Most Robust way of reading a file or stream using Java (to prevent DoS attacks)

Question

Currently I have the below code for reading an InputStream. I am storing the whole file into a StringBuilder variable and processing this string afterwards.

public static String getContentFromInputStream(InputStream inputStream)
// public static String getContentFromInputStream(InputStream inputStream,
// int maxLineSize, int maxFileSize)
{

    StringBuilder stringBuilder = new StringBuilder();
    BufferedReader bufferedReader = new BufferedReader(new InputStreamReader(inputStream));
    String lineSeparator = System.getProperty("line.separator");
    String fileLine;

    boolean firstLine = true;
    try {
        // Expect some function which checks for line size limit.
        // eg: reading character by character to an char array and checking for
        // linesize in a loop until line feed is encountered.
        // if max line size limit is passed then throw an exception
        // if a line feed is encountered append the char array to a StringBuilder
        // after appending check the size of the StringBuilder
        // if file size exceeds the max file limit then throw an exception

        fileLine = bufferedReader.readLine();

        while (fileLine != null) {
            if (!firstLine) stringBuilder.append(lineSeparator);
            stringBuilder.append(fileLine);
            fileLine = bufferedReader.readLine();
            firstLine = false;
        }
    } catch (IOException e) {
        //TODO : throw or handle the exception
    }
    //TODO : close the stream

    return stringBuilder.toString();

}

The code went for a review with the Security team and the following comments were received:

1. BufferedReader.readLine is susceptible to DOS (Denial of Service) attacks (line of infinite length, huge file containing no line feed/carriage return)

2. Resource exhaustion for the StringBuilder variable (cases when a file containing data greater than the available memory)

Below are the solutions I could think of:

1. Create an alternate implementation of readLine method (readLine(int limit)), which checks for the no. of bytes read and if it exceeds the specified limit, throw a custom exception.

2. Process the file line by line without loading the file in entirety. (pure non-Java solution :) )

Please suggest if there are any existing libraries which implement the above solutions. Also suggest any alternate solutions which offer more robustness or are more convenient to implement than the proposed ones. Though performance is also a major requirement, security comes first.

Answer 1

Updated Answer

You want to avoid all sorts of DOS attacks (on lines, on size of the file, etc). But in the end of the function, you're trying to convert the entire file into one single String!!! Assume that you limit the line to 8 KB, but what happens if somebody sends you a file with two 8 KB lines? The line reading part will pass, but when finally you combine everything into a single string, the String will choke all available memory.

So since finally you're converting everything into one single String, limiting line size doesn't matter, nor is safe. You have to limit the entire size of the file.

Secondly, what you're basically trying to do is, you're trying to read data in chunks. So you're using BufferedReader and reading it line-by-line. But what you're trying to do, and what you really want at the end - is some way of reading the file piece by piece. Instead of reading one line at a time, why not instead read 2 KB at a time?

BufferedReader - by its name - has a buffer inside it. You can configure that buffer. Let's say you create a BufferedReader with buffer size of 2 KB:

BufferedReader reader = new BufferedReader(..., 2048);

Now if the InputStream that you pass to BufferedReader has 100 KB of data, BufferedReader will automatically read it 2 KB at at time. So it will read the stream 50 times, 2 KB each (50x2KB = 100 KB). Similarly, if you create BufferedReader with a 10 KB buffer size, it will read the input 10 times (10x10KB = 100 KB).

BufferedReader already does the work of reading your file chunk-by-chunk. So you don't want to add an extra layer of line-by-line above it. Just focus on the end result - if your file at the end is too big (> available RAM) - how are you going to convert it into a String at the end?

One better way is to just pass things around as a CharSequence. That's what Android does. Throughout the Android APIs, you will see that they return CharSequence everywhere. Since StringBuilder is also a subclass of CharSequence, Android will internally use either a String, or a StringBuilder or some other optimized string class based on the size/nature of input. So you could rather directly return the StringBuilder object itself once you've read everything, rather than converting it to a String. This would be safer against large data. StringBuilder also maintains the same concept of buffers inside it, and it will internally allocate multiple buffers for large strings, rather than one long string.

