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/route.h>
66 #include <netinet/in.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip_divert.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/ip_fw.h>
75 #include <security/mac/mac_framework.h>
82 * Allocate enough space to hold a full IP packet
84 #define DIVSNDQ (65536 + 100)
85 #define DIVRCVQ (65536 + 100)
88 * Divert sockets work in conjunction with ipfw, see the divert(4)
89 * manpage for features.
90 * Internally, packets selected by ipfw in ip_input() or ip_output(),
91 * and never diverted before, are passed to the input queue of the
92 * divert socket with a given 'divert_port' number (as specified in
93 * the matching ipfw rule), and they are tagged with a 16 bit cookie
94 * (representing the rule number of the matching ipfw rule), which
95 * is passed to process reading from the socket.
97 * Packets written to the divert socket are again tagged with a cookie
98 * (usually the same as above) and a destination address.
99 * If the destination address is INADDR_ANY then the packet is
100 * treated as outgoing and sent to ip_output(), otherwise it is
101 * treated as incoming and sent to ip_input().
102 * In both cases, the packet is tagged with the cookie.
104 * On reinjection, processing in ip_input() and ip_output()
105 * will be exactly the same as for the original packet, except that
106 * ipfw processing will start at the rule number after the one
107 * written in the cookie (so, tagging a packet with a cookie of 0
108 * will cause it to be effectively considered as a standard packet).
111 /* Internal variables. */
112 static struct inpcbhead divcb;
113 static struct inpcbinfo divcbinfo;
115 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
116 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
119 * Initialize divert connection block queue.
122 div_zone_change(void *tag)
125 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
129 div_inpcb_init(void *mem, int size, int flags)
131 struct inpcb *inp = (struct inpcb *) mem;
132 INP_LOCK_INIT(inp, "inp", "divinp");
137 div_inpcb_fini(void *mem, int size)
139 struct inpcb *inp = (struct inpcb *) mem;
140 INP_LOCK_DESTROY(inp);
147 INP_INFO_LOCK_INIT(&divcbinfo, "div");
149 divcbinfo.listhead = &divcb;
151 * XXX We don't use the hash list for divert IP, but it's easier
152 * to allocate a one entry hash list than it is to check all
153 * over the place for hashbase == NULL.
155 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
156 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
157 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
158 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
159 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
160 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change,
161 NULL, EVENTHANDLER_PRI_ANY);
165 * IPPROTO_DIVERT is not in the real IP protocol number space; this
166 * function should never be called. Just in case, drop any packets.
169 div_input(struct mbuf *m, int off)
171 ipstat.ips_noproto++;
176 * Divert a packet by passing it up to the divert socket at port 'port'.
178 * Setup generic address and protocol structures for div_input routine,
179 * then pass them along with mbuf chain.
182 divert_packet(struct mbuf *m, int incoming)
188 struct sockaddr_in divsrc;
191 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
193 printf("%s: no divert tag\n", __func__);
198 if (m->m_len < sizeof(struct ip) &&
199 (m = m_pullup(m, sizeof(struct ip))) == 0)
201 ip = mtod(m, struct ip *);
203 /* Delayed checksums are currently not compatible with divert. */
204 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
205 ip->ip_len = ntohs(ip->ip_len);
207 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
208 ip->ip_len = htons(ip->ip_len);
212 * Record receive interface address, if any.
213 * But only for incoming packets.
215 bzero(&divsrc, sizeof(divsrc));
216 divsrc.sin_len = sizeof(divsrc);
217 divsrc.sin_family = AF_INET;
218 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
225 /* Find IP address for receive interface */
226 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
227 if (ifa->ifa_addr->sa_family != AF_INET)
230 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
235 * Record the incoming interface name whenever we have one.
237 if (m->m_pkthdr.rcvif) {
239 * Hide the actual interface name in there in the
240 * sin_zero array. XXX This needs to be moved to a
241 * different sockaddr type for divert, e.g.
