[Samba] Corrupted idmap...

[Ryan Ashley]

OK, so since it appears our only recourse is to build a new domain from
scratch, how can we prevent this from happening again? We have several
Gentoo workstations, a bunch of Windows 7 workstations, and a few NAS
devices which run Linux of some flavor. How do we use NIS attributes
without killing our domain? The Samba guide even has instructions for
using ADUC to set the UID/GID for users and groups. You stated I should
only set a GID for "Domain Users", but what about other AD security
groups we create? This is a tad confusing since I thought NIS was needed
for our Linux systems and the NAS devices.

Lead IT/IS Specialist
Reach Technology FP, Inc

On 01/17/2017 10:57 AM, Rowland Penny via samba wrote:
> On Tue, 17 Jan 2017 10:04:23 -0500
> Ryan Ashley via samba <samba at lists.samba.org> wrote:
> 
> Firstly , 'gencache_notrans.tdb' is a cache file and is recreated when
> Samba is restarted.
> 
>> Rowland, I was just reading over another thread on this list about the
>> inability to access group policy from client machines. The user did
>> not have the symlinks setup (I do) but one thing you mentioned was
>> using the NIS attributes to set UID/GID numbers for the domain. You
>> said we should not do this for certain users and groups, but there is
>> no mention of this in the guides to setting up an AD DC, so I have
>> always done it. We do this to make our Linux-based NAS devices work.
> 
> The only only windows group that needs a gidNumber attribute is Domain
> Users and then only when you use the windbind 'ad' backend on a domain
> member. the other windows groups don't need a gidNumber, in fact, as
> Domain Admins needs to own directories in sysvol, you definitely
> shouldn't give this group a gidNumber.
> If you have to set up Samba this way because of your NAS, I would look
> closely at your NAS ;-)
>  
>>
>> Furthermore, you recommended the user use the idmap lines to ensure
>> consistent UID/GID numbers across devices, yet you suggested I turn
>> the exact same lines off in my config. Why is this? I understand our
>> situations are different, but when should we set winbind to use the AD
>> backend and set UID/GID numbers? How do do this so Linux-base file
>> services can be accessed by users and come out the same?
> 
> You are mixing up idmap on a DC and a Unix domain member. On a DC,
> idmapping is done in idmap.ldb, users & groups are allocated an
> xidNumber in the '3000000' range, the number allocated is on the next
> number available basis, apart from 'Administrator', 'Domain Users' and
> 'nobody' which get '0', '100' and '65534'.
> 
> On a Unix domain member, the two main ways of setting up idmapping is
> with the winbind 'rid' and 'ad' backends. The 'rid' backends calculates
> an ID from the windows RID, so you don't have to add anything to AD.
> This means that whilst, by using the 'rid' backend, you will get the
> same ID on every Unix domain member, it will still be different from
> the ID on a DC (and the ID will probably be different on other DCs).
> 
> The only way to get the same ID everywhere is to use the 'ad' backend,
> If you give a user a uidNumber and run 'net cache flush', this will be
> used instead of the xidNumber without modifying smb.conf in any way.
> On a Unix domain member it is different, you need to add something
> like this:
> 
>     idmap config *:backend = tdb
>     idmap config *:range = 2000-9999
>     ## map ids from the domain  the ranges may not overlap !
>     idmap config SAMDOM : backend = ad
>     idmap config SAMDOM : schema_mode = rfc2307
>     idmap config SAMDOM : range = 10000-999999
> 
> Now provided that the uidNumber attributes you have added are between
> 10000 and 999999 AND you have given Domain users a gidNumber in the
> same range, getent will display info for your users.
> 
> Now somebody (and I know who) recommended adding those lines to the
> smb.conf, but they do nothing on a DC, well they didn't until 4.5.0
> came out and then they started causing errors, so bottom line, don't add
> them to a Samba AD DC smb.conf
> 
> Rowland
> 
>