Hashrat and libUseful are (C) 2014 Colum Paget. They are released under the GNU Public License, so you may do anything with them that the GPL allows.
Email: colums.projects@gmail.com
This is free software. It comes with no guarentees and I take no responsiblity if it makes your computer explode or opens a portal to the demon dimensions, or does anything at all, or doesn't.
Hashrat is a hash-generation utility that supports the md5, sha1, sha256, sha512, whirlpool, jh-244, jh256, jh-384 and jh-512 hash functions, and also the HMAC versions of those functions. It can output in 'traditional' format (same as md5sum and shasum and the like), or its own format. Hashes can be output in octal, decimal, hexadecimal, uppercase hexadecimal or base64. It supports directory recursion, hashing entire devices, and generating a hash for an entire directory. It has a 'CGI' mode that can be used as a web-page to lookup hashes.
Thanks for bug reports/advice to: Stephan Hegel, Michael Shigorin mike@altlinux.org and Joao Eriberto Mota Filho eriberto@debian.org
Thanks to the people who invented the hash functions!
- MD5: Ronald Rivest
- Whirlpool: Vincent Rijmen, Paulo S. L. M. Barreto
- JH: Hongjun Wu
- SHA: The NSA (thanks, but please stop reading my email. It's kinda creepy.).
Special thanks to Professor Hongjun Wu for taking the time to confirm that his JH algorithm is free for use in GPL programs. Special, special thanks to Joao Eriberto Mota Filho for doing a LOT of work to make hashrat debian ready!
hashrat [options] [paths]...
Hash things: hashrat [options] [paths to hash]
Check hashes: hashrat -c [options] [paths to hash]
Find files matching: hashrat -m [options] [paths to hash]
Find duplicate files: hashrat -dups [options] [paths to hash]
--help Print this help
-help Print this help
-? Print this help
--version Print program version
-version Print program version
-list-hashes List available hash functions, including those supported by openssl
-type <hash> Hash using supplied type. This supports chaining hashes like so: '-type sha512,md5'
-md5 Use md5 hash algorithmn
-sha1 Use sha1 hash algorithmn
-sha256 Use sha256 hash algorithmn
-sha384 Use sha256 hash algorithmn
-sha512 Use sha512 hash algorithmn
-whirl Use whirlpool hash algorithmn
-whirlpool Use whirlpool hash algorithmn
-jh224 Use jh-224 hash algorithmn
-jh256 Use jh-256 hash algorithmn
-jh384 Use jh-384 hash algorithmn
-jh512 Use jh-512 hash algorithmn
-hmac HMAC using specified hash algorithm
-totp <secret> TOTP code from supplied secret (defaults to google authenticator compatible code).
-totp <url> TOTP code from supplied otpauth url.
-digits <n> Produce TOTP code with <n> digits.
-period <n> Produce TOTP code with period/lifetime of <n> seconds.
-8 Encode with octal instead of hex
-10 Encode with decimal instead of hex
-H Encode with UPPERCASE hexadecimal instead of lowercase
-HEX Encode with UPPERCASE hexadecimal instead of lowercase
-32 Encode with base32 instead of hex
-base32 Encode with base32 instead of hex
-c32 Encode with Crockford base32 instead of hex
-w32 Encode with word-safe base32 instead of hex
-z32 Encode with zbase32 instead of hex
-64 Encode with base64 instead of hex
-base64 Encode with base64 instead of hex
-i64 Encode with base64 with rearranged characters
-p64 Encode with base64 with a-z,A-Z and _-, for best compatibility with 'allowed characters' in websites.
-r64 Encode with base64 with a-z,A-Z and _-, rfc4648 compatible.
-rfc4648 Encode with base64 with a-z,A-Z and _-, rfc4648 compatible.
-x64 Encode with XXencode style base64.
-u64 Encode with UUencode style base64.
-g64 Encode with GEDCOM style base64.
-a85 Encode with ASCII85.
-z85 Encode with ZEROMQ variant of ASCII85.
