Sweden-Number/dlls/setupapi/parser.c

1910 lines
58 KiB
C
Raw Normal View History

/*
* INF file parsing
*
* Copyright 2002 Alexandre Julliard for CodeWeavers
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
*/
#include "config.h"
#include "wine/port.h"
#include <assert.h>
#include <limits.h>
#include <stdarg.h>
#include <string.h>
#include <stdlib.h>
#include "windef.h"
#include "winbase.h"
#include "wingdi.h"
#include "winuser.h"
#include "winnls.h"
2003-09-11 04:58:46 +02:00
#include "winreg.h"
#include "winternl.h"
#include "winerror.h"
#include "setupapi.h"
#include "setupapi_private.h"
2003-09-11 04:58:46 +02:00
#include "wine/unicode.h"
#include "wine/debug.h"
WINE_DEFAULT_DEBUG_CHANNEL(setupapi);
#define CONTROL_Z '\x1a'
#define MAX_SECTION_NAME_LEN 255
#define MAX_FIELD_LEN 511 /* larger fields get silently truncated */
/* actual string limit is MAX_INF_STRING_LENGTH+1 (plus terminating null) under Windows */
#define MAX_STRING_LEN (MAX_INF_STRING_LENGTH+1)
/* inf file structure definitions */
struct field
{
const WCHAR *text; /* field text */
};
struct line
{
int first_field; /* index of first field in field array */
int nb_fields; /* number of fields in line */
int key_field; /* index of field for key or -1 if no key */
};
struct section
{
const WCHAR *name; /* section name */
unsigned int nb_lines; /* number of used lines */
unsigned int alloc_lines; /* total number of allocated lines in array below */
struct line lines[16]; /* lines information (grown dynamically, 16 is initial size) */
};
struct inf_file
{
struct inf_file *next; /* next appended file */
WCHAR *strings; /* buffer for string data (section names and field values) */
WCHAR *string_pos; /* position of next available string in buffer */
unsigned int nb_sections; /* number of used sections */
unsigned int alloc_sections; /* total number of allocated section pointers */
struct section **sections; /* section pointers array */
unsigned int nb_fields;
unsigned int alloc_fields;
struct field *fields;
int strings_section; /* index of [Strings] section or -1 if none */
WCHAR *filename; /* filename of the INF */
};
/* parser definitions */
enum parser_state
{
LINE_START, /* at beginning of a line */
SECTION_NAME, /* parsing a section name */
KEY_NAME, /* parsing a key name */
VALUE_NAME, /* parsing a value name */
EOL_BACKSLASH, /* backslash at end of line */
QUOTES, /* inside quotes */
LEADING_SPACES, /* leading spaces */
TRAILING_SPACES, /* trailing spaces */
COMMENT, /* inside a comment */
NB_PARSER_STATES
};
struct parser
{
const WCHAR *start; /* start position of item being parsed */
const WCHAR *end; /* end of buffer */
struct inf_file *file; /* file being built */
enum parser_state state; /* current parser state */
enum parser_state stack[4]; /* state stack */
int stack_pos; /* current pos in stack */
int cur_section; /* index of section being parsed*/
struct line *line; /* current line */
unsigned int line_pos; /* current line position in file */
unsigned int error; /* error code */
unsigned int token_len; /* current token len */
WCHAR token[MAX_FIELD_LEN+1]; /* current token */
};
typedef const WCHAR * (*parser_state_func)( struct parser *parser, const WCHAR *pos );
/* parser state machine functions */
static const WCHAR *line_start_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *section_name_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *key_name_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *value_name_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *eol_backslash_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *quotes_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *leading_spaces_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *trailing_spaces_state( struct parser *parser, const WCHAR *pos );
static const WCHAR *comment_state( struct parser *parser, const WCHAR *pos );
static const parser_state_func parser_funcs[NB_PARSER_STATES] =
{
line_start_state, /* LINE_START */
section_name_state, /* SECTION_NAME */
key_name_state, /* KEY_NAME */
value_name_state, /* VALUE_NAME */
eol_backslash_state, /* EOL_BACKSLASH */
quotes_state, /* QUOTES */
leading_spaces_state, /* LEADING_SPACES */
trailing_spaces_state, /* TRAILING_SPACES */
comment_state /* COMMENT */
};
/* Unicode string constants */
static const WCHAR Version[] = {'V','e','r','s','i','o','n',0};
static const WCHAR Signature[] = {'S','i','g','n','a','t','u','r','e',0};
static const WCHAR Chicago[] = {'$','C','h','i','c','a','g','o','$',0};
static const WCHAR WindowsNT[] = {'$','W','i','n','d','o','w','s',' ','N','T','$',0};
static const WCHAR Windows95[] = {'$','W','i','n','d','o','w','s',' ','9','5','$',0};
static const WCHAR LayoutFile[] = {'L','a','y','o','u','t','F','i','l','e',0};
/* extend an array, allocating more memory if necessary */
static void *grow_array( void *array, unsigned int *count, size_t elem )
{
void *new_array;
unsigned int new_count = *count + *count / 2;
if (new_count < 32) new_count = 32;
2003-10-24 02:29:43 +02:00
if (array)
new_array = HeapReAlloc( GetProcessHeap(), HEAP_ZERO_MEMORY, array, new_count * elem );
else
new_array = HeapAlloc( GetProcessHeap(), HEAP_ZERO_MEMORY, new_count * elem );
if (new_array)
*count = new_count;
else
HeapFree( GetProcessHeap(), 0, array );
return new_array;
}
/* get the directory of the inf file (as counted string, not null-terminated) */
2007-08-27 17:09:06 +02:00
static const WCHAR *get_inf_dir( const struct inf_file *file, unsigned int *len )
{
const WCHAR *p = strrchrW( file->filename, '\\' );
*len = p ? (p + 1 - file->filename) : 0;
return file->filename;
}
/* find a section by name */
2007-08-27 17:09:06 +02:00
static int find_section( const struct inf_file *file, const WCHAR *name )
{
unsigned int i;
for (i = 0; i < file->nb_sections; i++)
if (!strcmpiW( name, file->sections[i]->name )) return i;
return -1;
}
/* find a line by name */
static struct line *find_line( struct inf_file *file, int section_index, const WCHAR *name )
{
struct section *section;
struct line *line;
int i;
if (section_index < 0 || section_index >= file->nb_sections) return NULL;
section = file->sections[section_index];
for (i = 0, line = section->lines; i < section->nb_lines; i++, line++)
{
if (line->key_field == -1) continue;
if (!strcmpiW( name, file->fields[line->key_field].text )) return line;
}
return NULL;
}
/* add a section to the file and return the section index */
static int add_section( struct inf_file *file, const WCHAR *name )
{
