1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2008 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
8 * Copyright (c) 2014 Kevin Lo
9 * All rights reserved.
10 *
11 * Portions of this software were developed by Robert N. M. Watson under
12 * contract to Juniper Networks, Inc.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 */
38
39 #include <sys/cdefs.h>
40 #include "opt_inet.h"
41 #include "opt_inet6.h"
42 #include "opt_ipsec.h"
43 #include "opt_route.h"
44 #include "opt_rss.h"
45
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/sdt.h>
58 #include <sys/signalvar.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sx.h>
62 #include <sys/sysctl.h>
63 #include <sys/syslog.h>
64 #include <sys/systm.h>
65
66 #include <vm/uma.h>
67
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/route.h>
71 #include <net/route/nhop.h>
72 #include <net/rss_config.h>
73
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #endif
84 #include <netinet/ip_icmp.h>
85 #include <netinet/icmp_var.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_options.h>
88 #ifdef INET6
89 #include <netinet6/ip6_var.h>
90 #endif
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
93 #include <netinet/udplite.h>
94 #include <netinet/in_rss.h>
95
96 #include <netipsec/ipsec_support.h>
97
98 #include <machine/in_cksum.h>
99
100 #include <security/mac/mac_framework.h>
101
102 /*
103 * UDP and UDP-Lite protocols implementation.
104 * Per RFC 768, August, 1980.
105 * Per RFC 3828, July, 2004.
106 */
107
108 VNET_DEFINE(int, udp_bind_all_fibs) = 1;
109 SYSCTL_INT(_net_inet_udp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN,
110 &VNET_NAME(udp_bind_all_fibs), 0,
111 "Bound sockets receive traffic from all FIBs");
112
113 /*
114 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
115 * removes the only data integrity mechanism for packets and malformed
116 * packets that would otherwise be discarded due to bad checksums, and may
117 * cause problems (especially for NFS data blocks).
118 */
119 VNET_DEFINE(int, udp_cksum) = 1;
120 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
121 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
122
123 VNET_DEFINE(int, udp_log_in_vain) = 0;
124 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
125 &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets");
126
127 VNET_DEFINE(int, udp_blackhole) = 0;
128 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
129 &VNET_NAME(udp_blackhole), 0,
130 "Do not send port unreachables for refused connects");
131 VNET_DEFINE(bool, udp_blackhole_local) = false;
132 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
133 CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false,
134 "Enforce net.inet.udp.blackhole for locally originated packets");
135
136 u_long udp_sendspace = 9216; /* really max datagram size */
137 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
138 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
139
140 u_long udp_recvspace = 40 * (1024 +
141 #ifdef INET6
142 sizeof(struct sockaddr_in6)
143 #else
144 sizeof(struct sockaddr_in)
145 #endif
146 ); /* 40 1K datagrams */
147
148 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
149 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
150
151 VNET_DEFINE(struct inpcbinfo, udbinfo);
152 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
153
154 #ifndef UDBHASHSIZE
155 #define UDBHASHSIZE 128
156 #endif
157
158 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
159 VNET_PCPUSTAT_SYSINIT(udpstat);
160 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
161 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
162
163 #ifdef VIMAGE
164 VNET_PCPUSTAT_SYSUNINIT(udpstat);
165 #endif /* VIMAGE */
166 #ifdef INET
167 static void udp_detach(struct socket *so);
168 #endif
169
170 INPCBSTORAGE_DEFINE(udpcbstor, udpcb, "udpinp", "udp_inpcb", "udp", "udphash");
171 INPCBSTORAGE_DEFINE(udplitecbstor, udpcb, "udpliteinp", "udplite_inpcb",
172 "udplite", "udplitehash");
173
174 static void
udp_vnet_init(void * arg __unused)175 udp_vnet_init(void *arg __unused)
176 {
177
178 /*
179 * For now default to 2-tuple UDP hashing - until the fragment
180 * reassembly code can also update the flowid.
181 *
182 * Once we can calculate the flowid that way and re-establish
183 * a 4-tuple, flip this to 4-tuple.
184 */
185 in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE);
186 /* Additional pcbinfo for UDP-Lite */
187 in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE,
188 UDBHASHSIZE);
189 }
190 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
191 udp_vnet_init, NULL);
192
193 /*
194 * Kernel module interface for updating udpstat. The argument is an index
195 * into udpstat treated as an array of u_long. While this encodes the
196 * general layout of udpstat into the caller, it doesn't encode its location,
197 * so that future changes to add, for example, per-CPU stats support won't
198 * cause binary compatibility problems for kernel modules.
199 */
200 void
kmod_udpstat_inc(int statnum)201 kmod_udpstat_inc(int statnum)
202 {
203
204 counter_u64_add(VNET(udpstat)[statnum], 1);
205 }
206
207 #ifdef VIMAGE
208 static void
udp_destroy(void * unused __unused)209 udp_destroy(void *unused __unused)
210 {
211
212 in_pcbinfo_destroy(&V_udbinfo);
213 in_pcbinfo_destroy(&V_ulitecbinfo);
214 }
215 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
216 #endif
217
218 #ifdef INET
219 /*
220 * Subroutine of udp_input(), which appends the provided mbuf chain to the
221 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
222 * contains the source address. If the socket ends up being an IPv6 socket,
223 * udp_append() will convert to a sockaddr_in6 before passing the address
224 * into the socket code.
225 *
226 * In the normal case udp_append() will return 0, indicating that you
227 * must unlock the inp. However if a tunneling protocol is in place we increment
228 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
229 * then decrement the reference count. If the inp_rele returns 1, indicating the
230 * inp is gone, we return that to the caller to tell them *not* to unlock
231 * the inp. In the case of multi-cast this will cause the distribution
232 * to stop (though most tunneling protocols known currently do *not* use
233 * multicast).
234 */
235 static int
udp_append(struct inpcb * inp,struct ip * ip,struct mbuf * n,int off,struct sockaddr_in * udp_in)236 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
237 struct sockaddr_in *udp_in)
238 {
239 struct sockaddr *append_sa;
240 struct socket *so;
241 struct mbuf *tmpopts, *opts = NULL;
242 #ifdef INET6
243 struct sockaddr_in6 udp_in6;
244 #endif
245 struct udpcb *up;
246
247 INP_LOCK_ASSERT(inp);
248
249 /*
250 * Engage the tunneling protocol.
