xref: /openbsd/usr.sbin/dhcrelay/bpf.c (revision 17df1aa7)
1 /*	$OpenBSD: bpf.c,v 1.7 2009/09/03 11:56:49 reyk 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 "dhcpd.h"
44 #include <sys/ioctl.h>
45 #include <sys/uio.h>
46 
47 #include <net/bpf.h>
48 #include <net/if_types.h>
49 
50 #include <netinet/in_systm.h>
51 #include <netinet/ip.h>
52 #include <netinet/udp.h>
53 #include <netinet/if_ether.h>
54 
55 #define BPF_FORMAT "/dev/bpf%d"
56 
57 /*
58  * Called by get_interface_list for each interface that's discovered.
59  * Opens a packet filter for each interface and adds it to the select
60  * mask.
61  */
62 int
63 if_register_bpf(struct interface_info *info)
64 {
65 	char filename[50];
66 	int sock, b;
67 
68 	/* Open a BPF device */
69 	for (b = 0; 1; b++) {
70 		snprintf(filename, sizeof(filename), BPF_FORMAT, b);
71 		sock = open(filename, O_RDWR, 0);
72 		if (sock == -1) {
73 			if (errno == EBUSY)
74 				continue;
75 			else
76 				error("Can't find free bpf: %m");
77 		} else
78 			break;
79 	}
80 
81 	/* Set the BPF device to point at this interface. */
82 	if (ioctl(sock, BIOCSETIF, info->ifp) == -1)
83 		error("Can't attach interface %s to bpf device %s: %m",
84 		    info->name, filename);
85 
86 	return (sock);
87 }
88 
89 void
90 if_register_send(struct interface_info *info)
91 {
92 	/*
93 	 * If we're using the bpf API for sending and receiving, we
94 	 * don't need to register this interface twice.
95 	 */
96 	info->wfdesc = info->rfdesc;
97 }
98 
99 /*
100  * Packet filter program: 'ip and udp and dst port SERVER_PORT'
101  */
102 struct bpf_insn dhcp_bpf_filter[] = {
103 	/* Make sure this is an IP packet... */
104 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
105 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
106 
107 	/* Make sure it's a UDP packet... */
108 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
109 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
110 
111 	/* Make sure this isn't a fragment... */
112 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
113 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
114 
115 	/* Get the IP header length... */
116 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
117 
118 	/* Make sure it's to the right port... */
119 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 16),
120 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
121 
122 	/* If we passed all the tests, ask for the whole packet. */
123 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
124 
125 	/* Otherwise, drop it. */
126 	BPF_STMT(BPF_RET+BPF_K, 0),
127 };
128 
129 int dhcp_bpf_filter_len = sizeof(dhcp_bpf_filter) / sizeof(struct bpf_insn);
130 
131 /*
132  * Packet filter program: encapsulated 'ip and udp and dst port SERVER_PORT'
133  */
134 struct bpf_insn dhcp_bpf_efilter[] = {
135 	/* Make sure this is an encapsulated AF_INET packet... */
136 	BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 0),
137 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, AF_INET << 24, 0, 10),
138 
139 	/* Make sure it's an IPIP packet... */
140 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 21),
141 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_IPIP, 0, 8),
142 
143 	/* Make sure it's an encapsulated UDP packet... */
144 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 41),
145 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
146 
147 	/* Make sure this isn't a fragment... */
148 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 38),
149 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
150 
151 	/* Get the IP header length... */
152 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 32),
153 
154 	/* Make sure it's to the right port... */
155 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 34),
156 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
157 
158 	/* If we passed all the tests, ask for the whole packet. */
159 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
160 
161 	/* Otherwise, drop it. */
162 	BPF_STMT(BPF_RET+BPF_K, 0),
163 };
164 
165 int dhcp_bpf_efilter_len = sizeof(dhcp_bpf_efilter) / sizeof(struct bpf_insn);
166 
167 /*
168  * Packet write filter program: 'ip and udp and src port SERVER_PORT'
169  */
170 struct bpf_insn dhcp_bpf_wfilter[] = {
171 	/* Make sure this is an IP packet... */
172 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 12),
173 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ETHERTYPE_IP, 0, 8),
174 
175 	/* Make sure it's a UDP packet... */
176 	BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 23),
