Why not just name the cam-ata devices the same as the old names?

[Matthew Dillon]

:> Not sure if we have support for labels based on disk serial number
:> similar to /dev/serno/* from DragonFlyBSD but
:> 
:> /dev/serno/*
:> 	* no extra step to setup, e.g. `gpart create' or `newfs'
:> 	* survive wiping entire disk, no metadata stored on-disk
:> 	* available on every ata disk
:
:How is this number generated within DFBSD?  I have seen hard disks that
:have literally no serial number (field is blank, and not space-padded
:either).  Some systems vendors do this.  I'm wondering if the generated
:number is based on a combination of details (ex. device model string +
:serial number string + total drive capacity in bytes), rather than just
:pure drive serial number.
:
:CC'ing Matt as he probably knows.  :-)
:
:-- 
:| Jeremy Chadwick                                   jdc at parodius.com |

    Basically what DragonFly does is extract the serial number if the
    device reports one (and it isn't trivially blank) and makes the drives
    and partitions available via /dev/serno/<serialno>[.s#[a-o]],
    in addition to their device attachment point at /dev/daXX, /dev/adXX,
    etc.

    If no serial number is available the devices will not show up in
    /dev/serno/ but will still show up at their device attachment point.

    We try to discourage the use of device attachment points when the serial
    number is available.  That is, we encourage the use of serial number
    specifications in /etc/fstab and /etc/rc.conf whenever possible.  The
    installer will install w/ such specifications.  This way the hard drive
    can be connected via multiple avenues or even switched between the
    IDE (ata) driver and the AHCI driver, or hot-plugged via E-Sata,
    or a port-multiplier, and it just works.  Plus, of course, with multiple
    device drivers the probe order is not necessarily going to be consistent
    either for DAxx assignment, and PHY issues / drive dropouts or
    probe failures would renumber everything anyway.  Device attachment
    points are a non-starter if you have more than one hard drive.

    Once users get used to /dev/serno and/or LVM the issue of trying to
    represent a consistent set of device attach points for different drivers
    goes away.  I strongly discourage any attempt to make device attach
    points consistent... it's a lost cause from the start.  I even discourage
    the use of LVM for simple installations.  LVM labels (roughly equivalent
    to GEOM labels in FreeBSD) are useful for more complex situations but
    logical labels in simple situations (which is to say, 99% of the
    installations out there) are unnecessarily complex relative to the
    use of the drive serial number.

    --

    Any IDE, SATA, or SCSI attached disk will have a real serial number.
    I've never come across disks attached that way which do not.

    VMWare and QEmu will generate a dummy serial number based on the
    order the drives are configured.  VirtualBox does not support serial
    numbers.  I think (not 100% sure).

    USB-connected sticks and USB drives which run through USB/SATA bridges
    (which is all of them) usually do not support serial numbers.  Those
    bridge chips are broken in many ways :-(

    iSCSI drives... I don't know, it usually depends more on the mid-layer
    software.  They are supposed to.

    RAID/OTHER - I don't know.  I'd like to say that they had better support
    proper serial numbers but I don't have enough data points.

						-Matt