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)
36 #error "IPDIVERT requires INET."
40 #include <sys/param.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/signalvar.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
54 #include <sys/sysctl.h>
55 #include <sys/systm.h>
60 #include <net/route.h>
62 #include <netinet/in.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_divert.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_fw.h>
76 * Allocate enough space to hold a full IP packet
78 #define DIVSNDQ (65536 + 100)
79 #define DIVRCVQ (65536 + 100)
82 * Divert sockets work in conjunction with ipfw, see the divert(4)
83 * manpage for features.
84 * Internally, packets selected by ipfw in ip_input() or ip_output(),
85 * and never diverted before, are passed to the input queue of the
86 * divert socket with a given 'divert_port' number (as specified in
87 * the matching ipfw rule), and they are tagged with a 16 bit cookie
88 * (representing the rule number of the matching ipfw rule), which
89 * is passed to process reading from the socket.
91 * Packets written to the divert socket are again tagged with a cookie
92 * (usually the same as above) and a destination address.
93 * If the destination address is INADDR_ANY then the packet is
94 * treated as outgoing and sent to ip_output(), otherwise it is
95 * treated as incoming and sent to ip_input().
96 * In both cases, the packet is tagged with the cookie.
98 * On reinjection, processing in ip_input() and ip_output()
99 * will be exactly the same as for the original packet, except that
100 * ipfw processing will start at the rule number after the one
101 * written in the cookie (so, tagging a packet with a cookie of 0
102 * will cause it to be effectively considered as a standard packet).
105 /* Internal variables. */
106 static struct inpcbhead divcb;
107 static struct inpcbinfo divcbinfo;
109 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
110 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
113 * Initialize divert connection block queue.
118 INP_INFO_LOCK_INIT(&divcbinfo, "div");
120 divcbinfo.listhead = &divcb;
122 * XXX We don't use the hash list for divert IP, but it's easier
123 * to allocate a one entry hash list than it is to check all
124 * over the place for hashbase == NULL.
126 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
127 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
128 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
129 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
130 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
134 * IPPROTO_DIVERT is not in the real IP protocol number space; this
135 * function should never be called. Just in case, drop any packets.
138 div_input(struct mbuf *m, int off)
140 ipstat.ips_noproto++;
145 * Divert a packet by passing it up to the divert socket at port 'port'.
147 * Setup generic address and protocol structures for div_input routine,
148 * then pass them along with mbuf chain.
151 divert_packet(struct mbuf *m, int incoming)
157 struct sockaddr_in divsrc;
160 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
162 printf("%s: no divert tag\n", __func__);
167 if (m->m_len < sizeof(struct ip) &&
168 (m = m_pullup(m, sizeof(struct ip))) == 0)
170 ip = mtod(m, struct ip *);
172 /* Delayed checksums are currently not compatible with divert. */
173 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
174 ip->ip_len = ntohs(ip->ip_len);
176 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
177 ip->ip_len = htons(ip->ip_len);
181 * Record receive interface address, if any.
182 * But only for incoming packets.
184 bzero(&divsrc, sizeof(divsrc));
185 divsrc.sin_len = sizeof(divsrc);
186 divsrc.sin_family = AF_INET;
187 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
194 /* Find IP address for receive interface */
195 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
196 if (ifa->ifa_addr == NULL)
198 if (ifa->ifa_addr->sa_family != AF_INET)
201 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
206 * Record the incoming interface name whenever we have one.
208 if (m->m_pkthdr.rcvif) {
210 * Hide the actual interface name in there in the
211 * sin_zero array. XXX This needs to be moved to a
212 * different sockaddr type for divert, e.g.
213 * sockaddr_div with multiple fields like
214 * sockaddr_dl. Presently we have only 7 bytes
215 * but that will do for now as most interfaces
216 * are 4 or less + 2 or less bytes for unit.