-t Output hashes in traditional md5sum, shaXsum format
-trad Output hashes in traditional md5sum, shaXsum format
-bsd Output hashes in bsdsum format
-tag Output hashes in bsdsum format
--tag Output hashes in bsdsum format
-r Recurse into directories when hashing files
-hid Show hidden (starting with .) files
-hidden Show hidden (starting with .) files
-f <listfile> Hash files listed in <listfile>
-i <patterns> Only hash items matching a comma-seperated list of shell patterns
-x <patterns> Exclude items matching a comma-sepearted list of shell patterns
-X <file> Exclude items matching shell patters stored in <file>
-name <patterns> Only hash items matching a comma-seperated list of shell patterns (-name aka 'find')
-mtime <days> Only hash items <days> old. Has the same format as the find command, e.g. -10 is younger than ten days, +10 is older than ten, and 10 is ten days old
-mmin <mins> Only hash items <min> minutes old. Has the same format as the find command, e.g. -10 is younger than ten mins, +10 is older than ten, and 10 is ten mins old
-myear <years> Only hash items <years> old. Has the same format as the find command, e.g. -10 is younger than ten years, +10 is older than ten, and 10 is ten years old
-exec In CHECK or MATCH mode only examine executable files.
-dups Search for duplicate files.
-rename Rename files to include a hash in their name. Filename format is '<name>-<hash>.<extn>'
-n <length> Truncate hashes to <length> bytes
-segment <length> Break hash up into segments of <length> chars seperated by '-'
-c CHECK hashes against list from file (or stdin)
-cf CHECK hashes against list but only show failures
-C <dir> Recursively CHECK directory against list of files on stdin
-Cf <dir> Recursively CHECK directory against list but only show failures
-m MATCH files from a list read from stdin.
-lm Read hashes from stdin, upload them to a memcached server (requires the -memcached option).
-memcached <server> Specify memcached server. (Overrides reading list from stdin if used with -m, -c or -cf).
-mcd <server> Specify memcached server. (Overrides reading list from stdin if used with -m, -c or -cf).
-h <script> Script to run when a file fails CHECK mode, or is found in MATCH mode.
-hook <script> Script to run when a file fails CHECK mode, or is found in FIND mode
-color Use ANSI color codes on output when checking hashes.
-strict Strict mode: when checking, check file mtime, owner, group, and inode as well as its hash
-S Strict mode: when checking, check file mtime, owner, group, and inode as well as its hash
-d dereference (follow) symlinks
-fs Stay on one file system
-dir DirMode: Read all files in directory and create one hash for them!
-dirmode DirMode: Read all files in directory and create one hash for them!
-devmode DevMode: read from a file EVEN IF IT'S A DEVNODE
-lines Read lines from stdin and hash each line independantly.
-rawlines Read lines from stdin and hash each line independantly, INCLUDING any trailing whitespace. (This is compatible with 'echo text | md5sum')
-rl Read lines from stdin and hash each line independantly, INCLUDING any trailing whitespace. (This is compatible with 'echo text | md5sum')
-cgi Run in HTTP CGI mode
-cgi Run in HTTP CGI mode
-xdialog Run in 'xdialog' (zenity, yad or qarama) mode
-dialog-types <list> Specify a list of dialog commands and use the first found on the system. Default is 'yad,zenity,qarma'
-iprefix <prefix> String to prefix all input before hashing
-oprefix <prefix> Prefix to add to the front of output hashes
-net Treat 'file' arguments as either ssh or http URLs, and pull files over the network and then hash them (Allows hashing of files on remote machines).
URLs are in the format ssh://[username]:[password]@[host]:[port] or http://[username]:[password]@[host]:[port]..
-idfile <path> Path to an ssh private key file to use to authenticate INSTEAD OF A PASSWORD when pulling files via ssh.
-xattr Use eXtended file ATTRibutes. In hash mode, store hashes in the file attributes, in check mode compare against hashes stored in file attributes.
-txattr Use TRUSTED eXtended file ATTRibutes. In hash mode, store hashes in 'trusted' file attributes. 'trusted' attributes can only be read and written by root. Under freebsd this menas SYSTEM attributes.
-attrs comma-separated list of filesystem attribute names to be set to the value of the hash.
-cache Use hashes stored in 'user' xattr if they're younger than the mtime of the file. This speeds up outputting hashes. When storing hashes, do not update files that already have the correct hash.
-u <types> Update. In checking mode, update hashes for the files as you go. <types> is a comma-separated list of things to update, which can be 'xattr' 'memcached' or a file name. This will update these targets with the hash that was found at the time of checking.
