Permalink
Fetching contributors…
Cannot retrieve contributors at this time
360 lines (331 sloc) 9.48 KB
/***************************************************************************
* _ _ ____ _
* Project ___| | | | _ \| |
* / __| | | | |_) | |
* | (__| |_| | _ <| |___
* \___|\___/|_| \_\_____|
*
* Copyright (C) 1998 - 2018, Daniel Stenberg, <daniel@haxx.se>, et al.
*
* This software is licensed as described in the file COPYING, which
* you should have received as part of this distribution. The terms
* are also available at https://curl.haxx.se/docs/copyright.html.
*
* You may opt to use, copy, modify, merge, publish, distribute and/or sell
* copies of the Software, and permit persons to whom the Software is
* furnished to do so, under the terms of the COPYING file.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
***************************************************************************/
#include "curl_setup.h"
#include <curl/curl.h>
#include "curl_fnmatch.h"
#include "curl_memory.h"
/* The last #include file should be: */
#include "memdebug.h"
#define CURLFNM_CHARSET_LEN (sizeof(char) * 256)
#define CURLFNM_CHSET_SIZE (CURLFNM_CHARSET_LEN + 15)
#define CURLFNM_NEGATE CURLFNM_CHARSET_LEN
#define CURLFNM_ALNUM (CURLFNM_CHARSET_LEN + 1)
#define CURLFNM_DIGIT (CURLFNM_CHARSET_LEN + 2)
#define CURLFNM_XDIGIT (CURLFNM_CHARSET_LEN + 3)
#define CURLFNM_ALPHA (CURLFNM_CHARSET_LEN + 4)
#define CURLFNM_PRINT (CURLFNM_CHARSET_LEN + 5)
#define CURLFNM_BLANK (CURLFNM_CHARSET_LEN + 6)
#define CURLFNM_LOWER (CURLFNM_CHARSET_LEN + 7)
#define CURLFNM_GRAPH (CURLFNM_CHARSET_LEN + 8)
#define CURLFNM_SPACE (CURLFNM_CHARSET_LEN + 9)
#define CURLFNM_UPPER (CURLFNM_CHARSET_LEN + 10)
typedef enum {
CURLFNM_SCHS_DEFAULT = 0,
CURLFNM_SCHS_RIGHTBR,
CURLFNM_SCHS_RIGHTBRLEFTBR
} setcharset_state;
typedef enum {
CURLFNM_PKW_INIT = 0,
CURLFNM_PKW_DDOT
} parsekey_state;
typedef enum {
CCLASS_OTHER = 0,
CCLASS_DIGIT,
CCLASS_UPPER,
CCLASS_LOWER
} char_class;
#define SETCHARSET_OK 1
#define SETCHARSET_FAIL 0
static int parsekeyword(unsigned char **pattern, unsigned char *charset)
{
parsekey_state state = CURLFNM_PKW_INIT;
#define KEYLEN 10
char keyword[KEYLEN] = { 0 };
int found = FALSE;
int i;
unsigned char *p = *pattern;
for(i = 0; !found; i++) {
char c = *p++;
if(i >= KEYLEN)
return SETCHARSET_FAIL;
switch(state) {
case CURLFNM_PKW_INIT:
if(ISLOWER(c))
keyword[i] = c;
else if(c == ':')
state = CURLFNM_PKW_DDOT;
else
return SETCHARSET_FAIL;
break;
case CURLFNM_PKW_DDOT:
if(c == ']')
found = TRUE;
else
return SETCHARSET_FAIL;
}
}
#undef KEYLEN
*pattern = p; /* move caller's pattern pointer */
if(strcmp(keyword, "digit") == 0)
charset[CURLFNM_DIGIT] = 1;
else if(strcmp(keyword, "alnum") == 0)
charset[CURLFNM_ALNUM] = 1;
else if(strcmp(keyword, "alpha") == 0)
charset[CURLFNM_ALPHA] = 1;
else if(strcmp(keyword, "xdigit") == 0)
charset[CURLFNM_XDIGIT] = 1;
else if(strcmp(keyword, "print") == 0)
charset[CURLFNM_PRINT] = 1;
else if(strcmp(keyword, "graph") == 0)
charset[CURLFNM_GRAPH] = 1;
else if(strcmp(keyword, "space") == 0)
charset[CURLFNM_SPACE] = 1;
else if(strcmp(keyword, "blank") == 0)
charset[CURLFNM_BLANK] = 1;
else if(strcmp(keyword, "upper") == 0)
charset[CURLFNM_UPPER] = 1;
else if(strcmp(keyword, "lower") == 0)
charset[CURLFNM_LOWER] = 1;
else
return SETCHARSET_FAIL;
return SETCHARSET_OK;
}
/* Return the character class. */
static char_class charclass(unsigned char c)
{
if(ISUPPER(c))
return CCLASS_UPPER;
if(ISLOWER(c))
return CCLASS_LOWER;
if(ISDIGIT(c))
return CCLASS_DIGIT;
return CCLASS_OTHER;
}
/* Include a character or a range in set. */
static void setcharorrange(unsigned char **pp, unsigned char *charset)
{
unsigned char *p = (*pp)++;
unsigned char c = *p++;
charset[c] = 1;
if(ISALNUM(c) && *p++ == '-') {
char_class cc = charclass(c);
unsigned char endrange = *p++;
if(endrange == '\\')
endrange = *p++;
if(endrange >= c && charclass(endrange) == cc) {
while(c++ != endrange)
if(charclass(c) == cc) /* Chars in class may be not consecutive. */