217 * There is probably a faster way of doing this,
218 * possibly taking it from the sockaddr_dl on the iface.
219 * This solves the problem of a P2P link and a LAN interface
220 * having the same address, which can result in the wrong
221 * interface being assigned to the packet when fed back
222 * into the divert socket. Theoretically if the daemon saves
223 * and re-uses the sockaddr_in as suggested in the man pages,
224 * this iface name will come along for the ride.
225 * (see div_output for the other half of this.)
227 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
228 sizeof(divsrc.sin_zero));
231 /* Put packet on socket queue, if any */
233 nport = htons((u_int16_t)divert_info(mtag));
234 INP_INFO_RLOCK(&divcbinfo);
235 LIST_FOREACH(inp, &divcb, inp_list) {
237 /* XXX why does only one socket match? */
238 if (inp->inp_lport == nport) {
239 sa = inp->inp_socket;
240 SOCKBUF_LOCK(&sa->so_rcv);
241 if (sbappendaddr_locked(&sa->so_rcv,
242 (struct sockaddr *)&divsrc, m,
243 (struct mbuf *)0) == 0) {
244 SOCKBUF_UNLOCK(&sa->so_rcv);
245 sa = NULL; /* force mbuf reclaim below */
247 sorwakeup_locked(sa);
253 INP_INFO_RUNLOCK(&divcbinfo);
256 ipstat.ips_noproto++;
257 ipstat.ips_delivered--;
262 * Deliver packet back into the IP processing machinery.
264 * If no address specified, or address is 0.0.0.0, send to ip_output();
265 * otherwise, send to ip_input() and mark as having been received on
266 * the interface with that address.
269 div_output(struct socket *so, struct mbuf *m,
270 struct sockaddr_in *sin, struct mbuf *control)
273 struct divert_tag *dt;
276 KASSERT(m->m_pkthdr.rcvif == NULL, ("rcvif not null"));
279 m_freem(control); /* XXX */
281 mtag = m_tag_get(PACKET_TAG_DIVERT,
282 sizeof(struct divert_tag), M_NOWAIT);
287 dt = (struct divert_tag *)(mtag+1);
290 m_tag_prepend(m, mtag);
292 /* Loopback avoidance and state recovery */
296 dt->cookie = sin->sin_port;
298 * Find receive interface with the given name, stuffed
299 * (if it exists) in the sin_zero[] field.
300 * The name is user supplied data so don't trust its size
301 * or that it is zero terminated.
303 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
305 if ( i > 0 && i < sizeof(sin->sin_zero))
306 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
309 /* Reinject packet into the system as incoming or outgoing */
310 if (!sin || sin->sin_addr.s_addr == 0) {
311 struct ip *const ip = mtod(m, struct ip *);
314 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG;
315 INP_INFO_WLOCK(&divcbinfo);
319 * Don't allow both user specified and setsockopt options,
320 * and don't allow packet length sizes that will crash
322 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
323 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
327 /* Convert fields to host order for ip_output() */
328 ip->ip_len = ntohs(ip->ip_len);
329 ip->ip_off = ntohs(ip->ip_off);
331 /* Send packet to output processing */
332 ipstat.ips_rawout++; /* XXX */
335 mac_create_mbuf_from_inpcb(inp, m);
338 inp->inp_options, NULL,
339 ((so->so_options & SO_DONTROUTE) ?
341 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
342 inp->inp_moptions, NULL);
345 INP_INFO_WUNLOCK(&divcbinfo);
347 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG;
348 if (m->m_pkthdr.rcvif == NULL) {
350 * No luck with the name, check by IP address.
351 * Clear the port and the ifname to make sure
352 * there are no distractions for ifa_ifwithaddr.
