txtferret

Identify and classify data in your text files with Python.

Description

Definition: txtferret

• A weasel-like mammal that feasts on rodents... and apparently social security numbers, credit card numbers, or any other data that's in your text or gzipped text files.

Use custom regular expressions and sanity checks (ex: luhn algorithm for account numbers) to find sensitive data in virtually any size file via your command line.

Why use txtferret? See the How/why did this come about? section below.

Table of Contents

• Quick Start
• Configuration
• How/why did this come about?
• Development

Quick Start

PyPi

1. Install it

   $ pip3 install txtferret

Repo

1. Clone it.

   $ git clone git@github.com:krayzpipes/txt-ferret.git
   $ cd txt-ferret

2. Setup environment.

   $ python3.7 -m venv venv
   $ source venv/bin/activate

3. Install it.

   (venv) $ python setup.py install

Run it

• Example file size:

  # Decent sized file.
  
  $ ls -alh | grep my_test_file.dat
  -rw-r--r--  1 mrferret ferrets  19G May  7 11:15 my_test_file.dat

• Scanning the file.

  # Scan the file.
  # Default behavior is to mask the string that was matched.
  
  $ txtferret scan my_test_file.dat
  2019:05:20-22:18:01:-0400 Beginning scan for /home/mrferret/Documents/test_file_1.dat
  2019:05:20-22:18:18:-0400 PASSED sanity and matched regex - /home/mrferret/Documents/test_file_1.dat - Filter: fake_ccn_account_filter, Line 712567, String: 100102030405060708094
  2019:05:20-22:19:09:-0400 Finished scan for /home/mrferret/Documents/test_file_1.dat
  2019:05:20-22:19:09:-0400 SUMMARY:
  2019:05:20-22:19:09:-0400   - Matched regex, failed sanity: 2
  2019:05:20-22:19:09:-0400   - Matched regex, passed sanity: 1
  2019:05:20-22:19:09:-0400 Finished in 78 seconds (~1 minutes).

• Scanning the file with a delimiter.

  # Break up each line of a CSV into columns by declaring a comma for a delimiter.
  # Scan each field in the row and return column numbers as well as line numbers.
  
  $ txtferret scan --delimiter , test_file_1.csv
  2019:05:20-21:41:57:-0400 Beginning scan for /home/mrferret/Documents/test_file_1.csv
  2019:05:20-21:44:34:-0400 PASSED sanity and matched regex - /home/mrferret/Documents/test_file_1.csv - Filter: fake_ccn_account_filter, Line 712567, String: 100102030405060708094, Column: 171
  2019:05:20-21:49:16:-0400 Finished scan for /home/mrferret/Documents/test_file_1.csv
  2019:05:20-21:49:16:-0400 SUMMARY:
  2019:05:20-21:49:16:-0400   - Matched regex, failed sanity: 2
  2019:05:20-21:49:16:-0400   - Matched regex, passed sanity: 1
  2019:05:20-21:49:16:-0400 Finished in 439 seconds (~7 minutes).

