Home | History | Annotate | Line # | Download | only in net
bpf_filter.c revision 1.51
      1 /*	$NetBSD: bpf_filter.c,v 1.51 2012/08/01 23:24:29 rmind Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * This code is derived from the Stanford/CMU enet packet filter,
      8  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
      9  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
     10  * Berkeley Laboratory.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  *	@(#)bpf_filter.c	8.1 (Berkeley) 6/10/93
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 __KERNEL_RCSID(0, "$NetBSD: bpf_filter.c,v 1.51 2012/08/01 23:24:29 rmind Exp $");
     41 
     42 #if 0
     43 #if !(defined(lint) || defined(KERNEL))
     44 static const char rcsid[] =
     45     "@(#) Header: bpf_filter.c,v 1.33 97/04/26 13:37:18 leres Exp  (LBL)";
     46 #endif
     47 #endif
     48 
     49 #include <sys/param.h>
     50 #include <sys/time.h>
     51 #include <sys/kmem.h>
     52 #include <sys/endian.h>
     53 
     54 #define EXTRACT_SHORT(p)	be16dec(p)
     55 #define EXTRACT_LONG(p)		be32dec(p)
     56 
     57 #ifdef _KERNEL
     58 #include <sys/mbuf.h>
     59 #define MINDEX(len, m, k) 		\
     60 {					\
     61 	len = m->m_len; 		\
     62 	while (k >= len) { 		\
     63 		k -= len; 		\
     64 		m = m->m_next; 		\
     65 		if (m == 0) 		\
     66 			return 0; 	\
     67 		len = m->m_len; 	\
     68 	}				\
     69 }
     70 
     71 static int m_xword (const struct mbuf *, uint32_t, int *);
     72 static int m_xhalf (const struct mbuf *, uint32_t, int *);
     73 
     74 static int
     75 m_xword(const struct mbuf *m, uint32_t k, int *err)
     76 {
     77 	int len;
     78 	u_char *cp, *np;
     79 	struct mbuf *m0;
     80 
     81 	MINDEX(len, m, k);
     82 	cp = mtod(m, u_char *) + k;
     83 	if (len >= k + 4) {
     84 		*err = 0;
     85 		return EXTRACT_LONG(cp);
     86 	}
     87 	m0 = m->m_next;
     88 	if (m0 == 0 || m0->m_len + len - k < 4) {
     89 		*err = 1;
     90 		return 0;
     91 	}
     92 	*err = 0;
     93 	np = mtod(m0, u_char *);
     94 
     95 	switch (len - k) {
     96 	case 1:
     97 		return (cp[0] << 24) | (np[0] << 16) | (np[1] << 8) | np[2];
     98 	case 2:
     99 		return (cp[0] << 24) | (cp[1] << 16) | (np[0] << 8) | np[1];
    100 	default:
    101 		return (cp[0] << 24) | (cp[1] << 16) | (cp[2] << 8) | np[0];
    102 	}
    103 }
    104 
    105 static int
    106 m_xhalf(const struct mbuf *m, uint32_t k, int *err)
    107 {
    108 	int len;
    109 	u_char *cp;
    110 	struct mbuf *m0;
    111 
    112 	MINDEX(len, m, k);
    113 	cp = mtod(m, u_char *) + k;
    114 	if (len >= k + 2) {
    115 		*err = 0;
    116 		return EXTRACT_SHORT(cp);
    117 	}
    118 	m0 = m->m_next;
    119 	if (m0 == 0) {
    120 		*err = 1;
    121 		return 0;
    122 	}
    123 	*err = 0;
    124 	return (cp[0] << 8) | mtod(m0, u_char *)[0];
    125 }
    126 #else /* _KERNEL */
    127 #include <stdlib.h>
    128 #include <string.h>
    129 #endif /* !_KERNEL */
    130 
    131 #include <net/bpf.h>
    132 
    133 /*
    134  * Execute the filter program starting at pc on the packet p
    135  * wirelen is the length of the original packet
    136  * buflen is the amount of data present
    137  */
    138 u_int
    139 bpf_filter(const struct bpf_insn *pc, const u_char *p, u_int wirelen,
    140     u_int buflen)
    141 {
    142 	uint32_t A, X, k;
    143 	uint32_t mem[BPF_MEMWORDS];
    144 
    145 	if (pc == 0) {
    146 		/*
    147 		 * No filter means accept all.