struct section *section;
if (file->nb_sections >= file->alloc_sections)
{
if (!(file->sections = grow_array( file->sections, &file->alloc_sections,
sizeof(file->sections[0]) ))) return -1;
}
if (!(section = HeapAlloc( GetProcessHeap(), 0, sizeof(*section) ))) return -1;
section->name = name;
section->nb_lines = 0;
section->alloc_lines = sizeof(section->lines)/sizeof(section->lines[0]);
file->sections[file->nb_sections] = section;
return file->nb_sections++;
}
/* add a line to a given section */
static struct line *add_line( struct inf_file *file, int section_index )
{
struct section *section;
struct line *line;
assert( section_index >= 0 && section_index < file->nb_sections );
section = file->sections[section_index];
if (section->nb_lines == section->alloc_lines) /* need to grow the section */
{
int size = sizeof(*section) - sizeof(section->lines) + 2*section->alloc_lines*sizeof(*line);
if (!(section = HeapReAlloc( GetProcessHeap(), 0, section, size ))) return NULL;
section->alloc_lines *= 2;
file->sections[section_index] = section;
}
line = &section->lines[section->nb_lines++];
line->first_field = file->nb_fields;
line->nb_fields = 0;
line->key_field = -1;
return line;
}
/* retrieve a given line from section/line index */
static inline struct line *get_line( struct inf_file *file, unsigned int section_index,
unsigned int line_index )
{
struct section *section;
if (section_index >= file->nb_sections) return NULL;
section = file->sections[section_index];
if (line_index >= section->nb_lines) return NULL;
return &section->lines[line_index];
}
/* retrieve a given field from section/line/field index */
static struct field *get_field( struct inf_file *file, int section_index, int line_index,
int field_index )
{
struct line *line = get_line( file, section_index, line_index );
if (!line) return NULL;
if (!field_index) /* get the key */
{
if (line->key_field == -1) return NULL;
return &file->fields[line->key_field];
}
field_index--;
if (field_index >= line->nb_fields) return NULL;
return &file->fields[line->first_field + field_index];
}
/* allocate a new field, growing the array if necessary */
static struct field *add_field( struct inf_file *file, const WCHAR *text )
{
struct field *field;
if (file->nb_fields >= file->alloc_fields)
{
if (!(file->fields = grow_array( file->fields, &file->alloc_fields,
sizeof(file->fields[0]) ))) return NULL;
}
field = &file->fields[file->nb_fields++];
field->text = text;
return field;
}
/* retrieve the string substitution for a directory id */
2007-08-27 17:09:06 +02:00
static const WCHAR *get_dirid_subst( const struct inf_file *file, int dirid, unsigned int *len )
{
const WCHAR *ret;
if (dirid == DIRID_SRCPATH) return get_inf_dir( file, len );
ret = DIRID_get_string( dirid );
if (ret) *len = strlenW(ret);
return ret;
}
/* retrieve the string substitution for a given string, or NULL if not found */
/* if found, len is set to the substitution length */
2007-08-27 17:09:06 +02:00
static const WCHAR *get_string_subst( const struct inf_file *file, const WCHAR *str, unsigned int *len,
BOOL no_trailing_slash )
{
static const WCHAR percent = '%';
struct section *strings_section;
struct line *line;
struct field *field;
unsigned int i;
int dirid;
WCHAR *dirid_str, *end;
const WCHAR *ret = NULL;
if (!*len) /* empty string (%%) is replaced by single percent */
{
*len = 1;
return &percent;
}
if (file->strings_section == -1) goto not_found;
strings_section = file->sections[file->strings_section];
for (i = 0, line = strings_section->lines; i < strings_section->nb_lines; i++, line++)
{
if (line->key_field == -1) continue;
if (strncmpiW( str, file->fields[line->key_field].text, *len )) continue;
if (!file->fields[line->key_field].text[*len]) break;
}
if (i == strings_section->nb_lines || !line->nb_fields) goto not_found;
field = &file->fields[line->first_field];
*len = strlenW( field->text );
return field->text;
not_found: /* check for integer id */
if ((dirid_str = HeapAlloc( GetProcessHeap(), 0, (*len+1) * sizeof(WCHAR) )))
{
memcpy( dirid_str, str, *len * sizeof(WCHAR) );
dirid_str[*len] = 0;
dirid = strtolW( dirid_str, &end, 10 );
if (!*end) ret = get_dirid_subst( file, dirid, len );
if (no_trailing_slash && ret && *len && ret[*len - 1] == '\\') *len -= 1;
HeapFree( GetProcessHeap(), 0, dirid_str );
return ret;
}
return NULL;
}
/* do string substitutions on the specified text */
/* the buffer is assumed to be large enough */
/* returns necessary length not including terminating null */
2007-08-27 17:09:06 +02:00
unsigned int PARSER_string_substW( const struct inf_file *file, const WCHAR *text, WCHAR *buffer,
unsigned int size )
{
const WCHAR *start, *subst, *p;
unsigned int len, total = 0;
int inside = 0;
if (!buffer) size = MAX_STRING_LEN + 1;
for (p = start = text; *p; p++)
{
if (*p != '%') continue;
inside = !inside;
if (inside) /* start of a %xx% string */
{
len = p - start;
if (len > size - 1) len = size - 1;
if (buffer) memcpy( buffer + total, start, len * sizeof(WCHAR) );
total += len;
size -= len;
start = p;
}
else /* end of the %xx% string, find substitution */
{
len = p - start - 1;
subst = get_string_subst( file, start + 1, &len, p[1] == '\\' );
if (!subst)
{
subst = start;
len = p - start + 1;
}
if (len > size - 1) len = size - 1;
if (buffer) memcpy( buffer + total, subst, len * sizeof(WCHAR) );
total += len;
size -= len;
start = p + 1;
}
}
if (start != p) /* unfinished string, copy it */
{
len = p - start;
if (len > size - 1) len = size - 1;
if (buffer) memcpy( buffer + total, start, len * sizeof(WCHAR) );
total += len;
}
if (buffer && size) buffer[total] = 0;
return total;
}
/* do string substitutions on the specified text */
/* the buffer is assumed to be large enough */
/* returns necessary length not including terminating null */
2007-08-27 17:09:06 +02:00
unsigned int PARSER_string_substA( const struct inf_file *file, const WCHAR *text, char *buffer,
unsigned int size )
{
WCHAR buffW[MAX_STRING_LEN+1];
DWORD ret;
unsigned int len = PARSER_string_substW( file, text, buffW, sizeof(buffW)/sizeof(WCHAR) );
if (!buffer) RtlUnicodeToMultiByteSize( &ret, buffW, len * sizeof(WCHAR) );
else
{
RtlUnicodeToMultiByteN( buffer, size-1, &ret, buffW, len * sizeof(WCHAR) );
buffer[ret] = 0;
}
return ret;
}
/* push some string data into the strings buffer */
static WCHAR *push_string( struct inf_file *file, const WCHAR *string )
{
WCHAR *ret = file->string_pos;
strcpyW( ret, string );
file->string_pos += strlenW( ret ) + 1;
return ret;
}
/* push the current state on the parser stack */
