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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
34 * $DragonFly: src/sys/netinet/ip_divert.c,v 1.40 2008/10/21 13:51:01 sephe Exp $
41 #include "opt_ipdivert.h"
42 #include "opt_ipsec.h"
45 #error "IPDIVERT requires INET."
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/socket.h>
53 #include <sys/protosw.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <sys/systm.h>
58 #include <sys/thread2.h>
59 #include <sys/in_cksum.h>
62 #include <sys/msgport.h>
65 #include <vm/vm_zone.h>
68 #include <net/route.h>
70 #include <net/netmsg2.h>
73 #include <netinet/in.h>
74 #include <netinet/in_systm.h>
75 #include <netinet/ip.h>
76 #include <netinet/in_pcb.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip_var.h>
79 #include <netinet/ip_divert.h>
86 * Allocate enough space to hold a full IP packet
88 #define DIVSNDQ (65536 + 100)
89 #define DIVRCVQ (65536 + 100)
91 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
93 #define DIV_OUTPUT 0x10000
94 #define DIV_INPUT 0x20000
97 * Divert sockets work in conjunction with ipfw, see the divert(4)
98 * manpage for features.
99 * Internally, packets selected by ipfw in ip_input() or ip_output(),
100 * and never diverted before, are passed to the input queue of the
101 * divert socket with a given 'divert_port' number (as specified in
102 * the matching ipfw rule), and they are tagged with a 16 bit cookie
103 * (representing the rule number of the matching ipfw rule), which
104 * is passed to process reading from the socket.
106 * Packets written to the divert socket are again tagged with a cookie
107 * (usually the same as above) and a destination address.
108 * If the destination address is INADDR_ANY then the packet is
109 * treated as outgoing and sent to ip_output(), otherwise it is
110 * treated as incoming and sent to ip_input().
111 * In both cases, the packet is tagged with the cookie.
113 * On reinjection, processing in ip_input() and ip_output()
114 * will be exactly the same as for the original packet, except that
115 * ipfw processing will start at the rule number after the one
116 * written in the cookie (so, tagging a packet with a cookie of 0
117 * will cause it to be effectively considered as a standard packet).
120 /* Internal variables */
121 static struct inpcbinfo divcbinfo;
123 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
124 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
126 static struct mbuf *ip_divert(struct mbuf *, int, int);
129 * Initialize divert connection block queue.
134 in_pcbinfo_init(&divcbinfo);
136 * XXX We don't use the hash list for divert IP, but it's easier
137 * to allocate a one entry hash list than it is to check all
138 * over the place for hashbase == NULL.
140 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
141 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
142 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
143 &divcbinfo.wildcardhashmask);
144 divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb),
145 maxsockets, ZONE_INTERRUPT, 0);
146 ip_divert_p = ip_divert;
150 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
151 * with that protocol number to enter the system from the outside.
154 div_input(struct mbuf *m, ...)
156 ipstat.ips_noproto++;
161 div_soport(struct socket *so, struct sockaddr *nam,
162 struct mbuf **mptr, int req)
164 struct sockaddr_in *sin;
168 /* Except for send(), everything happens on CPU0 */
170 return cpu0_soport(so, nam, mptr, req);
172 sin = (struct sockaddr_in *)nam;
176 m->m_pkthdr.rcvif = NULL;
177 dir = DIV_IS_OUTPUT(sin) ? IP_MPORT_OUT : IP_MPORT_IN;
183 * Try locating the interface, if we originally had one.
184 * This is done even for outgoing packets, since for a
185 * forwarded packet, there must be an interface attached.
187 * Find receive interface with the given name, stuffed
188 * (if it exists) in the sin_zero[] field.
189 * The name is user supplied data so don't trust its size
190 * or that it is zero terminated.
192 for (i = 0; sin->sin_zero[i] && i < sizeof(sin->sin_zero); i++)
194 if (i > 0 && i < sizeof(sin->sin_zero))
195 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
198 if (dir == IP_MPORT_IN && m->m_pkthdr.rcvif == NULL) {
200 * No luck with the name, check by IP address.
201 * Clear the port and the ifname to make sure
202 * there are no distractions for ifa_ifwithaddr.
204 * Be careful not to trash sin->sin_port; it will
205 * be used later in div_output().
210 bzero(sin->sin_zero, sizeof(sin->sin_zero));
211 sin_port = sin->sin_port; /* save */
213 ifa = ifa_ifwithaddr((struct sockaddr *)sin);
219 sin->sin_port = sin_port; /* restore */
220 m->m_pkthdr.rcvif = ifa->ifa_ifp;
223 return ip_mport(mptr, dir);
227 * Divert a packet by passing it up to the divert socket at port 'port'.
229 * Setup generic address and protocol structures for div_input routine,
230 * then pass them along with mbuf chain.
233 div_packet(struct mbuf *m, int incoming, int port)
235 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
239 struct divert_info *divinfo;
242 /* Locate the divert info */
243 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
244 divinfo = m_tag_data(mtag);
245 divsrc.sin_port = divinfo->skipto;
248 * Record receive interface address, if any.
249 * But only for incoming packets.
