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 datgram.)
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_BCAST | M_MCAST)) {
658 UDPSTAT_INC(udps_noportbcast);
659 goto badunlocked;
660 }
661 if (V_udp_blackhole && (V_udp_blackhole_local ||
662 !in_localip(ip->ip_src)))
663 goto badunlocked;
664 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
665 goto badunlocked;
666 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
667 return (IPPROTO_DONE);
668 }
669
670 /*
671 * Check the minimum TTL for socket.
672 */
673 INP_RLOCK_ASSERT(inp);
674 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
675 if (proto == IPPROTO_UDPLITE)
676 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
677 else
678 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
679 INP_RUNLOCK(inp);
680 m_freem(m);
681 return (IPPROTO_DONE);
682 }
683 if (cscov_partial) {
684 struct udpcb *up;
685
686 up = intoudpcb(inp);
687 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
688 INP_RUNLOCK(inp);
689 m_freem(m);
690 return (IPPROTO_DONE);
691 }
692 }
693
694 if (proto == IPPROTO_UDPLITE)
695 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
696 else
697 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
698 if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
699 INP_RUNLOCK(inp);
700 return (IPPROTO_DONE);
701
702 badunlocked:
703 m_freem(m);
704 return (IPPROTO_DONE);
705 }
706 #endif /* INET */
707
708 /*
709 * Notify a udp user of an asynchronous error; just wake up so that they can
710 * collect error status.
711 */
712 struct inpcb *
udp_notify(struct inpcb * inp,int errno)713 udp_notify(struct inpcb *inp, int errno)
714 {
715
716 INP_WLOCK_ASSERT(inp);
717 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
718 errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
719 NH_FREE(inp->inp_route.ro_nh);
720 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
721 }
722
723 inp->inp_socket->so_error = errno;
724 sorwakeup(inp->inp_socket);
725 sowwakeup(inp->inp_socket);
726 return (inp);
727 }
728
729 #ifdef INET
730 static void
udp_common_ctlinput(struct icmp * icmp,struct inpcbinfo * pcbinfo)731 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo)
732 {
733 struct ip *ip = &icmp->icmp_ip;
734 struct udphdr *uh;
735 struct inpcb *inp;
736
737 if (icmp_errmap(icmp) == 0)
738 return;
739
740 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
741 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src,
742 uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
743 if (inp != NULL) {
744 INP_WLOCK_ASSERT(inp);
745 if (inp->inp_socket != NULL)
746 udp_notify(inp, icmp_errmap(icmp));
747 INP_WUNLOCK(inp);
748 } else {
749 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport,
750 ip->ip_src, uh->uh_sport,
751 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
752 if (inp != NULL) {
753 struct udpcb *up;
754 udp_tun_icmp_t *func;
755
756 up = intoudpcb(inp);
757 func = up->u_icmp_func;
758 INP_RUNLOCK(inp);
759 if (func != NULL)
760 func(icmp);
761 }
762 }
763 }
764
765 static void
udp_ctlinput(struct icmp * icmp)766 udp_ctlinput(struct icmp *icmp)
767 {
768
769 return (udp_common_ctlinput(icmp, &V_udbinfo));
770 }
771
772 static void
udplite_ctlinput(struct icmp * icmp)773 udplite_ctlinput(struct icmp *icmp)
774 {
775
776 return (udp_common_ctlinput(icmp, &V_ulitecbinfo));
777 }
778 #endif /* INET */
779
780 static int
udp_pcblist(SYSCTL_HANDLER_ARGS)781 udp_pcblist(SYSCTL_HANDLER_ARGS)
782 {
783 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo,
784 INPLOOKUP_RLOCKPCB);
785 struct xinpgen xig;
786 struct inpcb *inp;
787 int error;
788
789 if (req->newptr != 0)
790 return (EPERM);
791
792 if (req->oldptr == 0) {
793 int n;
794
795 n = V_udbinfo.ipi_count;
796 n += imax(n / 8, 10);
797 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
798 return (0);
799 }
800
801 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
802 return (error);
803
804 bzero(&xig, sizeof(xig));
805 xig.xig_len = sizeof xig;
806 xig.xig_count = V_udbinfo.ipi_count;
807 xig.xig_gen = V_udbinfo.ipi_gencnt;
808 xig.xig_sogen = so_gencnt;
809 error = SYSCTL_OUT(req, &xig, sizeof xig);
810 if (error)
811 return (error);
812
813 while ((inp = inp_next(&inpi)) != NULL) {
814 if (inp->inp_gencnt <= xig.xig_gen &&
815 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
816 struct xinpcb xi;
817
818 in_pcbtoxinpcb(inp, &xi);
819 error = SYSCTL_OUT(req, &xi, sizeof xi);
820 if (error) {
821 INP_RUNLOCK(inp);
822 break;
823 }
824 }
825 }
826
827 if (!error) {
828 /*
829 * Give the user an updated idea of our state. If the
830 * generation differs from what we told her before, she knows
831 * that something happened while we were processing this
832 * request, and it might be necessary to retry.
