freebsd-dev/contrib/file/softmagic.c

1264 lines
28 KiB
C
Raw Normal View History

/*
* Copyright (c) Ian F. Darwin 1986-1995.
* Software written by Ian F. Darwin and others;
* maintained 1995-present by Christos Zoulas and others.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* softmagic - interpret variable magic from MAGIC
*/
#include "file.h"
#include "magic.h"
#include <string.h>
2000-11-05 08:33:55 +00:00
#include <ctype.h>
#include <stdlib.h>
#include <time.h>
#include <regex.h>
#ifndef lint
FILE_RCSID("@(#)$Id: softmagic.c,v 1.72 2004/11/24 17:38:25 christos Exp $")
#endif /* lint */
private int match(struct magic_set *, struct magic *, uint32_t,
const unsigned char *, size_t);
private int mget(struct magic_set *, union VALUETYPE *, const unsigned char *,
struct magic *, size_t);
private int mcheck(struct magic_set *, union VALUETYPE *, struct magic *);
private int32_t mprint(struct magic_set *, union VALUETYPE *, struct magic *);
private void mdebug(uint32_t, const char *, size_t);
private int mcopy(struct magic_set *, union VALUETYPE *, int, int,
const unsigned char *, size_t, size_t);
private int mconvert(struct magic_set *, union VALUETYPE *, struct magic *);
private int check_mem(struct magic_set *, unsigned int);
/*
* softmagic - lookup one file in database
* (already read from MAGIC by apprentice.c).
* Passed the name and FILE * of one file to be typed.
*/
/*ARGSUSED1*/ /* nbytes passed for regularity, maybe need later */
protected int
file_softmagic(struct magic_set *ms, const unsigned char *buf, size_t nbytes)
{
struct mlist *ml;
for (ml = ms->mlist->next; ml != ms->mlist; ml = ml->next)
if (match(ms, ml->magic, ml->nmagic, buf, nbytes))
return 1;
return 0;
}
/*
* Go through the whole list, stopping if you find a match. Process all
* the continuations of that match before returning.
*
* We support multi-level continuations:
*
* At any time when processing a successful top-level match, there is a
* current continuation level; it represents the level of the last
* successfully matched continuation.
*
* Continuations above that level are skipped as, if we see one, it
* means that the continuation that controls them - i.e, the
* lower-level continuation preceding them - failed to match.
*
* Continuations below that level are processed as, if we see one,
* it means we've finished processing or skipping higher-level
* continuations under the control of a successful or unsuccessful
* lower-level continuation, and are now seeing the next lower-level
* continuation and should process it. The current continuation
* level reverts to the level of the one we're seeing.
*
* Continuations at the current level are processed as, if we see
* one, there's no lower-level continuation that may have failed.
*
* If a continuation matches, we bump the current continuation level
* so that higher-level continuations are processed.
*/
private int
match(struct magic_set *ms, struct magic *magic, uint32_t nmagic,
const unsigned char *s, size_t nbytes)
{
uint32_t magindex = 0;
unsigned int cont_level = 0;
int need_separator = 0;
union VALUETYPE p;
int32_t oldoff = 0;
2000-11-05 08:33:55 +00:00
int returnval = 0; /* if a match is found it is set to 1*/
int firstline = 1; /* a flag to print X\n X\n- X */
if (check_mem(ms, cont_level) == -1)
return -1;
for (magindex = 0; magindex < nmagic; magindex++) {
/* if main entry matches, print it... */
int flush = !mget(ms, &p, s, &magic[magindex], nbytes);
switch (mcheck(ms, &p, &magic[magindex])) {
case -1:
return -1;
case 0:
flush++;
break;
default:
break;
}
if (flush) {
/*
* main entry didn't match,
* flush its continuations
*/
while (magindex < nmagic - 1 &&
magic[magindex + 1].cont_level != 0)
magindex++;
continue;
}
if (!firstline) { /* we found another match */
2000-11-05 08:33:55 +00:00
/* put a newline and '-' to do some simple formatting*/
if (file_printf(ms, "\n- ") == -1)
return -1;
2000-11-05 08:33:55 +00:00
}
if ((ms->c.off[cont_level] = mprint(ms, &p, &magic[magindex]))
== -1)
return -1;
/*
* If we printed something, we'll need to print
* a blank before we print something else.
