joeware - never stop exploring... :)

1. Determine if your application has been configured to use a specific named Domain Controller, if so use it.

    a. For debugging purposes only

    b. Application specific (where do you keep your configuration info?)

2. Determine if your application has been configured to use a specific named Site at startup, if so do not Autodiscover site.

    a. For debugging purposes only

    b. Application specific (where do you keep your configuration info?)

3. Determine if your client has been configured (hardcoded) to use a specific AD Site.

    a. For debugging purposes only

    b. Platform specific

        i. Windows

           1. REG_SZ registry value "SiteName" under HKLM\System\CurrentControlSet\Services\NetLogon\Parameters

4. Determine if your client has "cached" a previously used AD Site.

    a. Used to improve efficiency

    b. Platform specific

        i. Windows

           1. REG_SZ registry value "DynamicSiteName" under HKLM\System\CurrentControlSet\Services\NetLogon\Parameters

5. If you have not ascertained the site from steps 1-4 then you need to dynamically determine the site the machine is in before you can request the DNS records for the closest site.

    a. Active Directory supports an anonymous UDP (optionally TCP) RootDSE query to return basic information from Active Directory. This is generally called an LDAP “Ping” and is part of the DC Locator process used by all Windows machines. This will be the subject of a future blog post.

    b. To find an initial domain controller to bootstrap this step you can either use the domain FQDN or perform a DNS SRV lookup of the _ldap._tcp.<DOMAIN FQDN> record as illustrated in step 6c. Be prepared to be able to use either/or because it is possible that some “unique” local design decisions could make one or the other unavailable with the domain group host record FQDN being the most likely to be unavailable. If both are unavailable you probably have no choice but to ask someone in the AD space for a bootstrap server.

    c. The LDAP “Ping” query will return useful information about the domain controller including:

        i. DC Flags

           1. Some examples:

               a. Is the PDC

               b. Is in the same site as client

               c. Is Writable

               d. Is RODC

               e. DS Major version (DS_6,DS_7,DS_10, etc)

        ii. Forest Name

        iii. Domain Name

        iv. DNS HostName of Domain Controller

        v. IP address of Domain Controller

        vi. NetBIOS Domain Name

        vii. NetBIOS Domain Controller Name

        viii. Site Name of Domain Controller

        ix. Site Name of Client

    d. LDAP Search parameters

        i. IP Protocol: Either but UDP is lighter/faster and way fewer packets (duh)

        ii. Search Credentials: Anonymous

        iii. Search Base: NULL

        iv. Search Scope: BASE

        v. Search Filter: (&(DnsDomain=FQDN)(NtVer=\FF\00\00\00))

            1. Really this value varies based on options.

        vi. Search Attributes to return: NetLogon

    e. The NetLogon attribute returned is a BLOB attribute. The structures of the BLOBs that can be returned are dependent on NtVer parameter in the LDAP query and details are available in the link in the next bullet.

    f. Additional Detail can be found in the MS-ADTS: Active Directory Technical Specification of the Windows Protocol documents, specifically section 6.3 under Structures and Constants and LDAP Ping

        i. https://msdn.microsoft.com/en-us/library/cc223799.aspx

6. Retrieve a list of the domain controllers or global catalog domain controllers servicing the site (previously determined) for the domain we need a domain controller for from DNS.

    a. Domain Controllers:

        i. Query DNS for SRV records

           1. _ldap._tcp.<SITENAME>._sites.dc._msdcs.<DOMAIN FQDN>.

           2. You want hostnames, priority, and weight.

           3. Example using nslookup

               a. nslookup -type=srv _ldap._tcp.site2._sites.dc._msdcs.child.k16tst.test.loc

               b. Global Catalogs:

                    i. Query DNS for SRV records

                       1. _gc._tcp.<SITENAME>._sites.<FOREST ROOT FQDN>.

