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
9 * Portions of this software were developed by Robert N. M. Watson under
10 * contract to Juniper Networks, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
36 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
48 #include <sys/param.h>
49 #include <sys/domain.h>
50 #include <sys/eventhandler.h>
52 #include <sys/kernel.h>
54 #include <sys/malloc.h>
58 #include <sys/protosw.h>
60 #include <sys/signalvar.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
64 #include <sys/sysctl.h>
65 #include <sys/syslog.h>
66 #include <sys/systm.h>
71 #include <net/if_var.h>
72 #include <net/route.h>
73 #include <net/rss_config.h>
75 #include <netinet/in.h>
76 #include <netinet/in_kdtrace.h>
77 #include <netinet/in_pcb.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
82 #include <netinet/ip6.h>
84 #include <netinet/ip_icmp.h>
85 #include <netinet/icmp_var.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_options.h>
89 #include <netinet6/ip6_var.h>
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
93 #include <netinet/udplite.h>
94 #include <netinet/in_rss.h>
97 #include <netipsec/ipsec.h>
98 #include <netipsec/esp.h>
101 #include <machine/in_cksum.h>
103 #include <security/mac/mac_framework.h>
106 * UDP and UDP-Lite protocols implementation.
107 * Per RFC 768, August, 1980.
108 * Per RFC 3828, July, 2004.
112 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums
113 * removes the only data integrity mechanism for packets and malformed
114 * packets that would otherwise be discarded due to bad checksums, and may
115 * cause problems (especially for NFS data blocks).
117 VNET_DEFINE(int, udp_cksum) = 1;
118 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
119 &VNET_NAME(udp_cksum), 0, "compute udp checksum");
121 int udp_log_in_vain = 0;
122 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
123 &udp_log_in_vain, 0, "Log all incoming UDP packets");
125 VNET_DEFINE(int, udp_blackhole) = 0;
126 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
127 &VNET_NAME(udp_blackhole), 0,
128 "Do not send port unreachables for refused connects");
130 u_long udp_sendspace = 9216; /* really max datagram size */
131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
132 &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
134 u_long udp_recvspace = 40 * (1024 +
136 sizeof(struct sockaddr_in6)
138 sizeof(struct sockaddr_in)
140 ); /* 40 1K datagrams */
142 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
143 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
145 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */
146 VNET_DEFINE(struct inpcbinfo, udbinfo);
147 VNET_DEFINE(struct inpcbhead, ulitecb);
148 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
149 static VNET_DEFINE(uma_zone_t, udpcb_zone);
150 #define V_udpcb_zone VNET(udpcb_zone)
153 #define UDBHASHSIZE 128
156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */
157 VNET_PCPUSTAT_SYSINIT(udpstat);
158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
159 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
162 VNET_PCPUSTAT_SYSUNINIT(udpstat);
165 static void udp_detach(struct socket *so);
166 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
167 struct mbuf *, struct thread *);
172 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
174 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
176 #endif /* IPSEC_NAT_T */
180 udp_zone_change(void *tag)
183 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
184 uma_zone_set_max(V_udpcb_zone, maxsockets);
188 udp_inpcb_init(void *mem, int size, int flags)
193 INP_LOCK_INIT(inp, "inp", "udpinp");
198 udplite_inpcb_init(void *mem, int size, int flags)
203 INP_LOCK_INIT(inp, "inp", "udpliteinp");
212 * For now default to 2-tuple UDP hashing - until the fragment
213 * reassembly code can also update the flowid.
215 * Once we can calculate the flowid that way and re-establish
216 * a 4-tuple, flip this to 4-tuple.
218 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
219 "udp_inpcb", udp_inpcb_init, NULL, 0,
220 IPI_HASHFIELDS_2TUPLE);
221 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
223 uma_zone_set_max(V_udpcb_zone, maxsockets);
224 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
225 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
226 EVENTHANDLER_PRI_ANY);
233 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
234 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
235 0, IPI_HASHFIELDS_2TUPLE);
239 * Kernel module interface for updating udpstat. The argument is an index
240 * into udpstat treated as an array of u_long. While this encodes the
241 * general layout of udpstat into the caller, it doesn't encode its location,
242 * so that future changes to add, for example, per-CPU stats support won't
243 * cause binary compatibility problems for kernel modules.
246 kmod_udpstat_inc(int statnum)
249 counter_u64_add(VNET(udpstat)[statnum], 1);
253 udp_newudpcb(struct inpcb *inp)
257 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
265 udp_discardcb(struct udpcb *up)
268 uma_zfree(V_udpcb_zone, up);
276 in_pcbinfo_destroy(&V_udbinfo);
277 uma_zdestroy(V_udpcb_zone);
281 udplite_destroy(void)
284 in_pcbinfo_destroy(&V_ulitecbinfo);
290 * Subroutine of udp_input(), which appends the provided mbuf chain to the
291 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that
292 * contains the source address. If the socket ends up being an IPv6 socket,
293 * udp_append() will convert to a sockaddr_in6 before passing the address
294 * into the socket code.
