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("$FreeBSD$");
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 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
671 UDPSTAT_INC(udps_noport);
672 if (m->m_flags & (M_BCAST | M_MCAST)) {
673 UDPSTAT_INC(udps_noportbcast);
674 goto badunlocked;
675 }
676 if (V_udp_blackhole)
677 goto badunlocked;
678 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
679 goto badunlocked;
680 *ip = save_ip;
681 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
682 return (IPPROTO_DONE);
683 }
684
685 /*
686 * Check the minimum TTL for socket.
687 */
688 INP_RLOCK_ASSERT(inp);
689 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
690 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
691 INP_RUNLOCK(inp);
692 m_freem(m);
693 return (IPPROTO_DONE);
694 }
695 if (cscov_partial) {
696 struct udpcb *up;
697
698 up = intoudpcb(inp);
699 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
700 INP_RUNLOCK(inp);
701 m_freem(m);
702 return (IPPROTO_DONE);
703 }
704 }
705
706 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
707 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
708 INP_RUNLOCK(inp);
709 return (IPPROTO_DONE);
710
711 badunlocked:
712 m_freem(m);
713 return (IPPROTO_DONE);
714 }
715 #endif /* INET */
716
717 /*
718 * Notify a udp user of an asynchronous error; just wake up so that they can
719 * collect error status.
720 */
721 struct inpcb *
722 udp_notify(struct inpcb *inp, int errno)
723 {
724
725 /*
726 * While udp_ctlinput() always calls udp_notify() with a read lock
727 * when invoking it directly, in_pcbnotifyall() currently uses write
728 * locks due to sharing code with TCP. For now, accept either a read
729 * or a write lock, but a read lock is sufficient.
730 */
731 INP_LOCK_ASSERT(inp);
732 if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
733 errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
734 RTFREE(inp->inp_route.ro_rt);
735 inp->inp_route.ro_rt = (struct rtentry *)NULL;
736 }
737
738 inp->inp_socket->so_error = errno;
739 sorwakeup(inp->inp_socket);
740 sowwakeup(inp->inp_socket);
741 return (inp);
742 }
743
744 #ifdef INET
745 static void
746 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
747 struct inpcbinfo *pcbinfo)
748 {
749 struct ip *ip = vip;
750 struct udphdr *uh;
751 struct in_addr faddr;
752 struct inpcb *inp;
753
754 faddr = ((struct sockaddr_in *)sa)->sin_addr;
755 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
756 return;
757
758 if (PRC_IS_REDIRECT(cmd)) {
759 /* signal EHOSTDOWN, as it flushes the cached route */
760 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
761 return;
762 }
763
764 /*
765 * Hostdead is ugly because it goes linearly through all PCBs.
766 *
767 * XXX: We never get this from ICMP, otherwise it makes an excellent
768 * DoS attack on machines with many connections.
769 */
770 if (cmd == PRC_HOSTDEAD)
771 ip = NULL;
772 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
773 return;
774 if (ip != NULL) {
775 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
776 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
777 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
778 if (inp != NULL) {
779 INP_RLOCK_ASSERT(inp);
780 if (inp->inp_socket != NULL) {
781 udp_notify(inp, inetctlerrmap[cmd]);
782 }
783 INP_RUNLOCK(inp);
784 } else {
785 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
786 ip->ip_src, uh->uh_sport,
787 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
788 if (inp != NULL) {
789 struct udpcb *up;
790
791 up = intoudpcb(inp);
792 if (up->u_icmp_func != NULL) {
793 INP_RUNLOCK(inp);
794 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
795 } else {
796 INP_RUNLOCK(inp);
797 }
798 }
799 }
800 } else
801 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
802 udp_notify);
803 }
804 void
805 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
806 {
807
808 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
809 }
810
811 void
812 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
813 {
814
815 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
816 }
817 #endif /* INET */
818
819 static int
820 udp_pcblist(SYSCTL_HANDLER_ARGS)
821 {
822 int error, i, n;
823 struct inpcb *inp, **inp_list;
824 inp_gen_t gencnt;
825 struct xinpgen xig;
826
827 /*
828 * The process of preparing the PCB list is too time-consuming and
829 * resource-intensive to repeat twice on every request.
830 */
831 if (req->oldptr == 0) {
832 n = V_udbinfo.ipi_count;
833 n += imax(n / 8, 10);
834 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
835 return (0);
836 }
837
838 if (req->newptr != 0)
839 return (EPERM);
840
841 /*
842 * OK, now we're committed to doing something.