  • Scan all files in a directory. Write results to a file and stdout

  # Uses multiprocessing to speed up scans of a bulk group of files
  
  $ txtferret scan -o bulk_testing.log --bulk ../test_files/
  2019:06:09-15:15:27:-0400 Detected non-text file '/home/mrferret/Documents/test_file_1.dat.gz'... attempting GZIP mode (slower).
  2019:06:09-15:15:27:-0400 Detected non-text file '/home/mrferret/Documents/test_file_2.dat.gz'... attempting GZIP mode (slower).
  2019:06:09-15:15:27:-0400 Beginning scan for /home/mrferret/Documents/test_file_1.dat.gz
  2019:06:09-15:15:27:-0400 Beginning scan for /home/mrferret/Documents/test_file_2.dat.gz
  2019:06:09-15:15:27:-0400 Beginning scan for /home/mrferret/Documents/test_file_3.dat
  2019:06:09-15:15:27:-0400 PASSED sanity and matched regex - /home/mrferret/Documents/test_file_2.dat.gz - Filter: fake_ccn_account_filter, Line 4, String: 100102030405060708094
  2019:06:09-15:15:27:-0400 Finished scan for /home/mrferret/Documents/test_file_2.dat.gz
  2019:06:09-15:16:04:-0400 PASSED sanity and matched regex - /home/mrferret/Documents/test_file_3.dat - Filter: fake_ccn_account_filter, Line 712567, String: 100102030405060708094
  2019:06:09-15:16:51:-0400 PASSED sanity and matched regex - /home/mrferret/Documents/test_file_1.dat.gz - Filter: fake_ccn_account_filter, Line 712567, String: 100102030405060708094
  2019:06:09-15:17:15:-0400 Finished scan for /home/mrferret/Documents/test_file_3.dat
  2019:06:09-15:19:24:-0400 Finished scan for /home/mrferret/Documents/test_file_1.dat.gz
  2019:06:09-15:19:24:-0400 SUMMARY:
  2019:06:09-15:19:24:-0400   - Scanned 3 file(s).
  2019:06:09-15:19:24:-0400   - Matched regex, failed sanity: 16
  2019:06:09-15:19:24:-0400   - Matched regex, passed sanity: 3
  2019:06:09-15:19:24:-0400   - Finished in 236 seconds (~3 minutes).
  2019:06:09-15:19:24:-0400 FILE SUMMARIES:
  2019:06:09-15:19:24:-0400 Matches: 1 passed sanity checks and 2 failed, Time Elapsed: 236 seconds / ~3 minutes - /home/mrferret/Documents/test_file_1.dat.gz
  2019:06:09-15:19:24:-0400 Matches: 1 passed sanity checks and 3 failed, Time Elapsed: 0 seconds / ~0 minutes - /home/mrferret/Documents/test_file_2.dat.gz
  2019:06:09-15:19:24:-0400 Matches: 1 passed sanity checks and 2 failed, Time Elapsed: 107 seconds / ~1 minutes - /home/mrferret/Documents/test_file_3.dat

Configuration

There are two ways to configure txt-ferret. You can make changes or add filters through making a custom configuration file (based on the default YAML file) or you can add some settings via CLI switches.

• CLI Switches will always win/take precedence.
• User-defined configuration file will always beat the default configuration.
• If any settings are not defined in a user-file configuration by cli switches, then the default setting will be applied.

Txt-ferret comes with a default config which you can dump into any directory you wish and change it or use it for reference. If you change the file, you have to specifiy it with the appropriate CLI switch in order for the script to use it. See the CLI section below.

(venv) $ txtferret dump-config /file/to/write/to.yaml

There are two sections of the config file: filters and settings.

Filters

Filters are regular expressions with some metadata. You can use this metadata to perform sanity checks on regex matches to sift out false positives. (Ex: luhn algorithm for credit card numbers). You can also mask the output of the matched string as it is logged to a file or displayed on a terminal.

filters:
- label: american_express_15_ccn
  pattern: '((?:34|37)\d{2}(?:(?:[\W_]\d{6}[\W_]\d{5})|\d{11}))'
  substitute: '[\W_]'
  exclude_patterns: ["dont_match_me", "dont_match_me_either"]
  sanity: luhn
  mask:
    index: 2,
    value: XXXXXXXX
  type: Credit Card Number