296 * In the normal case udp_append() will return 0, indicating that you
297 * must unlock the inp. However if a tunneling protocol is in place we increment
298 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
299 * then decrement the reference count. If the inp_rele returns 1, indicating the
300 * inp is gone, we return that to the caller to tell them *not* to unlock
301 * the inp. In the case of multi-cast this will cause the distribution
302 * to stop (though most tunneling protocols known currently do *not* use
306 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
307 struct sockaddr_in *udp_in)
309 struct sockaddr *append_sa;
311 struct mbuf *opts = 0;
313 struct sockaddr_in6 udp_in6;
317 INP_LOCK_ASSERT(inp);
320 * Engage the tunneling protocol.
323 if (up->u_tun_func != NULL) {
326 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
329 return (in_pcbrele_rlocked(inp));
332 off += sizeof(struct udphdr);
335 /* Check AH/ESP integrity. */
336 if (ipsec4_in_reject(n, inp)) {
342 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
343 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */
344 n = udp4_espdecap(inp, n, off);
345 if (n == NULL) /* Consumed. */
348 #endif /* IPSEC_NAT_T */
351 if (mac_inpcb_check_deliver(inp, n) != 0) {
356 if (inp->inp_flags & INP_CONTROLOPTS ||
357 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
359 if (inp->inp_vflag & INP_IPV6)
360 (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
363 ip_savecontrol(inp, &opts, ip, n);
366 if (inp->inp_vflag & INP_IPV6) {
367 bzero(&udp_in6, sizeof(udp_in6));
368 udp_in6.sin6_len = sizeof(udp_in6);
369 udp_in6.sin6_family = AF_INET6;
370 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
371 append_sa = (struct sockaddr *)&udp_in6;
374 append_sa = (struct sockaddr *)udp_in;
377 so = inp->inp_socket;
378 SOCKBUF_LOCK(&so->so_rcv);
379 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
380 SOCKBUF_UNLOCK(&so->so_rcv);
384 UDPSTAT_INC(udps_fullsock);
386 sorwakeup_locked(so);
391 udp_input(struct mbuf **mp, int *offp, int proto)
397 uint16_t len, ip_len;
398 struct inpcbinfo *pcbinfo;
400 struct sockaddr_in udp_in;
402 struct m_tag *fwd_tag;
403 int cscov_partial, iphlen;
407 ifp = m->m_pkthdr.rcvif;
409 UDPSTAT_INC(udps_ipackets);
412 * Strip IP options, if any; should skip this, make available to
413 * user, and use on returned packets, but we don't yet have a way to
414 * check the checksum with options still present.
416 if (iphlen > sizeof (struct ip)) {
418 iphlen = sizeof(struct ip);
422 * Get IP and UDP header together in first mbuf.
424 ip = mtod(m, struct ip *);
425 if (m->m_len < iphlen + sizeof(struct udphdr)) {
426 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
427 UDPSTAT_INC(udps_hdrops);
428 return (IPPROTO_DONE);
430 ip = mtod(m, struct ip *);
432 uh = (struct udphdr *)((caddr_t)ip + iphlen);
433 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
436 * Destination port of 0 is illegal, based on RFC768.
438 if (uh->uh_dport == 0)
442 * Construct sockaddr format source address. Stuff source address
443 * and datagram in user buffer.
445 bzero(&udp_in, sizeof(udp_in));
446 udp_in.sin_len = sizeof(udp_in);
447 udp_in.sin_family = AF_INET;
448 udp_in.sin_port = uh->uh_sport;
449 udp_in.sin_addr = ip->ip_src;
452 * Make mbuf data length reflect UDP length. If not enough data to
453 * reflect UDP length, drop.
455 len = ntohs((u_short)uh->uh_ulen);
456 ip_len = ntohs(ip->ip_len) - iphlen;
457 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
458 /* Zero means checksum over the complete packet. */
464 if (len > ip_len || len < sizeof(struct udphdr)) {
465 UDPSTAT_INC(udps_badlen);
468 if (proto == IPPROTO_UDP)
469 m_adj(m, len - ip_len);
473 * Save a copy of the IP header in case we want restore it for
474 * sending an ICMP error message in response.
476 if (!V_udp_blackhole)
479 memset(&save_ip, 0, sizeof(save_ip));
482 * Checksum extended UDP header and data.
487 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
489 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
490 uh_sum = m->m_pkthdr.csum_data;
492 uh_sum = in_pseudo(ip->ip_src.s_addr,
493 ip->ip_dst.s_addr, htonl((u_short)len +
494 m->m_pkthdr.csum_data + proto));
499 bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
500 bzero(((struct ipovly *)ip)->ih_x1, 9);
501 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
502 uh->uh_ulen : htons(ip_len);
503 uh_sum = in_cksum(m, len + sizeof (struct ip));
504 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
507 UDPSTAT_INC(udps_badsum);
509 return (IPPROTO_DONE);
512 if (proto == IPPROTO_UDP) {
513 UDPSTAT_INC(udps_nosum);
515 /* UDPLite requires a checksum */
516 /* XXX: What is the right UDPLite MIB counter here? */
518 return (IPPROTO_DONE);
522 pcbinfo = udp_get_inpcbinfo(proto);
523 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
524 in_broadcast(ip->ip_dst, ifp)) {
526 struct inpcbhead *pcblist;
527 struct ip_moptions *imo;
529 INP_INFO_RLOCK(pcbinfo);
530 pcblist = udp_get_pcblist(proto);
532 LIST_FOREACH(inp, pcblist, inp_list) {
533 if (inp->inp_lport != uh->uh_dport)
536 if ((inp->inp_vflag & INP_IPV4) == 0)
539 if (inp->inp_laddr.s_addr != INADDR_ANY &&
540 inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
542 if (inp->inp_faddr.s_addr != INADDR_ANY &&
543 inp->inp_faddr.s_addr != ip->ip_src.s_addr)
545 if (inp->inp_fport != 0 &&
546 inp->inp_fport != uh->uh_sport)
552 * XXXRW: Because we weren't holding either the inpcb
553 * or the hash lock when we checked for a match
554 * before, we should probably recheck now that the
555 * inpcb lock is held.