833 */
834 xig.xig_gen = V_udbinfo.ipi_gencnt;
835 xig.xig_sogen = so_gencnt;
836 xig.xig_count = V_udbinfo.ipi_count;
837 error = SYSCTL_OUT(req, &xig, sizeof xig);
838 }
839
840 return (error);
841 }
842
843 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
844 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
845 udp_pcblist, "S,xinpcb",
846 "List of active UDP sockets");
847
848 #ifdef INET
849 static int
udp_getcred(SYSCTL_HANDLER_ARGS)850 udp_getcred(SYSCTL_HANDLER_ARGS)
851 {
852 struct xucred xuc;
853 struct sockaddr_in addrs[2];
854 struct epoch_tracker et;
855 struct inpcb *inp;
856 int error;
857
858 if (req->newptr == NULL)
859 return (EINVAL);
860 error = priv_check(req->td, PRIV_NETINET_GETCRED);
861 if (error)
862 return (error);
863 error = SYSCTL_IN(req, addrs, sizeof(addrs));
864 if (error)
865 return (error);
866 NET_EPOCH_ENTER(et);
867 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
868 addrs[0].sin_addr, addrs[0].sin_port,
869 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
870 NET_EPOCH_EXIT(et);
871 if (inp != NULL) {
872 INP_RLOCK_ASSERT(inp);
873 if (inp->inp_socket == NULL)
874 error = ENOENT;
875 if (error == 0)
876 error = cr_canseeinpcb(req->td->td_ucred, inp);
877 if (error == 0)
878 cru2x(inp->inp_cred, &xuc);
879 INP_RUNLOCK(inp);
880 } else
881 error = ENOENT;
882 if (error == 0)
883 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
884 return (error);
885 }
886
887 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
888 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
889 0, 0, udp_getcred, "S,xucred",
890 "Get the xucred of a UDP connection");
891 #endif /* INET */
892
893 int
udp_ctloutput(struct socket * so,struct sockopt * sopt)894 udp_ctloutput(struct socket *so, struct sockopt *sopt)
895 {
896 struct inpcb *inp;
897 struct udpcb *up;
898 int isudplite, error, optval;
899
900 error = 0;
901 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
902 inp = sotoinpcb(so);
903 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
904 INP_WLOCK(inp);
905 if (sopt->sopt_level != so->so_proto->pr_protocol) {
906 #ifdef INET6
907 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
908 INP_WUNLOCK(inp);
909 error = ip6_ctloutput(so, sopt);
910 }
911 #endif
912 #if defined(INET) && defined(INET6)
913 else
914 #endif
915 #ifdef INET
916 {
917 INP_WUNLOCK(inp);
918 error = ip_ctloutput(so, sopt);
919 }
920 #endif
921 return (error);
922 }
923
924 switch (sopt->sopt_dir) {
925 case SOPT_SET:
926 switch (sopt->sopt_name) {
927 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
928 #if defined(INET) || defined(INET6)
929 case UDP_ENCAP:
930 #ifdef INET
931 if (INP_SOCKAF(so) == AF_INET) {
932 if (!IPSEC_ENABLED(ipv4)) {
933 INP_WUNLOCK(inp);
934 return (ENOPROTOOPT);
935 }
936 error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
937 break;
938 }
939 #endif /* INET */
940 #ifdef INET6
941 if (INP_SOCKAF(so) == AF_INET6) {
942 if (!IPSEC_ENABLED(ipv6)) {
943 INP_WUNLOCK(inp);
944 return (ENOPROTOOPT);
945 }
946 error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
947 break;
948 }
949 #endif /* INET6 */
950 INP_WUNLOCK(inp);
951 return (EINVAL);
952 #endif /* INET || INET6 */
953
954 #endif /* IPSEC */
955 case UDPLITE_SEND_CSCOV:
956 case UDPLITE_RECV_CSCOV:
957 if (!isudplite) {
958 INP_WUNLOCK(inp);
959 error = ENOPROTOOPT;
960 break;
961 }
962 INP_WUNLOCK(inp);
963 error = sooptcopyin(sopt, &optval, sizeof(optval),
964 sizeof(optval));
965 if (error != 0)
966 break;
967 inp = sotoinpcb(so);
968 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
969 INP_WLOCK(inp);
970 up = intoudpcb(inp);
971 KASSERT(up != NULL, ("%s: up == NULL", __func__));
972 if ((optval != 0 && optval < 8) || (optval > 65535)) {
973 INP_WUNLOCK(inp);
