xref: /freebsd/sys/netinet/ip_divert.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
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. 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.
20  *
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
31  * SUCH DAMAGE.
32  *
33  * $FreeBSD$
34  */
35 
36 #include "opt_inet.h"
37 #include "opt_ipfw.h"
38 #include "opt_ipdivert.h"
39 #include "opt_ipsec.h"
40 
41 #ifndef INET
42 #error "IPDIVERT requires INET."
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/protosw.h>
50 #include <sys/socketvar.h>
51 #include <sys/systm.h>
52 #include <sys/proc.h>
53 
54 #include <vm/vm_zone.h>
55 
56 #include <net/if.h>
57 #include <net/route.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_var.h>
65 
66 /*
67  * Divert sockets
68  */
69 
70 /*
71  * Allocate enough space to hold a full IP packet
72  */
73 #define	DIVSNDQ		(65536 + 100)
74 #define	DIVRCVQ		(65536 + 100)
75 
76 /*
77  * A 16 bit cookie is passed to and from the user process.
78  * The user process can send it back to help the caller know
79  * something about where the packet originally came from.
80  *
81  * In the case of ipfw, then the cookie is the rule that sent
82  * us here. On reinjection is is the rule after which processing
83  * should continue. Leaving it the same will make processing start
84  * at the rule number after that which sent it here. Setting it to
85  * 0 will restart processing at the beginning.
86  *
87  * For divert_packet(), ip_divert_cookie is an input value only.
88  * For div_output(), ip_divert_cookie is an output value only.
89  */
90 u_int16_t ip_divert_cookie;
91 
92 /* Internal variables */
93 static struct inpcbhead divcb;
94 static struct inpcbinfo divcbinfo;
95 
96 static u_long	div_sendspace = DIVSNDQ;	/* XXX sysctl ? */
97 static u_long	div_recvspace = DIVRCVQ;	/* XXX sysctl ? */
98 
99 /* Optimization: have this preinitialized */
100 static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET };
101 
102 /* Internal functions */
103 static int div_output(struct socket *so,
104 		struct mbuf *m, struct sockaddr *addr, struct mbuf *control);
105 
106 /*
107  * Initialize divert connection block queue.
108  */
109 void
110 div_init(void)
111 {
112 	LIST_INIT(&divcb);
113 	divcbinfo.listhead = &divcb;
114 	/*
115 	 * XXX We don't use the hash list for divert IP, but it's easier
116 	 * to allocate a one entry hash list than it is to check all
117 	 * over the place for hashbase == NULL.
118 	 */
119 	divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
120 	divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
121 	divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb),
122 				   maxsockets, ZONE_INTERRUPT, 0);
123 }
124 
125 /*
126  * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
127  * with that protocol number to enter the system from the outside.
128  */
129 void
130 div_input(struct mbuf *m, int off, int proto)
131 {
132 	ipstat.ips_noproto++;
133 	m_freem(m);
134 }
135 
136 /*
137  * Divert a packet by passing it up to the divert socket at port 'port'.
138  *
139  * Setup generic address and protocol structures for div_input routine,
140  * then pass them along with mbuf chain.
141  */
142 void
143 divert_packet(struct mbuf *m, int incoming, int port)
144 {
145 	struct ip *ip;
146 	struct inpcb *inp;
147 	struct socket *sa;
148 	u_int16_t nport;
149 
150 	/* Sanity check */
151 	KASSERT(port != 0, ("%s: port=0", __FUNCTION__));
152 
153 	/* Record and reset divert cookie */
154 	divsrc.sin_port = ip_divert_cookie;
155 	ip_divert_cookie = 0;
156 
157 	/* Assure header */
158 	if (m->m_len < sizeof(struct ip) &&
159 	    (m = m_pullup(m, sizeof(struct ip))) == 0) {
160 		return;
161 	}
162 	ip = mtod(m, struct ip *);
163 
164 	/*
165 	 * Record receive interface address, if any.
166 	 * But only for incoming packets.
167 	 */
168 	divsrc.sin_addr.s_addr = 0;
169 	if (incoming) {
170 		struct ifaddr *ifa;
171 
172 		/* Sanity check */
173 		KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __FUNCTION__));
174 
175 		/* Find IP address for receive interface */
176 		for (ifa = m->m_pkthdr.rcvif->if_addrhead.tqh_first;
177 		    ifa != NULL; ifa = ifa->ifa_link.tqe_next) {
178 			if (ifa->ifa_addr == NULL)
179 				continue;
180 			if (ifa->ifa_addr->sa_family != AF_INET)
181 				continue;
182 			divsrc.sin_addr =
183 			    ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
184 			break;
185 		}
186 	}
187 	/*
188 	 * Record the incoming interface name whenever we have one.
