Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
From:   Tavian Barnes <tavianator@tavianator.com>
Date:   Mon, 11 Feb 2019 14:59:29 -0500
Message-ID: <CABg4E-mbCk-jeRjETEtXk+ZHVub00_G5QaOnxbf5fgKd5MpD+A@mail.gmail.com>
Subject: d_type and bind mounts
To:     linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

It seems that readdir()/getdents() fill d_type from the underlying
filesystem, not respecting bind mounts of non-directories:

$ touch mount_point
$ sudo mount --bind /dev/null mount_point
$ find -name mount_point -type c
$ find -name mount_point -type f
./mount_point

(Requires a fairly recent GNU findutils to reproduce, older ones
always call stat().)  I've seen similar discussions about d_ino being
for the underlying file, not the mount point, which people have said
is technically a POSIX violation but also unlikely to be fixed.  Is
the same true of d_type?  And is there some workaround a program could
use to get the actual type without the overhead of a whole stat()
call?  For example, a way to tell whether a directory entry is a mount
point?

For reference, the relevant findutils bug is
https://savannah.gnu.org/bugs/?54913.

(Please keep me cc'd, I'm not subscribed to lkml or fsdevel.  Thanks!)

-- 
Tavian Barnes