[RFC][PATCH 0/4] fsfreeze: new API

[Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>]

There are a few with issues with how freeze works in current kernels:

- Unmounting: It is possible to umount a frozen filesystem; the freeze
code holds an active reference to the superblock so it does not go away.
Since it is not possible to thaw by block device, thawing an unmounted
filesystem involves mounting it again and using the current vfs freeze
API.

- There is no check API: There is no easy way to know whether a
filesystem is frozen or not.

- Foolproofness: Quite often the freeze/thaw process is handled from a
daemon. In such cases we have to make really sure that the process does
not go away or is killed while the filesystem is frozen; there is no
check API so it is not always easy to figure out what is going on. With
the advent of virtualization things got even funnier; in some cases not
even the root user is aware of the existence of such daemon (usually a
guest agent that freezes the guests filesystems before taking a storage
snapshot).

I cooked some patches that add the long missing check ioctls and a new
freeze API, which should (modulo bugs) the current situation:

- Check ioctls

  [1/4] fsfreeze: add vfs ioctl to check freeze state
        Adds a vfs ioctl to check the freeze state.

  [2/4] fsfreeze: add block device ioctl to check freeze state
        Adds a block device ioctl to check the freeze state. This is
        needed because it possible to umount frozen filesystems.

- New freeze API

  [3/4] fsfreeze: add ioctl to create a fd for freeze control
        Adds a new ioctl that freezes the indicated filesystem and
        returns a file descriptor; as long as that file descriptor is
        held open, the filesystem remains open.

  [4/4] fsfreeze: add freeze fd ioctls
        Adds new ioctls that operate on the file descriptor described in
        the previous patch; these can be used to freeze/thaw the
        filesystem and check the freeze state of the filesystem, thus
        avoiding the need to get new file descriptors constantly.