Bump __DragonFly_version for recent updates.
[dragonfly.git] / sys / netinet / ip_divert.c
blob410f8348e9c609607e33b763b00c95a80f35bdbf
1 /*
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
7 * are met:
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
27 * SUCH DAMAGE.
29 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
32 #define _IP_VHL
34 #include "opt_inet.h"
35 #include "opt_ipdivert.h"
37 #ifndef INET
38 #error "IPDIVERT requires INET."
39 #endif
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/socket.h>
46 #include <sys/protosw.h>
47 #include <sys/socketvar.h>
48 #include <sys/socketvar2.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/proc.h>
52 #include <sys/priv.h>
53 #include <sys/in_cksum.h>
54 #include <sys/lock.h>
55 #include <sys/msgport.h>
57 #include <net/if.h>
58 #include <net/route.h>
60 #include <net/netmsg2.h>
61 #include <net/netisr2.h>
62 #include <sys/mplock2.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_divert.h>
73 * Divert sockets
77 * Allocate enough space to hold a full IP packet
79 #define DIVSNDQ (65536 + 100)
80 #define DIVRCVQ (65536 + 100)
82 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0)
84 #define DIV_OUTPUT 0x10000
85 #define DIV_INPUT 0x20000
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 inpcbinfo divcbinfo;
113 static struct inpcbportinfo divcbportinfo;
115 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
116 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
118 static struct mbuf *ip_divert(struct mbuf *, int, int);
121 * Initialize divert connection block queue.
123 void
124 div_init(void)
126 in_pcbinfo_init(&divcbinfo, 0, FALSE);
127 in_pcbportinfo_init(&divcbportinfo, 1, 0);
129 * XXX We don't use the hash list for divert IP, but it's easier
130 * to allocate a one entry hash list than it is to check all
131 * over the place for hashbase == NULL.
133 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
134 in_pcbportinfo_set(&divcbinfo, &divcbportinfo, 1);
135 divcbinfo.wildcardhashbase = hashinit(1, M_PCB,
136 &divcbinfo.wildcardhashmask);
137 divcbinfo.ipi_size = sizeof(struct inpcb);
138 ip_divert_p = ip_divert;
142 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
143 * with that protocol number to enter the system from the outside.
146 div_input(struct mbuf **mp, int *offp, int proto)
148 struct mbuf *m = *mp;
150 ipstat.ips_noproto++;
151 m_freem(m);
152 return(IPPROTO_DONE);
156 * Divert a packet by passing it up to the divert socket at port 'port'.
158 * Setup generic address and protocol structures for div_input routine,
159 * then pass them along with mbuf chain.
161 static void
162 div_packet(struct mbuf *m, int incoming, int port)
164 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET };
165 struct inpcb *inp;
166 struct socket *sa;
167 struct m_tag *mtag;
168 struct divert_info *divinfo;
169 u_int16_t nport;
171 ASSERT_NETISR0;
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;
183 if (incoming) {
184 struct ifaddr_container *ifac;
186 /* Find IP address for receive interface */
187 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
188 ifa_link) {
189 struct ifaddr *ifa = ifac->ifa;
191 if (ifa->ifa_addr == NULL)
192 continue;
193 if (ifa->ifa_addr->sa_family != AF_INET)
194 continue;
195 divsrc.sin_addr =
196 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
197 break;
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 */
227 sa = NULL;
228 nport = htons((u_int16_t)port);
231 * Following loop to locate the inpcb is MPSAFE since the inpcb
232 * insertion/removal happens on the same CPU (CPU0).