559 * Handle socket delivery policy for any-source
560 * and source-specific multicast. [RFC3678]
562 imo = inp->inp_moptions;
563 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
564 struct sockaddr_in group;
570 bzero(&group, sizeof(struct sockaddr_in));
571 group.sin_len = sizeof(struct sockaddr_in);
572 group.sin_family = AF_INET;
573 group.sin_addr = ip->ip_dst;
575 blocked = imo_multi_filter(imo, ifp,
576 (struct sockaddr *)&group,
577 (struct sockaddr *)&udp_in);
578 if (blocked != MCAST_PASS) {
579 if (blocked == MCAST_NOTGMEMBER)
580 IPSTAT_INC(ips_notmember);
581 if (blocked == MCAST_NOTSMEMBER ||
582 blocked == MCAST_MUTED)
583 UDPSTAT_INC(udps_filtermcast);
591 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
592 UDP_PROBE(receive, NULL, last, ip,
594 if (udp_append(last, ip, n, iphlen,
603 * Don't look for additional matches if this one does
604 * not have either the SO_REUSEPORT or SO_REUSEADDR
605 * socket options set. This heuristic avoids
606 * searching through all pcbs in the common case of a
607 * non-shared port. It assumes that an application
608 * will never clear these options after setting them.
610 if ((last->inp_socket->so_options &
611 (SO_REUSEPORT|SO_REUSEADDR)) == 0)
617 * No matching pcb found; discard datagram. (No need
618 * to send an ICMP Port Unreachable for a broadcast
619 * or multicast datgram.)
621 UDPSTAT_INC(udps_noportbcast);
624 INP_INFO_RUNLOCK(pcbinfo);
627 UDP_PROBE(receive, NULL, last, ip, last, uh);
628 if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
631 INP_INFO_RUNLOCK(pcbinfo);
632 return (IPPROTO_DONE);
636 * Locate pcb for datagram.
640 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
642 if ((m->m_flags & M_IP_NEXTHOP) &&
643 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
644 struct sockaddr_in *next_hop;
646 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
649 * Transparently forwarded. Pretend to be the destination.
650 * Already got one like this?
652 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
653 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
656 * It's new. Try to find the ambushing socket.
657 * Because we've rewritten the destination address,
658 * any hardware-generated hash is ignored.
660 inp = in_pcblookup(pcbinfo, ip->ip_src,
661 uh->uh_sport, next_hop->sin_addr,
662 next_hop->sin_port ? htons(next_hop->sin_port) :
663 uh->uh_dport, INPLOOKUP_WILDCARD |
664 INPLOOKUP_RLOCKPCB, ifp);
666 /* Remove the tag from the packet. We don't need it anymore. */
667 m_tag_delete(m, fwd_tag);
668 m->m_flags &= ~M_IP_NEXTHOP;
670 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
671 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
672 INPLOOKUP_RLOCKPCB, ifp, m);
674 if (udp_log_in_vain) {
675 char buf[4*sizeof "123"];
677 strcpy(buf, inet_ntoa(ip->ip_dst));
679 "Connection attempt to UDP %s:%d from %s:%d\n",
680 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
681 ntohs(uh->uh_sport));
683 UDPSTAT_INC(udps_noport);
684 if (m->m_flags & (M_BCAST | M_MCAST)) {
685 UDPSTAT_INC(udps_noportbcast);
690 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
693 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
694 return (IPPROTO_DONE);
698 * Check the minimum TTL for socket.
700 INP_RLOCK_ASSERT(inp);
701 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
704 return (IPPROTO_DONE);
710 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
713 return (IPPROTO_DONE);
717 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
718 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
720 return (IPPROTO_DONE);
724 return (IPPROTO_DONE);
729 * Notify a udp user of an asynchronous error; just wake up so that they can
730 * collect error status.
733 udp_notify(struct inpcb *inp, int errno)
737 * While udp_ctlinput() always calls udp_notify() with a read lock
738 * when invoking it directly, in_pcbnotifyall() currently uses write
739 * locks due to sharing code with TCP. For now, accept either a read
740 * or a write lock, but a read lock is sufficient.
742 INP_LOCK_ASSERT(inp);
744 inp->inp_socket->so_error = errno;
745 sorwakeup(inp->inp_socket);
746 sowwakeup(inp->inp_socket);
752 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
753 struct inpcbinfo *pcbinfo)
757 struct in_addr faddr;
760 faddr = ((struct sockaddr_in *)sa)->sin_addr;
761 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
765 * Redirects don't need to be handled up here.
