Re: Where do I find the definitive man page for mdadm?
On Mon 15 Nov 2021 at 09:04:55 (+0100), Thomas Schmitt wrote:
> Greg Wooledge wrote:
> > > > ├─sda1 8:1 0 260M 0 part /boot/efi 4C30-7972
>
> i wrote:
> > > The universe must be small where this FAT UUID is unique.
>
> David Wright wrote:
> > I think I can live with odds of 1:4294967296.
>
> Don't forget the birthday paradox. For getting a first collision you
> need roughly as many random tries as the square root of the space size.
> https://en.wikipedia.org/wiki/Birthday_problem#Square_approximation
>
> I.e. a hash of 32 bit is good for roughly 2 exp 16 = 65336 tries before
> you have to expect two of them to match.
Sure, but birthdays are immutible. Though the idea behind lengthy
GUIDs is that you can choose one at random without needing to take
others into consideration, if a collision were to be created against
the odds, you could replace it easily enough.
> Ok. I confess that having ten thousands of partitions is unlikely on a
> single computer. But mathematically a 32 bit number is far from being a
> Universally Unique IDentifier.
As you said, its universe is very small; so too its significance.
> > What's more important
> > is that ID_PART_ENTRY_TYPE=c12a7328-f81f-11d2-ba4b-00a0c93ec93b (or
> > 0xef on MBRs), which of course is anything but unique.
>
> Yeah. The long one is on about any computer which boots by EFI from a
> disk that's partitioned by GPT. Partition type UUIDs are not random but
> rather intended to be the same for any partition of the same type. Even
> the MBR partition type 0xef is supposed not to be used for anything but
> EFI System Partitions.
> (We have 0xef partitions in the Debian installation ISOs. But many old
> EFI partitions on disk have MBR partition types like 0x06. Microsoft is
> always a bit stronger than public specs.)
And we even coped with linux fs partitions sharing their type-GUID
with that of Basic Data Partitions in Windows.
Cheers,
David.
Reply to: