This site is the archived OWASP Foundation Wiki and is no longer accepting Account Requests.
To view the new OWASP Foundation website, please visit

Difference between revisions of "PDF Attack Filter for Apache mod rewrite"

Jump to: navigation, search
(final touchups and removed stub as its ready!)
Line 304: Line 304:
[[Category:How To]]
[[Category:How To]]
[[Category:OWASP Java Project]]

Revision as of 08:01, 19 March 2007


This is a filter to block XSS attacks on pdf files, as originally discovered by Stefano Di Paola and Giorgio Fedon, served by Apache with mod_rewrite installed. After developing an initial implementation, I found it to be almost identical to an algorithm developed by Amit Klein. After further inspection, I was able to optimize the algorithm even more as explained below.

So far I've seen two real-world implementations of this algorithm. One for Java EE right here in OWASP and a second one developed by F5 iRules.

Adobe has published their official server-side workarounds. The Adobe solutions work by changing either the Content-Type or Content-Disposition headers in order to force the pdf to be downloaded avoiding the Adobe plug-in. I think the "Content-Type" change is less desirable as I've seen it working improperly in some cases. The downside of either implementation is that for long-established commercial websites, this changes the effective functionality of the site in that pdf's (on PC's configured with Adobe plug in), no longer open up in the browser.

Ideally, in order not to lose the functionality, we can clear the "anchor/fragment" off of the original URL by forcing a redirect to an URL containing a "hard to reproduce" but verifiable token in the query string (or even in the location).

More details of the attack are discussed elsewhere. The software in the public domain to make it easy for anyone to use for any purpose.


  1. OWASP Java EE entry for structure and skeleton contents.
  2. Algorithm developed by Amit Klein.
  3. Base64 code.
  4. OpenSSL encryption code. Vinayak Hegde.
  5. Adobe Acrobat Reader Plugin - Multiple Vulnerabilities. Stefano Di Paola
  6. Server-side workarounds. Adobe


My initial approach was almost the same as the approach described for the Java EE implementation. The main difference was that in the case that a request arrives with a query string, but one which doesn't match our generated token, instead of setting the Content-Disposition and forcing a download of the file, we redirect to the same URL containing a newly generated token in the query string.

On further study, I simplified the approach, hopefully maintaining the same level of security. This consists of at most a single rewrite based on two conditions, the file is a pdf and it has a correctly generated token in the query string. If the two conditions are not met, we redirect to the same pdf but with a valid token in the query string. This method skips step 2 and 3 of the Ami algorithm. Purposefully the query string is cleared out on the redirect and only the token is passed. The reason is to avoid possible other hacks playing around with the query string. Unless the pdf is dynamically created and depends on get parameters, this shouldn't be a problem. Certainly it is possible to implement a version of rewrite rules which will allow us also to maintain the original query string, without difficulty.


Apache mod_rewrite rules

	RewriteEngine   On
	RewriteLog "logs/rewrite.log"
	RewriteLogLevel 10

	RewriteMap tokenize prg:/home/jon/tokenize

        RewriteCond %{REQUEST_URI} .pdf
        RewriteCond ${tokenize:%{REMOTE_ADDR}%{QUERY_STRING}} !^$
        RewriteRule ^(.*)$ $1?${tokenize:%{REMOTE_ADDR}} [R,L]

The first rewrite condition matches all pdf files. The second condition matches all requests whose query string doesn't match our generated token, which is created by passing the client IP address %{REMOTE_ADDR} and the query string supplied in the request %{QUERY_STRING}.

Assuming we are requesting a pdf file, there are different cases that the second part of this rule handles.

  1. First time request without any query string, e.g.
    tokenize is called with the client ip address (the query string is empty), e.g.
    and returns a new token
    As the value is not zero, we send back a redirect to
  2. Request comes with a valid token, such as:
    tokenize is called with just client ip address and query string, e.g.
    and tokenize will return an empty string. At this point the second RewriteCond will fail and the request will pass through to normal Apache processing.
  3. Request comes in with a query string or an invalid token, such as:
    In this case, tokenize will be passed client ip address and query string, for example:
    and the return value will be the correct token
    RewriteCond will evaluate true and we will perform the redirect as in the first case.

Apache mod_rewrite rules - Amit version

Here is a previous version of rewrite rules, using the same tokenize, much more similar to Amit's algorithm even if I think that it changes nothing in respect to the actual functionality.

	RewriteEngine   On
	RewriteLog "logs/rewrite.log"
	RewriteLogLevel 10

	RewriteMap tokenize prg:/home/jon/tokenize

	RewriteCond %{REQUEST_URI} .pdf
	RewriteCond %{QUERY_STRING} ^$
	RewriteRule ^(.*)$ $1?${tokenize:%{REMOTE_ADDR}} [R,L]

	RewriteCond %{REQUEST_URI} .pdf
	RewriteRule ^(.*)$ $1?${tokenize:%{REMOTE_ADDR}%{QUERY_STRING}}

	RewriteCond %{REQUEST_URI} .pdf
	RewriteCond %{QUERY_STRING} !^$
	RewriteRule ^(.*)$ $1?${tokenize:%{REMOTE_ADDR}} [R,L]

tokenize - mod_rewrite prg

Now on to "tokenize". This is a mod_rewrite RewriteMap external program. It is run on initialization of the rewrite engine and loops on stdin where it receives text to transform, delimited by newlines. Output is on stdout, terminated by newline. Our version of tokenize interprets the input text as two pieces, some client-based info, such as client-ip or other, and a possible token to check, for example

Tokenize will generate a token using all of the text before "TOKENDELIMITER" + any extra info such as current time (maybe taking out minutes). If the token generated matches the stuff after "TOKENDELIMITER" it will write a blank line on stdout, otherwise it will write out the new token on stdout, e.g.


