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("$FreeBSD$");
33 #if !defined(KLD_MODULE)
38 #error "IPDIVERT requires INET."
41 #error "IPDIVERT requires IPFIREWALL"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/kernel.h>
54 #include <sys/protosw.h>
55 #include <sys/signalvar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
65 #include <net/netisr.h>
66 #include <net/route.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_divert.h>
74 #include <netinet/ip_var.h>
75 #include <netinet/ip_fw.h>
77 #include <security/mac/mac_framework.h>
84 * Allocate enough space to hold a full IP packet
86 #define DIVSNDQ (65536 + 100)
87 #define DIVRCVQ (65536 + 100)
90 * Divert sockets work in conjunction with ipfw, see the divert(4)
91 * manpage for features.
92 * Internally, packets selected by ipfw in ip_input() or ip_output(),
93 * and never diverted before, are passed to the input queue of the
94 * divert socket with a given 'divert_port' number (as specified in
95 * the matching ipfw rule), and they are tagged with a 16 bit cookie
96 * (representing the rule number of the matching ipfw rule), which
97 * is passed to process reading from the socket.
99 * Packets written to the divert socket are again tagged with a cookie
100 * (usually the same as above) and a destination address.
101 * If the destination address is INADDR_ANY then the packet is
102 * treated as outgoing and sent to ip_output(), otherwise it is
103 * treated as incoming and sent to ip_input().
104 * In both cases, the packet is tagged with the cookie.
106 * On reinjection, processing in ip_input() and ip_output()
107 * will be exactly the same as for the original packet, except that
108 * ipfw processing will start at the rule number after the one
109 * written in the cookie (so, tagging a packet with a cookie of 0
110 * will cause it to be effectively considered as a standard packet).
113 /* Internal variables. */
114 static struct inpcbhead divcb;
115 static struct inpcbinfo divcbinfo;
117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
121 * Initialize divert connection block queue.
124 div_zone_change(void *tag)
127 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
131 div_inpcb_init(void *mem, int size, int flags)
133 struct inpcb *inp = mem;
135 INP_LOCK_INIT(inp, "inp", "divinp");
140 div_inpcb_fini(void *mem, int size)
142 struct inpcb *inp = mem;
144 INP_LOCK_DESTROY(inp);
151 INP_INFO_LOCK_INIT(&divcbinfo, "div");
153 divcbinfo.ipi_listhead = &divcb;
155 * XXX We don't use the hash list for divert IP, but it's easier
156 * to allocate a one entry hash list than it is to check all
157 * over the place for hashbase == NULL.
159 divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask);
160 divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
161 &divcbinfo.ipi_porthashmask);
162 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
163 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
165 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
166 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
167 NULL, EVENTHANDLER_PRI_ANY);
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)
177 ipstat.ips_noproto++;
182 * Divert a packet by passing it up to the divert socket at port 'port'.
184 * Setup generic address and protocol structures for div_input routine,
185 * then pass them along with mbuf chain.
188 divert_packet(struct mbuf *m, int incoming)
194 struct sockaddr_in divsrc;
197 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
199 printf("%s: no divert tag\n", __func__);
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);
218 * Record receive interface address, if any.
219 * But only for incoming packets.
221 bzero(&divsrc, sizeof(divsrc));
222 divsrc.sin_len = sizeof(divsrc);
223 divsrc.sin_family = AF_INET;
224 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
231 /* Find IP address for receive interface */
232 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
233 if (ifa->ifa_addr->sa_family != AF_INET)
236 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
241 * Record the incoming interface name whenever we have one.
243 if (m->m_pkthdr.rcvif) {
245 * Hide the actual interface name in there in the
246 * sin_zero array. XXX This needs to be moved to a
247 * different sockaddr type for divert, e.g.
248 * sockaddr_div with multiple fields like
249 * sockaddr_dl. Presently we have only 7 bytes
250 * but that will do for now as most interfaces
251 * are 4 or less + 2 or less bytes for unit.