*/
if (magic[magindex].desc[0])
need_separator = 1;
/* and any continuations that match */
if (check_mem(ms, ++cont_level) == -1)
return -1;
while (magic[magindex+1].cont_level != 0 &&
++magindex < nmagic) {
if (cont_level < magic[magindex].cont_level)
continue;
if (cont_level > magic[magindex].cont_level) {
/*
* We're at the end of the level
* "cont_level" continuations.
*/
cont_level = magic[magindex].cont_level;
}
if (magic[magindex].flag & OFFADD) {
oldoff=magic[magindex].offset;
magic[magindex].offset += ms->c.off[cont_level-1];
}
if (!mget(ms, &p, s, &magic[magindex], nbytes))
goto done;
switch (mcheck(ms, &p, &magic[magindex])) {
case -1:
return -1;
case 0:
break;
default:
/*
* This continuation matched.
* Print its message, with
* a blank before it if
* the previous item printed
* and this item isn't empty.
*/
/* space if previous printed */
if (need_separator
&& (magic[magindex].nospflag == 0)
&& (magic[magindex].desc[0] != '\0')) {
if (file_printf(ms, " ") == -1)
return -1;
need_separator = 0;
}
if ((ms->c.off[cont_level] = mprint(ms, &p,
&magic[magindex])) == -1)
return -1;
if (magic[magindex].desc[0])
need_separator = 1;
/*
* If we see any continuations
* at a higher level,
* process them.
*/
if (check_mem(ms, ++cont_level) == -1)
return -1;
}
done:
if (magic[magindex].flag & OFFADD) {
magic[magindex].offset = oldoff;
}
}
2000-11-05 08:33:55 +00:00
firstline = 0;
returnval = 1;
if ((ms->flags & MAGIC_CONTINUE) == 0) {
2000-11-05 08:33:55 +00:00
return 1; /* don't keep searching */
}
}
2000-11-05 08:33:55 +00:00
return returnval; /* This is hit if -k is set or there is no match */
}
private int
check_mem(struct magic_set *ms, unsigned int level)
{
size_t len;
if (level < ms->c.len)
return 0;
len = (ms->c.len += 20) * sizeof(*ms->c.off);
ms->c.off = (ms->c.off == NULL) ? malloc(len) : realloc(ms->c.off, len);
if (ms->c.off != NULL)
return 0;
file_oomem(ms);
return -1;
}
private int32_t
mprint(struct magic_set *ms, union VALUETYPE *p, struct magic *m)
{
uint32_t v;
int32_t t=0 ;
switch (m->type) {
case FILE_BYTE:
v = file_signextend(ms, m, (size_t)p->b);
if (file_printf(ms, m->desc, (unsigned char) v) == -1)
return -1;
t = m->offset + sizeof(char);
break;
case FILE_SHORT:
case FILE_BESHORT:
case FILE_LESHORT:
v = file_signextend(ms, m, (size_t)p->h);
if (file_printf(ms, m->desc, (unsigned short) v) == -1)
return -1;
t = m->offset + sizeof(short);
break;
case FILE_LONG:
case FILE_BELONG:
case FILE_LELONG:
v = file_signextend(ms, m, p->l);
if (file_printf(ms, m->desc, (uint32_t) v) == -1)
return -1;
t = m->offset + sizeof(int32_t);
break;
case FILE_STRING:
case FILE_PSTRING:
case FILE_BESTRING16:
case FILE_LESTRING16:
if (m->reln == '=') {
if (file_printf(ms, m->desc, m->value.s) == -1)
return -1;
t = m->offset + strlen(m->value.s);
}
else {
if (*m->value.s == '\0') {
char *cp = strchr(p->s,'\n');
if (cp)
*cp = '\0';
}
if (file_printf(ms, m->desc, p->s) == -1)
return -1;
t = m->offset + strlen(p->s);
}
break;
case FILE_DATE:
case FILE_BEDATE:
case FILE_LEDATE:
if (file_printf(ms, m->desc, file_fmttime(p->l, 1)) == -1)
return -1;
t = m->offset + sizeof(time_t);
break;
case FILE_LDATE:
case FILE_BELDATE:
case FILE_LELDATE:
if (file_printf(ms, m->desc, file_fmttime(p->l, 0)) == -1)
return -1;
t = m->offset + sizeof(time_t);
break;
case FILE_REGEX:
if (file_printf(ms, m->desc, p->s) == -1)
