xref: /openbsd-src/usr.sbin/dhcrelay/bpf.c (revision 7b08a90aae262418a4b5fe9eae6b24d21fd1dcd4)
1 /*	$OpenBSD: bpf.c,v 1.19 2017/04/19 05:36:12 natano Exp $ */
2 
3 /* BPF socket interface code, originally contributed by Archie Cobbs. */
4 
5 /*
6  * Copyright (c) 1995, 1996, 1998, 1999
7  * The Internet Software Consortium.    All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of The Internet Software Consortium nor the names
19  *    of its contributors may be used to endorse or promote products derived
20  *    from this software without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE INTERNET SOFTWARE CONSORTIUM AND
23  * CONTRIBUTORS ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,
24  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26  * DISCLAIMED.  IN NO EVENT SHALL THE INTERNET SOFTWARE CONSORTIUM OR
27  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
29  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
30  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
31  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * This software has been written for the Internet Software Consortium
37  * by Ted Lemon <mellon@fugue.com> in cooperation with Vixie
38  * Enterprises.  To learn more about the Internet Software Consortium,
39  * see ``http://www.vix.com/isc''.  To learn more about Vixie
40  * Enterprises, see ``http://www.vix.com''.
41  */
42 
43 #include <sys/types.h>
44 #include <sys/ioctl.h>
45 #include <sys/socket.h>
46 
47 #include <net/bpf.h>
48 #include <net/if.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/if_ether.h>
52 
53 #include <errno.h>
54 #include <fcntl.h>
55 #include <stdio.h>
56 #include <stdlib.h>
57 #include <string.h>
58 #include <unistd.h>
59 
60 #include "dhcp.h"
61 #include "dhcpd.h"
62 #include "log.h"
63 
64 /*
65  * Called by get_interface_list for each interface that's discovered.
66  * Opens a packet filter for each interface and adds it to the select
67  * mask.
68  */
69 int
if_register_bpf(struct interface_info * info)70 if_register_bpf(struct interface_info *info)
71 {
72 	int sock;
73 
74 	/* Open the BPF device */
75 	if ((sock = open("/dev/bpf", O_RDWR)) == -1)
76 		fatal("Can't open bpf device");
77 
78 	/* Set the BPF device to point at this interface. */
79 	if (ioctl(sock, BIOCSETIF, &info->ifr) == -1)
80 		fatal("Can't attach interface %s to bpf device", info->name);
81 
82 	return (sock);
83 }
84 
85 void
if_register_send(struct interface_info * info)86 if_register_send(struct interface_info *info)
87 {
88 	/*
89 	 * If we're using the bpf API for sending and receiving, we
90 	 * don't need to register this interface twice.
91 	 */
92 	info->wfdesc = info->rfdesc;
93 }
94 
95 /*
96  * Packet filter program: 'ip and udp and dst port CLIENT_PORT'
97  */
98 struct bpf_insn dhcp_bpf_sfilter[] = {
99 	/* Make sure this is an IP packet... */
100 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
101 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
102 
103 	/* Make sure it's a UDP packet... */
104 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
105 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
106 
107 	/* Make sure this isn't a fragment... */
108 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
109 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
110 
111 	/* Get the IP header length... */
112 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
113 
114 	/* Make sure it's to the right port... */
115 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 16),
116 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, CLIENT_PORT, 0, 1),
117 
118 	/* If we passed all the tests, ask for the whole packet. */
119 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
120 
121 	/* Otherwise, drop it. */
122 	BPF_STMT(BPF_RET+BPF_K, 0),
123 };
124 
125 int dhcp_bpf_sfilter_len = sizeof(dhcp_bpf_sfilter) / sizeof(struct bpf_insn);
126 
127 /*
128  * Packet filter program: 'ip and udp and dst port SERVER_PORT'
129  */
130 struct bpf_insn dhcp_bpf_filter[] = {
131 	/* Make sure this is an IP packet... */
132 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
133 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
134 
135 	/* Make sure it's a UDP packet... */
136 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
137 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
138 
139 	/* Make sure this isn't a fragment... */
140 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
141 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
142 
143 	/* Get the IP header length... */
144 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
145 
146 	/* Make sure it's to the right port... */
147 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 16),
148 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
149 
150 	/* If we passed all the tests, ask for the whole packet. */