251 divsrc.sin_addr.s_addr = 0;
253 struct ifaddr_container *ifac;
255 /* Find IP address for receive interface */
256 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
258 struct ifaddr *ifa = ifac->ifa;
260 if (ifa->ifa_addr == NULL)
262 if (ifa->ifa_addr->sa_family != AF_INET)
265 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
270 * Record the incoming interface name whenever we have one.
272 if (m->m_pkthdr.rcvif) {
274 * Hide the actual interface name in there in the
275 * sin_zero array. XXX This needs to be moved to a
276 * different sockaddr type for divert, e.g.
277 * sockaddr_div with multiple fields like
278 * sockaddr_dl. Presently we have only 7 bytes
279 * but that will do for now as most interfaces
280 * are 4 or less + 2 or less bytes for unit.
281 * There is probably a faster way of doing this,
282 * possibly taking it from the sockaddr_dl on the iface.
283 * This solves the problem of a P2P link and a LAN interface
284 * having the same address, which can result in the wrong
285 * interface being assigned to the packet when fed back
286 * into the divert socket. Theoretically if the daemon saves
287 * and re-uses the sockaddr_in as suggested in the man pages,
288 * this iface name will come along for the ride.
289 * (see div_output for the other half of this.)
291 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero,
292 m->m_pkthdr.rcvif->if_xname);
295 /* Put packet on socket queue, if any */
297 nport = htons((u_int16_t)port);
301 * Following loop to locate the inpcb is MPSAFE since the inpcb
302 * insertion/removal happens on the same CPU (CPU0), however,
303 * saving/testing the socket pointer is not MPSAFE. So we still
304 * need to hold BGL here.
307 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
308 if (inp->inp_flags & INP_PLACEMARKER)
310 if (inp->inp_lport == nport)
311 sa = inp->inp_socket;
314 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m,
315 (struct mbuf *)NULL) == 0)
323 ipstat.ips_noproto++;
324 ipstat.ips_delivered--;
330 div_packet_handler(struct netmsg *nmsg)
332 struct netmsg_packet *nmp;
333 struct lwkt_msg *msg;
335 int port, incoming = 0;
337 nmp = (struct netmsg_packet *)nmsg;
340 msg = &nmsg->nm_lmsg;
341 port = msg->u.ms_result32 & 0xffff;
342 if (msg->u.ms_result32 & DIV_INPUT)
345 div_packet(m, incoming, port);
350 divert_packet(struct mbuf *m, int incoming)
353 struct divert_info *divinfo;
359 if (m->m_len < sizeof(struct ip) &&
360 (m = m_pullup(m, sizeof(struct ip))) == NULL)
363 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
364 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
365 divinfo = m_tag_data(mtag);
367 port = divinfo->port;
368 KASSERT(port != 0, ("%s: port=0", __func__));
372 struct netmsg_packet *nmp;
373 struct lwkt_msg *msg;
375 nmp = &m->m_hdr.mh_netmsg;
376 netmsg_init(&nmp->nm_netmsg, &netisr_apanic_rport, MSGF_MPSAFE,
380 msg = &nmp->nm_netmsg.nm_lmsg;
381 msg->u.ms_result32 = port; /* port is 16bits */
383 msg->u.ms_result32 |= DIV_INPUT;
385 msg->u.ms_result32 |= DIV_OUTPUT;
387 lwkt_sendmsg(cpu_portfn(0), &nmp->nm_netmsg.nm_lmsg);
390 div_packet(m, incoming, port);
394 * Deliver packet back into the IP processing machinery.
396 * If no address specified, or address is 0.0.0.0, send to ip_output();
397 * otherwise, send to ip_input() and mark as having been received on
398 * the interface with that address.
401 div_output(struct socket *so, struct mbuf *m,
402 struct sockaddr_in *sin, struct mbuf *control)
406 struct divert_info *divinfo;
409 m_freem(control); /* XXX */
412 * Prepare the tag for divert info. Note that a packet
413 * with a 0 tag in mh_data is effectively untagged,
414 * so we could optimize that case.
416 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
421 m_tag_prepend(m, mtag);
423 /* Loopback avoidance and state recovery */
424 divinfo = m_tag_data(mtag);
426 divinfo->skipto = sin->sin_port;
430 /* Reinject packet into the system as incoming or outgoing */
431 if (DIV_IS_OUTPUT(sin)) {
432 struct ip *const ip = mtod(m, struct ip *);
434 /* Don't allow packet length sizes that will crash */
435 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
440 /* Convert fields to host order for ip_output() */
441 ip->ip_len = ntohs(ip->ip_len);
442 ip->ip_off = ntohs(ip->ip_off);
444 /* Send packet to output processing */
445 ipstat.ips_rawout++; /* XXX */
446 error = ip_output(m, NULL, NULL,
447 (so->so_options & SO_DONTROUTE) |
448 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
461 div_attach(struct socket *so, int proto, struct pru_attach_info *ai)
469 if ((error = suser_cred(ai->p_ucred, NULL_CRED_OKAY)) != 0)
472 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
475 error = in_pcballoc(so, &divcbinfo);
478 inp = (struct inpcb *)so->so_pcb;
479 inp->inp_ip_p = proto;
480 inp->inp_vflag |= INP_IPV4;
481 inp->inp_flags |= INP_HDRINCL;
483 * The socket is always "connected" because
484 * we always know "where" to send the packet.