• label:
  • This will be displayed in the logs when the filter in question has matched a string.
• pattern:
  • The regular expression which will be used to find data in the file.
  • Regular expression must be compatible with the python re module in the standard library.
  • Be sure that your regular expression only contains ONE and ONLY ONE capture group. For example, if you are capturing a phone number:
    • Don't do this: '(555-(867|555)-5309)'
    • Do this: '(555-(?:867|555)-5309)'
    • The former example has two capture groups, and inner and an outer.
    • The latter has one capture group (the outer). The inner is a non-capturing group as defined by starting the capture group with ?:.
  • Note: If you run into issues with loading a custom filter, try adding single-quotes around your regular expression.
• exclude_patterns:
  • List of regular expressions.
  • If a pattern match also matches any of the exclude_patterns, it will not be included in the results.'
• substitute:
  • Allows you to define what characters are removed from a string before it is passed to the sanity check(s).
  • Must be a valid regular expression.
  • If missing or empty, the default substitute is [\W_].
• sanity:
  • This is the algorithm to use with this filter in order to validate the data is really what you're looking for. For example, 16 digits might just be a random number and not a credit card. Putting the numbers through the luhn algorithm will validate they could potentially be an account number and reduce false positives.
  • You can also pass through a list of strings that represent algorithms as long as the algorithm exists in the library. If not, please add it and make a pull request!
• mask:
  • index:
    • This is the position in the matched string in which the masking will begin.
  • value
    • The string which will be used to mask the matched string.
• type:
  • This is basically a description of the 'type' of data you're looking for with this filter.

Settings

settings:
  mask: No
  log_level: INFO
  summarize: No
  output_file:
  show_matches: Yes
  delimiter:
  ignore_columns: [1, 5, 6]
  file_encoding: 'utf-8'

• bulk
  • This setting is accessible via CLI arguments -b or --bulk.
  • When those switches are used, pass the directory to scan instead of a single file name.
  • Example:

  $ txtferret scan --bulk /home/mrferret/Documents

• mask
  • If set to true, the mask value defined in the filter will be used to mask the data during output.
  • If no mask is set for a filter, the program will mask with a default mask value.
  • This is set to 'true' by default.
  • CLI - The -m switch can be used to turn on masking'

  $ txtferret scan -m ../fake_ccn_data.txt
  2017:05:20-00:24:52:-0400 PASSED sanity and matched regex - Filter: fake_ccn_account_filter, Line 1, String: 10XXXXXXXXXXXXXXXXXXX
  
  $ txtferret scan ../fake_ccn_data.txt
  2017:05:20-00:26:18:-0400 PASSED sanity and matched regex - Filter: fake_ccn_account_filter, Line 1, String: 100102030405060708094

• summarize
  • If set to true, the script will only output a list of three data points:
    • The number of matches that did not pass sanity checks.
    • The number of matches that did pass sanity checks.
    • The time it took to finish searching the file.
  • CLI - The -s switch will kickoff the summary.

  $ txtferret scan ../fake_ccn_data.txt
  2019:05:20-00:36:00:-0400 PASSED sanity and matched regex - Filter: fake_ccn_account_filter, Line 1, String: 100102030405060708094
  
  $ txtferret scan -s ../fake_ccn_data.txt
  2019:05:20-01:05:29:-0400 SUMMARY:
  2019:05:20-01:05:29:-0400   - Matched regex, failed sanity: 1
  2019:05:20-01:05:29:-0400   - Matched regex, passed sanity: 1
  2019:05:20-01:05:29:-0400 Finished in 0 seconds (~0 minutes)
  

• output_file
  • Add the absolute path to a file for script output.
  • Note - File results are put into individual log files per file scanned.
  • CLI - Use the -o switch to set an output file.

  $ txtferret scan -o my_output.log file_to_scan.txt

• show_matches
  • If this is set to 'No', then it will redact the matched string all-together in output. Be careful that you do not override this setting via a CLI switch unless it is on purpose.
• delimiter
  • Define a delimiter. This delimiter will be used to split each line from the txt file into columns. The output of the script will provide you with the column in which the regex matched in addition to the line number.
  • CAUTION: Searching by columns GREATLY slows down the script since you are applying a regular expression to each column instead of the entire line.
  • You can define a byte-code delimiter by using b followed by the code. For example, b1 will use Start of Header as a delimiter (\x01 in hex)
  • CLI - Use the -d switch to set a delimiter and scan per column instead of line.