252 * There is probably a faster way of doing this,
253 * possibly taking it from the sockaddr_dl on the iface.
254 * This solves the problem of a P2P link and a LAN interface
255 * having the same address, which can result in the wrong
256 * interface being assigned to the packet when fed back
257 * into the divert socket. Theoretically if the daemon saves
258 * and re-uses the sockaddr_in as suggested in the man pages,
259 * this iface name will come along for the ride.
260 * (see div_output for the other half of this.)
262 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
263 sizeof(divsrc.sin_zero));
266 /* Put packet on socket queue, if any */
268 nport = htons((u_int16_t)divert_info(mtag));
269 INP_INFO_RLOCK(&divcbinfo);
270 LIST_FOREACH(inp, &divcb, inp_list) {
271 /* XXX why does only one socket match? */
272 if (inp->inp_lport == nport) {
274 sa = inp->inp_socket;
275 SOCKBUF_LOCK(&sa->so_rcv);
276 if (sbappendaddr_locked(&sa->so_rcv,
277 (struct sockaddr *)&divsrc, m,
278 (struct mbuf *)0) == 0) {
279 SOCKBUF_UNLOCK(&sa->so_rcv);
280 sa = NULL; /* force mbuf reclaim below */
282 sorwakeup_locked(sa);
287 INP_INFO_RUNLOCK(&divcbinfo);
290 ipstat.ips_noproto++;
291 ipstat.ips_delivered--;
296 * Deliver packet back into the IP processing machinery.
298 * If no address specified, or address is 0.0.0.0, send to ip_output();
299 * otherwise, send to ip_input() and mark as having been received on
300 * the interface with that address.
303 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin,
304 struct mbuf *control)
307 struct divert_tag *dt;
309 struct mbuf *options;
312 * An mbuf may hasn't come from userland, but we pretend
315 m->m_pkthdr.rcvif = NULL;
317 M_SETFIB(m, so->so_fibnum);
320 m_freem(control); /* XXX */
322 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
323 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
329 dt = (struct divert_tag *)(mtag+1);
330 m_tag_prepend(m, mtag);
332 dt = (struct divert_tag *)(mtag+1);
334 /* Loopback avoidance and state recovery */
338 dt->cookie = sin->sin_port;
340 * Find receive interface with the given name, stuffed
341 * (if it exists) in the sin_zero[] field.
342 * The name is user supplied data so don't trust its size
343 * or that it is zero terminated.
345 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
347 if ( i > 0 && i < sizeof(sin->sin_zero))
348 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
351 /* Reinject packet into the system as incoming or outgoing */
352 if (!sin || sin->sin_addr.s_addr == 0) {
353 struct ip *const ip = mtod(m, struct ip *);
356 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
357 INP_INFO_WLOCK(&divcbinfo);
361 * Don't allow both user specified and setsockopt options,
362 * and don't allow packet length sizes that will crash
364 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
365 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
368 INP_INFO_WUNLOCK(&divcbinfo);
371 /* Convert fields to host order for ip_output() */
372 ip->ip_len = ntohs(ip->ip_len);
373 ip->ip_off = ntohs(ip->ip_off);
375 /* Send packet to output processing */
376 ipstat.ips_rawout++; /* XXX */
379 mac_create_mbuf_from_inpcb(inp, m);
382 * Get ready to inject the packet into ip_output().
383 * Just in case socket options were specified on the
384 * divert socket, we duplicate them. This is done
385 * to avoid having to hold the PCB locks over the call
386 * to ip_output(), as doing this results in a number of
387 * lock ordering complexities.
389 * Note that we set the multicast options argument for
390 * ip_output() to NULL since it should be invariant that
391 * they are not present.
393 KASSERT(inp->inp_moptions == NULL,
394 ("multicast options set on a divert socket"));
397 * XXXCSJP: It is unclear to me whether or not it makes
398 * sense for divert sockets to have options. However,
399 * for now we will duplicate them with the INP locks
400 * held so we can use them in ip_output() without
401 * requring a reference to the pcb.
