1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California.
4 * Copyright (c) 2008 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Juniper Networks, Inc.
6 * Copyright (c) 2014 Kevin Lo
7 * All rights reserved.
8 *
9 * Portions of this software were developed by Robert N. M. Watson under
10 * contract to Juniper Networks, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD: releng/11.1/sys/netinet/udp_usrreq.c 315514 2017-03-18 22:04:20Z ae $");
41
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
45 #include "opt_rss.h"
46
47 #include <sys/param.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/sdt.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/systm.h>
66
67 #include <vm/uma.h>
68
69 #include <net/if.h>
70 #include <net/if_var.h>
71 #include <net/route.h>
72 #include <net/rss_config.h>
73
74 #include <netinet/in.h>
75 #include <netinet/in_kdtrace.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/in_systm.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip.h>
80 #ifdef INET6
81 #include <netinet/ip6.h>
82 #endif
83 #include <netinet/ip_icmp.h>
84 #include <netinet/icmp_var.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/ip_options.h>
87 #ifdef INET6
88 #include <netinet6/ip6_var.h>
89 #endif
90 #include <netinet/udp.h>
91 #include <netinet/udp_var.h>
92 #include <netinet/udplite.h>
93 #include <netinet/in_rss.h>
94
95 #include <netipsec/ipsec_support.h>
96
97 #include <machine/in_cksum.h>
98
99 #include <security/mac/mac_framework.h>
100
101 /*
102 * UDP and UDP-Lite protocols implementation.
103 * Per RFC 768, August, 1980.
104 * Per RFC 3828, July, 2004.
105 */
106
107 /*
108 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
109 * removes the only data integrity mechanism for packets and malformed
110 * packets that would otherwise be discarded due to bad checksums, and may
111 * cause problems (especially for NFS data blocks).
112 */
113 VNET_DEFINE(int, udp_cksum) = 1;
114 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
115 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
116
117 int udp_log_in_vain = 0;
118 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
119 &udp_log_in_vain, 0, "Log all incoming UDP packets");
120
121 VNET_DEFINE(int, udp_blackhole) = 0;
122 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
123 &VNET_NAME(udp_blackhole), 0,
124 "Do not send port unreachables for refused connects");
125
126 u_long udp_sendspace = 9216; /* really max datagram size */
127 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
128 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
129
130 u_long udp_recvspace = 40 * (1024 +
131 #ifdef INET6
132 sizeof(struct sockaddr_in6)
133 #else
134 sizeof(struct sockaddr_in)
135 #endif
136 ); /* 40 1K datagrams */
137
138 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
139 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
140
141 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
142 VNET_DEFINE(struct inpcbinfo, udbinfo);
143 VNET_DEFINE(struct inpcbhead, ulitecb);
144 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
145 static VNET_DEFINE(uma_zone_t, udpcb_zone);
146 #define V_udpcb_zone VNET(udpcb_zone)
147
148 #ifndef UDBHASHSIZE
149 #define UDBHASHSIZE 128
150 #endif
151
152 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
153 VNET_PCPUSTAT_SYSINIT(udpstat);
154 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
155 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
156
157 #ifdef VIMAGE
158 VNET_PCPUSTAT_SYSUNINIT(udpstat);
159 #endif /* VIMAGE */
160 #ifdef INET
161 static void udp_detach(struct socket *so);
162 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
163 struct mbuf *, struct thread *);
164 #endif
165
166 static void
167 udp_zone_change(void *tag)
168 {
169
170 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
171 uma_zone_set_max(V_udpcb_zone, maxsockets);
172 }
173
174 static int
175 udp_inpcb_init(void *mem, int size, int flags)
176 {
177 struct inpcb *inp;
178
179 inp = mem;
180 INP_LOCK_INIT(inp, "inp", "udpinp");
181 return (0);
182 }
183
184 static int
185 udplite_inpcb_init(void *mem, int size, int flags)
186 {
187 struct inpcb *inp;
188
189 inp = mem;
190 INP_LOCK_INIT(inp, "inp", "udpliteinp");
191 return (0);
192 }
193
194 void
195 udp_init(void)
196 {
197
198 /*
199 * For now default to 2-tuple UDP hashing - until the fragment
200 * reassembly code can also update the flowid.
201 *
202 * Once we can calculate the flowid that way and re-establish
203 * a 4-tuple, flip this to 4-tuple.
204 */
205 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
206 "udp_inpcb", udp_inpcb_init, NULL, 0,
207 IPI_HASHFIELDS_2TUPLE);
208 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
210 uma_zone_set_max(V_udpcb_zone, maxsockets);
211 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
212 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
213 EVENTHANDLER_PRI_ANY);
214 }
215
216 void
217 udplite_init(void)
218 {
219
220 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
221 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
222 0, IPI_HASHFIELDS_2TUPLE);
223 }
224
225 /*
226 * Kernel module interface for updating udpstat. The argument is an index
227 * into udpstat treated as an array of u_long. While this encodes the
228 * general layout of udpstat into the caller, it doesn't encode its location,
229 * so that future changes to add, for example, per-CPU stats support won't
230 * cause binary compatibility problems for kernel modules.
231 */
232 void
233 kmod_udpstat_inc(int statnum)
234 {
235
236 counter_u64_add(VNET(udpstat)[statnum], 1);
237 }
238
239 int
240 udp_newudpcb(struct inpcb *inp)
241 {
242 struct udpcb *up;
243
244 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
245 if (up == NULL)
246 return (ENOBUFS);
247 inp->inp_ppcb = up;
248 return (0);
249 }
250
251 void
252 udp_discardcb(struct udpcb *up)
253 {
254
255 uma_zfree(V_udpcb_zone, up);
256 }
257
258 #ifdef VIMAGE
259 static void
260 udp_destroy(void *unused __unused)
261 {
262
263 in_pcbinfo_destroy(&V_udbinfo);
264 uma_zdestroy(V_udpcb_zone);
265 }
266 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
267
268 static void
269 udplite_destroy(void *unused __unused)
270 {
271
272 in_pcbinfo_destroy(&V_ulitecbinfo);
273 }
274 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
275 NULL);
276 #endif
277
278 #ifdef INET
279 /*
280 * Subroutine of udp_input(), which appends the provided mbuf chain to the
281 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
282 * contains the source address. If the socket ends up being an IPv6 socket,
283 * udp_append() will convert to a sockaddr_in6 before passing the address
284 * into the socket code.
285 *
286 * In the normal case udp_append() will return 0, indicating that you
287 * must unlock the inp. However if a tunneling protocol is in place we increment
288 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
289 * then decrement the reference count. If the inp_rele returns 1, indicating the
290 * inp is gone, we return that to the caller to tell them *not* to unlock
291 * the inp. In the case of multi-cast this will cause the distribution
292 * to stop (though most tunneling protocols known currently do *not* use
293 * multicast).
