Internet JUNKBUSTER Technical Information

  Name

junkbuster - The Internet Junkbuster Proxy ^TM

  Synopsis

junkbuster configfile (Version 2.0 onwards)
junkbstr.exe configfile (Windows)
junkbuster [-a] [-y] [-s] [-c] [-v]
[-u user_agent] [-r referer] [-t from]
[-b blockfile] [-j jarfile] [-l logfile]
[-w NAME=VALUE] [-x Header_text]
[-h [bind_host_address][:bind_port]]
[-f forward_host[:port]] [-d N]
[-g gw_protocol[:[gw_host][:gw_port]]]
(Version 1.4 and earlier)

  Description

junkbuster is an instrumentable proxy that filters the HTTP stream between web servers and browsers. Its main purpose is to enhance privacy.

Versions before 2.0 used command-line options; Versions from 2.0 onward use a configuration file. The following descriptions of the options first give the older command-line usage, then the new configfile line.

In Versions 2.0.1 upwards on Windows, a start-up message is printed and the configuration is read from the file junkbstr.ini if it exists and no argument was given.

All files except the configfile are checked for changes before each page is fetched, so they may edited without restarting the proxy.

Options

-b blockfile
blockfile  blockfile

Block requests to URLs matching any pattern given in the lines of the blockfile. The junkbuster instead returns status 202, indicating that the request has been accepted (though not completed), and a message identifying itself (though the browser may display only a broken image icon). (Versions before 2.0 returned an error 403 (Forbidden).) The syntax of a pattern is [domain][:port][/path] (the http:// or https:// protocol part is omitted). To decide if a pattern matches a target, the domains are compared first, then the paths.

To compare the domains, the pattern domain and the target domain specified in the URL are each broken into their components. (Components are separated by the . (period) character.) Next each of the target components is compared with the corresponding pattern component: last with last, next-to-last with next-to-last, and so on. (This is called right-anchored matching.) If all of the pattern components find their match in the target, then the domains are considered a match. Case is irrelevant when comparing domain components.

A successfully matching pattern can be an anchored substring of a target, but not vice versa. Thus if a pattern doesn't specify a domain, it matches all domains. Furthermore, when comparing two components, the components must either match in their entirety or up to a wildcard * (star character) in the pattern. The wildcard feature implements only a "prefix" match capability ("abc*" vs. "abcdefg"), not suffix matching ("*efg" vs. "abcdefg") or infix matching ("abc*efg" vs. "abcdefg"). The feature is restricted to the domain component; it is unrelated to the optional regular expression feature in the path (described below).

If a numeric port is specified in the pattern domain, then the target port must match as well. The default port in a target is port 80.

If the domain and port match, then the target URL path is checked for a match against the path in the pattern. Paths are compared with a simple case-sensitive left-anchored substring comparison. Once again, the pattern can be an anchored substring of the target, but not vice versa. A path of / (slash) would match all paths. Wildcards are not considered in path comparisons.

For example, the target URL
   the.yellow-brick-road.com/TinMan/has_no_brain
would be matched (and blocked) by the following patterns
   yellow-brick-road.com
and
   Yellow*.COM
and
   /TinM
but not
   follow.the.yellow-brick-road.com
or
   /tinman

Comments in a blockfile start with a # (hash) character and end at a new line. Blank lines are also ignored.

Lines beginning with a ~ (tilde) character are taken to be exceptions: a URL blocked by previous patterns that matches the rest of the line is let through. (The last match wins.)

Patterns may contain POSIX regular expressions provided the junkbuster was compiled with this option (the default in Version 2.0 on). The idiom /*.*/ad can then be used to match any URL containing /ad (such as http://nomatterwhere.com/images/advert/g3487.gif for example). These expressions don't work in the domain part.

In version 1.3 and later the blockfile and cookiefile are checked for changes before each request.

-w NAME=VALUE
wafer  NAME=VALUE

Specifies a pair to be sent as a cookie with every request to the server. (Such boring cookies are called wafers.) This option may be called more than once to generate multiple wafers. The original Netscape specification prohibited semi-colons, commas and white space; these characters will be URL-encoded if used in wafers. The Path and Domain attributes are not currently supported.

-c cookiefile
cookiefile  cookiefile

Enforce the cookie management policy specified in the cookiefile. If this option is not used all cookies are silently crunched, so that users who never want cookies aren't bothered by browsers asking whether each cookie should be accepted. However, cookies can still get through via JavaScript and SSL, so alerts should be left on.

In Version 1.2 and later this option must be followed by a filename containing instructions on which sites are allowed to receive and set cookies. By default cookies are dropped in both the browser's request and the server's response, unless the URL requested matches an entry in the cookiefile. The matching algorithm is the same as for the blockfile. A leading > character allows server-bound cookies only; a < allows only browser-bound cookies; a ~ character stops cookies in both directions. Thus a cookiefile containing a single line with the two characters >* will pass on all cookies to servers but not give any new ones to the browser.