189 	 */
190 	bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero));
191 	if (m->m_pkthdr.rcvif) {
192 		/*
193 		 * Hide the actual interface name in there in the
194 		 * sin_zero array. XXX This needs to be moved to a
195 		 * different sockaddr type for divert, e.g.
196 		 * sockaddr_div with multiple fields like
197 		 * sockaddr_dl. Presently we have only 7 bytes
198 		 * but that will do for now as most interfaces
199 		 * are 4 or less + 2 or less bytes for unit.
200 		 * There is probably a faster way of doing this,
201 		 * possibly taking it from the sockaddr_dl on the iface.
202 		 * This solves the problem of a P2P link and a LAN interface
203 		 * having the same address, which can result in the wrong
204 		 * interface being assigned to the packet when fed back
205 		 * into the divert socket. Theoretically if the daemon saves
206 		 * and re-uses the sockaddr_in as suggested in the man pages,
207 		 * this iface name will come along for the ride.
208 		 * (see div_output for the other half of this.)
209 		 */
210 		snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero),
211 			"%s%d", m->m_pkthdr.rcvif->if_name,
212 			m->m_pkthdr.rcvif->if_unit);
213 	}
214 
215 	/* Put packet on socket queue, if any */
216 	sa = NULL;
217 	nport = htons((u_int16_t)port);
218 	for (inp = divcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
219 		if (inp->inp_lport == nport)
220 			sa = inp->inp_socket;
221 	}
222 	if (sa) {
223 		if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc,
224 				m, (struct mbuf *)0) == 0)
225 			m_freem(m);
226 		else
227 			sorwakeup(sa);
228 	} else {
229 		m_freem(m);
230 		ipstat.ips_noproto++;
231 		ipstat.ips_delivered--;
232         }
233 }
234 
235 /*
236  * Deliver packet back into the IP processing machinery.
237  *
238  * If no address specified, or address is 0.0.0.0, send to ip_output();
239  * otherwise, send to ip_input() and mark as having been received on
240  * the interface with that address.
241  */
242 static int
243 div_output(so, m, addr, control)
244 	struct socket *so;
245 	register struct mbuf *m;
246 	struct sockaddr *addr;
247 	struct mbuf *control;
248 {
249 	register struct inpcb *const inp = sotoinpcb(so);
250 	register struct ip *const ip = mtod(m, struct ip *);
251 	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
252 	int error = 0;
253 
254 	if (control)
255 		m_freem(control);		/* XXX */
256 
257 	/* Loopback avoidance and state recovery */
258 	if (sin) {
259 		int	len = 0;
260 		char	*c = sin->sin_zero;
261 
262 		ip_divert_cookie = sin->sin_port;
263 
264 		/*
265 		 * Find receive interface with the given name or IP address.
266 		 * The name is user supplied data so don't trust it's size or
267 		 * that it is zero terminated. The name has priority.
268 		 * We are presently assuming that the sockaddr_in
269 		 * has not been replaced by a sockaddr_div, so we limit it
270 		 * to 16 bytes in total. the name is stuffed (if it exists)
271 		 * in the sin_zero[] field.
272 		 */
273 		while (*c++ && (len++ < sizeof(sin->sin_zero)));
274 		if ((len > 0) && (len < sizeof(sin->sin_zero)))
275 			m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
276 	} else {
277 		ip_divert_cookie = 0;
278 	}
279 
280 	/* Reinject packet into the system as incoming or outgoing */
281 	if (!sin || sin->sin_addr.s_addr == 0) {
282 		/*
283 		 * Don't allow both user specified and setsockopt options,
284 		 * and don't allow packet length sizes that will crash
285 		 */
286 		if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
287 		     ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
288 			error = EINVAL;
289 			goto cantsend;
290 		}
291 
292 		/* Convert fields to host order for ip_output() */
293 		NTOHS(ip->ip_len);
294 		NTOHS(ip->ip_off);
295 
296 		/* Send packet to output processing */
297 		ipstat.ips_rawout++;			/* XXX */
298 		error = ip_output(m, inp->inp_options, &inp->inp_route,
299 			(so->so_options & SO_DONTROUTE) |
300 			IP_ALLOWBROADCAST | IP_RAWOUTPUT,
301 			inp->inp_moptions);
302 	} else {
303 		struct	ifaddr *ifa;
304 
305 		/* If no luck with the name above. check by IP address.  */
306 		if (m->m_pkthdr.rcvif == NULL) {
307 			/*
308 			 * Make sure there are no distractions
309 			 * for ifa_ifwithaddr. Clear the port and the ifname.