return -1;
t = m->offset + strlen(p->s);
break;
default:
file_error(ms, 0, "invalid m->type (%d) in mprint()", m->type);
return -1;
}
return(t);
}
/*
* Convert the byte order of the data we are looking at
* While we're here, let's apply the mask operation
* (unless you have a better idea)
*/
private int
mconvert(struct magic_set *ms, union VALUETYPE *p, struct magic *m)
{
switch (m->type) {
case FILE_BYTE:
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->b &= m->mask;
break;
case FILE_OPOR:
p->b |= m->mask;
break;
case FILE_OPXOR:
p->b ^= m->mask;
break;
case FILE_OPADD:
p->b += m->mask;
break;
case FILE_OPMINUS:
p->b -= m->mask;
break;
case FILE_OPMULTIPLY:
p->b *= m->mask;
break;
case FILE_OPDIVIDE:
p->b /= m->mask;
break;
case FILE_OPMODULO:
p->b %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->b = ~p->b;
return 1;
case FILE_SHORT:
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->h &= m->mask;
break;
case FILE_OPOR:
p->h |= m->mask;
break;
case FILE_OPXOR:
p->h ^= m->mask;
break;
case FILE_OPADD:
p->h += m->mask;
break;
case FILE_OPMINUS:
p->h -= m->mask;
break;
case FILE_OPMULTIPLY:
p->h *= m->mask;
break;
case FILE_OPDIVIDE:
p->h /= m->mask;
break;
case FILE_OPMODULO:
p->h %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->h = ~p->h;
return 1;
case FILE_LONG:
case FILE_DATE:
case FILE_LDATE:
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->l &= m->mask;
break;
case FILE_OPOR:
p->l |= m->mask;
break;
case FILE_OPXOR:
p->l ^= m->mask;
break;
case FILE_OPADD:
p->l += m->mask;
break;
case FILE_OPMINUS:
p->l -= m->mask;
break;
case FILE_OPMULTIPLY:
p->l *= m->mask;
break;
case FILE_OPDIVIDE:
p->l /= m->mask;
break;
case FILE_OPMODULO:
p->l %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->l = ~p->l;
return 1;
case FILE_STRING:
case FILE_BESTRING16:
case FILE_LESTRING16:
{
size_t len;
/* Null terminate and eat *trailing* return */
p->s[sizeof(p->s) - 1] = '\0';
len = strlen(p->s);
if (len-- && p->s[len] == '\n')
p->s[len] = '\0';
return 1;
}
case FILE_PSTRING:
{
char *ptr1 = p->s, *ptr2 = ptr1 + 1;
size_t len = *p->s;
if (len >= sizeof(p->s))
len = sizeof(p->s) - 1;
while (len--)
*ptr1++ = *ptr2++;
*ptr1 = '\0';
len = strlen(p->s);
if (len-- && p->s[len] == '\n')
p->s[len] = '\0';
return 1;
}
case FILE_BESHORT:
p->h = (short)((p->hs[0]<<8)|(p->hs[1]));
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->h &= m->mask;
break;
case FILE_OPOR:
p->h |= m->mask;
break;
case FILE_OPXOR:
p->h ^= m->mask;
break;
case FILE_OPADD:
p->h += m->mask;
break;
case FILE_OPMINUS:
p->h -= m->mask;
break;
case FILE_OPMULTIPLY:
p->h *= m->mask;
break;
case FILE_OPDIVIDE:
p->h /= m->mask;
break;
case FILE_OPMODULO:
p->h %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->h = ~p->h;
return 1;
case FILE_BELONG:
case FILE_BEDATE:
case FILE_BELDATE:
p->l = (int32_t)
((p->hl[0]<<24)|(p->hl[1]<<16)|(p->hl[2]<<8)|(p->hl[3]));
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->l &= m->mask;
break;
case FILE_OPOR:
p->l |= m->mask;
break;
case FILE_OPXOR:
p->l ^= m->mask;
break;
case FILE_OPADD:
p->l += m->mask;
break;
case FILE_OPMINUS:
p->l -= m->mask;
break;
case FILE_OPMULTIPLY:
p->l *= m->mask;
break;
case FILE_OPDIVIDE:
p->l /= m->mask;
break;
case FILE_OPMODULO:
p->l %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->l = ~p->l;
return 1;
case FILE_LESHORT:
p->h = (short)((p->hs[1]<<8)|(p->hs[0]));
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->h &= m->mask;
break;
case FILE_OPOR:
p->h |= m->mask;
break;
case FILE_OPXOR:
p->h ^= m->mask;
break;
case FILE_OPADD:
p->h += m->mask;
break;
case FILE_OPMINUS:
p->h -= m->mask;
break;
case FILE_OPMULTIPLY:
p->h *= m->mask;
break;
case FILE_OPDIVIDE:
p->h /= m->mask;
break;
case FILE_OPMODULO:
p->h %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->h = ~p->h;
return 1;
case FILE_LELONG:
case FILE_LEDATE:
case FILE_LELDATE:
p->l = (int32_t)
((p->hl[3]<<24)|(p->hl[2]<<16)|(p->hl[1]<<8)|(p->hl[0]));
if (m->mask)
switch (m->mask_op&0x7F) {
case FILE_OPAND:
p->l &= m->mask;
break;
case FILE_OPOR:
p->l |= m->mask;
break;
case FILE_OPXOR:
p->l ^= m->mask;
break;
case FILE_OPADD:
p->l += m->mask;
break;
case FILE_OPMINUS:
p->l -= m->mask;
break;
case FILE_OPMULTIPLY:
p->l *= m->mask;
break;
case FILE_OPDIVIDE:
p->l /= m->mask;
break;
case FILE_OPMODULO:
p->l %= m->mask;
break;
}
if (m->mask_op & FILE_OPINVERSE)
p->l = ~p->l;
return 1;
case FILE_REGEX:
return 1;
default:
file_error(ms, 0, "invalid type %d in mconvert()", m->type);
return 0;
}
}
private void
mdebug(uint32_t offset, const char *str, size_t len)
{
(void) fprintf(stderr, "mget @%d: ", offset);
file_showstr(stderr, str, len);
(void) fputc('\n', stderr);
(void) fputc('\n', stderr);
}
private int
mcopy(struct magic_set *ms, union VALUETYPE *p, int type, int indir,
const unsigned char *s, size_t offset, size_t nbytes)
{
if (type == FILE_REGEX && indir == 0) {
/*
* offset is interpreted as last line to search,
* (starting at 1), not as bytes-from start-of-file
*/
unsigned char *b, *last = NULL;
if ((p->buf = strdup((const char *)s)) == NULL) {
file_oomem(ms);
return -1;
}
for (b = (unsigned char *)p->buf; offset &&
(b = (unsigned char *)strchr((char *)b, '\n')) != NULL;
offset--, s++)
last = b;
if (last != NULL)
*last = '\0';
return 0;
1996-01-23 12:29:07 +00:00
}
if (indir == 0 && (type == FILE_BESTRING16 || type == FILE_LESTRING16))
{
const char *src = s + offset;
const char *esrc = s + nbytes;
char *dst = p->s, *edst = &p->s[sizeof(p->s) - 1];
if (type == FILE_BESTRING16)
src++;
for (;src < esrc; src++, dst++) {
if (dst < edst)
*dst = *src++;
else
break;
if (*dst == '\0')
*dst = ' ';
}
*edst = '\0';
return 0;
}
if (offset >= nbytes) {
(void)memset(p, '\0', sizeof(*p));
return 0;
}
if (nbytes - offset < sizeof(*p))
nbytes = nbytes - offset;
else
nbytes = sizeof(*p);
(void)memcpy(p, s + offset, nbytes);
/*
* the usefulness of padding with zeroes eludes me, it
* might even cause problems
*/
if (nbytes < sizeof(*p))
(void)memset(((char *)p) + nbytes, '\0', sizeof(*p) - nbytes);
return 0;
}
private int
mget(struct magic_set *ms, union VALUETYPE *p, const unsigned char *s,
struct magic *m, size_t nbytes)
{
uint32_t offset = m->offset;
if (mcopy(ms, p, m->type, m->flag & INDIR, s, offset, nbytes) == -1)
return -1;
/* Verify we have enough data to match magic type */
switch (m->type) {
case FILE_BYTE:
if (nbytes < (offset + 1)) /* should alway be true */
return 0;
break;
case FILE_SHORT:
case FILE_BESHORT:
case FILE_LESHORT:
if (nbytes < (offset + 2))
return 0;
break;
case FILE_LONG:
case FILE_BELONG:
case FILE_LELONG:
case FILE_DATE:
case FILE_BEDATE:
case FILE_LEDATE:
case FILE_LDATE:
case FILE_BELDATE:
case FILE_LELDATE:
if (nbytes < (offset + 4))
