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) {
211 ip->ip_len = ntohs(ip->ip_len);
213 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
214 ip->ip_len = htons(ip->ip_len);
217 if (m->m_pkthdr.csum_flags & CSUM_SCTP) {
218 ip->ip_len = ntohs(ip->ip_len);
219 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
220 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
221 ip->ip_len = htons(ip->ip_len);
224 bzero(&divsrc, sizeof(divsrc));
225 divsrc.sin_len = sizeof(divsrc);
226 divsrc.sin_family = AF_INET;
227 /* record matching rule, in host format */
228 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum;
230 * Record receive interface address, if any.
231 * But only for incoming packets.
240 /* Find IP address for receive interface */
241 ifp = m->m_pkthdr.rcvif;
243 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
244 if (ifa->ifa_addr->sa_family != AF_INET)
247 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
250 if_addr_runlock(ifp);
253 * Record the incoming interface name whenever we have one.
255 if (m->m_pkthdr.rcvif) {
257 * Hide the actual interface name in there in the
258 * sin_zero array. XXX This needs to be moved to a
259 * different sockaddr type for divert, e.g.
260 * sockaddr_div with multiple fields like
261 * sockaddr_dl. Presently we have only 7 bytes
262 * but that will do for now as most interfaces
263 * are 4 or less + 2 or less bytes for unit.
264 * There is probably a faster way of doing this,
265 * possibly taking it from the sockaddr_dl on the iface.
266 * This solves the problem of a P2P link and a LAN interface
267 * having the same address, which can result in the wrong
268 * interface being assigned to the packet when fed back
269 * into the divert socket. Theoretically if the daemon saves
270 * and re-uses the sockaddr_in as suggested in the man pages,
271 * this iface name will come along for the ride.
272 * (see div_output for the other half of this.)
274 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
275 sizeof(divsrc.sin_zero));
278 /* Put packet on socket queue, if any */
280 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info));
281 INP_INFO_RLOCK(&V_divcbinfo);
282 LIST_FOREACH(inp, &V_divcb, inp_list) {
283 /* XXX why does only one socket match? */
284 if (inp->inp_lport == nport) {
286 sa = inp->inp_socket;
287 SOCKBUF_LOCK(&sa->so_rcv);
288 if (sbappendaddr_locked(&sa->so_rcv,
289 (struct sockaddr *)&divsrc, m,
290 (struct mbuf *)0) == 0) {
291 SOCKBUF_UNLOCK(&sa->so_rcv);
292 sa = NULL; /* force mbuf reclaim below */
294 sorwakeup_locked(sa);
299 INP_INFO_RUNLOCK(&V_divcbinfo);
302 KMOD_IPSTAT_INC(ips_noproto);
303 KMOD_IPSTAT_DEC(ips_delivered);
308 * Deliver packet back into the IP processing machinery.
310 * If no address specified, or address is 0.0.0.0, send to ip_output();
311 * otherwise, send to ip_input() and mark as having been received on
312 * the interface with that address.
315 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
316 struct mbuf *control)
318 struct ip *const ip = mtod(m, struct ip *);
320 struct ipfw_rule_ref *dt;
324 * An mbuf may hasn't come from userland, but we pretend
327 m->m_pkthdr.rcvif = NULL;
329 M_SETFIB(m, so->so_fibnum);
332 m_freem(control); /* XXX */
334 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
336 /* this should be normal */
337 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
338 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
343 m_tag_prepend(m, mtag);
345 dt = (struct ipfw_rule_ref *)(mtag+1);
347 /* Loopback avoidance and state recovery */
351 /* set the starting point. We provide a non-zero slot,
352 * but a non_matching chain_id to skip that info and use
353 * the rulenum/rule_id.
355 dt->slot = 1; /* dummy, chain_id is invalid */
357 dt->rulenum = sin->sin_port+1; /* host format ? */
360 * Find receive interface with the given name, stuffed
361 * (if it exists) in the sin_zero[] field.
362 * The name is user supplied data so don't trust its size
363 * or that it is zero terminated.
365 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
367 if ( i > 0 && i < sizeof(sin->sin_zero))
368 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
371 /* Reinject packet into the system as incoming or outgoing */
372 if (!sin || sin->sin_addr.s_addr == 0) {
373 struct mbuf *options = NULL;
376 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT;
382 * Don't allow both user specified and setsockopt
383 * options, and don't allow packet length sizes that
386 if ((((ip->ip_hl << 2) != sizeof(struct ip)) &&
387 inp->inp_options != NULL) ||
388 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
394 /* Convert fields to host order for ip_output() */
395 ip->ip_len = ntohs(ip->ip_len);
396 ip->ip_off = ntohs(ip->ip_off);
399 case IPV6_VERSION >> 4:
401 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *);
403 /* Don't allow packet length sizes that will crash */
404 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) {
410 ip6->ip6_plen = ntohs(ip6->ip6_plen);
420 /* Send packet to output processing */
421 KMOD_IPSTAT_INC(ips_rawout); /* XXX */
424 mac_inpcb_create_mbuf(inp, m);
427 * Get ready to inject the packet into ip_output().