403 if (inp->inp_options != NULL) {
404 options = m_dup(inp->inp_options, M_DONTWAIT);
409 INP_INFO_WUNLOCK(&divcbinfo);
410 if (error == ENOBUFS) {
414 error = ip_output(m, options, NULL,
415 ((so->so_options & SO_DONTROUTE) ?
416 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
417 IP_RAWOUTPUT, NULL, NULL);
422 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
423 if (m->m_pkthdr.rcvif == NULL) {
425 * No luck with the name, check by IP address.
426 * Clear the port and the ifname to make sure
427 * there are no distractions for ifa_ifwithaddr.
431 bzero(sin->sin_zero, sizeof(sin->sin_zero));
433 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
435 error = EADDRNOTAVAIL;
438 m->m_pkthdr.rcvif = ifa->ifa_ifp;
442 mac_create_mbuf_from_socket(so, m);
445 /* Send packet to input processing via netisr */
446 netisr_queue(NETISR_IP, m);
457 div_attach(struct socket *so, int proto, struct thread *td)
463 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
465 error = priv_check(td, PRIV_NETINET_DIVERT);
469 error = soreserve(so, div_sendspace, div_recvspace);
472 INP_INFO_WLOCK(&divcbinfo);
473 error = in_pcballoc(so, &divcbinfo);
475 INP_INFO_WUNLOCK(&divcbinfo);
478 inp = (struct inpcb *)so->so_pcb;
479 INP_INFO_WUNLOCK(&divcbinfo);
480 inp->inp_ip_p = proto;
481 inp->inp_vflag |= INP_IPV4;
482 inp->inp_flags |= INP_HDRINCL;
488 div_detach(struct socket *so)
493 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
494 INP_INFO_WLOCK(&divcbinfo);
498 INP_INFO_WUNLOCK(&divcbinfo);
502 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
508 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
509 /* in_pcbbind assumes that nam is a sockaddr_in
510 * and in_pcbbind requires a valid address. Since divert
511 * sockets don't we need to make sure the address is
512 * filled in properly.
513 * XXX -- divert should not be abusing in_pcbind
514 * and should probably have its own family.
516 if (nam->sa_family != AF_INET)
518 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
519 INP_INFO_WLOCK(&divcbinfo);
521 error = in_pcbbind(inp, nam, td->td_ucred);
523 INP_INFO_WUNLOCK(&divcbinfo);
528 div_shutdown(struct socket *so)
533 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
541 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
542 struct mbuf *control, struct thread *td)
544 /* Packet must have a header (but that's about it) */
545 if (m->m_len < sizeof (struct ip) &&
546 (m = m_pullup(m, sizeof (struct ip))) == 0) {
547 ipstat.ips_toosmall++;
553 return div_output(so, m, (struct sockaddr_in *)nam, control);
557 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
559 struct in_addr faddr;
561 faddr = ((struct sockaddr_in *)sa)->sin_addr;
562 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
564 if (PRC_IS_REDIRECT(cmd))
569 div_pcblist(SYSCTL_HANDLER_ARGS)
572 struct inpcb *inp, **inp_list;
577 * The process of preparing the TCB list is too time-consuming and
578 * resource-intensive to repeat twice on every request.
580 if (req->oldptr == 0) {
581 n = divcbinfo.ipi_count;
582 req->oldidx = 2 * (sizeof xig)
583 + (n + n/8) * sizeof(struct xinpcb);
587 if (req->newptr != 0)
591 * OK, now we're committed to doing something.
