Once we distribute socket protocol processing requests to different
[dragonfly.git] / sys / netinet / ip_divert.c
CommitLineData
984263bc
MD
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: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $
e4700d00 34 * $DragonFly: src/sys/netinet/ip_divert.c,v 1.8 2004/03/05 16:57:15 hsu Exp $
984263bc
MD
35 */
36
37#include "opt_inet.h"
38#include "opt_ipfw.h"
39#include "opt_ipdivert.h"
40#include "opt_ipsec.h"
41
42#ifndef INET
43#error "IPDIVERT requires INET."
44#endif
45
46#include <sys/param.h>
47#include <sys/kernel.h>
48#include <sys/malloc.h>
49#include <sys/mbuf.h>
50#include <sys/socket.h>
51#include <sys/protosw.h>
52#include <sys/socketvar.h>
53#include <sys/sysctl.h>
54#include <sys/systm.h>
55#include <sys/proc.h>
56
57#include <vm/vm_zone.h>
58
59#include <net/if.h>
60#include <net/route.h>
61
62#include <netinet/in.h>
63#include <netinet/in_systm.h>
64#include <netinet/ip.h>
65#include <netinet/in_pcb.h>
66#include <netinet/in_var.h>
67#include <netinet/ip_var.h>
68
69/*
70 * Divert sockets
71 */
72
73/*
74 * Allocate enough space to hold a full IP packet
75 */
76#define DIVSNDQ (65536 + 100)
77#define DIVRCVQ (65536 + 100)
78
79/*
80 * Divert sockets work in conjunction with ipfw, see the divert(4)
81 * manpage for features.
82 * Internally, packets selected by ipfw in ip_input() or ip_output(),
83 * and never diverted before, are passed to the input queue of the
84 * divert socket with a given 'divert_port' number (as specified in
85 * the matching ipfw rule), and they are tagged with a 16 bit cookie
86 * (representing the rule number of the matching ipfw rule), which
87 * is passed to process reading from the socket.
88 *
89 * Packets written to the divert socket are again tagged with a cookie
90 * (usually the same as above) and a destination address.
91 * If the destination address is INADDR_ANY then the packet is
92 * treated as outgoing and sent to ip_output(), otherwise it is
93 * treated as incoming and sent to ip_input().
94 * In both cases, the packet is tagged with the cookie.
95 *
96 * On reinjection, processing in ip_input() and ip_output()
97 * will be exactly the same as for the original packet, except that
98 * ipfw processing will start at the rule number after the one
99 * written in the cookie (so, tagging a packet with a cookie of 0
100 * will cause it to be effectively considered as a standard packet).
101 */
102
103/* Internal variables */
104static struct inpcbhead divcb;
105static struct inpcbinfo divcbinfo;
106
107static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */
108static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */
109
110/* Optimization: have this preinitialized */
111static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET };
112
113/*
114 * Initialize divert connection block queue.
115 */
116void
117div_init(void)
118{
119 LIST_INIT(&divcb);
120 divcbinfo.listhead = &divcb;
121 /*
122 * XXX We don't use the hash list for divert IP, but it's easier
123 * to allocate a one entry hash list than it is to check all
124 * over the place for hashbase == NULL.
125 */
126 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask);
127 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask);
13d8907a 128 divcbinfo.bindhashbase = hashinit(1, M_PCB, &divcbinfo.bindhashmask);
984263bc
MD
129 divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb),
130 maxsockets, ZONE_INTERRUPT, 0);
131}
132
133/*
134 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets
135 * with that protocol number to enter the system from the outside.
136 */
137void
138div_input(struct mbuf *m, int off, int proto)
139{
140 ipstat.ips_noproto++;
141 m_freem(m);
142}
143
144/*
145 * Divert a packet by passing it up to the divert socket at port 'port'.
146 *
147 * Setup generic address and protocol structures for div_input routine,
148 * then pass them along with mbuf chain.
