19ac4b08f0
LOCALEDIR. * addr2line.c (main): Call setlocale, bindtextdomain, textdomain. * ar.c (main): Likewise. * coffdump.c (main): Likewise. * dlltool.c (main): Likewise. * nlmconv.c (main): Likewise. * nm.c (main): Likewise. * objcopy.c (main): Likewise. * objdump.c (main): Likewise. * size.c (main): Likewise. * srconv.c (main): Likewise. * strings.c (main): Likewise. * sysdump.c (main): Likewise. * windres.c (main): Likewise.
513 lines
11 KiB
C
513 lines
11 KiB
C
/* strings -- print the strings of printable characters in files
|
||
Copyright (C) 1993, 94, 95, 96, 97, 1998 Free Software Foundation, Inc.
|
||
|
||
This program is free software; you can redistribute it and/or modify
|
||
it under the terms of the GNU General Public License as published by
|
||
the Free Software Foundation; either version 2, or (at your option)
|
||
any later version.
|
||
|
||
This program 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 General Public License for more details.
|
||
|
||
You should have received a copy of the GNU General Public License
|
||
along with this program; if not, write to the Free Software
|
||
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
|
||
02111-1307, USA. */
|
||
|
||
/* Usage: strings [options] file...
|
||
|
||
Options:
|
||
--all
|
||
-a
|
||
- Do not scan only the initialized data section of object files.
|
||
|
||
--print-file-name
|
||
-f Print the name of the file before each string.
|
||
|
||
--bytes=min-len
|
||
-n min-len
|
||
-min-len Print graphic char sequences, MIN-LEN or more bytes long,
|
||
that are followed by a NUL or a newline. Default is 4.
|
||
|
||
--radix={o,x,d}
|
||
-t {o,x,d} Print the offset within the file before each string,
|
||
in octal/hex/decimal.
|
||
|
||
-o Like -to. (Some other implementations have -o like -to,
|
||
others like -td. We chose one arbitrarily.)
|
||
|
||
--target=BFDNAME
|
||
Specify a non-default object file format.
|
||
|
||
--help
|
||
-h Print the usage message on the standard output.
|
||
|
||
--version
|
||
-v Print the program version number.
|
||
|
||
Written by Richard Stallman <rms@gnu.ai.mit.edu>
|
||
and David MacKenzie <djm@gnu.ai.mit.edu>. */
|
||
|
||
#include "bfd.h"
|
||
#include <stdio.h>
|
||
#include <getopt.h>
|
||
#include <ctype.h>
|
||
#include <errno.h>
|
||
#include "bucomm.h"
|
||
#include "libiberty.h"
|
||
|
||
#ifdef isascii
|
||
#define isgraphic(c) (isascii (c) && isprint (c))
|
||
#else
|
||
#define isgraphic(c) (isprint (c))
|
||
#endif
|
||
|
||
#ifndef errno
|
||
extern int errno;
|
||
#endif
|
||
|
||
/* The BFD section flags that identify an initialized data section. */
|
||
#define DATA_FLAGS (SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS)
|
||
|
||
/* Radix for printing addresses (must be 8, 10 or 16). */
|
||
static int address_radix;
|
||
|
||
/* Minimum length of sequence of graphic chars to trigger output. */
|
||
static int string_min;
|
||
|
||
/* true means print address within file for each string. */
|
||
static boolean print_addresses;
|
||
|
||
/* true means print filename for each string. */
|
||
static boolean print_filenames;
|
||
|
||
/* true means for object files scan only the data section. */
|
||
static boolean datasection_only;
|
||
|
||
/* true if we found an initialized data section in the current file. */
|
||
static boolean got_a_section;
|
||
|
||
