> Its potentially going to stop umounts, and file write accesses.
>
> If you think this is the case, the following should find which device
> the problem is on (untested)
I will test that patch and see what I can get from it.
Something worth noting: I discovered this trying to debug
{get,put}_write_access, and I did some printing of dentry names. It
turned out that every time things went wrong, the dentry name was /
(and the parent was self).
Therefore, I suspect pipe.c:do_pipe() and socket.c:get_fd().
/Magnus
[EMAIL PROTECTED]