149 */
150void
151divert_packet(struct mbuf *m, int incoming, int port, int rule)
152{
153 struct ip *ip;
154 struct inpcb *inp;
155 struct socket *sa;
156 u_int16_t nport;
157
158 /* Sanity check */
159 KASSERT(port != 0, ("%s: port=0", __FUNCTION__));
160
161 divsrc.sin_port = rule; /* record matching rule */
162
163 /* Assure header */
164 if (m->m_len < sizeof(struct ip) &&
165 (m = m_pullup(m, sizeof(struct ip))) == 0)
166 return;
167 ip = mtod(m, struct ip *);
168
169 /*
170 * Record receive interface address, if any.
171 * But only for incoming packets.
172 */
173 divsrc.sin_addr.s_addr = 0;
174 if (incoming) {
175 struct ifaddr *ifa;
176
177 /* Sanity check */
178 KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __FUNCTION__));
179
180 /* Find IP address for receive interface */
181 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
182 if (ifa->ifa_addr == NULL)
183 continue;
184 if (ifa->ifa_addr->sa_family != AF_INET)
185 continue;
186 divsrc.sin_addr =
187 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr;
188 break;
189 }
190 }
191 /*
192 * Record the incoming interface name whenever we have one.
193 */
194 bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero));
195 if (m->m_pkthdr.rcvif) {
196 /*
197 * Hide the actual interface name in there in the
198 * sin_zero array. XXX This needs to be moved to a
199 * different sockaddr type for divert, e.g.
200 * sockaddr_div with multiple fields like
201 * sockaddr_dl. Presently we have only 7 bytes
202 * but that will do for now as most interfaces
203 * are 4 or less + 2 or less bytes for unit.
204 * There is probably a faster way of doing this,
205 * possibly taking it from the sockaddr_dl on the iface.
206 * This solves the problem of a P2P link and a LAN interface
207 * having the same address, which can result in the wrong
208 * interface being assigned to the packet when fed back
209 * into the divert socket. Theoretically if the daemon saves
210 * and re-uses the sockaddr_in as suggested in the man pages,
211 * this iface name will come along for the ride.
212 * (see div_output for the other half of this.)
213 */
bee64f81
MD
214 snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero),
215 m->m_pkthdr.rcvif->if_xname);
984263bc
MD
216 }
217
218 /* Put packet on socket queue, if any */
219 sa = NULL;
220 nport = htons((u_int16_t)port);
221 LIST_FOREACH(inp, &divcb, inp_list) {
222 if (inp->inp_lport == nport)
223 sa = inp->inp_socket;
224 }
225 if (sa) {
226 if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc,
227 m, (struct mbuf *)0) == 0)
228 m_freem(m);
229 else
230 sorwakeup(sa);
231 } else {
232 m_freem(m);
233 ipstat.ips_noproto++;
234 ipstat.ips_delivered--;
235 }
236}
237
238/*
239 * Deliver packet back into the IP processing machinery.
240 *
241 * If no address specified, or address is 0.0.0.0, send to ip_output();
242 * otherwise, send to ip_input() and mark as having been received on
243 * the interface with that address.
244 */
245static int
246div_output(struct socket *so, struct mbuf *m,
247 struct sockaddr_in *sin, struct mbuf *control)
248{
249 int error = 0;
250 struct m_hdr divert_tag;
251
252 /*
253 * Prepare the tag for divert info. Note that a packet
254 * with a 0 tag in mh_data is effectively untagged,
255 * so we could optimize that case.
256 */
257 divert_tag.mh_type = MT_TAG;
258 divert_tag.mh_flags = PACKET_TAG_DIVERT;
259 divert_tag.mh_next = m;
260 divert_tag.mh_data = 0; /* the matching rule # */
261 m->m_pkthdr.rcvif = NULL; /* XXX is it necessary ? */
262
263 if (control)
264 m_freem(control); /* XXX */
265
266 /* Loopback avoidance and state recovery */
267 if (sin) {
268 int i;
269
270 divert_tag.mh_data = (caddr_t)(int)sin->sin_port;
271 /*
272 * Find receive interface with the given name, stuffed
273 * (if it exists) in the sin_zero[] field.
274 * The name is user supplied data so don't trust its size
275 * or that it is zero terminated.