843 */
844 INP_INFO_RLOCK(&V_udbinfo);
845 gencnt = V_udbinfo.ipi_gencnt;
846 n = V_udbinfo.ipi_count;
847 INP_INFO_RUNLOCK(&V_udbinfo);
848
849 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
850 + n * sizeof(struct xinpcb));
851 if (error != 0)
852 return (error);
853
854 bzero(&xig, sizeof(xig));
855 xig.xig_len = sizeof xig;
856 xig.xig_count = n;
857 xig.xig_gen = gencnt;
858 xig.xig_sogen = so_gencnt;
859 error = SYSCTL_OUT(req, &xig, sizeof xig);
860 if (error)
861 return (error);
862
863 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
864 if (inp_list == NULL)
865 return (ENOMEM);
866
867 INP_INFO_RLOCK(&V_udbinfo);
868 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
869 inp = LIST_NEXT(inp, inp_list)) {
870 INP_WLOCK(inp);
871 if (inp->inp_gencnt <= gencnt &&
872 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
873 in_pcbref(inp);
874 inp_list[i++] = inp;
875 }
876 INP_WUNLOCK(inp);
877 }
878 INP_INFO_RUNLOCK(&V_udbinfo);
879 n = i;
880
881 error = 0;
882 for (i = 0; i < n; i++) {
883 inp = inp_list[i];
884 INP_RLOCK(inp);
885 if (inp->inp_gencnt <= gencnt) {
886 struct xinpcb xi;
887
888 bzero(&xi, sizeof(xi));
889 xi.xi_len = sizeof xi;
890 /* XXX should avoid extra copy */
891 bcopy(inp, &xi.xi_inp, sizeof *inp);
892 if (inp->inp_socket)
893 sotoxsocket(inp->inp_socket, &xi.xi_socket);
894 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
895 INP_RUNLOCK(inp);
896 error = SYSCTL_OUT(req, &xi, sizeof xi);
897 } else
898 INP_RUNLOCK(inp);
899 }
900 INP_INFO_WLOCK(&V_udbinfo);
901 for (i = 0; i < n; i++) {
902 inp = inp_list[i];
903 INP_RLOCK(inp);
904 if (!in_pcbrele_rlocked(inp))
905 INP_RUNLOCK(inp);
906 }
907 INP_INFO_WUNLOCK(&V_udbinfo);
908
909 if (!error) {
910 /*
911 * Give the user an updated idea of our state. If the
912 * generation differs from what we told her before, she knows
913 * that something happened while we were processing this
914 * request, and it might be necessary to retry.
915 */
916 INP_INFO_RLOCK(&V_udbinfo);
917 xig.xig_gen = V_udbinfo.ipi_gencnt;
918 xig.xig_sogen = so_gencnt;
919 xig.xig_count = V_udbinfo.ipi_count;
920 INP_INFO_RUNLOCK(&V_udbinfo);
921 error = SYSCTL_OUT(req, &xig, sizeof xig);
922 }
923 free(inp_list, M_TEMP);
924 return (error);
925 }
926
927 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
928 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
929 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
930
931 #ifdef INET
932 static int
933 udp_getcred(SYSCTL_HANDLER_ARGS)
934 {
935 struct xucred xuc;
936 struct sockaddr_in addrs[2];
937 struct inpcb *inp;
938 int error;
939
940 error = priv_check(req->td, PRIV_NETINET_GETCRED);
941 if (error)
942 return (error);
943 error = SYSCTL_IN(req, addrs, sizeof(addrs));
944 if (error)
945 return (error);
946 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
947 addrs[0].sin_addr, addrs[0].sin_port,
948 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
949 if (inp != NULL) {
950 INP_RLOCK_ASSERT(inp);
951 if (inp->inp_socket == NULL)
952 error = ENOENT;
953 if (error == 0)
954 error = cr_canseeinpcb(req->td->td_ucred, inp);
955 if (error == 0)
956 cru2x(inp->inp_cred, &xuc);
957 INP_RUNLOCK(inp);
958 } else
959 error = ENOENT;
960 if (error == 0)
961 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
962 return (error);
963 }
964
965 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
966 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
967 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
968 #endif /* INET */
969
970 int
971 udp_ctloutput(struct socket *so, struct sockopt *sopt)
972 {
973 struct inpcb *inp;
974 struct udpcb *up;