294 */
295 static int
296 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
297 struct sockaddr_in *udp_in)
298 {
299 struct sockaddr *append_sa;
300 struct socket *so;
301 struct mbuf *opts = NULL;
302 #ifdef INET6
303 struct sockaddr_in6 udp_in6;
304 #endif
305 struct udpcb *up;
306
307 INP_LOCK_ASSERT(inp);
308
309 /*
310 * Engage the tunneling protocol.
311 */
312 up = intoudpcb(inp);
313 if (up->u_tun_func != NULL) {
314 in_pcbref(inp);
315 INP_RUNLOCK(inp);
316 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
317 up->u_tun_ctx);
318 INP_RLOCK(inp);
319 return (in_pcbrele_rlocked(inp));
320 }
321
322 off += sizeof(struct udphdr);
323
324 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
325 /* Check AH/ESP integrity. */
326 if (IPSEC_ENABLED(ipv4) &&
327 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
328 m_freem(n);
329 return (0);
330 }
331 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
332 if (IPSEC_ENABLED(ipv4) &&
333 UDPENCAP_INPUT(n, off, AF_INET) != 0)
334 return (0); /* Consumed. */
335 }
336 #endif /* IPSEC */
337 #ifdef MAC
338 if (mac_inpcb_check_deliver(inp, n) != 0) {
339 m_freem(n);
340 return (0);
341 }
342 #endif /* MAC */
343 if (inp->inp_flags & INP_CONTROLOPTS ||
344 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
345 #ifdef INET6
346 if (inp->inp_vflag & INP_IPV6)
347 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
348 else
349 #endif /* INET6 */
350 ip_savecontrol(inp, &opts, ip, n);
351 }
352 #ifdef INET6
353 if (inp->inp_vflag & INP_IPV6) {
354 bzero(&udp_in6, sizeof(udp_in6));
355 udp_in6.sin6_len = sizeof(udp_in6);
356 udp_in6.sin6_family = AF_INET6;
357 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
358 append_sa = (struct sockaddr *)&udp_in6;
359 } else
360 #endif /* INET6 */
361 append_sa = (struct sockaddr *)udp_in;
362 m_adj(n, off);
363
364 so = inp->inp_socket;
365 SOCKBUF_LOCK(&so->so_rcv);
366 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
367 SOCKBUF_UNLOCK(&so->so_rcv);
368 m_freem(n);
369 if (opts)
370 m_freem(opts);
371 UDPSTAT_INC(udps_fullsock);
372 } else
373 sorwakeup_locked(so);
374 return (0);
375 }
376
377 int
378 udp_input(struct mbuf **mp, int *offp, int proto)
379 {
380 struct ip *ip;
381 struct udphdr *uh;
382 struct ifnet *ifp;
383 struct inpcb *inp;
384 uint16_t len, ip_len;
385 struct inpcbinfo *pcbinfo;
386 struct ip save_ip;
387 struct sockaddr_in udp_in;
388 struct mbuf *m;
389 struct m_tag *fwd_tag;
390 int cscov_partial, iphlen;
391
392 m = *mp;
393 iphlen = *offp;
394 ifp = m->m_pkthdr.rcvif;
395 *mp = NULL;
396 UDPSTAT_INC(udps_ipackets);
397
398 /*
399 * Strip IP options, if any; should skip this, make available to
400 * user, and use on returned packets, but we don't yet have a way to
401 * check the checksum with options still present.
402 */
403 if (iphlen > sizeof (struct ip)) {
404 ip_stripoptions(m);
405 iphlen = sizeof(struct ip);
406 }
407
408 /*
409 * Get IP and UDP header together in first mbuf.
410 */
411 ip = mtod(m, struct ip *);
412 if (m->m_len < iphlen + sizeof(struct udphdr)) {
413 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
414 UDPSTAT_INC(udps_hdrops);
415 return (IPPROTO_DONE);
416 }
417 ip = mtod(m, struct ip *);
418 }
419 uh = (struct udphdr *)((caddr_t)ip + iphlen);
420 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
421
422 /*
423 * Destination port of 0 is illegal, based on RFC768.
424 */
425 if (uh->uh_dport == 0)
426 goto badunlocked;
427
428 /*
429 * Construct sockaddr format source address. Stuff source address
430 * and datagram in user buffer.
431 */
432 bzero(&udp_in, sizeof(udp_in));
433 udp_in.sin_len = sizeof(udp_in);
434 udp_in.sin_family = AF_INET;
435 udp_in.sin_port = uh->uh_sport;
436 udp_in.sin_addr = ip->ip_src;
437
438 /*
439 * Make mbuf data length reflect UDP length. If not enough data to
440 * reflect UDP length, drop.
441 */
442 len = ntohs((u_short)uh->uh_ulen);
443 ip_len = ntohs(ip->ip_len) - iphlen;
444 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
445 /* Zero means checksum over the complete packet. */
446 if (len == 0)
447 len = ip_len;
448 cscov_partial = 0;
449 }
450 if (ip_len != len) {
451 if (len > ip_len || len < sizeof(struct udphdr)) {
452 UDPSTAT_INC(udps_badlen);
453 goto badunlocked;
454 }
455 if (proto == IPPROTO_UDP)
456 m_adj(m, len - ip_len);
457 }
458
459 /*
460 * Save a copy of the IP header in case we want restore it for
461 * sending an ICMP error message in response.
462 */
463 if (!V_udp_blackhole)
464 save_ip = *ip;
465 else
466 memset(&save_ip, 0, sizeof(save_ip));
467
468 /*
469 * Checksum extended UDP header and data.
470 */
471 if (uh->uh_sum) {
472 u_short uh_sum;
473
474 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
475 !cscov_partial) {
476 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
477 uh_sum = m->m_pkthdr.csum_data;
478 else
479 uh_sum = in_pseudo(ip->ip_src.s_addr,
480 ip->ip_dst.s_addr, htonl((u_short)len +
481 m->m_pkthdr.csum_data + proto));
482 uh_sum ^= 0xffff;
483 } else {
484 char b[9];
485
486 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
487 bzero(((struct ipovly *)ip)->ih_x1, 9);
488 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
489 uh->uh_ulen : htons(ip_len);
490 uh_sum = in_cksum(m, len + sizeof (struct ip));
491 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
492 }
493 if (uh_sum) {
494 UDPSTAT_INC(udps_badsum);
495 m_freem(m);
496 return (IPPROTO_DONE);
497 }
498 } else {
499 if (proto == IPPROTO_UDP) {
500 UDPSTAT_INC(udps_nosum);
501 } else {
502 /* UDPLite requires a checksum */
503 /* XXX: What is the right UDPLite MIB counter here? */
504 m_freem(m);
505 return (IPPROTO_DONE);
506 }
507 }
508
509 pcbinfo = udp_get_inpcbinfo(proto);
510 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
511 in_broadcast(ip->ip_dst, ifp)) {
512 struct inpcb *last;
513 struct inpcbhead *pcblist;
514 struct ip_moptions *imo;
515
516 INP_INFO_RLOCK(pcbinfo);
517 pcblist = udp_get_pcblist(proto);
518 last = NULL;
519 LIST_FOREACH(inp, pcblist, inp_list) {
520 if (inp->inp_lport != uh->uh_dport)
521 continue;
522 #ifdef INET6
523 if ((inp->inp_vflag & INP_IPV4) == 0)
524 continue;
525 #endif
526 if (inp->inp_laddr.s_addr != INADDR_ANY &&
527 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
528 continue;
529 if (inp->inp_faddr.s_addr != INADDR_ANY &&
530 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
531 continue;
532 if (inp->inp_fport != 0 &&
533 inp->inp_fport != uh->uh_sport)
534 continue;
535
536 INP_RLOCK(inp);
537
538 /*
539 * XXXRW: Because we weren't holding either the inpcb
540 * or the hash lock when we checked for a match
541 * before, we should probably recheck now that the
542 * inpcb lock is held.
