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
32 #if !defined(KLD_MODULE)
37 #error "IPDIVERT requires INET."
40 #error "IPDIVERT requires IPFIREWALL"
44 #include <sys/param.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/signalvar.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
64 #include <net/netisr.h>
65 #include <net/route.h>
67 #include <netinet/in.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip.h>
72 #include <netinet/ip_divert.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_fw.h>
76 #include <security/mac/mac_framework.h>
83 * Allocate enough space to hold a full IP packet
85 #define DIVSNDQ (65536 + 100)
86 #define DIVRCVQ (65536 + 100)
89 * Divert sockets work in conjunction with ipfw, see the divert(4)
90 * manpage for features.
91 * Internally, packets selected by ipfw in ip_input() or ip_output(),
92 * and never diverted before, are passed to the input queue of the
93 * divert socket with a given 'divert_port' number (as specified in
94 * the matching ipfw rule), and they are tagged with a 16 bit cookie
95 * (representing the rule number of the matching ipfw rule), which
96 * is passed to process reading from the socket.
98 * Packets written to the divert socket are again tagged with a cookie
99 * (usually the same as above) and a destination address.
100 * If the destination address is INADDR_ANY then the packet is
101 * treated as outgoing and sent to ip_output(), otherwise it is
102 * treated as incoming and sent to ip_input().
103 * In both cases, the packet is tagged with the cookie.
105 * On reinjection, processing in ip_input() and ip_output()
106 * will be exactly the same as for the original packet, except that
107 * ipfw processing will start at the rule number after the one
108 * written in the cookie (so, tagging a packet with a cookie of 0
109 * will cause it to be effectively considered as a standard packet).
112 /* Internal variables. */
113 static struct inpcbhead divcb;
114 static struct inpcbinfo divcbinfo;
116 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
117 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
120 * Initialize divert connection block queue.
123 div_zone_change(void *tag)
126 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
130 div_inpcb_init(void *mem, int size, int flags)
132 struct inpcb *inp = mem;
134 INP_LOCK_INIT(inp, "inp", "divinp");
139 div_inpcb_fini(void *mem, int size)
141 struct inpcb *inp = mem;
143 INP_LOCK_DESTROY(inp);
150 INP_INFO_LOCK_INIT(&divcbinfo, "div");
152 divcbinfo.ipi_listhead = &divcb;
154 * XXX We don't use the hash list for divert IP, but it's easier
155 * to allocate a one entry hash list than it is to check all
156 * over the place for hashbase == NULL.
158 divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask);
159 divcbinfo.ipi_porthashbase = hashinit(1, M_PCB,
160 &divcbinfo.ipi_porthashmask);
161 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
162 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR,
164 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
165 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
166 NULL, EVENTHANDLER_PRI_ANY);
170 * IPPROTO_DIVERT is not in the real IP protocol number space; this
171 * function should never be called. Just in case, drop any packets.
174 div_input(struct mbuf *m, int off)
176 ipstat.ips_noproto++;
181 * Divert a packet by passing it up to the divert socket at port 'port'.
183 * Setup generic address and protocol structures for div_input routine,
184 * then pass them along with mbuf chain.
187 divert_packet(struct mbuf *m, int incoming)
193 struct sockaddr_in divsrc;
196 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
198 printf("%s: no divert tag\n", __func__);
203 if (m->m_len < sizeof(struct ip) &&
204 (m = m_pullup(m, sizeof(struct ip))) == 0)
206 ip = mtod(m, struct ip *);
208 /* Delayed checksums are currently not compatible with divert. */
209 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
210 ip->ip_len = ntohs(ip->ip_len);
212 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
213 ip->ip_len = htons(ip->ip_len);
217 * Record receive interface address, if any.
218 * But only for incoming packets.