-hide-input When reading data from stdin in linemode, set the terminal to not echo characters, thus hiding typed input.
-star-input When reading data from stdin in linemode replace characters with stars.
-xsel Update X11 clipboard and primary selections to the current hash. This works using Xterm command sequences. The xterm resource 'allowWindowOps' must be set to 'true' for this to work.
-clip Update X11 clipboard to the current hash. This works using the 'xsel', 'xclip' or 'pbcopy' commands, or if none of those are installed falls back to Xterm clipboard as in the '-xsel' option .
-qr Display the current hash as a qrcode. This requires the 'qrencode' command to be installed, and also an image viewer like fim, feh, or imagemagick display to be installed.
-qrcode Display the current hash as a qrcode. This requires the 'qrencode' command to be installed, and also an image viewer like fim, feh, or imagemagick display to be installed.
-clipcmd <cmds> Comma separated list of clipboard-setter commands to use instead of the defaults.
-viewcmd <cmds> Comma separated list of image-viewer commands to use instead of the defaults.
Hashrat can also detect if it's being run under any of the following names (e.g., via symlinks)
md5sum run with '-trad -md5'
shasum run with '-trad -sha1'
sha1sum run with '-trad -sha1'
sha256sum run with '-trad -sha256'
sha384sum run with '-trad -sha256'
sha512sum run with '-trad -sha512'
jh224sum run with '-trad -jh224'
jh256sum run with '-trad -jh256'
jh384sum run with '-trad -jh384'
jh512sum run with '-trad -jh512'
whirlpoolsum run with '-trad -whirl'
hashrat.cgi run in web-enabled 'cgi mode'
hashrat
Generate an md5 hash of data read from stdin (default hash type is md5).
hashrat -jh256
Generate a jh-256 hash of data read from stdin
hashrat -sha256 -64
Generate an sha-256 hash of data read from stdin, output with base64 encoding.
hashrat -sha256 -64 -lines
Read lines from stdin, and generate an sha-256 with base64 encoding FOR EVERY LINE. This strips any whitespace from the end of the line (including \r and/or \n line terminators).
hashrat -md5 -trad -rawlines
Read lines from stdin, and generate an md5 hash in 'traditional' format for every line INCLUDING TRAILING WHITESPACE. This is compatible with 'echo text | md5sum' where 'text' is one line, as 'echo' adds a newline to the end of the text it outputs.
hashrat -type sha256,whirl,md5
Read data from stdin, hash it with sha256, then hash the resulting hash with whirlpool, then with md5
hashrat *
Generate a list of hashes for files in the current directory (default hash type is md5).
hashrat -r -sha1 * > hashes.sha1
Generate a list of hashes for files in the current directory, AND ALL SUBDIRECTORIES, using sha1 hashing.
cat hashes.sha1 > hashrat -c
Check hashes listed in 'hashes.sha1'
cat hashes.sha1 > hashrat -c -strict
Check hashes listed in 'hashes.sha1'. If hashes are NOT in 'traditional' format than the '-strict' flag will cause hashrat to check the files uid, gid, size, mtime and inode and print a failure message if any of those don't match.
cat hashes.sha1 > hashrat -cf
Check hashes listed in 'hashes.sha1' but only output failures
cat APT1.md5 | hashrat -m -r /
Read a list of hashes from stdin, and search recursively for files matching them.
cat APT1.md5 | hashrat -lm -memcached 127.0.0.1
Read a list of hashes from stdin, and register them in a memcached server for later use in a search/check.
cat APT1.ioc | hashrat -lm -memcached 127.0.0.1
Extract hashes from an Open IOC file and register them in a memcached server for later use in a search/check.
hashrat -m -memcached 127.0.0.1 -r /
Search recursively for files whose hashes are stored in a memcached server.
hashrat -devmode -whirlpool -64 /dev/sda1
Generate a whirlpool hash of the entire device /dev/sda1. Output result in base 64.
hashrat -sha1 -net ssh:user:password@myhost/bin/*
Generate sha1 hashes of files in /bin/* on the remote machine 'myhost'
hashrat -whirlpool -net http://myhost.com/webpage.html
Generate whirlpool hash for the listed URL. (Note, many webpages have dynamic content that changes every time, so this will only return the same hash over and over if the page is static and doesn't change.