251 */
252 up = intoudpcb(inp);
253 if (up->u_tun_func != NULL) {
254 bool filtered;
255
256 in_pcbref(inp);
257 INP_RUNLOCK(inp);
258 filtered = (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
259 up->u_tun_ctx);
260 INP_RLOCK(inp);
261 if (in_pcbrele_rlocked(inp))
262 return (1);
263 if (filtered) {
264 INP_RUNLOCK(inp);
265 return (1);
266 }
267 }
268
269 off += sizeof(struct udphdr);
270
271 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
272 /* Check AH/ESP integrity. */
273 if (IPSEC_ENABLED(ipv4) &&
274 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
275 m_freem(n);
276 return (0);
277 }
278 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
279 if (IPSEC_ENABLED(ipv4) &&
280 UDPENCAP_INPUT(ipv4, n, off, AF_INET) != 0)
281 return (0); /* Consumed. */
282 }
283 #endif /* IPSEC */
284 #ifdef MAC
285 if (mac_inpcb_check_deliver(inp, n) != 0) {
286 m_freem(n);
287 return (0);
288 }
289 #endif /* MAC */
290 if (inp->inp_flags & INP_CONTROLOPTS ||
291 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
292 #ifdef INET6
293 if (inp->inp_vflag & INP_IPV6)
294 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
295 else
296 #endif /* INET6 */
297 ip_savecontrol(inp, &opts, ip, n);
298 }
299 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
300 tmpopts = sbcreatecontrol(&udp_in[1],
301 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP,
302 M_NOWAIT);
303 if (tmpopts) {
304 if (opts) {
305 tmpopts->m_next = opts;
306 opts = tmpopts;
307 } else
308 opts = tmpopts;
309 }
310 }
311 #ifdef INET6
312 if (inp->inp_vflag & INP_IPV6) {
313 bzero(&udp_in6, sizeof(udp_in6));
314 udp_in6.sin6_len = sizeof(udp_in6);
315 udp_in6.sin6_family = AF_INET6;
316 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
317 append_sa = (struct sockaddr *)&udp_in6;
318 } else
319 #endif /* INET6 */
320 append_sa = (struct sockaddr *)&udp_in[0];
321 m_adj(n, off);
322
323 so = inp->inp_socket;
324 SOCKBUF_LOCK(&so->so_rcv);
325 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
326 soroverflow_locked(so);
327 m_freem(n);
328 if (opts)
329 m_freem(opts);
330 UDPSTAT_INC(udps_fullsock);
331 } else
332 sorwakeup_locked(so);
333 return (0);
334 }
335
336 static bool
udp_multi_match(const struct inpcb * inp,void * v)337 udp_multi_match(const struct inpcb *inp, void *v)
338 {
339 struct ip *ip = v;
340 struct udphdr *uh = (struct udphdr *)(ip + 1);
341
342 if (inp->inp_lport != uh->uh_dport)
343 return (false);
344 #ifdef INET6
345 if ((inp->inp_vflag & INP_IPV4) == 0)
346 return (false);
347 #endif
348 if (inp->inp_laddr.s_addr != INADDR_ANY &&
349 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
350 return (false);
351 if (inp->inp_faddr.s_addr != INADDR_ANY &&
352 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
353 return (false);
354 if (inp->inp_fport != 0 &&
355 inp->inp_fport != uh->uh_sport)
356 return (false);
357
358 return (true);
359 }
360
361 static int
udp_multi_input(struct mbuf * m,int proto,struct sockaddr_in * udp_in)362 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in)
363 {
364 struct ip *ip = mtod(m, struct ip *);
365 struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto),
366 INPLOOKUP_RLOCKPCB, udp_multi_match, ip);
367 #ifdef KDTRACE_HOOKS
368 struct udphdr *uh = (struct udphdr *)(ip + 1);
369 #endif
370 struct inpcb *inp;
371 struct mbuf *n;
372 int appends = 0, fib;
373
374 MPASS(ip->ip_hl == sizeof(struct ip) >> 2);
375
376 fib = M_GETFIB(m);
377
378 while ((inp = inp_next(&inpi)) != NULL) {
379 /*
380 * XXXRW: Because we weren't holding either the inpcb
381 * or the hash lock when we checked for a match
382 * before, we should probably recheck now that the
383 * inpcb lock is held.
384 */
385
386 if (V_udp_bind_all_fibs == 0 && fib != inp->inp_inc.inc_fibnum)
387 /*
388 * Sockets bound to a specific FIB can only receive
389 * packets from that FIB.
390 */
391 continue;
392
393 /*
394 * Handle socket delivery policy for any-source
395 * and source-specific multicast. [RFC3678]
396 */
397 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
398 struct ip_moptions *imo;
399 struct sockaddr_in group;
400 int blocked;
401
402 imo = inp->inp_moptions;
403 if (imo == NULL)
404 continue;
405 bzero(&group, sizeof(struct sockaddr_in));
406 group.sin_len = sizeof(struct sockaddr_in);
407 group.sin_family = AF_INET;
408 group.sin_addr = ip->ip_dst;
409
410 blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif,
411 (struct sockaddr *)&group,
412 (struct sockaddr *)&udp_in[0]);
413 if (blocked != MCAST_PASS) {
414 if (blocked == MCAST_NOTGMEMBER)
415 IPSTAT_INC(ips_notmember);
416 if (blocked == MCAST_NOTSMEMBER ||
417 blocked == MCAST_MUTED)
418 UDPSTAT_INC(udps_filtermcast);
419 continue;
420 }
421 }
422 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) {
423 if (proto == IPPROTO_UDPLITE)
424 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
425 else
426 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
427 if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) {
428 break;
429 } else
430 appends++;
431 }
432 /*
433 * Don't look for additional matches if this one does
434 * not have either the SO_REUSEPORT or SO_REUSEADDR
435 * socket options set. This heuristic avoids
436 * searching through all pcbs in the common case of a
437 * non-shared port. It assumes that an application
438 * will never clear these options after setting them.
439 */
440 if ((inp->inp_socket->so_options &
441 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) {
442 INP_RUNLOCK(inp);
443 break;
444 }
445 }
446
447 if (appends == 0) {
448 /*
449 * No matching pcb found; discard datagram. (No need
450 * to send an ICMP Port Unreachable for a broadcast
451 * or multicast datagram.)
452 */
453 UDPSTAT_INC(udps_noport);
454 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
455 UDPSTAT_INC(udps_noportmcast);
456 else
457 UDPSTAT_INC(udps_noportbcast);
458 }
459 m_freem(m);
460
461 return (IPPROTO_DONE);
462 }
463
464 static int
udp_input(struct mbuf ** mp,int * offp,int proto)465 udp_input(struct mbuf **mp, int *offp, int proto)
466 {
467 struct ip *ip;
468 struct udphdr *uh;
469 struct ifnet *ifp;
470 struct inpcb *inp;
471 uint16_t len, ip_len;
472 struct inpcbinfo *pcbinfo;
473 struct sockaddr_in udp_in[2];
474 struct mbuf *m;
475 struct m_tag *fwd_tag;
476 int cscov_partial, iphlen, lookupflags;
477
478 m = *mp;
479 iphlen = *offp;
480 ifp = m->m_pkthdr.rcvif;
481 *mp = NULL;
482 UDPSTAT_INC(udps_ipackets);
483
484 /*
485 * Strip IP options, if any; should skip this, make available to
486 * user, and use on returned packets, but we don't yet have a way to
487 * check the checksum with options still present.
488 */
489 if (iphlen > sizeof (struct ip)) {
490 ip_stripoptions(m);
491 iphlen = sizeof(struct ip);
492 }
493
494 /*
495 * Get IP and UDP header together in first mbuf.