543 */
544
545 /*
546 * Handle socket delivery policy for any-source
547 * and source-specific multicast. [RFC3678]
548 */
549 imo = inp->inp_moptions;
550 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
551 struct sockaddr_in group;
552 int blocked;
553 if (imo == NULL) {
554 INP_RUNLOCK(inp);
555 continue;
556 }
557 bzero(&group, sizeof(struct sockaddr_in));
558 group.sin_len = sizeof(struct sockaddr_in);
559 group.sin_family = AF_INET;
560 group.sin_addr = ip->ip_dst;
561
562 blocked = imo_multi_filter(imo, ifp,
563 (struct sockaddr *)&group,
564 (struct sockaddr *)&udp_in);
565 if (blocked != MCAST_PASS) {
566 if (blocked == MCAST_NOTGMEMBER)
567 IPSTAT_INC(ips_notmember);
568 if (blocked == MCAST_NOTSMEMBER ||
569 blocked == MCAST_MUTED)
570 UDPSTAT_INC(udps_filtermcast);
571 INP_RUNLOCK(inp);
572 continue;
573 }
574 }
575 if (last != NULL) {
576 struct mbuf *n;
577
578 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
579 UDP_PROBE(receive, NULL, last, ip,
580 last, uh);
581 if (udp_append(last, ip, n, iphlen,
582 &udp_in)) {
583 goto inp_lost;
584 }
585 }
586 INP_RUNLOCK(last);
587 }
588 last = inp;
589 /*
590 * Don't look for additional matches if this one does
591 * not have either the SO_REUSEPORT or SO_REUSEADDR
592 * socket options set. This heuristic avoids
593 * searching through all pcbs in the common case of a
594 * non-shared port. It assumes that an application
595 * will never clear these options after setting them.
596 */
597 if ((last->inp_socket->so_options &
598 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
599 break;
600 }
601
602 if (last == NULL) {
603 /*
604 * No matching pcb found; discard datagram. (No need
605 * to send an ICMP Port Unreachable for a broadcast
606 * or multicast datgram.)
607 */
608 UDPSTAT_INC(udps_noportbcast);
609 if (inp)
610 INP_RUNLOCK(inp);
611 INP_INFO_RUNLOCK(pcbinfo);
612 goto badunlocked;
613 }
614 UDP_PROBE(receive, NULL, last, ip, last, uh);
615 if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
616 INP_RUNLOCK(last);
617 inp_lost:
618 INP_INFO_RUNLOCK(pcbinfo);
619 return (IPPROTO_DONE);
620 }
621
622 /*
623 * Locate pcb for datagram.
624 */
625
626 /*
627 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
628 */
629 if ((m->m_flags & M_IP_NEXTHOP) &&
630 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
631 struct sockaddr_in *next_hop;
632
633 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
634
635 /*
636 * Transparently forwarded. Pretend to be the destination.
637 * Already got one like this?
638 */
639 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
640 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
641 if (!inp) {
642 /*
643 * It's new. Try to find the ambushing socket.
644 * Because we've rewritten the destination address,
645 * any hardware-generated hash is ignored.
646 */
647 inp = in_pcblookup(pcbinfo, ip->ip_src,
648 uh->uh_sport, next_hop->sin_addr,
649 next_hop->sin_port ? htons(next_hop->sin_port) :
650 uh->uh_dport, INPLOOKUP_WILDCARD |
651 INPLOOKUP_RLOCKPCB, ifp);
652 }
653 /* Remove the tag from the packet. We don't need it anymore. */
654 m_tag_delete(m, fwd_tag);
655 m->m_flags &= ~M_IP_NEXTHOP;
656 } else
657 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
658 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
659 INPLOOKUP_RLOCKPCB, ifp, m);
660 if (inp == NULL) {
661 if (udp_log_in_vain) {
662 char src[INET_ADDRSTRLEN];
663 char dst[INET_ADDRSTRLEN];
664
665 log(LOG_INFO,
666 "Connection attempt to UDP %s:%d from %s:%d\n",
667 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
668 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
669 }
670 UDPSTAT_INC(udps_noport);
671 if (m->m_flags & (M_BCAST | M_MCAST)) {
672 UDPSTAT_INC(udps_noportbcast);
673 goto badunlocked;
674 }
675 if (V_udp_blackhole)
676 goto badunlocked;
677 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
678 goto badunlocked;
679 *ip = save_ip;
680 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
681 return (IPPROTO_DONE);
682 }
683
684 /*
685 * Check the minimum TTL for socket.
686 */
687 INP_RLOCK_ASSERT(inp);
688 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
689 INP_RUNLOCK(inp);
690 m_freem(m);
691 return (IPPROTO_DONE);
692 }
693 if (cscov_partial) {
694 struct udpcb *up;
695
696 up = intoudpcb(inp);
697 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
698 INP_RUNLOCK(inp);
699 m_freem(m);
700 return (IPPROTO_DONE);
701 }
702 }
703
704 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
705 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
706 INP_RUNLOCK(inp);
707 return (IPPROTO_DONE);
708
709 badunlocked:
710 m_freem(m);
711 return (IPPROTO_DONE);
712 }
713 #endif /* INET */
714
715 /*
716 * Notify a udp user of an asynchronous error; just wake up so that they can
717 * collect error status.
718 */
719 struct inpcb *
720 udp_notify(struct inpcb *inp, int errno)
721 {
722
723 /*
724 * While udp_ctlinput() always calls udp_notify() with a read lock
725 * when invoking it directly, in_pcbnotifyall() currently uses write
726 * locks due to sharing code with TCP. For now, accept either a read
727 * or a write lock, but a read lock is sufficient.
