TODO - pub/scm/utils/util-linux/util-linux - Git at Google


  Note that items with (!) have high priority.

 dmesg
 -----

  - add --follow option
    http://www.spinics.net/lists/util-linux-ng/msg04453.html

  - add --color to colorize err and panic messages

 ldattach
 --------

   - write usage()

 minix (fsck, mkfs)
 ------------------

  - clean up types -- use ino_t, size_t and ssize_t

 libmount (mount/umount)
 -----------------------

  Note that the old mount/[u]mount.c code is in MAINTENANCE MODE only. All new
  features should be implemented to libmount or to the new mount/umount
  implementation in the libmount/samples/ directory.

  - (!) on systems with regular mtab file it is impossible to umount by "umount
    /dev/loop0" if the loop device has been created by "mount -o loop", because
    there is backing file in the mtab (instead of the device name).

    Now we have all necessary information in /sysfs so it should be possible to
    translate the device name to backing file and then search in mtab for the
    filename. See loopdev_get_loopfile().

  - (!!!) add libmount based umount(8) (see libmount/samples/ where is new mount(8)
    implementation.

  - (!) add --source and --target options to specify device and mountpoint

      fstab:
         /dev/sda1 /foo	auto	defaults 0 0
         /foo      /mnt  auto    bind     0 0
      command line:
 	mount -o remount,ro /foo

     ... this command is ambiguous. It would be better to have a way how specify
     target or source:

         mount --target /foo -o remount,ro

     Note that findmnt(8) already supports --target and --source.

  - add ---target-prefix option to specify prefix for mountpoints
    so then you can

 	LIBMOUNT_FSTAB=/etc/fstab.foo mount -a --target-prefix=$FOO_ROOT

    to create a new hierarchy of filesystems at $FOO_ROOT.

  - umount by label:
 	# mount LABEL=mylabel
 	# umount LABEL=mylabel

 losetup
 -------

  - (!!!) use new lib/sysfs.c code for losetup(8), the new implementation should
    be moved to sys-utils/losetup.c.

  - (!!!) add support for LOOP_CTL_ ioctls (probably will be in kernel 3.1 or 3.2)

 libblkd and libmountL
 ----------------------------

  - use __attribute__((notnull)) and __attribute__((warn_unused_result))
    Note that the code has to be usefull for non-gcc compilers too.

 partx
 -----

  - (!) add support loop devices:

 	partx -a /path/file.img

 	* associate the file with loop device (use lib/loopdev.c)
 	* map partitions by BLKPG_ADD_PARTITION

  - support mapping by device-mapper if argv[0] is "kpartx" or --dm option is used.

  - (!) add regression tests for partx, addpart and delpart


 docs
 ----

  - (!) use something better than gtk-doc for libmount and libblkid (doxyden?)

  - (!) add API documentation to libuuid

  - add Documentation/ directory with:
 	   - man page template
 	   - usage() HOW-TO
 	   - move README.devel to the Documentation/

 build-sys
 --------

  - use non-recursive build-sys, see
    http://thread.gmane.org/gmane.linux.utilities.util-linux-ng/3297

  - we use something like

 	AC_ARG_ENABLE(...., enable_foo=check)
 	build_foo=yes
 	if test "x$enable_foo" = xcheck; then
 		if test "x$linux_os" = xno; then
 			build_foo=no
 		fi
 	fi
 	AM_CONDITIONAL(BUILD_FOO, test "x$build_foo" = xyes)

    for Linux-only utils in configure.ac. It would be nice to set all defaults
    for all "$enable_" variables at the begin of the configure script according to
    $linux_os. Something like:

 	if test "x$linux_os" = xno
 		enable_mount=no
 		enable_libmount=no
 		enable_lsblk=no
 	fi

    then we can remove all "if test "x$enable_foo" = xcheck;" stuff from the rest
    of the configure script.

  - it would be nice to remove all "if BUILD_LIB{MOUNT,BLKID,UUID}" from the
    Makefiles (e.g. misc-utils/Makefile.am) and use BUILD_<utilname> only

    The configure.ac has to care about dependence between utils and librares and
    enable BUILD_<utilname>.

 lib/tt.c
 --------

  - allows to sort columns, for example sort lsblk(8) output by SIZE

 login-utils:
 -----------

  - use err() and warn() macros rather than fprintf(stderr, ...)

 libblkid
 --------

  - remove strerrr() from debug messages (use %m) to make BLKID_DEBUG= output
    thread-safe

  - (!) don't use internally blkid_loff_t, rather use off_t, size_t, ssize_t,
        stdint.h types and so on...

