Correct handling of isRecycled

[Matthieu Patou]

On 03/11/2011 18:15, Stefan (metze) Metzmacher wrote:
> Hi Matthieu,
>
>> I pushed here:
>> http://gitweb.samba.org/?p=mat/samba.git;a=shortlog;h=refs/heads/drs_recycled
>>
>> Windows 2008r2 insists on having a compatible schema when joining a
>> domain, and for all objects that are tombstones or becoming tombstones
>> it hads the isRecycled=TRUE attribute if FL<2K8R2 or if the recycle-bin
>> is not enabled. For already deleted objects or objects deleted on pre
>> Windows 2008r2 (ie. Windows 2003r2) this attribute is added while doing
>> the replication (initial or regular one).
>>
>> With this patches, Samba adds this attributes to tombstones objects if
>> the schema support the isDeleted attributes (commit a80863db) while
>> deleting objects or during replication (initial and regular).
> But with this patch we always add isRecycled: TRUE if it's not already
> there when we get a change from another dc, there's no check for the
> functional
> level.
Hum I think I forgot that isRecycled is not set even when the object is 
just in the DELETED status by Windows and Samba.

I changed to add a small test for the recycle-bin so that.
If FL <2K8R2 then the recycle-bin enabled is always false, if FL >= 
2K8R2 we test if the feature is enabled.
If not enabled that's the way to go (add isRecycled = TRUE)
If enabled we don't do anything while replicating.

>
>> In order to cope with current Samba setup and without requiring them to
>> run upgradeprovision or dbcheck, the kcc-deleted task has been modified
>> to add this attribute if supported by the schema to deleted objects that
>> do not have it (commit fad767b761f).
> This commit
> http://gitweb.samba.org/?p=mat/samba.git;a=commitdiff;h=fad767b761f641bc13824774144c85e5fe1f56d4
> has still stuff the belongs to the full scan commit.
>
> Also I'm not sure if this is the correct thing todo, as we never
> explicitly store 'isRecycled: FALSE' and cannot find out if adding
> isRecycled: TRUE
Fixed too
> is the right thing to do for the given object.
>
>> I also corrected the changed I made for link pointing to deleted objects
>> that should not be sent, so that it takes care of the forest level
>> (commit 21270a76b).
> This commit
> http://gitweb.samba.org/?p=mat/samba.git;a=commitdiff;h=21270a76bcb54599980dfd05643bf0351572bcf2
> should be the last one (not the first one) and you should use "TRUE"
> instead of "true",
it's a strncasecmp as true and TRUE are two valid value.
> as "TRUE" is stored.
>
>> I also included a change for the removal of deleted of objects that are
>> not in the Deleted Object container (commit c13e511da), I think that the
>> only case I know so far are site related objects and some child objects
>> that have DONOTMOVE flag (or similar).
>>
>> Let me know about them.
> I think we need a better way to detect if we want to fix an object which
> is stored broken,
> maybe look for attributes which should be there if the object is only
> DELETED and not an old style TOMBSTONE.
Well here it's not fixing it's just removing objects that are not in the 
Deleted Object container, for sites and such it's a correct behavior to 
have them not in the Deleted Objects container.

Ok I think I addressed your remarks in the new version of patches in my 
branch.

-- 
Matthieu Patou
Samba Team
http://samba.org