LWN: Comments on "revoke() and frevoke()"

Forced unmount

efexis — Fri, 04 Aug 2006 04:11:55 +0000

You don't always want to kill the process; many processes could handle a file being closed without having to terminate, and would want to if it was multithreaded, with one thread using that file (eg, disconnecting a file from a browser, you wouldn't want to lose all browser windows).

It could also be the kernel that has the file open (eg, mounted into the filesystem using loopback block device); killing the kernel to release the file isn't desirable =)

I think if you're detaching files, a signal should be sent to processes at the same time, so they know the file is closed before they try and fail a read/write.

Forced unmount

rvfh — Thu, 03 Aug 2006 17:40:09 +0000

Could not some of that be userspace?

find open files on filesystem, and there owning processes
kill the processes in question*
unmount filesystem

* This _needs_ to be in userspace, so the user (who most likely asked for the unmount in the first place) can do something about it (save unsaved files...)

But maybe my knowledge is just out-dated, and this is already the case, and the problem at hand is a different one...

revoke() and frevoke()

niner — Thu, 03 Aug 2006 16:44:45 +0000

Shouldn't this be a fairly simple shellscript?
Just lsof for the device files in question and kill all processes of the logging out user that still have them open.

revoke() and frevoke()

joey — Thu, 27 Jul 2006 01:23:06 +0000

Also very useful for not letting a user keep a daemon open recording from a system's sound card or camera when they're logged out, while still letting them legitimately access them when logged in at the console. AFAIK there's no secure implementation of that for linux currently, so here's looking forward to revoke()!