356 bzero(sin->sin_zero, sizeof(sin->sin_zero));
358 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
360 error = EADDRNOTAVAIL;
363 m->m_pkthdr.rcvif = ifa->ifa_ifp;
367 mac_create_mbuf_from_socket(so, m);
370 /* Send packet to input processing */
382 div_attach(struct socket *so, int proto, struct thread *td)
387 INP_INFO_WLOCK(&divcbinfo);
390 INP_INFO_WUNLOCK(&divcbinfo);
393 if (td && (error = suser(td)) != 0) {
394 INP_INFO_WUNLOCK(&divcbinfo);
397 error = soreserve(so, div_sendspace, div_recvspace);
399 INP_INFO_WUNLOCK(&divcbinfo);
402 error = in_pcballoc(so, &divcbinfo, "divinp");
404 INP_INFO_WUNLOCK(&divcbinfo);
407 inp = (struct inpcb *)so->so_pcb;
409 INP_INFO_WUNLOCK(&divcbinfo);
410 inp->inp_ip_p = proto;
411 inp->inp_vflag |= INP_IPV4;
412 inp->inp_flags |= INP_HDRINCL;
413 /* The socket is always "connected" because
414 we always know "where" to send the packet */
417 so->so_state |= SS_ISCONNECTED;
423 div_detach(struct socket *so)
427 INP_INFO_WLOCK(&divcbinfo);
430 INP_INFO_WUNLOCK(&divcbinfo);
435 INP_INFO_WUNLOCK(&divcbinfo);
440 div_abort(struct socket *so)
444 INP_INFO_WLOCK(&divcbinfo);
447 INP_INFO_WUNLOCK(&divcbinfo);
448 return EINVAL; /* ??? possible? panic instead? */
451 soisdisconnected(so);
453 INP_INFO_WUNLOCK(&divcbinfo);
458 div_disconnect(struct socket *so)
460 if ((so->so_state & SS_ISCONNECTED) == 0)
462 return div_abort(so);
466 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
471 INP_INFO_WLOCK(&divcbinfo);
474 INP_INFO_WUNLOCK(&divcbinfo);
477 /* in_pcbbind assumes that nam is a sockaddr_in
478 * and in_pcbbind requires a valid address. Since divert
479 * sockets don't we need to make sure the address is
480 * filled in properly.
481 * XXX -- divert should not be abusing in_pcbind
482 * and should probably have its own family.
484 if (nam->sa_family != AF_INET)
485 error = EAFNOSUPPORT;
487 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
489 error = in_pcbbind(inp, nam, td->td_ucred);
492 INP_INFO_WUNLOCK(&divcbinfo);
497 div_shutdown(struct socket *so)
501 INP_INFO_RLOCK(&divcbinfo);
504 INP_INFO_RUNLOCK(&divcbinfo);
508 INP_INFO_RUNLOCK(&divcbinfo);
515 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
516 struct mbuf *control, struct thread *td)
518 /* Packet must have a header (but that's about it) */
519 if (m->m_len < sizeof (struct ip) &&
520 (m = m_pullup(m, sizeof (struct ip))) == 0) {
521 ipstat.ips_toosmall++;
527 return div_output(so, m, (struct sockaddr_in *)nam, control);
531 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
533 struct in_addr faddr;
535 faddr = ((struct sockaddr_in *)sa)->sin_addr;
536 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
538 if (PRC_IS_REDIRECT(cmd))
543 div_pcblist(SYSCTL_HANDLER_ARGS)
546 struct inpcb *inp, **inp_list;
551 * The process of preparing the TCB list is too time-consuming and
552 * resource-intensive to repeat twice on every request.
554 if (req->oldptr == 0) {
555 n = divcbinfo.ipi_count;
556 req->oldidx = 2 * (sizeof xig)
557 + (n + n/8) * sizeof(struct xinpcb);
561 if (req->newptr != 0)
565 * OK, now we're committed to doing something.