728 */
729 INP_LOCK_ASSERT(inp);
730 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
731 errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
732 RTFREE(inp->inp_route.ro_rt);
733 inp->inp_route.ro_rt = (struct rtentry *)NULL;
734 }
735
736 inp->inp_socket->so_error = errno;
737 sorwakeup(inp->inp_socket);
738 sowwakeup(inp->inp_socket);
739 return (inp);
740 }
741
742 #ifdef INET
743 static void
744 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
745 struct inpcbinfo *pcbinfo)
746 {
747 struct ip *ip = vip;
748 struct udphdr *uh;
749 struct in_addr faddr;
750 struct inpcb *inp;
751
752 faddr = ((struct sockaddr_in *)sa)->sin_addr;
753 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
754 return;
755
756 if (PRC_IS_REDIRECT(cmd)) {
757 /* signal EHOSTDOWN, as it flushes the cached route */
758 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
759 return;
760 }
761
762 /*
763 * Hostdead is ugly because it goes linearly through all PCBs.
764 *
765 * XXX: We never get this from ICMP, otherwise it makes an excellent
766 * DoS attack on machines with many connections.
767 */
768 if (cmd == PRC_HOSTDEAD)
769 ip = NULL;
770 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
771 return;
772 if (ip != NULL) {
773 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
774 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
775 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
776 if (inp != NULL) {
777 INP_RLOCK_ASSERT(inp);
778 if (inp->inp_socket != NULL) {
779 udp_notify(inp, inetctlerrmap[cmd]);
780 }
781 INP_RUNLOCK(inp);
782 } else {
783 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
784 ip->ip_src, uh->uh_sport,
785 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
786 if (inp != NULL) {
787 struct udpcb *up;
788
789 up = intoudpcb(inp);
790 if (up->u_icmp_func != NULL) {
791 INP_RUNLOCK(inp);
792 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
793 } else {
794 INP_RUNLOCK(inp);
795 }
796 }
797 }
798 } else
799 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
800 udp_notify);
801 }
802 void
803 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
804 {
805
806 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
807 }
808
809 void
810 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
811 {
812
813 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
814 }
815 #endif /* INET */
816
817 static int
818 udp_pcblist(SYSCTL_HANDLER_ARGS)
819 {
820 int error, i, n;
821 struct inpcb *inp, **inp_list;
822 inp_gen_t gencnt;
823 struct xinpgen xig;
824
825 /*
826 * The process of preparing the PCB list is too time-consuming and
827 * resource-intensive to repeat twice on every request.
828 */
829 if (req->oldptr == 0) {
830 n = V_udbinfo.ipi_count;
831 n += imax(n / 8, 10);
832 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
833 return (0);
834 }
835
836 if (req->newptr != 0)
837 return (EPERM);
838
839 /*
840 * OK, now we're committed to doing something.
841 */
842 INP_INFO_RLOCK(&V_udbinfo);
843 gencnt = V_udbinfo.ipi_gencnt;
844 n = V_udbinfo.ipi_count;
845 INP_INFO_RUNLOCK(&V_udbinfo);
846
847 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
848 + n * sizeof(struct xinpcb));
849 if (error != 0)
850 return (error);
851
852 xig.xig_len = sizeof xig;
853 xig.xig_count = n;
854 xig.xig_gen = gencnt;
855 xig.xig_sogen = so_gencnt;
856 error = SYSCTL_OUT(req, &xig, sizeof xig);
857 if (error)
858 return (error);
859
860 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
861 if (inp_list == NULL)
862 return (ENOMEM);
863
864 INP_INFO_RLOCK(&V_udbinfo);
865 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
866 inp = LIST_NEXT(inp, inp_list)) {
867 INP_WLOCK(inp);
868 if (inp->inp_gencnt <= gencnt &&
869 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
870 in_pcbref(inp);
871 inp_list[i++] = inp;
872 }
873 INP_WUNLOCK(inp);
874 }
875 INP_INFO_RUNLOCK(&V_udbinfo);
876 n = i;
877
878 error = 0;
879 for (i = 0; i < n; i++) {
880 inp = inp_list[i];
881 INP_RLOCK(inp);
882 if (inp->inp_gencnt <= gencnt) {
883 struct xinpcb xi;
884
885 bzero(&xi, sizeof(xi));
886 xi.xi_len = sizeof xi;
887 /* XXX should avoid extra copy */
888 bcopy(inp, &xi.xi_inp, sizeof *inp);
889 if (inp->inp_socket)
890 sotoxsocket(inp->inp_socket, &xi.xi_socket);
891 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
892 INP_RUNLOCK(inp);
893 error = SYSCTL_OUT(req, &xi, sizeof xi);
894 } else
895 INP_RUNLOCK(inp);
896 }
897 INP_INFO_WLOCK(&V_udbinfo);
898 for (i = 0; i < n; i++) {
899 inp = inp_list[i];
900 INP_RLOCK(inp);
901 if (!in_pcbrele_rlocked(inp))
902 INP_RUNLOCK(inp);
903 }
904 INP_INFO_WUNLOCK(&V_udbinfo);
905
906 if (!error) {
907 /*
908 * Give the user an updated idea of our state. If the
909 * generation differs from what we told her before, she knows
910 * that something happened while we were processing this
911 * request, and it might be necessary to retry.