975 int isudplite, error, optval;
976
977 error = 0;
978 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
979 inp = sotoinpcb(so);
980 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
981 INP_WLOCK(inp);
982 if (sopt->sopt_level != so->so_proto->pr_protocol) {
983 #ifdef INET6
984 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
985 INP_WUNLOCK(inp);
986 error = ip6_ctloutput(so, sopt);
987 }
988 #endif
989 #if defined(INET) && defined(INET6)
990 else
991 #endif
992 #ifdef INET
993 {
994 INP_WUNLOCK(inp);
995 error = ip_ctloutput(so, sopt);
996 }
997 #endif
998 return (error);
999 }
1000
1001 switch (sopt->sopt_dir) {
1002 case SOPT_SET:
1003 switch (sopt->sopt_name) {
1004 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1005 #ifdef INET
1006 case UDP_ENCAP:
1007 if (!IPSEC_ENABLED(ipv4)) {
1008 INP_WUNLOCK(inp);
1009 return (ENOPROTOOPT);
1010 }
1011 error = UDPENCAP_PCBCTL(inp, sopt);
1012 break;
1013 #endif /* INET */
1014 #endif /* IPSEC */
1015 case UDPLITE_SEND_CSCOV:
1016 case UDPLITE_RECV_CSCOV:
1017 if (!isudplite) {
1018 INP_WUNLOCK(inp);
1019 error = ENOPROTOOPT;
1020 break;
1021 }
1022 INP_WUNLOCK(inp);
1023 error = sooptcopyin(sopt, &optval, sizeof(optval),
1024 sizeof(optval));
1025 if (error != 0)
1026 break;
1027 inp = sotoinpcb(so);
1028 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1029 INP_WLOCK(inp);
1030 up = intoudpcb(inp);
1031 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1032 if ((optval != 0 && optval < 8) || (optval > 65535)) {
1033 INP_WUNLOCK(inp);
1034 error = EINVAL;
1035 break;
1036 }
1037 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1038 up->u_txcslen = optval;
1039 else
1040 up->u_rxcslen = optval;
1041 INP_WUNLOCK(inp);
1042 break;
1043 default:
1044 INP_WUNLOCK(inp);
1045 error = ENOPROTOOPT;
1046 break;
1047 }
1048 break;
1049 case SOPT_GET:
1050 switch (sopt->sopt_name) {
1051 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1052 #ifdef INET
1053 case UDP_ENCAP:
1054 if (!IPSEC_ENABLED(ipv4)) {
1055 INP_WUNLOCK(inp);
1056 return (ENOPROTOOPT);
1057 }
1058 error = UDPENCAP_PCBCTL(inp, sopt);
1059 break;
1060 #endif /* INET */
1061 #endif /* IPSEC */
1062 case UDPLITE_SEND_CSCOV:
1063 case UDPLITE_RECV_CSCOV:
1064 if (!isudplite) {
1065 INP_WUNLOCK(inp);
1066 error = ENOPROTOOPT;
1067 break;
1068 }
1069 up = intoudpcb(inp);
1070 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1071 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1072 optval = up->u_txcslen;
1073 else
1074 optval = up->u_rxcslen;
1075 INP_WUNLOCK(inp);
1076 error = sooptcopyout(sopt, &optval, sizeof(optval));
1077 break;
1078 default:
1079 INP_WUNLOCK(inp);
1080 error = ENOPROTOOPT;
1081 break;
1082 }
1083 break;
1084 }
1085 return (error);
1086 }
1087
1088 #ifdef INET
1089 #define UH_WLOCKED 2
1090 #define UH_RLOCKED 1
1091 #define UH_UNLOCKED 0
1092 static int
1093 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1094 struct mbuf *control, struct thread *td)
1095 {
1096 struct udpiphdr *ui;
1097 int len = m->m_pkthdr.len;
1098 struct in_addr faddr, laddr;
1099 struct cmsghdr *cm;
1100 struct inpcbinfo *pcbinfo;
1101 struct sockaddr_in *sin, src;
1102 int cscov_partial = 0;
1103 int error = 0;
1104 int ipflags;
1105 u_short fport, lport;
1106 int unlock_udbinfo, unlock_inp;
1107 u_char tos;
1108 uint8_t pr;
1109 uint16_t cscov = 0;
1110 uint32_t flowid = 0;
1111 uint8_t flowtype = M_HASHTYPE_NONE;
1112
1113 /*
1114 * udp_output() may need to temporarily bind or connect the current
1115 * inpcb. As such, we don't know up front whether we will need the
1116 * pcbinfo lock or not. Do any work to decide what is needed up
1117 * front before acquiring any locks.