567 INP_INFO_RLOCK(&divcbinfo);
568 gencnt = divcbinfo.ipi_gencnt;
569 n = divcbinfo.ipi_count;
570 INP_INFO_RUNLOCK(&divcbinfo);
572 error = sysctl_wire_old_buffer(req,
573 2 * sizeof(xig) + n*sizeof(struct xinpcb));
577 xig.xig_len = sizeof xig;
579 xig.xig_gen = gencnt;
580 xig.xig_sogen = so_gencnt;
581 error = SYSCTL_OUT(req, &xig, sizeof xig);
585 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
589 INP_INFO_RLOCK(&divcbinfo);
590 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
591 inp = LIST_NEXT(inp, inp_list)) {
593 if (inp->inp_gencnt <= gencnt &&
594 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
598 INP_INFO_RUNLOCK(&divcbinfo);
602 for (i = 0; i < n; i++) {
604 if (inp->inp_gencnt <= gencnt) {
606 xi.xi_len = sizeof xi;
607 /* XXX should avoid extra copy */
608 bcopy(inp, &xi.xi_inp, sizeof *inp);
610 sotoxsocket(inp->inp_socket, &xi.xi_socket);
611 error = SYSCTL_OUT(req, &xi, sizeof xi);
616 * Give the user an updated idea of our state.
617 * If the generation differs from what we told
618 * her before, she knows that something happened
619 * while we were processing this request, and it
620 * might be necessary to retry.
622 INP_INFO_RLOCK(&divcbinfo);
623 xig.xig_gen = divcbinfo.ipi_gencnt;
624 xig.xig_sogen = so_gencnt;
625 xig.xig_count = divcbinfo.ipi_count;
626 INP_INFO_RUNLOCK(&divcbinfo);
627 error = SYSCTL_OUT(req, &xig, sizeof xig);
629 free(inp_list, M_TEMP);
634 * This is the wrapper function for in_setsockaddr. We just pass down
635 * the pcbinfo for in_setpeeraddr to lock.
638 div_sockaddr(struct socket *so, struct sockaddr **nam)
640 return (in_setsockaddr(so, nam, &divcbinfo));
644 * This is the wrapper function for in_setpeeraddr. We just pass down
645 * the pcbinfo for in_setpeeraddr to lock.
648 div_peeraddr(struct socket *so, struct sockaddr **nam)
650 return (in_setpeeraddr(so, nam, &divcbinfo));
654 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT");
655 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
656 div_pcblist, "S,xinpcb", "List of active divert sockets");
659 struct pr_usrreqs div_usrreqs = {
660 div_abort, pru_accept_notsupp, div_attach, div_bind,
661 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
662 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
663 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
664 div_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel
667 struct protosw div_protosw = {
668 SOCK_RAW, NULL, IPPROTO_DIVERT, PR_ATOMIC|PR_ADDR,
669 div_input, NULL, div_ctlinput, ip_ctloutput,
671 div_init, NULL, NULL, NULL,
676 div_modevent(module_t mod, int type, void *unused)
684 * Protocol will be initialized by pf_proto_register().
685 * We don't have to register ip_protox because we are not
686 * a true IP protocol that goes over the wire.
688 err = pf_proto_register(PF_INET, &div_protosw);
689 ip_divert_ptr = divert_packet;
693 * Module ipdivert can only be unloaded if no sockets are
694 * connected. Maybe this can be changed later to forcefully
695 * disconnect any open sockets.
697 * XXXRW: Note that there is a slight race here, as a new
698 * socket open request could be spinning on the lock and then
699 * we destroy the lock.
701 INP_INFO_WLOCK(&divcbinfo);
702 n = divcbinfo.ipi_count;
705 INP_INFO_WUNLOCK(&divcbinfo);
708 ip_divert_ptr = NULL;
709 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW);
710 INP_INFO_WUNLOCK(&divcbinfo);
711 INP_INFO_LOCK_DESTROY(&divcbinfo);
720 static moduledata_t ipdivertmod = {
726 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
727 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
728 MODULE_VERSION(ipdivert, 1);