  - add -<BE|LE> suffix to test images for native-endian filesystems (e.g. swap)
    and add support for such functionality to tests/ts/blkid/low-probe

  - add FSSIZE value  -- filesystem size (klibc requirement)

  - (!) add support for dasd PT (used for example on s390)

  - (!!!) support PARTUUID= tag

    * the partitions probing is already supported by low-level part of the
      library, but it's necessary add support for this tag also to high-level
      blkid_cache and blkid_evaluate_* APIs

    * add blkid -P <PARTUUID>

    * add PARTUUID to wipefs output

 wipefs
 ------

  - some filesystem (namely FAT) contains more magic strings, so if you erase
    one magic string the FS is still detectable by libblkid. We have to inform
    users that there is more valid magic string for the same FS.

  - (!!!) allow to wipe partition tables

 fdisk(s)
 --------

  - add "move end" command to move end of the last primary/extended partition.
    This feature seems very attractive to users who resizing their disks
    (for example in virtual machines).

  - sfdisk has to use rpmatch() for answers to y/n questions
    (e.g. "Are you satisfied with this? [ynq]")

  - sfdisk rounds to cylinders is -uM (megabyte units) is specified, this is
    pretty stupid feature. It has to round to sectors if -uS or -uM is specified.

  - Sun label support is completely useless for large disks, it uses number of
    cylinders from on-disk-label where the geometry is stored by int16 values.
    It seems better to completely ignore this stuff from the label and always
    use geometry + BLKGETSIZE64 from kernel.

  - use off_t instead "long long"

  - catch SIGINT (Ctrl-C) and return to main menu.
    From Red Hat bugzilla #545488:

    While using fdisk normally, if you accidentally pressed the wrong button (to
    start a sequence of questions for some operation, e.g. 'c' to create
    partition).  The tool tries too hard to keep asking you for valid input.  You
    can't provide a blank or invalid input to get it to break out of the current
    dialog sequence and get back to the main menu.

  - fdisk/* refactoring (probably implement libfdisk ???)

  - add GPT support (probably implement libfdisk ???)

 misc
 ----

  - switch_root:
      - move all mountpoints to the newroot (there are hardcoded /proc /sys and /dev paths now)
      - add --dont-move[=<list of dirs>] options

  - use TZ=UTC and LANG=en_US.UTF-8 for tests

  - add mllockall() and SCHED_FIFO to hwclock,
    see http://lkml.org/lkml/2008/10/12/132

  - use rpmatch() for all Y/N questions


 ---------------
 exotic requests
 ---------------

  - mount -a -- reorder fstab entries by paths before mount (just idea only)

  - mount -a  (just idea only)

         Date:   Sun, 3 Jun 2007 18:04:24 +0300 (MET DST)
         From: Szabolcs Szakacsits <szaka@sienet.hu>

 	In the past the right record order could be figured out easily by just
 	checking out fstab (if one knew what to look for) but considering the
 	fastly increasing number of user space file systems and their usage, with
 	their path, library, etc dependencies, it's getting trickier and is a black
 	magic for most users because they simply expect drives to be mounted
 	independently of their order in fstab.

 	One typical, wrongly edited fstab example is:

 	  /dev/hda2 /             ext3      defaults  1 1
 	  /dev/hda1 /mnt/windows  ntfs-3g   defaults  0 0
 	  /dev/hda3 /usr          ext3      defaults  0 0

 	The events:

 	  mount  ->  /sbin/mount.ntfs-3g  ->
 		 ->  resolves to <path1>/ntfs-3g via a symlink ->
 		 ->  ntfs-3g requires at least <path2>/libfuse*

 	There are many potential solutions. For example installing everything on
 	the root file system which may be needed for successful mount. But this
 	is not always feasible or practical since we could end up putting almost
 	everything on the root file system in the end.

 	Another idea is an improved mount strategy:

 	  do {
 		try to mount all unmounted entries
 	  } while (not all mounted && at least one new was successfully mounted)

  - rewrite ipcs to use /proc/sys/kernel rather than unreliable syscalls
    (there are problems with 32bit userspace on 64bit kernel)


  - add SELinux security contexts support to the 'ipcs' utility
    http://bugzilla.redhat.com/show_bug.cgi?id=225342

    Would be great to list the current system IPC Objects with their respective
    security labels (where allowed) with something like 'ipcs -Z' - following the
    way other tools reports those.