static inline void push_state( struct parser *parser, enum parser_state state )
{
assert( parser->stack_pos < sizeof(parser->stack)/sizeof(parser->stack[0]) );
parser->stack[parser->stack_pos++] = state;
}
/* pop the current state */
static inline void pop_state( struct parser *parser )
{
assert( parser->stack_pos );
parser->state = parser->stack[--parser->stack_pos];
}
/* set the parser state and return the previous one */
static inline enum parser_state set_state( struct parser *parser, enum parser_state state )
{
enum parser_state ret = parser->state;
parser->state = state;
return ret;
}
/* check if the pointer points to an end of file */
2007-08-27 17:09:06 +02:00
static inline int is_eof( const struct parser *parser, const WCHAR *ptr )
{
return (ptr >= parser->end || *ptr == CONTROL_Z);
}
/* check if the pointer points to an end of line */
2007-08-27 17:09:06 +02:00
static inline int is_eol( const struct parser *parser, const WCHAR *ptr )
{
return (ptr >= parser->end || *ptr == CONTROL_Z || *ptr == '\n');
}
/* push data from current token start up to pos into the current token */
static int push_token( struct parser *parser, const WCHAR *pos )
{
int len = pos - parser->start;
const WCHAR *src = parser->start;
WCHAR *dst = parser->token + parser->token_len;
if (len > MAX_FIELD_LEN - parser->token_len) len = MAX_FIELD_LEN - parser->token_len;
parser->token_len += len;
for ( ; len > 0; len--, dst++, src++) *dst = *src ? *src : ' ';
*dst = 0;
parser->start = pos;
return 0;
}
/* add a section with the current token as name */
static int add_section_from_token( struct parser *parser )
{
int section_index;
if (parser->token_len > MAX_SECTION_NAME_LEN)
{
parser->error = ERROR_SECTION_NAME_TOO_LONG;
return -1;
}
if ((section_index = find_section( parser->file, parser->token )) == -1)
{
/* need to create a new one */
const WCHAR *name = push_string( parser->file, parser->token );
if ((section_index = add_section( parser->file, name )) == -1)
{
parser->error = ERROR_NOT_ENOUGH_MEMORY;
return -1;
}
}
parser->token_len = 0;
parser->cur_section = section_index;
return section_index;
}
/* add a field containing the current token to the current line */
static struct field *add_field_from_token( struct parser *parser, int is_key )
{
struct field *field;
WCHAR *text;
if (!parser->line) /* need to start a new line */
{
if (parser->cur_section == -1) /* got a line before the first section */
{
parser->error = ERROR_EXPECTED_SECTION_NAME;
return NULL;
}
if (!(parser->line = add_line( parser->file, parser->cur_section ))) goto error;
}
else assert(!is_key);
text = push_string( parser->file, parser->token );
if ((field = add_field( parser->file, text )))
{
if (!is_key) parser->line->nb_fields++;
else
{
/* replace first field by key field */
parser->line->key_field = parser->line->first_field;
parser->line->first_field++;
}
parser->token_len = 0;
return field;
}
error:
parser->error = ERROR_NOT_ENOUGH_MEMORY;
return NULL;
}
/* close the current line and prepare for parsing a new one */
static void close_current_line( struct parser *parser )
{
struct line *cur_line = parser->line;
if (cur_line)
{
/* if line has a single field and no key, the field is the key too */
if (cur_line->nb_fields == 1 && cur_line->key_field == -1)
cur_line->key_field = cur_line->first_field;
}
parser->line = NULL;
}
/* handler for parser LINE_START state */
static const WCHAR *line_start_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p;
for (p = pos; !is_eof( parser, p ); p++)
{
switch(*p)
{
case '\n':
parser->line_pos++;
close_current_line( parser );
break;
case ';':
push_state( parser, LINE_START );
set_state( parser, COMMENT );
return p + 1;
case '[':
parser->start = p + 1;
set_state( parser, SECTION_NAME );
return p + 1;
default:
if (!isspaceW(*p))
{
parser->start = p;
set_state( parser, KEY_NAME );
return p;
}
break;
}
}
close_current_line( parser );
return NULL;
}
/* handler for parser SECTION_NAME state */
static const WCHAR *section_name_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p;
for (p = pos; !is_eol( parser, p ); p++)
{
if (*p == ']')
{
push_token( parser, p );
if (add_section_from_token( parser ) == -1) return NULL;
push_state( parser, LINE_START );
set_state( parser, COMMENT ); /* ignore everything else on the line */
return p + 1;
}
}
parser->error = ERROR_BAD_SECTION_NAME_LINE; /* unfinished section name */
return NULL;
}
/* handler for parser KEY_NAME state */
static const WCHAR *key_name_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p, *token_end = parser->start;
for (p = pos; !is_eol( parser, p ); p++)
{
if (*p == ',') break;
switch(*p)
{
case '=':
push_token( parser, token_end );
if (!add_field_from_token( parser, 1 )) return NULL;
parser->start = p + 1;
push_state( parser, VALUE_NAME );
set_state( parser, LEADING_SPACES );
return p + 1;
case ';':
push_token( parser, token_end );
if (!add_field_from_token( parser, 0 )) return NULL;
push_state( parser, LINE_START );
set_state( parser, COMMENT );
return p + 1;
case '"':
push_token( parser, p );
parser->start = p + 1;
push_state( parser, KEY_NAME );
set_state( parser, QUOTES );
return p + 1;
case '\\':
push_token( parser, token_end );
parser->start = p;
push_state( parser, KEY_NAME );
set_state( parser, EOL_BACKSLASH );
return p;
default:
if (!isspaceW(*p)) token_end = p + 1;
else
{
push_token( parser, p );
push_state( parser, KEY_NAME );
set_state( parser, TRAILING_SPACES );
return p;
}
break;
}
}
push_token( parser, token_end );
set_state( parser, VALUE_NAME );
return p;
}
/* handler for parser VALUE_NAME state */
static const WCHAR *value_name_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p, *token_end = parser->start;
for (p = pos; !is_eol( parser, p ); p++)
{
switch(*p)
{
case ';':
push_token( parser, token_end );
if (!add_field_from_token( parser, 0 )) return NULL;
push_state( parser, LINE_START );
set_state( parser, COMMENT );
return p + 1;
case ',':
push_token( parser, token_end );
if (!add_field_from_token( parser, 0 )) return NULL;
parser->start = p + 1;
push_state( parser, VALUE_NAME );
set_state( parser, LEADING_SPACES );
return p + 1;
case '"':
push_token( parser, p );
parser->start = p + 1;
push_state( parser, VALUE_NAME );
set_state( parser, QUOTES );
return p + 1;
case '\\':
push_token( parser, token_end );
parser->start = p;
push_state( parser, VALUE_NAME );
set_state( parser, EOL_BACKSLASH );
return p;
default:
if (!isspaceW(*p)) token_end = p + 1;
else
{
push_token( parser, p );
