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.
8 * Portions of this software were developed by Robert N. M. Watson under
9 * contract to Juniper Networks, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
46 #include <sys/param.h>
47 #include <sys/domain.h>
48 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
52 #include <sys/malloc.h>
56 #include <sys/protosw.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/systm.h>
68 #include <net/route.h>
70 #include <netinet/in.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/in_var.h>
74 #include <netinet/ip.h>
76 #include <netinet/ip6.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/ip_options.h>
83 #include <netinet6/ip6_var.h>
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
89 #include <netipsec/ipsec.h>
90 #include <netipsec/esp.h>
93 #include <machine/in_cksum.h>
95 #include <security/mac/mac_framework.h>
98 * UDP protocol implementation.
99 * Per RFC 768, August, 1980.
103 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
104 * removes the only data integrity mechanism for packets and malformed
105 * packets that would otherwise be discarded due to bad checksums, and may
106 * cause problems (especially for NFS data blocks).
108 VNET_DEFINE(int, udp_cksum) = 1;
109 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
110 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
112 int udp_log_in_vain = 0;
113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
114 &udp_log_in_vain, 0, "Log all incoming UDP packets");
116 VNET_DEFINE(int, udp_blackhole) = 0;
117 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
118 &VNET_NAME(udp_blackhole), 0,
119 "Do not send port unreachables for refused connects");
121 u_long udp_sendspace = 9216; /* really max datagram size */
122 /* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
126 u_long udp_recvspace = 40 * (1024 +
128 sizeof(struct sockaddr_in6)
130 sizeof(struct sockaddr_in)
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
137 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
138 VNET_DEFINE(struct inpcbinfo, udbinfo);
139 static VNET_DEFINE(uma_zone_t, udpcb_zone);
140 #define V_udpcb_zone VNET(udpcb_zone)
143 #define UDBHASHSIZE 128
146 VNET_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
147 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
148 &VNET_NAME(udpstat), udpstat,
149 "UDP statistics (struct udpstat, netinet/udp_var.h)");
152 static void udp_detach(struct socket *so);
153 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
154 struct mbuf *, struct thread *);
159 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
161 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
163 #endif /* IPSEC_NAT_T */
167 udp_zone_change(void *tag)
170 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
171 uma_zone_set_max(V_udpcb_zone, maxsockets);
175 udp_inpcb_init(void *mem, int size, int flags)
180 INP_LOCK_INIT(inp, "inp", "udpinp");
188 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
189 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
190 IPI_HASHFIELDS_2TUPLE);
191 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 uma_zone_set_max(V_udpcb_zone, maxsockets);
194 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
195 EVENTHANDLER_PRI_ANY);
199 * Kernel module interface for updating udpstat. The argument is an index
200 * into udpstat treated as an array of u_long. While this encodes the
201 * general layout of udpstat into the caller, it doesn't encode its location,
202 * so that future changes to add, for example, per-CPU stats support won't
203 * cause binary compatibility problems for kernel modules.
206 kmod_udpstat_inc(int statnum)
209 (*((u_long *)&V_udpstat + statnum))++;
213 udp_newudpcb(struct inpcb *inp)
217 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
225 udp_discardcb(struct udpcb *up)
228 uma_zfree(V_udpcb_zone, up);
236 in_pcbinfo_destroy(&V_udbinfo);
237 uma_zdestroy(V_udpcb_zone);
243 * Subroutine of udp_input(), which appends the provided mbuf chain to the
244 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
245 * contains the source address. If the socket ends up being an IPv6 socket,
246 * udp_append() will convert to a sockaddr_in6 before passing the address
247 * into the socket code.
250 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
251 struct sockaddr_in *udp_in)
253 struct sockaddr *append_sa;
255 struct mbuf *opts = 0;
257 struct sockaddr_in6 udp_in6;
261 INP_LOCK_ASSERT(inp);
264 * Engage the tunneling protocol.