242 * sockaddr_div with multiple fields like
243 * sockaddr_dl. Presently we have only 7 bytes
244 * but that will do for now as most interfaces
245 * are 4 or less + 2 or less bytes for unit.
246 * There is probably a faster way of doing this,
247 * possibly taking it from the sockaddr_dl on the iface.
248 * This solves the problem of a P2P link and a LAN interface
249 * having the same address, which can result in the wrong
250 * interface being assigned to the packet when fed back
251 * into the divert socket. Theoretically if the daemon saves
252 * and re-uses the sockaddr_in as suggested in the man pages,
253 * this iface name will come along for the ride.
254 * (see div_output for the other half of this.)
256 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
257 sizeof(divsrc.sin_zero));
260 /* Put packet on socket queue, if any */
262 nport = htons((u_int16_t)divert_info(mtag));
263 INP_INFO_RLOCK(&divcbinfo);
264 LIST_FOREACH(inp, &divcb, inp_list) {
266 /* XXX why does only one socket match? */
267 if (inp->inp_lport == nport) {
268 sa = inp->inp_socket;
269 SOCKBUF_LOCK(&sa->so_rcv);
270 if (sbappendaddr_locked(&sa->so_rcv,
271 (struct sockaddr *)&divsrc, m,
272 (struct mbuf *)0) == 0) {
273 SOCKBUF_UNLOCK(&sa->so_rcv);
274 sa = NULL; /* force mbuf reclaim below */
276 sorwakeup_locked(sa);
282 INP_INFO_RUNLOCK(&divcbinfo);
285 ipstat.ips_noproto++;
286 ipstat.ips_delivered--;
291 * Deliver packet back into the IP processing machinery.
293 * If no address specified, or address is 0.0.0.0, send to ip_output();
294 * otherwise, send to ip_input() and mark as having been received on
295 * the interface with that address.
298 div_output(struct socket *so, struct mbuf *m,
299 struct sockaddr_in *sin, struct mbuf *control)
302 struct divert_tag *dt;
306 * An mbuf may hasn't come from userland, but we pretend
309 m->m_pkthdr.rcvif = NULL;
313 m_freem(control); /* XXX */
315 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) {
316 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag),
322 dt = (struct divert_tag *)(mtag+1);
323 m_tag_prepend(m, mtag);
325 dt = (struct divert_tag *)(mtag+1);
327 /* Loopback avoidance and state recovery */
331 dt->cookie = sin->sin_port;
333 * Find receive interface with the given name, stuffed
334 * (if it exists) in the sin_zero[] field.
335 * The name is user supplied data so don't trust its size
336 * or that it is zero terminated.
338 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
340 if ( i > 0 && i < sizeof(sin->sin_zero))
341 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
344 /* Reinject packet into the system as incoming or outgoing */
345 if (!sin || sin->sin_addr.s_addr == 0) {
346 struct ip *const ip = mtod(m, struct ip *);
349 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
350 INP_INFO_WLOCK(&divcbinfo);
354 * Don't allow both user specified and setsockopt options,
355 * and don't allow packet length sizes that will crash
357 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
358 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
362 /* Convert fields to host order for ip_output() */
363 ip->ip_len = ntohs(ip->ip_len);
364 ip->ip_off = ntohs(ip->ip_off);
366 /* Send packet to output processing */
367 ipstat.ips_rawout++; /* XXX */
370 mac_create_mbuf_from_inpcb(inp, m);
373 inp->inp_options, NULL,
374 ((so->so_options & SO_DONTROUTE) ?
376 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
377 inp->inp_moptions, NULL);
380 INP_INFO_WUNLOCK(&divcbinfo);
382 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
383 if (m->m_pkthdr.rcvif == NULL) {
385 * No luck with the name, check by IP address.
386 * Clear the port and the ifname to make sure
387 * there are no distractions for ifa_ifwithaddr.