220 bzero(&divsrc, sizeof(divsrc));
221 divsrc.sin_len = sizeof(divsrc);
222 divsrc.sin_family = AF_INET;
223 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
230 /* Find IP address for receive interface */
231 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
232 if (ifa->ifa_addr->sa_family != AF_INET)
235 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
240 * Record the incoming interface name whenever we have one.
242 if (m->m_pkthdr.rcvif) {
244 * Hide the actual interface name in there in the
245 * sin_zero array. XXX This needs to be moved to a
246 * different sockaddr type for divert, e.g.
247 * sockaddr_div with multiple fields like
248 * sockaddr_dl. Presently we have only 7 bytes
249 * but that will do for now as most interfaces
250 * are 4 or less + 2 or less bytes for unit.
251 * There is probably a faster way of doing this,
252 * possibly taking it from the sockaddr_dl on the iface.
253 * This solves the problem of a P2P link and a LAN interface
254 * having the same address, which can result in the wrong
255 * interface being assigned to the packet when fed back
256 * into the divert socket. Theoretically if the daemon saves
257 * and re-uses the sockaddr_in as suggested in the man pages,
258 * this iface name will come along for the ride.
259 * (see div_output for the other half of this.)
261 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
262 sizeof(divsrc.sin_zero));
265 /* Put packet on socket queue, if any */
267 nport = htons((u_int16_t)divert_info(mtag));
268 INP_INFO_RLOCK(&divcbinfo);
269 LIST_FOREACH(inp, &divcb, inp_list) {
271 /* XXX why does only one socket match? */
272 if (inp->inp_lport == nport) {
273 sa = inp->inp_socket;
274 SOCKBUF_LOCK(&sa->so_rcv);
275 if (sbappendaddr_locked(&sa->so_rcv,
276 (struct sockaddr *)&divsrc, m,
277 (struct mbuf *)0) == 0) {
278 SOCKBUF_UNLOCK(&sa->so_rcv);
279 sa = NULL; /* force mbuf reclaim below */
281 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;
319 m_freem(control); /* XXX */
321 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
322 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
328 dt = (struct divert_tag *)(mtag+1);
329 m_tag_prepend(m, mtag);
331 dt = (struct divert_tag *)(mtag+1);
333 /* Loopback avoidance and state recovery */
337 dt->cookie = sin->sin_port;
339 * Find receive interface with the given name, stuffed
340 * (if it exists) in the sin_zero[] field.
341 * The name is user supplied data so don't trust its size
342 * or that it is zero terminated.
344 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
346 if ( i > 0 && i < sizeof(sin->sin_zero))
347 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
350 /* Reinject packet into the system as incoming or outgoing */
351 if (!sin || sin->sin_addr.s_addr == 0) {
352 struct ip *const ip = mtod(m, struct ip *);
355 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
356 INP_INFO_WLOCK(&divcbinfo);
360 * Don't allow both user specified and setsockopt options,
361 * and don't allow packet length sizes that will crash
363 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
364 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
367 INP_INFO_WUNLOCK(&divcbinfo);
370 /* Convert fields to host order for ip_output() */
371 ip->ip_len = ntohs(ip->ip_len);
372 ip->ip_off = ntohs(ip->ip_off);
374 /* Send packet to output processing */
375 ipstat.ips_rawout++; /* XXX */
378 mac_create_mbuf_from_inpcb(inp, m);
381 * Get ready to inject the packet into ip_output().
382 * Just in case socket options were specified on the
383 * divert socket, we duplicate them. This is done
384 * to avoid having to hold the PCB locks over the call
385 * to ip_output(), as doing this results in a number of
386 * lock ordering complexities.
388 * Note that we set the multicast options argument for
389 * ip_output() to NULL since it should be invariant that
390 * they are not present.
392 KASSERT(inp->inp_moptions == NULL,
393 ("multicast options set on a divert socket"));
396 * XXXCSJP: It is unclear to me whether or not it makes
397 * sense for divert sockets to have options. However,
398 * for now we will duplicate them with the INP locks
399 * held so we can use them in ip_output() without
400 * requring a reference to the pcb.