So overall:

• Limit the overall file size since you're going to deal with the entire content at some point. Forget about limiting or splitting lines
• Read in chunks

Using Apache Commons IO, here is how you would read data from a BoundedInputStream into a StringBuilder, splitting by 2 KB blocks instead of lines:

// import org.apache.commons.io.output.StringBuilderWriter;
// import org.apache.commons.io.input.BoundedInputStream;
// import org.apache.commons.io.IOUtils;

BoundedInputStream boundedInput = new BoundedInputStream(originalInput, <max-file-size>);
BufferedReader reader = new BufferedReader(new InputStreamReader(boundedInput), 2048);

StringBuilder output = new StringBuilder();
StringBuilderWriter writer = new StringBuilderWriter(output);

IOUtils.copy(reader, writer); // copies data from "reader" => "writer"
return output;

Original Answer

Use BoundedInputStream from Apache Commons IO library. Your work becomes much more easier.

The following code will do what you want:

public static String getContentFromInputStream(InputStream inputStream) {
  inputStream = new BoundedInputStream(inputStream, <number-of-bytes>);
  // Rest code are all same

You just simply wrap your InputStream with a BoundedInputStream and you specify a maximum size. BoundedInputStream will take care of limiting reads up to that maximum size.

Or you can do this when you're creating the reader:

BufferedReader bufferedReader = new BufferedReader(
  new InputStreamReader(
    new BoundedInputStream(inputStream, <no-of-bytes>)
  )
);

Basically what we're doing here is, we're limiting the read size at the InputStream layer itself, rather than doing that when reading lines. So you end up with a reusable component like BoundedInputStream which limits reading at the InputStream layer, and you can use that wherever you want.

Edit: Added footnote

Edit 2: Added updated answer based on comments

Answer 2

There are basically 4 ways to do file processing:

1. Stream-Based Processing (the java.io.InputStream model): Optionally put a bufferedReader around the stream, iterate & read the next available text from the stream (if no text is available, block until some becomes available), process each piece of text independently as it's read (catering for widely-varying sizes of text pieces)

2. Chunk-Based Non-Blocking Processing (the java.nio.channels.Channel model): Create a set of fixed-sized buffers (representing the "chunks" to be processed), read into each of the buffers in turn without blocking (nio API delegates to native IO, using fast O/S-level threads), your main processing thread picks each buffer in turn once it is filled and processes the fixed-size chunk, as other buffers continue to be asynchronously loaded.

3. Part File Processing (including line-by-line processing) (can leverage (1) or (2) to isolate or build up each "part"): break your file format down into semantically meaningful sub-parts (if possible! breaking into lines could be possible!), iterate through stream pieces or chunks and build-up content in memory until the next part is completely built, process each part as soon as it's built.

4. Entire File Processing (the java.nio.file.Files model): Read the entire file into memory in one operation, process the complete contents

Which one should you use?
It depends - on your file contents and the type of processing you require.
From a resource-use efficiency perspective (best to worst) is: 1,2,3,4.
From a processing speed & efficiency perspective (best to worst) is: 2,1,3,4.
From an ease of programming perspective (best to worst): 4,3,1,2.
However, some types of processing might require more than the smallest piece of text (ruling out 1, and maybe 2) and some file formats may not have internal parts (ruling out 3).

You're doing 4. I suggest you shift to 3 (or lower), if you can.

Under 4, there's only one way to avoid DOS - limit the size before it's read into memory, (or for that matter copied to your file system). It's too late once it's read in. If this is not possible, then try 3, 2 or 1.

Limiting File Size

Often the file is uploaded via a HTML form.

