Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 14 Apr 2023 03:55:15 +0100
From:   Al Viro <viro@zeniv.linux.org.uk>
To:     Richard Cochran <richardcochran@gmail.com>
Cc:     linux-kernel@vger.kernel.org
Subject: Re: Weirdness with bind mounts and moving files from tmpfs
Message-ID: <20230414025515.GG3390869@ZenIV>
References: <ZDic8fx1NQirB2+d@hoboy.vegasvil.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZDic8fx1NQirB2+d@hoboy.vegasvil.org>
Sender: Al Viro <viro@ftp.linux.org.uk>
Precedence: bulk

On Thu, Apr 13, 2023 at 05:23:13PM -0700, Richard Cochran wrote:

> So far, so good.  Now do the same, but with 'mv' instead of 'cp'...
> 
>     # echo three > /run/file
>     # mv /run/file /home/file
>     # cat /opt/file
>     two
> 
> At this point, the contents of /opt/file are stuck forever with "two"...
> 
>     # echo three > /run/file
>     # cp /run/file /home/file
>     # cat /opt/file
>     two
>     # echo three > /home/file
>     # cat /opt/file
>     two
> 
> What is going on here?  Is this a bug or a feature?

strace and compare the traces...  cp(1) opens the existing
target, truncates and writes to it; mv(1) tries rename(),
falling back to unlink()+creat()+write() in case rename()
failed with -EXDEV.

The same behaviour can be seen without any mount(2) use:

; echo x > a
; ln a b
; echo y > c
; cat a
x
; cat b
x
; cat c
y
; cp c a
; cat b
y
; echo x > a
; cat a
x
; cat b
x
; mv c a
; cat a
y
; cat b
x
;

Exact same underlying cause - writing down what happens
step-by-step might be enlightening...