metasploit-framework/data/john/doc/RULES

296 lines
13 KiB
Plaintext
Executable File

Wordlist rules syntax.
Each wordlist rule consists of optional rule reject flags followed by
one or more simple commands, listed all on one line and optionally
separated with spaces. There's also a preprocessor, which generates
multiple rules for a single source line. Below you will find
descriptions of the rule reject flags, the rule commands (many of them
are compatible with those of Crack 5.0a), and the preprocessor syntax.
Rule reject flags.
-: no-op: don't reject
-c reject this rule unless current hash type is case-sensitive
-8 reject this rule unless current hash type uses 8-bit characters
-s reject this rule unless some password hashes were split at loading
-p reject this rule unless word pair commands are currently allowed
-u reject this rule unless the --utf8 flag is used
-U reject this rule if the --utf8 flag is used
Numeric constants and variables.
Numeric constants may be specified and variables referred to with the
following characters:
0...9 for 0...9
A...Z for 10...35
* for max_length
- for (max_length - 1)
+ for (max_length + 1)
a...k user-defined numeric variables (with the "v" command)
l initial or updated word's length (updated whenever "v" is used)
m initial or memorized word's last character position
p position of the character last found with the "/" or "%" commands
z "infinite" position or length (beyond end of word)
Here max_length is the maximum plaintext length supported for the
current hash type.
These may be used to specify character positions, substring lengths, and
other numeric parameters to rule commands as appropriate for a given
command. Character positions are numbered starting with 0. Thus, for
example, the initial value of "m" (last character position) is one less
than that of "l" (length).
Character classes.
?? matches "?"
?v matches vowels: "aeiouAEIOU"
?c matches consonants: "bcdfghjklmnpqrstvwxyzBCDFGHJKLMNPQRSTVWXYZ"
?w matches whitespace: space and horizontal tabulation characters
?p matches punctuation: ".,:;'?!`" and the double quote character
?s matches symbols "$%^&*()-_+=|\<>[]{}#@/~"
?l matches lowercase letters [a-z]
?u matches uppercase letters [A-Z]
?d matches digits [0-9]
?a matches letters [a-zA-Z]
?x matches letters and digits [a-zA-Z0-9]
?z matches all characters
?b matches characters with 8th bit set (mnemonic "b for binary")
The complement of a class can be specified by uppercasing its name. For
example, "?D" matches everything but digits.
Simple commands.
: no-op: do nothing to the input word
l convert to lowercase
u convert to uppercase
c capitalize
C lowercase the first character, and uppercase the rest
t toggle case of all characters in the word
TN toggle case of the character in position N
r reverse: "Fred" -> "derF"
d duplicate: "Fred" -> "FredFred"
f reflect: "Fred" -> "FredderF"
{ rotate the word left: "jsmith" -> "smithj"
} rotate the word right: "smithj" -> "jsmith"
$X append character X to the word
^X prefix the word with character X
String commands.
AN"STR" insert string STR into the word at position N
To append a string, specify "z" for the position. To prefix the word
with a string, specify "0" for the position.
Although the use of the double-quote character is good for readability,
you may use any other character not found in STR instead. This is
particularly useful when STR contains the double-quote character.
There's no way to escape your quotation character of choice within a
string (preventing it from ending the string and the command), but you
may achieve the same effect by specifying multiple commands one after
another. For example, if you choose to use the forward slash as your
quotation character, yet it happens to be found in a string and you
don't want to reconsider your choice, you may write "Az/yes/$/Az/no/",
which will append the string "yes/no". Of course, it is simpler and
more efficient to use, say, "Az,yes/no," for the same effect.
Length control commands.
<N reject the word unless it is less than N characters long
>N reject the word unless it is greater than N characters long
'N truncate the word at length N
English grammar commands.
p pluralize: "crack" -> "cracks", etc. (lowercase only)
P "crack" -> "cracked", etc. (lowercase only)
I "crack" -> "cracking", etc. (lowercase only)
Insert/delete commands.
[ delete the first character
] delete the last character
DN delete the character in position N
xNM extract substring from position N for up to M characters
iNX insert character X in position N and shift the rest right
oNX overstrike character in position N with character X
Also see the "X" command (extract and insert substring) under "Memory
access commands" below.
Note that square brackets ("[" and "]") are special characters to the
preprocessor: you should escape them with a backslash ("\") if using
these commands.
Charset conversion commands.
S shift case: "Crack96" -> "cRACK(^"
V lowercase vowels, uppercase consonants: "Crack96" -> "CRaCK96"
R shift each character right, by keyboard: "Crack96" -> "Vtsvl07"
L shift each character left, by keyboard: "Crack96" -> "Xeaxj85"
Memory access commands.
M memorize the word (for use with "Q", "X", or to update "m")
Q query the memory and reject the word unless it has changed
XNMI extract substring NM from memory and insert into current word at I
If "Q" or "X" are used without a preceding "M", they read from the
initial "word". In other words, you may assume an implied "M" at the
start of each rule, and there's no need to ever start a rule with "M"
(that "M" would be a no-op). The only reasonable use for "M" is in the
middle of a rule, after some commands have possibly modified the word.
The intended use for the "Q" command is to help avoid duplicate
candidate passwords that could result from multiple similar rules. For
example, if you have the rule "l" (lowercase) somewhere in your ruleset
and you want to add the rule "lr" (lowercase and reverse), you could
instead write the latter as "lMrQ" in order to avoid producing duplicate
candidate passwords for palindromes.