974 error = EINVAL;
975 break;
976 }
977 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
978 up->u_txcslen = optval;
979 else
980 up->u_rxcslen = optval;
981 INP_WUNLOCK(inp);
982 break;
983 default:
984 INP_WUNLOCK(inp);
985 error = ENOPROTOOPT;
986 break;
987 }
988 break;
989 case SOPT_GET:
990 switch (sopt->sopt_name) {
991 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
992 #if defined(INET) || defined(INET6)
993 case UDP_ENCAP:
994 #ifdef INET
995 if (INP_SOCKAF(so) == AF_INET) {
996 if (!IPSEC_ENABLED(ipv4)) {
997 INP_WUNLOCK(inp);
998 return (ENOPROTOOPT);
999 }
1000 error = UDPENCAP_PCBCTL(ipv4, inp, sopt);
1001 break;
1002 }
1003 #endif /* INET */
1004 #ifdef INET6
1005 if (INP_SOCKAF(so) == AF_INET6) {
1006 if (!IPSEC_ENABLED(ipv6)) {
1007 INP_WUNLOCK(inp);
1008 return (ENOPROTOOPT);
1009 }
1010 error = UDPENCAP_PCBCTL(ipv6, inp, sopt);
1011 break;
1012 }
1013 #endif /* INET6 */
1014 INP_WUNLOCK(inp);
1015 return (EINVAL);
1016 #endif /* INET || INET6 */
1017
1018 #endif /* IPSEC */
1019 case UDPLITE_SEND_CSCOV:
1020 case UDPLITE_RECV_CSCOV:
1021 if (!isudplite) {
1022 INP_WUNLOCK(inp);
1023 error = ENOPROTOOPT;
1024 break;
1025 }
1026 up = intoudpcb(inp);
1027 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1028 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1029 optval = up->u_txcslen;
1030 else
1031 optval = up->u_rxcslen;
1032 INP_WUNLOCK(inp);
1033 error = sooptcopyout(sopt, &optval, sizeof(optval));
1034 break;
1035 default:
1036 INP_WUNLOCK(inp);
1037 error = ENOPROTOOPT;
1038 break;
1039 }
1040 break;
1041 }
1042 return (error);
1043 }
1044
1045 #ifdef INET
1046 #ifdef INET6
1047 /* The logic here is derived from ip6_setpktopt(). See comments there. */
1048 static int
udp_v4mapped_pktinfo(struct cmsghdr * cm,struct sockaddr_in * src,struct inpcb * inp,int flags)1049 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src,
1050 struct inpcb *inp, int flags)
1051 {
1052 struct ifnet *ifp;
1053 struct in6_pktinfo *pktinfo;
1054 struct in_addr ia;
1055
1056 NET_EPOCH_ASSERT();
1057
1058 if ((flags & PRUS_IPV6) == 0)
1059 return (0);
1060
1061 if (cm->cmsg_level != IPPROTO_IPV6)
1062 return (0);
1063
1064 if (cm->cmsg_type != IPV6_2292PKTINFO &&
1065 cm->cmsg_type != IPV6_PKTINFO)
1066 return (0);
1067
1068 if (cm->cmsg_len !=
1069 CMSG_LEN(sizeof(struct in6_pktinfo)))
1070 return (EINVAL);
1071
1072 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm);
1073 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) &&
1074 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr))
1075 return (EINVAL);
1076
1077 /* Validate the interface index if specified. */
1078 if (pktinfo->ipi6_ifindex) {
1079 ifp = ifnet_byindex(pktinfo->ipi6_ifindex);
1080 if (ifp == NULL)
1081 return (ENXIO);
1082 } else
1083 ifp = NULL;
1084 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) {
1085 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1086 if (!in_ifhasaddr(ifp, ia))
1087 return (EADDRNOTAVAIL);
1088 }
1089
1090 bzero(src, sizeof(*src));
1091 src->sin_family = AF_INET;
1092 src->sin_len = sizeof(*src);
1093 src->sin_port = inp->inp_lport;
1094 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3];
1095
1096 return (0);
1097 }
1098 #endif /* INET6 */
1099
1100 int
udp_send(struct socket * so,int flags,struct mbuf * m,struct sockaddr * addr,struct mbuf * control,struct thread * td)1101 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1102 struct mbuf *control, struct thread *td)
1103 {
1104 struct inpcb *inp;
1105 struct udpiphdr *ui;
1106 int len, error = 0;
1107 struct in_addr faddr, laddr;
1108 struct cmsghdr *cm;
1109 struct inpcbinfo *pcbinfo;
1110 struct sockaddr_in *sin, src;
1111 struct epoch_tracker et;
1112 int cscov_partial = 0;
1113 int ipflags = 0;