1118 */
1119 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1120 if (control)
1121 m_freem(control);
1122 m_freem(m);
1123 return (EMSGSIZE);
1124 }
1125
1126 src.sin_family = 0;
1127 sin = (struct sockaddr_in *)addr;
1128 if (sin == NULL ||
1129 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1130 INP_WLOCK(inp);
1131 unlock_inp = UH_WLOCKED;
1132 } else {
1133 INP_RLOCK(inp);
1134 unlock_inp = UH_RLOCKED;
1135 }
1136 tos = inp->inp_ip_tos;
1137 if (control != NULL) {
1138 /*
1139 * XXX: Currently, we assume all the optional information is
1140 * stored in a single mbuf.
1141 */
1142 if (control->m_next) {
1143 if (unlock_inp == UH_WLOCKED)
1144 INP_WUNLOCK(inp);
1145 else
1146 INP_RUNLOCK(inp);
1147 m_freem(control);
1148 m_freem(m);
1149 return (EINVAL);
1150 }
1151 for (; control->m_len > 0;
1152 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1153 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1154 cm = mtod(control, struct cmsghdr *);
1155 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1156 || cm->cmsg_len > control->m_len) {
1157 error = EINVAL;
1158 break;
1159 }
1160 if (cm->cmsg_level != IPPROTO_IP)
1161 continue;
1162
1163 switch (cm->cmsg_type) {
1164 case IP_SENDSRCADDR:
1165 if (cm->cmsg_len !=
1166 CMSG_LEN(sizeof(struct in_addr))) {
1167 error = EINVAL;
1168 break;
1169 }
1170 bzero(&src, sizeof(src));
1171 src.sin_family = AF_INET;
1172 src.sin_len = sizeof(src);
1173 src.sin_port = inp->inp_lport;
1174 src.sin_addr =
1175 *(struct in_addr *)CMSG_DATA(cm);
1176 break;
1177
1178 case IP_TOS:
1179 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1180 error = EINVAL;
1181 break;
1182 }
1183 tos = *(u_char *)CMSG_DATA(cm);
1184 break;
1185
1186 case IP_FLOWID:
1187 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1188 error = EINVAL;
1189 break;
1190 }
1191 flowid = *(uint32_t *) CMSG_DATA(cm);
1192 break;
1193
1194 case IP_FLOWTYPE:
1195 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1196 error = EINVAL;
1197 break;
1198 }
1199 flowtype = *(uint32_t *) CMSG_DATA(cm);
1200 break;
1201
1202 #ifdef RSS
1203 case IP_RSSBUCKETID:
1204 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1205 error = EINVAL;
1206 break;
1207 }
1208 /* This is just a placeholder for now */
1209 break;
1210 #endif /* RSS */
1211 default:
1212 error = ENOPROTOOPT;
1213 break;
1214 }
1215 if (error)
1216 break;
1217 }
1218 m_freem(control);
1219 }
1220 if (error) {
1221 if (unlock_inp == UH_WLOCKED)
1222 INP_WUNLOCK(inp);
1223 else
1224 INP_RUNLOCK(inp);
1225 m_freem(m);
1226 return (error);
1227 }
1228
1229 /*
1230 * Depending on whether or not the application has bound or connected
1231 * the socket, we may have to do varying levels of work. The optimal
1232 * case is for a connected UDP socket, as a global lock isn't
1233 * required at all.
1234 *
1235 * In order to decide which we need, we require stability of the
1236 * inpcb binding, which we ensure by acquiring a read lock on the
1237 * inpcb. This doesn't strictly follow the lock order, so we play
1238 * the trylock and retry game; note that we may end up with more
1239 * conservative locks than required the second time around, so later
1240 * assertions have to accept that. Further analysis of the number of
1241 * misses under contention is required.
1242 *
1243 * XXXRW: Check that hash locking update here is correct.
1244 */
1245 pr = inp->inp_socket->so_proto->pr_protocol;
1246 pcbinfo = udp_get_inpcbinfo(pr);
1247 sin = (struct sockaddr_in *)addr;
1248 if (sin != NULL &&
1249 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1250 INP_HASH_WLOCK(pcbinfo);
1251 unlock_udbinfo = UH_WLOCKED;
1252 } else if ((sin != NULL && (
1253 (sin->sin_addr.s_addr == INADDR_ANY) ||
1254 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1255 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1256 (inp->inp_lport == 0))) ||
1257 (src.sin_family == AF_INET)) {
1258 INP_HASH_RLOCK(pcbinfo);
1259 unlock_udbinfo = UH_RLOCKED;
1260 } else
1261 unlock_udbinfo = UH_UNLOCKED;
1262
1263 /*
1264 * If the IP_SENDSRCADDR control message was specified, override the
1265 * source address for this datagram. Its use is invalidated if the
1266 * address thus specified is incomplete or clobbers other inpcbs.
