Monday, September 17, 2012

util-linux 2.22

I released the latest util-linux version 2.22 one week ago. As usually the list of the changes in our ReleaseNotes is huge (it would be nice to release more often ;-)).

For me, the most important thing is that number of the project contributors is growing and we are able to coordinate our changes with another upstream projects like coreutils, procps-ng or systemd and maintainers from distributions contribute to the project.

The commands like mount(8), umount(8) or swapon(8) support new tags PARTLABEL=
and PARTUUID=.  It means that you can address partitions by name or UUID independently on the filesystem on the device.  You don't have to care about your fstab after mkfs or mkswap. The setting with PARTUUID= will be always valid.

Finally, we have dmesg --follow to print new kernel messages (like tail -f). This feature depends on readable /dev/kmsg (since kernel 3.5.0). I have also implemented a new dmesg output format --reltime (suggested by Linus on lkml):

$ dmesg --reltime
...
[Aug26 10:58] scsi_debug: host protection
[  +0.000004] scsi84 : scsi_debug, version 1.82 [20100324], dev_size_mb=50, opts=0x0
[  +0.000546] scsi 84:0:0:0: Direct-Access     Linux    scsi_debug       0004 PQ: 0 ANSI: 5
[  +0.000173] sd 84:0:0:0: Attached scsi generic sg1 type 0
[  +0.000356] sd 84:0:0:0: [sdb] 102400 512-byte logical blocks: (52.4 MB/50.0 MiB)
[  +0.000988] sd 84:0:0:0: [sdb] Write Protect is off
[  +0.000004] sd 84:0:0:0: [sdb] Mode Sense: 73 00 10 08

The low-level userspace tools consolidation continues:
  • sulogin(1) and utmpdump(1) from sysvinit merged into util-linux (the goal is to remove all init independent utils from sysvinit package)
  • eject(1) reimplemented to use /proc and /sys information and libmount and moved to util-linux
  • new command lslocks(8) as replacement for dead lslk(8)
The command findmnt(8) has initial support for per-process mount tables (namespaces), for example findmnt --task will print mount table for the PID. In the next release I'd like to have a new option --unshared-tasks to print all processes with unshared mounts.

The tool lsblk(8) supports reverse trees, it means that you can see whole stack of the block devices from top to down:

$ lsblk -s /dev/mapper/luks-10d813de-fa82-4f67-a86c-23d5d0e7c30e
NAME    MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
luks-10d813de-fa82-4f67-a86c-23d5d0e7c30e (dm-0)
        253:0    0  39.1G  0 crypt /home/kzak
└─sda6    8:6    0  39.1G  0 disk
  └─sda   8:0    0 149.1G  0 disk

The most invasive change is a new non-recursive build-system (just for the record: autotools are the best :-)). The result is faster build, binaries are in one top-level build directory rather than in many subdirectories, make distcheck calls our regression etc.

The another big change is fdisk refactoring. This is slow and painful work, but the result should be GPT support in release 2.23 (patches from Davidlohr Bueso are already in mailing list). I hope that one day the default fdisk will be nice, readable (colored?) low-level tool without some obsolete junk like CHS.

Note that in the next util-linux release 2.23 we're going to remove cryptoloop support. Yes, cryptoloop is bad and dead, use dm-crypt. (Note that util-linux upstream has never supported loop-AES.)

4 comments:

  1. Are you going to document how to dm-decrypt devices encrypted with cryptoloop? Current dmsetup(8) manual page is very brief.

    ReplyDelete
    Replies
    1. http://code.google.com/p/cryptsetup/wiki/FrequentlyAskedQuestions#7._Interoperability_with_other_Disk_Encryption_Tools

      Delete
  2. cryptoloop is NOT bad and dead.
    It's alive and well, except for what util-linux developers have thrown at it.

    dmcrypt is slower and inferior to loop-AES. That's still the way it is today. You guys are making the wrong decision and most people familiar with this issue know it. So I guess mask the updates until some sensible developer replaces you.

    ReplyDelete