push_state( parser, VALUE_NAME );
set_state( parser, TRAILING_SPACES );
return p;
}
break;
}
}
push_token( parser, token_end );
if (!add_field_from_token( parser, 0 )) return NULL;
set_state( parser, LINE_START );
return p;
}
/* handler for parser EOL_BACKSLASH state */
static const WCHAR *eol_backslash_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p;
for (p = pos; !is_eof( parser, p ); p++)
{
switch(*p)
{
case '\n':
parser->line_pos++;
parser->start = p + 1;
set_state( parser, LEADING_SPACES );
return p + 1;
case '\\':
continue;
case ';':
push_state( parser, EOL_BACKSLASH );
set_state( parser, COMMENT );
return p + 1;
default:
if (isspaceW(*p)) continue;
push_token( parser, p );
pop_state( parser );
return p;
}
}
parser->start = p;
pop_state( parser );
return p;
}
/* handler for parser QUOTES state */
static const WCHAR *quotes_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p, *token_end = parser->start;
for (p = pos; !is_eol( parser, p ); p++)
{
if (*p == '"')
{
if (p+1 < parser->end && p[1] == '"') /* double quotes */
{
push_token( parser, p + 1 );
parser->start = token_end = p + 2;
p++;
}
else /* end of quotes */
{
push_token( parser, p );
parser->start = p + 1;
pop_state( parser );
return p + 1;
}
}
}
push_token( parser, p );
pop_state( parser );
return p;
}
/* handler for parser LEADING_SPACES state */
static const WCHAR *leading_spaces_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p;
for (p = pos; !is_eol( parser, p ); p++)
{
if (*p == '\\')
{
parser->start = p;
set_state( parser, EOL_BACKSLASH );
return p;
}
if (!isspaceW(*p)) break;
}
parser->start = p;
pop_state( parser );
return p;
}
/* handler for parser TRAILING_SPACES state */
static const WCHAR *trailing_spaces_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p;
for (p = pos; !is_eol( parser, p ); p++)
{
if (*p == '\\')
{
set_state( parser, EOL_BACKSLASH );
return p;
}
if (!isspaceW(*p)) break;
}
pop_state( parser );
return p;
}
/* handler for parser COMMENT state */
static const WCHAR *comment_state( struct parser *parser, const WCHAR *pos )
{
const WCHAR *p = pos;
while (!is_eol( parser, p )) p++;
pop_state( parser );
return p;
}
/* parse a complete buffer */
static DWORD parse_buffer( struct inf_file *file, const WCHAR *buffer, const WCHAR *end,
UINT *error_line )
{
static const WCHAR Strings[] = {'S','t','r','i','n','g','s',0};
struct parser parser;
const WCHAR *pos = buffer;
parser.start = buffer;
parser.end = end;
parser.file = file;
parser.line = NULL;
parser.state = LINE_START;
parser.stack_pos = 0;
parser.cur_section = -1;
parser.line_pos = 1;
parser.error = 0;
parser.token_len = 0;
/* parser main loop */
while (pos) pos = (parser_funcs[parser.state])( &parser, pos );
/* trim excess buffer space */
if (file->alloc_sections > file->nb_sections)
{
file->sections = HeapReAlloc( GetProcessHeap(), 0, file->sections,
file->nb_sections * sizeof(file->sections[0]) );
file->alloc_sections = file->nb_sections;
}
if (file->alloc_fields > file->nb_fields)
{
file->fields = HeapReAlloc( GetProcessHeap(), 0, file->fields,
file->nb_fields * sizeof(file->fields[0]) );
file->alloc_fields = file->nb_fields;
}
file->strings = HeapReAlloc( GetProcessHeap(), HEAP_REALLOC_IN_PLACE_ONLY, file->strings,
(file->string_pos - file->strings) * sizeof(WCHAR) );
if (parser.error)
{
if (error_line) *error_line = parser.line_pos;
return parser.error;
}
/* find the [strings] section */
file->strings_section = find_section( file, Strings );
return 0;
}
/* append a child INF file to its parent list, in a thread-safe manner */
static void append_inf_file( struct inf_file *parent, struct inf_file *child )
{
struct inf_file **ppnext = &parent->next;
child->next = NULL;
for (;;)
{
struct inf_file *next = InterlockedCompareExchangePointer( (void **)ppnext, child, NULL );
if (!next) return;
ppnext = &next->next;
}
}
/***********************************************************************
* parse_file
*
* parse an INF file.
*/
static struct inf_file *parse_file( HANDLE handle, const WCHAR *class, DWORD style, UINT *error_line )
{
void *buffer;
DWORD err = 0;
struct inf_file *file;
DWORD size = GetFileSize( handle, NULL );
HANDLE mapping = CreateFileMappingW( handle, NULL, PAGE_READONLY, 0, size, NULL );
if (!mapping) return NULL;
buffer = MapViewOfFile( mapping, FILE_MAP_READ, 0, 0, size );
NtClose( mapping );
if (!buffer) return NULL;
if (class) FIXME( "class %s not supported yet\n", debugstr_w(class) );
if (!(file = HeapAlloc( GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*file) )))
{
err = ERROR_NOT_ENOUGH_MEMORY;
goto done;
}
/* we won't need more strings space than the size of the file,
* so we can preallocate it here
*/
if (!(file->strings = HeapAlloc( GetProcessHeap(), 0, size * sizeof(WCHAR) )))
{
err = ERROR_NOT_ENOUGH_MEMORY;
goto done;
}
file->string_pos = file->strings;
file->strings_section = -1;
if (!RtlIsTextUnicode( buffer, size, NULL ))
{
static const BYTE utf8_bom[3] = { 0xef, 0xbb, 0xbf };
WCHAR *new_buff;
UINT codepage = CP_ACP;
UINT offset = 0;
if (size > sizeof(utf8_bom) && !memcmp( buffer, utf8_bom, sizeof(utf8_bom) ))
{
codepage = CP_UTF8;
offset = sizeof(utf8_bom);
}
if ((new_buff = HeapAlloc( GetProcessHeap(), 0, size * sizeof(WCHAR) )))
{
DWORD len = MultiByteToWideChar( codepage, 0, (char *)buffer + offset,
size - offset, new_buff, size );
err = parse_buffer( file, new_buff, new_buff + len, error_line );
HeapFree( GetProcessHeap(), 0, new_buff );
}
}
else
{
WCHAR *new_buff = (WCHAR *)buffer;
/* UCS-16 files should start with the Unicode BOM; we should skip it */
if (*new_buff == 0xfeff)
new_buff++;
err = parse_buffer( file, new_buff, (WCHAR *)((char *)buffer + size), error_line );
}
if (!err) /* now check signature */
{
int version_index = find_section( file, Version );
if (version_index != -1)
{
struct line *line = find_line( file, version_index, Signature );
if (line && line->nb_fields > 0)
{
struct field *field = file->fields + line->first_field;
if (!strcmpiW( field->text, Chicago )) goto done;
if (!strcmpiW( field->text, WindowsNT )) goto done;
if (!strcmpiW( field->text, Windows95 )) goto done;
}
}
if (error_line) *error_line = 0;
if (style & INF_STYLE_WIN4) err = ERROR_WRONG_INF_STYLE;
}
done:
UnmapViewOfFile( buffer );
if (err)
{
HeapFree( GetProcessHeap(), 0, file );
SetLastError( err );
file = NULL;
}
return file;
}
/***********************************************************************
* PARSER_get_inf_filename
*
* Retrieve the filename of an inf file.