391 bzero(sin->sin_zero, sizeof(sin->sin_zero));
393 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
395 error = EADDRNOTAVAIL;
398 m->m_pkthdr.rcvif = ifa->ifa_ifp;
402 mac_create_mbuf_from_socket(so, m);
405 /* Send packet to input processing */
417 div_attach(struct socket *so, int proto, struct thread *td)
423 KASSERT(inp == NULL, ("div_attach: inp != NULL"));
425 error = priv_check(td, PRIV_NETINET_DIVERT);
429 error = soreserve(so, div_sendspace, div_recvspace);
432 INP_INFO_WLOCK(&divcbinfo);
433 error = in_pcballoc(so, &divcbinfo);
435 INP_INFO_WUNLOCK(&divcbinfo);
438 inp = (struct inpcb *)so->so_pcb;
439 INP_INFO_WUNLOCK(&divcbinfo);
440 inp->inp_ip_p = proto;
441 inp->inp_vflag |= INP_IPV4;
442 inp->inp_flags |= INP_HDRINCL;
448 div_detach(struct socket *so)
453 KASSERT(inp != NULL, ("div_detach: inp == NULL"));
454 INP_INFO_WLOCK(&divcbinfo);
458 INP_INFO_WUNLOCK(&divcbinfo);
462 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
468 KASSERT(inp != NULL, ("div_bind: inp == NULL"));
469 /* in_pcbbind assumes that nam is a sockaddr_in
470 * and in_pcbbind requires a valid address. Since divert
471 * sockets don't we need to make sure the address is
472 * filled in properly.
473 * XXX -- divert should not be abusing in_pcbind
474 * and should probably have its own family.
476 if (nam->sa_family != AF_INET)
478 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
479 INP_INFO_WLOCK(&divcbinfo);
481 error = in_pcbbind(inp, nam, td->td_ucred);
483 INP_INFO_WUNLOCK(&divcbinfo);
488 div_shutdown(struct socket *so)
493 KASSERT(inp != NULL, ("div_shutdown: inp == NULL"));
501 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
502 struct mbuf *control, struct thread *td)
504 /* Packet must have a header (but that's about it) */
505 if (m->m_len < sizeof (struct ip) &&
506 (m = m_pullup(m, sizeof (struct ip))) == 0) {
507 ipstat.ips_toosmall++;
513 return div_output(so, m, (struct sockaddr_in *)nam, control);
517 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
519 struct in_addr faddr;
521 faddr = ((struct sockaddr_in *)sa)->sin_addr;
522 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
524 if (PRC_IS_REDIRECT(cmd))
529 div_pcblist(SYSCTL_HANDLER_ARGS)
532 struct inpcb *inp, **inp_list;
537 * The process of preparing the TCB list is too time-consuming and
538 * resource-intensive to repeat twice on every request.
540 if (req->oldptr == 0) {
541 n = divcbinfo.ipi_count;
542 req->oldidx = 2 * (sizeof xig)
543 + (n + n/8) * sizeof(struct xinpcb);
547 if (req->newptr != 0)
551 * OK, now we're committed to doing something.
553 INP_INFO_RLOCK(&divcbinfo);
554 gencnt = divcbinfo.ipi_gencnt;
555 n = divcbinfo.ipi_count;
556 INP_INFO_RUNLOCK(&divcbinfo);
558 error = sysctl_wire_old_buffer(req,
559 2 * sizeof(xig) + n*sizeof(struct xinpcb));
563 xig.xig_len = sizeof xig;
565 xig.xig_gen = gencnt;
566 xig.xig_sogen = so_gencnt;
567 error = SYSCTL_OUT(req, &xig, sizeof xig);
571 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
575 INP_INFO_RLOCK(&divcbinfo);
576 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
577 inp = LIST_NEXT(inp, inp_list)) {
579 if (inp->inp_gencnt <= gencnt &&
580 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
584 INP_INFO_RUNLOCK(&divcbinfo);
588 for (i = 0; i < n; i++) {
591 if (inp->inp_gencnt <= gencnt) {
593 bzero(&xi, sizeof(xi));
594 xi.xi_len = sizeof xi;
595 /* XXX should avoid extra copy */
596 bcopy(inp, &xi.xi_inp, sizeof *inp);
598 sotoxsocket(inp->inp_socket, &xi.xi_socket);
600 error = SYSCTL_OUT(req, &xi, sizeof xi);
606 * Give the user an updated idea of our state.
