[cifs-protocol] 117060815865307 Question about DRS client behaviour with linked attributes

Sreekanth Nadendla srenaden at microsoft.com
Thu Jun 29 18:26:56 UTC 2017


Hello Tim,
GetDSNameFromAttrVal() only verifies that the DSNAME referenced in the attrval is well formed DSNAME. Getting a null from DSNameFromAttrVal means the replication partner sent malformed link val which is different from unknown object.  It just means we have invalidly formed data.

Here the intent of the protocol is to retrieve a syntactically correct DN from the attribute. Protocol does NOT say to validate that the name actually occurs in the local DIT. As you mentioned, if we get INVALID_ATTRIBUTE_SYNTAX here, the replication halts.

If you get a validly formed DSNAME that refers to an object that does not exist in the local directory, the implementation must proceed and handle that case. This means creating a DN valued attribute that does not refer to an actual object in the local DIT ( An implementation specific way of temporarily managing reference to actual object).

Hope this resolves your pending question.

Regards,
Sreekanth Nadendla
Microsoft Windows Open Specifications

From: Tim Beale [mailto:timbeale at catalyst.net.nz]
Sent: Monday, June 26, 2017 5:00 PM
To: Sreekanth Nadendla <srenaden at microsoft.com>
Cc: cifs-protocol at lists.samba.org; MSSolve Case Email <casemail at microsoft.com>
Subject: Re: 117060815865307 Question about DRS client behaviour with linked attributes


Hi Sreekanth,

Thanks for your reply.

How does this relate to where ProcessLinkValue() is called, in particular the setting of the DRS_GET_TGT flag? i.e. the block near the end of 4.1.10.6.1:

if (ulResult = ERROR_DS_DRA_MISSING_PARENT) then
    Send IDL_DRSGetNCChanges message again with the same input
    parameters specified in msgIn but this time with msgIn.ulFlags
    containing DRS_GET_ANC field set. It is an error for this
    condition to occur if (DRS_GET_ANC in msgIn.ulFlags) is true
else if (ulResult = ERROR_DS_DRA_RECYCLED_TARGET) then
    Send IDL_DRSGetNCChanges message again with the same input
    parameters specified in the msgIn but this time with msgIn.ulMoreFlags
    containing DRS_GET_TGT field set.

The documentation here talks about the ERROR_DS_DRA_RECYCLED_TARGET error, but not ERROR_DS_INVALID_ATTRIBUTE_SYNTAX error. But in ProcessLinkValue(), we're hitting the case where the target object is unknown/missing. The amended documentation would imply that in the case where the target object is missing/unknown (as opposed to deleted/recycled), then we fail the replication (instead of retrying with DRS_GET_TGT set).

In my testing with Windows I have seen cases where the Windows DC will send a linked attribute with an unknown target object (that hasn't been deleted/recycled at all). Resending the GetNCChanges request with the DRS_GET_TGT flag set will fix this problem. Am I misunderstanding how the protocol works? I was expecting the documentation to look more like:

if (targetObject = null or
    (IsRecycleBinEnabled() and targetObject!isRecycled) or
    (not IsRecycleBinEnabled() and targetObject!isDeleted)) then
    if (DRS_GET_TGT in ulMoreFlags) then
        /* nothing to do */
        return 0
    else
        return ERROR_DS_DRA_RECYCLED_TARGET
endif

Thanks,

Tim

On 27/06/17 07:17, Sreekanth Nadendla wrote:
Hello Tim, below is the corrected edit for issue #1 you’ve reported. MS-DRSR will be updated with this change. The correct error code returned would be ERROR_DS_INVALID_ATTRIBUTE_SYNTAX.


  targetObject = GetDSNameFromAttrVal( replValinf.attrTyp, replValInf.pAval)



  if (targetObject = null)

      return ERROR_DS_INVALID_ATTRIBUTE_SYNTAX



  if ((IsRecycleBinEnabled() and targetObject!isRecycled) or

      (not IsRecycleBinEnabled() and targetObject!isDeleted)) then

     if (DRS_GET_TGT in ulMoreFlags) then

         /* nothing to do */

        return 0

     else

        return ERROR_DS_DRA_RECYCLED_TARGET

   endif

Regards,
Sreekanth Nadendla
Microsoft Windows Open Specifications

From: Tim Beale [mailto:timbeale at catalyst.net.nz]
Sent: Wednesday, June 7, 2017 7:35 PM
To: Sreekanth Nadendla <srenaden at microsoft.com><mailto:srenaden at microsoft.com>
Cc: cifs-protocol at lists.samba.org<mailto:cifs-protocol at lists.samba.org>; MSSolve Case Email <casemail at microsoft.com><mailto:casemail at microsoft.com>
Subject: Re: 117060115827644 Question about DRS client behaviour with linked attributes


Hi Sreekanth,

Thanks for the link. I'd already seen that article (or one very similar), but it does provide a lot of useful information about replication.