1267 */
1268 laddr = inp->inp_laddr;
1269 lport = inp->inp_lport;
1270 if (src.sin_family == AF_INET) {
1271 INP_HASH_LOCK_ASSERT(pcbinfo);
1272 if ((lport == 0) ||
1273 (laddr.s_addr == INADDR_ANY &&
1274 src.sin_addr.s_addr == INADDR_ANY)) {
1275 error = EINVAL;
1276 goto release;
1277 }
1278 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1279 &laddr.s_addr, &lport, td->td_ucred);
1280 if (error)
1281 goto release;
1282 }
1283
1284 /*
1285 * If a UDP socket has been connected, then a local address/port will
1286 * have been selected and bound.
1287 *
1288 * If a UDP socket has not been connected to, then an explicit
1289 * destination address must be used, in which case a local
1290 * address/port may not have been selected and bound.
1291 */
1292 if (sin != NULL) {
1293 INP_LOCK_ASSERT(inp);
1294 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1295 error = EISCONN;
1296 goto release;
1297 }
1298
1299 /*
1300 * Jail may rewrite the destination address, so let it do
1301 * that before we use it.
1302 */
1303 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1304 if (error)
1305 goto release;
1306
1307 /*
1308 * If a local address or port hasn't yet been selected, or if
1309 * the destination address needs to be rewritten due to using
1310 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1311 * to do the heavy lifting. Once a port is selected, we
1312 * commit the binding back to the socket; we also commit the
1313 * binding of the address if in jail.
1314 *
1315 * If we already have a valid binding and we're not
1316 * requesting a destination address rewrite, use a fast path.
1317 */
1318 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1319 inp->inp_lport == 0 ||
1320 sin->sin_addr.s_addr == INADDR_ANY ||
1321 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1322 INP_HASH_LOCK_ASSERT(pcbinfo);
1323 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1324 &lport, &faddr.s_addr, &fport, NULL,
1325 td->td_ucred);
1326 if (error)
1327 goto release;
1328
1329 /*
1330 * XXXRW: Why not commit the port if the address is
1331 * !INADDR_ANY?
1332 */
1333 /* Commit the local port if newly assigned. */
1334 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1335 inp->inp_lport == 0) {
1336 INP_WLOCK_ASSERT(inp);
1337 INP_HASH_WLOCK_ASSERT(pcbinfo);
1338 /*
1339 * Remember addr if jailed, to prevent
1340 * rebinding.
1341 */
1342 if (prison_flag(td->td_ucred, PR_IP4))
1343 inp->inp_laddr = laddr;
1344 inp->inp_lport = lport;
1345 if (in_pcbinshash(inp) != 0) {
1346 inp->inp_lport = 0;
1347 error = EAGAIN;
1348 goto release;
1349 }
1350 inp->inp_flags |= INP_ANONPORT;
1351 }
1352 } else {
1353 faddr = sin->sin_addr;
1354 fport = sin->sin_port;
1355 }
1356 } else {
1357 INP_LOCK_ASSERT(inp);
1358 faddr = inp->inp_faddr;
1359 fport = inp->inp_fport;
1360 if (faddr.s_addr == INADDR_ANY) {
1361 error = ENOTCONN;
1362 goto release;
1363 }
1364 }
1365
1366 /*
1367 * Calculate data length and get a mbuf for UDP, IP, and possible
1368 * link-layer headers. Immediate slide the data pointer back forward
1369 * since we won't use that space at this layer.
1370 */
1371 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1372 if (m == NULL) {
1373 error = ENOBUFS;
1374 goto release;
1375 }
1376 m->m_data += max_linkhdr;
1377 m->m_len -= max_linkhdr;
1378 m->m_pkthdr.len -= max_linkhdr;
1379
1380 /*
1381 * Fill in mbuf with extended UDP header and addresses and length put
1382 * into network format.