912 */
913 INP_INFO_RLOCK(&V_udbinfo);
914 xig.xig_gen = V_udbinfo.ipi_gencnt;
915 xig.xig_sogen = so_gencnt;
916 xig.xig_count = V_udbinfo.ipi_count;
917 INP_INFO_RUNLOCK(&V_udbinfo);
918 error = SYSCTL_OUT(req, &xig, sizeof xig);
919 }
920 free(inp_list, M_TEMP);
921 return (error);
922 }
923
924 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
925 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
926 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
927
928 #ifdef INET
929 static int
930 udp_getcred(SYSCTL_HANDLER_ARGS)
931 {
932 struct xucred xuc;
933 struct sockaddr_in addrs[2];
934 struct inpcb *inp;
935 int error;
936
937 error = priv_check(req->td, PRIV_NETINET_GETCRED);
938 if (error)
939 return (error);
940 error = SYSCTL_IN(req, addrs, sizeof(addrs));
941 if (error)
942 return (error);
943 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
944 addrs[0].sin_addr, addrs[0].sin_port,
945 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
946 if (inp != NULL) {
947 INP_RLOCK_ASSERT(inp);
948 if (inp->inp_socket == NULL)
949 error = ENOENT;
950 if (error == 0)
951 error = cr_canseeinpcb(req->td->td_ucred, inp);
952 if (error == 0)
953 cru2x(inp->inp_cred, &xuc);
954 INP_RUNLOCK(inp);
955 } else
956 error = ENOENT;
957 if (error == 0)
958 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
959 return (error);
960 }
961
962 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
963 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
964 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
965 #endif /* INET */
966
967 int
968 udp_ctloutput(struct socket *so, struct sockopt *sopt)
969 {
970 struct inpcb *inp;
971 struct udpcb *up;
972 int isudplite, error, optval;
973
974 error = 0;
975 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
976 inp = sotoinpcb(so);
977 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
978 INP_WLOCK(inp);
979 if (sopt->sopt_level != so->so_proto->pr_protocol) {
980 #ifdef INET6
981 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
982 INP_WUNLOCK(inp);
983 error = ip6_ctloutput(so, sopt);
984 }
985 #endif
986 #if defined(INET) && defined(INET6)
987 else
988 #endif
989 #ifdef INET
990 {
991 INP_WUNLOCK(inp);
992 error = ip_ctloutput(so, sopt);
993 }
994 #endif
995 return (error);
996 }
997
998 switch (sopt->sopt_dir) {
999 case SOPT_SET:
1000 switch (sopt->sopt_name) {
1001 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1002 #ifdef INET
1003 case UDP_ENCAP:
1004 if (!IPSEC_ENABLED(ipv4)) {
1005 INP_WUNLOCK(inp);
1006 return (ENOPROTOOPT);
1007 }
1008 error = UDPENCAP_PCBCTL(inp, sopt);
1009 break;
1010 #endif /* INET */
1011 #endif /* IPSEC */
1012 case UDPLITE_SEND_CSCOV:
1013 case UDPLITE_RECV_CSCOV:
1014 if (!isudplite) {
1015 INP_WUNLOCK(inp);
1016 error = ENOPROTOOPT;
1017 break;
1018 }
1019 INP_WUNLOCK(inp);
1020 error = sooptcopyin(sopt, &optval, sizeof(optval),
1021 sizeof(optval));
1022 if (error != 0)
1023 break;
1024 inp = sotoinpcb(so);
1025 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1026 INP_WLOCK(inp);
1027 up = intoudpcb(inp);
1028 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1029 if ((optval != 0 && optval < 8) || (optval > 65535)) {
1030 INP_WUNLOCK(inp);
1031 error = EINVAL;
1032 break;
1033 }
1034 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1035 up->u_txcslen = optval;
1036 else
1037 up->u_rxcslen = optval;
1038 INP_WUNLOCK(inp);
1039 break;
1040 default:
1041 INP_WUNLOCK(inp);
1042 error = ENOPROTOOPT;
1043 break;
1044 }
1045 break;
1046 case SOPT_GET:
1047 switch (sopt->sopt_name) {
1048 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1049 #ifdef INET
1050 case UDP_ENCAP:
1051 if (!IPSEC_ENABLED(ipv4)) {
1052 INP_WUNLOCK(inp);
1053 return (ENOPROTOOPT);
1054 }
1055 error = UDPENCAP_PCBCTL(inp, sopt);
1056 break;
1057 #endif /* INET */
1058 #endif /* IPSEC */
1059 case UDPLITE_SEND_CSCOV:
1060 case UDPLITE_RECV_CSCOV:
1061 if (!isudplite) {
1062 INP_WUNLOCK(inp);
1063 error = ENOPROTOOPT;
1064 break;
1065 }
1066 up = intoudpcb(inp);
1067 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1068 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1069 optval = up->u_txcslen;
1070 else
1071 optval = up->u_rxcslen;
1072 INP_WUNLOCK(inp);
1073 error = sooptcopyout(sopt, &optval, sizeof(optval));
1074 break;
1075 default:
1076 INP_WUNLOCK(inp);
1077 error = ENOPROTOOPT;
1078 break;
1079 }
1080 break;
1081 }
1082 return (error);
1083 }
1084
1085 #ifdef INET
1086 #define UH_WLOCKED 2
1087 #define UH_RLOCKED 1
1088 #define UH_UNLOCKED 0
1089 static int
1090 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1091 struct mbuf *control, struct thread *td)
1092 {
1093 struct udpiphdr *ui;
1094 int len = m->m_pkthdr.len;
1095 struct in_addr faddr, laddr;
1096 struct cmsghdr *cm;
1097 struct inpcbinfo *pcbinfo;
1098 struct sockaddr_in *sin, src;
1099 int cscov_partial = 0;
1100 int error = 0;
1101 int ipflags;
1102 u_short fport, lport;
1103 int unlock_udbinfo, unlock_inp;
1104 u_char tos;
1105 uint8_t pr;
1106 uint16_t cscov = 0;
1107 uint32_t flowid = 0;
1108 uint8_t flowtype = M_HASHTYPE_NONE;
1109
1110 /*
1111 * udp_output() may need to temporarily bind or connect the current
1112 * inpcb. As such, we don't know up front whether we will need the
1113 * pcbinfo lock or not. Do any work to decide what is needed up
1114 * front before acquiring any locks.
1115 */
1116 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1117 if (control)
1118 m_freem(control);
1119 m_freem(m);
1120 return (EMSGSIZE);
1121 }
1122
1123 src.sin_family = 0;
1124 sin = (struct sockaddr_in *)addr;
1125 if (sin == NULL ||
1126 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1127 INP_WLOCK(inp);
1128 unlock_inp = UH_WLOCKED;
1129 } else {
1130 INP_RLOCK(inp);
1131 unlock_inp = UH_RLOCKED;
1132 }
1133 tos = inp->inp_ip_tos;
1134 if (control != NULL) {
1135 /*
1136 * XXX: Currently, we assume all the optional information is
1137 * stored in a single mbuf.
