2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 #include "opt_inet6.h"
37 #error "IPDIVERT requires INET"
40 #include <sys/param.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
56 #include <net/netisr.h>
58 #include <netinet/in.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/in_var.h>
62 #include <netinet/ip.h>
63 #include <netinet/ip_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
69 #include <netinet/sctp_crc32.h>
72 #include <security/mac/mac_framework.h>
79 * Allocate enough space to hold a full IP packet
81 #define DIVSNDQ (65536 + 100)
82 #define DIVRCVQ (65536 + 100)
85 * Divert sockets work in conjunction with ipfw or other packet filters,
86 * see the divert(4) manpage for features.
87 * Packets are selected by the packet filter and tagged with an
88 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by
89 * the packet filter) and information on the matching filter rule for
90 * subsequent reinjection. The divert_port is used to put the packet
91 * on the corresponding divert socket, while the rule number is passed
92 * up (at least partially) as the sin_port in the struct sockaddr.
94 * Packets written to the divert socket carry in sin_addr a
95 * destination address, and in sin_port the number of the filter rule
96 * after which to continue processing.
97 * If the destination address is INADDR_ANY, the packet is treated as
98 * as outgoing and sent to ip_output(); otherwise it is treated as
99 * incoming and sent to ip_input().
100 * Further, sin_zero carries some information on the interface,
101 * which can be used in the reinject -- see comments in the code.
103 * On reinjection, processing in ip_input() and ip_output()
104 * will be exactly the same as for the original packet, except that
105 * packet filter processing will start at the rule number after the one
106 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0
107 * will apply the entire ruleset to the packet).
110 /* Internal variables. */
111 static VNET_DEFINE(struct inpcbhead, divcb);
112 static VNET_DEFINE(struct inpcbinfo, divcbinfo);
114 #define V_divcb VNET(divcb)
115 #define V_divcbinfo VNET(divcbinfo)
117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
120 static eventhandler_tag ip_divert_event_tag;
123 * Initialize divert connection block queue.
126 div_zone_change(void *tag)
129 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets);
133 div_inpcb_init(void *mem, int size, int flags)
135 struct inpcb *inp = mem;
137 INP_LOCK_INIT(inp, "inp", "divinp");
142 div_inpcb_fini(void *mem, int size)
144 struct inpcb *inp = mem;
146 INP_LOCK_DESTROY(inp);
154 * XXX We don't use the hash list for divert IP, but it's easier to
155 * allocate one-entry hash lists than it is to check all over the
156 * place for hashbase == NULL.
158 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb",
159 div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE,
160 IPI_HASHFIELDS_NONE);
167 in_pcbinfo_destroy(&V_divcbinfo);
171 * IPPROTO_DIVERT is not in the real IP protocol number space; this
172 * function should never be called. Just in case, drop any packets.
175 div_input(struct mbuf *m, int off)
178 KMOD_IPSTAT_INC(ips_noproto);
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, int incoming)
195 struct sockaddr_in divsrc;
198 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
204 if (m->m_len < sizeof(struct ip) &&
205 (m = m_pullup(m, sizeof(struct ip))) == 0)
207 ip = mtod(m, struct ip *);
209 /* Delayed checksums are currently not compatible with divert. */
210 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
212 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
215 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
216 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
217 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
220 bzero(&divsrc, sizeof(divsrc));
221 divsrc.sin_len = sizeof(divsrc);
222 divsrc.sin_family = AF_INET;
223 /* record matching rule, in host format */
224 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
226 * Record receive interface address, if any.
227 * But only for incoming packets.
236 /* Find IP address for receive interface */
237 ifp = m->m_pkthdr.rcvif;
239 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
240 if (ifa->ifa_addr->sa_family != AF_INET)
243 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
246 if_addr_runlock(ifp);
249 * Record the incoming interface name whenever we have one.
251 if (m->m_pkthdr.rcvif) {
253 * Hide the actual interface name in there in the
254 * sin_zero array. XXX This needs to be moved to a
255 * different sockaddr type for divert, e.g.