    148 		 */
    149 		return (u_int)-1;
    150 	}
    151 
    152 	memset(mem, 0, sizeof(mem));
    153 	A = 0;
    154 	X = 0;
    155 	--pc;
    156 
    157 	for (;;) {
    158 		++pc;
    159 		switch (pc->code) {
    160 
    161 		default:
    162 #ifdef _KERNEL
    163 			return 0;
    164 #else
    165 			abort();
    166 			/*NOTREACHED*/
    167 #endif
    168 		case BPF_RET|BPF_K:
    169 			return (u_int)pc->k;
    170 
    171 		case BPF_RET|BPF_A:
    172 			return (u_int)A;
    173 
    174 		case BPF_LD|BPF_W|BPF_ABS:
    175 			k = pc->k;
    176 			if (k > buflen || sizeof(int32_t) > buflen - k) {
    177 #ifdef _KERNEL
    178 				int merr;
    179 
    180 				if (buflen != 0)
    181 					return 0;
    182 				A = m_xword((const struct mbuf *)p, k, &merr);
    183 				if (merr != 0)
    184 					return 0;
    185 				continue;
    186 #else
    187 				return 0;
    188 #endif
    189 			}
    190 			A = EXTRACT_LONG(&p[k]);
    191 			continue;
    192 
    193 		case BPF_LD|BPF_H|BPF_ABS:
    194 			k = pc->k;
    195 			if (k > buflen || sizeof(int16_t) > buflen - k) {
    196 #ifdef _KERNEL
    197 				int merr;
    198 
    199 				if (buflen != 0)
    200 					return 0;
    201 				A = m_xhalf((const struct mbuf *)p, k, &merr);
    202 				if (merr != 0)
    203 					return 0;
    204 				continue;
    205 #else
    206 				return 0;
    207 #endif
    208 			}
    209 			A = EXTRACT_SHORT(&p[k]);
    210 			continue;
    211 
    212 		case BPF_LD|BPF_B|BPF_ABS:
    213 			k = pc->k;
    214 			if (k >= buflen) {
    215 #ifdef _KERNEL
    216 				const struct mbuf *m;
    217 				int len;
    218 
    219 				if (buflen != 0)
    220 					return 0;
    221 				m = (const struct mbuf *)p;
    222 				MINDEX(len, m, k);
    223 				A = mtod(m, u_char *)[k];
    224 				continue;
    225 #else
    226 				return 0;
    227 #endif
    228 			}
    229 			A = p[k];
    230 			continue;
    231 
    232 		case BPF_LD|BPF_W|BPF_LEN:
    233 			A = wirelen;
    234 			continue;
    235 
    236 		case BPF_LDX|BPF_W|BPF_LEN:
    237 			X = wirelen;
    238 			continue;
    239 
    240 		case BPF_LD|BPF_W|BPF_IND:
    241 			k = X + pc->k;
    242 			if (pc->k > buflen || X > buflen - pc->k ||
    243 			    sizeof(int32_t) > buflen - k) {
    244 #ifdef _KERNEL
    245 				int merr;
    246 
    247 				if (buflen != 0)
    248 					return 0;
    249 				A = m_xword((const struct mbuf *)p, k, &merr);
    250 				if (merr != 0)
    251 					return 0;
    252 				continue;
    253 #else
    254 				return 0;
    255 #endif
    256 			}
    257 			A = EXTRACT_LONG(&p[k]);
    258 			continue;
    259 
    260 		case BPF_LD|BPF_H|BPF_IND:
    261 			k = X + pc->k;
    262 			if (pc->k > buflen || X > buflen - pc->k ||
    263 			    sizeof(int16_t) > buflen - k) {
    264 #ifdef _KERNEL
    265 				int merr;
    266 
    267 				if (buflen != 0)
    268 					return 0;
    269 				A = m_xhalf((const struct mbuf *)p, k, &merr);
    270 				if (merr != 0)
    271 					return 0;
    272 				continue;
    273 #else
    274 				return 0;
    275 #endif
    276 			}
    277 			A = EXTRACT_SHORT(&p[k]);
    278 			continue;
    279 