1138 */
1139 if (control->m_next) {
1140 if (unlock_inp == UH_WLOCKED)
1141 INP_WUNLOCK(inp);
1142 else
1143 INP_RUNLOCK(inp);
1144 m_freem(control);
1145 m_freem(m);
1146 return (EINVAL);
1147 }
1148 for (; control->m_len > 0;
1149 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1150 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1151 cm = mtod(control, struct cmsghdr *);
1152 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1153 || cm->cmsg_len > control->m_len) {
1154 error = EINVAL;
1155 break;
1156 }
1157 if (cm->cmsg_level != IPPROTO_IP)
1158 continue;
1159
1160 switch (cm->cmsg_type) {
1161 case IP_SENDSRCADDR:
1162 if (cm->cmsg_len !=
1163 CMSG_LEN(sizeof(struct in_addr))) {
1164 error = EINVAL;
1165 break;
1166 }
1167 bzero(&src, sizeof(src));
1168 src.sin_family = AF_INET;
1169 src.sin_len = sizeof(src);
1170 src.sin_port = inp->inp_lport;
1171 src.sin_addr =
1172 *(struct in_addr *)CMSG_DATA(cm);
1173 break;
1174
1175 case IP_TOS:
1176 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1177 error = EINVAL;
1178 break;
1179 }
1180 tos = *(u_char *)CMSG_DATA(cm);
1181 break;
1182
1183 case IP_FLOWID:
1184 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1185 error = EINVAL;
1186 break;
1187 }
1188 flowid = *(uint32_t *) CMSG_DATA(cm);
1189 break;
1190
1191 case IP_FLOWTYPE:
1192 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1193 error = EINVAL;
1194 break;
1195 }
1196 flowtype = *(uint32_t *) CMSG_DATA(cm);
1197 break;
1198
1199 #ifdef RSS
1200 case IP_RSSBUCKETID:
1201 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1202 error = EINVAL;
1203 break;
1204 }
1205 /* This is just a placeholder for now */
1206 break;
1207 #endif /* RSS */
1208 default:
1209 error = ENOPROTOOPT;
1210 break;
1211 }
1212 if (error)
1213 break;
1214 }
1215 m_freem(control);
1216 }
1217 if (error) {
1218 if (unlock_inp == UH_WLOCKED)
1219 INP_WUNLOCK(inp);
1220 else
1221 INP_RUNLOCK(inp);
1222 m_freem(m);
1223 return (error);
1224 }
1225
1226 /*
1227 * Depending on whether or not the application has bound or connected
1228 * the socket, we may have to do varying levels of work. The optimal
1229 * case is for a connected UDP socket, as a global lock isn't
1230 * required at all.
1231 *
1232 * In order to decide which we need, we require stability of the
1233 * inpcb binding, which we ensure by acquiring a read lock on the
1234 * inpcb. This doesn't strictly follow the lock order, so we play
1235 * the trylock and retry game; note that we may end up with more
1236 * conservative locks than required the second time around, so later
1237 * assertions have to accept that. Further analysis of the number of
1238 * misses under contention is required.
1239 *
1240 * XXXRW: Check that hash locking update here is correct.
1241 */
1242 pr = inp->inp_socket->so_proto->pr_protocol;
1243 pcbinfo = udp_get_inpcbinfo(pr);
1244 sin = (struct sockaddr_in *)addr;
1245 if (sin != NULL &&
1246 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1247 INP_HASH_WLOCK(pcbinfo);
1248 unlock_udbinfo = UH_WLOCKED;
1249 } else if ((sin != NULL && (
1250 (sin->sin_addr.s_addr == INADDR_ANY) ||
1251 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1252 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1253 (inp->inp_lport == 0))) ||
1254 (src.sin_family == AF_INET)) {
1255 INP_HASH_RLOCK(pcbinfo);
1256 unlock_udbinfo = UH_RLOCKED;
1257 } else
1258 unlock_udbinfo = UH_UNLOCKED;
1259
1260 /*
1261 * If the IP_SENDSRCADDR control message was specified, override the
1262 * source address for this datagram. Its use is invalidated if the
1263 * address thus specified is incomplete or clobbers other inpcbs.
1264 */
1265 laddr = inp->inp_laddr;
1266 lport = inp->inp_lport;
1267 if (src.sin_family == AF_INET) {
1268 INP_HASH_LOCK_ASSERT(pcbinfo);
1269 if ((lport == 0) ||
1270 (laddr.s_addr == INADDR_ANY &&
1271 src.sin_addr.s_addr == INADDR_ANY)) {
1272 error = EINVAL;
1273 goto release;
1274 }
1275 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1276 &laddr.s_addr, &lport, td->td_ucred);
1277 if (error)
1278 goto release;
1279 }
1280
1281 /*
1282 * If a UDP socket has been connected, then a local address/port will
1283 * have been selected and bound.
1284 *
1285 * If a UDP socket has not been connected to, then an explicit
1286 * destination address must be used, in which case a local
1287 * address/port may not have been selected and bound.
1288 */
1289 if (sin != NULL) {
1290 INP_LOCK_ASSERT(inp);
1291 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1292 error = EISCONN;
1293 goto release;
1294 }
1295
1296 /*
1297 * Jail may rewrite the destination address, so let it do
1298 * that before we use it.
1299 */
1300 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1301 if (error)
1302 goto release;
1303
1304 /*
1305 * If a local address or port hasn't yet been selected, or if
1306 * the destination address needs to be rewritten due to using
1307 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1308 * to do the heavy lifting. Once a port is selected, we
1309 * commit the binding back to the socket; we also commit the
1310 * binding of the address if in jail.
1311 *
1312 * If we already have a valid binding and we're not
1313 * requesting a destination address rewrite, use a fast path.
1314 */
1315 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1316 inp->inp_lport == 0 ||
1317 sin->sin_addr.s_addr == INADDR_ANY ||
1318 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1319 INP_HASH_LOCK_ASSERT(pcbinfo);
1320 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1321 &lport, &faddr.s_addr, &fport, NULL,
1322 td->td_ucred);
1323 if (error)
1324 goto release;
1325
1326 /*
1327 * XXXRW: Why not commit the port if the address is
1328 * !INADDR_ANY?
1329 */
1330 /* Commit the local port if newly assigned. */
1331 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1332 inp->inp_lport == 0) {
1333 INP_WLOCK_ASSERT(inp);
1334 INP_HASH_WLOCK_ASSERT(pcbinfo);
1335 /*
1336 * Remember addr if jailed, to prevent
1337 * rebinding.
1338 */
1339 if (prison_flag(td->td_ucred, PR_IP4))
1340 inp->inp_laddr = laddr;
1341 inp->inp_lport = lport;
1342 if (in_pcbinshash(inp) != 0) {
1343 inp->inp_lport = 0;
1344 error = EAGAIN;
1345 goto release;
1346 }
1347 inp->inp_flags |= INP_ANONPORT;
1348 }
1349 } else {
1350 faddr = sin->sin_addr;
1351 fport = sin->sin_port;
1352 }
1353 } else {
1354 INP_LOCK_ASSERT(inp);
1355 faddr = inp->inp_faddr;
1356 fport = inp->inp_fport;
1357 if (faddr.s_addr == INADDR_ANY) {
1358 error = ENOTCONN;
1359 goto release;
1360 }
1361 }
1362
1363 /*
1364 * Calculate data length and get a mbuf for UDP, IP, and possible
1365 * link-layer headers. Immediate slide the data pointer back forward
1366 * since we won't use that space at this layer.
1367 */
1368 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1369 if (m == NULL) {
1370 error = ENOBUFS;
1371 goto release;
1372 }
1373 m->m_data += max_linkhdr;
1374 m->m_len -= max_linkhdr;
1375 m->m_pkthdr.len -= max_linkhdr;
1376
1377 /*
1378 * Fill in mbuf with extended UDP header and addresses and length put
1379 * into network format.
