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
34 #include "opt_ipdivert.h"
35 #include "opt_ipsec.h"
39 #error "IPDIVERT requires INET."
42 #include <sys/param.h>
43 #include <sys/kernel.h>
45 #include <sys/malloc.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>
75 * Allocate enough space to hold a full IP packet
77 #define DIVSNDQ (65536 + 100)
78 #define DIVRCVQ (65536 + 100)
81 * Divert sockets work in conjunction with ipfw, see the divert(4)
82 * manpage for features.
83 * Internally, packets selected by ipfw in ip_input() or ip_output(),
84 * and never diverted before, are passed to the input queue of the
85 * divert socket with a given 'divert_port' number (as specified in
86 * the matching ipfw rule), and they are tagged with a 16 bit cookie
87 * (representing the rule number of the matching ipfw rule), which
88 * is passed to process reading from the socket.
90 * Packets written to the divert socket are again tagged with a cookie
91 * (usually the same as above) and a destination address.
92 * If the destination address is INADDR_ANY then the packet is
93 * treated as outgoing and sent to ip_output(), otherwise it is
94 * treated as incoming and sent to ip_input().
95 * In both cases, the packet is tagged with the cookie.
97 * On reinjection, processing in ip_input() and ip_output()
98 * will be exactly the same as for the original packet, except that
99 * ipfw processing will start at the rule number after the one
100 * written in the cookie (so, tagging a packet with a cookie of 0
101 * will cause it to be effectively considered as a standard packet).
104 /* Internal variables */
105 static struct inpcbhead divcb;
106 static struct inpcbinfo divcbinfo;
108 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
109 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
112 * Initialize divert connection block queue.
117 INP_INFO_LOCK_INIT(&divcbinfo, "div");
119 divcbinfo.listhead = &divcb;
121 * XXX We don't use the hash list for divert IP, but it's easier
122 * to allocate a one entry hash list than it is to check all
123 * over the place for hashbase == NULL.
125 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
126 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
127 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb),
128 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
129 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets);
133 * IPPROTO_DIVERT is not in the real IP protocol number space; this
134 * function should never be called. Just in case, drop any packets.
137 div_input(struct mbuf *m, int off)
139 ipstat.ips_noproto++;
144 * Divert a packet by passing it up to the divert socket at port 'port'.
146 * Setup generic address and protocol structures for div_input routine,
147 * then pass them along with mbuf chain.
150 divert_packet(struct mbuf *m, int incoming)
156 struct sockaddr_in divsrc;
159 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
161 printf("%s: no divert tag\n", __func__);
166 if (m->m_len < sizeof(struct ip) &&
167 (m = m_pullup(m, sizeof(struct ip))) == 0)
169 ip = mtod(m, struct ip *);
171 /* Delayed checksums are currently not compatible with divert. */
172 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
173 ip->ip_len = ntohs(ip->ip_len);
175 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
176 ip->ip_len = htons(ip->ip_len);
180 * Record receive interface address, if any.
181 * But only for incoming packets.
183 bzero(&divsrc, sizeof(divsrc));
184 divsrc.sin_len = sizeof(divsrc);
185 divsrc.sin_family = AF_INET;
186 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */
193 /* Find IP address for receive interface */
194 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
195 if (ifa->ifa_addr == NULL)
197 if (ifa->ifa_addr->sa_family != AF_INET)
200 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
205 * Record the incoming interface name whenever we have one.
207 if (m->m_pkthdr.rcvif) {
209 * Hide the actual interface name in there in the
210 * sin_zero array. XXX This needs to be moved to a
211 * different sockaddr type for divert, e.g.
212 * sockaddr_div with multiple fields like
213 * sockaddr_dl. Presently we have only 7 bytes
214 * but that will do for now as most interfaces
215 * are 4 or less + 2 or less bytes for unit.
216 * There is probably a faster way of doing this,
217 * possibly taking it from the sockaddr_dl on the iface.
218 * This solves the problem of a P2P link and a LAN interface
219 * having the same address, which can result in the wrong
220 * interface being assigned to the packet when fed back
221 * into the divert socket. Theoretically if the daemon saves
222 * and re-uses the sockaddr_in as suggested in the man pages,
223 * this iface name will come along for the ride.
224 * (see div_output for the other half of this.)