return 0;
break;
case FILE_STRING:
case FILE_PSTRING:
if (nbytes < (offset + m->vallen))
return 0;
break;
}
if ((ms->flags & MAGIC_DEBUG) != 0) {
mdebug(offset, (char *)(void *)p, sizeof(union VALUETYPE));
file_mdump(m);
}
if (m->flag & INDIR) {
switch (m->in_type) {
case FILE_BYTE:
if (m->in_offset) {
switch (m->in_op&0x7F) {
case FILE_OPAND:
offset = p->b & m->in_offset;
break;
case FILE_OPOR:
offset = p->b | m->in_offset;
break;
case FILE_OPXOR:
offset = p->b ^ m->in_offset;
break;
case FILE_OPADD:
offset = p->b + m->in_offset;
break;
case FILE_OPMINUS:
offset = p->b - m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = p->b * m->in_offset;
break;
case FILE_OPDIVIDE:
offset = p->b / m->in_offset;
break;
case FILE_OPMODULO:
offset = p->b % m->in_offset;
break;
}
} else
offset = p->b;
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
break;
case FILE_BESHORT:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) &
m->in_offset;
break;
case FILE_OPOR:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) |
m->in_offset;
break;
case FILE_OPXOR:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) ^
m->in_offset;
break;
case FILE_OPADD:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) +
m->in_offset;
break;
case FILE_OPMINUS:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) -
m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) *
m->in_offset;
break;
case FILE_OPDIVIDE:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) /
m->in_offset;
break;
case FILE_OPMODULO:
offset = (short)((p->hs[0]<<8)|
(p->hs[1])) %
m->in_offset;
break;
}
} else
offset = (short)((p->hs[0]<<8)|
(p->hs[1]));
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
2000-11-05 08:33:55 +00:00
break;
case FILE_LESHORT:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) &
m->in_offset;
break;
case FILE_OPOR:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) |
m->in_offset;
break;
case FILE_OPXOR:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) ^
m->in_offset;
break;
case FILE_OPADD:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) +
m->in_offset;
break;
case FILE_OPMINUS:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) -
m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) *
m->in_offset;
break;
case FILE_OPDIVIDE:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) /
m->in_offset;
break;
case FILE_OPMODULO:
offset = (short)((p->hs[1]<<8)|
(p->hs[0])) %
m->in_offset;
break;
}
} else
offset = (short)((p->hs[1]<<8)|
(p->hs[0]));
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
2000-11-05 08:33:55 +00:00
break;
case FILE_SHORT:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = p->h & m->in_offset;
break;
case FILE_OPOR:
offset = p->h | m->in_offset;
break;
case FILE_OPXOR:
offset = p->h ^ m->in_offset;
break;
case FILE_OPADD:
offset = p->h + m->in_offset;
break;
case FILE_OPMINUS:
offset = p->h - m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = p->h * m->in_offset;
break;
case FILE_OPDIVIDE:
offset = p->h / m->in_offset;
break;
case FILE_OPMODULO:
offset = p->h % m->in_offset;
break;
}
}
else
offset = p->h;
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
break;
case FILE_BELONG:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) &
m->in_offset;
break;
case FILE_OPOR:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) |
m->in_offset;
break;