    280 		case BPF_LD|BPF_B|BPF_IND:
    281 			k = X + pc->k;
    282 			if (pc->k >= buflen || X >= buflen - pc->k) {
    283 #ifdef _KERNEL
    284 				const struct mbuf *m;
    285 				int len;
    286 
    287 				if (buflen != 0)
    288 					return 0;
    289 				m = (const struct mbuf *)p;
    290 				MINDEX(len, m, k);
    291 				A = mtod(m, u_char *)[k];
    292 				continue;
    293 #else
    294 				return 0;
    295 #endif
    296 			}
    297 			A = p[k];
    298 			continue;
    299 
    300 		case BPF_LDX|BPF_MSH|BPF_B:
    301 			k = pc->k;
    302 			if (k >= buflen) {
    303 #ifdef _KERNEL
    304 				const struct mbuf *m;
    305 				int len;
    306 
    307 				if (buflen != 0)
    308 					return 0;
    309 				m = (const struct mbuf *)p;
    310 				MINDEX(len, m, k);
    311 				X = (mtod(m, char *)[k] & 0xf) << 2;
    312 				continue;
    313 #else
    314 				return 0;
    315 #endif
    316 			}
    317 			X = (p[pc->k] & 0xf) << 2;
    318 			continue;
    319 
    320 		case BPF_LD|BPF_IMM:
    321 			A = pc->k;
    322 			continue;
    323 
    324 		case BPF_LDX|BPF_IMM:
    325 			X = pc->k;
    326 			continue;
    327 
    328 		case BPF_LD|BPF_MEM:
    329 			A = mem[pc->k];
    330 			continue;
    331 
    332 		case BPF_LDX|BPF_MEM:
    333 			X = mem[pc->k];
    334 			continue;
    335 
    336 		case BPF_ST:
    337 			mem[pc->k] = A;
    338 			continue;
    339 
    340 		case BPF_STX:
    341 			mem[pc->k] = X;
    342 			continue;
    343 
    344 		case BPF_JMP|BPF_JA:
    345 			pc += pc->k;
    346 			continue;
    347 
    348 		case BPF_JMP|BPF_JGT|BPF_K:
    349 			pc += (A > pc->k) ? pc->jt : pc->jf;
    350 			continue;
    351 
    352 		case BPF_JMP|BPF_JGE|BPF_K:
    353 			pc += (A >= pc->k) ? pc->jt : pc->jf;
    354 			continue;
    355 
    356 		case BPF_JMP|BPF_JEQ|BPF_K:
    357 			pc += (A == pc->k) ? pc->jt : pc->jf;
    358 			continue;
    359 
    360 		case BPF_JMP|BPF_JSET|BPF_K:
    361 			pc += (A & pc->k) ? pc->jt : pc->jf;
    362 			continue;
    363 
    364 		case BPF_JMP|BPF_JGT|BPF_X:
    365 			pc += (A > X) ? pc->jt : pc->jf;
    366 			continue;
    367 
    368 		case BPF_JMP|BPF_JGE|BPF_X:
    369 			pc += (A >= X) ? pc->jt : pc->jf;
    370 			continue;
    371 
    372 		case BPF_JMP|BPF_JEQ|BPF_X:
    373 			pc += (A == X) ? pc->jt : pc->jf;
    374 			continue;
    375 
    376 		case BPF_JMP|BPF_JSET|BPF_X:
    377 			pc += (A & X) ? pc->jt : pc->jf;
    378 			continue;
    379 
    380 		case BPF_ALU|BPF_ADD|BPF_X:
    381 			A += X;
    382 			continue;
    383 
    384 		case BPF_ALU|BPF_SUB|BPF_X:
    385 			A -= X;
    386 			continue;
    387 
    388 		case BPF_ALU|BPF_MUL|BPF_X:
    389 			A *= X;
    390 			continue;
    391 
    392 		case BPF_ALU|BPF_DIV|BPF_X:
    393 			if (X == 0)
    394 				return 0;
    395 			A /= X;
    396 			continue;
    397 
    398 		case BPF_ALU|BPF_AND|BPF_X:
    399 			A &= X;
    400 			continue;
    401 
    402 		case BPF_ALU|BPF_OR|BPF_X:
    403 			A |= X;
    404 			continue;
    405 
    406 		case BPF_ALU|BPF_LSH|BPF_X:
    407 			A <<= X;
    408 			continue;
    409 
    410 		case BPF_ALU|BPF_RSH|BPF_X:
    411 			A >>= X;