*/
const WCHAR *PARSER_get_inf_filename( HINF hinf )
{
struct inf_file *file = hinf;
return file->filename;
}
/***********************************************************************
* PARSER_get_src_root
*
* Retrieve the source directory of an inf file.
*/
WCHAR *PARSER_get_src_root( HINF hinf )
{
unsigned int len;
const WCHAR *dir = get_inf_dir( hinf, &len );
WCHAR *ret = HeapAlloc( GetProcessHeap(), 0, (len + 1) * sizeof(WCHAR) );
if (ret)
{
memcpy( ret, dir, len * sizeof(WCHAR) );
ret[len] = 0;
}
return ret;
}
/***********************************************************************
* PARSER_get_dest_dir
*
* retrieve a destination dir of the form "dirid,relative_path" in the given entry.
* returned buffer must be freed by caller.
*/
WCHAR *PARSER_get_dest_dir( INFCONTEXT *context )
{
const WCHAR *dir;
WCHAR *ptr, *ret;
INT dirid;
unsigned int len1;
DWORD len2;
if (!SetupGetIntField( context, 1, &dirid )) return NULL;
if (!(dir = get_dirid_subst( context->Inf, dirid, &len1 ))) return NULL;
if (!SetupGetStringFieldW( context, 2, NULL, 0, &len2 )) len2 = 0;
if (!(ret = HeapAlloc( GetProcessHeap(), 0, (len1+len2+1) * sizeof(WCHAR) ))) return NULL;
memcpy( ret, dir, len1 * sizeof(WCHAR) );
ptr = ret + len1;
if (len2 && ptr > ret && ptr[-1] != '\\') *ptr++ = '\\';
if (!SetupGetStringFieldW( context, 2, ptr, len2, NULL )) *ptr = 0;
return ret;
}
/***********************************************************************
* SetupOpenInfFileA (SETUPAPI.@)
*/
HINF WINAPI SetupOpenInfFileA( PCSTR name, PCSTR class, DWORD style, UINT *error )
{
UNICODE_STRING nameW, classW;
2008-01-19 15:56:39 +01:00
HINF ret = INVALID_HANDLE_VALUE;
classW.Buffer = NULL;
if (class && !RtlCreateUnicodeStringFromAsciiz( &classW, class ))
{
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
return ret;
}
if (RtlCreateUnicodeStringFromAsciiz( &nameW, name ))
{
ret = SetupOpenInfFileW( nameW.Buffer, classW.Buffer, style, error );
RtlFreeUnicodeString( &nameW );
}
RtlFreeUnicodeString( &classW );
return ret;
}
/***********************************************************************
* SetupOpenInfFileW (SETUPAPI.@)
*/
HINF WINAPI SetupOpenInfFileW( PCWSTR name, PCWSTR class, DWORD style, UINT *error )
{
struct inf_file *file = NULL;
HANDLE handle;
WCHAR *path, *p;
UINT len;
if (strchrW( name, '\\' ) || strchrW( name, '/' ))
{
2008-01-19 15:56:39 +01:00
if (!(len = GetFullPathNameW( name, 0, NULL, NULL ))) return INVALID_HANDLE_VALUE;
if (!(path = HeapAlloc( GetProcessHeap(), 0, len * sizeof(WCHAR) )))
{
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
2008-01-19 15:56:39 +01:00
return INVALID_HANDLE_VALUE;
}
GetFullPathNameW( name, len, path, NULL );
handle = CreateFileW( path, GENERIC_READ, FILE_SHARE_READ, NULL, OPEN_EXISTING, 0, 0 );
}
else /* try Windows directory */
{
static const WCHAR Inf[] = {'\\','i','n','f','\\',0};
static const WCHAR System32[] = {'\\','s','y','s','t','e','m','3','2','\\',0};
len = GetWindowsDirectoryW( NULL, 0 ) + strlenW(name) + 12;
if (!(path = HeapAlloc( GetProcessHeap(), 0, len * sizeof(WCHAR) )))
{
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
2008-01-19 15:56:39 +01:00
return INVALID_HANDLE_VALUE;
}
GetWindowsDirectoryW( path, len );
p = path + strlenW(path);
strcpyW( p, Inf );
strcatW( p, name );
handle = CreateFileW( path, GENERIC_READ, FILE_SHARE_READ, NULL, OPEN_EXISTING, 0, 0 );
if (handle == INVALID_HANDLE_VALUE)
{
strcpyW( p, System32 );
strcatW( p, name );
handle = CreateFileW( path, GENERIC_READ, FILE_SHARE_READ, NULL, OPEN_EXISTING, 0, 0 );
}
}
if (handle != INVALID_HANDLE_VALUE)
{
file = parse_file( handle, class, style, error );
CloseHandle( handle );
}
if (!file)
{
HeapFree( GetProcessHeap(), 0, path );
2008-01-19 15:56:39 +01:00
return INVALID_HANDLE_VALUE;
}
TRACE( "%s -> %p\n", debugstr_w(path), file );
file->filename = path;
SetLastError( 0 );
return (HINF)file;
}
/***********************************************************************
* SetupOpenAppendInfFileA (SETUPAPI.@)
*/
BOOL WINAPI SetupOpenAppendInfFileA( PCSTR name, HINF parent_hinf, UINT *error )
{
HINF child_hinf;
if (!name) return SetupOpenAppendInfFileW( NULL, parent_hinf, error );
child_hinf = SetupOpenInfFileA( name, NULL, INF_STYLE_WIN4, error );
2008-01-19 15:56:39 +01:00
if (child_hinf == INVALID_HANDLE_VALUE) return FALSE;
append_inf_file( parent_hinf, child_hinf );
TRACE( "%p: appended %s (%p)\n", parent_hinf, debugstr_a(name), child_hinf );
return TRUE;
}
/***********************************************************************
* SetupOpenAppendInfFileW (SETUPAPI.@)
*/
BOOL WINAPI SetupOpenAppendInfFileW( PCWSTR name, HINF parent_hinf, UINT *error )
{
HINF child_hinf;
if (!name)
{
INFCONTEXT context;
WCHAR filename[MAX_PATH];
int idx = 1;
if (!SetupFindFirstLineW( parent_hinf, Version, LayoutFile, &context )) return FALSE;
while (SetupGetStringFieldW( &context, idx++, filename,
sizeof(filename)/sizeof(WCHAR), NULL ))
{
child_hinf = SetupOpenInfFileW( filename, NULL, INF_STYLE_WIN4, error );
2008-01-19 15:56:39 +01:00
if (child_hinf == INVALID_HANDLE_VALUE) return FALSE;
append_inf_file( parent_hinf, child_hinf );
TRACE( "%p: appended %s (%p)\n", parent_hinf, debugstr_w(filename), child_hinf );
}
return TRUE;
}
child_hinf = SetupOpenInfFileW( name, NULL, INF_STYLE_WIN4, error );
2008-01-19 15:56:39 +01:00
if (child_hinf == INVALID_HANDLE_VALUE) return FALSE;
append_inf_file( parent_hinf, child_hinf );
TRACE( "%p: appended %s (%p)\n", parent_hinf, debugstr_w(name), child_hinf );
return TRUE;
}
/***********************************************************************
* SetupOpenMasterInf (SETUPAPI.@)
*/
HINF WINAPI SetupOpenMasterInf( VOID )
{
static const WCHAR Layout[] = {'\\','i','n','f','\\', 'l', 'a', 'y', 'o', 'u', 't', '.', 'i', 'n', 'f', 0};
WCHAR Buffer[MAX_PATH];
GetWindowsDirectoryW( Buffer, MAX_PATH );
strcatW( Buffer, Layout );
return SetupOpenInfFileW( Buffer, NULL, INF_STYLE_WIN4, NULL);
}
/***********************************************************************
* SetupCloseInfFile (SETUPAPI.@)
*/
void WINAPI SetupCloseInfFile( HINF hinf )
{
struct inf_file *file = hinf;
unsigned int i;
if (!hinf || (hinf == INVALID_HANDLE_VALUE)) return;
for (i = 0; i < file->nb_sections; i++) HeapFree( GetProcessHeap(), 0, file->sections[i] );
HeapFree( GetProcessHeap(), 0, file->filename );
HeapFree( GetProcessHeap(), 0, file->sections );
HeapFree( GetProcessHeap(), 0, file->fields );
HeapFree( GetProcessHeap(), 0, file->strings );
HeapFree( GetProcessHeap(), 0, file );
}
/***********************************************************************
* SetupGetLineCountA (SETUPAPI.@)
*/
LONG WINAPI SetupGetLineCountA( HINF hinf, PCSTR name )
{
UNICODE_STRING sectionW;
LONG ret = -1;
if (!RtlCreateUnicodeStringFromAsciiz( &sectionW, name ))
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
else
{
ret = SetupGetLineCountW( hinf, sectionW.Buffer );
RtlFreeUnicodeString( &sectionW );
}
return ret;
}
/***********************************************************************
* SetupGetLineCountW (SETUPAPI.@)
*/
LONG WINAPI SetupGetLineCountW( HINF hinf, PCWSTR section )
{
struct inf_file *file = hinf;
int section_index;
LONG ret = -1;
for (file = hinf; file; file = file->next)
{
if ((section_index = find_section( file, section )) == -1) continue;
if (ret == -1) ret = 0;
ret += file->sections[section_index]->nb_lines;
}
TRACE( "(%p,%s) returning %d\n", hinf, debugstr_w(section), ret );
SetLastError( (ret == -1) ? ERROR_SECTION_NOT_FOUND : 0 );
return ret;
}
/***********************************************************************
* SetupGetLineByIndexA (SETUPAPI.@)
*/
BOOL WINAPI SetupGetLineByIndexA( HINF hinf, PCSTR section, DWORD index, INFCONTEXT *context )
{
UNICODE_STRING sectionW;
BOOL ret = FALSE;
if (!RtlCreateUnicodeStringFromAsciiz( &sectionW, section ))
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
else
{
ret = SetupGetLineByIndexW( hinf, sectionW.Buffer, index, context );
RtlFreeUnicodeString( &sectionW );
}
return ret;
}
/***********************************************************************
* SetupGetLineByIndexW (SETUPAPI.@)
*/
BOOL WINAPI SetupGetLineByIndexW( HINF hinf, PCWSTR section, DWORD index, INFCONTEXT *context )
{
struct inf_file *file = hinf;
int section_index;
for (file = hinf; file; file = file->next)
{
if ((section_index = find_section( file, section )) == -1) continue;
if (index < file->sections[section_index]->nb_lines)
{
context->Inf = hinf;
context->CurrentInf = file;
context->Section = section_index;
context->Line = index;
SetLastError( 0 );
TRACE( "(%p,%s): returning %d/%d\n",
hinf, debugstr_w(section), section_index, index );
return TRUE;
}
index -= file->sections[section_index]->nb_lines;
}
TRACE( "(%p,%s) not found\n", hinf, debugstr_w(section) );
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
/***********************************************************************
* SetupFindFirstLineA (SETUPAPI.@)
*/
BOOL WINAPI SetupFindFirstLineA( HINF hinf, PCSTR section, PCSTR key, INFCONTEXT *context )
{
UNICODE_STRING sectionW, keyW;
BOOL ret = FALSE;
if (!RtlCreateUnicodeStringFromAsciiz( &sectionW, section ))
{
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
return FALSE;
}
if (!key) ret = SetupFindFirstLineW( hinf, sectionW.Buffer, NULL, context );
else
{
if (RtlCreateUnicodeStringFromAsciiz( &keyW, key ))
{
ret = SetupFindFirstLineW( hinf, sectionW.Buffer, keyW.Buffer, context );
RtlFreeUnicodeString( &keyW );
}
else SetLastError( ERROR_NOT_ENOUGH_MEMORY );
}
RtlFreeUnicodeString( &sectionW );
return ret;
}
/***********************************************************************
* SetupFindFirstLineW (SETUPAPI.@)
*/
BOOL WINAPI SetupFindFirstLineW( HINF hinf, PCWSTR section, PCWSTR key, INFCONTEXT *context )
{
struct inf_file *file;
int section_index;
for (file = hinf; file; file = file->next)
{
if ((section_index = find_section( file, section )) == -1) continue;
if (key)
{
INFCONTEXT ctx;
ctx.Inf = hinf;
ctx.CurrentInf = file;
ctx.Section = section_index;
ctx.Line = -1;
return SetupFindNextMatchLineW( &ctx, key, context );
}
if (file->sections[section_index]->nb_lines)
{
context->Inf = hinf;
context->CurrentInf = file;
context->Section = section_index;
context->Line = 0;
SetLastError( 0 );
TRACE( "(%p,%s,%s): returning %d/0\n",
hinf, debugstr_w(section), debugstr_w(key), section_index );
return TRUE;
}
}
TRACE( "(%p,%s,%s): not found\n", hinf, debugstr_w(section), debugstr_w(key) );
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
/***********************************************************************
* SetupFindNextLine (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupFindNextLine( PINFCONTEXT context_in, PINFCONTEXT context_out )
{
struct inf_file *file = context_in->CurrentInf;
struct section *section;
if (context_in->Section >= file->nb_sections) goto error;
section = file->sections[context_in->Section];
if (context_in->Line+1 < section->nb_lines)
{
if (context_out != context_in) *context_out = *context_in;
context_out->Line++;
SetLastError( 0 );
return TRUE;
}
/* now search the appended files */
for (file = file->next; file; file = file->next)
{
int section_index = find_section( file, section->name );
if (section_index == -1) continue;
if (file->sections[section_index]->nb_lines)
{
context_out->Inf = context_in->Inf;
context_out->CurrentInf = file;
context_out->Section = section_index;
context_out->Line = 0;
SetLastError( 0 );
return TRUE;
}
}
error:
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
/***********************************************************************
* SetupFindNextMatchLineA (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupFindNextMatchLineA( PINFCONTEXT context_in, PCSTR key,
PINFCONTEXT context_out )
{
UNICODE_STRING keyW;
BOOL ret = FALSE;