If uploading using Servlet @MultipartConfig annotation and request.getPart().getInputStream(), you have control over how much data you read from the stream. Also, request.getPart().getSize() returns the file size in advance and if it's small enough, you can do request.getPart().write(path) to write the file to disk.

If uploading using JSF, then JSF 2.2 (very new) has the standard html component <h:inputFile> (javax.faces.component.html.InputFile), which has an attribute for maxLength; pre-JSF 2.2 implementations have similar custom components (e.g. Tomahawk has <t:InputFileUpload> with maxLength attribute; PrimeFaces has <p:FileUpload> with sizeLimit attribute).

Alternatives to Read Entire File

Your code which uses InputStream, StringBuilder, etc, is an efficient way to read the entire file, but is not necessarily the simplest way (least lines of code).

Junior/average developers could get the misapprehension that you're doing efficient stream-based processing, when you're processing the entire file - so include appropriate comments.

If you want less code, you could try one of the following:

 List<String> stringList = java.nio.file.Files.readAllLines(path, charset);

 or 

 byte[] byteContents =  java.nio.file.Files.readAllBytes(path);

But they require care, or they could be inefficient in resource usage. If you use readAllLines and then concatenate the List elements into a single String, then you would consume double the memory (for the List elements + the concatenated String). Similarly, if you use readAllBytes, followed by encoding to String (new String(byteContents, charset)), then again, you're using "double" the memory. So best to process directly against List<String> or byte[], unless you limit your files to a small enough size.

Answer 3

instead of readLine use read which reads a given amount of chars.

in each loop check how much data has been read, if it's more then a certain amount, more then the maximum of an expected input, stop it and return an error and log it.

Answer 4

I faced a similar issue when copying a huge binary file (which generally does not contain newline character). doing a readline() leads to reading the entire binary file into one single string causing OutOfMemory on Heap space.

Here is a simple JDK alternative:

public static void main(String[] args) throws Exception
{
    byte[] array = new byte[1024];
    FileInputStream fis = new FileInputStream(new File("<Path-to-input-file>"));
    FileOutputStream fos = new FileOutputStream(new File("<Path-to-output-file>"));
    int length = 0;
    while((length = fis.read(array)) != -1)
    {
        fos.write(array, 0, length);
    }
    fis.close();
    fos.close();
}

Things to note:

• The above example copies the file using a buffer of 1K bytes. However, if you are doing this copy over network, you may want to tweak the buffer size.

• If you would like to use FileChannel or libraries like Commons IO, just make sure that the implementation boils down to something like above

Answer 5

This worked for me without any problems.

    char charArray[] = new char[ MAX_BUFFER_SIZE ];
    int i = 0;
    int c = 0;
    while((c = br.read()) != -1 && i < MAX_BUFFER_SIZE) {
        char character = (char) c;
        charArray[i++] = character;
   }
   return Arrays.copyOfRange(charArray,0,i); 

Answer 6

I cannot think a soloution other than Apache Commons IO FileUtils. Its pretty simple with FileUtils class, as the so called DOS attack wont come directly from the top layer. Reading and writing a file is very much simple as you can do it with just one line of code like

String content =FileUtils.readFileToString(new File(filePath));

You can explore more about this.

Answer 7

There is class EntityUtils under Apache httpCore. Use getString() method of this class to get the String from Response content.

Answer 8

Recommendations from Fortify Scan. You can adapt the InputStream to other resources such as HTTP request InputStream.

InputStream zipInput = zipFile.getInputStream(zipEntry);
Reader zipReader = new InputStreamReader(zipInput);
BufferedReader br = new BufferedReader(zipReader);
StringBuffer sb = new StringBuffer();
int intC;
while ((intC = br.read()) != -1){
    char c = (char)intC;
    if (c == "\n"){
       break;
    }
    if (sb.length >= MAX_STR_LEN){
       throw new Exception("Input too long");
    }
    sb.append(c);
}
String line = sb.toString();