1114 u_short fport, lport;
1115 u_char tos, vflagsav;
1116 uint8_t pr;
1117 uint16_t cscov = 0;
1118 uint32_t flowid = 0;
1119 uint8_t flowtype = M_HASHTYPE_NONE;
1120 bool use_cached_route;
1121
1122 inp = sotoinpcb(so);
1123 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1124
1125 if (addr != NULL) {
1126 if (addr->sa_family != AF_INET)
1127 error = EAFNOSUPPORT;
1128 else if (addr->sa_len != sizeof(struct sockaddr_in))
1129 error = EINVAL;
1130 if (__predict_false(error != 0)) {
1131 m_freem(control);
1132 m_freem(m);
1133 return (error);
1134 }
1135 }
1136
1137 len = m->m_pkthdr.len;
1138 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1139 if (control)
1140 m_freem(control);
1141 m_freem(m);
1142 return (EMSGSIZE);
1143 }
1144
1145 src.sin_family = 0;
1146 sin = (struct sockaddr_in *)addr;
1147
1148 /*
1149 * udp_send() may need to bind the current inpcb. As such, we don't
1150 * know up front whether we will need the pcbinfo lock or not. Do any
1151 * work to decide what is needed up front before acquiring any locks.
1152 *
1153 * We will need network epoch in either case, to safely lookup into
1154 * pcb hash.
1155 */
1156 use_cached_route = sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0);
1157 if (use_cached_route || (flags & PRUS_IPV6) != 0)
1158 INP_WLOCK(inp);
1159 else
1160 INP_RLOCK(inp);
1161 NET_EPOCH_ENTER(et);
1162 tos = inp->inp_ip_tos;
1163 if (control != NULL) {
1164 /*
1165 * XXX: Currently, we assume all the optional information is
1166 * stored in a single mbuf.
1167 */
1168 if (control->m_next) {
1169 m_freem(control);
1170 error = EINVAL;
1171 goto release;
1172 }
1173 for (; control->m_len > 0;
1174 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1175 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1176 cm = mtod(control, struct cmsghdr *);
1177 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1178 || cm->cmsg_len > control->m_len) {
1179 error = EINVAL;
1180 break;
1181 }
1182 #ifdef INET6
1183 error = udp_v4mapped_pktinfo(cm, &src, inp, flags);
1184 if (error != 0)
1185 break;
1186 #endif
1187 if (cm->cmsg_level != IPPROTO_IP)
1188 continue;
1189
1190 switch (cm->cmsg_type) {
1191 case IP_SENDSRCADDR:
1192 if (cm->cmsg_len !=
1193 CMSG_LEN(sizeof(struct in_addr))) {
1194 error = EINVAL;
1195 break;
1196 }
1197 bzero(&src, sizeof(src));
1198 src.sin_family = AF_INET;
1199 src.sin_len = sizeof(src);
1200 src.sin_port = inp->inp_lport;
1201 src.sin_addr =
1202 *(struct in_addr *)CMSG_DATA(cm);
1203 break;
1204
1205 case IP_TOS:
1206 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1207 error = EINVAL;
1208 break;
1209 }
1210 tos = *(u_char *)CMSG_DATA(cm);
1211 break;
1212
1213 case IP_FLOWID:
1214 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1215 error = EINVAL;
1216 break;
1217 }
1218 flowid = *(uint32_t *) CMSG_DATA(cm);
1219 break;
1220
1221 case IP_FLOWTYPE:
1222 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1223 error = EINVAL;
1224 break;
1225 }
1226 flowtype = *(uint32_t *) CMSG_DATA(cm);
1227 break;
1228
1229 #ifdef RSS
1230 case IP_RSSBUCKETID:
1231 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1232 error = EINVAL;
1233 break;
1234 }
1235 /* This is just a placeholder for now */
1236 break;
1237 #endif /* RSS */
1238 default:
1239 error = ENOPROTOOPT;
1240 break;
1241 }
1242 if (error)
1243 break;
1244 }
1245 m_freem(control);
1246 control = NULL;
1247 }
1248 if (error)
1249 goto release;
1250
1251 pr = inp->inp_socket->so_proto->pr_protocol;
1252 pcbinfo = udp_get_inpcbinfo(pr);
1253
1254 /*
1255 * If the IP_SENDSRCADDR control message was specified, override the
1256 * source address for this datagram. Its use is invalidated if the
1257 * address thus specified is incomplete or clobbers other inpcbs.