hashrat -dups -r /home -u xattr
Search for duplicate files under /home. Update hashes stored in filesystem attributes as you go
From v1.21 hashrat supports using hash functions supplied by openssl (provided it's been compiled with --enable-openssl). A list of available hashes can be viewed with hashrat -list-hashes and then any listed hash function can be used via the type option.
e.g.
hashrat -type openssl:shake128
Hashes can be 'chained' (fed into each other) using the '-type' option and a comma-separated list of hash names.
e.g.
hashrat -type sha512,sha512,whirl,md5
this would pipe any input into sha512, then the output of that goes into another sha512 round, then into whirlpool, and finally md5
- Strong Passwords I mostly use hashrat to generate strong passwords for websites. As I always have access to hashrat, I don't have to remember the strong password, as I can always regenerate them with hashrat. So, I remember a handful of moderately decent passwords (i.e. things that I can't find by grepping in the '10,000 most popular passwords' list https://github.com/discourse/discourse/blob/master/lib/common_passwords/10k-common-passwords.txt), and I also remember a 'personal pin'. I then combine the website name, one of my passwords, and my personal pin, into a string and feed them into hashrat:
echo "facebook.com password 1234" | hashrat -sha1 -64
Obviously, my password isn't 'password' and my pin isn't '1234', but you get the idea. This gives me a 28-character string that should take "8.02 trillion trillion centuries" to crack with a "massive cracking array" according to Steve Gibson's 'Password haystacks' utility, https://www.grc.com/haystack.htm. This is what I then use as my password. Unfortunately some websites won't take a 28-character password, and for these I have to truncate to the appropriate length (using the -n flag), but the results are still stronger than anything I could remember, and nothing needs storing on disk (as with password managers).
There are some dangers to using the 'echo' method shown above if you are on a shared machine, or if someone gets hold of your computer/harddrive. On a shared machine someone could type 'ps ax' to see all commands running, and if they time it right, they might see your command-line with your password in it. Another danger lies in using a shell (like bash) that will record your typed commands so you can recall them later. Bash stores this information on disk in the file .bash_history, so if you use the 'echo' method shown above your password will be saved on disk. To combat this hashrat has 'line mode'
hashrat -sha1 -64 -lines
This reads lines from stdin, so type into hashrat and then press 'enter', and you'll be given the hash of the line you typed. By this method your password is neither visible in 'ps ax', nor is ever stored on disk.
'-lines' will produce a different hash to the 'echo' method listed above, because it strips any trailing whiespace off the lines read. If you want strict compatiblity with 'echo' (by default echo adds a 'newline' to the end of the text to output) then use 'rawlines' mode:
hashrat -sha1 -64 -rawlines
Finally, you can prevent shoulder-surfers seeing you type your password by using the '-hide-input' or '-star-input' options to hide what you type.
- Watching for file changes
Like md5sum/shasum etc, hashrat can be used to detect changes in files that might indicate malicious activity. For instance, in order to get early warning of malware like 'cryptolocker' (that encrypts files on a users disk, or on network shares, and then demands a ransom for file recovery) I scatter about the disk a number of 'Canary files' that should not change. I record their hashes and regularly check them. If they change, I know something is going on.
Hashes generated by hashrat can be output to a file, or stored in extended file attributes, or in a memcached server.
hashrat -sha256 -r . > /tmp/files.sha256
hashrat -sha256 -r . -xattr
hashrat -sha256 -r . -memcached
Similarly these can then be used to check files later
cat /tmp/files.sha256 | hashrat -c -sha256
cat /tmp/files.sha256 | hashrat -sha256 -C /tmp
hashrat -C -sha256 -xattr
hashrat -C -sha256 -memcached
Hashrat has four 'check' options:
-c check from list
-cf check from list and only show failures
-C check directory
-Cf check directory and only show failures
the -c forms only check the files listed on stdin. The -C forms read a list of files on stdin, but they expect to be given a directory to check, and they check every file in that directory, outputing any that weren't in the list, or that are in the list but not on disk, as well as any that have changed. Thus the -C forms allow one to detect new files, changed files, and deleted files, where the -c form only detects changed and deleted files. The -c and -C forms both work as -C if the hashes are supplied via memcached or file system attributes (using -xattr or -memcached)
Note that -C implies -r, so you don't need to supply -r
- Finding files that match hashes.