767 if (PRC_IS_REDIRECT(cmd))
771 * Hostdead is ugly because it goes linearly through all PCBs.
773 * XXX: We never get this from ICMP, otherwise it makes an excellent
774 * DoS attack on machines with many connections.
776 if (cmd == PRC_HOSTDEAD)
778 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
781 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
782 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
783 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
785 INP_RLOCK_ASSERT(inp);
786 if (inp->inp_socket != NULL) {
787 udp_notify(inp, inetctlerrmap[cmd]);
792 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
796 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
799 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
803 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
806 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
811 udp_pcblist(SYSCTL_HANDLER_ARGS)
814 struct inpcb *inp, **inp_list;
819 * The process of preparing the PCB list is too time-consuming and
820 * resource-intensive to repeat twice on every request.
822 if (req->oldptr == 0) {
823 n = V_udbinfo.ipi_count;
824 n += imax(n / 8, 10);
825 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
829 if (req->newptr != 0)
833 * OK, now we're committed to doing something.
835 INP_INFO_RLOCK(&V_udbinfo);
836 gencnt = V_udbinfo.ipi_gencnt;
837 n = V_udbinfo.ipi_count;
838 INP_INFO_RUNLOCK(&V_udbinfo);
840 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
841 + n * sizeof(struct xinpcb));
845 xig.xig_len = sizeof xig;
847 xig.xig_gen = gencnt;
848 xig.xig_sogen = so_gencnt;
849 error = SYSCTL_OUT(req, &xig, sizeof xig);
853 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
857 INP_INFO_RLOCK(&V_udbinfo);
858 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
859 inp = LIST_NEXT(inp, inp_list)) {
861 if (inp->inp_gencnt <= gencnt &&
862 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
868 INP_INFO_RUNLOCK(&V_udbinfo);
872 for (i = 0; i < n; i++) {
875 if (inp->inp_gencnt <= gencnt) {
878 bzero(&xi, sizeof(xi));
879 xi.xi_len = sizeof xi;
880 /* XXX should avoid extra copy */
881 bcopy(inp, &xi.xi_inp, sizeof *inp);
883 sotoxsocket(inp->inp_socket, &xi.xi_socket);
884 xi.xi_inp.inp_gencnt = inp->inp_gencnt;
886 error = SYSCTL_OUT(req, &xi, sizeof xi);
890 INP_INFO_WLOCK(&V_udbinfo);
891 for (i = 0; i < n; i++) {
894 if (!in_pcbrele_rlocked(inp))
897 INP_INFO_WUNLOCK(&V_udbinfo);
901 * Give the user an updated idea of our state. If the
902 * generation differs from what we told her before, she knows
903 * that something happened while we were processing this
904 * request, and it might be necessary to retry.
906 INP_INFO_RLOCK(&V_udbinfo);
907 xig.xig_gen = V_udbinfo.ipi_gencnt;
908 xig.xig_sogen = so_gencnt;
909 xig.xig_count = V_udbinfo.ipi_count;
910 INP_INFO_RUNLOCK(&V_udbinfo);
911 error = SYSCTL_OUT(req, &xig, sizeof xig);
913 free(inp_list, M_TEMP);
917 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
918 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
919 udp_pcblist, "S,xinpcb", "List of active UDP sockets");
923 udp_getcred(SYSCTL_HANDLER_ARGS)
926 struct sockaddr_in addrs[2];
930 error = priv_check(req->td, PRIV_NETINET_GETCRED);
933 error = SYSCTL_IN(req, addrs, sizeof(addrs));
936 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
937 addrs[0].sin_addr, addrs[0].sin_port,
938 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
940 INP_RLOCK_ASSERT(inp);
941 if (inp->inp_socket == NULL)
944 error = cr_canseeinpcb(req->td->td_ucred, inp);
946 cru2x(inp->inp_cred, &xuc);
951 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
955 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
956 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
957 udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
961 udp_ctloutput(struct socket *so, struct sockopt *sopt)
965 int isudplite, error, optval;
968 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
970 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
972 if (sopt->sopt_level != so->so_proto->pr_protocol) {
974 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
976 error = ip6_ctloutput(so, sopt);
979 #if defined(INET) && defined(INET6)
985 error = ip_ctloutput(so, sopt);
991 switch (sopt->sopt_dir) {
993 switch (sopt->sopt_name) {
996 error = sooptcopyin(sopt, &optval, sizeof optval,
1000 inp = sotoinpcb(so);
1001 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1004 up = intoudpcb(inp);
1005 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1009 /* Clear all UDP encap. */
1011 up->u_flags &= ~UF_ESPINUDP_ALL;
1015 case UDP_ENCAP_ESPINUDP:
1016 case UDP_ENCAP_ESPINUDP_NON_IKE:
1017 up->u_flags &= ~UF_ESPINUDP_ALL;
1018 if (optval == UDP_ENCAP_ESPINUDP)
1019 up->u_flags |= UF_ESPINUDP;