267 if (up->u_tun_func != NULL) {
268 (*up->u_tun_func)(n, off, inp);
275 off += sizeof(struct udphdr);
278 /* Check AH/ESP integrity. */
279 if (ipsec4_in_reject(n, inp)) {
281 IPSECSTAT_INC(in_polvio);
286 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
287 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
288 n = udp4_espdecap(inp, n, off);
289 if (n == NULL) /* Consumed. */
292 #endif /* IPSEC_NAT_T */
295 if (mac_inpcb_check_deliver(inp, n) != 0) {
300 if (inp->inp_flags & INP_CONTROLOPTS ||
301 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
303 if (inp->inp_vflag & INP_IPV6)
304 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
307 ip_savecontrol(inp, &opts, ip, n);
310 if (inp->inp_vflag & INP_IPV6) {
311 bzero(&udp_in6, sizeof(udp_in6));
312 udp_in6.sin6_len = sizeof(udp_in6);
313 udp_in6.sin6_family = AF_INET6;
314 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
315 append_sa = (struct sockaddr *)&udp_in6;
318 append_sa = (struct sockaddr *)udp_in;
321 so = inp->inp_socket;
322 SOCKBUF_LOCK(&so->so_rcv);
323 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
324 SOCKBUF_UNLOCK(&so->so_rcv);
328 UDPSTAT_INC(udps_fullsock);
330 sorwakeup_locked(so);
334 udp_input(struct mbuf *m, int off)
343 struct sockaddr_in udp_in;
344 struct m_tag *fwd_tag;
346 ifp = m->m_pkthdr.rcvif;
347 UDPSTAT_INC(udps_ipackets);
350 * Strip IP options, if any; should skip this, make available to
351 * user, and use on returned packets, but we don't yet have a way to
352 * check the checksum with options still present.
354 if (iphlen > sizeof (struct ip)) {
355 ip_stripoptions(m, (struct mbuf *)0);
356 iphlen = sizeof(struct ip);
360 * Get IP and UDP header together in first mbuf.
362 ip = mtod(m, struct ip *);
363 if (m->m_len < iphlen + sizeof(struct udphdr)) {
364 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
365 UDPSTAT_INC(udps_hdrops);
368 ip = mtod(m, struct ip *);
370 uh = (struct udphdr *)((caddr_t)ip + iphlen);
373 * Destination port of 0 is illegal, based on RFC768.
375 if (uh->uh_dport == 0)
379 * Construct sockaddr format source address. Stuff source address
380 * and datagram in user buffer.
382 bzero(&udp_in, sizeof(udp_in));
383 udp_in.sin_len = sizeof(udp_in);
384 udp_in.sin_family = AF_INET;
385 udp_in.sin_port = uh->uh_sport;
386 udp_in.sin_addr = ip->ip_src;
389 * Make mbuf data length reflect UDP length. If not enough data to
390 * reflect UDP length, drop.
392 len = ntohs((u_short)uh->uh_ulen);
393 if (ip->ip_len != len) {
394 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
395 UDPSTAT_INC(udps_badlen);
398 m_adj(m, len - ip->ip_len);
399 /* ip->ip_len = len; */
403 * Save a copy of the IP header in case we want restore it for
404 * sending an ICMP error message in response.
406 if (!V_udp_blackhole)
409 memset(&save_ip, 0, sizeof(save_ip));
412 * Checksum extended UDP header and data.
417 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
418 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
419 uh_sum = m->m_pkthdr.csum_data;
421 uh_sum = in_pseudo(ip->ip_src.s_addr,
422 ip->ip_dst.s_addr, htonl((u_short)len +
423 m->m_pkthdr.csum_data + IPPROTO_UDP));
428 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
429 bzero(((struct ipovly *)ip)->ih_x1, 9);
430 ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
431 uh_sum = in_cksum(m, len + sizeof (struct ip));
432 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
435 UDPSTAT_INC(udps_badsum);
440 UDPSTAT_INC(udps_nosum);
442 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
443 in_broadcast(ip->ip_dst, ifp)) {
445 struct ip_moptions *imo;
447 INP_INFO_RLOCK(&V_udbinfo);
449 LIST_FOREACH(inp, &V_udb, inp_list) {
450 if (inp->inp_lport != uh->uh_dport)
453 if ((inp->inp_vflag & INP_IPV4) == 0)
456 if (inp->inp_laddr.s_addr != INADDR_ANY &&
457 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
459 if (inp->inp_faddr.s_addr != INADDR_ANY &&
460 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
462 if (inp->inp_fport != 0 &&
463 inp->inp_fport != uh->uh_sport)
469 * XXXRW: Because we weren't holding either the inpcb
470 * or the hash lock when we checked for a match
471 * before, we should probably recheck now that the
472 * inpcb lock is held.
