Error Handling in n4
$Id: errors.txt,v 1.1 2001/05/17 05:46:08 mbp Exp $

The biggest complication in error handling is that errors may occur at
several different layers of the protocol stack, and they have to be
reported at the corresponding layer.  Sun make this process more
difficult by using the same low numbers for different errors depending
on context.

In every case, we want to log the error and roll the server back to an
appropriate enclosing scope.

Also, if there is an error in a response, then we can't send an error
back to the server.  We need to pass an appropriate description to
higher level local software, which might decide for example to retry
the request.

The layers are, from the bottom up:

 * Networking error: 

    Can't even receive a whole packet correctly.  There's not much we
    can do except close the socket, since we suspect the client is not
    doing NFS, or badly desynchronized, or there is an error in some
    underlying system.

 * RPC error:

    Examples: bad RPC version or NFS version, or an incomplete
    packet, or garbage arguments.

    ONC-RPC actually has two cases here: accepted and 

    Some cases, such as truncated arguments, may not be detected until
    parsing is almost complete, but we still need to be able to report
    the error at this level.  This requires throwing away the
    partially-generated response and replacing it with an error
    message.  This should be simple if caught at the right level,
    since the output packet is in a pool and can just be discarded.

 * NFS operation failed:

    Examples: file not found, readonly filesystem, stale filehandle,
    lock expired, ...
    
    There are many of these.  Some are only supported by later
    versions of NFS, and we might have to fold them downwards for
    older versions.

    They're generated by particular sub-operations inside a compound,
    and indicate partial success.  We want to send back a completion
    packet, which has all the results up to that point.

Since the most numerous error codes at the nfsstat4 codes, we'll use
them, with extended error codes for internal use.  If we get an
internal error after processing of packet contents has started, then
we need to translate it into a lower-level error.