severity 487758 normal
thanks
On Tue, Jun 24, 2008 at 12:10:35AM +0200, Holger Fischer wrote:
> Package: e2fsprogs
> Version: 1.40.11-1
> Severity: critical
> File: /sbin/blkid
> Justification: causes serious data loss
>
> Normally my root device is /dev/sda6. When I connect an ext. sata
> before next boot my root device becomes /dev/sdb6 and the ext. sata is
> sda, because the external is connected to sata port 1 on the mainboard
> and the system disk is connected to sata port 4 on the mainboard.
>
> But not today:
> sda is obvious the external disk and sdb the internal, but mount tells
> me that sda6 is my root device. gparted tells me that there is no sda6,
> because my external disk doesn't contain a 6th partition
Mount sometimes lies; if /etc/mtab has the wrong information, it's
going report the wrong information.
> blkid tells me that there is a sda6 and a sdb6 with the same uuid (there
> should only be sdb6)
>
> What could it be related to? Where is that salad from?
blkid deliberately doesn't delete cached information for non-existent
devices. Since /dev/sda6 doesn't exist, it didn't remove it from the
blkid cache. If /dev/sda6 existed as a device, it would tried to
validate it, and removed the cache entry if the validation failed.
In the case where a device is missing, this is not a problem. In the
case where device names are bouncing around, it can mean that that the
when you lookup a device by LABEL or UUID, you will get multiple
names, and when using the lookup, you can get the "wrong" (i.e.,
stale) device name.
This will cause a mount to fail, but it won't necessarily lead to
"serious data loss", unless the script is seriously fragile (i.e., not
checking error returns, etc.,). By your usage, "it *could* possibly
lead to serious data loss if a script depends on the program in some
bad/wierd way", any bug could be classified as critical. Blkid isn't
itself causing serious data loss, so it's not critical.
It is a bug in that we should return a validated device in preference
to the non-validated device, and I'll work to fix it quickly. But I
don't consider this a RC bug.
- Ted
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