1383 */
1384 ui = mtod(m, struct udpiphdr *);
1385 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1386 ui->ui_v = IPVERSION << 4;
1387 ui->ui_pr = pr;
1388 ui->ui_src = laddr;
1389 ui->ui_dst = faddr;
1390 ui->ui_sport = lport;
1391 ui->ui_dport = fport;
1392 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1393 if (pr == IPPROTO_UDPLITE) {
1394 struct udpcb *up;
1395 uint16_t plen;
1396
1397 up = intoudpcb(inp);
1398 cscov = up->u_txcslen;
1399 plen = (u_short)len + sizeof(struct udphdr);
1400 if (cscov >= plen)
1401 cscov = 0;
1402 ui->ui_len = htons(plen);
1403 ui->ui_ulen = htons(cscov);
1404 /*
1405 * For UDP-Lite, checksum coverage length of zero means
1406 * the entire UDPLite packet is covered by the checksum.
1407 */
1408 cscov_partial = (cscov == 0) ? 0 : 1;
1409 }
1410
1411 /*
1412 * Set the Don't Fragment bit in the IP header.
1413 */
1414 if (inp->inp_flags & INP_DONTFRAG) {
1415 struct ip *ip;
1416
1417 ip = (struct ip *)&ui->ui_i;
1418 ip->ip_off |= htons(IP_DF);
1419 }
1420
1421 ipflags = 0;
1422 if (inp->inp_socket->so_options & SO_DONTROUTE)
1423 ipflags |= IP_ROUTETOIF;
1424 if (inp->inp_socket->so_options & SO_BROADCAST)
1425 ipflags |= IP_ALLOWBROADCAST;
1426 if (inp->inp_flags & INP_ONESBCAST)
1427 ipflags |= IP_SENDONES;
1428
1429 #ifdef MAC
1430 mac_inpcb_create_mbuf(inp, m);
1431 #endif
1432
1433 /*
1434 * Set up checksum and output datagram.
1435 */
1436 ui->ui_sum = 0;
1437 if (pr == IPPROTO_UDPLITE) {
1438 if (inp->inp_flags & INP_ONESBCAST)
1439 faddr.s_addr = INADDR_BROADCAST;
1440 if (cscov_partial) {
1441 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1442 ui->ui_sum = 0xffff;
1443 } else {
1444 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1445 ui->ui_sum = 0xffff;
1446 }
1447 } else if (V_udp_cksum) {
1448 if (inp->inp_flags & INP_ONESBCAST)
1449 faddr.s_addr = INADDR_BROADCAST;
1450 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1451 htons((u_short)len + sizeof(struct udphdr) + pr));
1452 m->m_pkthdr.csum_flags = CSUM_UDP;
1453 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1454 }
1455 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1456 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1457 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1458 UDPSTAT_INC(udps_opackets);
1459
1460 /*
1461 * Setup flowid / RSS information for outbound socket.
1462 *
1463 * Once the UDP code decides to set a flowid some other way,
1464 * this allows the flowid to be overridden by userland.
1465 */
1466 if (flowtype != M_HASHTYPE_NONE) {
1467 m->m_pkthdr.flowid = flowid;
1468 M_HASHTYPE_SET(m, flowtype);
1469 #ifdef RSS
1470 } else {
1471 uint32_t hash_val, hash_type;
1472 /*
1473 * Calculate an appropriate RSS hash for UDP and
1474 * UDP Lite.
1475 *
1476 * The called function will take care of figuring out
1477 * whether a 2-tuple or 4-tuple hash is required based
1478 * on the currently configured scheme.
1479 *
1480 * Later later on connected socket values should be
1481 * cached in the inpcb and reused, rather than constantly
1482 * re-calculating it.
1483 *
1484 * UDP Lite is a different protocol number and will
1485 * likely end up being hashed as a 2-tuple until
1486 * RSS / NICs grow UDP Lite protocol awareness.
1487 */
1488 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1489 pr, &hash_val, &hash_type) == 0) {
1490 m->m_pkthdr.flowid = hash_val;
1491 M_HASHTYPE_SET(m, hash_type);
1492 }
1493 #endif
1494 }
1495
1496 #ifdef RSS
1497 /*
1498 * Don't override with the inp cached flowid value.
1499 *
1500 * Depending upon the kind of send being done, the inp
1501 * flowid/flowtype values may actually not be appropriate
1502 * for this particular socket send.
1503 *
1504 * We should either leave the flowid at zero (which is what is
1505 * currently done) or set it to some software generated
1506 * hash value based on the packet contents.