                       2. You want hostnames, priority, and weight.

                       3. Example using nslookup

                           a. nslookup -type=srv _gc._tcp.site2._sites.k16tst.test.loc

               c. If no domain controllers are found in the site you need that fit your criteria you will either have to select a different site OR open your location process to ANY domain controller in the domain.

                   i. Query DNS for SRV records

                      1. _ldap._tcp.<DOMAIN FQDN>

                      2. You want hostnames, priority, and weight.

                      3. Example using nslookup

                          a. nslookup -type=srv _ldap._tcp. child.k16tst.test.loc

               d. If no Global Catalogs are found in the site you need that fit your criteria you will either have to select a different site OR open your location process to ANY global catalog in the entire forest.

                   i. Query DNS for SRV records

                      1. _gc._tcp.<DOMAIN FQDN>

                      2. You want hostnames, priority, and weight.

                      3. Example using nslookup

                          a. nslookup -type=srv _gc._tcp.k16tst.test.loc

7. Find the PDC for the domain of the domain controllers you are looking at and exclude it from your list of domain controllers for consideration UNLESS that is the only domain controller available in the site that meets your criteria.

    a. Query DNS for SRV records

         i. _ldap._tcp.pdc._msdcs.<DOMAIN FQDN>

         ii. Example using nslookup

            1. nslookup -type=srv _ldap._tcp.pdc._msdcs.k16tst.test.loc

8. Validate the list of domain controllers to produce a final list of functioning validated domain controllers sorted by validation performance and DNS SRV record priority.

    a. Validation can vary greatly based on needs. Some minimal validation steps to consider:

        i. Anonymous UDP LDAP Ping or other basic RootDSE query with short timeout (1 second – configurable)

           1. This should be a configuration that can optionally be disabled in case UDP 389 is blocked.

           2. Better than ICMP ping because ping is often blocked by firewalls.

           3. Track response time.

        ii. Anonymous TCP RootDSE query with short timeout (1 second – configurable)

            1. TCP should be used because all normal requests will be TCP based.

            2. Track response time.

            3. Validate isSynchronized is TRUE

            4. Validate isGlobalCatalogReady is TRUE if looking for GC.

            5. Validate defaultNamingContext is what you need if not looking for GC.

            6. Validate supportedControl if any special needs.

            7. Validate forestFunctionality if specific forest mode is needed

            8. Validate domainFunctionality if specific domain mode is needed

            9. Validate domainControllerFunctionality if specific domain controller mode is needed

        iii. If using LDAPS for binds/queries perform an LDAPS RootDSE query with short timeout (1 second – configurable)

            1. Validates that LDAPS is functioning.

                a. LDAP could be working fine between client and server and LDAPS not be working because of a multitude of reasons

                b. Great for troubleshooting (LDAP works but LDAPS doesn’t)

            2. Performance of LDAPS varies from LDAP, more overhead on client and domain controller.

            3. Track response time.

    b. If no domain controllers (or Global Catalogs if that is what you are looking for) pass your criteria validation checks then you need to go back to step 6 and either use a different site OR look up machines for the whole domain or forest as indicated.

9. Merge the list of the highest performing domain controllers with the DNS SRV record priority.

10. Select as many of the domain controllers as you need (current use and spares) from the highest performing domain controllers.

11. Use domain controller(s) from list based on previous sorting and if using multiple LDAP connections also by DNS SRV record weight (if available).

12. Repeat the process regularly (every few hours) or anytime you hit a failure to connect or return a result set or performance drags.

1. Determine if your application has been configured to use a specific named Domain Controller, use it.

    a. For debugging purposes only

    b. Application specific (where do you keep your configuration info?)

2. Determine if your application has been configured to use a specific named Site at startup, do not Autodiscover site.

    a. For debugging purposes only

    b. Application specific (where do you keep your configuration info?)

3. If you do not have a hard-coded site and want to retrieve the name of site that your machine is in you can call DsGetSiteName

    a. Note that this is NOT required, the API calls later do not need to have the site name fed to them. If no sitename is specified the API will find the current site under the covers for you.

    b. Useful for logging

    c. DWORD dwRV=DsGetSiteName(strComputerName, &strSiteName);

    d. strComputerName can be NULL if you are running from a machine joined to any domain in the forest that you are working on. If you are trying to get information for a domain in a different forest you will need to specify a machine (or domain FQDN) in the foreign forest.

    e. If the machine has a cached site or is hardcoded to use a specific site this API call will honor that configuration.

        i. This hardcoding of the site on Windows is done via the REG_SZ registry value named "SiteName" under the following registry key:

           1. HKLM\System\CurrentControlSet\Services\NetLogon\Parameters.

4. Retrieve a single domain controller for the site using DsGetDcName.

    a. Use this if you trust Windows to give you the perfect domain controller right off the bat.

    b. DWORD dwRV=DsGetDcName(strComputerName, strDomainName, NULL, strSiteName, ulFlags, &DCInfo);

        i. strComputerName can be NULL if you are running from a machine joined to any domain in the forest that you are working on. If you are trying to get information for a domain in a different forest you will need to specify a machine (or domain FQDN) in the foreign forest.

        ii. strSiteName is optional, if it is not specified the API will look at any client level hardcoding and/or caching or even back out to any domain controller in the entire domain if nothing else is available. If you do specify a site and you get to step 10 without any valid functioning domain controllers execute this call without specifying a site.

        iii. ulFlags is used to specify the requirements for the domain controller functionality.