226 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname,
227 sizeof(divsrc.sin_zero));
230 /* Put packet on socket queue, if any */
232 nport = htons((u_int16_t)divert_info(mtag));
233 INP_INFO_RLOCK(&divcbinfo);
234 LIST_FOREACH(inp, &divcb, inp_list) {
236 /* XXX why does only one socket match? */
237 if (inp->inp_lport == nport) {
238 sa = inp->inp_socket;
239 SOCKBUF_LOCK(&sa->so_rcv);
240 if (sbappendaddr_locked(&sa->so_rcv,
241 (struct sockaddr *)&divsrc, m,
242 (struct mbuf *)0) == 0) {
243 SOCKBUF_UNLOCK(&sa->so_rcv);
244 sa = NULL; /* force mbuf reclaim below */
246 sorwakeup_locked(sa);
252 INP_INFO_RUNLOCK(&divcbinfo);
255 ipstat.ips_noproto++;
256 ipstat.ips_delivered--;
261 * Deliver packet back into the IP processing machinery.
263 * If no address specified, or address is 0.0.0.0, send to ip_output();
264 * otherwise, send to ip_input() and mark as having been received on
265 * the interface with that address.
268 div_output(struct socket *so, struct mbuf *m,
269 struct sockaddr_in *sin, struct mbuf *control)
273 KASSERT(m->m_pkthdr.rcvif == NULL, ("rcvif not null"));
276 m_freem(control); /* XXX */
278 /* Loopback avoidance and state recovery */
281 struct divert_tag *dt;
284 mtag = m_tag_get(PACKET_TAG_DIVERT,
285 sizeof(struct divert_tag), M_NOWAIT);
290 dt = (struct divert_tag *)(mtag+1);
292 dt->cookie = sin->sin_port;
293 m_tag_prepend(m, mtag);
296 * Find receive interface with the given name, stuffed
297 * (if it exists) in the sin_zero[] field.
298 * The name is user supplied data so don't trust its size
299 * or that it is zero terminated.
301 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++)
303 if ( i > 0 && i < sizeof(sin->sin_zero))
304 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
307 /* Reinject packet into the system as incoming or outgoing */
308 if (!sin || sin->sin_addr.s_addr == 0) {
309 struct ip *const ip = mtod(m, struct ip *);
312 INP_INFO_WLOCK(&divcbinfo);
316 * Don't allow both user specified and setsockopt options,
317 * and don't allow packet length sizes that will crash
319 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
320 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
324 /* Convert fields to host order for ip_output() */
325 ip->ip_len = ntohs(ip->ip_len);
326 ip->ip_off = ntohs(ip->ip_off);
328 /* Send packet to output processing */
329 ipstat.ips_rawout++; /* XXX */
332 mac_create_mbuf_from_inpcb(inp, m);
335 inp->inp_options, NULL,
336 ((so->so_options & SO_DONTROUTE) ?
338 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
339 inp->inp_moptions, NULL);
342 INP_INFO_WUNLOCK(&divcbinfo);
344 if (m->m_pkthdr.rcvif == NULL) {
346 * No luck with the name, check by IP address.
347 * Clear the port and the ifname to make sure
348 * there are no distractions for ifa_ifwithaddr.
352 bzero(sin->sin_zero, sizeof(sin->sin_zero));
354 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
356 error = EADDRNOTAVAIL;
359 m->m_pkthdr.rcvif = ifa->ifa_ifp;
363 mac_create_mbuf_from_socket(so, m);
366 /* Send packet to input processing */
378 div_attach(struct socket *so, int proto, struct thread *td)
383 INP_INFO_WLOCK(&divcbinfo);
386 INP_INFO_WUNLOCK(&divcbinfo);
389 if (td && (error = suser(td)) != 0) {
390 INP_INFO_WUNLOCK(&divcbinfo);
393 error = soreserve(so, div_sendspace, div_recvspace);
395 INP_INFO_WUNLOCK(&divcbinfo);
398 error = in_pcballoc(so, &divcbinfo, "divinp");
400 INP_INFO_WUNLOCK(&divcbinfo);
403 inp = (struct inpcb *)so->so_pcb;
405 INP_INFO_WUNLOCK(&divcbinfo);
406 inp->inp_ip_p = proto;
407 inp->inp_vflag |= INP_IPV4;
408 inp->inp_flags |= INP_HDRINCL;
409 /* The socket is always "connected" because
410 we always know "where" to send the packet */
413 so->so_state |= SS_ISCONNECTED;
419 div_detach(struct socket *so)
423 INP_INFO_WLOCK(&divcbinfo);
426 INP_INFO_WUNLOCK(&divcbinfo);
431 INP_INFO_WUNLOCK(&divcbinfo);
436 div_abort(struct socket *so)
440 INP_INFO_WLOCK(&divcbinfo);
443 INP_INFO_WUNLOCK(&divcbinfo);
444 return EINVAL; /* ??? possible? panic instead? */
447 soisdisconnected(so);
449 INP_INFO_WUNLOCK(&divcbinfo);
454 div_disconnect(struct socket *so)
456 if ((so->so_state & SS_ISCONNECTED) == 0)
458 return div_abort(so);
462 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
467 INP_INFO_WLOCK(&divcbinfo);
470 INP_INFO_WUNLOCK(&divcbinfo);
473 /* in_pcbbind assumes that nam is a sockaddr_in
474 * and in_pcbbind requires a valid address. Since divert
475 * sockets don't we need to make sure the address is
476 * filled in properly.