486 so->so_state |= SS_ISCONNECTED;
491 div_detach(struct socket *so)
503 div_abort(struct socket *so)
505 soisdisconnected(so);
506 return div_detach(so);
510 div_disconnect(struct socket *so)
512 if (!(so->so_state & SS_ISCONNECTED))
514 return div_abort(so);
518 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
523 * in_pcbbind assumes that nam is a sockaddr_in
524 * and in_pcbbind requires a valid address. Since divert
525 * sockets don't we need to make sure the address is
526 * filled in properly.
527 * XXX -- divert should not be abusing in_pcbind
528 * and should probably have its own family.
530 if (nam->sa_family != AF_INET) {
531 error = EAFNOSUPPORT;
533 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
534 error = in_pcbbind(so->so_pcb, nam, td);
540 div_shutdown(struct socket *so)
547 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
548 struct mbuf *control, struct thread *td)
550 /* Length check already done in ip_mport() */
551 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
554 return div_output(so, m, (struct sockaddr_in *)nam, control);
557 SYSCTL_DECL(_net_inet_divert);
558 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
559 in_pcblist_global, "S,xinpcb", "List of active divert sockets");
561 struct pr_usrreqs div_usrreqs = {
562 .pru_abort = div_abort,
563 .pru_accept = pru_accept_notsupp,
564 .pru_attach = div_attach,
565 .pru_bind = div_bind,
566 .pru_connect = pru_connect_notsupp,
567 .pru_connect2 = pru_connect2_notsupp,
568 .pru_control = in_control,
569 .pru_detach = div_detach,
570 .pru_disconnect = div_disconnect,
571 .pru_listen = pru_listen_notsupp,
572 .pru_peeraddr = in_setpeeraddr,
573 .pru_rcvd = pru_rcvd_notsupp,
574 .pru_rcvoob = pru_rcvoob_notsupp,
575 .pru_send = div_send,
576 .pru_sense = pru_sense_null,
577 .pru_shutdown = div_shutdown,
578 .pru_sockaddr = in_setsockaddr,
579 .pru_sosend = sosend,
580 .pru_soreceive = soreceive,
585 ip_divert_out(struct mbuf *m, int tee)
587 struct mbuf *clone = NULL;
588 struct ip *ip = mtod(m, struct ip *);
590 /* Clone packet if we're doing a 'tee' */
592 clone = m_dup(m, MB_DONTWAIT);
596 * delayed checksums are not currently compatible
597 * with divert sockets.
599 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
601 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
604 /* Restore packet header fields to original values */
605 ip->ip_len = htons(ip->ip_len);
606 ip->ip_off = htons(ip->ip_off);
608 /* Deliver packet to divert input routine */
611 /* If 'tee', continue with original packet */
616 ip_divert_in(struct mbuf *m, int tee)
618 struct mbuf *clone = NULL;
619 struct ip *ip = mtod(m, struct ip *);
622 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
623 const struct divert_info *divinfo;
628 * Only trust divert info in the fragment
631 frag_off = ip->ip_off << 3;
633 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
634 m_tag_delete(m, mtag);
638 * Attempt reassembly; if it succeeds, proceed.
639 * ip_reass() will return a different mbuf.
644 ip = mtod(m, struct ip *);
646 /* Caller need to redispatch the packet, if it is for us */
647 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
650 * Get the header length of the reassembled
653 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
656 * Restore original checksum before diverting
660 ip->ip_len = htons(ip->ip_len);
661 ip->ip_off = htons(ip->ip_off);
663 if (hlen == sizeof(struct ip))
664 ip->ip_sum = in_cksum_hdr(ip);
666 ip->ip_sum = in_cksum(m, hlen);
667 ip->ip_off = ntohs(ip->ip_off);
668 ip->ip_len = ntohs(ip->ip_len);
671 * Only use the saved divert info
673 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
675 /* Wrongly configured ipfw */
676 kprintf("ip_input no divert info\n");
680 divinfo = m_tag_data(mtag);
685 * Divert or tee packet to the divert protocol if
689 /* Clone packet if we're doing a 'tee' */
691 clone = m_dup(m, MB_DONTWAIT);
694 * Restore packet header fields to original
697 ip->ip_len = htons(ip->ip_len);
698 ip->ip_off = htons(ip->ip_off);
700 /* Deliver packet to divert input routine */
703 /* Catch invalid reference */
707 ipstat.ips_delivered++;
709 /* If 'tee', continue with original packet */
712 * Complete processing of the packet.
713 * XXX Better safe than sorry, remove the DIVERT tag.
715 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
716 KKASSERT(mtag != NULL);
717 m_tag_delete(clone, mtag);
723 ip_divert(struct mbuf *m, int tee, int incoming)
728 ret = ip_divert_in(m, tee);
730 ret = ip_divert_out(m, tee);