2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 #include <sys/cdefs.h>
34 #include "opt_inet6.h"
37 #error "IPDIVERT requires INET"
40 #include <sys/param.h>
41 #include <sys/eventhandler.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/kernel.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
57 #include <net/if_var.h>
58 #include <net/netisr.h>
60 #include <netinet/in.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_var.h>
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
70 #if defined(SCTP) || defined(SCTP_SUPPORT)
71 #include <netinet/sctp_crc32.h>
74 #include <security/mac/mac_framework.h>
80 * Allocate enough space to hold a full IP packet
82 #define DIVSNDQ (65536 + 100)
83 #define DIVRCVQ (65536 + 100)
86 * Divert sockets work in conjunction with ipfw or other packet filters,
87 * see the divert(4) manpage for features.
88 * Packets are selected by the packet filter and tagged with an
89 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
90 * the packet filter) and information on the matching filter rule for
91 * subsequent reinjection. The divert_port is used to put the packet
92 * on the corresponding divert socket, while the rule number is passed
93 * up (at least partially) as the sin_port in the struct sockaddr.
95 * Packets written to the divert socket carry in sin_addr a
96 * destination address, and in sin_port the number of the filter rule
97 * after which to continue processing.
98 * If the destination address is INADDR_ANY, the packet is treated as
99 * as outgoing and sent to ip_output(); otherwise it is treated as
100 * incoming and sent to ip_input().
101 * Further, sin_zero carries some information on the interface,
102 * which can be used in the reinject -- see comments in the code.
104 * On reinjection, processing in ip_input() and ip_output()
105 * will be exactly the same as for the original packet, except that
106 * packet filter processing will start at the rule number after the one
107 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
108 * will apply the entire ruleset to the packet).
111 /* Internal variables. */
112 VNET_DEFINE_STATIC(struct inpcbhead, divcb);
113 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo);
115 #define V_divcb VNET(divcb)
116 #define V_divcbinfo VNET(divcbinfo)
118 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
119 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
121 static eventhandler_tag ip_divert_event_tag;
123 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m,
124 struct sockaddr_in *sin);
125 static int div_output_outbound(int family, struct socket *so, struct mbuf *m);
128 * Initialize divert connection block queue.
131 div_zone_change(void *tag)
134 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
138 div_inpcb_init(void *mem, int size, int flags)
140 struct inpcb *inp = mem;
142 INP_LOCK_INIT(inp, "inp", "divinp");
151 * XXX We don't use the hash list for divert IP, but it's easier to
152 * allocate one-entry hash lists than it is to check all over the
153 * place for hashbase == NULL.
155 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
156 div_inpcb_init, IPI_HASHFIELDS_NONE);
160 div_destroy(void *unused __unused)
163 in_pcbinfo_destroy(&V_divcbinfo);
165 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
169 * IPPROTO_DIVERT is not in the real IP protocol number space; this
170 * function should never be called. Just in case, drop any packets.
173 div_input(struct mbuf **mp, int *offp, int proto)
175 struct mbuf *m = *mp;
177 KMOD_IPSTAT_INC(ips_noproto);
179 return (IPPROTO_DONE);
183 * Divert a packet by passing it up to the divert socket at port 'port'.
185 * Setup generic address and protocol structures for div_input routine,
186 * then pass them along with mbuf chain.
189 divert_packet(struct mbuf *m, bool incoming)
195 struct sockaddr_in divsrc;
200 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
206 if (m->m_len < sizeof(struct ip) &&
207 (m = m_pullup(m, sizeof(struct ip))) == NULL)
209 ip = mtod(m, struct ip *);
211 /* Delayed checksums are currently not compatible with divert. */
212 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
214 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
216 #if defined(SCTP) || defined(SCTP_SUPPORT)
217 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
218 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
219 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
223 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
224 in6_delayed_cksum(m, m->m_pkthdr.len -
225 sizeof(struct ip6_hdr), sizeof(struct ip6_hdr));
226 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
228 #if defined(SCTP) || defined(SCTP_SUPPORT)
229 if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) {
230 sctp_delayed_cksum(m, sizeof(struct ip6_hdr));
231 m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6;
235 bzero(&divsrc, sizeof(divsrc));
236 divsrc.sin_len = sizeof(divsrc);
237 divsrc.sin_family = AF_INET;
238 /* record matching rule, in host format */
239 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
241 * Record receive interface address, if any.