476 * Handle socket delivery policy for any-source
477 * and source-specific multicast. [RFC3678]
479 imo = inp->inp_moptions;
480 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
481 struct sockaddr_in group;
487 bzero(&group, sizeof(struct sockaddr_in));
488 group.sin_len = sizeof(struct sockaddr_in);
489 group.sin_family = AF_INET;
490 group.sin_addr = ip->ip_dst;
492 blocked = imo_multi_filter(imo, ifp,
493 (struct sockaddr *)&group,
494 (struct sockaddr *)&udp_in);
495 if (blocked != MCAST_PASS) {
496 if (blocked == MCAST_NOTGMEMBER)
497 IPSTAT_INC(ips_notmember);
498 if (blocked == MCAST_NOTSMEMBER ||
499 blocked == MCAST_MUTED)
500 UDPSTAT_INC(udps_filtermcast);
508 n = m_copy(m, 0, M_COPYALL);
509 udp_append(last, ip, n, iphlen, &udp_in);
514 * Don't look for additional matches if this one does
515 * not have either the SO_REUSEPORT or SO_REUSEADDR
516 * socket options set. This heuristic avoids
517 * searching through all pcbs in the common case of a
518 * non-shared port. It assumes that an application
519 * will never clear these options after setting them.
521 if ((last->inp_socket->so_options &
522 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
528 * No matching pcb found; discard datagram. (No need
529 * to send an ICMP Port Unreachable for a broadcast
530 * or multicast datgram.)
532 UDPSTAT_INC(udps_noportbcast);
535 INP_INFO_RUNLOCK(&V_udbinfo);
538 udp_append(last, ip, m, iphlen, &udp_in);
540 INP_INFO_RUNLOCK(&V_udbinfo);
545 * Locate pcb for datagram.
549 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
551 if ((m->m_flags & M_IP_NEXTHOP) &&
552 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
553 struct sockaddr_in *next_hop;
555 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
558 * Transparently forwarded. Pretend to be the destination.
559 * Already got one like this?
561 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
562 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
565 * It's new. Try to find the ambushing socket.
566 * Because we've rewritten the destination address,
567 * any hardware-generated hash is ignored.
569 inp = in_pcblookup(&V_udbinfo, ip->ip_src,
570 uh->uh_sport, next_hop->sin_addr,
571 next_hop->sin_port ? htons(next_hop->sin_port) :
572 uh->uh_dport, INPLOOKUP_WILDCARD |
573 INPLOOKUP_RLOCKPCB, ifp);
575 /* Remove the tag from the packet. We don't need it anymore. */
576 m_tag_delete(m, fwd_tag);
577 m->m_flags &= ~M_IP_NEXTHOP;
579 inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
580 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
581 INPLOOKUP_RLOCKPCB, ifp, m);
583 if (udp_log_in_vain) {
584 char buf[4*sizeof "123"];
586 strcpy(buf, inet_ntoa(ip->ip_dst));
588 "Connection attempt to UDP %s:%d from %s:%d\n",
589 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
590 ntohs(uh->uh_sport));
592 UDPSTAT_INC(udps_noport);
593 if (m->m_flags & (M_BCAST | M_MCAST)) {
594 UDPSTAT_INC(udps_noportbcast);
599 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
602 ip->ip_len += iphlen;
603 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
608 * Check the minimum TTL for socket.
610 INP_RLOCK_ASSERT(inp);
611 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
616 udp_append(inp, ip, m, iphlen, &udp_in);
626 * Notify a udp user of an asynchronous error; just wake up so that they can
627 * collect error status.
630 udp_notify(struct inpcb *inp, int errno)
634 * While udp_ctlinput() always calls udp_notify() with a read lock
635 * when invoking it directly, in_pcbnotifyall() currently uses write
636 * locks due to sharing code with TCP. For now, accept either a read
637 * or a write lock, but a read lock is sufficient.