1507 */
1508 ipflags |= IP_NODEFAULTFLOWID;
1509 #endif /* RSS */
1510
1511 if (unlock_udbinfo == UH_WLOCKED)
1512 INP_HASH_WUNLOCK(pcbinfo);
1513 else if (unlock_udbinfo == UH_RLOCKED)
1514 INP_HASH_RUNLOCK(pcbinfo);
1515 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1516 error = ip_output(m, inp->inp_options,
1517 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1518 inp->inp_moptions, inp);
1519 if (unlock_inp == UH_WLOCKED)
1520 INP_WUNLOCK(inp);
1521 else
1522 INP_RUNLOCK(inp);
1523 return (error);
1524
1525 release:
1526 if (unlock_udbinfo == UH_WLOCKED) {
1527 KASSERT(unlock_inp == UH_WLOCKED,
1528 ("%s: excl udbinfo lock, shared inp lock", __func__));
1529 INP_HASH_WUNLOCK(pcbinfo);
1530 INP_WUNLOCK(inp);
1531 } else if (unlock_udbinfo == UH_RLOCKED) {
1532 KASSERT(unlock_inp == UH_RLOCKED,
1533 ("%s: shared udbinfo lock, excl inp lock", __func__));
1534 INP_HASH_RUNLOCK(pcbinfo);
1535 INP_RUNLOCK(inp);
1536 } else if (unlock_inp == UH_WLOCKED)
1537 INP_WUNLOCK(inp);
1538 else
1539 INP_RUNLOCK(inp);
1540 m_freem(m);
1541 return (error);
1542 }
1543
1544 static void
1545 udp_abort(struct socket *so)
1546 {
1547 struct inpcb *inp;
1548 struct inpcbinfo *pcbinfo;
1549
1550 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1551 inp = sotoinpcb(so);
1552 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1553 INP_WLOCK(inp);
1554 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1555 INP_HASH_WLOCK(pcbinfo);
1556 in_pcbdisconnect(inp);
1557 inp->inp_laddr.s_addr = INADDR_ANY;
1558 INP_HASH_WUNLOCK(pcbinfo);
1559 soisdisconnected(so);
1560 }
1561 INP_WUNLOCK(inp);
1562 }
1563
1564 static int
1565 udp_attach(struct socket *so, int proto, struct thread *td)
1566 {
1567 struct inpcb *inp;
1568 struct inpcbinfo *pcbinfo;
1569 int error;
1570
1571 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1572 inp = sotoinpcb(so);
1573 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1574 error = soreserve(so, udp_sendspace, udp_recvspace);
1575 if (error)
1576 return (error);
1577 INP_INFO_WLOCK(pcbinfo);
1578 error = in_pcballoc(so, pcbinfo);
1579 if (error) {
1580 INP_INFO_WUNLOCK(pcbinfo);
1581 return (error);
1582 }
1583
1584 inp = sotoinpcb(so);
1585 inp->inp_vflag |= INP_IPV4;
1586 inp->inp_ip_ttl = V_ip_defttl;
1587
1588 error = udp_newudpcb(inp);
1589 if (error) {
1590 in_pcbdetach(inp);
1591 in_pcbfree(inp);
1592 INP_INFO_WUNLOCK(pcbinfo);
1593 return (error);
1594 }
1595
1596 INP_WUNLOCK(inp);
1597 INP_INFO_WUNLOCK(pcbinfo);
1598 return (0);
1599 }
1600 #endif /* INET */
1601
1602 int
1603 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1604 {
1605 struct inpcb *inp;
1606 struct udpcb *up;
1607
1608 KASSERT(so->so_type == SOCK_DGRAM,
1609 ("udp_set_kernel_tunneling: !dgram"));
1610 inp = sotoinpcb(so);
1611 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1612 INP_WLOCK(inp);
1613 up = intoudpcb(inp);
1614 if ((up->u_tun_func != NULL) ||
1615 (up->u_icmp_func != NULL)) {
1616 INP_WUNLOCK(inp);
1617 return (EBUSY);
1618 }
1619 up->u_tun_func = f;
1620 up->u_icmp_func = i;
1621 up->u_tun_ctx = ctx;
1622 INP_WUNLOCK(inp);
1623 return (0);
1624 }
1625
1626 #ifdef INET
1627 static int
1628 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1629 {
1630 struct inpcb *inp;
1631 struct inpcbinfo *pcbinfo;
1632 int error;
1633
1634 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1635 inp = sotoinpcb(so);
1636 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1637 INP_WLOCK(inp);
1638 INP_HASH_WLOCK(pcbinfo);
1639 error = in_pcbbind(inp, nam, td->td_ucred);
1640 INP_HASH_WUNLOCK(pcbinfo);
1641 INP_WUNLOCK(inp);
1642 return (error);
1643 }
1644
1645 static void
1646 udp_close(struct socket *so)
1647 {
1648 struct inpcb *inp;
1649 struct inpcbinfo *pcbinfo;
1650