496 */
497 if (m->m_len < iphlen + sizeof(struct udphdr)) {
498 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
499 UDPSTAT_INC(udps_hdrops);
500 return (IPPROTO_DONE);
501 }
502 }
503 ip = mtod(m, struct ip *);
504 uh = (struct udphdr *)((caddr_t)ip + iphlen);
505 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
506
507 /*
508 * Destination port of 0 is illegal, based on RFC768.
509 */
510 if (uh->uh_dport == 0)
511 goto badunlocked;
512
513 /*
514 * Construct sockaddr format source address. Stuff source address
515 * and datagram in user buffer.
516 */
517 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
518 udp_in[0].sin_len = sizeof(struct sockaddr_in);
519 udp_in[0].sin_family = AF_INET;
520 udp_in[0].sin_port = uh->uh_sport;
521 udp_in[0].sin_addr = ip->ip_src;
522 udp_in[1].sin_len = sizeof(struct sockaddr_in);
523 udp_in[1].sin_family = AF_INET;
524 udp_in[1].sin_port = uh->uh_dport;
525 udp_in[1].sin_addr = ip->ip_dst;
526
527 /*
528 * Make mbuf data length reflect UDP length. If not enough data to
529 * reflect UDP length, drop.
530 */
531 len = ntohs((u_short)uh->uh_ulen);
532 ip_len = ntohs(ip->ip_len) - iphlen;
533 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
534 /* Zero means checksum over the complete packet. */
535 if (len == 0)
536 len = ip_len;
537 cscov_partial = 0;
538 }
539 if (ip_len != len) {
540 if (len > ip_len || len < sizeof(struct udphdr)) {
541 UDPSTAT_INC(udps_badlen);
542 goto badunlocked;
543 }
544 if (proto == IPPROTO_UDP)
545 m_adj(m, len - ip_len);
546 }
547
548 /*
549 * Checksum extended UDP header and data.
550 */
551 if (uh->uh_sum) {
552 u_short uh_sum;
553
554 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
555 !cscov_partial) {
556 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
557 uh_sum = m->m_pkthdr.csum_data;
558 else
559 uh_sum = in_pseudo(ip->ip_src.s_addr,
560 ip->ip_dst.s_addr, htonl((u_short)len +
561 m->m_pkthdr.csum_data + proto));
562 uh_sum ^= 0xffff;
563 } else if (m->m_pkthdr.csum_flags & CSUM_IP_UDP) {
564 /*
565 * Packet from local host (maybe from a VM).
566 * Checksum not required.
567 */
568 uh_sum = 0;
569 } else {
570 char b[offsetof(struct ipovly, ih_src)];
571 struct ipovly *ipov = (struct ipovly *)ip;
572
573 memcpy(b, ipov, sizeof(b));
574 bzero(ipov, sizeof(ipov->ih_x1));
575 ipov->ih_len = (proto == IPPROTO_UDP) ?
576 uh->uh_ulen : htons(ip_len);
577 uh_sum = in_cksum(m, len + sizeof (struct ip));
578 memcpy(ipov, b, sizeof(b));
579 }
580 if (uh_sum) {
581 UDPSTAT_INC(udps_badsum);
582 m_freem(m);
583 return (IPPROTO_DONE);
584 }
585 } else {
586 if (proto == IPPROTO_UDP) {
587 UDPSTAT_INC(udps_nosum);
588 } else {
589 /* UDPLite requires a checksum */
590 /* XXX: What is the right UDPLite MIB counter here? */
591 m_freem(m);
592 return (IPPROTO_DONE);
593 }
594 }
595
596 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
597 in_ifnet_broadcast(ip->ip_dst, ifp))
598 return (udp_multi_input(m, proto, udp_in));
599
600 pcbinfo = udp_get_inpcbinfo(proto);
601
602 /*
603 * Locate pcb for datagram.
604 */
605 lookupflags = INPLOOKUP_RLOCKPCB |
606 (V_udp_bind_all_fibs ? 0 : INPLOOKUP_FIB);
607
608 /*
609 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
610 */
611 if ((m->m_flags & M_IP_NEXTHOP) &&
612 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
613 struct sockaddr_in *next_hop;
614
615 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
616
617 /*
618 * Transparently forwarded. Pretend to be the destination.
619 * Already got one like this?
620 */
621 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
622 ip->ip_dst, uh->uh_dport, lookupflags, ifp, m);
623 if (!inp) {
624 /*
625 * It's new. Try to find the ambushing socket.
626 * Because we've rewritten the destination address,
627 * any hardware-generated hash is ignored.
628 */
629 inp = in_pcblookup(pcbinfo, ip->ip_src,
630 uh->uh_sport, next_hop->sin_addr,
631 next_hop->sin_port ? htons(next_hop->sin_port) :
632 uh->uh_dport, INPLOOKUP_WILDCARD | lookupflags,
633 ifp);
634 }
635 /* Remove the tag from the packet. We don't need it anymore. */
636 m_tag_delete(m, fwd_tag);
637 m->m_flags &= ~M_IP_NEXTHOP;
638 } else
639 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
640 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
641 lookupflags, ifp, m);
642 if (inp == NULL) {
643 if (V_udp_log_in_vain) {
644 char src[INET_ADDRSTRLEN];
645 char dst[INET_ADDRSTRLEN];
646
647 log(LOG_INFO,
648 "Connection attempt to UDP %s:%d from %s:%d\n",
649 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
650 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
651 }
652 if (proto == IPPROTO_UDPLITE)
653 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
654 else
655 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
656 UDPSTAT_INC(udps_noport);
657 if (m->m_flags & M_MCAST) {
658 UDPSTAT_INC(udps_noportmcast);
659 goto badunlocked;
660 }
661 if (m->m_flags & M_BCAST) {
662 UDPSTAT_INC(udps_noportbcast);
663 goto badunlocked;
664 }
665 if (V_udp_blackhole && (V_udp_blackhole_local ||
666 !in_localip(ip->ip_src)))
667 goto badunlocked;
668 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
669 goto badunlocked;
670 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
671 return (IPPROTO_DONE);
672 }
673
674 /*
675 * Check the minimum TTL for socket.
676 */
677 INP_RLOCK_ASSERT(inp);
678 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
679 if (proto == IPPROTO_UDPLITE)
680 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
681 else
682 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
683 INP_RUNLOCK(inp);
684 m_freem(m);
685 return (IPPROTO_DONE);
686 }
687 if (cscov_partial) {
688 struct udpcb *up;
689
690 up = intoudpcb(inp);
691 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
692 INP_RUNLOCK(inp);
693 m_freem(m);
694 return (IPPROTO_DONE);
695 }
696 }
697
698 if (proto == IPPROTO_UDPLITE)
699 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
700 else
701 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
702 if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
703 INP_RUNLOCK(inp);
704 return (IPPROTO_DONE);
705
706 badunlocked:
707 m_freem(m);
708 return (IPPROTO_DONE);
709 }
710 #endif /* INET */
711
712 /*
713 * Notify a udp user of an asynchronous error; just wake up so that they can
714 * collect error status.