639 INP_LOCK_ASSERT(inp);
641 inp->inp_socket->so_error = errno;
642 sorwakeup(inp->inp_socket);
643 sowwakeup(inp->inp_socket);
649 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
653 struct in_addr faddr;
656 faddr = ((struct sockaddr_in *)sa)->sin_addr;
657 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
661 * Redirects don't need to be handled up here.
663 if (PRC_IS_REDIRECT(cmd))
667 * Hostdead is ugly because it goes linearly through all PCBs.
669 * XXX: We never get this from ICMP, otherwise it makes an excellent
670 * DoS attack on machines with many connections.
672 if (cmd == PRC_HOSTDEAD)
674 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
677 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
678 inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
679 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
681 INP_RLOCK_ASSERT(inp);
682 if (inp->inp_socket != NULL) {
683 udp_notify(inp, inetctlerrmap[cmd]);
688 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
694 udp_pcblist(SYSCTL_HANDLER_ARGS)
697 struct inpcb *inp, **inp_list;
702 * The process of preparing the PCB list is too time-consuming and
703 * resource-intensive to repeat twice on every request.
705 if (req->oldptr == 0) {
706 n = V_udbinfo.ipi_count;
707 n += imax(n / 8, 10);
708 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
712 if (req->newptr != 0)
716 * OK, now we're committed to doing something.
718 INP_INFO_RLOCK(&V_udbinfo);
719 gencnt = V_udbinfo.ipi_gencnt;
720 n = V_udbinfo.ipi_count;
721 INP_INFO_RUNLOCK(&V_udbinfo);
723 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
724 + n * sizeof(struct xinpcb));
728 xig.xig_len = sizeof xig;
730 xig.xig_gen = gencnt;
731 xig.xig_sogen = so_gencnt;
732 error = SYSCTL_OUT(req, &xig, sizeof xig);
736 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
740 INP_INFO_RLOCK(&V_udbinfo);
741 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
742 inp = LIST_NEXT(inp, inp_list)) {
744 if (inp->inp_gencnt <= gencnt &&
745 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
751 INP_INFO_RUNLOCK(&V_udbinfo);
755 for (i = 0; i < n; i++) {
758 if (inp->inp_gencnt <= gencnt) {
761 bzero(&xi, sizeof(xi));
762 xi.xi_len = sizeof xi;
763 /* XXX should avoid extra copy */
764 bcopy(inp, &xi.xi_inp, sizeof *inp);
766 sotoxsocket(inp->inp_socket, &xi.xi_socket);
767 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
769 error = SYSCTL_OUT(req, &xi, sizeof xi);
773 INP_INFO_WLOCK(&V_udbinfo);
774 for (i = 0; i < n; i++) {
777 if (!in_pcbrele_rlocked(inp))
780 INP_INFO_WUNLOCK(&V_udbinfo);
784 * Give the user an updated idea of our state. If the
785 * generation differs from what we told her before, she knows
786 * that something happened while we were processing this
787 * request, and it might be necessary to retry.