607 * If the generation differs from what we told
608 * her before, she knows that something happened
609 * while we were processing this request, and it
610 * might be necessary to retry.
612 INP_INFO_RLOCK(&divcbinfo);
613 xig.xig_gen = divcbinfo.ipi_gencnt;
614 xig.xig_sogen = so_gencnt;
615 xig.xig_count = divcbinfo.ipi_count;
616 INP_INFO_RUNLOCK(&divcbinfo);
617 error = SYSCTL_OUT(req, &xig, sizeof xig);
619 free(inp_list, M_TEMP);
624 * This is the wrapper function for in_setsockaddr. We just pass down
625 * the pcbinfo for in_setpeeraddr to lock.
628 div_sockaddr(struct socket *so, struct sockaddr **nam)
630 return (in_setsockaddr(so, nam, &divcbinfo));
634 * This is the wrapper function for in_setpeeraddr. We just pass down
635 * the pcbinfo for in_setpeeraddr to lock.
638 div_peeraddr(struct socket *so, struct sockaddr **nam)
640 return (in_setpeeraddr(so, nam, &divcbinfo));
644 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
645 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
646 div_pcblist, "S,xinpcb", "List of active divert sockets");
649 struct pr_usrreqs div_usrreqs = {
650 .pru_attach = div_attach,
651 .pru_bind = div_bind,
652 .pru_control = in_control,
653 .pru_detach = div_detach,
654 .pru_peeraddr = div_peeraddr,
655 .pru_send = div_send,
656 .pru_shutdown = div_shutdown,
657 .pru_sockaddr = div_sockaddr,
658 .pru_sosetlabel = in_pcbsosetlabel
661 struct protosw div_protosw = {
663 .pr_protocol = IPPROTO_DIVERT,
664 .pr_flags = PR_ATOMIC|PR_ADDR,
665 .pr_input = div_input,
666 .pr_ctlinput = div_ctlinput,
667 .pr_ctloutput = ip_ctloutput,
669 .pr_usrreqs = &div_usrreqs
673 div_modevent(module_t mod, int type, void *unused)
681 * Protocol will be initialized by pf_proto_register().
682 * We don't have to register ip_protox because we are not
683 * a true IP protocol that goes over the wire.
685 err = pf_proto_register(PF_INET, &div_protosw);
686 ip_divert_ptr = divert_packet;
690 * IPDIVERT may normally not be unloaded because of the
691 * potential race conditions. Tell kldunload we can't be
692 * unloaded unless the unload is forced.
700 * Module ipdivert can only be unloaded if no sockets are
701 * connected. Maybe this can be changed later to forcefully
702 * disconnect any open sockets.
704 * XXXRW: Note that there is a slight race here, as a new
705 * socket open request could be spinning on the lock and then
706 * we destroy the lock.
708 INP_INFO_WLOCK(&divcbinfo);
709 n = divcbinfo.ipi_count;
712 INP_INFO_WUNLOCK(&divcbinfo);
715 ip_divert_ptr = NULL;
716 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
717 INP_INFO_WUNLOCK(&divcbinfo);
718 INP_INFO_LOCK_DESTROY(&divcbinfo);
719 uma_zdestroy(divcbinfo.ipi_zone);
728 static moduledata_t ipdivertmod = {
734 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
735 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
736 MODULE_VERSION(ipdivert, 1);