The "X" command extracts a substring from memory (or from the initial
word if "M" was never used) starting at position N (in the memorized or
initial word) and going for up to M characters. It inserts the
substring into the current word at position I. The target position may
be "z" for appending the substring, "0" for prefixing the word with it,
or it may be any other valid numeric constant or variable. Some example
uses, assuming that we're at the start of a rule or after an "M", would
be "X011" (duplicate the first character), "Xm1z" (duplicate the last
character), "dX0zz" (triplicate the word), "<4X011X113X215" (duplicate
every character in a short word), ">9x5zX05z" (rotate long words left by
5 characters, same as ">9{{{{{" but faster due to fewer commands),
">9vam4Xa50'l" (rotate right by 5 characters, same as ">9}}}}}").
Numeric commands.
vVNM update "l" (length), then subtract M from N and assign to variable V
V must be one of "a" through "k". N and M may be any valid numeric
constants or initialized variables. It is OK to refer to the same
variable in the same command more than once, even three times. For
example, "va00" and "vaaa" will both set the variable "a" to zero (but
the latter will require "a" to have been previously initialized),
whereas "vil2" will set the variable "i" to the current word's length
minus 2. If "i" is then used as a character position before the word is
modified further, it will refer to the second character from the end.
It is OK for intermediate variable values to become negative, but such
values should not be directly used as positions or lengths. For
example, if we follow our "vil2" somewhere later in the same rule with
"vj02vjij", we'll set "j" to "i" plus 2, or to the word's length as of
the time of processing of the "vil2" command earlier in the rule.
Character class commands.
sXY replace all characters X in the word with Y
s?CY replace all characters of class C in the word with Y
@X purge all characters X from the word
@?C purge all characters of class C from the word
!X reject the word if it contains character X
!?C reject the word if it contains a character in class C
/X reject the word unless it contains character X
/?C reject the word unless it contains a character in class C
=NX reject the word unless character in position N is equal to X
=N?C reject the word unless character in position N is in class C
(X reject the word unless its first character is X
(?C reject the word unless its first character is in class C
)X reject the word unless its last character is X
)?C reject the word unless its last character is in class C
%NX reject the word unless it contains at least N instances of X
%N?C reject the word unless it contains at least N characters of class C
Extra "single crack" mode commands.
When defining "single crack" mode rules, extra commands are available
for word pairs support, to control if other commands are applied to the
first, the second, or to both words:
1 first word only
2 second word only
+ the concatenation of both (should only be used after a "1" or "2")
If you use some of the above commands in a rule, it will only process
word pairs (e.g., full names from the GECOS field) and reject single
words. A "+" is assumed at the end of any rule that uses some of these
commands, unless you specify it manually. For example, "1l2u" will
convert the first word to lowercase, the second one to uppercase, and
use the concatenation of both. The use for a "+" might be to apply some
more commands: "1l2u+r" will reverse the concatenation of both words,
after applying some commands to them separately.
The rule preprocessor.
The preprocessor is used to combine similar rules into one source line.
For example, if you need to make John try lowercased words with digits
appended, you could write a rule for each digit, 10 rules total. Now
imagine appending two-digit numbers - the configuration file would get
large and ugly.
With the preprocessor you can do these things easier. Simply write one
source line containing the common part of these rules followed by the
list of characters you would have put into separate rules, in square
brackets (the way you would do in a regexp). The preprocessor will then
generate the rules for you (at John startup for syntax checking, and
once again while cracking, but never keeping all of the expanded rules
in memory). For the examples above, the source lines will be "l$[0-9]"
(lowercase and append a digit) and "l$[0-9]$[0-9]" (lowercase and append
two digits). These source lines will be expanded to 10 and 100 rules,
respectively. By the way, preprocessor commands are processed
right-to-left while character lists are processed left-to-right, which
results in natural ordering of numbers in the above examples and in
other typical cases. Note that arbitrary combinations of character
ranges and character lists are valid. For example, "[aeiou]" will use
vowels, whereas "[aeiou0-9]" will use vowels and digits. If you need to
have John try vowels followed by all other letters, you can use
"[aeioua-z]" - the preprocessor is smart enough not to produce duplicate
rules in such cases (although this behavior may be disabled with the
"\r" magic escape sequence described below).
There are some special characters in rules ("[" starts a preprocessor
character list, "-" marks a range inside the list, etc.) You should
prefix them with a backslash ("\") if you want to put them inside a rule
without using their special meaning. Of course, the same applies to "\"
itself. Also, if you need to start a preprocessor character list at the
very beginning of a line, you'll have to prefix it with a ":" (the no-op
rule command), or it would be treated as a new section start.
Finally, the preprocessor supports some magic escape sequences. These
start with a backslash and use characters that you would not normally
need to escape. In the following paragraph describing the escapes, the
word "range" refers to a single instance of a mix of character lists
and/or ranges placed in square brackets as illustrated above.
Currently supported are "\1" through "\9" for back-references to prior
ranges (these will be substituted by the same character that is
currently substituted for the referenced range, with ranges numbered
from 1, left-to-right), "\0" for back-reference to the immediately
preceding range, "\p" before a range to have that range processed "in
parallel" with preceding ranges, "\p1" through "\p9" to have the range
processed "in parallel" with the specific referenced range, "\p0" to
have the range processed "in parallel" with the immediately preceding
range, and "\r" to allow the range to produce repeated characters. The
"\r" escape is only useful if the range is "parallel" to another one or
if there's at least one other range "parallel" to this one, because you
should not want to actually produce duplicate rules.
Please refer to the default configuration file for John the Ripper for
many example uses of the features described in here.
$Owl: Owl/packages/john/john/doc/RULES,v 1.11 2010/02/26 01:13:37 solar Exp $