Just for reference and your entertainment, another friend of ours
truncates to 16 bit: CASTOR RFIO. I looked at the library at some
point, it neatly and silently slices off the top 16 bits.
Which means that you can only access CASTOR via RFIO if you come from a
system which stays inside the allocation scheme that Chris referred to.
Hey ho.
--jens
Kostas Georgiou wrote:
> On Fri, Aug 03, 2007 at 03:06:15PM +0100, Brew, CAJ (Chris) wrote:
>
>> Hi All,
>>
>> Does anyone know the maximum permissible UID in Linux of the top of
>> their head?
>>
>> I'm thinking of redoing my pool accounts to make them a bit more
>> sensible with some sort of systematic numbering scheme with plenty of
>> space for whatever crazy ideas the developers come up with next.
>>
>> Since there is (theoretically at least) sitewide allocation of UID/GID
>> blocks here at RAL I'd like to get well out of the way of that. So how
>> high can I go?
>
> The kernel uses 32bit uids since the 2.4 days and I think glibc got the
> support before 2.2
>
> The only problem that I've seen so far is with the accounting which is still
> stuck at 16bit :( This means that for example lastcomm gives you either
> wrapped uids or nobody for uids about the 64k limit depending on which
> version of EL you are running. I am pretty sure that there is a ticket
> open for it in bugzilla, from what I remember it requires changes to the
> accounting format that the kernel writes and the tools to parse it so I
> would not expect a fix for it in existing rhel versions.
>
> Cheers,
> Kostas
|