if (!key) return SetupFindNextLine( context_in, context_out );
if (!RtlCreateUnicodeStringFromAsciiz( &keyW, key ))
SetLastError( ERROR_NOT_ENOUGH_MEMORY );
else
{
ret = SetupFindNextMatchLineW( context_in, keyW.Buffer, context_out );
RtlFreeUnicodeString( &keyW );
}
return ret;
}
/***********************************************************************
* SetupFindNextMatchLineW (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupFindNextMatchLineW( PINFCONTEXT context_in, PCWSTR key,
PINFCONTEXT context_out )
{
struct inf_file *file = context_in->CurrentInf;
struct section *section;
struct line *line;
unsigned int i;
if (!key) return SetupFindNextLine( context_in, context_out );
if (context_in->Section >= file->nb_sections) goto error;
section = file->sections[context_in->Section];
for (i = context_in->Line+1, line = &section->lines[i]; i < section->nb_lines; i++, line++)
{
if (line->key_field == -1) continue;
if (!strcmpiW( key, file->fields[line->key_field].text ))
{
if (context_out != context_in) *context_out = *context_in;
context_out->Line = i;
SetLastError( 0 );
TRACE( "(%p,%s,%s): returning %d\n",
file, debugstr_w(section->name), debugstr_w(key), i );
return TRUE;
}
}
/* now search the appended files */
for (file = file->next; file; file = file->next)
{
int section_index = find_section( file, section->name );
if (section_index == -1) continue;
section = file->sections[section_index];
for (i = 0, line = section->lines; i < section->nb_lines; i++, line++)
{
if (line->key_field == -1) continue;
if (!strcmpiW( key, file->fields[line->key_field].text ))
{
context_out->Inf = context_in->Inf;
context_out->CurrentInf = file;
context_out->Section = section_index;
context_out->Line = i;
SetLastError( 0 );
TRACE( "(%p,%s,%s): returning %d/%d\n",
file, debugstr_w(section->name), debugstr_w(key), section_index, i );
return TRUE;
}
}
}
TRACE( "(%p,%s,%s): not found\n",
context_in->CurrentInf, debugstr_w(section->name), debugstr_w(key) );
error:
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
/***********************************************************************
* SetupGetLineTextW (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetLineTextW( PINFCONTEXT context, HINF hinf, PCWSTR section_name,
PCWSTR key_name, PWSTR buffer, DWORD size, PDWORD required )
{
struct inf_file *file;
struct line *line;
struct field *field;
int i;
DWORD total = 0;
if (!context)
{
INFCONTEXT new_context;
if (!SetupFindFirstLineW( hinf, section_name, key_name, &new_context )) return FALSE;
file = new_context.CurrentInf;
line = get_line( file, new_context.Section, new_context.Line );
}
else
{
file = context->CurrentInf;
if (!(line = get_line( file, context->Section, context->Line )))
{
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
}
for (i = 0, field = &file->fields[line->first_field]; i < line->nb_fields; i++, field++)
total += PARSER_string_substW( file, field->text, NULL, 0 ) + 1;
if (required) *required = total;
if (buffer)
{
if (total > size)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
for (i = 0, field = &file->fields[line->first_field]; i < line->nb_fields; i++, field++)
{
unsigned int len = PARSER_string_substW( file, field->text, buffer, size );
if (i+1 < line->nb_fields) buffer[len] = ',';
buffer += len + 1;
}
}
return TRUE;
}
/***********************************************************************
* SetupGetLineTextA (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetLineTextA( PINFCONTEXT context, HINF hinf, PCSTR section_name,
PCSTR key_name, PSTR buffer, DWORD size, PDWORD required )
{
struct inf_file *file;
struct line *line;
struct field *field;
int i;
DWORD total = 0;
if (!context)
{
INFCONTEXT new_context;
if (!SetupFindFirstLineA( hinf, section_name, key_name, &new_context )) return FALSE;
file = new_context.CurrentInf;
line = get_line( file, new_context.Section, new_context.Line );
}
else
{
file = context->CurrentInf;
if (!(line = get_line( file, context->Section, context->Line )))
{
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
}
for (i = 0, field = &file->fields[line->first_field]; i < line->nb_fields; i++, field++)
total += PARSER_string_substA( file, field->text, NULL, 0 ) + 1;
if (required) *required = total;
if (buffer)
{
if (total > size)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
for (i = 0, field = &file->fields[line->first_field]; i < line->nb_fields; i++, field++)
{
unsigned int len = PARSER_string_substA( file, field->text, buffer, size );
if (i+1 < line->nb_fields) buffer[len] = ',';
buffer += len + 1;
}
}
return TRUE;
}
/***********************************************************************
* SetupGetFieldCount (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
DWORD WINAPI SetupGetFieldCount( PINFCONTEXT context )
{
struct inf_file *file = context->CurrentInf;
struct line *line = get_line( file, context->Section, context->Line );
if (!line) return 0;
return line->nb_fields;
}
/***********************************************************************
* SetupGetStringFieldA (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetStringFieldA( PINFCONTEXT context, DWORD index, PSTR buffer,
DWORD size, PDWORD required )
{
struct inf_file *file = context->CurrentInf;
struct field *field = get_field( file, context->Section, context->Line, index );
unsigned int len;
SetLastError(0);
if (!field) return FALSE;
len = PARSER_string_substA( file, field->text, NULL, 0 );
if (required) *required = len + 1;
if (buffer)
{
if (size <= len)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
PARSER_string_substA( file, field->text, buffer, size );
TRACE( "context %p/%p/%d/%d index %d returning %s\n",
context->Inf, context->CurrentInf, context->Section, context->Line,
index, debugstr_a(buffer) );
}
return TRUE;
}
/***********************************************************************
* SetupGetStringFieldW (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetStringFieldW( PINFCONTEXT context, DWORD index, PWSTR buffer,
