This is version 0.5 of ncpfs, a free NetWare client for Linux. For me
it works with 1.3.39, although this version has severe problems with
the socket layer. Your connection will block after you have sent 64k
of requests to the server. Alan Cox told me he would like to have that
fixed in 1.3.42 or so.

I know that this piece of software is VERY incomplete, I have to
apologize for that. But I thought I should make it publically
available, because I have tried to ask several people about the legal
status of the code. I did not get very satisfying answers. So I publish
ncpfs to open it for criticism. If nobody complains, I will go on
working.

To install ncpfs, just type 'make', 'insmod ncpfs.o' and then
'ncpmount server mount-point'. 

Please note that your IPX system has to be configured correctly. There
has to be a route to the internal network of your server. Please see
the file start_ipx for an example.

I use tools written by Greg Page, Caldera. I hope I did not do too
much harm to their business. For your convenience I included the file
ipx.tar made available by Caldera.

My main source of information is a book written in german by Manfred
Hill and Ralf Zessin, "Netzwerkprogrammierung in C", IWT Verlag GmbH,
1995, ISBN 3-88322-491-X. It contains quite a lot of typographical and
other errors, but I find it very valuable as an introduction to NCP
programming. If you know about the concepts and possibilities of NCP,
Ralph Brown's interrupt list becomes much more readable. It's much
easier to find undocumented information if you know what to look for!

For the curious: the file ncplib_user.[ch] is a library that makes it
possible to send NCP requests to the server over a mounted
directory. I use it to keep the encryption stuff out of the kernel by
logging in from user space. Look at the file ncptest.c for other
possible uses. I use ncptest to check my assumptions about the widely
undocumented NetWare Core Protocol.  Maybe this is the beginning of a
free NetWare API for Linux! I would be happy to receive your comments
on this.


LIMITATIONS (compare these with smbfs :-)

The limitations ncpfs has are the natural limitations of the NCP
protocol, which was designed with MS-DOS based PCs in mind. The first
limitation is the lack of uid, gid and permission information per
file. You have to assign those values once for a complete mounted
directory.

The second limitation is just as annoying as the first: You cannot
re-export a ncp-mounted directory by nfs. It is not possible because
the NFS protocol defines access to files through unique file handles,
which can be mapped to the device and inode numbers in unix NFS
servers. NCP does not have unique numbers per file, you only have the
path name. I implemented a caching scheme for inode numbers, which
gives unique inode numbers for every open file in the system. This is
just sufficient for local use of the files, because you can tell when
an inode number can be discarded. With NFS the situation is
different. You can never know when the client will access the file-id
you offered, so you would have to cache the inode numbers
indefinitely long. I think this should not be done in kernel mode, as
it would require an unlimited amount of RAM.
