On 3/8/22 18:44, Paul Eggert wrote:
> On 3/8/22 04:29, Bernhard Voelker wrote:
>> I'm not so sure about that: at least in container environment it seems to be
>> common practice to mount regular files somewhere into the container, e.g.:
>
> Oh, I wasn't aware of that. However, don't they mount a regular file
> atop an already-existing regular file? In that case, the optimization
> would still be valid.
Indeed, the GNU/Linux kernel does not allow to bind-mount a file atop of a
directory:
$ strace -ve mount mount --bind /etc/hosts dir
mount("/etc/hosts", "/root/dir", 0x557e7ed77570, MS_BIND, NULL) = -1 ENOTDIR
(Not a directory)
mount: /root/dir: mount(2) system call failed: Not a directory.
+++ exited with 32 +++
The kernel doesn't seem so picky about other target file types;
e.g. it allows to bind-mount atop of a character device:
$ mknod char c 1 3
$ ls -log x
crw-r--r-- 1 1, 3 Mar 9 00:51 char
$ mount --bind /etc/hosts char
$ ls -log char
-rw-r--r-- 1 1515 Jan 16 14:55 char
Have a nice day,
Berny