"C" source code

This code has been only minimally tested. Please help verify the approach and the implementation used here.

Note: In order to use, you must fill in key and initial value (key & iv), and of course these should be kept secret. Also you may want to modify how time/date are used as well as any other system variables you might want to add in.

 *  This software is in the public domain with no warranty.
 * @author     Jon Zaid 
 * @created    January 14, 2007

#include <time.h>
#include <openssl/blowfish.h>
#include <openssl/evp.h>
#include <openssl/blowfish.h>
#include <openssl/evp.h>
#include <fcntl.h>
#include <stdio.h>
#include <sys/stat.h>
#include <sys/types.h>

#define IP_SIZE 1024
#define OP_SIZE 1032
#include <stdio.h>

#include <string.h>
#include <errno.h>
#include <stdlib.h>


encode a buffer using base64 - simple and slow algorithm. null terminates
the result.
Code taken from
static void base64_encode(char *buf, int len, char *out)
	char *b64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
	int bit_offset, byte_offset, idx, i;
	unsigned char *d = (unsigned char *)buf;
	int bytes = (len*8 + 5)/6;

	memset(out, 0, bytes+1);

	for (i=0;i<bytes;i++) {
		byte_offset = (i*6)/8;
		bit_offset = (i*6)%8;
		if (bit_offset < 3) {
			idx = (d[byte_offset] >> (2-bit_offset)) & 0x3F;
		} else {
			idx = (d[byte_offset] << (bit_offset-2)) & 0x3F;
			if (byte_offset+1 < len) {
				idx |= (d[byte_offset+1] >> (8-(bit_offset-2)));
		out[i] = b64[idx];

unsigned char key[16] = { xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx};
unsigned char iv[8] = {xx, xx, xx, xx, xx, xx, xx, xx};

/* code taken from

encrypt_str (char *in, char *out)
	int olen, tlen, n;
	char outbuf[IP_SIZE];
	EVP_CIPHER_CTX_init (&ctx);
	EVP_EncryptInit (&ctx, EVP_bf_cbc (), key, iv);

	if (EVP_EncryptUpdate (&ctx, outbuf, &olen, in, strlen(in)) != 1)
	    printf ("error in encrypt update\n");
	    return 0;

	if (EVP_EncryptFinal (&ctx, outbuf + olen, &tlen) != 1)
	    printf ("error in encrypt final\n");
	    return 0;
	olen += tlen;
	base64_encode(outbuf, olen, out);

	EVP_CIPHER_CTX_cleanup (&ctx);
	return 1;

#define MAX_TOKEN_LEN	1024
/* in contains the request buffer which is an arbitrary string possibly
   followed by a delimited token (delimted "TOKEN")
   we use the time as well as the string up to TOKEN to delimit it
gen_token(char *in, char *token64, int token64_len)
time_t now;
char token[MAX_TOKEN_LEN];
int len;
char *pToken;

	/* copy first part of buffer to token */
	if (pToken = strstr(in, TOKEN_PATTERN))
		len = pToken - in;
		len = strlen(in);
	if (len >= MAX_TOKEN_LEN)
		len = MAX_TOKEN_LEN-1;
	strncpy(token, in, len);
	token[len] = '\0';

	/* now add time to it */
	now = time(NULL);
	now =  now - (now % TOKEN_EXPIRY_TIME);
	snprintf(&token[len], sizeof(token)-len-1, "%d",  now);
	encrypt_str(token, token64);

mod_rewrite prg handler
Requests arrive on stdin delimter by \n
Single line reply for every request

	generate a token based on input buffer contents and any other sysinfo
	we want;
	if request buffer already contains a token compare it with the 
	generated one;
	is (generated same as passed token)
		return emptyline;
		return generated token;


    while (1)
	{ /* for all requests */
	char reqBuffer[4096];
	char c;
	int len;
	int charsRead;
	char token64[MAX_TOKEN_LEN*2];
	char *pToken = reqBuffer;

	len = 0;
	while (    ((charsRead = read(0, &c, 1)) == 1) 
		&& (len < sizeof(reqBuffer)-2)) 
	    if (c == '\n') /* end of single request? */
	    reqBuffer[len++] = c;
	if (charsRead < 0) break; /* exiting from prg/apache? */
	reqBuffer[len] = '\0';

	/* based on contents of request, generate a string token. Must be
	   zero-delimted and ASCII printable chars, e.g. base64 */
	gen_token(reqBuffer, token64, sizeof(token64));

	/* find pointer to starting of actual encrypted token in 
	   request buffer, e.g. 
	   we will point to "encryptedtoken".
	if (pToken = strstr(reqBuffer, TOKEN_PATTERN))
		pToken += strlen(TOKEN_PATTERN);

	/* compare newly generated token with one that is passed
	   and if not same output the newly generated token, else
	   don't output anything */
	if (!pToken || strcmp(token64, pToken))
		printf("%s %s\n", token64, pToken?pToken:"NULL");
		write(1, TOKEN_PATTERN, strlen(TOKEN_PATTERN));
		write(1, token64, strlen(token64));
	write(1, "\n", 1);
	} /* for all requests */