715 */
716 struct inpcb *
udp_notify(struct inpcb * inp,int errno)717 udp_notify(struct inpcb *inp, int errno)
718 {
719
720 INP_WLOCK_ASSERT(inp);
721 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
722 errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
723 NH_FREE(inp->inp_route.ro_nh);
724 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
725 }
726
727 inp->inp_socket->so_error = errno;
728 sorwakeup(inp->inp_socket);
729 sowwakeup(inp->inp_socket);
730 return (inp);
731 }
732
733 #ifdef INET
734 static void
udp_common_ctlinput(struct icmp * icmp,struct inpcbinfo * pcbinfo)735 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo)
736 {
737 struct ip *ip = &icmp->icmp_ip;
738 struct udphdr *uh;
739 struct inpcb *inp;
740
741 if (icmp_errmap(icmp) == 0)
742 return;
743
744 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
745 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src,
746 uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
747 if (inp != NULL) {
748 INP_WLOCK_ASSERT(inp);
749 if (inp->inp_socket != NULL)
750 udp_notify(inp, icmp_errmap(icmp));
751 INP_WUNLOCK(inp);
752 } else {
753 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport,
754 ip->ip_src, uh->uh_sport,
755 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
756 if (inp != NULL) {
757 struct udpcb *up;
758 udp_tun_icmp_t *func;
759
760 up = intoudpcb(inp);
761 func = up->u_icmp_func;
762 INP_RUNLOCK(inp);
763 if (func != NULL)
764 func(icmp);
765 }
766 }
767 }
768
769 static void
udp_ctlinput(struct icmp * icmp)770 udp_ctlinput(struct icmp *icmp)
771 {
772
773 return (udp_common_ctlinput(icmp, &V_udbinfo));
774 }
775
776 static void
udplite_ctlinput(struct icmp * icmp)777 udplite_ctlinput(struct icmp *icmp)
778 {
779
780 return (udp_common_ctlinput(icmp, &V_ulitecbinfo));
781 }
782 #endif /* INET */
783
784 static int
udp_pcblist(SYSCTL_HANDLER_ARGS)785 udp_pcblist(SYSCTL_HANDLER_ARGS)
786 {
787 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
788 INPLOOKUP_RLOCKPCB);
789 struct xinpgen xig;
790 struct inpcb *inp;
791 int error;
792
793 if (req->newptr != 0)
794 return (EPERM);
795
796 if (req->oldptr == 0) {
797 int n;
798
799 n = V_udbinfo.ipi_count;
800 n += imax(n / 8, 10);
801 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
802 return (0);
803 }
804
805 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
806 return (error);
807
808 bzero(&xig, sizeof(xig));
809 xig.xig_len = sizeof xig;
810 xig.xig_count = V_udbinfo.ipi_count;
811 xig.xig_gen = V_udbinfo.ipi_gencnt;
812 xig.xig_sogen = so_gencnt;
813 error = SYSCTL_OUT(req, &xig, sizeof xig);
814 if (error)
815 return (error);
816
817 while ((inp = inp_next(&inpi)) != NULL) {
818 if (inp->inp_gencnt <= xig.xig_gen &&
819 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
820 struct xinpcb xi;
821
822 in_pcbtoxinpcb(inp, &xi);
823 error = SYSCTL_OUT(req, &xi, sizeof xi);
824 if (error) {
825 INP_RUNLOCK(inp);
826 break;
827 }
828 }
829 }
830
831 if (!error) {
832 /*
833 * Give the user an updated idea of our state. If the
834 * generation differs from what we told her before, she knows
835 * that something happened while we were processing this
836 * request, and it might be necessary to retry.
837 */
838 xig.xig_gen = V_udbinfo.ipi_gencnt;
839 xig.xig_sogen = so_gencnt;
840 xig.xig_count = V_udbinfo.ipi_count;
841 error = SYSCTL_OUT(req, &xig, sizeof xig);
842 }
843
844 return (error);
845 }
846
847 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
848 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
849 udp_pcblist, "S,xinpcb",
850 "List of active UDP sockets");
851
852 #ifdef INET
853 static int
udp_getcred(SYSCTL_HANDLER_ARGS)854 udp_getcred(SYSCTL_HANDLER_ARGS)
855 {
856 struct xucred xuc;
857 struct sockaddr_in addrs[2];
858 struct epoch_tracker et;
859 struct inpcb *inp;
860 int error;
861
862 if (req->newptr == NULL)
863 return (EINVAL);
864 error = priv_check(req->td, PRIV_NETINET_GETCRED);
865 if (error)
866 return (error);
867 error = SYSCTL_IN(req, addrs, sizeof(addrs));
868 if (error)
869 return (error);
870 NET_EPOCH_ENTER(et);
871 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
872 addrs[0].sin_addr, addrs[0].sin_port,
873 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
874 NET_EPOCH_EXIT(et);
875 if (inp != NULL) {
876 INP_RLOCK_ASSERT(inp);
877 if (inp->inp_socket == NULL)
878 error = ENOENT;
879 if (error == 0)
880 error = cr_canseeinpcb(req->td->td_ucred, inp);
881 if (error == 0)
882 cru2x(inp->inp_cred, &xuc);
883 INP_RUNLOCK(inp);
884 } else
885 error = ENOENT;
886 if (error == 0)
887 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
888 return (error);
889 }
890
891 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
892 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
893 0, 0, udp_getcred, "S,xucred",
894 "Get the xucred of a UDP connection");
895 #endif /* INET */
896
897 int
udp_ctloutput(struct socket * so,struct sockopt * sopt)898 udp_ctloutput(struct socket *so, struct sockopt *sopt)
899 {
900 struct inpcb *inp;
901 struct udpcb *up;
902 int isudplite, error, optval;
903
904 error = 0;
905 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
906 inp = sotoinpcb(so);
907 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
908 INP_WLOCK(inp);
909 if (sopt->sopt_level != so->so_proto->pr_protocol) {
910 #ifdef INET6
911 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
912 INP_WUNLOCK(inp);
913 error = ip6_ctloutput(so, sopt);
914 }
915 #endif
916 #if defined(INET) && defined(INET6)
917 else
918 #endif
919 #ifdef INET
920 {
921 INP_WUNLOCK(inp);
922 error = ip_ctloutput(so, sopt);
923 }
924 #endif
925 return (error);
926 }
927
928 switch (sopt->sopt_dir) {
929 case SOPT_SET:
930 switch (sopt->sopt_name) {
931 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
932 #if defined(INET) || defined(INET6)
933 case UDP_ENCAP:
934 #ifdef INET
935 if (INP_SOCKAF(so) == AF_INET) {
936 if (!IPSEC_ENABLED(ipv4)) {