402 if (inp->inp_options != NULL) {
403 options = m_dup(inp->inp_options, M_DONTWAIT);
408 INP_INFO_WUNLOCK(&divcbinfo);
409 if (error == ENOBUFS) {
413 error = ip_output(m, options, NULL,
414 ((so->so_options & SO_DONTROUTE) ?
415 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST |
416 IP_RAWOUTPUT, NULL, NULL);
421 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
422 if (m->m_pkthdr.rcvif == NULL) {
424 * No luck with the name, check by IP address.
425 * Clear the port and the ifname to make sure
426 * there are no distractions for ifa_ifwithaddr.
430 bzero(sin->sin_zero, sizeof(sin->sin_zero));
432 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
434 error = EADDRNOTAVAIL;
437 m->m_pkthdr.rcvif = ifa->ifa_ifp;
441 mac_create_mbuf_from_socket(so, m);
444 /* Send packet to input processing via netisr */
445 netisr_queue(NETISR_IP, m);
456 div_attach(struct socket *so, int proto, struct thread *td)
462 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
464 error = priv_check(td, PRIV_NETINET_DIVERT);
468 error = soreserve(so, div_sendspace, div_recvspace);
471 INP_INFO_WLOCK(&divcbinfo);
472 error = in_pcballoc(so, &divcbinfo);
474 INP_INFO_WUNLOCK(&divcbinfo);
477 inp = (struct inpcb *)so->so_pcb;
478 INP_INFO_WUNLOCK(&divcbinfo);
479 inp->inp_ip_p = proto;
480 inp->inp_vflag |= INP_IPV4;
481 inp->inp_flags |= INP_HDRINCL;
487 div_detach(struct socket *so)
492 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
493 INP_INFO_WLOCK(&divcbinfo);
497 INP_INFO_WUNLOCK(&divcbinfo);
501 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
507 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
508 /* in_pcbbind assumes that nam is a sockaddr_in
509 * and in_pcbbind requires a valid address. Since divert
510 * sockets don't we need to make sure the address is
511 * filled in properly.
512 * XXX -- divert should not be abusing in_pcbind
513 * and should probably have its own family.
515 if (nam->sa_family != AF_INET)
517 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
518 INP_INFO_WLOCK(&divcbinfo);
520 error = in_pcbbind(inp, nam, td->td_ucred);
522 INP_INFO_WUNLOCK(&divcbinfo);
527 div_shutdown(struct socket *so)
532 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
540 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
541 struct mbuf *control, struct thread *td)
543 /* Packet must have a header (but that's about it) */
544 if (m->m_len < sizeof (struct ip) &&
545 (m = m_pullup(m, sizeof (struct ip))) == 0) {
546 ipstat.ips_toosmall++;
552 return div_output(so, m, (struct sockaddr_in *)nam, control);
556 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
558 struct in_addr faddr;
560 faddr = ((struct sockaddr_in *)sa)->sin_addr;
561 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
563 if (PRC_IS_REDIRECT(cmd))
568 div_pcblist(SYSCTL_HANDLER_ARGS)
571 struct inpcb *inp, **inp_list;
576 * The process of preparing the TCB list is too time-consuming and
577 * resource-intensive to repeat twice on every request.
579 if (req->oldptr == 0) {
580 n = divcbinfo.ipi_count;
581 req->oldidx = 2 * (sizeof xig)
582 + (n + n/8) * sizeof(struct xinpcb);
586 if (req->newptr != 0)
590 * OK, now we're committed to doing something.