151 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
152 
153 	/* Otherwise, drop it. */
154 	BPF_STMT(BPF_RET+BPF_K, 0),
155 };
156 
157 int dhcp_bpf_filter_len = sizeof(dhcp_bpf_filter) / sizeof(struct bpf_insn);
158 
159 /*
160  * Packet filter program: encapsulated 'ip and udp and dst port SERVER_PORT'
161  */
162 struct bpf_insn dhcp_bpf_efilter[] = {
163 	/* Make sure this is an encapsulated AF_INET packet... */
164 	BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 0),
165 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, AF_INET << 24, 0, 10),
166 
167 	/* Make sure it's an IPIP packet... */
168 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 21),
169 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_IPIP, 0, 8),
170 
171 	/* Make sure it's an encapsulated UDP packet... */
172 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 41),
173 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
174 
175 	/* Make sure this isn't a fragment... */
176 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 38),
177 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
178 
179 	/* Get the IP header length... */
180 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 32),
181 
182 	/* Make sure it's to the right port... */
183 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 34),
184 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
185 
186 	/* If we passed all the tests, ask for the whole packet. */
187 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
188 
189 	/* Otherwise, drop it. */
190 	BPF_STMT(BPF_RET+BPF_K, 0),
191 };
192 
193 int dhcp_bpf_efilter_len = sizeof(dhcp_bpf_efilter) / sizeof(struct bpf_insn);
194 
195 /*
196  * Packet write filter program: 'ip and udp and src port CLIENT_PORT'
197  */
198 struct bpf_insn dhcp_bpf_swfilter[] = {
199 	/* Make sure this is an IP packet... */
200 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
201 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
202 
203 	/* Make sure it's a UDP packet... */
204 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
205 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
206 
207 	/* Make sure this isn't a fragment... */
208 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
209 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
210 
211 	/* Get the IP header length... */
212 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
213 
214 	/* Make sure it's from the right port... */
215 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 14),
216 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, CLIENT_PORT, 0, 1),
217 
218 	/* If we passed all the tests, ask for the whole packet. */
219 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
220 
221 	/* Otherwise, drop it. */
222 	BPF_STMT(BPF_RET+BPF_K, 0),
223 };
224 
225 int dhcp_bpf_swfilter_len = sizeof(dhcp_bpf_swfilter) /
226 	sizeof(struct bpf_insn);
227 
228 /*
229  * Packet write filter program: 'ip and udp and src port SERVER_PORT'
230  */
231 struct bpf_insn dhcp_bpf_wfilter[] = {
232 	/* Make sure this is an IP packet... */
233 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
234 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
235 
236 	/* Make sure it's a UDP packet... */
237 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
238 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
239 
240 	/* Make sure this isn't a fragment... */
241 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
242 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
243 
244 	/* Get the IP header length... */
245 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
246 
247 	/* Make sure it's from the right port... */
248 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 14),
249 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
250 
251 	/* If we passed all the tests, ask for the whole packet. */
252 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
253 
254 	/* Otherwise, drop it. */
255 	BPF_STMT(BPF_RET+BPF_K, 0),
256 };
257 
258 int dhcp_bpf_wfilter_len = sizeof(dhcp_bpf_wfilter) / sizeof(struct bpf_insn);
259 
260 void
if_register_receive(struct interface_info * info,int isserver)261 if_register_receive(struct interface_info *info, int isserver)
262 {
263 	struct bpf_version v;
264 	struct bpf_program p;
265 	int flag = 1, sz, cmplt = 0;
266 
267 	/* Open a BPF device and hang it on this interface... */
268 	info->rfdesc = if_register_bpf(info);
269 
270 	/* Make sure the BPF version is in range... */
271 	if (ioctl(info->rfdesc, BIOCVERSION, &v) == -1)
272 		fatal("Can't get BPF version");
273 
274 	if (v.bv_major != BPF_MAJOR_VERSION ||
275 	    v.bv_minor < BPF_MINOR_VERSION)
276 		fatalx("Kernel BPF version out of range - recompile dhcpd!");
277 
278 	/*
279 	 * Set immediate mode so that reads return as soon as a packet
280 	 * comes in, rather than waiting for the input buffer to fill
281 	 * with packets.