310 			 * Maybe zap all 8 bytes at once using a 64bit write?
311 			 */
312 			bzero(sin->sin_zero, sizeof(sin->sin_zero));
313 			/* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */
314 			sin->sin_port = 0;
315 			if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) {
316 				error = EADDRNOTAVAIL;
317 				goto cantsend;
318 			}
319 			m->m_pkthdr.rcvif = ifa->ifa_ifp;
320 		}
321 
322 		/* Send packet to input processing */
323 		ip_input(m);
324 	}
325 
326 	/* paranoid: Reset for next time (and other packets) */
327 	/* almost definitly already done in the ipfw filter but.. */
328 	ip_divert_cookie = 0;
329 	return error;
330 
331 cantsend:
332 	m_freem(m);
333 	ip_divert_cookie = 0;
334 	return error;
335 }
336 
337 static int
338 div_attach(struct socket *so, int proto, struct proc *p)
339 {
340 	struct inpcb *inp;
341 	int error, s;
342 
343 	inp  = sotoinpcb(so);
344 	if (inp)
345 		panic("div_attach");
346 	if (p && (error = suser(p)) != 0)
347 		return error;
348 
349 	error = soreserve(so, div_sendspace, div_recvspace);
350 	if (error)
351 		return error;
352 	s = splnet();
353 	error = in_pcballoc(so, &divcbinfo, p);
354 	splx(s);
355 	if (error)
356 		return error;
357 	inp = (struct inpcb *)so->so_pcb;
358 	inp->inp_ip_p = proto;
359 	inp->inp_flags |= INP_HDRINCL;
360 	/* The socket is always "connected" because
361 	   we always know "where" to send the packet */
362 	so->so_state |= SS_ISCONNECTED;
363 #ifdef IPSEC
364 	error = ipsec_init_policy(so, &inp->inp_sp);
365 	if (error != 0) {
366 		in_pcbdetach(inp);
367 		return error;
368 	}
369 #endif /*IPSEC*/
370 	return 0;
371 }
372 
373 static int
374 div_detach(struct socket *so)
375 {
376 	struct inpcb *inp;
377 
378 	inp = sotoinpcb(so);
379 	if (inp == 0)
380 		panic("div_detach");
381 	in_pcbdetach(inp);
382 	return 0;
383 }
384 
385 static int
386 div_abort(struct socket *so)
387 {
388 	soisdisconnected(so);
389 	return div_detach(so);
390 }
391 
392 static int
393 div_disconnect(struct socket *so)
394 {
395 	if ((so->so_state & SS_ISCONNECTED) == 0)
396 		return ENOTCONN;
397 	return div_abort(so);
398 }
399 
400 static int
401 div_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
402 {
403 	struct inpcb *inp;
404 	int s;
405 	int error;
406 
407 	s = splnet();
408 	inp = sotoinpcb(so);
409 	error = in_pcbbind(inp, nam, p);
410 	splx(s);
411 	return 0;
412 }
413 
414 static int
415 div_shutdown(struct socket *so)
416 {
417 	socantsendmore(so);
418 	return 0;
419 }
420 
421 static int
422 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
423 	 struct mbuf *control, struct proc *p)
424 {
425 	/* Packet must have a header (but that's about it) */
426 	if (m->m_len < sizeof (struct ip) &&
427 	    (m = m_pullup(m, sizeof (struct ip))) == 0) {
428 		ipstat.ips_toosmall++;
429 		m_freem(m);
430 		return EINVAL;
431 	}
432 
433 	/* Send packet */
434 	return div_output(so, m, nam, control);
435 }
436 
437 struct pr_usrreqs div_usrreqs = {
438 	div_abort, pru_accept_notsupp, div_attach, div_bind,
439 	pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
440 	div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp,
441 	pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
442 	in_setsockaddr, sosend, soreceive, sopoll
443 };
444