937 INP_WUNLOCK(inp);
938 return (ENOPROTOOPT);
939 }
940 error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
941 break;
942 }
943 #endif /* INET */
944 #ifdef INET6
945 if (INP_SOCKAF(so) == AF_INET6) {
946 if (!IPSEC_ENABLED(ipv6)) {
947 INP_WUNLOCK(inp);
948 return (ENOPROTOOPT);
949 }
950 error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
951 break;
952 }
953 #endif /* INET6 */
954 INP_WUNLOCK(inp);
955 return (EINVAL);
956 #endif /* INET || INET6 */
957
958 #endif /* IPSEC */
959 case UDPLITE_SEND_CSCOV:
960 case UDPLITE_RECV_CSCOV:
961 if (!isudplite) {
962 INP_WUNLOCK(inp);
963 error = ENOPROTOOPT;
964 break;
965 }
966 INP_WUNLOCK(inp);
967 error = sooptcopyin(sopt, &optval, sizeof(optval),
968 sizeof(optval));
969 if (error != 0)
970 break;
971 inp = sotoinpcb(so);
972 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
973 INP_WLOCK(inp);
974 up = intoudpcb(inp);
975 KASSERT(up != NULL, ("%s: up == NULL", __func__));
976 if ((optval != 0 && optval < 8) || (optval > 65535)) {
977 INP_WUNLOCK(inp);
978 error = EINVAL;
979 break;
980 }
981 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
982 up->u_txcslen = optval;
983 else
984 up->u_rxcslen = optval;
985 INP_WUNLOCK(inp);
986 break;
987 default:
988 INP_WUNLOCK(inp);
989 error = ENOPROTOOPT;
990 break;
991 }
992 break;
993 case SOPT_GET:
994 switch (sopt->sopt_name) {
995 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
996 #if defined(INET) || defined(INET6)
997 case UDP_ENCAP:
998 #ifdef INET
999 if (INP_SOCKAF(so) == AF_INET) {
1000 if (!IPSEC_ENABLED(ipv4)) {
1001 INP_WUNLOCK(inp);
1002 return (ENOPROTOOPT);
1003 }
1004 error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
1005 break;
1006 }
1007 #endif /* INET */
1008 #ifdef INET6
1009 if (INP_SOCKAF(so) == AF_INET6) {
1010 if (!IPSEC_ENABLED(ipv6)) {
1011 INP_WUNLOCK(inp);
1012 return (ENOPROTOOPT);
1013 }
1014 error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
1015 break;
1016 }
1017 #endif /* INET6 */
1018 INP_WUNLOCK(inp);
1019 return (EINVAL);
1020 #endif /* INET || INET6 */
1021
1022 #endif /* IPSEC */
1023 case UDPLITE_SEND_CSCOV:
1024 case UDPLITE_RECV_CSCOV:
1025 if (!isudplite) {
1026 INP_WUNLOCK(inp);
1027 error = ENOPROTOOPT;
1028 break;
1029 }
1030 up = intoudpcb(inp);
1031 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1032 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1033 optval = up->u_txcslen;
1034 else
1035 optval = up->u_rxcslen;
1036 INP_WUNLOCK(inp);
1037 error = sooptcopyout(sopt, &optval, sizeof(optval));
1038 break;
1039 default:
1040 INP_WUNLOCK(inp);
1041 error = ENOPROTOOPT;
1042 break;
1043 }
1044 break;
1045 }
1046 return (error);
1047 }
1048
1049 #ifdef INET
1050 #ifdef INET6
1051 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1052 static int
udp_v4mapped_pktinfo(struct cmsghdr * cm,struct sockaddr_in * src,struct inpcb * inp,int flags)1053 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1054 struct inpcb *inp, int flags)
1055 {
1056 struct ifnet *ifp;
1057 struct in6_pktinfo *pktinfo;
1058 struct in_addr ia;
1059
1060 NET_EPOCH_ASSERT();
1061
1062 if ((flags & PRUS_IPV6) == 0)
1063 return (0);
1064
1065 if (cm->cmsg_level != IPPROTO_IPV6)
1066 return (0);
1067
1068 if (cm->cmsg_type != IPV6_2292PKTINFO &&
1069 cm->cmsg_type != IPV6_PKTINFO)
1070 return (0);
1071
1072 if (cm->cmsg_len !=
1073 CMSG_LEN(sizeof(struct in6_pktinfo)))
1074 return (EINVAL);
1075
1076 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1077 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1078 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1079 return (EINVAL);
1080
1081 /* Validate the interface index if specified. */
1082 if (pktinfo->ipi6_ifindex) {
1083 ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1084 if (ifp == NULL)
1085 return (ENXIO);
1086 } else
1087 ifp = NULL;
1088 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1089 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1090 if (!in_ifhasaddr(ifp, ia))
1091 return (EADDRNOTAVAIL);
1092 }
1093
1094 bzero(src, sizeof(*src));
1095 src->sin_family = AF_INET;
1096 src->sin_len = sizeof(*src);
1097 src->sin_port = inp->inp_lport;
1098 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1099
1100 return (0);
1101 }
1102 #endif /* INET6 */
1103
1104 int
udp_send(struct socket * so,int flags,struct mbuf * m,struct sockaddr * addr,struct mbuf * control,struct thread * td)1105 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1106 struct mbuf *control, struct thread *td)
1107 {
1108 struct inpcb *inp;
1109 struct udpiphdr *ui;
1110 int len, error = 0;
1111 struct in_addr faddr, laddr;
1112 struct cmsghdr *cm;
1113 struct inpcbinfo *pcbinfo;
1114 struct sockaddr_in *sin, src;
1115 struct epoch_tracker et;
1116 int cscov_partial = 0;
1117 int ipflags = 0;
1118 u_short fport, lport;
1119 u_char tos, vflagsav;
1120 uint8_t pr;
1121 uint16_t cscov = 0;
1122 uint32_t flowid = 0;
1123 uint8_t flowtype = M_HASHTYPE_NONE;
1124 bool use_cached_route;
1125
1126 inp = sotoinpcb(so);
1127 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1128
1129 if (addr != NULL) {
1130 if (addr->sa_family != AF_INET)
1131 error = EAFNOSUPPORT;
1132 else if (addr->sa_len != sizeof(struct sockaddr_in))
1133 error = EINVAL;
1134 if (__predict_false(error != 0)) {
1135 m_freem(control);
1136 m_freem(m);
1137 return (error);
1138 }
1139 }
1140
1141 len = m->m_pkthdr.len;
1142 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1143 if (control)
1144 m_freem(control);
1145 m_freem(m);
1146 return (EMSGSIZE);
1147 }
1148
1149 src.sin_family = 0;
1150 sin = (struct sockaddr_in *)addr;
1151
1152 /*
1153 * udp_send() may need to bind the current inpcb. As such, we don't
1154 * know up front whether we will need the pcbinfo lock or not. Do any
1155 * work to decide what is needed up front before acquiring any locks.
1156 *
1157 * We will need network epoch in either case, to safely lookup into
1158 * pcb hash.