477 * XXX -- divert should not be abusing in_pcbind
478 * and should probably have its own family.
480 if (nam->sa_family != AF_INET)
481 error = EAFNOSUPPORT;
483 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
485 error = in_pcbbind(inp, nam, td->td_ucred);
488 INP_INFO_WUNLOCK(&divcbinfo);
493 div_shutdown(struct socket *so)
497 INP_INFO_RLOCK(&divcbinfo);
500 INP_INFO_RUNLOCK(&divcbinfo);
504 INP_INFO_RUNLOCK(&divcbinfo);
511 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
512 struct mbuf *control, struct thread *td)
514 /* Packet must have a header (but that's about it) */
515 if (m->m_len < sizeof (struct ip) &&
516 (m = m_pullup(m, sizeof (struct ip))) == 0) {
517 ipstat.ips_toosmall++;
523 return div_output(so, m, (struct sockaddr_in *)nam, control);
527 div_ctlinput(int cmd, struct sockaddr *sa, void *vip)
529 struct in_addr faddr;
531 faddr = ((struct sockaddr_in *)sa)->sin_addr;
532 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
534 if (PRC_IS_REDIRECT(cmd))
539 div_pcblist(SYSCTL_HANDLER_ARGS)
542 struct inpcb *inp, **inp_list;
547 * The process of preparing the TCB list is too time-consuming and
548 * resource-intensive to repeat twice on every request.
550 if (req->oldptr == 0) {
551 n = divcbinfo.ipi_count;
552 req->oldidx = 2 * (sizeof xig)
553 + (n + n/8) * sizeof(struct xinpcb);
557 if (req->newptr != 0)
561 * OK, now we're committed to doing something.
563 INP_INFO_RLOCK(&divcbinfo);
564 gencnt = divcbinfo.ipi_gencnt;
565 n = divcbinfo.ipi_count;
566 INP_INFO_RUNLOCK(&divcbinfo);
568 error = sysctl_wire_old_buffer(req,
569 2 * sizeof(xig) + n*sizeof(struct xinpcb));
573 xig.xig_len = sizeof xig;
575 xig.xig_gen = gencnt;
576 xig.xig_sogen = so_gencnt;
577 error = SYSCTL_OUT(req, &xig, sizeof xig);
581 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
585 INP_INFO_RLOCK(&divcbinfo);
586 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
587 inp = LIST_NEXT(inp, inp_list)) {
589 if (inp->inp_gencnt <= gencnt &&
590 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0)
594 INP_INFO_RUNLOCK(&divcbinfo);
598 for (i = 0; i < n; i++) {
600 if (inp->inp_gencnt <= gencnt) {
602 xi.xi_len = sizeof xi;
603 /* XXX should avoid extra copy */
604 bcopy(inp, &xi.xi_inp, sizeof *inp);
606 sotoxsocket(inp->inp_socket, &xi.xi_socket);
607 error = SYSCTL_OUT(req, &xi, sizeof xi);
612 * Give the user an updated idea of our state.
613 * If the generation differs from what we told
614 * her before, she knows that something happened
615 * while we were processing this request, and it
616 * might be necessary to retry.
618 INP_INFO_RLOCK(&divcbinfo);
619 xig.xig_gen = divcbinfo.ipi_gencnt;
620 xig.xig_sogen = so_gencnt;
621 xig.xig_count = divcbinfo.ipi_count;
622 INP_INFO_RUNLOCK(&divcbinfo);
623 error = SYSCTL_OUT(req, &xig, sizeof xig);
625 free(inp_list, M_TEMP);
630 * This is the wrapper function for in_setsockaddr. We just pass down
631 * the pcbinfo for in_setpeeraddr to lock.
634 div_sockaddr(struct socket *so, struct sockaddr **nam)
636 return (in_setsockaddr(so, nam, &divcbinfo));
640 * This is the wrapper function for in_setpeeraddr. We just pass down
641 * the pcbinfo for in_setpeeraddr to lock.
644 div_peeraddr(struct socket *so, struct sockaddr **nam)
646 return (in_setpeeraddr(so, nam, &divcbinfo));
650 SYSCTL_DECL(_net_inet_divert);
651 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
652 div_pcblist, "S,xinpcb", "List of active divert sockets");
654 struct pr_usrreqs div_usrreqs = {
655 div_abort, pru_accept_notsupp, div_attach, div_bind,
656 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
657 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp,
658 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
659 div_sockaddr, sosend, soreceive, sopoll, in_pcbsosetlabel