5. Retrieve a set of domain controllers for the site using DsGetDcOpen and DsGetDcNext to produce a list of possible candidates.

    a. Use this to get a pool of domain controllers to perform your own validation on or to spread connections across

    b. You need to call DsGetDcOpen to get a handle and then loop through the returned domain controllers with DsGetDcNext.

        i. DWORD dwRV=DsGetDcOpen(strDomainName, DS_ONLY_DO_SITE_NAME, strSiteName, NULL, NULL, ulFlags, &getDCCxt);

        ii. DWPRD dwRV=DsGetDcNext(getDCCxt, NULL, NULL, &DnsHostName);

        iii. strSiteName is optional, if it is not specified the API will look at any client level hardcoding and/or caching or even back out to any domain controller in the entire domain if nothing else is available. If you do specify a site and you get to step 10 without any valid functioning domain controllers execute this call without specifying a site.

        iv. ulFlags is used to specify the requirements for the domain controller functionality.

6. If you only retrieved a single domain controller in Step 4 continue with your trusting model, log it and just start using it.

7. Retrieve the PDC for the domain you are trying to find a domain controller for and exclude it from your list of domain controllers unless it is the only domain controller in your list.

    a. As in step 5 use DsGetDCOpen and use the Options Flag DS_PDC_REQUIRED.

8. Validate the list of domain controllers to produce a final list of functioning validated domain controllers sorted by validation performance and DNS SRV record priority.

    a. Validation can vary greatly based on needs. Some minimal validation steps to consider:

        i. Anonymous UDP LDAP Ping or other basic RootDSE query with short timeout (1 second – configurable)

           1. This should be a configuration that can optionally be disabled in case UDP 389 is blocked.

           2. Better than ICMP ping because ping is often blocked by firewalls.

           3. Track response time.

        ii. Anonymous TCP RootDSE query with short timeout (1 second – configurable)

            1. TCP should be tested because all normal requests will be TCP based.

            2. Track response time.

            3. Validate isSynchronized is TRUE

            4. Validate isGlobalCatalogReady is TRUE if looking for GC.

            5. Validate defaultNamingContext is what you need if not looking for GC.

            6. Validate supportedControl if any special needs.

            7. Validate forestFunctionality if specific forest mode is needed

            8. Validate domainFunctionality if specific domain mode is needed

            9. Validate domainControllerFunctionality if specific domain controller mode is needed

        iii. If using LDAPS for binds/queries perform an LDAPS RootDSE query with short timeout (1 second – configurable)

            1. Validates that LDAPS is functioning.

                a. LDAP could be working fine between client and server and LDAPS not be working because of a multitude of reasons

                b. Great for troubleshooting (LDAP works but LDAPS doesn’t)

            2. Performance of LDAPS varies from LDAP, more overhead on client and domain controller.

            3. Track response time.

9. Optionally look up the SRV records for the domain controllers you are using to determine their priority and merge that info with the list of best performing domain controllers.

10. Select as many of the domain controllers as you need (current use and spares) from the prioritized list.

      a. If you get here without any valid functional domain controllers loop back up to step 4 or 5 and don’t specify a site.

11. Use domain controller(s) from list based on previous sorting and if using multiple LDAP connections by DNS SRV record weight (if available).

12. Repeat the process regularly (every few hours) or anytime you hit a failure to connect or return a result set or performance drags.

    joe

0. This does not follow the previously mentioned high level steps, this method is “I hand my problem over to Microsoft to figure out on my behalf”.

