jboss url decoding - apache

Question

We have a servlet hosted on jboss which works on HttpServletRequest. But sometimes we receieve requests that do not get decoded by jboss, and when we do getQueryParam on HttpServletRequest, we get null. The jboss access log shows the url in encoded form. Normally, when everything works smooth, url is shown decoded in access log.
e.g.:
This was a problematic request:
127.0.0.1 [13/Apr/2009:14:18:53 +0000] GET /redirectService//%3Fclient_id=3&redirect_url=http%253A%252F%252Fwww.amazon.de%252Fgp%252Fsearch%253Fie%253DUTF8%2526keywords%253DMicrosoft+Office+2007%2526search-alias%253Dsoftware%2526 HTTP/1.1 'null' 'Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.12)'
This was a proper request:
127.0.0.1 [13/Apr/2009:14:19:37 +0000] GET /redirectService//?client_id=3&redirect_url=http%3A%2F%2Fwww.amazon.de%2Fgp%2Fsearch%3Fie%3DUTF8%26keywords%3DMAGIX+Video+deluxe+2008%26search-alias%3Dsoftware%26 HTTP/1.1 'http://www.google.de/search?hl=de&q=magix+video+deluxe+2008&meta=&aq=3&oq=%22magix%22' 'Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 6.0; SLCC1; .NET CLR 2.0.50727; Media Center PC 5.0; .NET CLR 3.0.04506; .NET CLR 1.1.4322)'
Could we be missing some jboss decode settings, or is it just a case of malicious user?

Answer 1

Hard to tell, really.
The client seems to be decoding the question mark into "%3F" but not the ampersand. Suspicious, isn't it?. This looks like a buggy client IMO. Maybe nonportable javascript, maybe some URL-rewriting bug on the web server side, or a more esoteric cause ... a malfunctioning browser plugin.
To rule out nonportable javascript, log the user-agent and compare results. To rule out url-rewriting bug, log referer.
AFAIK, the URL decoder behavior is hardcoded. The string encoding can change if uri's get written in non-ascii or non-iso88591, but that's not what you're after. What encodes question marks but fail to encode ampersands escapes me.

Answer 2

We logged the user-agent, it is some suspicious "XXXagentXXX" in most cases, but a genuine Mozilla (as above) in others. Referrer is "-" for all these requests. However, there is one curious thing I noticed today. We redirect our requests from apache (80) to jboss. Apache access log shows above request as completely encoded:
GET /r/%3Fclient_id%3D3%26redirect_url%3Dhttp%253A%252F%252Fwww.amazon.de%252Fgp%252Fsearch%253Fie%253DUTF8%2526keywords%253DCyberlink%2BPower%2BDirector%2526search-alias%253Dsoftware HTTP/1.0" 400 965 "-" "Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.8.1.10)"
while jboss access log has everything except %3F decoded. Now this makes me think apache is screwing up somewhere in the decoding?

Answer 3

I had problem decoding URL too with JBoss 13.
I added the last line in JBoss configuration and it works now.
/subsystem=undertow/servlet-container=default:write-attribute(name=default-encoding,value="ISO-8859-15")
/subsystem=undertow/server=default-server/http-listener=default:write-attribute(name=url-charset,value="ISO-8859-15")
Doc is here if more needed : https://wildscribe.github.io/WildFly/13.0/subsystem/undertow/server/http-listener/index.html