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. 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
29 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
35 #include "opt_ipdivert.h"
36 #include "opt_ipsec.h"
39 #error "IPDIVERT requires INET."
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/socket.h>
47 #include <sys/protosw.h>
48 #include <sys/socketvar.h>
49 #include <sys/socketvar2.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
54 #include <sys/in_cksum.h>
56 #include <sys/msgport.h>
59 #include <net/route.h>
61 #include <net/netmsg2.h>
62 #include <net/netisr2.h>
63 #include <sys/thread2.h>
64 #include <sys/mplock2.h>
66 #include <netinet/in.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_divert.h>
79 * Allocate enough space to hold a full IP packet
81 #define DIVSNDQ (65536 + 100)
82 #define DIVRCVQ (65536 + 100)
84 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
86 #define DIV_OUTPUT 0x10000
87 #define DIV_INPUT 0x20000
90 * Divert sockets work in conjunction with ipfw, see the divert(4)
91 * manpage for features.
92 * Internally, packets selected by ipfw in ip_input() or ip_output(),
93 * and never diverted before, are passed to the input queue of the
94 * divert socket with a given 'divert_port' number (as specified in
95 * the matching ipfw rule), and they are tagged with a 16 bit cookie
96 * (representing the rule number of the matching ipfw rule), which
97 * is passed to process reading from the socket.
99 * Packets written to the divert socket are again tagged with a cookie
100 * (usually the same as above) and a destination address.
101 * If the destination address is INADDR_ANY then the packet is
102 * treated as outgoing and sent to ip_output(), otherwise it is
103 * treated as incoming and sent to ip_input().
104 * In both cases, the packet is tagged with the cookie.
106 * On reinjection, processing in ip_input() and ip_output()
107 * will be exactly the same as for the original packet, except that
108 * ipfw processing will start at the rule number after the one
109 * written in the cookie (so, tagging a packet with a cookie of 0
110 * will cause it to be effectively considered as a standard packet).
113 /* Internal variables */
114 static struct inpcbinfo divcbinfo;
116 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
117 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
119 static struct mbuf *ip_divert(struct mbuf *, int, int);
121 static struct lwkt_token div_token = LWKT_TOKEN_INITIALIZER(div_token);
124 * Initialize divert connection block queue.
129 in_pcbinfo_init(&divcbinfo);
131 * XXX We don't use the hash list for divert IP, but it's easier
132 * to allocate a one entry hash list than it is to check all
133 * over the place for hashbase == NULL.
135 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
136 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
137 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
138 &divcbinfo.wildcardhashmask);
139 divcbinfo.ipi_size = sizeof(struct inpcb);
140 ip_divert_p = ip_divert;
144 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
145 * with that protocol number to enter the system from the outside.
148 div_input(struct mbuf **mp, int *offp, int proto)
150 struct mbuf *m = *mp;
152 ipstat.ips_noproto++;
154 return(IPPROTO_DONE);
158 * Divert a packet by passing it up to the divert socket at port 'port'.
160 * Setup generic address and protocol structures for div_input routine,
161 * then pass them along with mbuf chain.
164 div_packet(struct mbuf *m, int incoming, int port)
166 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
170 struct divert_info *divinfo;
173 /* Locate the divert info */
174 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
175 divinfo = m_tag_data(mtag);
176 divsrc.sin_port = divinfo->skipto;
179 * Record receive interface address, if any.
180 * But only for incoming packets.
182 divsrc.sin_addr.s_addr = 0;
184 struct ifaddr_container *ifac;
186 /* Find IP address for receive interface */
187 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
189 struct ifaddr *ifa = ifac->ifa;
191 if (ifa->ifa_addr == NULL)
193 if (ifa->ifa_addr->sa_family != AF_INET)
196 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
201 * Record the incoming interface name whenever we have one.
203 if (m->m_pkthdr.rcvif) {
205 * Hide the actual interface name in there in the
206 * sin_zero array. XXX This needs to be moved to a
207 * different sockaddr type for divert, e.g.
208 * sockaddr_div with multiple fields like
209 * sockaddr_dl. Presently we have only 7 bytes
210 * but that will do for now as most interfaces
211 * are 4 or less + 2 or less bytes for unit.
212 * There is probably a faster way of doing this,
213 * possibly taking it from the sockaddr_dl on the iface.
214 * This solves the problem of a P2P link and a LAN interface
215 * having the same address, which can result in the wrong
216 * interface being assigned to the packet when fed back
217 * into the divert socket. Theoretically if the daemon saves
218 * and re-uses the sockaddr_in as suggested in the man pages,
219 * this iface name will come along for the ride.
220 * (see div_output for the other half of this.)