1. Call ldap_sslinit specifying domain name or forest name (for Global Catalog)

    a. LDAP port

        i. LDAP* pLdap=ldap_sslinit(L"domain.forestroot.com", LDAP_PORT, false);

    b. LDAPS (SSL) port

        i. LDAP* pLdap=ldap_sslinit(L"domain.forestroot.com", LDAP_SSL_PORT, true);

    c. Global Catalog port

        i. LDAP* pLdap=ldap_sslinit(L"forestroot..com", LDAP_GC_PORT, false);

    d. Global Catalog SSL port

        i. LDAP* pLdap=ldap_sslinit(L"forestroot..com", LDAP_SSL_GC_PORT, true);

2. That was Easy™

0. If you are on Windows use the Windows LDAP Library and let it handle all of this for you.

1. Determine if your application has been configured to use a specific named Domain Controller, use it.

    a. For debugging purposes only

2. Determine if your application has been configured (hardcoded) to use a specific AD Site, do not Autodiscover site.

    a. For debugging purposes only

3. Determine if your client has been configured (hardcoded) to use a specific AD Site, do not Autodiscover site.

    a. For debugging purposes only

4. Determine if your client has "cached" a previously used AD Site.

    a. Used to improve efficiency especially between reboots, app restarts.

5. If you do not have a site from the following steps, determine the site (Autodiscover) the machine is in.

6. Retrieve a list of the domain controllers which are servicing the site (previously determined) for the domain we need a domain controller for.

7. [Optional but recommended] Find the PDC for the domain (or domains) of the domain controllers you are looking at and exclude it (them) from your list of domain controllers for consideration UNLESS that is (those are) the only domain controller available.

8. Validate the list of domain controllers to produce a final list of functioning validated domain controllers sorted by validation performance and DNS SRV record priority.

9. If no valid functioning domain controllers make it through steps 1-8 then you either need to select another site (hopefully “close” to the first site) to look in for domain controllers or you need to process steps 6-8 again but with a wider focus of any domain controllers in the entire domain.

10. Use domain controller(s) from list based on previous sorting and if using multiple LDAP connections distribute LDAP requests by DNS SRV record weight.

11. Repeat the process regularly (every few hours) or anytime you hit a failure to connect or get a result set or if you detect performance is dragging.

Coming Soon: Additional posts with details.

   joe

Active Directory location capability is all based on open standards based DNS SRV records which are designed to offer location capability for ANY services. The DNS SRV record RFC is RFC2782 which you can find at https://www.ietf.org/rfc/rfc2782.txt. There are two main components of the SRV process for domain controllers; registration and lookup.

First the domain controllers figure out what SRV records need to be registered for its services depending on various configurations in Active Directory and the registry of each domain controller. Applications aren’t involved in this process at all, they simply need to be able to lookup the results in DNS. The main issues that can occur on this side are DNS systems that aren’t properly allowing dynamic registrations or Active Directory admins misconfiguring sites and subnets and/or registry keys (directly or via GPO).

Second the clients that need to access Active Directory query DNS for the service’s SRV records in the specific sites that they need domain controllers OR look at the global set of service SRV records for all of Active Directory.

The service SRV records are significantly different from other well-known DNS record types such as A/HOST records or CNAME records in that there is a bunch of information packed into the records that allow for a fairly robust high availability service location system. They are exactly the same in that they can be dynamically updated and queried using open standards based DNS APIs.

SRV Record Components

Service SRV records can have multiple hosts and the following components are the publicly available pieces in DNS that make up each SRV record:

Record Name

• The actual name of the service record in DNS that you specify to look up the record.
• This is broken up of several components itself

• _<SERVICE>.<PROTOCOL>.<NAME>

• SERVICE: The service prefix for specified service such as LDAP.

• Note that there is no requirement for a service name to be prefixed with an underscore but they usually are. All of the SRV records published by AD are prefixed with an underscore.

PROTOCOL: The protocol the record is for such as TCP or UDP.

NAME: The DNS Zone the record lives in such as domain.com

Ex: _ldap._tcp.k16tst.test.loc

Priority

• The relative priority of the specified host for the record. The lower the value the more preferred the host. These values are for picking which hosts should be targeted first.
• Ex: 0

Weight

• The relative weight of the specified host for the record. The higher the value the more preferred. These values are for balancing load between multiple hosts with the same priority.
• Ex: 100

Port

• What port the service is available on for this specific host.
• Ex: 389

Svr HostName

• The canonical hostname of the target of the record.
• Ex: k16tst-dc1.k16tst.test.loc.