-j jarfile
jarfile  jarfile

All Set-cookie attempts by the server are logged to jarfile. If no wafer is specified, one containing a canned notice (the vanilla wafer) is added as an alert to the server unless the suppress-vanilla-wafer option is invoked.

-v
suppress-vanilla-wafer

Suppress the vanilla wafer.

-t from
from  from

If the browser discloses an email address in the FROM header (most don't), replace it with from. If from is set to . (the period character) the FROM is passed to the server unchanged. The default is to delete the FROM header.

-r referer
referer  referer

Whenever the browser discloses the URL that led to the current request, replace it with referer. If referer is set to . (period) the URL is passed to the server unchanged. In Version 1.4 and later, if referer is set to @ (at) the URL is sent in cases where the cookiefile specifies that a cookie would be sent. (No way to send bogus referers selectively is provided.) The default is to delete Referer.

Version 2.0 also accepts the spelling referrer, which most dictionaries consider correct.

-u user-agent
user-agent  user-agent

Information disclosed by the browser about itself is replaced with the value user-agent. If user-agent is set to . (period) the User-Agent header is passed to the server unchanged, along with any UA headers produced by MS-IE (which would otherwise be deleted). In Version 1.4 and later, if user-agent is set to @ (at) these headers are sent unchanged in cases where the cookiefile specifies that a cookie would be sent, otherwise only default User-Agent header is sent. That default is Mozilla/3.0 (Netscape) with an unremarkable Macintosh configuration. If used with a browser less advanced than Mozilla/3.0 or IE-3, the default may encourage pages containing extensions that confuse the browser.

-h [host][:port]
listen-address  [host][:port]

If host is specified, bind the junkbuster to that IP address. If a port is specified, use it. The default port is 8000; the default host is localhost. Before Version 2.0.2, the default was to bind to all IP addresses (INADDR_ANY); but this has been restricted to localhost to avoid unintended security breaches. (To open the proxy to all, use the line
   listen-address :8000
in the configuration file.)

-f forward_host[:port]
forwardfile  forwardfile

Version 1.X required all HTTP requests from the client to be forwarded to the same destination. Version 2.0 takes its routing specification from a forwardfile, allowing selection of the proxy (a.k.a. forwarding host) and gateway according to the URL. Here is a typical line.

*         lpwa.com:8000      .      .

Each line contains four fields: target, forward_to, via_gateway_type and gateway. As usual, the last target domain that matches the requested URL wins, and the * character alone matches any domain. The target domain need not be a fully qualified hostname; it can be a general domain such as com or co.uk or even just a port number. For example, because LPWA does not handle SSL, the line above will typically be followed by a line such as

:443  .      .      .

to allow SSL transactions to proceed directly. The cautious would also add an entry in their blockfile to stop transactions to port 443 for all but specified trusted sites.

If the winning forward_to field is . (the dot character) the proxy connects directly to the server given in the URL, otherwise it forwards to the host and port number specified. The default port is 8000. The via_gateway_type and gateway fields also use a dot to indicate no gateway protocol. The gateway protocols are explained below.

The example line above in a forwardfile alone would send everything through port 8000 at lpwa.com with no gateway protocol, and is equivalent to the old -f lpwa.com:8000 with no -g option. For more information see the example file provided with the distribution.

Configure with care: no loop detection is performed. When setting up chains of proxies that might loop back, try adding Squid.

-g gw_protocol[:[gw_host][:gw_port]]

Use gw_protocol as the gateway protocol. This option was introduced in Version 1.4, but was folded into the forwardfile option in Version 2.0. The default is to use no gateway protocol; this may be explicitly specified as direct on the command line or the dot character in the forwardfile. The SOCKS4 protocol may be specified as socks or socks4. The SOCKS4A protocol is specified as socks4a. The SOCKS5 protocol is not currently supported. The default SOCKS gw_port is 1080.

The user's browser should not be configured to use SOCKS; the proxy conducts the negotiations, not the browser.

The user identification capabilities of SOCKS4 are deliberately not used; the user is always identified to the SOCKS server as userid=anonymous. If the server's policy is to reject requests from anonymous, the proxy will not work. Use a debug value of 3 to see the status returned by the server.

-d N
debug  N

Set debug mode. The most common value is 1, to pinpoint offensive URLs, so they can be added to the blockfile. The value of N is a bitwise logical-OR of the following values:
1 = URLs (show each URL requested by the browser);
2 = Connections (show each connection to or from the proxy);
4 = I/O (log I/O errors);
8 = Headers (as each header is scanned, show the header and what is done to it);
16 = Log everything (including debugging traces and the contents of the pages).
Multiple debug lines are permitted; they are logical OR-ed together.