1258 */
1259 laddr = inp->inp_laddr;
1260 lport = inp->inp_lport;
1261 if (src.sin_family == AF_INET) {
1262 if ((lport == 0) ||
1263 (laddr.s_addr == INADDR_ANY &&
1264 src.sin_addr.s_addr == INADDR_ANY)) {
1265 error = EINVAL;
1266 goto release;
1267 }
1268 if ((flags & PRUS_IPV6) != 0) {
1269 vflagsav = inp->inp_vflag;
1270 inp->inp_vflag |= INP_IPV4;
1271 inp->inp_vflag &= ~INP_IPV6;
1272 }
1273 INP_HASH_WLOCK(pcbinfo);
1274 error = in_pcbbind_setup(inp, &src, &laddr.s_addr, &lport,
1275 V_udp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred);
1276 INP_HASH_WUNLOCK(pcbinfo);
1277 if ((flags & PRUS_IPV6) != 0)
1278 inp->inp_vflag = vflagsav;
1279 if (error)
1280 goto release;
1281 }
1282
1283 /*
1284 * If a UDP socket has been connected, then a local address/port will
1285 * have been selected and bound.
1286 *
1287 * If a UDP socket has not been connected to, then an explicit
1288 * destination address must be used, in which case a local
1289 * address/port may not have been selected and bound.
1290 */
1291 if (sin != NULL) {
1292 INP_LOCK_ASSERT(inp);
1293 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1294 error = EISCONN;
1295 goto release;
1296 }
1297
1298 /*
1299 * Jail may rewrite the destination address, so let it do
1300 * that before we use it.
1301 */
1302 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1303 if (error)
1304 goto release;
1305 /*
1306 * sendto(2) on unconnected UDP socket results in implicit
1307 * binding to INADDR_ANY and anonymous port. This has two
1308 * side effects:
1309 * 1) after first sendto(2) the socket will receive datagrams
1310 * destined to the selected port.
1311 * 2) subsequent sendto(2) calls will use the same source port.
1312 */
1313 if (inp->inp_lport == 0) {
1314 struct sockaddr_in wild = {
1315 .sin_family = AF_INET,
1316 .sin_len = sizeof(struct sockaddr_in),
1317 };
1318
1319 INP_HASH_WLOCK(pcbinfo);
1320 error = in_pcbbind(inp, &wild, V_udp_bind_all_fibs ?
1321 0 : INPBIND_FIB, td->td_ucred);
1322 INP_HASH_WUNLOCK(pcbinfo);
1323 if (error)
1324 goto release;
1325 lport = inp->inp_lport;
1326 laddr = inp->inp_laddr;
1327 }
1328 if (laddr.s_addr == INADDR_ANY) {
1329 error = in_pcbladdr(inp, &sin->sin_addr, &laddr,
1330 td->td_ucred);
1331 if (error)
1332 goto release;
1333 }
1334 faddr = sin->sin_addr;
1335 fport = sin->sin_port;
1336 } else {
1337 INP_LOCK_ASSERT(inp);
1338 faddr = inp->inp_faddr;
1339 fport = inp->inp_fport;
1340 if (faddr.s_addr == INADDR_ANY) {
1341 error = ENOTCONN;
1342 goto release;
1343 }
1344 }
1345
1346 /*
1347 * Calculate data length and get a mbuf for UDP, IP, and possible
1348 * link-layer headers. Immediate slide the data pointer back forward
1349 * since we won't use that space at this layer.
1350 */
1351 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1352 if (m == NULL) {
1353 error = ENOBUFS;
1354 goto release;
1355 }
1356 m->m_data += max_linkhdr;
1357 m->m_len -= max_linkhdr;
1358 m->m_pkthdr.len -= max_linkhdr;
1359
1360 /*
1361 * Fill in mbuf with extended UDP header and addresses and length put
1362 * into network format.
1363 */
1364 ui = mtod(m, struct udpiphdr *);
1365 /*
1366 * Filling only those fields of udpiphdr that participate in the
1367 * checksum calculation. The rest must be zeroed and will be filled
1368 * later.
1369 */
1370 bzero(ui->ui_x1, sizeof(ui->ui_x1));
1371 ui->ui_pr = pr;
1372 ui->ui_src = laddr;
1373 ui->ui_dst = faddr;
1374 ui->ui_sport = lport;
1375 ui->ui_dport = fport;
1376 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1377 if (pr == IPPROTO_UDPLITE) {
1378 struct udpcb *up;
1379 uint16_t plen;
1380
1381 up = intoudpcb(inp);
1382 cscov = up->u_txcslen;
1383 plen = (u_short)len + sizeof(struct udphdr);
1384 if (cscov >= plen)
1385 cscov = 0;
1386 ui->ui_len = htons(plen);
1387 ui->ui_ulen = htons(cscov);
1388 /*
1389 * For UDP-Lite, checksum coverage length of zero means
1390 * the entire UDPLite packet is covered by the checksum.