789 INP_INFO_RLOCK(&V_udbinfo);
790 xig.xig_gen = V_udbinfo.ipi_gencnt;
791 xig.xig_sogen = so_gencnt;
792 xig.xig_count = V_udbinfo.ipi_count;
793 INP_INFO_RUNLOCK(&V_udbinfo);
794 error = SYSCTL_OUT(req, &xig, sizeof xig);
796 free(inp_list, M_TEMP);
800 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
801 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
802 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
806 udp_getcred(SYSCTL_HANDLER_ARGS)
809 struct sockaddr_in addrs[2];
813 error = priv_check(req->td, PRIV_NETINET_GETCRED);
816 error = SYSCTL_IN(req, addrs, sizeof(addrs));
819 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
820 addrs[0].sin_addr, addrs[0].sin_port,
821 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
823 INP_RLOCK_ASSERT(inp);
824 if (inp->inp_socket == NULL)
827 error = cr_canseeinpcb(req->td->td_ucred, inp);
829 cru2x(inp->inp_cred, &xuc);
834 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
838 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
839 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
840 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
844 udp_ctloutput(struct socket *so, struct sockopt *sopt)
846 int error = 0, optval;
853 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
855 if (sopt->sopt_level != IPPROTO_UDP) {
857 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
859 error = ip6_ctloutput(so, sopt);
862 #if defined(INET) && defined(INET6)
868 error = ip_ctloutput(so, sopt);
874 switch (sopt->sopt_dir) {
876 switch (sopt->sopt_name) {
879 error = sooptcopyin(sopt, &optval, sizeof optval,
884 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
888 KASSERT(up != NULL, ("%s: up == NULL", __func__));
892 /* Clear all UDP encap. */
894 up->u_flags &= ~UF_ESPINUDP_ALL;
898 case UDP_ENCAP_ESPINUDP:
899 case UDP_ENCAP_ESPINUDP_NON_IKE:
900 up->u_flags &= ~UF_ESPINUDP_ALL;
901 if (optval == UDP_ENCAP_ESPINUDP)
902 up->u_flags |= UF_ESPINUDP;
903 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
904 up->u_flags |= UF_ESPINUDP_NON_IKE;
920 switch (sopt->sopt_name) {
924 KASSERT(up != NULL, ("%s: up == NULL", __func__));
925 optval = up->u_flags & UF_ESPINUDP_ALL;
927 error = sooptcopyout(sopt, &optval, sizeof optval);
943 #define UH_UNLOCKED 0
945 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
946 struct mbuf *control, struct thread *td)
949 int len = m->m_pkthdr.len;
950 struct in_addr faddr, laddr;
952 struct sockaddr_in *sin, src;
955 u_short fport, lport;
960 * udp_output() may need to temporarily bind or connect the current
961 * inpcb. As such, we don't know up front whether we will need the
962 * pcbinfo lock or not. Do any work to decide what is needed up
963 * front before acquiring any locks.
965 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
974 tos = inp->inp_ip_tos;
975 if (control != NULL) {
977 * XXX: Currently, we assume all the optional information is
978 * stored in a single mbuf.
980 if (control->m_next) {
986 for (; control->m_len > 0;
987 control->m_data += CMSG_ALIGN(cm->cmsg_len),
988 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
989 cm = mtod(control, struct cmsghdr *);
990 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
991 || cm->cmsg_len > control->m_len) {
995 if (cm->cmsg_level != IPPROTO_IP)
998 switch (cm->cmsg_type) {
1001 CMSG_LEN(sizeof(struct in_addr))) {
1005 bzero(&src, sizeof(src));
1006 src.sin_family = AF_INET;
1007 src.sin_len = sizeof(src);
1008 src.sin_port = inp->inp_lport;
1010 *(struct in_addr *)CMSG_DATA(cm);
1014 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1018 tos = *(u_char *)CMSG_DATA(cm);
1022 error = ENOPROTOOPT;
1037 * Depending on whether or not the application has bound or connected
1038 * the socket, we may have to do varying levels of work. The optimal
1039 * case is for a connected UDP socket, as a global lock isn't
1042 * In order to decide which we need, we require stability of the
1043 * inpcb binding, which we ensure by acquiring a read lock on the
1044 * inpcb. This doesn't strictly follow the lock order, so we play
1045 * the trylock and retry game; note that we may end up with more
1046 * conservative locks than required the second time around, so later
1047 * assertions have to accept that. Further analysis of the number of
1048 * misses under contention is required.
1050 * XXXRW: Check that hash locking update here is correct.
1052 sin = (struct sockaddr_in *)addr;
1054 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1057 INP_HASH_WLOCK(&V_udbinfo);
1058 unlock_udbinfo = UH_WLOCKED;
1059 } else if ((sin != NULL && (
1060 (sin->sin_addr.s_addr == INADDR_ANY) ||
1061 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1062 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1063 (inp->inp_lport == 0))) ||
1064 (src.sin_family == AF_INET)) {
1065 INP_HASH_RLOCK(&V_udbinfo);
1066 unlock_udbinfo = UH_RLOCKED;
1068 unlock_udbinfo = UH_UNLOCKED;
1071 * If the IP_SENDSRCADDR control message was specified, override the
1072 * source address for this datagram. Its use is invalidated if the
1073 * address thus specified is incomplete or clobbers other inpcbs.