1159 */
1160 use_cached_route = sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0);
1161 if (use_cached_route || (flags & PRUS_IPV6) != 0)
1162 INP_WLOCK(inp);
1163 else
1164 INP_RLOCK(inp);
1165 NET_EPOCH_ENTER(et);
1166 tos = inp->inp_ip_tos;
1167 if (control != NULL) {
1168 /*
1169 * XXX: Currently, we assume all the optional information is
1170 * stored in a single mbuf.
1171 */
1172 if (control->m_next) {
1173 m_freem(control);
1174 error = EINVAL;
1175 goto release;
1176 }
1177 for (; control->m_len > 0;
1178 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1179 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1180 cm = mtod(control, struct cmsghdr *);
1181 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1182 || cm->cmsg_len > control->m_len) {
1183 error = EINVAL;
1184 break;
1185 }
1186 #ifdef INET6
1187 error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1188 if (error != 0)
1189 break;
1190 #endif
1191 if (cm->cmsg_level != IPPROTO_IP)
1192 continue;
1193
1194 switch (cm->cmsg_type) {
1195 case IP_SENDSRCADDR:
1196 if (cm->cmsg_len !=
1197 CMSG_LEN(sizeof(struct in_addr))) {
1198 error = EINVAL;
1199 break;
1200 }
1201 bzero(&src, sizeof(src));
1202 src.sin_family = AF_INET;
1203 src.sin_len = sizeof(src);
1204 src.sin_port = inp->inp_lport;
1205 src.sin_addr =
1206 *(struct in_addr *)CMSG_DATA(cm);
1207 break;
1208
1209 case IP_TOS:
1210 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1211 error = EINVAL;
1212 break;
1213 }
1214 tos = *(u_char *)CMSG_DATA(cm);
1215 break;
1216
1217 case IP_FLOWID:
1218 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1219 error = EINVAL;
1220 break;
1221 }
1222 flowid = *(uint32_t *) CMSG_DATA(cm);
1223 break;
1224
1225 case IP_FLOWTYPE:
1226 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1227 error = EINVAL;
1228 break;
1229 }
1230 flowtype = *(uint32_t *) CMSG_DATA(cm);
1231 break;
1232
1233 #ifdef RSS
1234 case IP_RSSBUCKETID:
1235 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1236 error = EINVAL;
1237 break;
1238 }
1239 /* This is just a placeholder for now */
1240 break;
1241 #endif /* RSS */
1242 default:
1243 error = ENOPROTOOPT;
1244 break;
1245 }
1246 if (error)
1247 break;
1248 }
1249 m_freem(control);
1250 control = NULL;
1251 }
1252 if (error)
1253 goto release;
1254
1255 pr = inp->inp_socket->so_proto->pr_protocol;
1256 pcbinfo = udp_get_inpcbinfo(pr);
1257
1258 /*
1259 * If the IP_SENDSRCADDR control message was specified, override the
1260 * source address for this datagram. Its use is invalidated if the
1261 * address thus specified is incomplete or clobbers other inpcbs.
1262 */
1263 laddr = inp->inp_laddr;
1264 lport = inp->inp_lport;
1265 if (src.sin_family == AF_INET) {
1266 if ((lport == 0) ||
1267 (laddr.s_addr == INADDR_ANY &&
1268 src.sin_addr.s_addr == INADDR_ANY)) {
1269 error = EINVAL;
1270 goto release;
1271 }
1272 if ((flags & PRUS_IPV6) != 0) {
1273 vflagsav = inp->inp_vflag;
1274 inp->inp_vflag |= INP_IPV4;
1275 inp->inp_vflag &= ~INP_IPV6;
1276 }
1277 INP_HASH_WLOCK(pcbinfo);
1278 error = in_pcbbind_setup(inp, &src, &laddr.s_addr, &lport,
1279 V_udp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred);
1280 INP_HASH_WUNLOCK(pcbinfo);
1281 if ((flags & PRUS_IPV6) != 0)
1282 inp->inp_vflag = vflagsav;
1283 if (error)
1284 goto release;
1285 }
1286
1287 /*
1288 * If a UDP socket has been connected, then a local address/port will
1289 * have been selected and bound.
1290 *
1291 * If a UDP socket has not been connected to, then an explicit
1292 * destination address must be used, in which case a local
1293 * address/port may not have been selected and bound.
1294 */
1295 if (sin != NULL) {
1296 INP_LOCK_ASSERT(inp);
1297 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1298 error = EISCONN;
1299 goto release;
1300 }
1301
1302 /*
1303 * Jail may rewrite the destination address, so let it do
1304 * that before we use it.
1305 */
1306 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1307 if (error)
1308 goto release;
1309 /*
1310 * sendto(2) on unconnected UDP socket results in implicit
1311 * binding to INADDR_ANY and anonymous port. This has two
1312 * side effects:
1313 * 1) after first sendto(2) the socket will receive datagrams
1314 * destined to the selected port.
1315 * 2) subsequent sendto(2) calls will use the same source port.
1316 */
1317 if (inp->inp_lport == 0) {
1318 struct sockaddr_in wild = {
1319 .sin_family = AF_INET,
1320 .sin_len = sizeof(struct sockaddr_in),
1321 };
1322
1323 INP_HASH_WLOCK(pcbinfo);
1324 error = in_pcbbind(inp, &wild, V_udp_bind_all_fibs ?
1325 0 : INPBIND_FIB, td->td_ucred);
1326 INP_HASH_WUNLOCK(pcbinfo);
1327 if (error)
1328 goto release;
1329 lport = inp->inp_lport;
1330 laddr = inp->inp_laddr;
1331 }
1332 if (laddr.s_addr == INADDR_ANY) {
1333 error = in_pcbladdr(inp, &sin->sin_addr, &laddr,
1334 td->td_ucred);
1335 if (error)
1336 goto release;
1337 }
1338 faddr = sin->sin_addr;
1339 fport = sin->sin_port;
1340 } else {
1341 INP_LOCK_ASSERT(inp);
1342 faddr = inp->inp_faddr;
1343 fport = inp->inp_fport;
1344 if (faddr.s_addr == INADDR_ANY) {
1345 error = ENOTCONN;
1346 goto release;
1347 }
1348 }
1349
1350 /*
1351 * Calculate data length and get a mbuf for UDP, IP, and possible
1352 * link-layer headers. Immediate slide the data pointer back forward
1353 * since we won't use that space at this layer.
1354 */
1355 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1356 if (m == NULL) {
1357 error = ENOBUFS;
1358 goto release;
1359 }
1360 m->m_data += max_linkhdr;
1361 m->m_len -= max_linkhdr;
1362 m->m_pkthdr.len -= max_linkhdr;
1363
1364 /*
1365 * Fill in mbuf with extended UDP header and addresses and length put
1366 * into network format.
1367 */
1368 ui = mtod(m, struct udpiphdr *);
1369 /*
1370 * Filling only those fields of udpiphdr that participate in the
1371 * checksum calculation. The rest must be zeroed and will be filled
1372 * later.