1380 */
1381 ui = mtod(m, struct udpiphdr *);
1382 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1383 ui->ui_pr = pr;
1384 ui->ui_src = laddr;
1385 ui->ui_dst = faddr;
1386 ui->ui_sport = lport;
1387 ui->ui_dport = fport;
1388 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1389 if (pr == IPPROTO_UDPLITE) {
1390 struct udpcb *up;
1391 uint16_t plen;
1392
1393 up = intoudpcb(inp);
1394 cscov = up->u_txcslen;
1395 plen = (u_short)len + sizeof(struct udphdr);
1396 if (cscov >= plen)
1397 cscov = 0;
1398 ui->ui_len = htons(plen);
1399 ui->ui_ulen = htons(cscov);
1400 /*
1401 * For UDP-Lite, checksum coverage length of zero means
1402 * the entire UDPLite packet is covered by the checksum.
1403 */
1404 cscov_partial = (cscov == 0) ? 0 : 1;
1405 } else
1406 ui->ui_v = IPVERSION << 4;
1407
1408 /*
1409 * Set the Don't Fragment bit in the IP header.
1410 */
1411 if (inp->inp_flags & INP_DONTFRAG) {
1412 struct ip *ip;
1413
1414 ip = (struct ip *)&ui->ui_i;
1415 ip->ip_off |= htons(IP_DF);
1416 }
1417
1418 ipflags = 0;
1419 if (inp->inp_socket->so_options & SO_DONTROUTE)
1420 ipflags |= IP_ROUTETOIF;
1421 if (inp->inp_socket->so_options & SO_BROADCAST)
1422 ipflags |= IP_ALLOWBROADCAST;
1423 if (inp->inp_flags & INP_ONESBCAST)
1424 ipflags |= IP_SENDONES;
1425
1426 #ifdef MAC
1427 mac_inpcb_create_mbuf(inp, m);
1428 #endif
1429
1430 /*
1431 * Set up checksum and output datagram.
1432 */
1433 ui->ui_sum = 0;
1434 if (pr == IPPROTO_UDPLITE) {
1435 if (inp->inp_flags & INP_ONESBCAST)
1436 faddr.s_addr = INADDR_BROADCAST;
1437 if (cscov_partial) {
1438 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1439 ui->ui_sum = 0xffff;
1440 } else {
1441 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1442 ui->ui_sum = 0xffff;
1443 }
1444 } else if (V_udp_cksum) {
1445 if (inp->inp_flags & INP_ONESBCAST)
1446 faddr.s_addr = INADDR_BROADCAST;
1447 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1448 htons((u_short)len + sizeof(struct udphdr) + pr));
1449 m->m_pkthdr.csum_flags = CSUM_UDP;
1450 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1451 }
1452 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1453 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1454 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1455 UDPSTAT_INC(udps_opackets);
1456
1457 /*
1458 * Setup flowid / RSS information for outbound socket.
1459 *
1460 * Once the UDP code decides to set a flowid some other way,
1461 * this allows the flowid to be overridden by userland.
1462 */
1463 if (flowtype != M_HASHTYPE_NONE) {
1464 m->m_pkthdr.flowid = flowid;
1465 M_HASHTYPE_SET(m, flowtype);
1466 #ifdef RSS
1467 } else {
1468 uint32_t hash_val, hash_type;
1469 /*
1470 * Calculate an appropriate RSS hash for UDP and
1471 * UDP Lite.
1472 *
1473 * The called function will take care of figuring out
1474 * whether a 2-tuple or 4-tuple hash is required based
1475 * on the currently configured scheme.
1476 *
1477 * Later later on connected socket values should be
1478 * cached in the inpcb and reused, rather than constantly
1479 * re-calculating it.
1480 *
1481 * UDP Lite is a different protocol number and will
1482 * likely end up being hashed as a 2-tuple until
1483 * RSS / NICs grow UDP Lite protocol awareness.
1484 */
1485 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1486 pr, &hash_val, &hash_type) == 0) {
1487 m->m_pkthdr.flowid = hash_val;
1488 M_HASHTYPE_SET(m, hash_type);
1489 }
1490 #endif
1491 }
1492
1493 #ifdef RSS
1494 /*
1495 * Don't override with the inp cached flowid value.
1496 *
1497 * Depending upon the kind of send being done, the inp
1498 * flowid/flowtype values may actually not be appropriate
1499 * for this particular socket send.
1500 *
1501 * We should either leave the flowid at zero (which is what is
1502 * currently done) or set it to some software generated
1503 * hash value based on the packet contents.
1504 */
1505 ipflags |= IP_NODEFAULTFLOWID;
1506 #endif /* RSS */
1507
1508 if (unlock_udbinfo == UH_WLOCKED)
1509 INP_HASH_WUNLOCK(pcbinfo);
1510 else if (unlock_udbinfo == UH_RLOCKED)
1511 INP_HASH_RUNLOCK(pcbinfo);
1512 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1513 error = ip_output(m, inp->inp_options,
1514 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1515 inp->inp_moptions, inp);
1516 if (unlock_inp == UH_WLOCKED)
1517 INP_WUNLOCK(inp);
1518 else
1519 INP_RUNLOCK(inp);
1520 return (error);
1521
1522 release:
1523 if (unlock_udbinfo == UH_WLOCKED) {
1524 KASSERT(unlock_inp == UH_WLOCKED,
1525 ("%s: excl udbinfo lock, shared inp lock", __func__));
1526 INP_HASH_WUNLOCK(pcbinfo);
1527 INP_WUNLOCK(inp);
1528 } else if (unlock_udbinfo == UH_RLOCKED) {
1529 KASSERT(unlock_inp == UH_RLOCKED,
1530 ("%s: shared udbinfo lock, excl inp lock", __func__));
1531 INP_HASH_RUNLOCK(pcbinfo);
1532 INP_RUNLOCK(inp);
1533 } else if (unlock_inp == UH_WLOCKED)
1534 INP_WUNLOCK(inp);
1535 else
1536 INP_RUNLOCK(inp);
1537 m_freem(m);
1538 return (error);
1539 }
1540
1541 static void
1542 udp_abort(struct socket *so)
1543 {
1544 struct inpcb *inp;
1545 struct inpcbinfo *pcbinfo;
1546
1547 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1548 inp = sotoinpcb(so);
1549 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1550 INP_WLOCK(inp);
1551 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1552 INP_HASH_WLOCK(pcbinfo);
1553 in_pcbdisconnect(inp);
1554 inp->inp_laddr.s_addr = INADDR_ANY;
1555 INP_HASH_WUNLOCK(pcbinfo);
1556 soisdisconnected(so);
1557 }
1558 INP_WUNLOCK(inp);
1559 }
1560
1561 static int
1562 udp_attach(struct socket *so, int proto, struct thread *td)
1563 {
1564 struct inpcb *inp;
1565 struct inpcbinfo *pcbinfo;
1566 int error;
1567
1568 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1569 inp = sotoinpcb(so);