/* The BFD object file format. */
|
||
static char *target;
|
||
|
||
static struct option long_options[] =
|
||
{
|
||
{"all", no_argument, NULL, 'a'},
|
||
{"print-file-name", no_argument, NULL, 'f'},
|
||
{"bytes", required_argument, NULL, 'n'},
|
||
{"radix", required_argument, NULL, 't'},
|
||
{"target", required_argument, NULL, 'T'},
|
||
{"help", no_argument, NULL, 'h'},
|
||
{"version", no_argument, NULL, 'v'},
|
||
{NULL, 0, NULL, 0}
|
||
};
|
||
|
||
static void strings_a_section PARAMS ((bfd *, asection *, PTR));
|
||
static boolean strings_object_file PARAMS ((const char *));
|
||
static boolean strings_file PARAMS ((char *file));
|
||
static int integer_arg PARAMS ((char *s));
|
||
static void print_strings PARAMS ((const char *filename, FILE *stream,
|
||
file_ptr address, int stop_point,
|
||
int magiccount, char *magic));
|
||
static void usage PARAMS ((FILE *stream, int status));
|
||
|
||
int
|
||
main (argc, argv)
|
||
int argc;
|
||
char **argv;
|
||
{
|
||
int optc;
|
||
int exit_status = 0;
|
||
boolean files_given = false;
|
||
|
||
setlocale (LC_MESSAGES, "");
|
||
bindtextdomain (PACKAGE, LOCALEDIR);
|
||
textdomain (PACKAGE);
|
||
|
||
program_name = argv[0];
|
||
xmalloc_set_program_name (program_name);
|
||
string_min = -1;
|
||
print_addresses = false;
|
||
print_filenames = false;
|
||
datasection_only = true;
|
||
target = NULL;
|
||
|
||
while ((optc = getopt_long (argc, argv, "afn:ot:v0123456789",
|
||
long_options, (int *) 0)) != EOF)
|
||
{
|
||
switch (optc)
|
||
{
|
||
case 'a':
|
||
datasection_only = false;
|
||
break;
|
||
|
||
case 'f':
|
||
print_filenames = true;
|
||
break;
|
||
|
||
case 'h':
|
||
usage (stdout, 0);
|
||
|
||
case 'n':
|
||
string_min = integer_arg (optarg);
|
||
if (string_min < 1)
|
||
{
|
||
fprintf (stderr, _("%s: invalid number %s\n"),
|
||
program_name, optarg);
|
||
exit (1);
|
||
}
|
||
break;
|
||
|
||
case 'o':
|
||
print_addresses = true;
|
||
address_radix = 8;
|
||
break;
|
||
|
||
case 't':
|
||
print_addresses = true;
|
||
if (optarg[1] != '\0')
|
||
usage (stderr, 1);
|
||
switch (optarg[0])
|
||
{
|
||
case 'o':
|
||
address_radix = 8;
|
||
break;
|
||
|
||
case 'd':
|
||
address_radix = 10;
|
||
break;
|
||
|
||
case 'x':
|
||
address_radix = 16;
|
||
break;
|
||
|
||
default:
|
||
usage (stderr, 1);
|
||
}
|
||
break;
|
||
|
||
case 'T':
|
||
target = optarg;
|
||
break;
|
||
|
||
case 'v':
|
||
print_version ("strings");
|
||
break;
|
||
|
||
case '?':
|
||
usage (stderr, 1);
|
||
|
||
default:
|
||
if (string_min < 0)
|
||
string_min = optc;
|
||
else
|
||
string_min = string_min * 10 + optc - '0';
|
||
break;
|
||
}
|
||
}
|
||
|
||
if (string_min < 0)
|
||
string_min = 4;
|
||
|
||
bfd_init ();
|
||
set_default_bfd_target ();
|
||
|
||
if (optind >= argc)
|
||
{
|
||
datasection_only = false;
|
||
print_strings ("{standard input}", stdin, 0, 0, 0, (char *) NULL);
|
||
files_given = true;
|
||
}
|
||
else
|
||
{
|
||
for (; optind < argc; ++optind)
|
||
{
|
||
if (strcmp (argv[optind], "-") == 0)
|
||
datasection_only = false;
|
||
else
|
||
{
|
||
files_given = true;
|
||
exit_status |= (strings_file (argv[optind]) == false);
|
||
}
|
||
}
|
||
}
|
||
|
||
if (files_given == false)
|
||
usage (stderr, 1);
|
||
|
||
return (exit_status);
|
||
}
|
||
|
||
/* Scan section SECT of the file ABFD, whose printable name is FILE.
