Cannot umount /proc after using update-binfmts in a chroot
Bug #534211 reported by
Alkis Georgopoulos
This bug affects 2 people
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
binfmt-support (Ubuntu) |
Fix Released
|
Low
|
Colin Watson | ||
ltsp (Ubuntu) |
Fix Released
|
Undecided
|
Alkis Georgopoulos |
Bug Description
Binary package hint: binfmt-support
As part of LTSP fat client installations, the following instructions are performed (cut down version):
$ chroot "$ROOT" mount -t proc proc /proc
$ chroot "$ROOT" update-binfmts --import wine
$ umount "$ROOT/proc"
At this point we get:
$ umount: /opt/ltsp/
$ (In some cases useful info about processes that use
$ the device is found by lsof(8) or fuser(1))
...and we're forced to resort to `umount -l` to work around this problem.
Is the any way to call update-binfmts in a chroot (e.g. to install wine) without having to resort to `umount -l`?
Changed in ltsp (Ubuntu): | |
assignee: | nobody → Alkis Georgopoulos (alkisg) |
Changed in binfmt-support (Ubuntu): | |
status: | Triaged → Fix Committed |
assignee: | nobody → Colin Watson (cjwatson) |
To post a comment you must log in.
If, before calling `umount "$ROOT/proc"`, I run:
$ chroot "$ROOT" invoke-rc.d binfmt-support stop
then umount works cleanly.
BUT I'm losing "cli" and "wine" from /proc/sys/ fs/binfmt_ misc/ on the *real* fs tree, so I then have to run
$ invoke-rc.d binfmt-support start
on the real tree to get them back... so I don't think that qualifies as an acceptable solution.
Would it be possible to completely divert update-binfmts in the chroot (and replace it with `true`), and still have everything working properly?
I.e. is the update-binfmts call in wine1.2.postinst (and in the java postinst etc) required, or is putting the "cli", "jar", "wine" etc files in /usr/share/binfmts enough?