case FILE_OPXOR:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) ^
m->in_offset;
break;
case FILE_OPADD:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) +
m->in_offset;
break;
case FILE_OPMINUS:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) -
m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) *
m->in_offset;
break;
case FILE_OPDIVIDE:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) /
m->in_offset;
break;
case FILE_OPMODULO:
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3])) %
m->in_offset;
break;
}
} else
offset = (int32_t)((p->hl[0]<<24)|
(p->hl[1]<<16)|
(p->hl[2]<<8)|
(p->hl[3]));
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
2000-11-05 08:33:55 +00:00
break;
case FILE_LELONG:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) &
m->in_offset;
break;
case FILE_OPOR:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) |
m->in_offset;
break;
case FILE_OPXOR:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) ^
m->in_offset;
break;
case FILE_OPADD:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) +
m->in_offset;
break;
case FILE_OPMINUS:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) -
m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) *
m->in_offset;
break;
case FILE_OPDIVIDE:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) /
m->in_offset;
break;
case FILE_OPMODULO:
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0])) %
m->in_offset;
break;
}
} else
offset = (int32_t)((p->hl[3]<<24)|
(p->hl[2]<<16)|
(p->hl[1]<<8)|
(p->hl[0]));
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
2000-11-05 08:33:55 +00:00
break;
case FILE_LONG:
if (m->in_offset) {
switch (m->in_op & 0x7F) {
case FILE_OPAND:
offset = p->l & m->in_offset;
break;
case FILE_OPOR:
offset = p->l | m->in_offset;
break;
case FILE_OPXOR:
offset = p->l ^ m->in_offset;
break;
case FILE_OPADD:
offset = p->l + m->in_offset;
break;
case FILE_OPMINUS:
offset = p->l - m->in_offset;
break;
case FILE_OPMULTIPLY:
offset = p->l * m->in_offset;
break;
case FILE_OPDIVIDE:
offset = p->l / m->in_offset;
break;
case FILE_OPMODULO:
offset = p->l % m->in_offset;
break;
/* case TOOMANYSWITCHBLOCKS:
* ugh = p->eye % m->strain;
* rub;
* case BEER:
* off = p->tab & m->in_gest;
* sleep;
*/
}
} else
offset = p->l;
if (m->in_op & FILE_OPINVERSE)
offset = ~offset;
break;
}
if (mcopy(ms, p, m->type, 0, s, offset, nbytes) == -1)
return -1;
if ((ms->flags & MAGIC_DEBUG) != 0) {
mdebug(offset, (char *)(void *)p,
sizeof(union VALUETYPE));
file_mdump(m);
}
}
if (!mconvert(ms, p, m))
2000-11-05 08:33:55 +00:00
return 0;
return 1;
}
private int
mcheck(struct magic_set *ms, union VALUETYPE *p, struct magic *m)
{
uint32_t l = m->value.l;
uint32_t v;
int matched;
if ( (m->value.s[0] == 'x') && (m->value.s[1] == '\0') ) {
return 1;
}
switch (m->type) {
case FILE_BYTE:
v = p->b;
break;
case FILE_SHORT:
case FILE_BESHORT:
case FILE_LESHORT:
v = p->h;
break;
case FILE_LONG:
case FILE_BELONG:
case FILE_LELONG:
case FILE_DATE:
case FILE_BEDATE:
case FILE_LEDATE:
case FILE_LDATE:
case FILE_BELDATE:
case FILE_LELDATE:
v = p->l;
break;
case FILE_STRING:
case FILE_BESTRING16:
case FILE_LESTRING16:
case FILE_PSTRING:
{
2000-11-05 08:33:55 +00:00
/*
* What we want here is:
* v = strncmp(m->value.s, p->s, m->vallen);
* but ignoring any nulls. bcmp doesn't give -/+/0
* and isn't universally available anyway.