234 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) {
235 if (inp->inp_flags & INP_PLACEMARKER)
236 continue;
237 if (inp->inp_lport == nport)
238 sa = inp->inp_socket;
240 if (sa) {
241 lwkt_gettoken(&sa->so_rcv.ssb_token);
242 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0)
243 m_freem(m);
244 else
245 sorwakeup(sa);
246 lwkt_reltoken(&sa->so_rcv.ssb_token);
247 } else {
248 m_freem(m);
249 ipstat.ips_noproto++;
250 ipstat.ips_delivered--;
254 static void
255 div_packet_handler(netmsg_t msg)
257 struct mbuf *m;
258 int port, incoming = 0;
260 m = msg->packet.nm_packet;
262 port = msg->lmsg.u.ms_result32 & 0xffff;
263 if (msg->lmsg.u.ms_result32 & DIV_INPUT)
264 incoming = 1;
265 div_packet(m, incoming, port);
266 /* no reply, msg embedded in mbuf */
269 static void
270 divert_packet(struct mbuf *m, int incoming)
272 struct m_tag *mtag;
273 struct divert_info *divinfo;
274 int port;
276 M_ASSERTPKTHDR(m);
278 /* Assure header */
279 if (m->m_len < sizeof(struct ip) &&
280 (m = m_pullup(m, sizeof(struct ip))) == NULL)
281 return;
283 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
284 KASSERT(mtag != NULL, ("%s no divert tag!", __func__));
285 divinfo = m_tag_data(mtag);
287 port = divinfo->port;
288 KASSERT(port != 0, ("%s: port=0", __func__));
290 if (mycpuid != 0) {
291 struct netmsg_packet *nmp;
293 nmp = &m->m_hdr.mh_netmsg;
294 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport,
295 0, div_packet_handler);
296 nmp->nm_packet = m;
298 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */
299 if (incoming)
300 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT;
301 else
302 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT;
304 lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg);
305 } else {
306 div_packet(m, incoming, port);
311 * Deliver packet back into the IP processing machinery.
313 * If no address specified, or address is 0.0.0.0, send to ip_output();
314 * otherwise, send to ip_input() and mark as having been received on
315 * the interface with that address.
317 static int
318 div_output(struct socket *so, struct mbuf *m,
319 struct sockaddr_in *sin, struct mbuf *control)
321 int error = 0;
322 struct m_tag *mtag;
323 struct divert_info *divinfo;
325 ASSERT_NETISR0;
327 if (control)
328 m_freem(control); /* XXX */
331 * Prepare the tag for divert info. Note that a packet
332 * with a 0 tag in mh_data is effectively untagged,
333 * so we could optimize that case.
335 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), M_NOWAIT);
336 if (mtag == NULL) {
337 error = ENOBUFS;
338 goto cantsend;
340 m_tag_prepend(m, mtag);
342 /* Loopback avoidance and state recovery */
343 divinfo = m_tag_data(mtag);
344 if (sin)
345 divinfo->skipto = sin->sin_port;
346 else
347 divinfo->skipto = 0;
349 /* Reinject packet into the system as incoming or outgoing */
350 if (DIV_IS_OUTPUT(sin)) {
351 struct ip *const ip = mtod(m, struct ip *);
353 /* Don't allow packet length sizes that will crash */
354 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) {
355 error = EINVAL;
356 goto cantsend;
359 /* Convert fields to host order for ip_output() */
360 ip->ip_len = ntohs(ip->ip_len);
361 ip->ip_off = ntohs(ip->ip_off);
363 /* Send packet to output processing */
364 ipstat.ips_rawout++; /* XXX */
365 error = ip_output(m, NULL, NULL,
366 (so->so_options & SO_DONTROUTE) |
367 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
368 NULL, NULL);
369 } else {
370 ip_input(m);
372 return error;
374 cantsend:
375 m_freem(m);
376 return error;
379 static void
380 div_attach(netmsg_t msg)
382 struct socket *so = msg->attach.base.nm_so;
383 int proto = msg->attach.nm_proto;
384 struct pru_attach_info *ai = msg->attach.nm_ai;
385 struct inpcb *inp;
386 int error;
388 ASSERT_NETISR0;
390 inp = so->so_pcb;
391 if (inp)
392 panic("div_attach");
393 error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY);
394 if (error)
395 goto out;
397 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
398 if (error)
399 goto out;
400 error = in_pcballoc(so, &divcbinfo);
401 if (error)
402 goto out;
403 inp = (struct inpcb *)so->so_pcb;
404 inp->inp_ip_p = proto;
405 inp->inp_flags |= INP_HDRINCL;
407 * The socket is always "connected" because
408 * we always know "where" to send the packet.
