IDS08-J. Sanitize untrusted data passed to a regex

Regular expressions are widely used to match strings of text. For example, the POSIX grep utility supports regular expressions for finding patterns in the specified text. For introductory information on regular expressions, see the Java Tutorials [[Tutorials 08]]. The java.util.regex package provides the Pattern class that encapsulates a compiled representation of a regular expression and the Matcher class, which is an engine that uses a Pattern to perform matching operations on a CharSequence.

Java's powerful regular expression (regex) facilities must be protected from misuse. An attacker may supply a malicious input that modifies the original regular expression in such a way that the regex fails to comply with the program's specification. This attack vector, called a regex injection, might affect control flow, cause information leaks, or result in denial-of-service (DoS) vulnerabilities.

Certain constructs and properties of Java regular expressions are susceptible to exploitation:

• Matching flags: Untrusted inputs may override matching options that may or may not have been passed to the Pattern.compile() method.
• Greediness: An untrusted input may attempt to inject a regex that changes the original regex to match as much of the string as possible, exposing sensitive information.
• Grouping: The programmer can enclose parts of a regular expression in parentheses to perform some common action on the group. An attacker may be able to change the groupings by supplying untrusted input.

Untrusted input should be sanitized before use to prevent regex injection. When the user must specify a regex as input, care must be taken to ensure that the original regex cannot be modified without restriction. Whitelisting characters (such as letters and digits) before delivering the user-supplied string to the regex parser is a good input sanitization strategy. A programmer must provide only a very limited subset of regular expression functionality to the user to minimize any chance of misuse.

Regex Injection Example

Suppose a system log file contains messages output by various system processes. Some processes produce public messages and some processes produce sensitive messages marked "private." Here is an example log file:

10:47:03 private[423] Successful logout  name: usr1 ssn: 111223333
10:47:04 public[48964] Failed to resolve network service
10:47:04 public[1] (public.message[49367]) Exited with exit code: 255
10:47:43 private[423] Successful login  name: usr2 ssn: 444556666
10:48:08 public[48964] Backup failed with error: 19

A user wishes to search the log file for interesting messages but must be prevented from seeing the private messages. A program might accomplish this by permitting the user to provide search text that becomes part of the following regex:

(.*? +public\[\d+\] +.*<SEARCHTEXT>.*)

However, if an attacker can substitute any string for <SEARCHTEXT>, he can perform a regex injection with the following text:

.*)|(.*

When injected into the regex, the regex becomes:

(.*? +public\[\d+\] +.*.*)|(.*.*)

This regex will match any line in the log file, including the private ones.

Noncompliant Code Example

This noncompliant code example periodically loads the log file into memory and allows clients to obtain keyword search suggestions by passing the keyword as an argument to suggestSearches().

public class Keywords {
  private static ScheduledExecutorService scheduler
      = Executors.newSingleThreadScheduledExecutor();
  private static CharBuffer log;
  private static final Object lock = new Object();

  // Map log file into memory, and periodically reload
  static
    try {
      FileChannel channel = new FileInputStream(
          "path").getChannel();

      // Get the file's size and map it into memory
      int size = (int) channel.size();
      final MappedByteBuffer mappedBuffer = channel.map(
          FileChannel.MapMode.READ_ONLY, 0, size);

      Charset charset = Charset.forName("ISO-8859-15");
      final CharsetDecoder decoder = charset.newDecoder();

      log = decoder.decode(mappedBuffer); // Read file into char buffer

      Runnable periodicLogRead = new Runnable() {
        @Override public void run() {
          synchronized(lock) { 
            try {
              log = decoder.decode(mappedBuffer);
            } catch (CharacterCodingException e) {
              // Forward to handler 
            } 
          }
        }
      };
      scheduler.scheduleAtFixedRate(periodicLogRead, 
                                    0, 5, TimeUnit.SECONDS);
    } catch (Throwable t) {
      // Forward to handler
    }
  }


  public static Set<String> suggestSearches(String search) {
    synchronized(lock) {
      Set<String> searches = new HashSet<String>();

      // Construct regex dynamically from user string
      String regex = "(.*? +public\\[\\d+\\] +.*" + search + ".*)";
  
      Pattern keywordPattern = Pattern.compile(regex);
      Matcher logMatcher = keywordPattern.matcher(log);
      while (logMatcher.find()) {
        String found = logMatcher.group(1);
        searches.add(found);
      }
      return searches;
    }  
  }

}

This code permits a trusted user to search for public log messages such as "error." However, it also allows a malicious attacker to perform the regex injection previously described.

Compliant Solution (Whitelisting)

This compliant solution filters out nonalphanumeric characters (except space and single quote) from the search string, which prevents regex injection.

public class Keywords {
  // ...
  public static Set<String> suggestSearches(String search) {
    synchronized (lock) {
      Set<String> searches = new HashSet<String>();

      StringBuilder sb = new StringBuilder(search.length());
      for (int i = 0; i < search.length(); ++i) {
        char ch = search.charAt(i);
        if (Character.isLetterOrDigit(ch) ||
            ch == ' ' ||
            ch == '\'') {
          sb.append(ch);
        }
      }
      search = sb.toString();

      // Construct regex dynamically from user string
      String regex = "(.*? +public\\[\\d+\\] +.*" + search + ".*)";
      // ...
    }
  }
}

This solution also limits the set of valid search terms. For instance, a user may no longer search for "name =" because the = character would be sanitized out of the regex.

Compliant Solution

Another method of mitigating this vulnerability is to filter out the sensitive information prior to matching. Such a solution would require the filtering to be done every time the log file is periodically refreshed, incurring extra complexity and a performance penalty. Sensitive information may still be exposed if the log format changes but the class is not also refactored to accommodate these changes.

Risk Assessment

Failing to sanitize untrusted data included as part of a regular expression can result in the disclosure of sensitive information.

Related Guidelines

Bibliography

IDS07-J. Do not pass untrusted, unsanitized data to the Runtime.exec() method            IDS09-J. Do not use locale-dependent methods on locale-dependent data without specifying the appropriate locale