242 * But only for incoming packets.
251 /* Find IP address for receive interface */
252 ifp = m->m_pkthdr.rcvif;
253 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
254 if (ifa->ifa_addr->sa_family != AF_INET)
257 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
262 * Record the incoming interface name whenever we have one.
264 if (m->m_pkthdr.rcvif) {
266 * Hide the actual interface name in there in the
267 * sin_zero array. XXX This needs to be moved to a
268 * different sockaddr type for divert, e.g.
269 * sockaddr_div with multiple fields like
270 * sockaddr_dl. Presently we have only 7 bytes
271 * but that will do for now as most interfaces
272 * are 4 or less + 2 or less bytes for unit.
273 * There is probably a faster way of doing this,
274 * possibly taking it from the sockaddr_dl on the iface.
275 * This solves the problem of a P2P link and a LAN interface
276 * having the same address, which can result in the wrong
277 * interface being assigned to the packet when fed back
278 * into the divert socket. Theoretically if the daemon saves
279 * and re-uses the sockaddr_in as suggested in the man pages,
280 * this iface name will come along for the ride.
281 * (see div_output for the other half of this.)
283 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
284 sizeof(divsrc.sin_zero));
287 /* Put packet on socket queue, if any */
289 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
290 CK_LIST_FOREACH(inp, &V_divcb, inp_list) {
291 /* XXX why does only one socket match? */
292 if (inp->inp_lport == nport) {
294 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
298 sa = inp->inp_socket;
299 SOCKBUF_LOCK(&sa->so_rcv);
300 if (sbappendaddr_locked(&sa->so_rcv,
301 (struct sockaddr *)&divsrc, m,
302 (struct mbuf *)0) == 0) {
303 soroverflow_locked(sa);
304 sa = NULL; /* force mbuf reclaim below */
306 sorwakeup_locked(sa);
313 KMOD_IPSTAT_INC(ips_noproto);
314 KMOD_IPSTAT_DEC(ips_delivered);
319 * Deliver packet back into the IP processing machinery.
321 * If no address specified, or address is 0.0.0.0, send to ip_output();
322 * otherwise, send to ip_input() and mark as having been received on
323 * the interface with that address.
326 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
327 struct mbuf *control)
329 struct epoch_tracker et;
332 struct ipfw_rule_ref *dt;
336 m_freem(control); /* XXX */
341 if (sin->sin_family != AF_INET) {
343 return (EAFNOSUPPORT);
345 if (sin->sin_len != sizeof(*sin)) {
352 * An mbuf may hasn't come from userland, but we pretend
355 m->m_pkthdr.rcvif = NULL;
357 M_SETFIB(m, so->so_fibnum);
359 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
361 /* this should be normal */
362 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
363 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
368 m_tag_prepend(m, mtag);
370 dt = (struct ipfw_rule_ref *)(mtag+1);
372 /* Loopback avoidance and state recovery */
376 /* set the starting point. We provide a non-zero slot,
377 * but a non_matching chain_id to skip that info and use
378 * the rulenum/rule_id.
380 dt->slot = 1; /* dummy, chain_id is invalid */
382 dt->rulenum = sin->sin_port+1; /* host format ? */
384 /* XXX: broken for IPv6 */
386 * Find receive interface with the given name, stuffed
387 * (if it exists) in the sin_zero[] field.
388 * The name is user supplied data so don't trust its size
389 * or that it is zero terminated.
391 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
393 if ( i > 0 && i < sizeof(sin->sin_zero))
394 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
397 ip = mtod(m, struct ip *);
403 case IPV6_VERSION >> 4:
409 return (EAFNOSUPPORT);
412 /* Reinject packet into the system as incoming or outgoing */
414 if (!sin || sin->sin_addr.s_addr == 0) {
415 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
416 error = div_output_outbound(family, so, m);
418 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
419 error = div_output_inbound(family, so, m, sin);
427 * Sends mbuf @m to the wire via ip[6]_output().