1075 laddr = inp->inp_laddr;
1076 lport = inp->inp_lport;
1077 if (src.sin_family == AF_INET) {
1078 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1080 (laddr.s_addr == INADDR_ANY &&
1081 src.sin_addr.s_addr == INADDR_ANY)) {
1085 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1086 &laddr.s_addr, &lport, td->td_ucred);
1092 * If a UDP socket has been connected, then a local address/port will
1093 * have been selected and bound.
1095 * If a UDP socket has not been connected to, then an explicit
1096 * destination address must be used, in which case a local
1097 * address/port may not have been selected and bound.
1100 INP_LOCK_ASSERT(inp);
1101 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1107 * Jail may rewrite the destination address, so let it do
1108 * that before we use it.
1110 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1115 * If a local address or port hasn't yet been selected, or if
1116 * the destination address needs to be rewritten due to using
1117 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1118 * to do the heavy lifting. Once a port is selected, we
1119 * commit the binding back to the socket; we also commit the
1120 * binding of the address if in jail.
1122 * If we already have a valid binding and we're not
1123 * requesting a destination address rewrite, use a fast path.
1125 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1126 inp->inp_lport == 0 ||
1127 sin->sin_addr.s_addr == INADDR_ANY ||
1128 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1129 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1130 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1131 &lport, &faddr.s_addr, &fport, NULL,
1137 * XXXRW: Why not commit the port if the address is
1140 /* Commit the local port if newly assigned. */
1141 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1142 inp->inp_lport == 0) {
1143 INP_WLOCK_ASSERT(inp);
1144 INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1146 * Remember addr if jailed, to prevent
1149 if (prison_flag(td->td_ucred, PR_IP4))
1150 inp->inp_laddr = laddr;
1151 inp->inp_lport = lport;
1152 if (in_pcbinshash(inp) != 0) {
1157 inp->inp_flags |= INP_ANONPORT;
1160 faddr = sin->sin_addr;
1161 fport = sin->sin_port;
1164 INP_LOCK_ASSERT(inp);
1165 faddr = inp->inp_faddr;
1166 fport = inp->inp_fport;
1167 if (faddr.s_addr == INADDR_ANY) {
1174 * Calculate data length and get a mbuf for UDP, IP, and possible
1175 * link-layer headers. Immediate slide the data pointer back forward
1176 * since we won't use that space at this layer.
1178 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1183 m->m_data += max_linkhdr;
1184 m->m_len -= max_linkhdr;
1185 m->m_pkthdr.len -= max_linkhdr;
1188 * Fill in mbuf with extended UDP header and addresses and length put
1189 * into network format.
1191 ui = mtod(m, struct udpiphdr *);
1192 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
1193 ui->ui_pr = IPPROTO_UDP;
1196 ui->ui_sport = lport;
1197 ui->ui_dport = fport;
1198 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1201 * Set the Don't Fragment bit in the IP header.
1203 if (inp->inp_flags & INP_DONTFRAG) {
1206 ip = (struct ip *)&ui->ui_i;
1207 ip->ip_off |= IP_DF;
1211 if (inp->inp_socket->so_options & SO_DONTROUTE)
1212 ipflags |= IP_ROUTETOIF;
1213 if (inp->inp_socket->so_options & SO_BROADCAST)
1214 ipflags |= IP_ALLOWBROADCAST;
1215 if (inp->inp_flags & INP_ONESBCAST)
1216 ipflags |= IP_SENDONES;
1219 mac_inpcb_create_mbuf(inp, m);
1223 * Set up checksum and output datagram.