1391 */
1392 cscov_partial = (cscov == 0) ? 0 : 1;
1393 }
1394
1395 if (inp->inp_socket->so_options & SO_DONTROUTE)
1396 ipflags |= IP_ROUTETOIF;
1397 if (inp->inp_socket->so_options & SO_BROADCAST)
1398 ipflags |= IP_ALLOWBROADCAST;
1399 if (inp->inp_flags & INP_ONESBCAST)
1400 ipflags |= IP_SENDONES;
1401
1402 #ifdef MAC
1403 mac_inpcb_create_mbuf(inp, m);
1404 #endif
1405
1406 /*
1407 * Set up checksum and output datagram.
1408 */
1409 ui->ui_sum = 0;
1410 if (pr == IPPROTO_UDPLITE) {
1411 if (inp->inp_flags & INP_ONESBCAST)
1412 faddr.s_addr = INADDR_BROADCAST;
1413 if (cscov_partial) {
1414 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1415 ui->ui_sum = 0xffff;
1416 } else {
1417 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1418 ui->ui_sum = 0xffff;
1419 }
1420 } else if (V_udp_cksum) {
1421 if (inp->inp_flags & INP_ONESBCAST)
1422 faddr.s_addr = INADDR_BROADCAST;
1423 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1424 htons((u_short)len + sizeof(struct udphdr) + pr));
1425 m->m_pkthdr.csum_flags = CSUM_UDP;
1426 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1427 }
1428 /*
1429 * After finishing the checksum computation, fill the remaining fields
1430 * of udpiphdr.
1431 */
1432 ((struct ip *)ui)->ip_v = IPVERSION;
1433 ((struct ip *)ui)->ip_tos = tos;
1434 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1435 if (inp->inp_flags & INP_DONTFRAG)
1436 ((struct ip *)ui)->ip_off |= htons(IP_DF);
1437 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;
1438 UDPSTAT_INC(udps_opackets);
1439
1440 /*
1441 * Setup flowid / RSS information for outbound socket.
1442 *
1443 * Once the UDP code decides to set a flowid some other way,
1444 * this allows the flowid to be overridden by userland.
1445 */
1446 if (flowtype != M_HASHTYPE_NONE) {
1447 m->m_pkthdr.flowid = flowid;
1448 M_HASHTYPE_SET(m, flowtype);
1449 }
1450 #if defined(ROUTE_MPATH) || defined(RSS)
1451 else if (CALC_FLOWID_OUTBOUND_SENDTO) {
1452 uint32_t hash_val, hash_type;
1453
1454 hash_val = fib4_calc_packet_hash(laddr, faddr,
1455 lport, fport, pr, &hash_type);
1456 m->m_pkthdr.flowid = hash_val;
1457 M_HASHTYPE_SET(m, hash_type);
1458 }
1459
1460 /*
1461 * Don't override with the inp cached flowid value.
1462 *
1463 * Depending upon the kind of send being done, the inp
1464 * flowid/flowtype values may actually not be appropriate
1465 * for this particular socket send.
1466 *
1467 * We should either leave the flowid at zero (which is what is
1468 * currently done) or set it to some software generated
1469 * hash value based on the packet contents.