Because most browsers send several requests in parallel the debugging output may appear intermingled, so the single-threaded option is recommended when using debug with N greater than 1.

-y
add-forwarded-header

Add X-Forwarded-For headers to the server-bound HTTP stream indicating the client IP address to the server, in the new style of Squid 1.1.4. If you want the traditional HTTP_FORWARDED response header, add it manually with the -x option.

-x HeaderText
add-header  HeaderText

Add the HeaderText verbatim to requests to the server. Typical uses include adding old-style forwarding notices such as Forwarded: by http://pro-privacy-isp.net and reinstating the Proxy-Connection: Keep-Alive header (which the junkbuster deletes so as not to reveal its existence). No checking is done for correctness or plausibility, so it can be used to throw any old trash into the server-bound HTTP stream. Please don't litter.

-s
single-threaded

Doesn't fork() a separate process (or create a separate thread) to handle each connection. Useful when debugging to keep the process single threaded.

-l logfile
logfile  logfile

Write all debugging data into logfile. The default logfile is the standard output.

aclfile  aclfile

Unless this option is used, the proxy talks to anyone who can connect to it, and everyone who can has equal permissions on where they can go. An access file allows restrictions to be placed on these two policies, by distinguishing some source IP addresses and/or some destination addresses. (If a forwarder or a gateway is being used, its address is considered the destination address, not the ultimate IP address of the URL requested.)

Each line of the access file begins with either the word permit or deny followed by source and (optionally) destination addresses to be matched against those of the HTTP request. The last matching line specifies the result: if it was a deny line or if no line matched, the request will be refused.

A source or destination can be specified as a single numeric IP address, or with a hostname, provided that the host's name can be resolved to a numeric address: this cannot be used to block all .mil domains for example, because there is no single address associated with that domain name. Either form may be followed by a slash and an integer N, specifying a subnet mask of N bits. For example, permit 207.153.200.72/24 matches the entire Class-C subnet from 207.153.200.0 through 207.153.200.255. (A netmask of 255.255.255.0 corresponds to 24 bits of ones in the netmask, as with *_MASKLEN=24.) A value of 16 would be used for a Class-B subnet. A value of zero for N in the subnet mask length will cause any address to match; this can be used to express a default rule. For more information see the example file provided with the distribution.

If you like these access controls you should probably have firewall; they are not intended to replace one.

trustfile  trustfile

This feature is experimental, has not been fully documented and is very subject to change. The goal is for parents to be able to choose a page or site whose links they regard suitable for their young children and for the proxy to allow access only to sites mentioned there. To do this the proxy examines the referer variable on each page request to check they resulted from a click on the ``trusted referer'' site: if so the referred site is added to a list of trusted sites, so that the child can then move around that site. There are several uncertainties in this scheme that experience may be able to iron out; check back in the months ahead.

trust_info_url  trust_info_url

When access is denied due to lack of a trusted referer, this URL is displayed with a message pointing the user to it for further information.

hide-console

In the Windows version only, instructs the program to disconnect from and hide the command console after starting.

-a

(Obsolete) Accept the server's Set-cookie headers, passing them through to the browser. This option was removed in Version 1.2 and replaced by an improvement to the -c option.

  Installation and Use

Browsers must be told where to find the junkbuster (e.g. localhost port 8000). To set the HTTP proxy in Netscape 3.0, go through: Options; Network Preferences; Proxies; Manual Proxy Configuration; View. See the FAQ for other browsers. The Security Proxy should also be set to the same values, otherwise shttp: URLs won't work.

Note the limitations explained in the FAQ.

  Checking Options

To allow users to check that a junkbuster is running and how it is configured, it intercepts requests for any URL ending in /show-proxy-args and blocks it, returning instead returns information on its version number and current configuration including the contents of its blockfile. To get an explicit warning that no junkbuster intervened if the proxy was not configured, it's best to point it to a URL that does this, such as http://internet.junkbuster.com/cgi-bin/show-proxy-args on Junkbusters's website.

  See Also

http://www.junkbusters.com/ht/en/ijbfaq.html
http://www.junkbusters.com/ht/en/cookies.html
http://internet.junkbuster.com/cgi-bin/show-proxy-args
http://www.cis.ohio-state.edu/htbin/rfc/rfc2109.html
http://squid.nlanr.net/Squid/
http://www-math.uni-paderborn.de/~axel/

  Copyright and GPL

Written and copyright by the Anonymous Coders and Junkbusters Corporation and made available under the GNU General Public License (GPL). This software comes with NO WARRANTY. Internet Junkbuster Proxy is a trademark of Junkbusters Corporation.