1226 if (inp->inp_flags & INP_ONESBCAST)
1227 faddr.s_addr = INADDR_BROADCAST;
1228 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1229 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1230 m->m_pkthdr.csum_flags = CSUM_UDP;
1231 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1234 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1235 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1236 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1237 UDPSTAT_INC(udps_opackets);
1239 if (unlock_udbinfo == UH_WLOCKED)
1240 INP_HASH_WUNLOCK(&V_udbinfo);
1241 else if (unlock_udbinfo == UH_RLOCKED)
1242 INP_HASH_RUNLOCK(&V_udbinfo);
1243 error = ip_output(m, inp->inp_options, NULL, ipflags,
1244 inp->inp_moptions, inp);
1245 if (unlock_udbinfo == UH_WLOCKED)
1252 if (unlock_udbinfo == UH_WLOCKED) {
1253 INP_HASH_WUNLOCK(&V_udbinfo);
1255 } else if (unlock_udbinfo == UH_RLOCKED) {
1256 INP_HASH_RUNLOCK(&V_udbinfo);
1265 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1267 * Potentially decap ESP in UDP frame. Check for an ESP header
1268 * and optional marker; if present, strip the UDP header and
1269 * push the result through IPSec.
1271 * Returns mbuf to be processed (potentially re-allocated) or
1272 * NULL if consumed and/or processed.
1274 static struct mbuf *
1275 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1277 size_t minlen, payload, skip, iphlen;
1281 struct udphdr *udphdr;
1284 INP_RLOCK_ASSERT(inp);
1287 * Pull up data so the longest case is contiguous:
1288 * IP/UDP hdr + non ESP marker + ESP hdr.
1290 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1291 if (minlen > m->m_pkthdr.len)
1292 minlen = m->m_pkthdr.len;
1293 if ((m = m_pullup(m, minlen)) == NULL) {
1294 IPSECSTAT_INC(in_inval);
1295 return (NULL); /* Bypass caller processing. */
1297 data = mtod(m, caddr_t); /* Points to ip header. */
1298 payload = m->m_len - off; /* Size of payload. */
1300 if (payload == 1 && data[off] == '\xff')
1301 return (m); /* NB: keepalive packet, no decap. */
1303 up = intoudpcb(inp);
1304 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1305 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1306 ("u_flags 0x%x", up->u_flags));
1309 * Check that the payload is large enough to hold an
1310 * ESP header and compute the amount of data to remove.
1312 * NB: the caller has already done a pullup for us.
1313 * XXX can we assume alignment and eliminate bcopys?
1315 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1317 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1318 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1319 * possible AH mode non-IKE marker+non-ESP marker
1320 * from draft-ietf-ipsec-udp-encaps-00.txt.
1324 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1325 return (m); /* NB: no decap. */
1326 bcopy(data + off, &marker, sizeof(uint64_t));
1327 if (marker != 0) /* Non-IKE marker. */
1328 return (m); /* NB: no decap. */
1329 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1333 if (payload <= sizeof(struct esp)) {
1334 IPSECSTAT_INC(in_inval);
1336 return (NULL); /* Discard. */
1338 bcopy(data + off, &spi, sizeof(uint32_t));
1339 if (spi == 0) /* Non-ESP marker. */
1340 return (m); /* NB: no decap. */
1341 skip = sizeof(struct udphdr);
1345 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1346 * the UDP ports. This is required if we want to select
1347 * the right SPD for multiple hosts behind same NAT.
1349 * NB: ports are maintained in network byte order everywhere
1350 * in the NAT-T code.
1352 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1353 2 * sizeof(uint16_t), M_NOWAIT);
1355 IPSECSTAT_INC(in_nomem);
1357 return (NULL); /* Discard. */
1359 iphlen = off - sizeof(struct udphdr);
1360 udphdr = (struct udphdr *)(data + iphlen);
1361 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1362 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1363 m_tag_prepend(m, tag);
1366 * Remove the UDP header (and possibly the non ESP marker)
1367 * IP header length is iphlen
1370 * +----+------+-----+
1371 * | IP | UDP | ESP |
1372 * +----+------+-----+
1380 ovbcopy(data, data + skip, iphlen);
1383 ip = mtod(m, struct ip *);
1385 ip->ip_p = IPPROTO_ESP;
1388 * We cannot yet update the cksums so clear any
1389 * h/w cksum flags as they are no longer valid.