276 */
277 for (i = 0; sin->sin_zero[i] && i < sizeof(sin->sin_zero); i++)
278 ;
279 if ( i > 0 && i < sizeof(sin->sin_zero))
280 m->m_pkthdr.rcvif = ifunit(sin->sin_zero);
281 }
282
283 /* Reinject packet into the system as incoming or outgoing */
284 if (!sin || sin->sin_addr.s_addr == 0) {
285 struct inpcb *const inp = sotoinpcb(so);
286 struct ip *const ip = mtod(m, struct ip *);
287
288 /*
289 * Don't allow both user specified and setsockopt options,
290 * and don't allow packet length sizes that will crash
291 */
292 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) ||
293 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) {
294 error = EINVAL;
295 goto cantsend;
296 }
297
298 /* Convert fields to host order for ip_output() */
299 ip->ip_len = ntohs(ip->ip_len);
300 ip->ip_off = ntohs(ip->ip_off);
301
302 /* Send packet to output processing */
303 ipstat.ips_rawout++; /* XXX */
304 error = ip_output((struct mbuf *)&divert_tag,
305 inp->inp_options, &inp->inp_route,
306 (so->so_options & SO_DONTROUTE) |
307 IP_ALLOWBROADCAST | IP_RAWOUTPUT,
308 inp->inp_moptions, NULL);
309 } else {
310 if (m->m_pkthdr.rcvif == NULL) {
311 /*
312 * No luck with the name, check by IP address.
313 * Clear the port and the ifname to make sure
314 * there are no distractions for ifa_ifwithaddr.
315 */
316 struct ifaddr *ifa;
317
318 bzero(sin->sin_zero, sizeof(sin->sin_zero));
319 sin->sin_port = 0;
320 ifa = ifa_ifwithaddr((struct sockaddr *) sin);
321 if (ifa == NULL) {
322 error = EADDRNOTAVAIL;
323 goto cantsend;
324 }
325 m->m_pkthdr.rcvif = ifa->ifa_ifp;
326 }
327 /* Send packet to input processing */
328 ip_input((struct mbuf *)&divert_tag);
329 }
330
331 return error;
332
333cantsend:
334 m_freem(m);
335 return error;
336}
337
338static int
e4700d00 339div_attach(struct socket *so, int proto, struct pru_attach_info *ai)
984263bc
MD
340{
341 struct inpcb *inp;
342 int error, s;
343
344 inp = sotoinpcb(so);
345 if (inp)
346 panic("div_attach");
e4700d00 347 if ((error = suser_cred(ai->p_ucred, NULL_CRED_OKAY)) != 0)
984263bc
MD
348 return error;
349
e4700d00 350 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit);
984263bc
MD
351 if (error)
352 return error;
353 s = splnet();
e4700d00 354 error = in_pcballoc(so, &divcbinfo);
984263bc
MD
355 splx(s);
356 if (error)
357 return error;
358 inp = (struct inpcb *)so->so_pcb;
359 inp->inp_ip_p = proto;
360 inp->inp_vflag |= INP_IPV4;
361 inp->inp_flags |= INP_HDRINCL;
362 /* The socket is always "connected" because
363 we always know "where" to send the packet */
364 so->so_state |= SS_ISCONNECTED;
365 return 0;
366}
367
368static int
369div_detach(struct socket *so)
370{
371 struct inpcb *inp;
372
373 inp = sotoinpcb(so);
374 if (inp == 0)
375 panic("div_detach");
376 in_pcbdetach(inp);
377 return 0;
378}
379
380static int
381div_abort(struct socket *so)
382{
383 soisdisconnected(so);
384 return div_detach(so);
385}
386
387static int
388div_disconnect(struct socket *so)
389{
390 if ((so->so_state & SS_ISCONNECTED) == 0)
391 return ENOTCONN;
392 return div_abort(so);
393}
394
395static int
dadab5e9 396div_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
397{
398 struct inpcb *inp;
399 int s;
400 int error;
401
402 s = splnet();
403 inp = sotoinpcb(so);
404 /* in_pcbbind assumes that nam is a sockaddr_in
405 * and in_pcbbind requires a valid address. Since divert
406 * sockets don't we need to make sure the address is
407 * filled in properly.