    412 			continue;
    413 
    414 		case BPF_ALU|BPF_ADD|BPF_K:
    415 			A += pc->k;
    416 			continue;
    417 
    418 		case BPF_ALU|BPF_SUB|BPF_K:
    419 			A -= pc->k;
    420 			continue;
    421 
    422 		case BPF_ALU|BPF_MUL|BPF_K:
    423 			A *= pc->k;
    424 			continue;
    425 
    426 		case BPF_ALU|BPF_DIV|BPF_K:
    427 			A /= pc->k;
    428 			continue;
    429 
    430 		case BPF_ALU|BPF_AND|BPF_K:
    431 			A &= pc->k;
    432 			continue;
    433 
    434 		case BPF_ALU|BPF_OR|BPF_K:
    435 			A |= pc->k;
    436 			continue;
    437 
    438 		case BPF_ALU|BPF_LSH|BPF_K:
    439 			A <<= pc->k;
    440 			continue;
    441 
    442 		case BPF_ALU|BPF_RSH|BPF_K:
    443 			A >>= pc->k;
    444 			continue;
    445 
    446 		case BPF_ALU|BPF_NEG:
    447 			A = -A;
    448 			continue;
    449 
    450 		case BPF_MISC|BPF_TAX:
    451 			X = A;
    452 			continue;
    453 
    454 		case BPF_MISC|BPF_TXA:
    455 			A = X;
    456 			continue;
    457 		}
    458 	}
    459 }
    460 
    461 /*
    462  * Return true if the 'fcode' is a valid filter program.
    463  * The constraints are that each jump be forward and to a valid
    464  * code, that memory accesses are within valid ranges (to the
    465  * extent that this can be checked statically; loads of packet
    466  * data have to be, and are, also checked at run time), and that
    467  * the code terminates with either an accept or reject.
    468  *
    469  * The kernel needs to be able to verify an application's filter code.
    470  * Otherwise, a bogus program could easily crash the system.
    471  */
    472 __CTASSERT(BPF_MEMWORDS == sizeof(uint16_t) * NBBY);
    473 
    474 int
    475 bpf_validate(const struct bpf_insn *f, int signed_len)
    476 {
    477 	u_int i, from, len, ok = 0;
    478 	const struct bpf_insn *p;
    479 #if defined(KERNEL) || defined(_KERNEL)
    480 	uint16_t *mem, invalid;
    481 	size_t size;
    482 #endif
    483 
    484 	len = (u_int)signed_len;
    485 	if (len < 1)
    486 		return 0;
    487 #if defined(KERNEL) || defined(_KERNEL)
    488 	if (len > BPF_MAXINSNS)
    489 		return 0;
    490 #endif
    491 	if (BPF_CLASS(f[len - 1].code) != BPF_RET)
    492 		return 0;
    493 
    494 #if defined(KERNEL) || defined(_KERNEL)
    495 	mem = kmem_zalloc(size = sizeof(*mem) * len, KM_SLEEP);
    496 	invalid = ~0;	/* All is invalid on startup */
    497 #endif
    498 
    499 	for (i = 0; i < len; ++i) {
    500 #if defined(KERNEL) || defined(_KERNEL)
    501 		/* blend in any invalid bits for current pc */
    502 		invalid |= mem[i];
    503 #endif
    504 		p = &f[i];
    505 		switch (BPF_CLASS(p->code)) {
    506 		/*
    507 		 * Check that memory operations use valid addresses.
    508 		 */
    509 		case BPF_LD:
    510 		case BPF_LDX:
    511 			switch (BPF_MODE(p->code)) {
    512 			case BPF_MEM:
    513 				/*
    514 				 * There's no maximum packet data size
    515 				 * in userland.  The runtime packet length
    516 				 * check suffices.
    517 				 */
    518 #if defined(KERNEL) || defined(_KERNEL)
    519 				/*
    520 				 * More strict check with actual packet length
    521 				 * is done runtime.