592 INP_INFO_RLOCK(&divcbinfo);
593 gencnt = divcbinfo.ipi_gencnt;
594 n = divcbinfo.ipi_count;
595 INP_INFO_RUNLOCK(&divcbinfo);
597 error = sysctl_wire_old_buffer(req,
598 2 * sizeof(xig) + n*sizeof(struct xinpcb));
602 xig.xig_len = sizeof xig;
604 xig.xig_gen = gencnt;
605 xig.xig_sogen = so_gencnt;
606 error = SYSCTL_OUT(req, &xig, sizeof xig);
610 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
614 INP_INFO_RLOCK(&divcbinfo);
615 for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n;
616 inp = LIST_NEXT(inp, inp_list)) {
618 if (inp->inp_gencnt <= gencnt &&
619 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
623 INP_INFO_RUNLOCK(&divcbinfo);
627 for (i = 0; i < n; i++) {
630 if (inp->inp_gencnt <= gencnt) {
632 bzero(&xi, sizeof(xi));
633 xi.xi_len = sizeof xi;
634 /* XXX should avoid extra copy */
635 bcopy(inp, &xi.xi_inp, sizeof *inp);
637 sotoxsocket(inp->inp_socket, &xi.xi_socket);
639 error = SYSCTL_OUT(req, &xi, sizeof xi);
645 * Give the user an updated idea of our state.
646 * If the generation differs from what we told
647 * her before, she knows that something happened
648 * while we were processing this request, and it
649 * might be necessary to retry.
651 INP_INFO_RLOCK(&divcbinfo);
652 xig.xig_gen = divcbinfo.ipi_gencnt;
653 xig.xig_sogen = so_gencnt;
654 xig.xig_count = divcbinfo.ipi_count;
655 INP_INFO_RUNLOCK(&divcbinfo);
656 error = SYSCTL_OUT(req, &xig, sizeof xig);
658 free(inp_list, M_TEMP);
663 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
664 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
665 div_pcblist, "S,xinpcb", "List of active divert sockets");
668 struct pr_usrreqs div_usrreqs = {
669 .pru_attach = div_attach,
670 .pru_bind = div_bind,
671 .pru_control = in_control,
672 .pru_detach = div_detach,
673 .pru_peeraddr = in_getpeeraddr,
674 .pru_send = div_send,
675 .pru_shutdown = div_shutdown,
676 .pru_sockaddr = in_getsockaddr,
677 .pru_sosetlabel = in_pcbsosetlabel
680 struct protosw div_protosw = {
682 .pr_protocol = IPPROTO_DIVERT,
683 .pr_flags = PR_ATOMIC|PR_ADDR,
684 .pr_input = div_input,
685 .pr_ctlinput = div_ctlinput,
686 .pr_ctloutput = ip_ctloutput,
688 .pr_usrreqs = &div_usrreqs
692 div_modevent(module_t mod, int type, void *unused)
700 * Protocol will be initialized by pf_proto_register().
701 * We don't have to register ip_protox because we are not
702 * a true IP protocol that goes over the wire.
704 err = pf_proto_register(PF_INET, &div_protosw);
705 ip_divert_ptr = divert_packet;
709 * IPDIVERT may normally not be unloaded because of the
710 * potential race conditions. Tell kldunload we can't be
711 * unloaded unless the unload is forced.
719 * Module ipdivert can only be unloaded if no sockets are
720 * connected. Maybe this can be changed later to forcefully
721 * disconnect any open sockets.
723 * XXXRW: Note that there is a slight race here, as a new
724 * socket open request could be spinning on the lock and then
725 * we destroy the lock.
727 INP_INFO_WLOCK(&divcbinfo);
728 n = divcbinfo.ipi_count;
731 INP_INFO_WUNLOCK(&divcbinfo);
734 ip_divert_ptr = NULL;
735 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
736 INP_INFO_WUNLOCK(&divcbinfo);
737 INP_INFO_LOCK_DESTROY(&divcbinfo);
738 uma_zdestroy(divcbinfo.ipi_zone);
747 static moduledata_t ipdivertmod = {
753 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
754 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
755 MODULE_VERSION(ipdivert, 1);