1391 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1392 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1394 (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1395 return (NULL); /* NB: consumed, bypass processing. */
1397 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1400 udp_abort(struct socket *so)
1404 inp = sotoinpcb(so);
1405 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1407 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1408 INP_HASH_WLOCK(&V_udbinfo);
1409 in_pcbdisconnect(inp);
1410 inp->inp_laddr.s_addr = INADDR_ANY;
1411 INP_HASH_WUNLOCK(&V_udbinfo);
1412 soisdisconnected(so);
1418 udp_attach(struct socket *so, int proto, struct thread *td)
1423 inp = sotoinpcb(so);
1424 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1425 error = soreserve(so, udp_sendspace, udp_recvspace);
1428 INP_INFO_WLOCK(&V_udbinfo);
1429 error = in_pcballoc(so, &V_udbinfo);
1431 INP_INFO_WUNLOCK(&V_udbinfo);
1435 inp = sotoinpcb(so);
1436 inp->inp_vflag |= INP_IPV4;
1437 inp->inp_ip_ttl = V_ip_defttl;
1439 error = udp_newudpcb(inp);
1443 INP_INFO_WUNLOCK(&V_udbinfo);
1448 INP_INFO_WUNLOCK(&V_udbinfo);
1454 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1459 KASSERT(so->so_type == SOCK_DGRAM,
1460 ("udp_set_kernel_tunneling: !dgram"));
1461 inp = sotoinpcb(so);
1462 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1464 up = intoudpcb(inp);
1465 if (up->u_tun_func != NULL) {
1476 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1481 inp = sotoinpcb(so);
1482 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1484 INP_HASH_WLOCK(&V_udbinfo);
1485 error = in_pcbbind(inp, nam, td->td_ucred);
1486 INP_HASH_WUNLOCK(&V_udbinfo);
1492 udp_close(struct socket *so)
1496 inp = sotoinpcb(so);
1497 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1499 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1500 INP_HASH_WLOCK(&V_udbinfo);
1501 in_pcbdisconnect(inp);
1502 inp->inp_laddr.s_addr = INADDR_ANY;
1503 INP_HASH_WUNLOCK(&V_udbinfo);
1504 soisdisconnected(so);
1510 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1514 struct sockaddr_in *sin;
1516 inp = sotoinpcb(so);
1517 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1519 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1523 sin = (struct sockaddr_in *)nam;
1524 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1529 INP_HASH_WLOCK(&V_udbinfo);
1530 error = in_pcbconnect(inp, nam, td->td_ucred);
1531 INP_HASH_WUNLOCK(&V_udbinfo);
1539 udp_detach(struct socket *so)
1544 inp = sotoinpcb(so);
1545 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1546 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1547 ("udp_detach: not disconnected"));
1548 INP_INFO_WLOCK(&V_udbinfo);
1550 up = intoudpcb(inp);
1551 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1552 inp->inp_ppcb = NULL;
1555 INP_INFO_WUNLOCK(&V_udbinfo);
1560 udp_disconnect(struct socket *so)
1564 inp = sotoinpcb(so);
1565 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1567 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1571 INP_HASH_WLOCK(&V_udbinfo);
1572 in_pcbdisconnect(inp);
1573 inp->inp_laddr.s_addr = INADDR_ANY;
1574 INP_HASH_WUNLOCK(&V_udbinfo);
1576 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1583 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1584 struct mbuf *control, struct thread *td)
1588 inp = sotoinpcb(so);
1589 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1590 return (udp_output(inp, m, addr, control, td));
1595 udp_shutdown(struct socket *so)
1599 inp = sotoinpcb(so);
1600 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1608 struct pr_usrreqs udp_usrreqs = {
1609 .pru_abort = udp_abort,
1610 .pru_attach = udp_attach,
1611 .pru_bind = udp_bind,
1612 .pru_connect = udp_connect,
1613 .pru_control = in_control,
1614 .pru_detach = udp_detach,
1615 .pru_disconnect = udp_disconnect,
1616 .pru_peeraddr = in_getpeeraddr,
1617 .pru_send = udp_send,
1618 .pru_soreceive = soreceive_dgram,
1619 .pru_sosend = sosend_dgram,
1620 .pru_shutdown = udp_shutdown,
1621 .pru_sockaddr = in_getsockaddr,
1622 .pru_sosetlabel = in_pcbsosetlabel,
1623 .pru_close = udp_close,