In addition to the above, each record also has a TTL specified for it. This controls how fast the records age out and changes will get updated down through the hierarchy of DNS servers and client caches. The lower the value the more “dynamic” the records can be to offer up different options, etc. Additionally the lower the value the higher the DNS Lookup and replication load there is on the systems as well.

Priority and Weight

Most of the components of a service SRV record should, generally speaking, be self-explanatory. The priority and weight are a little different as their proper use may not be obvious.

Each service record can have multiple SRV entries associated with it for each unique instance of the service. The priority and weight give hints on how the entries should be used.

The priority is a numeric value where the lowest value has the great preference. Use all of the entries with a priority of 0 before all of the entries with a priority of 1 before all of the entries with a priority of 10 before all of the entries with a priority of 100, etc. If none of the instances with the lowest priority are responding, drop to the next lowest priority, etc.

The weight is a numeric value where the highest value has the greatest preference. Unlike with priority all of the weights of the similar priority entries that are available are collected together and normalized to an overall value of 100% and that gives a ratio / percentage of how requests to each service instance should be balanced. Obviously, this should also be used dynamically in terms of which records are actually for available services at the time of use. This becomes more clear with the examples.

Ex 1: Say you have three instances of the service each with a priority of 0 and weight of 100 then you should balance the requests across all three instances equally, 33.333% per instance. If one of those instances becomes unavailable then you should balance the requests across the two remaining instances at 50% per instance.

Ex 2: Say you have three instances of the service each with a priority of 0 but two have a weight of 40 and one has a weight of 20 then out of every 10 requests 4 should go to service instance 1, 4 should go to service instance 2, and 2 should go to service instance 3. If service instance 2 with a weight of 40 becomes unavailable then for every 10 requests 7 should go to service instance 1 and 3 should go to service instance 2.

Ex 3: Say you have three instances of the service each with priority of 0 and weight of 100 and one instance of the service with a priority of 1 and a weight of 100. Requests to the service should be split three ways between the instances with a priority of 0. If all three instances become unavailable and ONLY after all three instances become unavailable then all request should go to the service instance with a priority of 1.

AD Service SRV Records

The SRV records you will see for AD include

• _ldap – LDAP service SRV records including normal LDAP and Global Catalog LDAP.
• _gc – LDAP server records used only for Global Catalog LDAP.
• _kerberos – Kerberos KDC service SRV records.
• _kpasswd – Kerberos Password Change service SRV records.

Here is an example of a complete set of records for the PDC of the root domain in a multi-domain forest with multiple sites. You can see the same information for any specific domain controller by looking at the C:\Windows\System32\Config\netlogon.dns file on each domain controller. In fact if you are missing AD SRV records in DNS this is the first place to look to troubleshoot.

_ldap._tcp.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.Default-First-Site-Name._sites.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.pdc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.98fd1190-e167-4734-a585-7981238a135e.domains._msdcs.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
b306bddc-2945-4a7d-b7ce-0bc829c55c5a._msdcs.k16tst.test.loc. 600 IN CNAME K16TST-DC1.k16tst.test.loc.
_ldap._tcp.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.Default-First-Site-Name._sites.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.gc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.Default-First-Site-Name._sites.gc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_gc._tcp.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_gc._tcp.Default-First-Site-Name._sites.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.DomainDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.Default-First-Site-Name._sites.DomainDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.ForestDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.Default-First-Site-Name._sites.ForestDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.RODCSite._sites.gc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_gc._tcp.RODCSite._sites.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.RODCSite._sites.DomainDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.RODCSite._sites.ForestDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.Default-First-Site-Name._sites.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.Default-First-Site-Name._sites.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_kerberos._udp.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_kpasswd._tcp.k16tst.test.loc. 600 IN SRV 0 100 464 K16TST-DC1.k16tst.test.loc.
_kpasswd._udp.k16tst.test.loc. 600 IN SRV 0 100 464 K16TST-DC1.k16tst.test.loc.
k16tst.test.loc. 600 IN A 192.168.0.75
gc._msdcs.k16tst.test.loc. 600 IN A 192.168.0.75
DomainDnsZones.k16tst.test.loc. 600 IN A 192.168.0.75
ForestDnsZones.k16tst.test.loc. 600 IN A 192.168.0.75
_ldap._tcp.joenetlogontestsite._sites.gc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_gc._tcp.joenetlogontestsite._sites.k16tst.test.loc. 600 IN SRV 0 100 3268 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.joenetlogontestsite._sites.DomainDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.joenetlogontestsite._sites.ForestDnsZones.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.joenetlogontestsite._sites.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.joenetlogontestsite._sites.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.
_ldap._tcp.joenetlogontestsite._sites.dc._msdcs.k16tst.test.loc. 600 IN SRV 0 100 389 K16TST-DC1.k16tst.test.loc.
_kerberos._tcp.joenetlogontestsite._sites.k16tst.test.loc. 600 IN SRV 0 100 88 K16TST-DC1.k16tst.test.loc.

