Hi all,
I found the cause and would like to post my findings:
The problem seems to be caused by the way Jetty and Jackrabbit interoperates
with each other conditioned by a peculiarity of the MAC OS Webdav client:
0) Jetty resuses the request input-stream instance:
org.mortbay.jetty.HttpParser$Input for new requests.
1) A PROPFIND arrives with Content-Length=123 for example. It seems however
that Jackrabbit is not interested into the xml content at all.
2) A PUT arrives with Content-Length=0 (seem to be a peculiarity of the MAC OS
Webdav client, I have no clue why MAC OS is doing this. The real payload will
be send a "few" requests later within another PUT).
3) Jetty reuses the InputStream which contains the xml content from the
previous PROPFIND request!
4) Jackrabbit uses Apache Tika in order to sleuth the content type and finds
the xml from the previous request (ignoring the fact the the Content-Length is
0)
The workaround I found proves my findings. I extended Jackrabbit's
org.apache.jackrabbit.j2ee.SimpleWebdavServlet
containing an overwritten service() method:
@Override
protected void service(HttpServletRequest request, HttpServletResponse
response) throws ServletException, IOException {
super.service(request, response);
// make sure input stream is eaten up
ServletInputStream ins = request.getInputStream();
while (ins.read() != -1);
}
Now the input-stream is always eaten up so that the following request will get
a clean input-stream.
Best regards
udo
Am 17.02.2012 um 15:48 schrieb Subscriber:
> Hi all,
>
> Uploading files via the Webdav servlet results to wrong content types under
> Mac OS X.
>
> I mount a Jackrabbit (version 2.2.5) workspace under Mac OS X and copy files
> (e.g. images) using the Finder.
> The problem is that the file is stored in the repository always as
> application/xml (javax.jcr.Property.JCR_MIMETYPE) regardless what file is
> copied.
>
> As a workaround I wrote a Servlet-Filter that sets the Content-Type according
> to the http path. This filter is processed just before the WebdavServlet.
> While the PUT operation is processed the content type is read from the http
> request header inside
>
> org.apache.jackrabbit.server.io.ImportContextImpl.ImportContextImpl(Item,
> String, InputContext, InputStream, IOListener, Detector) line 81.
>
> However the "real" mimetype is determined using Apache Tika (it seems the
> Content-Type is just a hint or so...).
>
> I have no idea why this leads always to "application/xml" when using MAC OS
> as client (maybe the stream is missing some magic bytes).
>
> Does anybody have an idea how to fix this? Is there a way to force Jackrabbit
> to use a certain content-type.
> A workaround using a filter or so would be fine for me...
>
> Thanks in advance and best regards
>
> udo
>