256 * sockaddr_div with multiple fields like
257 * sockaddr_dl. Presently we have only 7 bytes
258 * but that will do for now as most interfaces
259 * are 4 or less + 2 or less bytes for unit.
260 * There is probably a faster way of doing this,
261 * possibly taking it from the sockaddr_dl on the iface.
262 * This solves the problem of a P2P link and a LAN interface
263 * having the same address, which can result in the wrong
264 * interface being assigned to the packet when fed back
265 * into the divert socket. Theoretically if the daemon saves
266 * and re-uses the sockaddr_in as suggested in the man pages,
267 * this iface name will come along for the ride.
268 * (see div_output for the other half of this.)
270 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
271 sizeof(divsrc.sin_zero));
274 /* Put packet on socket queue, if any */
276 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
277 INP_INFO_RLOCK(&V_divcbinfo);
278 LIST_FOREACH(inp, &V_divcb, inp_list) {
279 /* XXX why does only one socket match? */
280 if (inp->inp_lport == nport) {
282 sa = inp->inp_socket;
283 SOCKBUF_LOCK(&sa->so_rcv);
284 if (sbappendaddr_locked(&sa->so_rcv,
285 (struct sockaddr *)&divsrc, m,
286 (struct mbuf *)0) == 0) {
287 SOCKBUF_UNLOCK(&sa->so_rcv);
288 sa = NULL; /* force mbuf reclaim below */
290 sorwakeup_locked(sa);
295 INP_INFO_RUNLOCK(&V_divcbinfo);
298 KMOD_IPSTAT_INC(ips_noproto);
299 KMOD_IPSTAT_DEC(ips_delivered);
304 * Deliver packet back into the IP processing machinery.
306 * If no address specified, or address is 0.0.0.0, send to ip_output();
307 * otherwise, send to ip_input() and mark as having been received on
308 * the interface with that address.
311 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
312 struct mbuf *control)
314 struct ip *const ip = mtod(m, struct ip *);
316 struct ipfw_rule_ref *dt;
320 * An mbuf may hasn't come from userland, but we pretend
323 m->m_pkthdr.rcvif = NULL;
325 M_SETFIB(m, so->so_fibnum);
328 m_freem(control); /* XXX */
330 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
332 /* this should be normal */
333 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
334 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
339 m_tag_prepend(m, mtag);
341 dt = (struct ipfw_rule_ref *)(mtag+1);
343 /* Loopback avoidance and state recovery */
347 /* set the starting point. We provide a non-zero slot,
348 * but a non_matching chain_id to skip that info and use
349 * the rulenum/rule_id.
351 dt->slot = 1; /* dummy, chain_id is invalid */
353 dt->rulenum = sin->sin_port+1; /* host format ? */
356 * Find receive interface with the given name, stuffed
357 * (if it exists) in the sin_zero[] field.
358 * The name is user supplied data so don't trust its size
359 * or that it is zero terminated.
361 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
363 if ( i > 0 && i < sizeof(sin->sin_zero))
364 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
367 /* Reinject packet into the system as incoming or outgoing */
368 if (!sin || sin->sin_addr.s_addr == 0) {
369 struct mbuf *options = NULL;
372 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
378 * Don't allow both user specified and setsockopt
379 * options, and don't allow packet length sizes that
382 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
383 inp->inp_options != NULL) ||
384 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
391 case IPV6_VERSION >> 4:
393 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
395 /* Don't allow packet length sizes that will crash */
396 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
410 /* Send packet to output processing */
411 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
414 mac_inpcb_create_mbuf(inp, m);
417 * Get ready to inject the packet into ip_output().
418 * Just in case socket options were specified on the
419 * divert socket, we duplicate them. This is done
420 * to avoid having to hold the PCB locks over the call
421 * to ip_output(), as doing this results in a number of
422 * lock ordering complexities.
424 * Note that we set the multicast options argument for
425 * ip_output() to NULL since it should be invariant that
426 * they are not present.