1470 */
1471 ipflags |= IP_NODEFAULTFLOWID;
1472 #endif /* RSS */
1473
1474 if (pr == IPPROTO_UDPLITE)
1475 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1476 else
1477 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1478 error = ip_output(m, inp->inp_options,
1479 use_cached_route ? &inp->inp_route : NULL, ipflags,
1480 inp->inp_moptions, inp);
1481 INP_UNLOCK(inp);
1482 NET_EPOCH_EXIT(et);
1483 return (error);
1484
1485 release:
1486 INP_UNLOCK(inp);
1487 NET_EPOCH_EXIT(et);
1488 m_freem(m);
1489 return (error);
1490 }
1491
1492 void
udp_abort(struct socket * so)1493 udp_abort(struct socket *so)
1494 {
1495 struct inpcb *inp;
1496 struct inpcbinfo *pcbinfo;
1497
1498 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1499 inp = sotoinpcb(so);
1500 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1501 INP_WLOCK(inp);
1502 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1503 INP_HASH_WLOCK(pcbinfo);
1504 in_pcbdisconnect(inp);
1505 INP_HASH_WUNLOCK(pcbinfo);
1506 soisdisconnected(so);
1507 }
1508 INP_WUNLOCK(inp);
1509 }
1510
1511 static int
udp_attach(struct socket * so,int proto,struct thread * td)1512 udp_attach(struct socket *so, int proto, struct thread *td)
1513 {
1514 static uint32_t udp_flowid;
1515 struct inpcbinfo *pcbinfo;
1516 struct inpcb *inp;
1517 struct udpcb *up;
1518 int error;
1519
1520 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1521 inp = sotoinpcb(so);
1522 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1523 error = soreserve(so, udp_sendspace, udp_recvspace);
1524 if (error)
1525 return (error);
1526 error = in_pcballoc(so, pcbinfo);
1527 if (error)
1528 return (error);
1529
1530 inp = sotoinpcb(so);
1531 inp->inp_ip_ttl = V_ip_defttl;
1532 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1533 inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1534 up = intoudpcb(inp);
1535 bzero(&up->u_start_zero, u_zero_size);
1536 INP_WUNLOCK(inp);
1537
1538 return (0);
1539 }
1540 #endif /* INET */
1541
1542 int
udp_set_kernel_tunneling(struct socket * so,udp_tun_func_t f,udp_tun_icmp_t i,void * ctx)1543 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1544 {
1545 struct inpcb *inp;
1546 struct udpcb *up;
1547
1548 KASSERT(so->so_type == SOCK_DGRAM,
1549 ("udp_set_kernel_tunneling: !dgram"));
1550 inp = sotoinpcb(so);
1551 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1552 INP_WLOCK(inp);
1553 up = intoudpcb(inp);
1554 if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) ||
1555 (up->u_icmp_func != NULL))) {
1556 INP_WUNLOCK(inp);
1557 return (EBUSY);
1558 }
1559 up->u_tun_func = f;
1560 up->u_icmp_func = i;
1561 up->u_tun_ctx = ctx;
1562 INP_WUNLOCK(inp);
1563 return (0);
1564 }
1565
1566 #ifdef INET
1567 static int
udp_bind(struct socket * so,struct sockaddr * nam,struct thread * td)1568 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1569 {
1570 struct inpcb *inp;
1571 struct inpcbinfo *pcbinfo;
1572 struct sockaddr_in *sinp;
1573 int error;
1574
1575 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1576 inp = sotoinpcb(so);
1577 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1578
1579 sinp = (struct sockaddr_in *)nam;
1580 if (nam->sa_family != AF_INET) {
1581 /*
1582 * Preserve compatibility with old programs.
1583 */
1584 if (nam->sa_family != AF_UNSPEC ||
1585 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
1586 sinp->sin_addr.s_addr != INADDR_ANY)
1587 return (EAFNOSUPPORT);
1588 nam->sa_family = AF_INET;
1589 }
1590 if (nam->sa_len != sizeof(struct sockaddr_in))
1591 return (EINVAL);
1592
1593 INP_WLOCK(inp);
1594 INP_HASH_WLOCK(pcbinfo);
1595 error = in_pcbbind(inp, sinp, V_udp_bind_all_fibs ? 0 : INPBIND_FIB,
1596 td->td_ucred);
1597 INP_HASH_WUNLOCK(pcbinfo);
1598 INP_WUNLOCK(inp);
1599 return (error);
1600 }
1601
1602 static void
udp_close(struct socket * so)1603 udp_close(struct socket *so)
1604 {
1605 struct inpcb *inp;
1606 struct inpcbinfo *pcbinfo;
1607
1608 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1609 inp = sotoinpcb(so);
1610 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1611 INP_WLOCK(inp);
1612 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1613 INP_HASH_WLOCK(pcbinfo);
1614 in_pcbdisconnect(inp);
1615 INP_HASH_WUNLOCK(pcbinfo);
1616 soisdisconnected(so);
1617 }
1618 INP_WUNLOCK(inp);
1619 }
1620
1621 static int
udp_connect(struct socket * so,struct sockaddr * nam,struct thread * td)1622 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1623 {