  $ txtferret scan ../fake_ccn_data.txt
  2019:05:20-00:36:00:-0400 PASSED sanity and matched regex - Filter: fake_ccn_account_filter, Line 1, String: 100102030405060708094
  
  # Comma delimiter
  
  $ txtferret scan -d , ../fake_ccn_CSV_file.csv
  2019:05:20-01:12:18:-0400 PASSED sanity and matched regex - Filter: fake_ccn_account_filter, Line 1, String: 100102030405060708094, Column: 3

• ignore_columns
  • This setting is ignored if the delimiter setting or switch is not set.
  • Add a list of integers and txtferret will skip those columns.
  • If ignore_columns: [2, 6] is configured and a csv row is hello,world,how,are,you,doing,today, then world and doing will not be scanned but will be ignored.
  • This is particularly useful in columnar datasets when you know there is a column that is full of false positives.
• file_encoding
  • Two uses:
    • Used to encode your delimiter value to the appropriate encoding of your file.
    • Used to encode the data matched in the file before being applied to sanity check.
  • Default value is 'utf-8'

How/why did this come about?

There are a few shortcomings with commercial Data Loss Prevention (DLP) products:

• They often rely on context. It would be too noisy for a commercial solution to alert every time it matched a sixteen digit string as not all of their customers handle credit card data.
• Many vendors don't have a cheap method of handling large files over 1 GB. Even less can handle files that are many GBs in size.
• Some DLP solutions will only classify data files based on the first so many megabytes.
• Many DLP solutions are black boxes of magic that do not give you say in what they are looking for in your data or how they know what they're looking at.

Txtferret was born out after realizing some of these limitations. It isn't perfect, but it's a great sanity check which can be paired with a DLP solution. Here are some things it was designed to do:

• Virtually no size limitation
  • Can run against any size file as long as you can fit it on a drive.
  • It's python... so... no speed guarantee on huge files, but at least it will eventually get it done. We've found that txtferret can scan a ~20 GB file between 1 and 3 minutes on our systems.
• Customizable
  • Define your own regular expressions and pair them with a sanity check. For example, using the built-in luhn algorithm will sift out many false positives for credit card numbers. The matched credit card number will be run through the luhn algorithm first. If it doesn't pass, it is discarded.
• No context needed
  • Yeah, this can cause a lot of false positives with certain files; However, if you're dealing with a file that doesn't contain context like 'VISA' or 'CVE', then you need to start somewhere.
• Helpful output
  • Indicates which line the string was found.
  • Indicates which column if you've defined a delimiter (ex: comma for CSV files).
  • You can choose to mask your output data to make sure you're not putting sensitive data into your log files or outputting them to your terminal.
• It's free
  • No contracts
  • No outrageous licensing per GB of data scanned.
• You can contribute!

Releases

Version 0.3.0a - 2019-09-05

• Removed log level switch. Only matches are shown now.
• Added exclude_patterns to filters
• Adjusted result format to tab-delimited output.
• Bulk scans now write results to separate files instead of all in one file.

Version 0.2.1 - 2019-08-11

• Fixed allowed keys bug introduced in v0.2.0

Version 0.2.0 - 2019-08-05

• Changed tokenize to mask because tokenize was a lie.. it's masking.
  • Replaced --no-tokenize switch with --mask switch.
• Turned masking off by default.

Version 0.1.3 - 2019-08-05

• Added file_encoding setting for multi-encoding support.
  • Reads in bytes and assumes 'utf-8' encoding by default.

Version 0.1.2 - 2019-08-01

• Fixed bug with regex when reading gzipped files.

Version 0.1.1 - 2019-07-30

• Added substitute option to filters.

Version 0.1.0 - 2019-07-30

• Removed the config-override option.
• Added ignore_columns setting.

Version 0.0.4 - 2019-06-09

• Added bulk file scanning by the --bulk switch.
• Added multiprocessing for bulk scanning.

Version 0.0.3 - 2019-06-01

• Added gzip detection and support.

Development

Some info about development.

Running Tests

$ pytest txt-ferret/tests/

Contributing

Process

1. Create an issue.
2. Fork the repo.
3. Do your work.
4. WRITE TESTS
5. Make a pull request.
   • Preferably, include the issue # in the pull request.

Style

• Black for formatting.
• Pylint for linting.

License

See License