logging Formatter to sanitize CRLF errors (CWE-93, some forms of CWE-117)

This class is a drop-in replacement for logging.Formatter, and has the exact same construction arguments. However, as a final step of formatting a log line, it escapes carriage returns (\r) and linefeeds (\n).

By default, these are replaced with their escaped equivalents (see Examples), but the replacements dictionary can be modified to change this behavior.

This sanitization should solve CWE-93 errors and CRLF-based versions of CWE-117. Some CWE-117 errors are concerns about e.g. XSS flaws in logs that are likely to be viewed in a browser; this formatter can't handle every form of CWE-117.

pip install logging-formatter-anticrlf

import sys
import logging
import anticrlf

handler = logging.StreamHandler(sys.stderr)
handler.setFormatter(anticrlf.LogFormatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s'))

logger = logging.getLogger(__name__)
logger.addHandler(handler)
logger.setLevel(logging.INFO)

logger.info("Example text with a newline\nhere")

This results in:

2017-02-03 08:43:52,557 - __main__ - INFO - Example text with a newline\nhere

Whereas with the default Formatter, it would be:

2017-02-03 08:43:52,557 - __main__ - INFO - Example text with a newline
here

If you wanted newlines to be replaced with \x0A instead, you could:

formatter = anticrlf.LogFormatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
formatter.replacements["\n"] = "\\x0A"  # Note the double backslash for literal!
handler.setFormatter(formatter)

The replacements field of the formatter is a SubstitutionMap object that behaves like a dict with a few exceptions designed to help developers avoid making insecure mistakes.

Specifically:

• an "empty" SubstitutionMap object will contain the default mappings for CR and LF chars
• calling del on either the CR or LF key will reset the value rather than delete the key
• any attempt to create a key-value pair that results in any value containing any of the keys will raise an UnsafeSubstitutionError

The rationale for the last item is that the keys of the replacements field are strings that are considered unsafe. Replacing one unsafe string with another defeats the purpose of using this module.

Additionally, if you assign a regular dict to the replacements field, and try to log something using that configuration, anticrlf.LogFormatter will reset the replacements field to its default value and issue a UserWarning to that effect.

That means the following:

formatter.replacements["\n"] = "\\x0A"  # replace LF chars with '\x0A'
del formatter.replacements["\n"]        # return to replacing LF with '\n'
formatter.replacements["\t"] = "\\t"    # replace tabs with '\t'
formatter.replacements["\n"] = "<\t>"   # raises UnsafeSubstitutionError

The last occurs because the value <\t> contains \t, which was previously created as a key.

And:

formatter.replacements = { "\n": "\r" }  # this is a mistake!
logger.info("example")

Will result, if that logger is using that formatter, in replacements being returned to its safe default value and a UserWarning about that being issued.