1020 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1021 up->u_flags |= UF_ESPINUDP_NON_IKE;
1030 case UDPLITE_SEND_CSCOV:
1031 case UDPLITE_RECV_CSCOV:
1034 error = ENOPROTOOPT;
1038 error = sooptcopyin(sopt, &optval, sizeof(optval),
1042 inp = sotoinpcb(so);
1043 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1045 up = intoudpcb(inp);
1046 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1047 if ((optval != 0 && optval < 8) || (optval > 65535)) {
1052 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1053 up->u_txcslen = optval;
1055 up->u_rxcslen = optval;
1060 error = ENOPROTOOPT;
1065 switch (sopt->sopt_name) {
1068 up = intoudpcb(inp);
1069 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1070 optval = up->u_flags & UF_ESPINUDP_ALL;
1072 error = sooptcopyout(sopt, &optval, sizeof optval);
1075 case UDPLITE_SEND_CSCOV:
1076 case UDPLITE_RECV_CSCOV:
1079 error = ENOPROTOOPT;
1082 up = intoudpcb(inp);
1083 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1084 if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1085 optval = up->u_txcslen;
1087 optval = up->u_rxcslen;
1089 error = sooptcopyout(sopt, &optval, sizeof(optval));
1093 error = ENOPROTOOPT;
1102 #define UH_WLOCKED 2
1103 #define UH_RLOCKED 1
1104 #define UH_UNLOCKED 0
1106 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1107 struct mbuf *control, struct thread *td)
1109 struct udpiphdr *ui;
1110 int len = m->m_pkthdr.len;
1111 struct in_addr faddr, laddr;
1113 struct inpcbinfo *pcbinfo;
1114 struct sockaddr_in *sin, src;
1115 int cscov_partial = 0;
1118 u_short fport, lport;
1123 uint32_t flowid = 0;
1124 uint8_t flowtype = M_HASHTYPE_NONE;
1127 * udp_output() may need to temporarily bind or connect the current
1128 * inpcb. As such, we don't know up front whether we will need the
1129 * pcbinfo lock or not. Do any work to decide what is needed up
1130 * front before acquiring any locks.
1132 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1141 tos = inp->inp_ip_tos;
1142 if (control != NULL) {
1144 * XXX: Currently, we assume all the optional information is
1145 * stored in a single mbuf.
1147 if (control->m_next) {
1153 for (; control->m_len > 0;
1154 control->m_data += CMSG_ALIGN(cm->cmsg_len),
1155 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1156 cm = mtod(control, struct cmsghdr *);
1157 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1158 || cm->cmsg_len > control->m_len) {
1162 if (cm->cmsg_level != IPPROTO_IP)
1165 switch (cm->cmsg_type) {
1166 case IP_SENDSRCADDR:
1168 CMSG_LEN(sizeof(struct in_addr))) {
1172 bzero(&src, sizeof(src));
1173 src.sin_family = AF_INET;
1174 src.sin_len = sizeof(src);
1175 src.sin_port = inp->inp_lport;
1177 *(struct in_addr *)CMSG_DATA(cm);
1181 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1185 tos = *(u_char *)CMSG_DATA(cm);
1189 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1193 flowid = *(uint32_t *) CMSG_DATA(cm);
1197 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1201 flowtype = *(uint32_t *) CMSG_DATA(cm);
1205 case IP_RSSBUCKETID:
1206 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1210 /* This is just a placeholder for now */
1214 error = ENOPROTOOPT;
1229 * Depending on whether or not the application has bound or connected
1230 * the socket, we may have to do varying levels of work. The optimal
1231 * case is for a connected UDP socket, as a global lock isn't
1234 * In order to decide which we need, we require stability of the
1235 * inpcb binding, which we ensure by acquiring a read lock on the
1236 * inpcb. This doesn't strictly follow the lock order, so we play
1237 * the trylock and retry game; note that we may end up with more
1238 * conservative locks than required the second time around, so later
1239 * assertions have to accept that. Further analysis of the number of
1240 * misses under contention is required.
1242 * XXXRW: Check that hash locking update here is correct.
1244 pr = inp->inp_socket->so_proto->pr_protocol;
1245 pcbinfo = udp_get_inpcbinfo(pr);
1246 sin = (struct sockaddr_in *)addr;
1248 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1251 INP_HASH_WLOCK(pcbinfo);
1252 unlock_udbinfo = UH_WLOCKED;
1253 } else if ((sin != NULL && (
1254 (sin->sin_addr.s_addr == INADDR_ANY) ||
1255 (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1256 (inp->inp_laddr.s_addr == INADDR_ANY) ||
1257 (inp->inp_lport == 0))) ||
1258 (src.sin_family == AF_INET)) {
1259 INP_HASH_RLOCK(pcbinfo);
1260 unlock_udbinfo = UH_RLOCKED;
1262 unlock_udbinfo = UH_UNLOCKED;
1265 * If the IP_SENDSRCADDR control message was specified, override the
1266 * source address for this datagram. Its use is invalidated if the
1267 * address thus specified is incomplete or clobbers other inpcbs.