428 KASSERT(inp->inp_moptions == NULL,
429 ("multicast options set on a divert socket"));
431 * XXXCSJP: It is unclear to me whether or not it makes
432 * sense for divert sockets to have options. However,
433 * for now we will duplicate them with the INP locks
434 * held so we can use them in ip_output() without
435 * requring a reference to the pcb.
437 if (inp->inp_options != NULL) {
438 options = m_dup(inp->inp_options, M_NOWAIT);
439 if (options == NULL) {
449 error = ip_output(m, options, NULL,
450 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
451 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
454 case IPV6_VERSION >> 4:
455 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
462 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
463 if (m->m_pkthdr.rcvif == NULL) {
465 * No luck with the name, check by IP address.
466 * Clear the port and the ifname to make sure
467 * there are no distractions for ifa_ifwithaddr.
471 bzero(sin->sin_zero, sizeof(sin->sin_zero));
473 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
475 error = EADDRNOTAVAIL;
478 m->m_pkthdr.rcvif = ifa->ifa_ifp;
482 mac_socket_create_mbuf(so, m);
484 /* Send packet to input processing via netisr */
487 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
490 case IPV6_VERSION >> 4:
491 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
508 div_attach(struct socket *so, int proto, struct thread *td)
514 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
516 error = priv_check(td, PRIV_NETINET_DIVERT);
520 error = soreserve(so, div_sendspace, div_recvspace);
523 INP_INFO_WLOCK(&V_divcbinfo);
524 error = in_pcballoc(so, &V_divcbinfo);
526 INP_INFO_WUNLOCK(&V_divcbinfo);
529 inp = (struct inpcb *)so->so_pcb;
530 INP_INFO_WUNLOCK(&V_divcbinfo);
531 inp->inp_ip_p = proto;
532 inp->inp_vflag |= INP_IPV4;
533 inp->inp_flags |= INP_HDRINCL;
539 div_detach(struct socket *so)
544 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
545 INP_INFO_WLOCK(&V_divcbinfo);
549 INP_INFO_WUNLOCK(&V_divcbinfo);
553 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
559 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
560 /* in_pcbbind assumes that nam is a sockaddr_in
561 * and in_pcbbind requires a valid address. Since divert
562 * sockets don't we need to make sure the address is
563 * filled in properly.
564 * XXX -- divert should not be abusing in_pcbind
565 * and should probably have its own family.
567 if (nam->sa_family != AF_INET)
569 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
570 INP_INFO_WLOCK(&V_divcbinfo);
572 INP_HASH_WLOCK(&V_divcbinfo);
573 error = in_pcbbind(inp, nam, td->td_ucred);
574 INP_HASH_WUNLOCK(&V_divcbinfo);
576 INP_INFO_WUNLOCK(&V_divcbinfo);
581 div_shutdown(struct socket *so)
586 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
594 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
595 struct mbuf *control, struct thread *td)
598 /* Packet must have a header (but that's about it) */
599 if (m->m_len < sizeof (struct ip) &&
600 (m = m_pullup(m, sizeof (struct ip))) == 0) {
601 KMOD_IPSTAT_INC(ips_toosmall);
607 return div_output(so, m, (struct sockaddr_in *)nam, control);
611 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
613 struct in_addr faddr;
615 faddr = ((struct sockaddr_in *)sa)->sin_addr;
616 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
618 if (PRC_IS_REDIRECT(cmd))
623 div_pcblist(SYSCTL_HANDLER_ARGS)
626 struct inpcb *inp, **inp_list;
631 * The process of preparing the TCB list is too time-consuming and
632 * resource-intensive to repeat twice on every request.
634 if (req->oldptr == 0) {
635 n = V_divcbinfo.ipi_count;
636 n += imax(n / 8, 10);
637 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
641 if (req->newptr != 0)
645 * OK, now we're committed to doing something.