1624 struct epoch_tracker et;
1625 struct inpcb *inp;
1626 struct inpcbinfo *pcbinfo;
1627 struct sockaddr_in *sin;
1628 int error;
1629
1630 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1631 inp = sotoinpcb(so);
1632 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1633
1634 sin = (struct sockaddr_in *)nam;
1635 if (sin->sin_family != AF_INET)
1636 return (EAFNOSUPPORT);
1637 if (sin->sin_len != sizeof(*sin))
1638 return (EINVAL);
1639
1640 INP_WLOCK(inp);
1641 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1642 INP_WUNLOCK(inp);
1643 return (EISCONN);
1644 }
1645 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1646 if (error != 0) {
1647 INP_WUNLOCK(inp);
1648 return (error);
1649 }
1650 NET_EPOCH_ENTER(et);
1651 INP_HASH_WLOCK(pcbinfo);
1652 error = in_pcbconnect(inp, sin, td->td_ucred);
1653 INP_HASH_WUNLOCK(pcbinfo);
1654 NET_EPOCH_EXIT(et);
1655 if (error == 0)
1656 soisconnected(so);
1657 INP_WUNLOCK(inp);
1658 return (error);
1659 }
1660
1661 static void
udp_detach(struct socket * so)1662 udp_detach(struct socket *so)
1663 {
1664 struct inpcb *inp;
1665
1666 inp = sotoinpcb(so);
1667 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1668 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1669 ("udp_detach: not disconnected"));
1670 INP_WLOCK(inp);
1671 in_pcbfree(inp);
1672 }
1673
1674 int
udp_disconnect(struct socket * so)1675 udp_disconnect(struct socket *so)
1676 {
1677 struct inpcb *inp;
1678 struct inpcbinfo *pcbinfo;
1679
1680 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1681 inp = sotoinpcb(so);
1682 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1683 INP_WLOCK(inp);
1684 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1685 INP_WUNLOCK(inp);
1686 return (ENOTCONN);
1687 }
1688 INP_HASH_WLOCK(pcbinfo);
1689 in_pcbdisconnect(inp);
1690 INP_HASH_WUNLOCK(pcbinfo);
1691 SOCK_LOCK(so);
1692 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1693 SOCK_UNLOCK(so);
1694 INP_WUNLOCK(inp);
1695 return (0);
1696 }
1697 #endif /* INET */
1698
1699 int
udp_shutdown(struct socket * so,enum shutdown_how how)1700 udp_shutdown(struct socket *so, enum shutdown_how how)
1701 {
1702 int error;
1703
1704 SOCK_LOCK(so);
1705 if (!(so->so_state & SS_ISCONNECTED))
1706 /*
1707 * POSIX mandates us to just return ENOTCONN when shutdown(2) is
1708 * invoked on a datagram sockets, however historically we would
1709 * actually tear socket down. This is known to be leveraged by
1710 * some applications to unblock process waiting in recv(2) by
1711 * other process that it shares that socket with. Try to meet
1712 * both backward-compatibility and POSIX requirements by forcing
1713 * ENOTCONN but still flushing buffers and performing wakeup(9).
1714 *
1715 * XXXGL: it remains unknown what applications expect this
1716 * behavior and is this isolated to unix/dgram or inet/dgram or
1717 * both. See: D10351, D3039.
1718 */
1719 error = ENOTCONN;
1720 else
1721 error = 0;
1722 SOCK_UNLOCK(so);
1723
1724 switch (how) {
1725 case SHUT_RD:
1726 sorflush(so);
1727 break;
1728 case SHUT_RDWR:
1729 sorflush(so);
1730 /* FALLTHROUGH */
1731 case SHUT_WR:
1732 socantsendmore(so);
1733 }
1734
1735 return (error);
1736 }
1737
1738 #ifdef INET
1739 #define UDP_PROTOSW \
1740 .pr_type = SOCK_DGRAM, \
1741 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH, \
1742 .pr_ctloutput = udp_ctloutput, \
1743 .pr_abort = udp_abort, \
1744 .pr_attach = udp_attach, \
1745 .pr_bind = udp_bind, \
1746 .pr_connect = udp_connect, \
1747 .pr_control = in_control, \
1748 .pr_detach = udp_detach, \
1749 .pr_disconnect = udp_disconnect, \
1750 .pr_peeraddr = in_getpeeraddr, \
1751 .pr_send = udp_send, \
1752 .pr_soreceive = soreceive_dgram, \
1753 .pr_sosend = sosend_dgram, \
1754 .pr_shutdown = udp_shutdown, \
1755 .pr_sockaddr = in_getsockaddr, \
1756 .pr_sosetlabel = in_pcbsosetlabel, \
1757 .pr_close = udp_close
1758
1759 struct protosw udp_protosw = {
1760 .pr_protocol = IPPROTO_UDP,
1761 UDP_PROTOSW
1762 };
1763
1764 struct protosw udplite_protosw = {
1765 .pr_protocol = IPPROTO_UDPLITE,
1766 UDP_PROTOSW
1767 };
1768
1769 static void
udp_init(void * arg __unused)1770 udp_init(void *arg __unused)
1771 {
1772
1773 IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput);
1774 IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput);
1775 }
1776 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL);
1777 #endif /* INET */
1778