One Last Thing…

I love this model. I think it is extremely intelligent and useful. Microsoft was brilliant for their involvement in SRV records and use of it in this way. It takes you out of the hole you can be in by depending on any given machine to always be available whether that machine is a domain controller, a network switch, a Virtual IP / Load Balancer, whatever. This is an inexpensive globally redundant mechanism using functionality available in every network that when it is used properly is very useful and just outright awesome.

That being said I am also very disappointed because Microsoft didn’t use it for LDAPS or Global Catalog LDAPS records nor have an option to use it for ADLDS or even for the ADWS service that now runs on Domain Controllers and ADLDS servers for the AD PowerShell Cmdlets. Come on Microsoft. On the positive side, because it is all based on open standards, you can (and I have) write scripts/tools to add/remove additional records as you see fit.

If you haven’t checked it out before, check out DNSSrvRec which is pretty much a quick and dirty tool that I wrote over a decade ago that allows you to quickly add and/or delete SRV records. You can find it at https://www.joeware.net/freetools/tools/dnssrvrec. It is so QnD there is a super obvious typo bug that is seen as soon as you run it, but don’t worry, it doesn’t impact its functionality. You will note that the usage examples illustrate how to add _LDAPS records but I use this to this day for troubleshooting and temporally removing or fixing the normal AD SRV records when things are broken.

   joe

While working on some posts about writing code for leveraging Active Directory I realized that a very weak point I have run into with many (perhaps most) apps is the logging, particularly for use in troubleshooting and/or debugging of issues. I don’t care how good of a coder you are (or think you are), your code will eventually be smack dab in the middle of a problem and someone is going to have to troubleshoot it. So just assume that that is going to happen right from the jump and do something about it as you write your very first lines of code and you won’t be called “a moron coder” when someone runs into the issue that needs that debugging… NB: I myself have been called a moron coder, often by myself, sometimes multiple times in a single sitting.

Personally, I have always been a fan of plain text file logging because it is simple to implement and simple to use requiring no special viewers, is easily searchable, etc but in the end, ANY logging is better than no logging. If instead of text logging you would rather use binary or XML log entries and instead of a text file you would rather send logging events to SYSLOG or the Windows Event Log or Windows Debug Stream or a sweet MariaDB SQL Server or all of them or something else entirely please do so if it gets you to implement logging. There is a caveat in that the “something else” shouldn’t be console logging unless your console is a teletype of some sort because otherwise the logs will just scroll off the screen and never be seen again. That makes it too easy to lose salient details that way. Actually, on second thought, even if you do have a teletype console available don’t use console logging as your sole logging method. I used to have to look through reams of paper with console logging entries and I regularly missed things in it.

I realize that logging can impact performance so if you want to allow the consumer to control the impact they are experiencing but still allow them to get data that is useful for troubleshooting set up logging levels that can be configured such that the higher the logging level that is configured the more verbose the logging becomes. Visualize a sliding dial, less logging/higher performance to super verbose logging/lower performance. There should probably be at least 3-5 levels of verbosity from normal regular running and tracking high level events to full on debugging tracking everything that is happening. If you can figure out a way to do it, you could even log the performance hit currently being experienced because the level of logging. That would require some baselining in any given environment but could be done such that perhaps you occasionally run a “transaction”, whatever that is for your app, raw without logging and then run the same “transaction” with the logging and see what the delta is. The more you do that the closer you can be to defining the performance hit involved.