647 INP_INFO_RLOCK(&V_divcbinfo);
648 gencnt = V_divcbinfo.ipi_gencnt;
649 n = V_divcbinfo.ipi_count;
650 INP_INFO_RUNLOCK(&V_divcbinfo);
652 error = sysctl_wire_old_buffer(req,
653 2 * sizeof(xig) + n*sizeof(struct xinpcb));
657 xig.xig_len = sizeof xig;
659 xig.xig_gen = gencnt;
660 xig.xig_sogen = so_gencnt;
661 error = SYSCTL_OUT(req, &xig, sizeof xig);
665 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
669 INP_INFO_RLOCK(&V_divcbinfo);
670 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
671 inp = LIST_NEXT(inp, inp_list)) {
673 if (inp->inp_gencnt <= gencnt &&
674 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
680 INP_INFO_RUNLOCK(&V_divcbinfo);
684 for (i = 0; i < n; i++) {
687 if (inp->inp_gencnt <= gencnt) {
689 bzero(&xi, sizeof(xi));
690 xi.xi_len = sizeof xi;
691 /* XXX should avoid extra copy */
692 bcopy(inp, &xi.xi_inp, sizeof *inp);
694 sotoxsocket(inp->inp_socket, &xi.xi_socket);
696 error = SYSCTL_OUT(req, &xi, sizeof xi);
700 INP_INFO_WLOCK(&V_divcbinfo);
701 for (i = 0; i < n; i++) {
704 if (!in_pcbrele_rlocked(inp))
707 INP_INFO_WUNLOCK(&V_divcbinfo);
711 * Give the user an updated idea of our state.
712 * If the generation differs from what we told
713 * her before, she knows that something happened
714 * while we were processing this request, and it
715 * might be necessary to retry.
717 INP_INFO_RLOCK(&V_divcbinfo);
718 xig.xig_gen = V_divcbinfo.ipi_gencnt;
719 xig.xig_sogen = so_gencnt;
720 xig.xig_count = V_divcbinfo.ipi_count;
721 INP_INFO_RUNLOCK(&V_divcbinfo);
722 error = SYSCTL_OUT(req, &xig, sizeof xig);
724 free(inp_list, M_TEMP);
729 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
731 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
732 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
735 struct pr_usrreqs div_usrreqs = {
736 .pru_attach = div_attach,
737 .pru_bind = div_bind,
738 .pru_control = in_control,
739 .pru_detach = div_detach,
740 .pru_peeraddr = in_getpeeraddr,
741 .pru_send = div_send,
742 .pru_shutdown = div_shutdown,
743 .pru_sockaddr = in_getsockaddr,
744 .pru_sosetlabel = in_pcbsosetlabel
747 struct protosw div_protosw = {
749 .pr_protocol = IPPROTO_DIVERT,
750 .pr_flags = PR_ATOMIC|PR_ADDR,
751 .pr_input = div_input,
752 .pr_ctlinput = div_ctlinput,
753 .pr_ctloutput = ip_ctloutput,
756 .pr_destroy = div_destroy,
758 .pr_usrreqs = &div_usrreqs
762 div_modevent(module_t mod, int type, void *unused)
772 * Protocol will be initialized by pf_proto_register().
773 * We don't have to register ip_protox because we are not
774 * a true IP protocol that goes over the wire.
776 err = pf_proto_register(PF_INET, &div_protosw);
779 ip_divert_ptr = divert_packet;
780 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
781 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
785 * IPDIVERT may normally not be unloaded because of the
786 * potential race conditions. Tell kldunload we can't be
787 * unloaded unless the unload is forced.
799 * Module ipdivert can only be unloaded if no sockets are
800 * connected. Maybe this can be changed later to forcefully
801 * disconnect any open sockets.
803 * XXXRW: Note that there is a slight race here, as a new
804 * socket open request could be spinning on the lock and then
805 * we destroy the lock.
807 INP_INFO_WLOCK(&V_divcbinfo);
808 n = V_divcbinfo.ipi_count;
811 INP_INFO_WUNLOCK(&V_divcbinfo);
814 ip_divert_ptr = NULL;
815 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
816 INP_INFO_WUNLOCK(&V_divcbinfo);
818 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
828 static moduledata_t ipdivertmod = {
834 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
835 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
836 MODULE_VERSION(ipdivert, 1);