Using the -m flag hashrat can be told to read a range of hashes from stdin, and then search for files matching those hashes. For Example:
cat APT1-AppendixE-MD5s.txt | hashrat -r -m /usr
Will search recursively under /usr for files with hashes matching those in APT1-AppendixE-MD5s.txt. The input on stdin must begin with a hash, anything written after the hash will be treated as a comment to be displayed if a file matching the hash is found.
Hashtypes other than md5 can be used thusly:
cat sha1-list.lst | hashrat -r -sha1 -m /usr
The input file can be in hashrat format, 'traditional' md5sum/shasum format, bsdsum format, or can be an Open IOC file from which any hashes will be extracted.
Hashes can also be loaded into a memcached server, so that the same file list can be checked on a number of machines, without needing to store the hashlist on those machines. First we load the hashes:
cat APT1-AppendixE-MD5s.txt | hashrat -lm -memcached 192.168.1.5
Loads the hashes to a memcached server at 192.168.1.5. We can then search against the memcached server by:
hashrat -r -m -memcached 192.168.1.5 /usr
- Find duplicate files
Using the -dups flag (usually in combination with the '-r' recursive flag) hashrat can be set to search for duplicate files and output any found to stdout.
- As an 'ls'
Hashrat outputs a file's name, type, mode, mtime, uid, gid and size, along with a hash. This allows it to be used as a kind of 'ls' by ftp style programs, listing all the details of a file, but with the added feature of a hash.
- Hashing files on remote machines
If run with the '-net' option, hashrat will treat paths starting with 'http://' or 'ssh://' differently, connecting to the target machine and pulling files off it (hashrat assumes there is no hashing program on the remote machine, and that it must therefore download the files to hash them). For ssh paths wildcars are supported:
hashrat -net ssh://username:password@server/usr/bin/*
- As a TOTP authenticator
Hashrat can be used as a TOTP authenticator, and defaults to google-authenticator compatible codes.
hashrat -totp 3EK4LIB2553CUPA7DBXJMMKDCYTEA2IZA
Hookscripts, defined using the -h or -hook command-line options, are scripts that are run for each hashed item. -h and -hook options take an argument that is the path to the script. Hookscripts behave differently in different modes:
- 'Check Hashes' mode: The hookscript is called if a file doesn't match its expected hash, or is not listed in the expected hashes. It is passed the path of the file.
- 'Locate files' mode: The hookscript is called if a file matches the hash to locate. It is passed the path of the file.
- 'Find duplicates' mode: The hookscript is called if a file is a duplicate of another file. It is passed the paths of both files.
The '-m' function allows matching a list of file hashes input on stdin. It differs from check mode in that, instead of checking a specific file against a hash supplied for the file, it accepts a hash and goes looking for any files that match that hash. However, this causes problems if different hash functions are used in the file input. Hashrat's native format allows specifying the hash-type that is supplied on a hash-by-hash basis. This works fine for 'check' mode, but in 'match' mode hashrat does not know which fingerprint/hash a file will match before it hashes it. Therefore hashrat will expect all the supplied hashes to be the same hash format when in match mode. This if suppling native-format hashes like this:
hash='sha1:c04b4a87e789eceb7c10c2d6a785f91820bf9db4' type='file' mode='100664' uid='1000' gid='1000' size='5759' mtime='1718700145' inode='8436275' path='tests/help.txt'
hash='md5:10fc712fd1298e2cb7a34303242d8ae4' type='file' mode='100664' uid='1000' gid='1000' size='621' mtime='1718700145' inode='8436276' path='tests/quotes.txt'
Hashrat will honor the LAST format given (in this case md5) and use it for matching all files.
Basically, don't mix hash types when using 'match' mode.
If hashrat is run with the -cgi flag, or if it's run with a name of 'hashrat.cgi' (either by renaming the 'hashrat' executable, or via a symbolic link) it will output a webpage that allows users to look up hashes over the web. This allows me to look-up my strong passwords even if I don't have access to a local version of hashrat.
CGI Mode can be configured using an options file. The file path is passed in the url, like this:
hashrat.cgi?OptionsFile=/etc/hashrat.options
If an options file is used, then CGI mode uses the options in the file as its defaults. The options file contains the following :
HashType <type> Type of hash to generate
Encoding <type> Type of encoding to use for outputted hash
Line Ending <type> Line ending to append to input text. This is for compatability with command-line usage with '-rawlines'. Options are 'none', 'lf', 'crlf' or 'cr', meaning 'none
', 'newline', 'carriage-return newline' and 'carriage-return' respectively.