408 * XXX -- divert should not be abusing in_pcbind
409 * and should probably have its own family.
410 */
411 if (nam->sa_family != AF_INET)
412 error = EAFNOSUPPORT;
413 else {
414 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY;
dadab5e9 415 error = in_pcbbind(inp, nam, td);
984263bc
MD
416 }
417 splx(s);
418 return error;
419}
420
421static int
422div_shutdown(struct socket *so)
423{
424 socantsendmore(so);
425 return 0;
426}
427
428static int
429div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
dadab5e9 430 struct mbuf *control, struct thread *td)
984263bc
MD
431{
432 /* Packet must have a header (but that's about it) */
433 if (m->m_len < sizeof (struct ip) &&
434 (m = m_pullup(m, sizeof (struct ip))) == 0) {
435 ipstat.ips_toosmall++;
436 m_freem(m);
437 return EINVAL;
438 }
439
440 /* Send packet */
441 return div_output(so, m, (struct sockaddr_in *)nam, control);
442}
443
444static int
445div_pcblist(SYSCTL_HANDLER_ARGS)
446{
447 int error, i, n, s;
448 struct inpcb *inp, **inp_list;
449 inp_gen_t gencnt;
450 struct xinpgen xig;
451
452 /*
453 * The process of preparing the TCB list is too time-consuming and
454 * resource-intensive to repeat twice on every request.
455 */
456 if (req->oldptr == 0) {
457 n = divcbinfo.ipi_count;
458 req->oldidx = 2 * (sizeof xig)
459 + (n + n/8) * sizeof(struct xinpcb);
460 return 0;
461 }
462
463 if (req->newptr != 0)
464 return EPERM;
465
466 /*
467 * OK, now we're committed to doing something.
468 */
469 s = splnet();
470 gencnt = divcbinfo.ipi_gencnt;
471 n = divcbinfo.ipi_count;
472 splx(s);
473
474 xig.xig_len = sizeof xig;
475 xig.xig_count = n;
476 xig.xig_gen = gencnt;
477 xig.xig_sogen = so_gencnt;
478 error = SYSCTL_OUT(req, &xig, sizeof xig);
479 if (error)
480 return error;
481
482 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
483 if (inp_list == 0)
484 return ENOMEM;
485
486 s = splnet();
487 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n;
488 inp = LIST_NEXT(inp, inp_list)) {
dadab5e9 489 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp))
984263bc
MD
490 inp_list[i++] = inp;
491 }
492 splx(s);
493 n = i;
494
495 error = 0;
496 for (i = 0; i < n; i++) {
497 inp = inp_list[i];
498 if (inp->inp_gencnt <= gencnt) {
499 struct xinpcb xi;
500 xi.xi_len = sizeof xi;
501 /* XXX should avoid extra copy */
502 bcopy(inp, &xi.xi_inp, sizeof *inp);
503 if (inp->inp_socket)
504 sotoxsocket(inp->inp_socket, &xi.xi_socket);
505 error = SYSCTL_OUT(req, &xi, sizeof xi);
506 }
507 }
508 if (!error) {
509 /*
510 * Give the user an updated idea of our state.
511 * If the generation differs from what we told
512 * her before, she knows that something happened
513 * while we were processing this request, and it
514 * might be necessary to retry.
515 */
516 s = splnet();
517 xig.xig_gen = divcbinfo.ipi_gencnt;
518 xig.xig_sogen = so_gencnt;
519 xig.xig_count = divcbinfo.ipi_count;
520 splx(s);
521 error = SYSCTL_OUT(req, &xig, sizeof xig);
522 }
523 free(inp_list, M_TEMP);
524 return error;
525}
526
527SYSCTL_DECL(_net_inet_divert);
528SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0,
529 div_pcblist, "S,xinpcb", "List of active divert sockets");
530
531struct pr_usrreqs div_usrreqs = {
532 div_abort, pru_accept_notsupp, div_attach, div_bind,
533 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach,
534 div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp,
535 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown,
536 in_setsockaddr, sosend, soreceive, sopoll
537};