429 * Returns 0 on success or an errno value on failure. @m is always consumed.
432 div_output_outbound(int family, struct socket *so, struct mbuf *m)
434 struct ip *const ip = mtod(m, struct ip *);
435 struct mbuf *options;
444 * Don't allow both user specified and setsockopt
445 * options, and don't allow packet length sizes that
448 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
449 inp->inp_options != NULL) ||
450 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
459 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
461 /* Don't allow packet length sizes that will crash */
462 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
472 /* Send packet to output processing */
473 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
476 mac_inpcb_create_mbuf(inp, m);
479 * Get ready to inject the packet into ip_output().
480 * Just in case socket options were specified on the
481 * divert socket, we duplicate them. This is done
482 * to avoid having to hold the PCB locks over the call
483 * to ip_output(), as doing this results in a number of
484 * lock ordering complexities.
486 * Note that we set the multicast options argument for
487 * ip_output() to NULL since it should be invariant that
488 * they are not present.
490 KASSERT(inp->inp_moptions == NULL,
491 ("multicast options set on a divert socket"));
493 * XXXCSJP: It is unclear to me whether or not it makes
494 * sense for divert sockets to have options. However,
495 * for now we will duplicate them with the INP locks
496 * held so we can use them in ip_output() without
497 * requring a reference to the pcb.
500 if (inp->inp_options != NULL) {
501 options = m_dup(inp->inp_options, M_NOWAIT);
502 if (options == NULL) {
513 error = ip_output(m, options, NULL,
514 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
515 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
519 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
530 * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue.
532 * Returns 0 on success or an errno value on failure. @m is always consumed.
535 div_output_inbound(int family, struct socket *so, struct mbuf *m,
536 struct sockaddr_in *sin)
541 if (m->m_pkthdr.rcvif == NULL) {
543 * No luck with the name, check by IP address.
544 * Clear the port and the ifname to make sure
545 * there are no distractions for ifa_ifwithaddr.
548 /* XXX: broken for IPv6 */
549 bzero(sin->sin_zero, sizeof(sin->sin_zero));
551 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
554 return (EADDRNOTAVAIL);
556 m->m_pkthdr.rcvif = ifa->ifa_ifp;
559 mac_socket_create_mbuf(so, m);
561 /* Send packet to input processing via netisr */
564 ip = mtod(m, struct ip *);
566 * Restore M_BCAST flag when destination address is
567 * broadcast. It is expected by ip_tryforward().
569 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)))
570 m->m_flags |= M_MCAST;
571 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
572 m->m_flags |= M_BCAST;
573 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
577 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
589 div_attach(struct socket *so, int proto, struct thread *td)
595 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
597 error = priv_check(td, PRIV_NETINET_DIVERT);
601 error = soreserve(so, div_sendspace, div_recvspace);
604 INP_INFO_WLOCK(&V_divcbinfo);
605 error = in_pcballoc(so, &V_divcbinfo);
607 INP_INFO_WUNLOCK(&V_divcbinfo);
610 inp = (struct inpcb *)so->so_pcb;
611 INP_INFO_WUNLOCK(&V_divcbinfo);
612 inp->inp_ip_p = proto;
613 inp->inp_vflag |= INP_IPV4;
614 inp->inp_flags |= INP_HDRINCL;
620 div_detach(struct socket *so)
625 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
626 INP_INFO_WLOCK(&V_divcbinfo);
630 INP_INFO_WUNLOCK(&V_divcbinfo);
634 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
640 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
641 /* in_pcbbind assumes that nam is a sockaddr_in
642 * and in_pcbbind requires a valid address. Since divert
643 * sockets don't we need to make sure the address is
644 * filled in properly.
645 * XXX -- divert should not be abusing in_pcbind
646 * and should probably have its own family.