1269 laddr = inp->inp_laddr;
1270 lport = inp->inp_lport;
1271 if (src.sin_family == AF_INET) {
1272 INP_HASH_LOCK_ASSERT(pcbinfo);
1274 (laddr.s_addr == INADDR_ANY &&
1275 src.sin_addr.s_addr == INADDR_ANY)) {
1279 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1280 &laddr.s_addr, &lport, td->td_ucred);
1286 * If a UDP socket has been connected, then a local address/port will
1287 * have been selected and bound.
1289 * If a UDP socket has not been connected to, then an explicit
1290 * destination address must be used, in which case a local
1291 * address/port may not have been selected and bound.
1294 INP_LOCK_ASSERT(inp);
1295 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1301 * Jail may rewrite the destination address, so let it do
1302 * that before we use it.
1304 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1309 * If a local address or port hasn't yet been selected, or if
1310 * the destination address needs to be rewritten due to using
1311 * a special INADDR_ constant, invoke in_pcbconnect_setup()
1312 * to do the heavy lifting. Once a port is selected, we
1313 * commit the binding back to the socket; we also commit the
1314 * binding of the address if in jail.
1316 * If we already have a valid binding and we're not
1317 * requesting a destination address rewrite, use a fast path.
1319 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1320 inp->inp_lport == 0 ||
1321 sin->sin_addr.s_addr == INADDR_ANY ||
1322 sin->sin_addr.s_addr == INADDR_BROADCAST) {
1323 INP_HASH_LOCK_ASSERT(pcbinfo);
1324 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1325 &lport, &faddr.s_addr, &fport, NULL,
1331 * XXXRW: Why not commit the port if the address is
1334 /* Commit the local port if newly assigned. */
1335 if (inp->inp_laddr.s_addr == INADDR_ANY &&
1336 inp->inp_lport == 0) {
1337 INP_WLOCK_ASSERT(inp);
1338 INP_HASH_WLOCK_ASSERT(pcbinfo);
1340 * Remember addr if jailed, to prevent
1343 if (prison_flag(td->td_ucred, PR_IP4))
1344 inp->inp_laddr = laddr;
1345 inp->inp_lport = lport;
1346 if (in_pcbinshash(inp) != 0) {
1351 inp->inp_flags |= INP_ANONPORT;
1354 faddr = sin->sin_addr;
1355 fport = sin->sin_port;
1358 INP_LOCK_ASSERT(inp);
1359 faddr = inp->inp_faddr;
1360 fport = inp->inp_fport;
1361 if (faddr.s_addr == INADDR_ANY) {
1368 * Calculate data length and get a mbuf for UDP, IP, and possible
1369 * link-layer headers. Immediate slide the data pointer back forward
1370 * since we won't use that space at this layer.
1372 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1377 m->m_data += max_linkhdr;
1378 m->m_len -= max_linkhdr;
1379 m->m_pkthdr.len -= max_linkhdr;
1382 * Fill in mbuf with extended UDP header and addresses and length put
1383 * into network format.
1385 ui = mtod(m, struct udpiphdr *);
1386 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
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) {
1397 up = intoudpcb(inp);
1398 cscov = up->u_txcslen;
1399 plen = (u_short)len + sizeof(struct udphdr);
1402 ui->ui_len = htons(plen);
1403 ui->ui_ulen = htons(cscov);
1405 * For UDP-Lite, checksum coverage length of zero means
1406 * the entire UDPLite packet is covered by the checksum.
1408 cscov_partial = (cscov == 0) ? 0 : 1;
1410 ui->ui_v = IPVERSION << 4;
1413 * Set the Don't Fragment bit in the IP header.
1415 if (inp->inp_flags & INP_DONTFRAG) {
1418 ip = (struct ip *)&ui->ui_i;
1419 ip->ip_off |= htons(IP_DF);
1423 if (inp->inp_socket->so_options & SO_DONTROUTE)
1424 ipflags |= IP_ROUTETOIF;
1425 if (inp->inp_socket->so_options & SO_BROADCAST)
1426 ipflags |= IP_ALLOWBROADCAST;
1427 if (inp->inp_flags & INP_ONESBCAST)
1428 ipflags |= IP_SENDONES;
1431 mac_inpcb_create_mbuf(inp, m);
1435 * Set up checksum and output datagram.
1438 if (pr == IPPROTO_UDPLITE) {
1439 if (inp->inp_flags & INP_ONESBCAST)
1440 faddr.s_addr = INADDR_BROADCAST;
1441 if (cscov_partial) {
1442 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1443 ui->ui_sum = 0xffff;
1445 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1446 ui->ui_sum = 0xffff;
1448 } else if (V_udp_cksum) {
1449 if (inp->inp_flags & INP_ONESBCAST)
1450 faddr.s_addr = INADDR_BROADCAST;
1451 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1452 htons((u_short)len + sizeof(struct udphdr) + pr));
1453 m->m_pkthdr.csum_flags = CSUM_UDP;
1454 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1456 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1457 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
1458 ((struct ip *)ui)->ip_tos = tos; /* XXX */
1459 UDPSTAT_INC(udps_opackets);
1462 * Setup flowid / RSS information for outbound socket.
1464 * Once the UDP code decides to set a flowid some other way,
1465 * this allows the flowid to be overridden by userland.
