Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 30 Mar 2021 10:13:59 -0400
From:   "J. Bruce Fields" <bfields@fieldses.org>
To:     Benjamin Maynard <benmaynard@google.com>
Cc:     linux-nfs@vger.kernel.org
Subject: Re: Input/output errors when mounting re-exported directories that
 use crossmounts
Message-ID: <20210330141359.GA8841@fieldses.org>
References: <CA+QRt4vr+CUgP-4xkVQwLWNZMHo-w6TwU8bFwzuZcUc1RPi-RA@mail.gmail.com>
 <20210325213421.GC18351@fieldses.org>
 <CA+QRt4srUGGgqYPs6bufihOrC8wUZ0-B5yzXjN7rrh=EdTatAA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CA+QRt4srUGGgqYPs6bufihOrC8wUZ0-B5yzXjN7rrh=EdTatAA@mail.gmail.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Precedence: bulk

One of my pie-in-the-sky ideas here is to allow running the server in a
mode where it re-exports one server and has no other exports whatsoever.

That would mean that when the re-exporting server gets a filehandle, it
knows it originated from that one server, and if necessary it can do
things like issue a getattr on that filehandle for the fsid to find out
which filesystem it belongs to.

It's a major restriction, but I figure that might be what you're doing
anyway in a case like yours.  Perhaps containers or VMs would be a
workaround if you want to server more exports from the same physical
hardware.

But, anyway, I'm not actually sure how to make that work and it's not
currently on the todo list.

--b.