410 sosetstate(so, SS_ISCONNECTED);
411 error = 0;
412 out:
413 lwkt_replymsg(&msg->attach.base.lmsg, error);
416 static void
417 div_detach(netmsg_t msg)
419 struct socket *so = msg->detach.base.nm_so;
420 struct inpcb *inp;
422 ASSERT_NETISR0;
424 inp = so->so_pcb;
425 if (inp == NULL)
426 panic("div_detach");
427 in_pcbdetach(inp);
428 lwkt_replymsg(&msg->detach.base.lmsg, 0);
431 static void
432 div_abort(netmsg_t msg)
435 * Divert socket does not support listen(2),
436 * so this should never be called.
438 panic("div_abort is called");
441 static void
442 div_disconnect(netmsg_t msg)
444 struct socket *so = msg->disconnect.base.nm_so;
445 int error;
447 ASSERT_NETISR0;
449 if (so->so_state & SS_ISCONNECTED) {
450 soisdisconnected(so);
451 error = 0;
452 } else {
453 error = ENOTCONN;
455 lwkt_replymsg(&msg->disconnect.base.lmsg, error);
458 static void
459 div_bind(netmsg_t msg)
461 struct socket *so = msg->bind.base.nm_so;
462 struct sockaddr *nam = msg->bind.nm_nam;
463 int error;
465 ASSERT_NETISR0;
468 * in_pcbbind assumes that nam is a sockaddr_in
469 * and in_pcbbind requires a valid address. Since divert
470 * sockets don't we need to make sure the address is
471 * filled in properly.
472 * XXX -- divert should not be abusing in_pcbind
473 * and should probably have its own family.
475 if (nam->sa_family != AF_INET) {
476 error = EAFNOSUPPORT;
477 } else {
478 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
479 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td);
481 lwkt_replymsg(&msg->bind.base.lmsg, error);
484 static void
485 div_shutdown(netmsg_t msg)
487 struct socket *so = msg->shutdown.base.nm_so;
489 ASSERT_NETISR0;
491 socantsendmore(so);
493 lwkt_replymsg(&msg->shutdown.base.lmsg, 0);
496 static void
497 div_send(netmsg_t msg)
499 struct socket *so = msg->send.base.nm_so;
500 struct mbuf *m = msg->send.nm_m;
501 struct sockaddr *nam = msg->send.nm_addr;
502 struct mbuf *control = msg->send.nm_control;
503 int error;
505 /* Length check already done in ip_hashfn() */
506 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
508 /* Send packet */
509 error = div_output(so, m, (struct sockaddr_in *)nam, control);
510 lwkt_replymsg(&msg->send.base.lmsg, error);
513 SYSCTL_DECL(_net_inet_divert);
514 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 1,
515 in_pcblist_range, "S,xinpcb", "List of active divert sockets");
517 struct pr_usrreqs div_usrreqs = {
518 .pru_abort = div_abort,
519 .pru_accept = pr_generic_notsupp,
520 .pru_attach = div_attach,
521 .pru_bind = div_bind,
522 .pru_connect = pr_generic_notsupp,
523 .pru_connect2 = pr_generic_notsupp,
524 .pru_control = in_control_dispatch,
525 .pru_detach = div_detach,
526 .pru_disconnect = div_disconnect,
527 .pru_listen = pr_generic_notsupp,
528 .pru_peeraddr = in_setpeeraddr_dispatch,
529 .pru_rcvd = pr_generic_notsupp,
530 .pru_rcvoob = pr_generic_notsupp,
531 .pru_send = div_send,
532 .pru_sense = pru_sense_null,
533 .pru_shutdown = div_shutdown,
534 .pru_sockaddr = in_setsockaddr_dispatch,
535 .pru_sosend = sosend,
536 .pru_soreceive = soreceive
539 static struct mbuf *
540 ip_divert_out(struct mbuf *m, int tee)
542 struct mbuf *clone = NULL;
543 struct ip *ip = mtod(m, struct ip *);
545 /* Clone packet if we're doing a 'tee' */
546 if (tee)
547 clone = m_dup(m, M_NOWAIT);
550 * XXX
551 * delayed checksums are not currently compatible
552 * with divert sockets.