1373 */
1374 bzero(ui->ui_x1, sizeof(ui->ui_x1));
1375 ui->ui_pr = pr;
1376 ui->ui_src = laddr;
1377 ui->ui_dst = faddr;
1378 ui->ui_sport = lport;
1379 ui->ui_dport = fport;
1380 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1381 if (pr == IPPROTO_UDPLITE) {
1382 struct udpcb *up;
1383 uint16_t plen;
1384
1385 up = intoudpcb(inp);
1386 cscov = up->u_txcslen;
1387 plen = (u_short)len + sizeof(struct udphdr);
1388 if (cscov >= plen)
1389 cscov = 0;
1390 ui->ui_len = htons(plen);
1391 ui->ui_ulen = htons(cscov);
1392 /*
1393 * For UDP-Lite, checksum coverage length of zero means
1394 * the entire UDPLite packet is covered by the checksum.
1395 */
1396 cscov_partial = (cscov == 0) ? 0 : 1;
1397 }
1398
1399 if (inp->inp_socket->so_options & SO_DONTROUTE)
1400 ipflags |= IP_ROUTETOIF;
1401 if (inp->inp_socket->so_options & SO_BROADCAST)
1402 ipflags |= IP_ALLOWBROADCAST;
1403 if (inp->inp_flags & INP_ONESBCAST)
1404 ipflags |= IP_SENDONES;
1405
1406 #ifdef MAC
1407 mac_inpcb_create_mbuf(inp, m);
1408 #endif
1409
1410 /*
1411 * Set up checksum and output datagram.
1412 */
1413 ui->ui_sum = 0;
1414 if (pr == IPPROTO_UDPLITE) {
1415 if (inp->inp_flags & INP_ONESBCAST)
1416 faddr.s_addr = INADDR_BROADCAST;
1417 if (cscov_partial) {
1418 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1419 ui->ui_sum = 0xffff;
1420 } else {
1421 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1422 ui->ui_sum = 0xffff;
1423 }
1424 } else if (V_udp_cksum) {
1425 if (inp->inp_flags & INP_ONESBCAST)
1426 faddr.s_addr = INADDR_BROADCAST;
1427 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1428 htons((u_short)len + sizeof(struct udphdr) + pr));
1429 m->m_pkthdr.csum_flags = CSUM_UDP;
1430 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1431 }
1432 /*
1433 * After finishing the checksum computation, fill the remaining fields
1434 * of udpiphdr.
1435 */
1436 ((struct ip *)ui)->ip_v = IPVERSION;
1437 ((struct ip *)ui)->ip_tos = tos;
1438 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1439 if (inp->inp_flags & INP_DONTFRAG)
1440 ((struct ip *)ui)->ip_off |= htons(IP_DF);
1441 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;
1442 UDPSTAT_INC(udps_opackets);
1443
1444 /*
1445 * Setup flowid / RSS information for outbound socket.
1446 *
1447 * Once the UDP code decides to set a flowid some other way,
1448 * this allows the flowid to be overridden by userland.
1449 */
1450 if (flowtype != M_HASHTYPE_NONE) {
1451 m->m_pkthdr.flowid = flowid;
1452 M_HASHTYPE_SET(m, flowtype);
1453 }
1454 #if defined(ROUTE_MPATH) || defined(RSS)
1455 else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1456 uint32_t hash_val, hash_type;
1457
1458 hash_val = fib4_calc_packet_hash(laddr, faddr,
1459 lport, fport, pr, &hash_type);
1460 m->m_pkthdr.flowid = hash_val;
1461 M_HASHTYPE_SET(m, hash_type);
1462 }
1463
1464 /*
1465 * Don't override with the inp cached flowid value.
1466 *
1467 * Depending upon the kind of send being done, the inp
1468 * flowid/flowtype values may actually not be appropriate
1469 * for this particular socket send.
1470 *
1471 * We should either leave the flowid at zero (which is what is
1472 * currently done) or set it to some software generated
1473 * hash value based on the packet contents.
1474 */
1475 ipflags |= IP_NODEFAULTFLOWID;
1476 #endif /* RSS */
1477
1478 if (pr == IPPROTO_UDPLITE)
1479 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1480 else
1481 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1482 error = ip_output(m, inp->inp_options,
1483 use_cached_route ? &inp->inp_route : NULL, ipflags,
1484 inp->inp_moptions, inp);
1485 INP_UNLOCK(inp);
1486 NET_EPOCH_EXIT(et);
1487 return (error);
1488
1489 release:
1490 INP_UNLOCK(inp);
1491 NET_EPOCH_EXIT(et);
1492 m_freem(m);
1493 return (error);
1494 }
1495
1496 void
udp_abort(struct socket * so)1497 udp_abort(struct socket *so)
1498 {
1499 struct inpcb *inp;
1500 struct inpcbinfo *pcbinfo;
1501
1502 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1503 inp = sotoinpcb(so);
1504 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1505 INP_WLOCK(inp);
1506 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1507 INP_HASH_WLOCK(pcbinfo);
1508 in_pcbdisconnect(inp);
1509 INP_HASH_WUNLOCK(pcbinfo);
1510 soisdisconnected(so);
1511 }
1512 INP_WUNLOCK(inp);
1513 }
1514
1515 static int
udp_attach(struct socket * so,int proto,struct thread * td)1516 udp_attach(struct socket *so, int proto, struct thread *td)
1517 {
1518 static uint32_t udp_flowid;
1519 struct inpcbinfo *pcbinfo;
1520 struct inpcb *inp;
1521 struct udpcb *up;
1522 int error;
1523
1524 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1525 inp = sotoinpcb(so);
1526 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1527 error = soreserve(so, udp_sendspace, udp_recvspace);
1528 if (error)
1529 return (error);
1530 error = in_pcballoc(so, pcbinfo);
1531 if (error)
1532 return (error);
1533
1534 inp = sotoinpcb(so);
1535 inp->inp_ip_ttl = V_ip_defttl;
1536 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1537 inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1538 up = intoudpcb(inp);
1539 bzero(&up->u_start_zero, u_zero_size);
1540 INP_WUNLOCK(inp);
1541
1542 return (0);
1543 }
1544 #endif /* INET */
1545
1546 int
udp_set_kernel_tunneling(struct socket * so,udp_tun_func_t f,udp_tun_icmp_t i,void * ctx)1547 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1548 {
1549 struct inpcb *inp;
1550 struct udpcb *up;
1551
1552 KASSERT(so->so_type == SOCK_DGRAM,
1553 ("udp_set_kernel_tunneling: !dgram"));
1554 inp = sotoinpcb(so);
1555 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1556 INP_WLOCK(inp);
1557 up = intoudpcb(inp);
1558 if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1559 (up->u_icmp_func != NULL))) {
1560 INP_WUNLOCK(inp);
1561 return (EBUSY);
1562 }
1563 up->u_tun_func = f;
1564 up->u_icmp_func = i;
1565 up->u_tun_ctx = ctx;
1566 INP_WUNLOCK(inp);
1567 return (0);
1568 }
1569
1570 #ifdef INET
1571 static int
udp_bind(struct socket * so,struct sockaddr * nam,struct thread * td)1572 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1573 {
1574 struct inpcb *inp;
1575 struct inpcbinfo *pcbinfo;
1576 struct sockaddr_in *sinp;
1577 int error;
1578
1579 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1580 inp = sotoinpcb(so);
1581 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1582
1583 sinp = (struct sockaddr_in *)nam;
1584 if (nam->sa_family != AF_INET) {
1585 /*
1586 * Preserve compatibility with old programs.