charset[c] = 1;
*pp = p;
}
}
}
/* returns 1 (true) if pattern is OK, 0 if is bad ("p" is pattern pointer) */
static int setcharset(unsigned char **p, unsigned char *charset)
{
setcharset_state state = CURLFNM_SCHS_DEFAULT;
bool something_found = FALSE;
unsigned char c;
memset(charset, 0, CURLFNM_CHSET_SIZE);
for(;;) {
c = **p;
if(!c)
return SETCHARSET_FAIL;
switch(state) {
case CURLFNM_SCHS_DEFAULT:
if(c == ']') {
if(something_found)
return SETCHARSET_OK;
something_found = TRUE;
state = CURLFNM_SCHS_RIGHTBR;
charset[c] = 1;
(*p)++;
}
else if(c == '[') {
unsigned char *pp = *p + 1;
if(*pp++ == ':' && parsekeyword(&pp, charset))
*p = pp;
else {
charset[c] = 1;
(*p)++;
}
something_found = TRUE;
}
else if(c == '^' || c == '!') {
if(!something_found) {
if(charset[CURLFNM_NEGATE]) {
charset[c] = 1;
something_found = TRUE;
}
else
charset[CURLFNM_NEGATE] = 1; /* negate charset */
}
else
charset[c] = 1;
(*p)++;
}
else if(c == '\\') {
c = *(++(*p));
if(c)
setcharorrange(p, charset);
else
charset['\\'] = 1;
something_found = TRUE;
}
else {
setcharorrange(p, charset);
something_found = TRUE;
}
break;
case CURLFNM_SCHS_RIGHTBR:
if(c == '[') {
state = CURLFNM_SCHS_RIGHTBRLEFTBR;
charset[c] = 1;
(*p)++;
}
else if(c == ']') {
return SETCHARSET_OK;
}
else if(ISPRINT(c)) {
charset[c] = 1;
(*p)++;
state = CURLFNM_SCHS_DEFAULT;
}
else
/* used 'goto fail' instead of 'return SETCHARSET_FAIL' to avoid a
* nonsense warning 'statement not reached' at end of the fnc when
* compiling on Solaris */
goto fail;
break;
case CURLFNM_SCHS_RIGHTBRLEFTBR:
if(c == ']')
return SETCHARSET_OK;
state = CURLFNM_SCHS_DEFAULT;
charset[c] = 1;
(*p)++;
break;
}
}
fail:
return SETCHARSET_FAIL;
}
static int loop(const unsigned char *pattern, const unsigned char *string,
int maxstars)
{
unsigned char *p = (unsigned char *)pattern;
unsigned char *s = (unsigned char *)string;
unsigned char charset[CURLFNM_CHSET_SIZE] = { 0 };
for(;;) {
unsigned char *pp;
switch(*p) {
case '*':
if(!maxstars)
return CURL_FNMATCH_NOMATCH;
/* Regroup consecutive stars and question marks. This can be done because
'*?*?*' can be expressed as '??*'. */
for(;;) {
if(*++p == '\0')
return CURL_FNMATCH_MATCH;
if(*p == '?') {
if(!*s++)
return CURL_FNMATCH_NOMATCH;
}
else if(*p != '*')
break;
}
/* Skip string characters until we find a match with pattern suffix. */
for(maxstars--; *s; s++) {
if(loop(p, s, maxstars) == CURL_FNMATCH_MATCH)
return CURL_FNMATCH_MATCH;
}
return CURL_FNMATCH_NOMATCH;
case '?':
if(!*s)
return CURL_FNMATCH_NOMATCH;
s++;
p++;
break;
case '\0':
return *s? CURL_FNMATCH_NOMATCH: CURL_FNMATCH_MATCH;
case '\\':
if(p[1])
p++;
if(*s++ != *p++)
return CURL_FNMATCH_NOMATCH;
break;
case '[':
pp = p + 1; /* Copy in case of syntax error in set. */
if(setcharset(&pp, charset)) {
int found = FALSE;
if(!*s)
return CURL_FNMATCH_NOMATCH;
if(charset[(unsigned int)*s])
found = TRUE;
else if(charset[CURLFNM_ALNUM])
found = ISALNUM(*s);
else if(charset[CURLFNM_ALPHA])
found = ISALPHA(*s);
else if(charset[CURLFNM_DIGIT])
found = ISDIGIT(*s);
else if(charset[CURLFNM_XDIGIT])
found = ISXDIGIT(*s);
else if(charset[CURLFNM_PRINT])
found = ISPRINT(*s);
else if(charset[CURLFNM_SPACE])
found = ISSPACE(*s);
else if(charset[CURLFNM_UPPER])
found = ISUPPER(*s);
else if(charset[CURLFNM_LOWER])
found = ISLOWER(*s);
else if(charset[CURLFNM_BLANK])
found = ISBLANK(*s);
else if(charset[CURLFNM_GRAPH])
found = ISGRAPH(*s);
if(charset[CURLFNM_NEGATE])
found = !found;
if(!found)
return CURL_FNMATCH_NOMATCH;
p = pp + 1;
s++;
break;
}
/* Syntax error in set: this must be taken as a regular character. */
/* FALLTHROUGH */
default:
if(*p++ != *s++)
return CURL_FNMATCH_NOMATCH;
break;
}
}
}
/*
* @unittest: 1307
*/
int Curl_fnmatch(void *ptr, const char *pattern, const char *string)
{
(void)ptr; /* the argument is specified by the curl_fnmatch_callback
prototype, but not used by Curl_fnmatch() */
if(!pattern || !string) {
return CURL_FNMATCH_FAIL;
}
return loop((unsigned char *)pattern, (unsigned char *)string, 5);
}