	Note that items with (!) have high priority.

	dmesg
	-----

	- add --follow option
	http://www.spinics.net/lists/util-linux-ng/msg04453.html

	- add --color to colorize err and panic messages

	ldattach
	--------

	- write usage()

	minix (fsck, mkfs)
	------------------

	- clean up types -- use ino_t, size_t and ssize_t

	libmount (mount/umount)
	-----------------------

	Note that the old mount/[u]mount.c code is in MAINTENANCE MODE only. All new
	features should be implemented to libmount or to the new mount/umount
	implementation in the libmount/samples/ directory.

	- (!) on systems with regular mtab file it is impossible to umount by "umount
	/dev/loop0" if the loop device has been created by "mount -o loop", because
	there is backing file in the mtab (instead of the device name).

	Now we have all necessary information in /sysfs so it should be possible to
	translate the device name to backing file and then search in mtab for the
	filename. See loopdev_get_loopfile().

	- (!!!) add libmount based umount(8) (see libmount/samples/ where is new mount(8)
	implementation.

	- (!) add --source and --target options to specify device and mountpoint

	fstab:
	/dev/sda1 /foo auto defaults 0 0
	/foo /mnt auto bind 0 0
	command line:
	mount -o remount,ro /foo

	... this command is ambiguous. It would be better to have a way how specify
	target or source:

	mount --target /foo -o remount,ro

	Note that findmnt(8) already supports --target and --source.

	- add ---target-prefix option to specify prefix for mountpoints
	so then you can

	LIBMOUNT_FSTAB=/etc/fstab.foo mount -a --target-prefix=$FOO_ROOT

	to create a new hierarchy of filesystems at $FOO_ROOT.

	- umount by label:
	# mount LABEL=mylabel
	# umount LABEL=mylabel

	losetup
	-------

	- (!!!) use new lib/sysfs.c code for losetup(8), the new implementation should
	be moved to sys-utils/losetup.c.

	- (!!!) add support for LOOP_CTL_ ioctls (probably will be in kernel 3.1 or 3.2)

	libblkd and libmountL
	----------------------------

	- use __attribute__((notnull)) and __attribute__((warn_unused_result))
	Note that the code has to be usefull for non-gcc compilers too.

	partx
	-----

	- (!) add support loop devices:

	partx -a /path/file.img

	* associate the file with loop device (use lib/loopdev.c)
	* map partitions by BLKPG_ADD_PARTITION

	- support mapping by device-mapper if argv[0] is "kpartx" or --dm option is used.

	- (!) add regression tests for partx, addpart and delpart


	docs
	----

	- (!) use something better than gtk-doc for libmount and libblkid (doxyden?)

	- (!) add API documentation to libuuid

	- add Documentation/ directory with:
	- man page template
	- usage() HOW-TO
	- move README.devel to the Documentation/

	build-sys
	--------

	- use non-recursive build-sys, see
	http://thread.gmane.org/gmane.linux.utilities.util-linux-ng/3297

	- we use something like

	AC_ARG_ENABLE(...., enable_foo=check)
	build_foo=yes
	if test "x$enable_foo" = xcheck; then
	if test "x$linux_os" = xno; then
	build_foo=no
	fi
	fi
	AM_CONDITIONAL(BUILD_FOO, test "x$build_foo" = xyes)

	for Linux-only utils in configure.ac. It would be nice to set all defaults
	for all "$enable_" variables at the begin of the configure script according to
	$linux_os. Something like:

	if test "x$linux_os" = xno
	enable_mount=no
	enable_libmount=no
	enable_lsblk=no
	fi

	then we can remove all "if test "x$enable_foo" = xcheck;" stuff from the rest
	of the configure script.

	- it would be nice to remove all "if BUILD_LIB{MOUNT,BLKID,UUID}" from the
	Makefiles (e.g. misc-utils/Makefile.am) and use BUILD_<utilname> only

	The configure.ac has to care about dependence between utils and librares and
	enable BUILD_<utilname>.

	lib/tt.c
	--------

	- allows to sort columns, for example sort lsblk(8) output by SIZE

	login-utils:
	-----------

	- use err() and warn() macros rather than fprintf(stderr, ...)

	libblkid
	--------

	- remove strerrr() from debug messages (use %m) to make BLKID_DEBUG= output
	thread-safe

	- (!) don't use internally blkid_loff_t, rather use off_t, size_t, ssize_t,
	stdint.h types and so on...