|
||
If it contains initialized data,
|
||
set `got_a_section' and print the strings in it. */
|
||
|
||
static void
|
||
strings_a_section (abfd, sect, filearg)
|
||
bfd *abfd;
|
||
asection *sect;
|
||
PTR filearg;
|
||
{
|
||
const char *file = (const char *) filearg;
|
||
|
||
if ((sect->flags & DATA_FLAGS) == DATA_FLAGS)
|
||
{
|
||
bfd_size_type sz = bfd_get_section_size_before_reloc (sect);
|
||
PTR mem = xmalloc (sz);
|
||
if (bfd_get_section_contents (abfd, sect, mem, (file_ptr) 0, sz))
|
||
{
|
||
got_a_section = true;
|
||
print_strings (file, (FILE *) NULL, sect->filepos, 0, sz, mem);
|
||
}
|
||
free (mem);
|
||
}
|
||
}
|
||
|
||
/* Scan all of the sections in FILE, and print the strings
|
||
in the initialized data section(s).
|
||
|
||
Return true if successful,
|
||
false if not (such as if FILE is not an object file). */
|
||
|
||
static boolean
|
||
strings_object_file (file)
|
||
const char *file;
|
||
{
|
||
bfd *abfd = bfd_openr (file, target);
|
||
|
||
if (abfd == NULL)
|
||
{
|
||
/* Treat the file as a non-object file. */
|
||
return false;
|
||
}
|
||
|
||
/* This call is mainly for its side effect of reading in the sections.
|
||
We follow the traditional behavior of `strings' in that we don't
|
||
complain if we don't recognize a file to be an object file. */
|
||
if (bfd_check_format (abfd, bfd_object) == false)
|
||
{
|
||
bfd_close (abfd);
|
||
return false;
|
||
}
|
||
|
||
got_a_section = false;
|
||
bfd_map_over_sections (abfd, strings_a_section, (PTR) file);
|
||
|
||
if (!bfd_close (abfd))
|
||
{
|
||
bfd_nonfatal (file);
|
||
return false;
|
||
}
|
||
|
||
return got_a_section;
|
||
}
|
||
|
||
/* Print the strings in FILE. Return true if ok, false if an error occurs. */
|
||
|
||
static boolean
|
||
strings_file (file)
|
||
char *file;
|
||
{
|
||
/* If we weren't told to scan the whole file,
|
||
try to open it as an object file and only look at
|
||
initialized data sections. If that fails, fall back to the
|
||
whole file. */
|
||
if (!datasection_only || !strings_object_file (file))
|
||
{
|
||
FILE *stream;
|
||
|
||
stream = fopen (file, "rb");
|
||
/* Not all systems permit "rb", so try "r" if it failed. */
|
||
if (stream == NULL)
|
||
stream = fopen (file, "r");
|
||
if (stream == NULL)
|
||
{
|
||
fprintf (stderr, "%s: ", program_name);
|
||
perror (file);
|
||
return false;
|
||
}
|
||
|
||
print_strings (file, stream, (file_ptr) 0, 0, 0, (char *) 0);
|
||
|
||
if (fclose (stream) == EOF)
|
||
{
|
||
fprintf (stderr, "%s: ", program_name);
|
||
perror (file);
|
||
return false;
|
||
}
|
||
}
|
||
|
||
return true;
|
||
}
|
||
|
||
/* Find the strings in file FILENAME, read from STREAM.
|
||
Assume that STREAM is positioned so that the next byte read
|
||
is at address ADDRESS in the file.
|
||
Stop reading at address STOP_POINT in the file, if nonzero.
|
||
|
||
If STREAM is NULL, do not read from it.
|
||
The caller can supply a buffer of characters
|
||
to be processed before the data in STREAM.
|
||
MAGIC is the address of the buffer and
|
||
MAGICCOUNT is how many characters are in it.