428 * Just in case socket options were specified on the
429 * divert socket, we duplicate them. This is done
430 * to avoid having to hold the PCB locks over the call
431 * to ip_output(), as doing this results in a number of
432 * lock ordering complexities.
434 * Note that we set the multicast options argument for
435 * ip_output() to NULL since it should be invariant that
436 * they are not present.
438 KASSERT(inp->inp_moptions == NULL,
439 ("multicast options set on a divert socket"));
441 * XXXCSJP: It is unclear to me whether or not it makes
442 * sense for divert sockets to have options. However,
443 * for now we will duplicate them with the INP locks
444 * held so we can use them in ip_output() without
445 * requring a reference to the pcb.
447 if (inp->inp_options != NULL) {
448 options = m_dup(inp->inp_options, M_NOWAIT);
449 if (options == NULL) {
459 error = ip_output(m, options, NULL,
460 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
461 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
464 case IPV6_VERSION >> 4:
465 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
472 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN;
473 if (m->m_pkthdr.rcvif == NULL) {
475 * No luck with the name, check by IP address.
476 * Clear the port and the ifname to make sure
477 * there are no distractions for ifa_ifwithaddr.
481 bzero(sin->sin_zero, sizeof(sin->sin_zero));
483 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
485 error = EADDRNOTAVAIL;
488 m->m_pkthdr.rcvif = ifa->ifa_ifp;
492 mac_socket_create_mbuf(so, m);
494 /* Send packet to input processing via netisr */
497 netisr_queue_src(NETISR_IP, (uintptr_t)so, m);
500 case IPV6_VERSION >> 4:
501 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m);
518 div_attach(struct socket *so, int proto, struct thread *td)
524 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
526 error = priv_check(td, PRIV_NETINET_DIVERT);
530 error = soreserve(so, div_sendspace, div_recvspace);
533 INP_INFO_WLOCK(&V_divcbinfo);
534 error = in_pcballoc(so, &V_divcbinfo);
536 INP_INFO_WUNLOCK(&V_divcbinfo);
539 inp = (struct inpcb *)so->so_pcb;
540 INP_INFO_WUNLOCK(&V_divcbinfo);
541 inp->inp_ip_p = proto;
542 inp->inp_vflag |= INP_IPV4;
543 inp->inp_flags |= INP_HDRINCL;
549 div_detach(struct socket *so)
554 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
555 INP_INFO_WLOCK(&V_divcbinfo);
559 INP_INFO_WUNLOCK(&V_divcbinfo);
563 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
569 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
570 /* in_pcbbind assumes that nam is a sockaddr_in
571 * and in_pcbbind requires a valid address. Since divert
572 * sockets don't we need to make sure the address is
573 * filled in properly.
574 * XXX -- divert should not be abusing in_pcbind
575 * and should probably have its own family.
577 if (nam->sa_family != AF_INET)
579 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
580 INP_INFO_WLOCK(&V_divcbinfo);
582 INP_HASH_WLOCK(&V_divcbinfo);
583 error = in_pcbbind(inp, nam, td->td_ucred);
584 INP_HASH_WUNLOCK(&V_divcbinfo);
586 INP_INFO_WUNLOCK(&V_divcbinfo);
591 div_shutdown(struct socket *so)
596 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
604 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
605 struct mbuf *control, struct thread *td)
608 /* Packet must have a header (but that's about it) */
609 if (m->m_len < sizeof (struct ip) &&
610 (m = m_pullup(m, sizeof (struct ip))) == 0) {
611 KMOD_IPSTAT_INC(ips_toosmall);
617 return div_output(so, m, (struct sockaddr_in *)nam, control);
621 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
623 struct in_addr faddr;
625 faddr = ((struct sockaddr_in *)sa)->sin_addr;
626 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
628 if (PRC_IS_REDIRECT(cmd))
633 div_pcblist(SYSCTL_HANDLER_ARGS)
636 struct inpcb *inp, **inp_list;
641 * The process of preparing the TCB list is too time-consuming and
642 * resource-intensive to repeat twice on every request.