1467 if (flowtype != M_HASHTYPE_NONE) {
1468 m->m_pkthdr.flowid = flowid;
1469 M_HASHTYPE_SET(m, flowtype);
1472 uint32_t hash_val, hash_type;
1474 * Calculate an appropriate RSS hash for UDP and
1477 * The called function will take care of figuring out
1478 * whether a 2-tuple or 4-tuple hash is required based
1479 * on the currently configured scheme.
1481 * Later later on connected socket values should be
1482 * cached in the inpcb and reused, rather than constantly
1483 * re-calculating it.
1485 * UDP Lite is a different protocol number and will
1486 * likely end up being hashed as a 2-tuple until
1487 * RSS / NICs grow UDP Lite protocol awareness.
1489 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1490 pr, &hash_val, &hash_type) == 0) {
1491 m->m_pkthdr.flowid = hash_val;
1492 M_HASHTYPE_SET(m, hash_type);
1499 * Don't override with the inp cached flowid value.
1501 * Depending upon the kind of send being done, the inp
1502 * flowid/flowtype values may actually not be appropriate
1503 * for this particular socket send.
1505 * We should either leave the flowid at zero (which is what is
1506 * currently done) or set it to some software generated
1507 * hash value based on the packet contents.
1509 ipflags |= IP_NODEFAULTFLOWID;
1512 if (unlock_udbinfo == UH_WLOCKED)
1513 INP_HASH_WUNLOCK(pcbinfo);
1514 else if (unlock_udbinfo == UH_RLOCKED)
1515 INP_HASH_RUNLOCK(pcbinfo);
1516 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1517 error = ip_output(m, inp->inp_options, NULL, ipflags,
1518 inp->inp_moptions, inp);
1519 if (unlock_udbinfo == UH_WLOCKED)
1526 if (unlock_udbinfo == UH_WLOCKED) {
1527 INP_HASH_WUNLOCK(pcbinfo);
1529 } else if (unlock_udbinfo == UH_RLOCKED) {
1530 INP_HASH_RUNLOCK(pcbinfo);
1539 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1541 * Potentially decap ESP in UDP frame. Check for an ESP header
1542 * and optional marker; if present, strip the UDP header and
1543 * push the result through IPSec.
1545 * Returns mbuf to be processed (potentially re-allocated) or
1546 * NULL if consumed and/or processed.
1548 static struct mbuf *
1549 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1551 size_t minlen, payload, skip, iphlen;
1555 struct udphdr *udphdr;
1558 INP_RLOCK_ASSERT(inp);
1561 * Pull up data so the longest case is contiguous:
1562 * IP/UDP hdr + non ESP marker + ESP hdr.
1564 minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1565 if (minlen > m->m_pkthdr.len)
1566 minlen = m->m_pkthdr.len;
1567 if ((m = m_pullup(m, minlen)) == NULL) {
1568 IPSECSTAT_INC(ips_in_inval);
1569 return (NULL); /* Bypass caller processing. */
1571 data = mtod(m, caddr_t); /* Points to ip header. */
1572 payload = m->m_len - off; /* Size of payload. */
1574 if (payload == 1 && data[off] == '\xff')
1575 return (m); /* NB: keepalive packet, no decap. */
1577 up = intoudpcb(inp);
1578 KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1579 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1580 ("u_flags 0x%x", up->u_flags));
1583 * Check that the payload is large enough to hold an
1584 * ESP header and compute the amount of data to remove.
1586 * NB: the caller has already done a pullup for us.
1587 * XXX can we assume alignment and eliminate bcopys?
1589 if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1591 * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1592 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1593 * possible AH mode non-IKE marker+non-ESP marker
1594 * from draft-ietf-ipsec-udp-encaps-00.txt.
1598 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1599 return (m); /* NB: no decap. */
1600 bcopy(data + off, &marker, sizeof(uint64_t));
1601 if (marker != 0) /* Non-IKE marker. */
1602 return (m); /* NB: no decap. */
1603 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1607 if (payload <= sizeof(struct esp)) {
1608 IPSECSTAT_INC(ips_in_inval);
1610 return (NULL); /* Discard. */
1612 bcopy(data + off, &spi, sizeof(uint32_t));
1613 if (spi == 0) /* Non-ESP marker. */
1614 return (m); /* NB: no decap. */
1615 skip = sizeof(struct udphdr);
1619 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1620 * the UDP ports. This is required if we want to select
1621 * the right SPD for multiple hosts behind same NAT.
1623 * NB: ports are maintained in network byte order everywhere
1624 * in the NAT-T code.
1626 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1627 2 * sizeof(uint16_t), M_NOWAIT);
1629 IPSECSTAT_INC(ips_in_nomem);
1631 return (NULL); /* Discard. */
1633 iphlen = off - sizeof(struct udphdr);
1634 udphdr = (struct udphdr *)(data + iphlen);
1635 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1636 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1637 m_tag_prepend(m, tag);
1640 * Remove the UDP header (and possibly the non ESP marker)
1641 * IP header length is iphlen
1644 * +----+------+-----+
1645 * | IP | UDP | ESP |
1646 * +----+------+-----+
1654 ovbcopy(data, data + skip, iphlen);
1657 ip = mtod(m, struct ip *);
1658 ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1659 ip->ip_p = IPPROTO_ESP;
1662 * We cannot yet update the cksums so clear any
1663 * h/w cksum flags as they are no longer valid.