648 if (nam->sa_family != AF_INET)
650 if (nam->sa_len != sizeof(struct sockaddr_in))
652 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
653 INP_INFO_WLOCK(&V_divcbinfo);
655 INP_HASH_WLOCK(&V_divcbinfo);
656 error = in_pcbbind(inp, nam, td->td_ucred);
657 INP_HASH_WUNLOCK(&V_divcbinfo);
659 INP_INFO_WUNLOCK(&V_divcbinfo);
664 div_shutdown(struct socket *so)
669 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
677 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
678 struct mbuf *control, struct thread *td)
681 /* Packet must have a header (but that's about it) */
682 if (m->m_len < sizeof (struct ip) &&
683 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
684 KMOD_IPSTAT_INC(ips_toosmall);
692 return div_output(so, m, (struct sockaddr_in *)nam, control);
696 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
698 struct in_addr faddr;
700 faddr = ((struct sockaddr_in *)sa)->sin_addr;
701 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
703 if (PRC_IS_REDIRECT(cmd))
708 div_pcblist(SYSCTL_HANDLER_ARGS)
711 struct epoch_tracker et;
715 if (req->newptr != 0)
718 if (req->oldptr == 0) {
721 n = V_divcbinfo.ipi_count;
722 n += imax(n / 8, 10);
723 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
727 if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
730 bzero(&xig, sizeof(xig));
731 xig.xig_len = sizeof xig;
732 xig.xig_count = V_divcbinfo.ipi_count;
733 xig.xig_gen = V_divcbinfo.ipi_gencnt;
734 xig.xig_sogen = so_gencnt;
735 error = SYSCTL_OUT(req, &xig, sizeof xig);
740 for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead);
742 inp = CK_LIST_NEXT(inp, inp_list)) {
744 if (inp->inp_gencnt <= xig.xig_gen) {
747 in_pcbtoxinpcb(inp, &xi);
749 error = SYSCTL_OUT(req, &xi, sizeof xi);
757 * Give the user an updated idea of our state.
758 * If the generation differs from what we told
759 * her before, she knows that something happened
760 * while we were processing this request, and it
761 * might be necessary to retry.
763 xig.xig_gen = V_divcbinfo.ipi_gencnt;
764 xig.xig_sogen = so_gencnt;
765 xig.xig_count = V_divcbinfo.ipi_count;
766 error = SYSCTL_OUT(req, &xig, sizeof xig);
773 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert,
774 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
776 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist,
777 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
778 NULL, 0, div_pcblist, "S,xinpcb",
779 "List of active divert sockets");
782 struct pr_usrreqs div_usrreqs = {
783 .pru_attach = div_attach,
784 .pru_bind = div_bind,
785 .pru_control = in_control,
786 .pru_detach = div_detach,
787 .pru_peeraddr = in_getpeeraddr,
788 .pru_send = div_send,
789 .pru_shutdown = div_shutdown,
790 .pru_sockaddr = in_getsockaddr,
791 .pru_sosetlabel = in_pcbsosetlabel
794 struct protosw div_protosw = {
796 .pr_protocol = IPPROTO_DIVERT,
797 .pr_flags = PR_ATOMIC|PR_ADDR,
798 .pr_input = div_input,
799 .pr_ctlinput = div_ctlinput,
800 .pr_ctloutput = ip_ctloutput,
802 .pr_usrreqs = &div_usrreqs
806 div_modevent(module_t mod, int type, void *unused)
813 * Protocol will be initialized by pf_proto_register().
814 * We don't have to register ip_protox because we are not
815 * a true IP protocol that goes over the wire.
817 err = pf_proto_register(PF_INET, &div_protosw);
820 ip_divert_ptr = divert_packet;
821 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
822 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
826 * IPDIVERT may normally not be unloaded because of the
827 * potential race conditions. Tell kldunload we can't be
828 * unloaded unless the unload is forced.
836 * Module ipdivert can only be unloaded if no sockets are
837 * connected. Maybe this can be changed later to forcefully
838 * disconnect any open sockets.
840 * XXXRW: Note that there is a slight race here, as a new
841 * socket open request could be spinning on the lock and then
842 * we destroy the lock.
844 INP_INFO_WLOCK(&V_divcbinfo);
845 if (V_divcbinfo.ipi_count != 0) {
847 INP_INFO_WUNLOCK(&V_divcbinfo);
850 ip_divert_ptr = NULL;
851 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
852 INP_INFO_WUNLOCK(&V_divcbinfo);
856 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
865 static moduledata_t ipdivertmod = {
871 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
872 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3);
873 MODULE_VERSION(ipdivert, 1);