OutputLength <len> Crop output hash to length len 'len'
SegmentLength <len> Break output up into segments of length 'len'
SegmentChar <char> Seperate output segments with character 'char'
NoOptions Do not offer the user the options so they can change them. Just show an entry box to enter text.
HideText Hide inputted text (overrides any other config)
ShowText Show inputted text (overrides any other config)
Example:
HashType=sha256
Encoding=base64
LineEnding=none
OutputLength=12
SegmentLength=4
SegmentChar=+
NoOptions=Y
Hashrat can be used as a TOTP (Time-based One Time Password) authenticator and defaults to google-authenticator compatible codes. The simplest use case is:
hashrat -totp <secret>
It's possible to change the hash, period/lifetime and number of digits in the TOTP code like so:
hashrat -totp 3EK4LIB2553CUPA7DB -sha256 -period 90 -digits 8
Hashes produced in standard-in input mode and TOTP codes can be pushed to the system clipboard using the -clip option. This option first tries to find a command that can set the clipboard, searching for one of 'xsel', 'xclip' or 'pbcopy'. If it can't find any of these, it falls back to using xterm's built in clipboard setting method.
Alternatively the -xsel option only attempts to use the xterm clipboard setting method.
The default list of clipboard commands can be overridden using the -clipcmd option.
Hashes produced in standard-in input mode and TOTP codes can be displayed as qrcodes using the -qr or -qrcode options. These options require the "qrencode" utility to be installed, and also an image viewer that can be used to display the qrcode image.
By default hashrat searches for the following image viewers:
imlib2_view,fim,feh,display,xv,phototonic,qimageviewer,pix,sxiv,qimgv,qview,nomacs,geeqie,ristretto,mirage,fotowall,links -g
The default list of image viewers can be overridden using the -viewcmd option.
The '-rename' option allows renaming files to include a hash in their filename. The filename format is '-.'. For example:
hashrat -p64 -md5 -rename main.c
Will rename 'main.c' to 'main-gAmHI2oarRmpiXkE7MVK90.c', where 'gAmHI2oarRmpiXkE7MVK90' is it's MD5 has encoded with the 'p64' base-64 encoding scheme.
The standard '-64' encoding scheme uses the '/' character, which is not allowed in a filename and results in renames failing. Thus the '-p64' or '-r64' encoding schemes should be used.
Using the '-r' flag entire directories of files can be renamed in this manner:
hashrat -p64 -md5 -rename -r myfiles
Hashrat can use extended filesystem attributes where these are supported. This allows a hash to be stored in the filesystem metadata of the target file. This can then be used for checking hashes, or for caching hashes to produce faster output during hashing runs. There are two types of filesystem attribute, 'trusted' attributes, which can only be set and read by root, and 'user' attributes, which can be set and read by any user that has the appropriate permissions for the file.
Hashes can be stored against files by using the -xattr option to set 'user' attributes...
hashrat -sha256 -r . -xattr
... and using the '-txattr' flag to set trusted attributes (you must be root to set trusted attributes)
hashrat -sha256 -r . -txattr
When checking either flag can be used, but hashrat will always use 'trusted' attributes when running as root, if those are available, otherwise it will fall back to 'user' attributes.
hashrat -c -sha256 -r . -xattr
The -cache option allows using stored hashes rather than regenerating hashes. It only considers hashes stored in 'user' attributes at current.
hashrat -r . -cache
This makes getting a report of hashes considerably faster, but it runs the risk that the hashes may not be accurate. Hashrat will only output a hash stored in file attributes if the storage time of the hash is younger than the modify time (mtime) of the file, however, this means an attacker could change the modify time of the file to hide changes they've made. Thus this feature should not be used for security checking purposes (but should be safe for uses like finding files that have changed and need to be backed up, for instance).
The '-u' option allows filesystem attributes to be updated as we do checks (in check mode -xattr means 'read from xattr', so we need the '-u' (update) flag to tell hashrat to also update the stored hash of any files who's hash has changed).
hashrat -c -r . -xattr -u xattr