I'm seeing the need for /etc/mtab (as a plain file) with loop-back mounts too now.
Apologies for being somewhat slow to report these things. I do try to stay quite current with Coreutils, but I don't run full regression testing on it. On 03/04/2016 11:17 AM, Bernhard Voelker wrote: > Well, the kernel always knows what's mounted where, so it seems that > mtab has always been redundant. > > See also: "Yeah, mtab is evil.", from > http://karelzak.blogspot.de/2011/04/bind-mounts-mtab-and-read-only.html > > Regarding your case: I don't see /dev/root in df's output inside > a chroot here. Bind mount can also take you to a sub-directory, for which case showing the device may not be what the sysadmin wants. (I'm sure Karel knows about sub-dirs. The article simply doesn't mention them.) Similarly (now) with loop-back, the pseudo device is less interesting than the backing file. The name of the file is available from /sys/dev/block/<dev>:<inode>/loop/backing_file which might not be visible under 'chroot'. The whole problem goes away with /etc/mtab (as a file). I did run some comparisons of 'df' from 8.24 and 8.25. Did not immediately see any differences (for bind mounts). Should I? -- R; <><