DWORD size, PDWORD required )
{
struct inf_file *file = context->CurrentInf;
struct field *field = get_field( file, context->Section, context->Line, index );
unsigned int len;
SetLastError(0);
if (!field) return FALSE;
len = PARSER_string_substW( file, field->text, NULL, 0 );
if (required) *required = len + 1;
if (buffer)
{
if (size <= len)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
PARSER_string_substW( file, field->text, buffer, size );
TRACE( "context %p/%p/%d/%d index %d returning %s\n",
context->Inf, context->CurrentInf, context->Section, context->Line,
index, debugstr_w(buffer) );
}
return TRUE;
}
/***********************************************************************
* SetupGetIntField (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetIntField( PINFCONTEXT context, DWORD index, PINT result )
{
char localbuff[20];
char *end, *buffer = localbuff;
DWORD required;
INT res;
BOOL ret;
if (!(ret = SetupGetStringFieldA( context, index, localbuff, sizeof(localbuff), &required )))
{
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER) return FALSE;
if (!(buffer = HeapAlloc( GetProcessHeap(), 0, required ))) return FALSE;
if (!(ret = SetupGetStringFieldA( context, index, buffer, required, NULL ))) goto done;
}
/* The call to SetupGetStringFieldA succeeded. If buffer is empty we have an optional field */
if (!*buffer) *result = 0;
else
{
res = strtol( buffer, &end, 0 );
if (end != buffer && !*end) *result = res;
else
{
SetLastError( ERROR_INVALID_DATA );
ret = FALSE;
}
}
done:
if (buffer != localbuff) HeapFree( GetProcessHeap(), 0, buffer );
return ret;
}
/***********************************************************************
* SetupGetBinaryField (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetBinaryField( PINFCONTEXT context, DWORD index, BYTE *buffer,
DWORD size, LPDWORD required )
{
struct inf_file *file = context->CurrentInf;
struct line *line = get_line( file, context->Section, context->Line );
struct field *field;
int i;
if (!line)
{
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
if (!index || index > line->nb_fields)
{
SetLastError( ERROR_INVALID_PARAMETER );
return FALSE;
}
index--; /* fields start at 0 */
if (required) *required = line->nb_fields - index;
if (!buffer) return TRUE;
if (size < line->nb_fields - index)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
field = &file->fields[line->first_field + index];
for (i = index; i < line->nb_fields; i++, field++)
{
const WCHAR *p;
DWORD value = 0;
for (p = field->text; *p && isxdigitW(*p); p++)
{
2002-08-13 20:11:15 +02:00
if ((value <<= 4) > 255)
{
SetLastError( ERROR_INVALID_DATA );
return FALSE;
}
if (*p <= '9') value |= (*p - '0');
else value |= (tolowerW(*p) - 'a' + 10);
}
buffer[i - index] = value;
}
if (TRACE_ON(setupapi))
{
TRACE( "%p/%p/%d/%d index %d returning",
context->Inf, context->CurrentInf, context->Section, context->Line, index );
2003-03-15 01:12:42 +01:00
for (i = index; i < line->nb_fields; i++) TRACE( " %02x", buffer[i - index] );
TRACE( "\n" );
}
return TRUE;
}
/***********************************************************************
* SetupGetMultiSzFieldA (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetMultiSzFieldA( PINFCONTEXT context, DWORD index, PSTR buffer,
DWORD size, LPDWORD required )
{
struct inf_file *file = context->CurrentInf;
struct line *line = get_line( file, context->Section, context->Line );
struct field *field;
unsigned int len;
int i;
DWORD total = 1;
if (!line)
{
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
if (!index || index > line->nb_fields)
{
SetLastError( ERROR_INVALID_PARAMETER );
return FALSE;
}
index--; /* fields start at 0 */
field = &file->fields[line->first_field + index];
for (i = index; i < line->nb_fields; i++, field++)
{
if (!(len = PARSER_string_substA( file, field->text, NULL, 0 ))) break;
total += len + 1;
}
if (required) *required = total;
if (!buffer) return TRUE;
if (total > size)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
field = &file->fields[line->first_field + index];
for (i = index; i < line->nb_fields; i++, field++)
{
if (!(len = PARSER_string_substA( file, field->text, buffer, size ))) break;
buffer += len + 1;
}
*buffer = 0; /* add final null */
return TRUE;
}
/***********************************************************************
* SetupGetMultiSzFieldW (SETUPAPI.@)
*/
2004-01-05 22:12:21 +01:00
BOOL WINAPI SetupGetMultiSzFieldW( PINFCONTEXT context, DWORD index, PWSTR buffer,
DWORD size, LPDWORD required )
{
struct inf_file *file = context->CurrentInf;
struct line *line = get_line( file, context->Section, context->Line );
struct field *field;
unsigned int len;
int i;
DWORD total = 1;
if (!line)
{
SetLastError( ERROR_LINE_NOT_FOUND );
return FALSE;
}
if (!index || index > line->nb_fields)
{
SetLastError( ERROR_INVALID_PARAMETER );
return FALSE;
}
index--; /* fields start at 0 */
field = &file->fields[line->first_field + index];
for (i = index; i < line->nb_fields; i++, field++)
{
if (!(len = PARSER_string_substW( file, field->text, NULL, 0 ))) break;
total += len + 1;
}
if (required) *required = total;
if (!buffer) return TRUE;
if (total > size)
{
SetLastError( ERROR_INSUFFICIENT_BUFFER );
return FALSE;
}
field = &file->fields[line->first_field + index];
for (i = index; i < line->nb_fields; i++, field++)
{
if (!(len = PARSER_string_substW( file, field->text, buffer, size ))) break;
buffer += len + 1;
}
*buffer = 0; /* add final null */
return TRUE;
}
/***********************************************************************
* pSetupGetField (SETUPAPI.@)
*/
LPCWSTR WINAPI pSetupGetField( PINFCONTEXT context, DWORD index )
{
struct inf_file *file = context->CurrentInf;
struct field *field = get_field( file, context->Section, context->Line, index );
if (!field)
{
SetLastError( ERROR_INVALID_PARAMETER );
return NULL;
}
return field->text;
}