1570 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1571 error = soreserve(so, udp_sendspace, udp_recvspace);
1572 if (error)
1573 return (error);
1574 INP_INFO_WLOCK(pcbinfo);
1575 error = in_pcballoc(so, pcbinfo);
1576 if (error) {
1577 INP_INFO_WUNLOCK(pcbinfo);
1578 return (error);
1579 }
1580
1581 inp = sotoinpcb(so);
1582 inp->inp_vflag |= INP_IPV4;
1583 inp->inp_ip_ttl = V_ip_defttl;
1584
1585 error = udp_newudpcb(inp);
1586 if (error) {
1587 in_pcbdetach(inp);
1588 in_pcbfree(inp);
1589 INP_INFO_WUNLOCK(pcbinfo);
1590 return (error);
1591 }
1592
1593 INP_WUNLOCK(inp);
1594 INP_INFO_WUNLOCK(pcbinfo);
1595 return (0);
1596 }
1597 #endif /* INET */
1598
1599 int
1600 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1601 {
1602 struct inpcb *inp;
1603 struct udpcb *up;
1604
1605 KASSERT(so->so_type == SOCK_DGRAM,
1606 ("udp_set_kernel_tunneling: !dgram"));
1607 inp = sotoinpcb(so);
1608 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1609 INP_WLOCK(inp);
1610 up = intoudpcb(inp);
1611 if ((up->u_tun_func != NULL) ||
1612 (up->u_icmp_func != NULL)) {
1613 INP_WUNLOCK(inp);
1614 return (EBUSY);
1615 }
1616 up->u_tun_func = f;
1617 up->u_icmp_func = i;
1618 up->u_tun_ctx = ctx;
1619 INP_WUNLOCK(inp);
1620 return (0);
1621 }
1622
1623 #ifdef INET
1624 static int
1625 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1626 {
1627 struct inpcb *inp;
1628 struct inpcbinfo *pcbinfo;
1629 int error;
1630
1631 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1632 inp = sotoinpcb(so);
1633 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1634 INP_WLOCK(inp);
1635 INP_HASH_WLOCK(pcbinfo);
1636 error = in_pcbbind(inp, nam, td->td_ucred);
1637 INP_HASH_WUNLOCK(pcbinfo);
1638 INP_WUNLOCK(inp);
1639 return (error);
1640 }
1641
1642 static void
1643 udp_close(struct socket *so)
1644 {
1645 struct inpcb *inp;
1646 struct inpcbinfo *pcbinfo;
1647
1648 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1649 inp = sotoinpcb(so);
1650 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1651 INP_WLOCK(inp);
1652 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1653 INP_HASH_WLOCK(pcbinfo);
1654 in_pcbdisconnect(inp);
1655 inp->inp_laddr.s_addr = INADDR_ANY;
1656 INP_HASH_WUNLOCK(pcbinfo);
1657 soisdisconnected(so);
1658 }
1659 INP_WUNLOCK(inp);
1660 }
1661
1662 static int
1663 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1664 {
1665 struct inpcb *inp;
1666 struct inpcbinfo *pcbinfo;
1667 struct sockaddr_in *sin;
1668 int error;
1669
1670 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1671 inp = sotoinpcb(so);
1672 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1673 INP_WLOCK(inp);
1674 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1675 INP_WUNLOCK(inp);
1676 return (EISCONN);
1677 }
1678 sin = (struct sockaddr_in *)nam;
1679 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1680 if (error != 0) {
1681 INP_WUNLOCK(inp);
1682 return (error);
1683 }
1684 INP_HASH_WLOCK(pcbinfo);
1685 error = in_pcbconnect(inp, nam, td->td_ucred);
1686 INP_HASH_WUNLOCK(pcbinfo);
1687 if (error == 0)
1688 soisconnected(so);
1689 INP_WUNLOCK(inp);
1690 return (error);
1691 }
1692
1693 static void
1694 udp_detach(struct socket *so)
1695 {
1696 struct inpcb *inp;
1697 struct inpcbinfo *pcbinfo;
1698 struct udpcb *up;
1699
1700 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1701 inp = sotoinpcb(so);
1702 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1703 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1704 ("udp_detach: not disconnected"));
1705 INP_INFO_WLOCK(pcbinfo);
1706 INP_WLOCK(inp);
1707 up = intoudpcb(inp);
1708 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1709 inp->inp_ppcb = NULL;
1710 in_pcbdetach(inp);
1711 in_pcbfree(inp);
1712 INP_INFO_WUNLOCK(pcbinfo);
1713 udp_discardcb(up);
1714 }
1715
1716 static int
1717 udp_disconnect(struct socket *so)
1718 {
1719 struct inpcb *inp;
1720 struct inpcbinfo *pcbinfo;
1721
1722 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1723 inp = sotoinpcb(so);
1724 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1725 INP_WLOCK(inp);
1726 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1727 INP_WUNLOCK(inp);
1728 return (ENOTCONN);
1729 }
1730 INP_HASH_WLOCK(pcbinfo);
1731 in_pcbdisconnect(inp);
1732 inp->inp_laddr.s_addr = INADDR_ANY;
1733 INP_HASH_WUNLOCK(pcbinfo);
1734 SOCK_LOCK(so);
1735 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1736 SOCK_UNLOCK(so);
1737 INP_WUNLOCK(inp);
1738 return (0);
1739 }
1740
1741 static int
1742 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1743 struct mbuf *control, struct thread *td)
1744 {
1745 struct inpcb *inp;
1746
1747 inp = sotoinpcb(so);
1748 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1749 return (udp_output(inp, m, addr, control, td));
1750 }
1751 #endif /* INET */
1752
1753 int
1754 udp_shutdown(struct socket *so)
1755 {
1756 struct inpcb *inp;
1757
1758 inp = sotoinpcb(so);
1759 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1760 INP_WLOCK(inp);
1761 socantsendmore(so);
1762 INP_WUNLOCK(inp);
1763 return (0);
1764 }
1765
1766 #ifdef INET
1767 struct pr_usrreqs udp_usrreqs = {
1768 .pru_abort = udp_abort,
1769 .pru_attach = udp_attach,
1770 .pru_bind = udp_bind,
1771 .pru_connect = udp_connect,
1772 .pru_control = in_control,
1773 .pru_detach = udp_detach,
1774 .pru_disconnect = udp_disconnect,
1775 .pru_peeraddr = in_getpeeraddr,
1776 .pru_send = udp_send,
1777 .pru_soreceive = soreceive_dgram,
1778 .pru_sosend = sosend_dgram,
1779 .pru_shutdown = udp_shutdown,
1780 .pru_sockaddr = in_getsockaddr,
1781 .pru_sosetlabel = in_pcbsosetlabel,
1782 .pru_close = udp_close,
1783 };
1784 #endif /* INET */
Cache object: 277448034b3008964632ab55121933d3
|