For AD interaction logging in particular, when configured for the most verbose logging you should log entries for the whole process to select the domain controller such as what DC was used to figure out the bootstrap information, anything read in from config files/registry/etc, DNS results, possible DCs, selected DCs and some of the information that is used to help you determine which DC to use (like performance, capabilities of the DCs, etc). You want to track all connection attempts, the parameters, controls, etc and their results (but don’t put in clear text passwords!) and performance of same. All queries again including parameters, controls, etc and the results and ta da… Performance of same. All updates again including parameters, controls, etc and the results and performance of same. You really want to include performance information (how long it takes to do various things – response time for connections and queries are usually quite useful), etc because some of the most common issues as you scale up environments is that something is “running slow” and you don’t want to make people have to break their arms trying to find convoluted ways to determine performance issues. If you are using multiple threads/processes then they should be uniquely identified in the logs as well so you can track the various streams that are likely occurring simultaneously. Oh, time stamps, lots and lots of time stamps. Perhaps the machine name the code is actually running on especially if you could have a pool of application servers running and want to consolidate the logging somehow. Pretty much ANYTHING that would help you troubleshoot issues with connecting and/or returning information and/or updating information in Active Directory or any other LDAP directory.

Also, and this is a bit of Psych 101, tie in non-configurable logging to the most verbose performance impacting logging levels that indicates what debug/hardcoding flags are set. Track everything put into the code to use for troubleshooting or debugging that someone may enable and actually forget about and leave running for long periods of time, this could include verbose debug logging levels being configured, hardcoding, etc. I see nothing wrong in stamping the log every 15 minutes, for example, when someone has hardcoded a specific machine or other resource to use when the application supports dynamic resource detection. No one wants to see errors popping up in their logs over and over again so when you do that for the debug configuration items you make it far more likely that someone will notice what is happening and will get it corrected. If there is a good reason for the debug logging then the admin/integrator/dev will understand and be able to properly look past those alert entries in the log for the limited period of time they are debugging and once they are done it will hopefully be so noisy and painful they will fix it. You can also put in regular warning reminders in the logs for perhaps some less than optimal configurations, especially configurations that impact security or performance. For example, a log entry once an hour saying “Hey goofball, you are sending passwords across an LDAP connection with no encryption, start using LDAPS or STARTTLS or LDAP Signing/Sealing!” That little bit could help keep your application from being used insecurely and jeopardizing a company.

This should be able to go unsaid but unfortunately I absolutely know better as I have seen regular occurrences of this rule being broken year over year but as I specifically called out above… Do NOT output clear text security secrets like passwords into log files. If you want to put passwords into the log files then a few basic rules:

1. It should only be secrets/passwords that you already know and control. I.E. Application ID passwords for YOUR application. DO NOT EVER output in any way shape or form the passwords of anyone using your application. You can and should log the IDs, but don’t even think about logging the passwords or even an encrypted form of the passwords. The user passwords should never be available outside of the memory space of the currently running application and it should be in memory only for a very short period of time as well, milliseconds at most, the time taken to authenticate the user.

2. The secrets that are ok for you to log you should encrypt in some way shape or form so a casual glance cannot pick up on them. And when I say encrypt I mean actually encrypt, don’t do something stupid like MIME encode them such that anyone who can grab the text can put it into one of a thousand different pages on the internet to revert back to clear text. Better than encryption is to use a hash. It still isn’t full proof but you can easily create the hash of your password that you think it is and then compare to the hash that is in the log to make sure it is the same.

3. If you ever gain knowledge or have a feeling that your secrets have been exposed to someone they shouldn’t be exposed to immediately change those secrets. Secrets like application / process passwords should be changed frequently, at least anytime someone leaves a team that knows the password and at a minimum annually though more preferably monthly, weekly, or even daily. I have even seen applications that changed their own passwords every 8-12 hours. That last one may be a bit excessive but it really depends on how critical or sensitive the information is that the secrets/passwords are protecting.

To wrap up logging… Produce good logging, you will thank yourself and operations will thank you as well when they are forced to use it. Be clear, be concise, be complete. Think about what information someone may need from your application when it is 2AM and it isn’t doing what it is supposed to be doing and someone needs to figure out what is wrong and they have called you to ask you to fix it. Things WILL go wrong, period. No one writes perfect code, no systems run 100% perfectly 100% of the time. It may not be your system or code that is failing but your code could still be blamed. You can NOT depend on other people, especially the people who support the underlying infrastructure that you use, to be able to tell you why your application is failing, it isn’t their job and very often, if the environment has any serious scale, they have no real capability to help you. It is all on you.

    joe