Walk(9P) says: "The walk request carries as arguments an existing fid and a
proposed newfid (which must not be in use unless it is the same as fid"

This patch returns an error if the newfid is already in use, unless it
is the same as fid.

Signed-off-by: David Scott <dave.scott@unikernel.com>

The Remove(9P) man page says: "The remove request asks the file server
both to remove the file represented by fid and to clunk the fid, even if
the remove fails."

This patch includes a unit test which checks this behaviour.

Signed-off-by: David Scott <dave.scott@unikernel.com>

This is the start of a series of patches where we associate more
server-side state with fids.

Signed-off-by: David Scott <dave.scott@unikernel.com>

Note we are not yet storing these open handles.

Signed-off-by: David Scott <dave.scott@unikernel.com>

Signed-off-by: David Scott <dave.scott@unikernel.com>

Note this means that the unit tests need to create the directories with
`Read permission, for them to be successfully opened.

This also implies that we perform an implicit mode check on `open` which
we didn't do before.

Signed-off-by: David Scott <dave.scott@unikernel.com>

Rather than opening fresh handles each time, we now use the handles from
`open` or `create`. Note this means we have to hold a mutex around
`seek`+`read` and `seek`+`write` to prevent parallel operations
conflicting over the file position.

Signed-off-by: David Scott <dave.scott@unikernel.com>

If the client reads beyond the end of the directory, return with a
length of 0 rather than an `Invalid_argument "Cstruct.sub..."`

Signed-off-by: David Scott <dave.scott@unikernel.com>

A symlink can be dangling, which means that `stat` will return `ENOENT`.
When removing we want to remove the symlink itself so `lstat` is
appropriate.

Signed-off-by: David Scott <dave.scott@unikernel.com>

Signed-off-by: David Scott <dave.scott@unikernel.com>

This avoids duplicating the `clunk` internals (i.e. closing the
associated Resource)

Signed-off-by: David Scott <dave.scott@unikernel.com>