593 INP_INFO_RLOCK(&divcbinfo);
594 gencnt = divcbinfo.ipi_gencnt;
595 n = divcbinfo.ipi_count;
596 INP_INFO_RUNLOCK(&divcbinfo);
598 error = sysctl_wire_old_buffer(req,
599 2 * sizeof(xig) + n*sizeof(struct xinpcb));
603 xig.xig_len = sizeof xig;
605 xig.xig_gen = gencnt;
606 xig.xig_sogen = so_gencnt;
607 error = SYSCTL_OUT(req, &xig, sizeof xig);
611 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
615 INP_INFO_RLOCK(&divcbinfo);
616 for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n;
617 inp = LIST_NEXT(inp, inp_list)) {
619 if (inp->inp_gencnt <= gencnt &&
620 cr_canseeinpcb(req->td->td_ucred, inp) == 0)
624 INP_INFO_RUNLOCK(&divcbinfo);
628 for (i = 0; i < n; i++) {
631 if (inp->inp_gencnt <= gencnt) {
633 bzero(&xi, sizeof(xi));
634 xi.xi_len = sizeof xi;
635 /* XXX should avoid extra copy */
636 bcopy(inp, &xi.xi_inp, sizeof *inp);
638 sotoxsocket(inp->inp_socket, &xi.xi_socket);
640 error = SYSCTL_OUT(req, &xi, sizeof xi);
646 * Give the user an updated idea of our state.
647 * If the generation differs from what we told
648 * her before, she knows that something happened
649 * while we were processing this request, and it
650 * might be necessary to retry.
652 INP_INFO_RLOCK(&divcbinfo);
653 xig.xig_gen = divcbinfo.ipi_gencnt;
654 xig.xig_sogen = so_gencnt;
655 xig.xig_count = divcbinfo.ipi_count;
656 INP_INFO_RUNLOCK(&divcbinfo);
657 error = SYSCTL_OUT(req, &xig, sizeof xig);
659 free(inp_list, M_TEMP);
664 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
665 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
666 div_pcblist, "S,xinpcb", "List of active divert sockets");
669 struct pr_usrreqs div_usrreqs = {
670 .pru_attach = div_attach,
671 .pru_bind = div_bind,
672 .pru_control = in_control,
673 .pru_detach = div_detach,
674 .pru_peeraddr = in_getpeeraddr,
675 .pru_send = div_send,
676 .pru_shutdown = div_shutdown,
677 .pru_sockaddr = in_getsockaddr,
678 .pru_sosetlabel = in_pcbsosetlabel
681 struct protosw div_protosw = {
683 .pr_protocol = IPPROTO_DIVERT,
684 .pr_flags = PR_ATOMIC|PR_ADDR,
685 .pr_input = div_input,
686 .pr_ctlinput = div_ctlinput,
687 .pr_ctloutput = ip_ctloutput,
689 .pr_usrreqs = &div_usrreqs
693 div_modevent(module_t mod, int type, void *unused)
701 * Protocol will be initialized by pf_proto_register().
702 * We don't have to register ip_protox because we are not
703 * a true IP protocol that goes over the wire.
705 err = pf_proto_register(PF_INET, &div_protosw);
706 ip_divert_ptr = divert_packet;
710 * IPDIVERT may normally not be unloaded because of the
711 * potential race conditions. Tell kldunload we can't be
712 * unloaded unless the unload is forced.
720 * Module ipdivert can only be unloaded if no sockets are
721 * connected. Maybe this can be changed later to forcefully
722 * disconnect any open sockets.
724 * XXXRW: Note that there is a slight race here, as a new
725 * socket open request could be spinning on the lock and then
726 * we destroy the lock.
728 INP_INFO_WLOCK(&divcbinfo);
729 n = divcbinfo.ipi_count;
732 INP_INFO_WUNLOCK(&divcbinfo);
735 ip_divert_ptr = NULL;
736 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
737 INP_INFO_WUNLOCK(&divcbinfo);
738 INP_INFO_LOCK_DESTROY(&divcbinfo);
739 uma_zdestroy(divcbinfo.ipi_zone);
748 static moduledata_t ipdivertmod = {
754 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
755 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
756 MODULE_VERSION(ipdivert, 1);