1651 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1652 inp = sotoinpcb(so);
1653 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1654 INP_WLOCK(inp);
1655 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1656 INP_HASH_WLOCK(pcbinfo);
1657 in_pcbdisconnect(inp);
1658 inp->inp_laddr.s_addr = INADDR_ANY;
1659 INP_HASH_WUNLOCK(pcbinfo);
1660 soisdisconnected(so);
1661 }
1662 INP_WUNLOCK(inp);
1663 }
1664
1665 static int
1666 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1667 {
1668 struct inpcb *inp;
1669 struct inpcbinfo *pcbinfo;
1670 struct sockaddr_in *sin;
1671 int error;
1672
1673 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1674 inp = sotoinpcb(so);
1675 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1676 INP_WLOCK(inp);
1677 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1678 INP_WUNLOCK(inp);
1679 return (EISCONN);
1680 }
1681 sin = (struct sockaddr_in *)nam;
1682 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1683 if (error != 0) {
1684 INP_WUNLOCK(inp);
1685 return (error);
1686 }
1687 INP_HASH_WLOCK(pcbinfo);
1688 error = in_pcbconnect(inp, nam, td->td_ucred);
1689 INP_HASH_WUNLOCK(pcbinfo);
1690 if (error == 0)
1691 soisconnected(so);
1692 INP_WUNLOCK(inp);
1693 return (error);
1694 }
1695
1696 static void
1697 udp_detach(struct socket *so)
1698 {
1699 struct inpcb *inp;
1700 struct inpcbinfo *pcbinfo;
1701 struct udpcb *up;
1702
1703 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1704 inp = sotoinpcb(so);
1705 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1706 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1707 ("udp_detach: not disconnected"));
1708 INP_INFO_WLOCK(pcbinfo);
1709 INP_WLOCK(inp);
1710 up = intoudpcb(inp);
1711 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1712 inp->inp_ppcb = NULL;
1713 in_pcbdetach(inp);
1714 in_pcbfree(inp);
1715 INP_INFO_WUNLOCK(pcbinfo);
1716 udp_discardcb(up);
1717 }
1718
1719 static int
1720 udp_disconnect(struct socket *so)
1721 {
1722 struct inpcb *inp;
1723 struct inpcbinfo *pcbinfo;
1724
1725 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1726 inp = sotoinpcb(so);
1727 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1728 INP_WLOCK(inp);
1729 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1730 INP_WUNLOCK(inp);
1731 return (ENOTCONN);
1732 }
1733 INP_HASH_WLOCK(pcbinfo);
1734 in_pcbdisconnect(inp);
1735 inp->inp_laddr.s_addr = INADDR_ANY;
1736 INP_HASH_WUNLOCK(pcbinfo);
1737 SOCK_LOCK(so);
1738 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1739 SOCK_UNLOCK(so);
1740 INP_WUNLOCK(inp);
1741 return (0);
1742 }
1743
1744 static int
1745 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1746 struct mbuf *control, struct thread *td)
1747 {
1748 struct inpcb *inp;
1749
1750 inp = sotoinpcb(so);
1751 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1752 return (udp_output(inp, m, addr, control, td));
1753 }
1754 #endif /* INET */
1755
1756 int
1757 udp_shutdown(struct socket *so)
1758 {
1759 struct inpcb *inp;
1760
1761 inp = sotoinpcb(so);
1762 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1763 INP_WLOCK(inp);
1764 socantsendmore(so);
1765 INP_WUNLOCK(inp);
1766 return (0);
1767 }
1768
1769 #ifdef INET
1770 struct pr_usrreqs udp_usrreqs = {
1771 .pru_abort = udp_abort,
1772 .pru_attach = udp_attach,
1773 .pru_bind = udp_bind,
1774 .pru_connect = udp_connect,
1775 .pru_control = in_control,
1776 .pru_detach = udp_detach,
1777 .pru_disconnect = udp_disconnect,
1778 .pru_peeraddr = in_getpeeraddr,
1779 .pru_send = udp_send,
1780 .pru_soreceive = soreceive_dgram,
1781 .pru_sosend = sosend_dgram,
1782 .pru_shutdown = udp_shutdown,
1783 .pru_sockaddr = in_getsockaddr,
1784 .pru_sosetlabel = in_pcbsosetlabel,
1785 .pru_close = udp_close,
1786 };
1787 #endif /* INET */
Cache object: 9849d62f70f0de1a63cea822a2f7f114
|