TODO

TODO

* fstransform: preserve filesystem LABEL and UUID. To collect them,
       run "lsblk -r -o maj:min,label,uuid" warning: special
       characters in label and uuid will be escaped as \xAB

* all: test on more file systems

       nilfs: ok as target.
              as source it gets practically full very quickly, free space fluctuates
              between zero and 1GB while the GC works to avoid "disk full" errors.
              But it works :)

       ntfs:  ntfs-3g is VERY slow both as source and target because it's userspace.
              it supports only UTF-8 filenames, not arbitrary bytes in the range 128-255.
              fsmove must check beforehand!
              use mount.ntfs option big_writes... enough or still too slow?

       logfs: as target: "mkfs -t logfs" does NOT accept option -q,
              fsmove and umount crash with kernel bugs (see tests/)

       others: btrfs, gfs, gfs2, hfs, jffs2, ocfs2, qnx4, qnx6, ubifs...

       not yet: tux3

* fstransform, fsmove: add fsremap-like persistence, i.e. the ability to resume
       a transformation after CTRL-C, power failure or similar interruptions

* fsmove: use a larger buffer (1MB) to speed up backward copy.
       Worth it? beware of low free space...

* fsremap: add a graphical interface ? probably not so worth it...

* all: port to FreeBSD. see this:
       Procedure to mount file system images under FreeBSD
       First you have to create virtual device, using this command:
       # mdconfig -a -t vnode -f /path/to/image.iso -u 1
       The "mdconfig" command creates a device and prints the name. Just
       append that to "/dev/" and use it for whatever you need:
       # newfs /dev/md1
       # mount /dev/md1 /mnt
       To unmount the image use command:
       # umount /mnt
       Once unmounted, you can unconfigure the device with
       # mdconfig -d -u 1

       Also, many system files, as /lib/libc.so.7 are set as
       'simmutable' (cannot open O_RDWR) and 'sunlink' (cannot unlink)
       so fsmove fails

       Furthermore, fsmove df-like stats do not work on FreeBSD:
       they seem to report usage of root file system

       Additionally, fsremap wraps calls to ff_posix_mkdir() with
       #ifdef __USE_POSIX, which is undefined by default on FreeBSD,
       causing ff_mkdir() to fail with ENOSYS.

       Finally, 'umount' wants mount-point instead of device.
       And how to 'mount -o remount,ro <mount-point>' ?

WONTDO
* fsmove: if file system free space is critically low, launch 'xfs_fsr'
          to try and free some space before giving up.
  why not:
  LIMITED USEFULNESS: only for 'xfs' file system
  DANGEROUS: at least on linux with an almost-full source device
      xfs_fsr can WORSEN the problem by triggering 'loop write error' kernel errors,
      which mean the source device has not enough space to accommodate the loop file contents.
      typically this CORRUPTS the file system inside target (loop) device!


DONE in 0.9.3
* all: check FHS (Filesystem Hierarchy Standard) and decide where to
       put persistence data. decision: /var/tmp/fstransform/

* fsremap: add persistence, i.e. the ability to resume a transformation
       after CTRL-C, power failure or similar interruptions

* fstransform: also allow device to be UNMOUNTED instead of MOUNTED.
       gives easier integration with gparted, and guarantees
       that nobody is accessing the file-system inside device

DONE in 0.9.2
* fsremap: preemptively check that last device block is writable.
       avoids troubles on linux with file-systems significantly smaller than device