282 	 */
283 	if (ioctl(info->rfdesc, BIOCIMMEDIATE, &flag) == -1)
284 		fatal("Can't set immediate mode on bpf device");
285 
286 	/* make sure kernel fills in the source ethernet address */
287 	if (ioctl(info->rfdesc, BIOCSHDRCMPLT, &cmplt) == -1)
288 		fatal("Can't set header complete flag on bpf device");
289 
290 	/* Get the required BPF buffer length from the kernel. */
291 	if (ioctl(info->rfdesc, BIOCGBLEN, &sz) == -1)
292 		fatal("Can't get bpf buffer length");
293 	info->rbuf_max = sz;
294 	info->rbuf = malloc(info->rbuf_max);
295 	if (!info->rbuf)
296 		fatalx("Can't allocate %lu bytes for bpf input buffer.",
297 		    (unsigned long)info->rbuf_max);
298 	info->rbuf_offset = 0;
299 	info->rbuf_len = 0;
300 
301 	/* Set up the bpf filter program structure. */
302 	if (isserver) {
303 		p.bf_len = dhcp_bpf_sfilter_len;
304 		p.bf_insns = dhcp_bpf_sfilter;
305 	} else if (info->hw_address.htype == HTYPE_IPSEC_TUNNEL) {
306 		p.bf_len = dhcp_bpf_efilter_len;
307 		p.bf_insns = dhcp_bpf_efilter;
308 	} else {
309 		p.bf_len = dhcp_bpf_filter_len;
310 		p.bf_insns = dhcp_bpf_filter;
311 	}
312 	if (ioctl(info->rfdesc, BIOCSETF, &p) == -1)
313 		fatal("Can't install packet filter program");
314 
315 	/* Set up the bpf write filter program structure. */
316 	if (isserver) {
317 		p.bf_len = dhcp_bpf_swfilter_len;
318 		p.bf_insns = dhcp_bpf_swfilter;
319 	} else {
320 		p.bf_len = dhcp_bpf_wfilter_len;
321 		p.bf_insns = dhcp_bpf_wfilter;
322 	}
323 
324 	if (ioctl(info->rfdesc, BIOCSETWF, &p) == -1)
325 		fatal("Can't install write filter program");
326 
327 	/* make sure these settings cannot be changed after dropping privs */
328 	if (ioctl(info->rfdesc, BIOCLOCK) == -1)
329 		fatal("Failed to lock bpf descriptor");
330 }
331 
332 ssize_t
send_packet(struct interface_info * interface,struct dhcp_packet * raw,size_t len,struct packet_ctx * pc)333 send_packet(struct interface_info *interface,
334     struct dhcp_packet *raw, size_t len, struct packet_ctx *pc)
335 {
336 	unsigned char buf[256];
337 	struct iovec iov[2];
338 	ssize_t bufp;
339 	int result;
340 
341 	result = -1;
342 
343 	if (interface->hw_address.htype == HTYPE_IPSEC_TUNNEL) {
344 		socklen_t slen = pc->pc_dst.ss_len;
345 		result = sendto(server_fd, raw, len, 0,
346 		    (struct sockaddr *)&pc->pc_dst, slen);
347 		goto done;
348 	}
349 
350 	/* Assemble the headers... */
351 	if ((bufp = assemble_hw_header(buf, sizeof(buf), 0, pc,
352 	    interface->hw_address.htype)) == -1)
353 		goto done;
354 	if ((bufp = assemble_udp_ip_header(buf, sizeof(buf), bufp, pc,
355 	    (unsigned char *)raw, len)) == -1)
356 		goto done;
357 
358 	/* Fire it off */
359 	iov[0].iov_base = (char *)buf;
360 	iov[0].iov_len = bufp;
361 	iov[1].iov_base = (char *)raw;
362 	iov[1].iov_len = len;
363 
364 	result = writev(interface->wfdesc, iov, 2);
365  done:
366 	if (result == -1)
367 		log_warn("send_packet");
368 	return (result);
369 }
370 
371 ssize_t
receive_packet(struct interface_info * interface,unsigned char * buf,size_t len,struct packet_ctx * pc)372 receive_packet(struct interface_info *interface, unsigned char *buf,
373     size_t len, struct packet_ctx *pc)
374 {
375 	int length = 0;
376 	ssize_t offset = 0;
377 	struct bpf_hdr hdr;
378 
379 	/*
380 	 * All this complexity is because BPF doesn't guarantee that
381 	 * only one packet will be returned at a time.  We're getting
382 	 * what we deserve, though - this is a terrible abuse of the BPF
383 	 * interface.  Sigh.