222 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s",
223 m->m_pkthdr.rcvif->if_xname);
226 /* Put packet on socket queue, if any */
228 nport = htons((u_int16_t)port);
232 * Following loop to locate the inpcb is MPSAFE since the inpcb
233 * insertion/removal happens on the same CPU (CPU0), however,
234 * saving/testing the socket pointer is not MPSAFE. So we still
235 * need to hold BGL here.
237 lwkt_gettoken(&div_token);
238 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
239 if (inp->inp_flags & INP_PLACEMARKER)
241 if (inp->inp_lport == nport)
242 sa = inp->inp_socket;
245 lwkt_gettoken(&sa->so_rcv.ssb_token);
246 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
250 lwkt_reltoken(&sa->so_rcv.ssb_token);
253 ipstat.ips_noproto++;
254 ipstat.ips_delivered--;
256 lwkt_reltoken(&div_token);
260 div_packet_handler(netmsg_t msg)
263 int port, incoming = 0;
265 m = msg->packet.nm_packet;
267 port = msg->lmsg.u.ms_result32 & 0xffff;
268 if (msg->lmsg.u.ms_result32 & DIV_INPUT)
270 div_packet(m, incoming, port);
271 /* no reply, msg embedded in mbuf */
275 divert_packet(struct mbuf *m, int incoming)
278 struct divert_info *divinfo;
284 if (m->m_len < sizeof(struct ip) &&
285 (m = m_pullup(m, sizeof(struct ip))) == NULL)
288 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
289 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
290 divinfo = m_tag_data(mtag);
292 port = divinfo->port;
293 KASSERT(port != 0, ("%s: port=0", __func__));
296 struct netmsg_packet *nmp;
298 nmp = &m->m_hdr.mh_netmsg;
299 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
300 0, div_packet_handler);
303 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
305 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
307 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
309 lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg);
311 div_packet(m, incoming, port);
316 * Deliver packet back into the IP processing machinery.
318 * If no address specified, or address is 0.0.0.0, send to ip_output();
319 * otherwise, send to ip_input() and mark as having been received on
320 * the interface with that address.
323 div_output(struct socket *so, struct mbuf *m,
324 struct sockaddr_in *sin, struct mbuf *control)
328 struct divert_info *divinfo;
331 m_freem(control); /* XXX */
334 * Prepare the tag for divert info. Note that a packet
335 * with a 0 tag in mh_data is effectively untagged,
336 * so we could optimize that case.
338 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), MB_DONTWAIT);
343 m_tag_prepend(m, mtag);
345 /* Loopback avoidance and state recovery */
346 divinfo = m_tag_data(mtag);
348 divinfo->skipto = sin->sin_port;
352 /* Reinject packet into the system as incoming or outgoing */
353 if (DIV_IS_OUTPUT(sin)) {
354 struct ip *const ip = mtod(m, struct ip *);
356 /* Don't allow packet length sizes that will crash */
357 if ((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 */
368 error = ip_output(m, NULL, NULL,
369 (so->so_options & SO_DONTROUTE) |
370 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
383 div_attach(netmsg_t msg)
385 struct socket *so = msg->attach.base.nm_so;
386 int proto = msg->attach.nm_proto;
387 struct pru_attach_info *ai = msg->attach.nm_ai;
394 error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
398 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
401 lwkt_gettoken(&div_token);
402 sosetport(so, netisr_cpuport(0));
403 error = in_pcballoc(so, &divcbinfo);
405 lwkt_reltoken(&div_token);
408 inp = (struct inpcb *)so->so_pcb;
409 inp->inp_ip_p = proto;
410 inp->inp_vflag |= INP_IPV4;
411 inp->inp_flags |= INP_HDRINCL;
413 * The socket is always "connected" because
414 * we always know "where" to send the packet.
416 sosetstate(so, SS_ISCONNECTED);
417 lwkt_reltoken(&div_token);
420 lwkt_replymsg(&msg->attach.base.lmsg, error);
424 div_detach(netmsg_t msg)
426 struct socket *so = msg->detach.base.nm_so;
433 lwkt_replymsg(&msg->detach.base.lmsg, 0);
437 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
438 * will sofree() it when we return.
441 div_abort(netmsg_t msg)
443 struct socket *so = msg->abort.base.nm_so;
445 soisdisconnected(so);
447 /* msg invalid now */
451 div_disconnect(netmsg_t msg)
453 struct socket *so = msg->disconnect.base.nm_so;
456 if (so->so_state & SS_ISCONNECTED) {
459 /* msg invalid now */
464 lwkt_replymsg(&msg->disconnect.base.lmsg, error);
468 div_bind(netmsg_t msg)
470 struct socket *so = msg->bind.base.nm_so;
471 struct sockaddr *nam = msg->bind.nm_nam;
475 * in_pcbbind assumes that nam is a sockaddr_in
476 * and in_pcbbind requires a valid address. Since divert
477 * sockets don't we need to make sure the address is
478 * filled in properly.
