[BusyBox] Banging on mount...
Rob Landley
rob at landley.net
Sun May 15 04:29:00 UTC 2005
You know, we call erase_mtab() when we successfully mount a new block device
to delete any "stale entries", except that's not right. You can have the
same block device mounted in many different places on a modern linux kernel,
and this works fine. (It's a subset of --bind mounts, really.)
I'm treating this as one more bug my rewrite fixes, although I'd like to point
out that using a file-based mtab at all these days is probably inadviseable
since we've got most of the infrastructure in place for per-process
namespaces. Meaning the symlink to /proc/mounts could become required at
some point in the future so that you.
Not to mention the fact I've mounted stuff in chroot environments and file
based mtab falls out of sync, and I overlaid a usermode linux instance
(running /bin/sh as init) with the root filesystem mounted read/write as
hostfs and its mounts didn't show up in the parent kernel at all (find with
UML's procfs overmounted on /proc, but it causes lots of grief if /etc/mtab
is a file)... I've probably broken it other ways too, I just don't remember
what they were at the moment...
Meanwhile /proc/mounts never gets out of sync because it _is_ the kernel view
of the filesystem. Oh well. (Yes, I fixed it so loopback mounts umount
properly with /etc/mtab symlinked to /proc/mmounts. :)
Rob
More information about the busybox
mailing list