	- add -<BE\|LE> suffix to test images for native-endian filesystems (e.g. swap)
	and add support for such functionality to tests/ts/blkid/low-probe

	- add FSSIZE value -- filesystem size (klibc requirement)

	- (!) add support for dasd PT (used for example on s390)

	- (!!!) support PARTUUID= tag

	* the partitions probing is already supported by low-level part of the
	library, but it's necessary add support for this tag also to high-level
	blkid_cache and blkid_evaluate_* APIs

	* add blkid -P <PARTUUID>

	* add PARTUUID to wipefs output

	wipefs
	------

	- some filesystem (namely FAT) contains more magic strings, so if you erase
	one magic string the FS is still detectable by libblkid. We have to inform
	users that there is more valid magic string for the same FS.

	- (!!!) allow to wipe partition tables

	fdisk(s)
	--------

	- add "move end" command to move end of the last primary/extended partition.
	This feature seems very attractive to users who resizing their disks
	(for example in virtual machines).

	- sfdisk has to use rpmatch() for answers to y/n questions
	(e.g. "Are you satisfied with this? [ynq]")

	- sfdisk rounds to cylinders is -uM (megabyte units) is specified, this is
	pretty stupid feature. It has to round to sectors if -uS or -uM is specified.

	- Sun label support is completely useless for large disks, it uses number of
	cylinders from on-disk-label where the geometry is stored by int16 values.
	It seems better to completely ignore this stuff from the label and always
	use geometry + BLKGETSIZE64 from kernel.

	- use off_t instead "long long"

	- catch SIGINT (Ctrl-C) and return to main menu.
	From Red Hat bugzilla #545488:

	While using fdisk normally, if you accidentally pressed the wrong button (to
	start a sequence of questions for some operation, e.g. 'c' to create
	partition). The tool tries too hard to keep asking you for valid input. You
	can't provide a blank or invalid input to get it to break out of the current
	dialog sequence and get back to the main menu.

	- fdisk/* refactoring (probably implement libfdisk ???)

	- add GPT support (probably implement libfdisk ???)

	misc
	----

	- switch_root:
	- move all mountpoints to the newroot (there are hardcoded /proc /sys and /dev paths now)
	- add --dont-move[=<list of dirs>] options

	- use TZ=UTC and LANG=en_US.UTF-8 for tests

	- add mllockall() and SCHED_FIFO to hwclock,
	see http://lkml.org/lkml/2008/10/12/132

	- use rpmatch() for all Y/N questions


	---------------
	exotic requests
	---------------

	- mount -a -- reorder fstab entries by paths before mount (just idea only)

	- mount -a (just idea only)

	Date: Sun, 3 Jun 2007 18:04:24 +0300 (MET DST)
	From: Szabolcs Szakacsits <szaka@sienet.hu>

	In the past the right record order could be figured out easily by just
	checking out fstab (if one knew what to look for) but considering the
	fastly increasing number of user space file systems and their usage, with
	their path, library, etc dependencies, it's getting trickier and is a black
	magic for most users because they simply expect drives to be mounted
	independently of their order in fstab.

	One typical, wrongly edited fstab example is:

	/dev/hda2 / ext3 defaults 1 1
	/dev/hda1 /mnt/windows ntfs-3g defaults 0 0
	/dev/hda3 /usr ext3 defaults 0 0

	The events:

	mount -> /sbin/mount.ntfs-3g ->
	-> resolves to <path1>/ntfs-3g via a symlink ->
	-> ntfs-3g requires at least <path2>/libfuse*

	There are many potential solutions. For example installing everything on
	the root file system which may be needed for successful mount. But this
	is not always feasible or practical since we could end up putting almost
	everything on the root file system in the end.

	Another idea is an improved mount strategy:

	do {
	try to mount all unmounted entries
	} while (not all mounted && at least one new was successfully mounted)

	- rewrite ipcs to use /proc/sys/kernel rather than unreliable syscalls
	(there are problems with 32bit userspace on 64bit kernel)


	- add SELinux security contexts support to the 'ipcs' utility
	http://bugzilla.redhat.com/show_bug.cgi?id=225342

	Would be great to list the current system IPC Objects with their respective
	security labels (where allowed) with something like 'ipcs -Z' - following the
	way other tools reports those.