The Docs issues I mentioned in the conference call are as follows:

1). A fortnight ago, I raised a minor issue with MS-DRSR 4.1.10.6.14 ProcessLinkValue(). i.e. https://msdn.microsoft.com/en-us/library/dd207759.aspx<https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmsdn.microsoft.com%2Fen-us%2Flibrary%2Fdd207759.aspx&data=02%7C01%7Csrenaden%40microsoft.com%7C46a7a120036b469503df08d4adfdd01e%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636324753009373564&sdata=paOhTFuE9pSC5yfXLlZhP9UZWBdUWzVR%2FleL0TVLWXk%3D&reserved=0>

In the following block of code that checks the link target, it looks like it's missing a null check for targetObject.

targetObject = GetDSNameFromAttrVal(replValinf.attrTyp, replValInf.pAval)
if ((IsRecycleBinEnabled() and targetObject!isRecycled) or
    (not IsRecycleBinEnabled() and targetObject!isDeleted)) then
    if (DRS_GET_TGT in ulMoreFlags) then
        /* nothing to do */
        return 0
    else
        return ERROR_DS_DRA_RECYCLED_TARGET
endif

Presumably it should also return ERROR_DS_DRA_RECYCLED_TARGET if targetObject is null, as this is what triggers the use of the DRS_GET_TGT flag.

2). Related to this, the error code used here isn't documented, i.e. it isn't listed on https://msdn.microsoft.com/en-us/library/cc231199.aspx<https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmsdn.microsoft.com%2Fen-us%2Flibrary%2Fcc231199.aspx&data=02%7C01%7Csrenaden%40microsoft.com%7C46a7a120036b469503df08d4adfdd01e%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636324753009373564&sdata=rM3m%2Bv6oVnAKrGM%2Frzu8PV8hX8eCbPfMe59TQdgB8Oc%3D&reserved=0>

I think it should be something like:

ERROR_DS_DRA_RECYCLED_TARGET            0x000021BF

Thanks for your help.

Tim
On 08/06/17 09:56, Sreekanth Nadendla wrote:

Hello Tim, below is the article I mentioned and it provides some useful details about Linked value replication scenarios.

https://technet.microsoft.com/en-us/library/cc772726(v=ws.10).aspx<https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftechnet.microsoft.com%2Fen-us%2Flibrary%2Fcc772726%28v%3Dws.10%29.aspx&data=02%7C01%7Csrenaden%40microsoft.com%7C46a7a120036b469503df08d4adfdd01e%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636324753009373564&sdata=hxJIraSwimp3qi%2BA6Sr7cRTqCPAitKfiZ2CmucuG4VM%3D&reserved=0>





Regards,

Sreekanth Nadendla

Microsoft Windows Open Specifications



-----Original Message-----

From: Sreekanth Nadendla

Sent: Thursday, June 1, 2017 5:03 PM

To: 'Tim Beale' <timbeale at catalyst.net.nz><mailto:timbeale at catalyst.net.nz>

Cc: cifs-protocol at lists.samba.org<mailto:cifs-protocol at lists.samba.org>; MSSolve Case Email <casemail at microsoft.com><mailto:casemail at microsoft.com>

Subject: 117060115827644 Question about DRS client behaviour with linked attributes



Hello Tim, I will be assisting you with this issue.



Regards,

Sreekanth Nadendla

Microsoft Windows Open Specifications



-----Original Message-----

From: Tim Beale [mailto:timbeale at catalyst.net.nz]

Sent: Wednesday, May 31, 2017 8:18 PM

To: Interoperability Documentation Help <dochelp at microsoft.com><mailto:dochelp at microsoft.com>; cifs-protocol at lists.samba.org<mailto:cifs-protocol at lists.samba.org>

Subject: Question about DRS client behaviour with linked attributes



Hi,



We're trying to add DRS GET_TGT support to Samba and have a question about the behaviour of linked attributes. We noticed that you can sometimes get linked attributes that span a partition.



For example you can have an object in the Configuration partition with a linked attribute to an object in the Domain partition:

dn:

CN=RWDC,CN=Servers,CN=Default-First-Site-Name,CN=Sites,CN=Configuration,DC=SAMDOM,DC=EXAMPLE,DC=COM

serverReference: CN=RWDC,OU=Domain Controllers,DC=SAMDOM,DC=EXAMPLE,DC=COM



And you can also have an object in the Domain partition with a linked attribute to an object in the Configuration partition:

dn: CN=RWDC,CN=Topology,CN=Domain System Volume,CN=DFSR-GlobalSettings,CN=System,DC=SAMDOM,DC=EXAMPLE,DC=COM

serverReference: CN=NTDS

Settings,CN=RWDC,CN=Servers,CN=Default-First-Site-Name,CN=Sites,CN=Configuration,DC=SAMDOM,DC=EXAMPLE,DC=COM



From what I could see from testing, when the DRS_GET_TGT flag is set in the ReplicateNCRequestMsg, Windows does not resolve a target object that is in another partition.



This results in a cyclical dependency between the partitions being replicated, E.g. you can't commit all the linked attributes in the Domain partition until all the objects in the Configuration partition have been received, but you can't commit all the linked attributes in the Configuration partition until all the objects in the Domain partition have been received. This problem could be resolved if you manage to replicate the target object before the link is replicated.



My question is: how does the Windows replication behaviour resolve this problem?

- Does this mean the client always has to replicate the partitions in parallel? i.e. it can't replicate them in series because it can't commit an entire partition until it has received portions of another partition.

- What happens when the client encounters a linked attribute with an unknown target object? ProcessLinkValue() in the MS-DRSR seems to imply the client will ignore it (if it's already tried with GET_TGT). So how does the client ensure that it receives the linked attribute eventually?



Thanks for your help.



Tim




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.samba.org/pipermail/cifs-protocol/attachments/20170629/d73fc4e9/attachment-0001.html>


More information about the cifs-protocol mailing list