554 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
555 in_delayed_cksum(m);
556 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
559 /* Restore packet header fields to original values */
560 ip->ip_len = htons(ip->ip_len);
561 ip->ip_off = htons(ip->ip_off);
563 /* Deliver packet to divert input routine */
564 divert_packet(m, 0);
566 /* If 'tee', continue with original packet */
567 return clone;
570 static struct mbuf *
571 ip_divert_in(struct mbuf *m, int tee)
573 struct mbuf *clone = NULL;
574 struct ip *ip = mtod(m, struct ip *);
575 struct m_tag *mtag;
577 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
578 const struct divert_info *divinfo;
579 u_short frag_off;
580 int hlen;
583 * Only trust divert info in the fragment
584 * at offset 0.
586 frag_off = ip->ip_off << 3;
587 if (frag_off != 0) {
588 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
589 m_tag_delete(m, mtag);
593 * Attempt reassembly; if it succeeds, proceed.
594 * ip_reass() will return a different mbuf.
596 m = ip_reass(m);
597 if (m == NULL)
598 return NULL;
599 ip = mtod(m, struct ip *);
601 /* Caller need to redispatch the packet, if it is for us */
602 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH;
605 * Get the header length of the reassembled
606 * packet
608 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
611 * Restore original checksum before diverting
612 * packet
614 ip->ip_len += hlen;
615 ip->ip_len = htons(ip->ip_len);
616 ip->ip_off = htons(ip->ip_off);
617 ip->ip_sum = 0;
618 if (hlen == sizeof(struct ip))
619 ip->ip_sum = in_cksum_hdr(ip);
620 else
621 ip->ip_sum = in_cksum(m, hlen);
622 ip->ip_off = ntohs(ip->ip_off);
623 ip->ip_len = ntohs(ip->ip_len);
626 * Only use the saved divert info
628 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
629 if (mtag == NULL) {
630 /* Wrongly configured ipfw */
631 kprintf("ip_input no divert info\n");
632 m_freem(m);
633 return NULL;
635 divinfo = m_tag_data(mtag);
636 tee = divinfo->tee;
640 * Divert or tee packet to the divert protocol if
641 * required.
644 /* Clone packet if we're doing a 'tee' */
645 if (tee)
646 clone = m_dup(m, M_NOWAIT);
649 * Restore packet header fields to original
650 * values
652 ip->ip_len = htons(ip->ip_len);
653 ip->ip_off = htons(ip->ip_off);
655 /* Deliver packet to divert input routine */
656 divert_packet(m, 1);
658 /* Catch invalid reference */
659 m = NULL;
660 ip = NULL;
662 ipstat.ips_delivered++;
664 /* If 'tee', continue with original packet */
665 if (clone != NULL) {
667 * Complete processing of the packet.
668 * XXX Better safe than sorry, remove the DIVERT tag.
670 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL);
671 KKASSERT(mtag != NULL);
672 m_tag_delete(clone, mtag);
674 return clone;
677 static struct mbuf *
678 ip_divert(struct mbuf *m, int tee, int incoming)
680 struct mbuf *ret;
682 if (incoming)
683 ret = ip_divert_in(m, tee);
684 else
685 ret = ip_divert_out(m, tee);
686 return ret;