|
||
Those characters come at address ADDRESS and the data in STREAM follow. */
|
||
|
||
static void
|
||
print_strings (filename, stream, address, stop_point, magiccount, magic)
|
||
const char *filename;
|
||
FILE *stream;
|
||
file_ptr address;
|
||
int stop_point;
|
||
int magiccount;
|
||
char *magic;
|
||
{
|
||
char *buf = (char *) xmalloc (string_min + 1);
|
||
|
||
while (1)
|
||
{
|
||
file_ptr start;
|
||
int i;
|
||
int c;
|
||
|
||
/* See if the next `string_min' chars are all graphic chars. */
|
||
tryline:
|
||
if (stop_point && address >= stop_point)
|
||
break;
|
||
start = address;
|
||
for (i = 0; i < string_min; i++)
|
||
{
|
||
if (magiccount)
|
||
{
|
||
magiccount--;
|
||
c = *magic++;
|
||
}
|
||
else
|
||
{
|
||
if (stream == NULL)
|
||
return;
|
||
c = getc (stream);
|
||
if (c == EOF)
|
||
return;
|
||
}
|
||
address++;
|
||
if (!isgraphic (c))
|
||
/* Found a non-graphic. Try again starting with next char. */
|
||
goto tryline;
|
||
buf[i] = c;
|
||
}
|
||
|
||
/* We found a run of `string_min' graphic characters. Print up
|
||
to the next non-graphic character. */
|
||
|
||
if (print_filenames)
|
||
printf ("%s: ", filename);
|
||
if (print_addresses)
|
||
switch (address_radix)
|
||
{
|
||
case 8:
|
||
printf ("%7lo ", (unsigned long) start);
|
||
break;
|
||
|
||
case 10:
|
||
printf ("%7ld ", (long) start);
|
||
break;
|
||
|
||
case 16:
|
||
printf ("%7lx ", (unsigned long) start);
|
||
break;
|
||
}
|
||
|
||
buf[i] = '\0';
|
||
fputs (buf, stdout);
|
||
|
||
while (1)
|
||
{
|
||
if (magiccount)
|
||
{
|
||
magiccount--;
|
||
c = *magic++;
|
||
}
|
||
else
|
||
{
|
||
if (stream == NULL)
|
||
break;
|
||
c = getc (stream);
|
||
if (c == EOF)
|
||
break;
|
||
}
|
||
address++;
|
||
if (! isgraphic (c))
|
||
break;
|
||
putchar (c);
|
||
}
|
||
|
||
putchar ('\n');
|
||
}
|
||
}
|
||
|
||
/* Parse string S as an integer, using decimal radix by default,
|
||
but allowing octal and hex numbers as in C. */
|
||
|
||
static int
|
||
integer_arg (s)
|
||
char *s;
|
||
{
|
||
int value;
|
||
int radix = 10;
|
||
char *p = s;
|
||
int c;
|
||
|
||
if (*p != '0')
|
||
radix = 10;
|
||
else if (*++p == 'x')
|
||
{
|
||
radix = 16;
|
||
p++;
|
||
}
|
||
else
|
||
radix = 8;
|
||
|
||
value = 0;
|
||
while (((c = *p++) >= '0' && c <= '9')
|
||
|| (radix == 16 && (c & ~40) >= 'A' && (c & ~40) <= 'Z'))
|
||
{
|
||
value *= radix;
|
||
if (c >= '0' && c <= '9')
|
||
value += c - '0';
|
||
else
|
||
value += (c & ~40) - 'A';
|
||
}
|
||
|
||
if (c == 'b')
|
||
value *= 512;
|
||
else if (c == 'B')
|
||
value *= 1024;
|
||
else
|
||
p--;
|
||
|
||
if (*p)
|
||
{
|
||
fprintf (stderr, _("%s: invalid integer argument %s\n"), program_name, s);
|
||
exit (1);
|
||
}
|
||
return value;
|
||
}
|
||
|
||
static void
|
||
usage (stream, status)
|
||
FILE *stream;
|
||
int status;
|
||
{
|
||
fprintf (stream, _("\
|
||
Usage: %s [-afov] [-n min-len] [-min-len] [-t {o,x,d}] [-]\n\
|
||
[--all] [--print-file-name] [--bytes=min-len] [--radix={o,x,d}]\n\
|
||
[--target=bfdname] [--help] [--version] file...\n"),
|
||
program_name);
|
||
list_supported_targets (program_name, stream);
|
||
if (status == 0)
|
||
fprintf (stream, _("Report bugs to bug-gnu-utils@gnu.org\n"));
|
||
exit (status);
|
||
}
|