384 	 */
385 
386 	/* Process packets until we get one we can return or until we've
387 	 * done a read and gotten nothing we can return...
388 	 */
389 	do {
390 		/* If the buffer is empty, fill it. */
391 		if (interface->rbuf_offset == interface->rbuf_len) {
392 			length = read(interface->rfdesc, interface->rbuf,
393 			    interface->rbuf_max);
394 			if (length <= 0)
395 				return (length);
396 			interface->rbuf_offset = 0;
397 			interface->rbuf_len = length;
398 		}
399 
400 		/*
401 		 * If there isn't room for a whole bpf header, something
402 		 * went wrong, but we'll ignore it and hope it goes
403 		 * away... XXX
404 		 */
405 		if (interface->rbuf_len - interface->rbuf_offset <
406 		    sizeof(hdr)) {
407 			interface->rbuf_offset = interface->rbuf_len;
408 			continue;
409 		}
410 
411 		/* Copy out a bpf header... */
412 		memcpy(&hdr, &interface->rbuf[interface->rbuf_offset],
413 		    sizeof(hdr));
414 
415 		/*
416 		 * If the bpf header plus data doesn't fit in what's
417 		 * left of the buffer, stick head in sand yet again...
418 		 */
419 		if (interface->rbuf_offset + hdr.bh_hdrlen + hdr.bh_caplen >
420 		    interface->rbuf_len) {
421 			interface->rbuf_offset = interface->rbuf_len;
422 			continue;
423 		}
424 
425 		/*
426 		 * If the captured data wasn't the whole packet, or if
427 		 * the packet won't fit in the input buffer, all we can
428 		 * do is drop it.
429 		 */
430 		if (hdr.bh_caplen != hdr.bh_datalen) {
431 			interface->rbuf_offset += hdr.bh_hdrlen =
432 			    hdr.bh_caplen;
433 			continue;
434 		}
435 
436 		/* Skip over the BPF header... */
437 		interface->rbuf_offset += hdr.bh_hdrlen;
438 
439 		/* Decode the physical header... */
440 		offset = decode_hw_header(interface->rbuf,
441 		    interface->rbuf_len, interface->rbuf_offset, pc,
442 		    interface->hw_address.htype);
443 
444 		/*
445 		 * If decoding or a physical layer checksum failed
446 		 * (dunno of any physical layer that supports this, but WTH),
447 		 * skip this packet.
448 		 */
449 		if (offset < 0) {
450 			interface->rbuf_offset += hdr.bh_caplen;
451 			continue;
452 		}
453 
454 		/* Decode the IP and UDP headers... */
455 		offset = decode_udp_ip_header(interface->rbuf,
456 		    interface->rbuf_len, offset, pc);
457 
458 		/* If the IP or UDP checksum was bad, skip the packet... */
459 		if (offset < 0) {
460 			interface->rbuf_offset += hdr.bh_caplen;
461 			continue;
462 		}
463 
464 		hdr.bh_caplen -= offset - interface->rbuf_offset;
465 		interface->rbuf_offset = offset;
466 
467 		/*
468 		 * If there's not enough room to stash the packet data,
469 		 * we have to skip it (this shouldn't happen in real
470 		 * life, though).
471 		 */
472 		if (hdr.bh_caplen > len) {
473 			interface->rbuf_offset += hdr.bh_caplen;
474 			continue;
475 		}
476 
477 		/* Copy out the data in the packet... */
478 		memcpy(buf, interface->rbuf + interface->rbuf_offset,
479 		    hdr.bh_caplen);
480 		interface->rbuf_offset += hdr.bh_caplen;
481 		return (hdr.bh_caplen);
482 	} while (!length);
483 	return (0);
484 }
485