1665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1666 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1668 (void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1670 return (NULL); /* NB: consumed, bypass processing. */
1672 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1675 udp_abort(struct socket *so)
1678 struct inpcbinfo *pcbinfo;
1680 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1681 inp = sotoinpcb(so);
1682 KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1684 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1685 INP_HASH_WLOCK(pcbinfo);
1686 in_pcbdisconnect(inp);
1687 inp->inp_laddr.s_addr = INADDR_ANY;
1688 INP_HASH_WUNLOCK(pcbinfo);
1689 soisdisconnected(so);
1695 udp_attach(struct socket *so, int proto, struct thread *td)
1698 struct inpcbinfo *pcbinfo;
1701 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1702 inp = sotoinpcb(so);
1703 KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1704 error = soreserve(so, udp_sendspace, udp_recvspace);
1707 INP_INFO_WLOCK(pcbinfo);
1708 error = in_pcballoc(so, pcbinfo);
1710 INP_INFO_WUNLOCK(pcbinfo);
1714 inp = sotoinpcb(so);
1715 inp->inp_vflag |= INP_IPV4;
1716 inp->inp_ip_ttl = V_ip_defttl;
1718 error = udp_newudpcb(inp);
1722 INP_INFO_WUNLOCK(pcbinfo);
1727 INP_INFO_WUNLOCK(pcbinfo);
1733 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, void *ctx)
1738 KASSERT(so->so_type == SOCK_DGRAM,
1739 ("udp_set_kernel_tunneling: !dgram"));
1740 inp = sotoinpcb(so);
1741 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1743 up = intoudpcb(inp);
1744 if (up->u_tun_func != NULL) {
1749 up->u_tun_ctx = ctx;
1756 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1759 struct inpcbinfo *pcbinfo;
1762 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1763 inp = sotoinpcb(so);
1764 KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1766 INP_HASH_WLOCK(pcbinfo);
1767 error = in_pcbbind(inp, nam, td->td_ucred);
1768 INP_HASH_WUNLOCK(pcbinfo);
1774 udp_close(struct socket *so)
1777 struct inpcbinfo *pcbinfo;
1779 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1780 inp = sotoinpcb(so);
1781 KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1783 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1784 INP_HASH_WLOCK(pcbinfo);
1785 in_pcbdisconnect(inp);
1786 inp->inp_laddr.s_addr = INADDR_ANY;
1787 INP_HASH_WUNLOCK(pcbinfo);
1788 soisdisconnected(so);
1794 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1797 struct inpcbinfo *pcbinfo;
1798 struct sockaddr_in *sin;
1801 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1802 inp = sotoinpcb(so);
1803 KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1805 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1809 sin = (struct sockaddr_in *)nam;
1810 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1815 INP_HASH_WLOCK(pcbinfo);
1816 error = in_pcbconnect(inp, nam, td->td_ucred);
1817 INP_HASH_WUNLOCK(pcbinfo);
1825 udp_detach(struct socket *so)
1828 struct inpcbinfo *pcbinfo;
1831 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1832 inp = sotoinpcb(so);
1833 KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1834 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1835 ("udp_detach: not disconnected"));
1836 INP_INFO_WLOCK(pcbinfo);
1838 up = intoudpcb(inp);
1839 KASSERT(up != NULL, ("%s: up == NULL", __func__));
1840 inp->inp_ppcb = NULL;
1843 INP_INFO_WUNLOCK(pcbinfo);
1848 udp_disconnect(struct socket *so)
1851 struct inpcbinfo *pcbinfo;
1853 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1854 inp = sotoinpcb(so);
1855 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1857 if (inp->inp_faddr.s_addr == INADDR_ANY) {
1861 INP_HASH_WLOCK(pcbinfo);
1862 in_pcbdisconnect(inp);
1863 inp->inp_laddr.s_addr = INADDR_ANY;
1864 INP_HASH_WUNLOCK(pcbinfo);
1866 so->so_state &= ~SS_ISCONNECTED; /* XXX */
1873 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1874 struct mbuf *control, struct thread *td)
1878 inp = sotoinpcb(so);
1879 KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1880 return (udp_output(inp, m, addr, control, td));
1885 udp_shutdown(struct socket *so)
1889 inp = sotoinpcb(so);
1890 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1898 struct pr_usrreqs udp_usrreqs = {
1899 .pru_abort = udp_abort,
1900 .pru_attach = udp_attach,
1901 .pru_bind = udp_bind,
1902 .pru_connect = udp_connect,
1903 .pru_control = in_control,
1904 .pru_detach = udp_detach,
1905 .pru_disconnect = udp_disconnect,
1906 .pru_peeraddr = in_getpeeraddr,
1907 .pru_send = udp_send,
1908 .pru_soreceive = soreceive_dgram,
1909 .pru_sosend = sosend_dgram,
1910 .pru_shutdown = udp_shutdown,
1911 .pru_sockaddr = in_getsockaddr,
1912 .pru_sosetlabel = in_pcbsosetlabel,
1913 .pru_close = udp_close,