*/
2000-11-05 08:33:55 +00:00
unsigned char *a = (unsigned char*)m->value.s;
unsigned char *b = (unsigned char*)p->s;
int len = m->vallen;
l = 0;
v = 0;
2000-11-05 08:33:55 +00:00
if (0L == m->mask) { /* normal string: do it fast */
while (--len >= 0)
if ((v = *b++ - *a++) != '\0')
2000-11-05 08:33:55 +00:00
break;
} else { /* combine the others */
while (--len >= 0) {
if ((m->mask & STRING_IGNORE_LOWERCASE) &&
islower(*a)) {
if ((v = tolower(*b++) - *a++) != '\0')
break;
} else if ((m->mask & STRING_COMPACT_BLANK) &&
isspace(*a)) {
a++;
if (isspace(*b++)) {
while (isspace(*b))
b++;
} else {
v = 1;
break;
}
} else if (isspace(*a) &&
(m->mask & STRING_COMPACT_OPTIONAL_BLANK)) {
a++;
while (isspace(*b))
b++;
} else {
if ((v = *b++ - *a++) != '\0')
break;
}
}
}
break;
2000-11-05 08:33:55 +00:00
}
case FILE_REGEX:
{
int rc;
regex_t rx;
char errmsg[512];
rc = regcomp(&rx, m->value.s, REG_EXTENDED|REG_NOSUB);
if (rc) {
free(p->buf);
regerror(rc, &rx, errmsg, sizeof(errmsg));
file_error(ms, 0, "regex error %d, (%s)", rc, errmsg);
return -1;
} else {
rc = regexec(&rx, p->buf, 0, 0, 0);
regfree(&rx);
free(p->buf);
return !rc;
}
}
default:
file_error(ms, 0, "invalid type %d in mcheck()", m->type);
return -1;
}
if (m->type != FILE_STRING && m->type != FILE_PSTRING)
v = file_signextend(ms, m, v);
switch (m->reln) {
case 'x':
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%u == *any* = 1\n", v);
matched = 1;
break;
case '!':
matched = v != l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%u != %u = %d\n",
v, l, matched);
break;
case '=':
matched = v == l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%u == %u = %d\n",
v, l, matched);
break;
case '>':
if (m->flag & UNSIGNED) {
matched = v > l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%u > %u = %d\n",
v, l, matched);
}
else {
matched = (int32_t) v > (int32_t) l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%d > %d = %d\n",
v, l, matched);
}
break;
case '<':
if (m->flag & UNSIGNED) {
matched = v < l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%u < %u = %d\n",
v, l, matched);
}
else {
matched = (int32_t) v < (int32_t) l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "%d < %d = %d\n",
v, l, matched);
}
break;
case '&':
matched = (v & l) == l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "((%x & %x) == %x) = %d\n",
v, l, l, matched);
break;
case '^':
matched = (v & l) != l;
if ((ms->flags & MAGIC_DEBUG) != 0)
(void) fprintf(stderr, "((%x & %x) != %x) = %d\n",
v, l, l, matched);
break;
default:
matched = 0;
file_error(ms, 0, "cannot happen: invalid relation `%c'",
m->reln);
return -1;
}
return matched;
}