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1191 if (error != 0) {
1192 INP_WUNLOCK(inp);
1193 INP_INFO_WUNLOCK(&udbinfo);
1194 return (error);
1195 }
1196 error = in_pcbconnect(inp, nam, td->td_ucred);
1197 if (error == 0)
1198 soisconnected(so);
1199 INP_WUNLOCK(inp);
1200 INP_INFO_WUNLOCK(&udbinfo);
1201 return (error);
1202 }
1203
1204 static void
1205 udp_detach(struct socket *so)
1206 {
1207 struct inpcb *inp;
1208
1209 inp = sotoinpcb(so);
1210 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1211 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1212 ("udp_detach: not disconnected"));
1213 INP_INFO_WLOCK(&udbinfo);
1214 INP_WLOCK(inp);
1215 in_pcbdetach(inp);
1216 in_pcbfree(inp);
1217 INP_INFO_WUNLOCK(&udbinfo);
1218 }
1219
1220 static int
1221 udp_disconnect(struct socket *so)
1222 {
1223 struct inpcb *inp;
1224
1225 inp = sotoinpcb(so);
1226 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1227 INP_INFO_WLOCK(&udbinfo);
1228 INP_WLOCK(inp);
1229 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1230 INP_WUNLOCK(inp);
1231 INP_INFO_WUNLOCK(&udbinfo);
1232 return (ENOTCONN);
1233 }
1234
1235 in_pcbdisconnect(inp);
1236 inp->inp_laddr.s_addr = INADDR_ANY;
1237 SOCK_LOCK(so);
1238 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1239 SOCK_UNLOCK(so);
1240 INP_WUNLOCK(inp);
1241 INP_INFO_WUNLOCK(&udbinfo);
1242 return (0);
1243 }
1244
1245 static int
1246 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1247 struct mbuf *control, struct thread *td)
1248 {
1249 struct inpcb *inp;
1250
1251 inp = sotoinpcb(so);
1252 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1253 return (udp_output(inp, m, addr, control, td));
1254 }
1255
1256 int
1257 udp_shutdown(struct socket *so)
1258 {
1259 struct inpcb *inp;
1260
1261 inp = sotoinpcb(so);
1262 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1263 INP_WLOCK(inp);
1264 socantsendmore(so);
1265 INP_WUNLOCK(inp);
1266 return (0);
1267 }
1268
1269 struct pr_usrreqs udp_usrreqs = {
1270 .pru_abort = udp_abort,
1271 .pru_attach = udp_attach,
1272 .pru_bind = udp_bind,
1273 .pru_connect = udp_connect,
1274 .pru_control = in_control,
1275 .pru_detach = udp_detach,
1276 .pru_disconnect = udp_disconnect,
1277 .pru_peeraddr = in_getpeeraddr,
1278 .pru_send = udp_send,
1279 .pru_sosend = sosend_dgram,
1280 .pru_shutdown = udp_shutdown,
1281 .pru_sockaddr = in_getsockaddr,
1282 .pru_sosetlabel = in_pcbsosetlabel,
1283 .pru_close = udp_close,
1284 };
Cache object: ca2603a9a6c374c96ac2ce2515450bea
|