177 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 0, 6),
178 
179 	/* Make sure this isn't a fragment... */
180 	BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 20),
181 	BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x1fff, 4, 0),
182 
183 	/* Get the IP header length... */
184 	BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 14),
185 
186 	/* Make sure it's from the right port... */
187 	BPF_STMT(BPF_LD + BPF_H + BPF_IND, 14),
188 	BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, SERVER_PORT, 0, 1),
189 
190 	/* If we passed all the tests, ask for the whole packet. */
191 	BPF_STMT(BPF_RET+BPF_K, (u_int)-1),
192 
193 	/* Otherwise, drop it. */
194 	BPF_STMT(BPF_RET+BPF_K, 0),
195 };
196 
197 int dhcp_bpf_wfilter_len = sizeof(dhcp_bpf_wfilter) / sizeof(struct bpf_insn);
198 
199 void
200 if_register_receive(struct interface_info *info)
201 {
202 	struct bpf_version v;
203 	struct bpf_program p;
204 	int flag = 1, sz, cmplt = 0;
205 
206 	/* Open a BPF device and hang it on this interface... */
207 	info->rfdesc = if_register_bpf(info);
208 
209 	/* Make sure the BPF version is in range... */
210 	if (ioctl(info->rfdesc, BIOCVERSION, &v) == -1)
211 		error("Can't get BPF version: %m");
212 
213 	if (v.bv_major != BPF_MAJOR_VERSION ||
214 	    v.bv_minor < BPF_MINOR_VERSION)
215 		error("Kernel BPF version out of range - recompile dhcpd!");
216 
217 	/*
218 	 * Set immediate mode so that reads return as soon as a packet
219 	 * comes in, rather than waiting for the input buffer to fill
220 	 * with packets.
221 	 */
222 	if (ioctl(info->rfdesc, BIOCIMMEDIATE, &flag) == -1)
223 		error("Can't set immediate mode on bpf device: %m");
224 
225 	/* make sure kernel fills in the source ethernet address */
226 	if (ioctl(info->rfdesc, BIOCSHDRCMPLT, &cmplt) == -1)
227 		error("Can't set header complete flag on bpf device: %m");
228 
229 	/* Get the required BPF buffer length from the kernel. */
230 	if (ioctl(info->rfdesc, BIOCGBLEN, &sz) == -1)
231 		error("Can't get bpf buffer length: %m");
232 	info->rbuf_max = sz;
233 	info->rbuf = malloc(info->rbuf_max);
234 	if (!info->rbuf)
235 		error("Can't allocate %lu bytes for bpf input buffer.",
236 		    (unsigned long)info->rbuf_max);
237 	info->rbuf_offset = 0;
238 	info->rbuf_len = 0;
239 
240 	/* Set up the bpf filter program structure. */
241 	if (info->hw_address.htype == HTYPE_IPSEC_TUNNEL) {
242 		p.bf_len = dhcp_bpf_efilter_len;
243 		p.bf_insns = dhcp_bpf_efilter;
244 	} else {
245 		p.bf_len = dhcp_bpf_filter_len;
246 		p.bf_insns = dhcp_bpf_filter;
247 	}
248 	if (ioctl(info->rfdesc, BIOCSETF, &p) == -1)
249 		error("Can't install packet filter program: %m");
250 
251 	/* Set up the bpf write filter program structure. */
252 	p.bf_len = dhcp_bpf_wfilter_len;
253 	p.bf_insns = dhcp_bpf_wfilter;
254 
255 	if (ioctl(info->rfdesc, BIOCSETWF, &p) == -1)
256 		error("Can't install write filter program: %m");
257 
258 	/* make sure these settings cannot be changed after dropping privs */
259 	if (ioctl(info->rfdesc, BIOCLOCK) == -1)
260 		error("Failed to lock bpf descriptor: %m");
261 }
262 
263 ssize_t
264 send_packet(struct interface_info *interface,
265     struct dhcp_packet *raw, size_t len, struct in_addr from,
266     struct sockaddr_in *to, struct hardware *hto)
267 {
268 	unsigned char buf[256];
269 	struct iovec iov[2];
270 	int result, bufp = 0;
271 
272 	if (interface->hw_address.htype == HTYPE_IPSEC_TUNNEL) {
273 		socklen_t slen = sizeof(*to);
274 		result = sendto(server_fd, raw, len, 0,
275 		    (struct sockaddr *)to, slen);
276 		goto done;
277 	}
278 
279 	/* Assemble the headers... */
280 	assemble_hw_header(interface, buf, &bufp, hto);
281 	assemble_udp_ip_header(interface, buf, &bufp, from.s_addr,
282 	    to->sin_addr.s_addr, to->sin_port, (unsigned char *)raw, len);
283 
284 	/* Fire it off */
285 	iov[0].iov_base = (char *)buf;
286 	iov[0].iov_len = bufp;
287 	iov[1].iov_base = (char *)raw;
288 	iov[1].iov_len = len;
289 
290 	result = writev(interface->wfdesc, iov, 2);
291  done:
292 	if (result == -1)
293 		warning("send_packet: %m");
294 	return (result);
295 }
296 
297 ssize_t
298 receive_packet(struct interface_info *interface, unsigned char *buf,
299     size_t len, struct sockaddr_in *from, struct hardware *hfrom)
300 {
301 	int length = 0, offset = 0;
302 	struct bpf_hdr hdr;
303 
304 	/*
305 	 * All this complexity is because BPF doesn't guarantee that
306 	 * only one packet will be returned at a time.  We're getting
307 	 * what we deserve, though - this is a terrible abuse of the BPF
308 	 * interface.  Sigh.