    522 				 */
    523 				if (p->k >= BPF_MEMWORDS)
    524 					goto out;
    525 				/* check for current memory invalid */
    526 				if (invalid & (1 << p->k))
    527 					goto out;
    528 #endif
    529 				break;
    530 			case BPF_ABS:
    531 			case BPF_IND:
    532 			case BPF_MSH:
    533 			case BPF_IMM:
    534 			case BPF_LEN:
    535 				break;
    536 			default:
    537 				goto out;
    538 			}
    539 			break;
    540 		case BPF_ST:
    541 		case BPF_STX:
    542 			if (p->k >= BPF_MEMWORDS)
    543 				goto out;
    544 #if defined(KERNEL) || defined(_KERNEL)
    545 			/* validate the memory word */
    546 			invalid &= ~(1 << p->k);
    547 #endif
    548 			break;
    549 		case BPF_ALU:
    550 			switch (BPF_OP(p->code)) {
    551 			case BPF_ADD:
    552 			case BPF_SUB:
    553 			case BPF_MUL:
    554 			case BPF_OR:
    555 			case BPF_AND:
    556 			case BPF_LSH:
    557 			case BPF_RSH:
    558 			case BPF_NEG:
    559 				break;
    560 			case BPF_DIV:
    561 				/*
    562 				 * Check for constant division by 0.
    563 				 */
    564 				if (BPF_SRC(p->code) == BPF_K && p->k == 0)
    565 					goto out;
    566 				break;
    567 			default:
    568 				goto out;
    569 			}
    570 			break;
    571 		case BPF_JMP:
    572 			/*
    573 			 * Check that jumps are within the code block,
    574 			 * and that unconditional branches don't go
    575 			 * backwards as a result of an overflow.
    576 			 * Unconditional branches have a 32-bit offset,
    577 			 * so they could overflow; we check to make
    578 			 * sure they don't.  Conditional branches have
    579 			 * an 8-bit offset, and the from address is <=
    580 			 * BPF_MAXINSNS, and we assume that BPF_MAXINSNS
    581 			 * is sufficiently small that adding 255 to it
    582 			 * won't overflow.
    583 			 *
    584 			 * We know that len is <= BPF_MAXINSNS, and we
    585 			 * assume that BPF_MAXINSNS is < the maximum size
    586 			 * of a u_int, so that i + 1 doesn't overflow.
    587 			 *
    588 			 * For userland, we don't know that the from
    589 			 * or len are <= BPF_MAXINSNS, but we know that
    590 			 * from <= len, and, except on a 64-bit system,
    591 			 * it's unlikely that len, if it truly reflects
    592 			 * the size of the program we've been handed,
    593 			 * will be anywhere near the maximum size of
    594 			 * a u_int.  We also don't check for backward
    595 			 * branches, as we currently support them in
    596 			 * userland for the protochain operation.
    597 			 */
    598 			from = i + 1;
    599 			switch (BPF_OP(p->code)) {
    600 			case BPF_JA:
    601 				if (from + p->k >= len)
    602 					goto out;
    603 #if defined(KERNEL) || defined(_KERNEL)
    604 				if (from + p->k < from)
    605 					goto out;
    606 				/*
    607 				 * mark the currently invalid bits for the
    608 				 * destination
    609 				 */
    610 				mem[from + p->k] |= invalid;
    611 				invalid = 0;
    612 #endif
    613 				break;
    614 			case BPF_JEQ:
    615 			case BPF_JGT:
    616 			case BPF_JGE:
    617 			case BPF_JSET:
    618 				if (from + p->jt >= len || from + p->jf >= len)
    619 					goto out;
    620 #if defined(KERNEL) || defined(_KERNEL)
    621 				/*
    622 				 * mark the currently invalid bits for both
    623 				 * possible jump destinations
    624 				 */
    625 				mem[from + p->jt] |= invalid;
    626 				mem[from + p->jf] |= invalid;
    627 				invalid = 0;
    628 #endif
    629 				break;
    630 			default:
    631 				goto out;
    632 			}
    633 			break;
    634 		case BPF_RET:
    635 			break;
    636 		case BPF_MISC:
    637 			break;
    638 		default:
    639 			goto out;
    640 		}
    641 	}
    642 	ok = 1;
    643 out:
    644 #if defined(KERNEL) || defined(_KERNEL)
    645 	kmem_free(mem, size);
    646 #endif
    647 	return ok;
    648 }
    649