1587 */
1588 if (nam->sa_family != AF_UNSPEC ||
1589 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1590 sinp->sin_addr.s_addr != INADDR_ANY)
1591 return (EAFNOSUPPORT);
1592 nam->sa_family = AF_INET;
1593 }
1594 if (nam->sa_len != sizeof(struct sockaddr_in))
1595 return (EINVAL);
1596
1597 INP_WLOCK(inp);
1598 INP_HASH_WLOCK(pcbinfo);
1599 error = in_pcbbind(inp, sinp, V_udp_bind_all_fibs ? 0 : INPBIND_FIB,
1600 td->td_ucred);
1601 INP_HASH_WUNLOCK(pcbinfo);
1602 INP_WUNLOCK(inp);
1603 return (error);
1604 }
1605
1606 static void
udp_close(struct socket * so)1607 udp_close(struct socket *so)
1608 {
1609 struct inpcb *inp;
1610 struct inpcbinfo *pcbinfo;
1611
1612 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1613 inp = sotoinpcb(so);
1614 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1615 INP_WLOCK(inp);
1616 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1617 INP_HASH_WLOCK(pcbinfo);
1618 in_pcbdisconnect(inp);
1619 INP_HASH_WUNLOCK(pcbinfo);
1620 soisdisconnected(so);
1621 }
1622 INP_WUNLOCK(inp);
1623 }
1624
1625 static int
udp_connect(struct socket * so,struct sockaddr * nam,struct thread * td)1626 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1627 {
1628 struct epoch_tracker et;
1629 struct inpcb *inp;
1630 struct inpcbinfo *pcbinfo;
1631 struct sockaddr_in *sin;
1632 int error;
1633
1634 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1635 inp = sotoinpcb(so);
1636 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1637
1638 sin = (struct sockaddr_in *)nam;
1639 if (sin->sin_family != AF_INET)
1640 return (EAFNOSUPPORT);
1641 if (sin->sin_len != sizeof(*sin))
1642 return (EINVAL);
1643
1644 INP_WLOCK(inp);
1645 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1646 INP_WUNLOCK(inp);
1647 return (EISCONN);
1648 }
1649 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1650 if (error != 0) {
1651 INP_WUNLOCK(inp);
1652 return (error);
1653 }
1654 NET_EPOCH_ENTER(et);
1655 INP_HASH_WLOCK(pcbinfo);
1656 error = in_pcbconnect(inp, sin, td->td_ucred);
1657 INP_HASH_WUNLOCK(pcbinfo);
1658 NET_EPOCH_EXIT(et);
1659 if (error == 0)
1660 soisconnected(so);
1661 INP_WUNLOCK(inp);
1662 return (error);
1663 }
1664
1665 static void
udp_detach(struct socket * so)1666 udp_detach(struct socket *so)
1667 {
1668 struct inpcb *inp;
1669
1670 inp = sotoinpcb(so);
1671 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1672 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1673 ("udp_detach: not disconnected"));
1674 INP_WLOCK(inp);
1675 in_pcbfree(inp);
1676 }
1677
1678 int
udp_disconnect(struct socket * so)1679 udp_disconnect(struct socket *so)
1680 {
1681 struct inpcb *inp;
1682 struct inpcbinfo *pcbinfo;
1683
1684 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1685 inp = sotoinpcb(so);
1686 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1687 INP_WLOCK(inp);
1688 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1689 INP_WUNLOCK(inp);
1690 return (ENOTCONN);
1691 }
1692 INP_HASH_WLOCK(pcbinfo);
1693 in_pcbdisconnect(inp);
1694 INP_HASH_WUNLOCK(pcbinfo);
1695 SOCK_LOCK(so);
1696 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1697 SOCK_UNLOCK(so);
1698 INP_WUNLOCK(inp);
1699 return (0);
1700 }
1701 #endif /* INET */
1702
1703 int
udp_shutdown(struct socket * so,enum shutdown_how how)1704 udp_shutdown(struct socket *so, enum shutdown_how how)
1705 {
1706 int error;
1707
1708 SOCK_LOCK(so);
1709 if (!(so->so_state & SS_ISCONNECTED))
1710 /*
1711 * POSIX mandates us to just return ENOTCONN when shutdown(2) is
1712 * invoked on a datagram sockets, however historically we would
1713 * actually tear socket down. This is known to be leveraged by
1714 * some applications to unblock process waiting in recv(2) by
1715 * other process that it shares that socket with. Try to meet
1716 * both backward-compatibility and POSIX requirements by forcing
1717 * ENOTCONN but still flushing buffers and performing wakeup(9).
1718 *
1719 * XXXGL: it remains unknown what applications expect this
1720 * behavior and is this isolated to unix/dgram or inet/dgram or
1721 * both. See: D10351, D3039.
1722 */
1723 error = ENOTCONN;
1724 else
1725 error = 0;
1726 SOCK_UNLOCK(so);
1727
1728 switch (how) {
1729 case SHUT_RD:
1730 sorflush(so);
1731 break;
1732 case SHUT_RDWR:
1733 sorflush(so);
1734 /* FALLTHROUGH */
1735 case SHUT_WR:
1736 socantsendmore(so);
1737 }
1738
1739 return (error);
1740 }
1741
1742 #ifdef INET
1743 #define UDP_PROTOSW \
1744 .pr_type = SOCK_DGRAM, \
1745 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH, \
1746 .pr_ctloutput = udp_ctloutput, \
1747 .pr_abort = udp_abort, \
1748 .pr_attach = udp_attach, \
1749 .pr_bind = udp_bind, \
1750 .pr_connect = udp_connect, \
1751 .pr_control = in_control, \
1752 .pr_detach = udp_detach, \
1753 .pr_disconnect = udp_disconnect, \
1754 .pr_peeraddr = in_getpeeraddr, \
1755 .pr_send = udp_send, \
1756 .pr_soreceive = soreceive_dgram, \
1757 .pr_sosend = sosend_dgram, \
1758 .pr_shutdown = udp_shutdown, \
1759 .pr_sockaddr = in_getsockaddr, \
1760 .pr_sosetlabel = in_pcbsosetlabel, \
1761 .pr_close = udp_close
1762
1763 struct protosw udp_protosw = {
1764 .pr_protocol = IPPROTO_UDP,
1765 UDP_PROTOSW
1766 };
1767
1768 struct protosw udplite_protosw = {
1769 .pr_protocol = IPPROTO_UDPLITE,
1770 UDP_PROTOSW
1771 };
1772
1773 static void
udp_init(void * arg __unused)1774 udp_init(void *arg __unused)
1775 {
1776
1777 IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput);
1778 IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput);
1779 }
1780 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
1781 #endif /* INET */
1782