309 	 */
310 
311 	/* Process packets until we get one we can return or until we've
312 	 * done a read and gotten nothing we can return...
313 	 */
314 	do {
315 		/* If the buffer is empty, fill it. */
316 		if (interface->rbuf_offset == interface->rbuf_len) {
317 			length = read(interface->rfdesc, interface->rbuf,
318 			    interface->rbuf_max);
319 			if (length <= 0)
320 				return (length);
321 			interface->rbuf_offset = 0;
322 			interface->rbuf_len = length;
323 		}
324 
325 		/*
326 		 * If there isn't room for a whole bpf header, something
327 		 * went wrong, but we'll ignore it and hope it goes
328 		 * away... XXX
329 		 */
330 		if (interface->rbuf_len - interface->rbuf_offset <
331 		    sizeof(hdr)) {
332 			interface->rbuf_offset = interface->rbuf_len;
333 			continue;
334 		}
335 
336 		/* Copy out a bpf header... */
337 		memcpy(&hdr, &interface->rbuf[interface->rbuf_offset],
338 		    sizeof(hdr));
339 
340 		/*
341 		 * If the bpf header plus data doesn't fit in what's
342 		 * left of the buffer, stick head in sand yet again...
343 		 */
344 		if (interface->rbuf_offset + hdr.bh_hdrlen + hdr.bh_caplen >
345 		    interface->rbuf_len) {
346 			interface->rbuf_offset = interface->rbuf_len;
347 			continue;
348 		}
349 
350 		/*
351 		 * If the captured data wasn't the whole packet, or if
352 		 * the packet won't fit in the input buffer, all we can
353 		 * do is drop it.
354 		 */
355 		if (hdr.bh_caplen != hdr.bh_datalen) {
356 			interface->rbuf_offset += hdr.bh_hdrlen = hdr.bh_caplen;
357 			continue;
358 		}
359 
360 		/* Skip over the BPF header... */
361 		interface->rbuf_offset += hdr.bh_hdrlen;
362 
363 		/* Decode the physical header... */
364 		offset = decode_hw_header(interface,
365 		    interface->rbuf, interface->rbuf_offset, hfrom);
366 
367 		/*
368 		 * If a physical layer checksum failed (dunno of any
369 		 * physical layer that supports this, but WTH), skip
370 		 * this packet.
371 		 */
372 		if (offset < 0) {
373 			interface->rbuf_offset += hdr.bh_caplen;
374 			continue;
375 		}
376 		interface->rbuf_offset += offset;
377 		hdr.bh_caplen -= offset;
378 
379 		/* Decode the IP and UDP headers... */
380 		offset = decode_udp_ip_header(interface, interface->rbuf,
381 		    interface->rbuf_offset, from, NULL, hdr.bh_caplen);
382 
383 		/* If the IP or UDP checksum was bad, skip the packet... */
384 		if (offset < 0) {
385 			interface->rbuf_offset += hdr.bh_caplen;
386 			continue;
387 		}
388 		interface->rbuf_offset += offset;
389 		hdr.bh_caplen -= offset;
390 
391 		/*
392 		 * If there's not enough room to stash the packet data,
393 		 * we have to skip it (this shouldn't happen in real
394 		 * life, though).
395 		 */
396 		if (hdr.bh_caplen > len) {
397 			interface->rbuf_offset += hdr.bh_caplen;
398 			continue;
399 		}
400 
401 		/* Copy out the data in the packet... */
402 		memcpy(buf, interface->rbuf + interface->rbuf_offset,
403 		    hdr.bh_caplen);
404 		interface->rbuf_offset += hdr.bh_caplen;
405 		return (hdr.bh_caplen);
406 	} while (!length);
407 	return (0);
408 }
409