mirror of
https://github.com/RfidResearchGroup/proxmark3.git
synced 2024-11-15 14:20:51 +08:00
119 lines
2.5 KiB
Perl
Executable file
119 lines
2.5 KiB
Perl
Executable file
#!/usr/bin/perl
|
|
# -samy kamkar, rfid@samy.pl
|
|
|
|
use strict;
|
|
|
|
die "usage: $0 <file with data> <binary to search for>\n" unless @ARGV == 2;
|
|
|
|
my ($file, $search) = @ARGV;
|
|
$search =~ s/\s//g;
|
|
|
|
# sure, these aren't perfect, but simplifies usability if you know what you're doing
|
|
# if in doubt, use binary
|
|
|
|
# binary, cool
|
|
if ($search =~ /^[01]+$/) { }
|
|
# decimal
|
|
elsif ($search =~ /^\d+$/)
|
|
{
|
|
$search = unpack("B*", pack("N", $search));
|
|
$search =~ s/^0*//;
|
|
}
|
|
# hex
|
|
elsif ($search =~ /^[\da-fA-F]+$/)
|
|
{
|
|
$search = unpack("B*", pack("H*", $search));
|
|
$search =~ s/^0*//;
|
|
}
|
|
# ascii
|
|
else
|
|
{
|
|
$search = unpack("B*", $search);
|
|
$search =~ s/^0*//;
|
|
}
|
|
|
|
|
|
# read file contents
|
|
open(F, "<$file") || die "Can't read $file: $!";
|
|
my $data = join("", <F>);
|
|
close(F);
|
|
|
|
# convert to binary
|
|
$data =~ s/\s//g;
|
|
# binary, great
|
|
if ($data =~ /^[01]+$/) { }
|
|
elsif ($data =~ /^[\da-fA-F]+$/)
|
|
{
|
|
$data = unpack("B*", pack("H*", $data));
|
|
$search =~ s/^0*//;
|
|
}
|
|
else
|
|
{
|
|
die "Seriously. What sort of data is this file? Binary or hex only please.\n";
|
|
}
|
|
|
|
|
|
# search every method we know how
|
|
print "Testing normally...\n";
|
|
test_all($data, $search);
|
|
|
|
print "Testing with flipped bits...\n";
|
|
test_all($data, $search, 1);
|
|
|
|
# now try manchester demodulating
|
|
my @bits = split(//, $data);
|
|
my $man;
|
|
my $last = 0;
|
|
for (my $i = 1; $i < @bits; $i++)
|
|
{
|
|
# if we changed, flip our bit
|
|
if ($bits[$i-1] == 1)
|
|
{
|
|
$last ^= 1;
|
|
}
|
|
$man .= $last;
|
|
}
|
|
|
|
print "Testing with manchester demodulation...\n";
|
|
test_all($man, $search);
|
|
|
|
print "Testing with flipped manchester demodulation...\n";
|
|
test_all($man, $search, 1);
|
|
|
|
|
|
sub test_all
|
|
{
|
|
my ($data, $search, $flip) = @_;
|
|
|
|
if ($flip)
|
|
{
|
|
$data =~ s/(.)/$1 ^ 1/eg;
|
|
}
|
|
|
|
# first just see if our data is in the stream
|
|
if ($data =~ /$search/)
|
|
{
|
|
print "Found $search in our stream ($data)\n";
|
|
}
|
|
|
|
# try removing parity every 4 and 8 bits
|
|
foreach my $parity (4, 8)
|
|
{
|
|
# try removing a parity bit every $parity bits
|
|
# test by cutting off a bit at a time in case we're in the wrong bit position
|
|
my $tmp = $data;
|
|
foreach (1 .. $parity)
|
|
{
|
|
my $test = $tmp;
|
|
$test =~ s/(.{$parity})./$1/g;
|
|
|
|
if ($test =~ /$search/)
|
|
{
|
|
print "Found $search with parity every " . ($parity + 1) . "th bit, round $_ out of $parity ($test)\n";
|
|
}
|
|
|
|
# chop of a bit to change our bit position next round
|
|
$tmp =~ s/^.//;
|
|
}
|
|
}
|
|
}
|