479 * XXX -- divert should not be abusing in_pcbind
480 * and should probably have its own family.
482 if (nam->sa_family != AF_INET) {
483 error = EAFNOSUPPORT;
485 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
486 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
488 lwkt_replymsg(&msg->bind.base.lmsg, error);
492 div_shutdown(netmsg_t msg)
494 struct socket *so = msg->shutdown.base.nm_so;
498 lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
502 div_send(netmsg_t msg)
504 struct socket *so = msg->send.base.nm_so;
505 struct mbuf *m = msg->send.nm_m;
506 struct sockaddr *nam = msg->send.nm_addr;
507 struct mbuf *control = msg->send.nm_control;
510 /* Length check already done in ip_hashfn() */
511 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
514 error = div_output(so, m, (struct sockaddr_in *)nam, control);
515 lwkt_replymsg(&msg->send.base.lmsg, error);
518 SYSCTL_DECL(_net_inet_divert);
519 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 0,
520 in_pcblist_global, "S,xinpcb", "List of active divert sockets");
522 struct pr_usrreqs div_usrreqs = {
523 .pru_abort = div_abort,
524 .pru_accept = pr_generic_notsupp,
525 .pru_attach = div_attach,
526 .pru_bind = div_bind,
527 .pru_connect = pr_generic_notsupp,
528 .pru_connect2 = pr_generic_notsupp,
529 .pru_control = in_control_dispatch,
530 .pru_detach = div_detach,
531 .pru_disconnect = div_disconnect,
532 .pru_listen = pr_generic_notsupp,
533 .pru_peeraddr = in_setpeeraddr_dispatch,
534 .pru_rcvd = pr_generic_notsupp,
535 .pru_rcvoob = pr_generic_notsupp,
536 .pru_send = div_send,
537 .pru_sense = pru_sense_null,
538 .pru_shutdown = div_shutdown,
539 .pru_sockaddr = in_setsockaddr_dispatch,
540 .pru_sosend = sosend,
541 .pru_soreceive = soreceive
545 ip_divert_out(struct mbuf *m, int tee)
547 struct mbuf *clone = NULL;
548 struct ip *ip = mtod(m, struct ip *);
550 /* Clone packet if we're doing a 'tee' */
552 clone = m_dup(m, MB_DONTWAIT);
556 * delayed checksums are not currently compatible
557 * with divert sockets.
559 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
561 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
564 /* Restore packet header fields to original values */
565 ip->ip_len = htons(ip->ip_len);
566 ip->ip_off = htons(ip->ip_off);
568 /* Deliver packet to divert input routine */
571 /* If 'tee', continue with original packet */
576 ip_divert_in(struct mbuf *m, int tee)
578 struct mbuf *clone = NULL;
579 struct ip *ip = mtod(m, struct ip *);
582 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
583 const struct divert_info *divinfo;
588 * Only trust divert info in the fragment
591 frag_off = ip->ip_off << 3;
593 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
594 m_tag_delete(m, mtag);
598 * Attempt reassembly; if it succeeds, proceed.
599 * ip_reass() will return a different mbuf.
604 ip = mtod(m, struct ip *);
606 /* Caller need to redispatch the packet, if it is for us */
607 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
610 * Get the header length of the reassembled
613 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
616 * Restore original checksum before diverting
620 ip->ip_len = htons(ip->ip_len);
621 ip->ip_off = htons(ip->ip_off);
623 if (hlen == sizeof(struct ip))
624 ip->ip_sum = in_cksum_hdr(ip);
626 ip->ip_sum = in_cksum(m, hlen);
627 ip->ip_off = ntohs(ip->ip_off);
628 ip->ip_len = ntohs(ip->ip_len);
631 * Only use the saved divert info
633 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
635 /* Wrongly configured ipfw */
636 kprintf("ip_input no divert info\n");
640 divinfo = m_tag_data(mtag);
645 * Divert or tee packet to the divert protocol if
649 /* Clone packet if we're doing a 'tee' */
651 clone = m_dup(m, MB_DONTWAIT);
654 * Restore packet header fields to original
657 ip->ip_len = htons(ip->ip_len);
658 ip->ip_off = htons(ip->ip_off);
660 /* Deliver packet to divert input routine */
663 /* Catch invalid reference */
667 ipstat.ips_delivered++;
669 /* If 'tee', continue with original packet */
672 * Complete processing of the packet.
673 * XXX Better safe than sorry, remove the DIVERT tag.
675 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
676 KKASSERT(mtag != NULL);
677 m_tag_delete(clone, mtag);
683 ip_divert(struct mbuf *m, int tee, int incoming)
688 ret = ip_divert_in(m, tee);
690 ret = ip_divert_out(m, tee);