644 if (req->oldptr == 0) {
645 n = V_divcbinfo.ipi_count;
646 n += imax(n / 8, 10);
647 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
651 if (req->newptr != 0)
655 * OK, now we're committed to doing something.
657 INP_INFO_RLOCK(&V_divcbinfo);
658 gencnt = V_divcbinfo.ipi_gencnt;
659 n = V_divcbinfo.ipi_count;
660 INP_INFO_RUNLOCK(&V_divcbinfo);
662 error = sysctl_wire_old_buffer(req,
663 2 * sizeof(xig) + n*sizeof(struct xinpcb));
667 xig.xig_len = sizeof xig;
669 xig.xig_gen = gencnt;
670 xig.xig_sogen = so_gencnt;
671 error = SYSCTL_OUT(req, &xig, sizeof xig);
675 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
679 INP_INFO_RLOCK(&V_divcbinfo);
680 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n;
681 inp = LIST_NEXT(inp, inp_list)) {
683 if (inp->inp_gencnt <= gencnt &&
684 cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
690 INP_INFO_RUNLOCK(&V_divcbinfo);
694 for (i = 0; i < n; i++) {
697 if (inp->inp_gencnt <= gencnt) {
699 bzero(&xi, sizeof(xi));
700 xi.xi_len = sizeof xi;
701 /* XXX should avoid extra copy */
702 bcopy(inp, &xi.xi_inp, sizeof *inp);
704 sotoxsocket(inp->inp_socket, &xi.xi_socket);
706 error = SYSCTL_OUT(req, &xi, sizeof xi);
710 INP_INFO_WLOCK(&V_divcbinfo);
711 for (i = 0; i < n; i++) {
714 if (!in_pcbrele_rlocked(inp))
717 INP_INFO_WUNLOCK(&V_divcbinfo);
721 * Give the user an updated idea of our state.
722 * If the generation differs from what we told
723 * her before, she knows that something happened
724 * while we were processing this request, and it
725 * might be necessary to retry.
727 INP_INFO_RLOCK(&V_divcbinfo);
728 xig.xig_gen = V_divcbinfo.ipi_gencnt;
729 xig.xig_sogen = so_gencnt;
730 xig.xig_count = V_divcbinfo.ipi_count;
731 INP_INFO_RUNLOCK(&V_divcbinfo);
732 error = SYSCTL_OUT(req, &xig, sizeof xig);
734 free(inp_list, M_TEMP);
739 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0,
741 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
742 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets");
745 struct pr_usrreqs div_usrreqs = {
746 .pru_attach = div_attach,
747 .pru_bind = div_bind,
748 .pru_control = in_control,
749 .pru_detach = div_detach,
750 .pru_peeraddr = in_getpeeraddr,
751 .pru_send = div_send,
752 .pru_shutdown = div_shutdown,
753 .pru_sockaddr = in_getsockaddr,
754 .pru_sosetlabel = in_pcbsosetlabel
757 struct protosw div_protosw = {
759 .pr_protocol = IPPROTO_DIVERT,
760 .pr_flags = PR_ATOMIC|PR_ADDR,
761 .pr_input = div_input,
762 .pr_ctlinput = div_ctlinput,
763 .pr_ctloutput = ip_ctloutput,
766 .pr_destroy = div_destroy,
768 .pr_usrreqs = &div_usrreqs
772 div_modevent(module_t mod, int type, void *unused)
782 * Protocol will be initialized by pf_proto_register().
783 * We don't have to register ip_protox because we are not
784 * a true IP protocol that goes over the wire.
786 err = pf_proto_register(PF_INET, &div_protosw);
789 ip_divert_ptr = divert_packet;
790 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change,
791 div_zone_change, NULL, EVENTHANDLER_PRI_ANY);
795 * IPDIVERT may normally not be unloaded because of the
796 * potential race conditions. Tell kldunload we can't be
797 * unloaded unless the unload is forced.
809 * Module ipdivert can only be unloaded if no sockets are
810 * connected. Maybe this can be changed later to forcefully
811 * disconnect any open sockets.
813 * XXXRW: Note that there is a slight race here, as a new
814 * socket open request could be spinning on the lock and then
815 * we destroy the lock.
817 INP_INFO_WLOCK(&V_divcbinfo);
818 n = V_divcbinfo.ipi_count;
821 INP_INFO_WUNLOCK(&V_divcbinfo);
824 ip_divert_ptr = NULL;
825 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
826 INP_INFO_WUNLOCK(&V_divcbinfo);
828 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag);
838 static moduledata_t ipdivertmod = {
844 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
845 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2);
846 MODULE_VERSION(ipdivert, 1);