On Fri, Jan 11, 2002 at 10:09:33AM +0100, Steffen Neumann wrote:
> AFAICT 2.4 and coda have still some issues,
> e.g. around symlinks (StarOffice 6 won't run
> due to failed opening of a library, and someone 
> else has reported a problem along these lines),
> and there was another major incident I forgot 
> details about around/after 2.4.4 Kernel

There is clear indication that something is wrong with 2.4 kernels. It
is hard to get in the first place, but once a testcase is found it seems
to be reproducable (like Ivan is doing with the symlink traversal).

There are no inherent limitations in Venus that should cause the problem
he is seeing. The readlink path is really relatively simple. 
 - kernel gets a page.
 - kernel gets upcall buffer.
 - kernel makes upcall.
 - venus possibly fetches the link contents from a server
 - venus copies the RVM copy of the link contents into the downcall
   buffer and returns it to the kernel.
 - kernel copies link contents into the allocated page and returns the
   page back to the VFS layer.

So either the page or the upcall buffer allocations fail. Or venus is
unable to fetch the link contents from the server, in which case 2.2
kernels and FreeBSD/Solaris/NetBSD should see the same problem. I cannot
see how the RVM->buffer or buffer->page copies can go wrong (so
according to Murphy's Law that is probably where the problem is ;)

The 2.4.4 incident was actually caused by the 2.4.2-RedHat kernel that
was included in the RH 7.2 release. They patched ext3 into the kernel
and set up the filesystem as ext3 by default. I had never tested the
Coda kernel module with ext3 as it hadn't been part of the official
kernel tree.

Jan