logo_kerberos.gif

Difference between revisions of "Projects/Samba4 Port"

From K5Wiki
Jump to: navigation, search
(Appendix 2: windc KDC Plugin for Account-AuthZ)
 
(458 intermediate revisions by 2 users not shown)
Line 4: Line 4:
 
Samba4, like earlier versions of Samba, uses Heimdal Kerberos.
 
Samba4, like earlier versions of Samba, uses Heimdal Kerberos.
 
The Samba4 Port project proposes to enable Samba4 to use MIT kerberos
 
The Samba4 Port project proposes to enable Samba4 to use MIT kerberos
instead. The near-term goal is that mixed krb5+AD deployments could
+
as an alternative. The near-term goal is that mixed krb5+AD deployments
use Samba4 to provide better interoperation between AD realms and krb5
+
could use Samba4 to provide better interoperation between AD realms
realms.
+
and MIT-krb5 realms.
   
The Samba4 team, the MIT Krb Consortium, RedHat, Ubuntu, and Sun all have
 
  +
Use case: For example, suppose a kerberos customer is deploying a network
shown some interest in this Samba4 Port project.
 
  +
with mixed operating systems using kerberos and would want to use one KDC
 
  +
for all of them. In this case, a single MIT Kerberos deployment should
== To do list ==
 
  +
be able to support both traditonal UNIX clients and servers, intermixed
 
  +
with Windows clients and Samba servers:
This is a task-list offered by Samba4's Andrew Bartlett,
 
but Andrew is unsure of how much of this list is already
 
available in MIT's 1.7 release.
 
 
=== Replace the MIT KDC's LDAP driver ===
 
 
<ol>
 
<ol>
<li> Our LDAP driver for the KDB needs to know how to do
 
  +
<li> The Windows clients should be able to use the MIT KDC(s) as AD servers,
Samba4's intricate canonicalization of server names,
 
  +
so as to authenticate themselves to Samba file-servers and to Windows
user-names, and realm names. </li>
 
  +
servers;
<li> AD-style aliases for HOST/ service names. </li>
 
<li> Implicit names for Win2k accounts. </li>
 
<li> Principal "types": client / server / krbtgs
 
<li> Most or all of this code is in 3 samba4 source files,
 
~1000 lines in all. </li>
 
</ol>
 
 
----
 
 
=== Use 1.7's AD-support features ===
 
This stuff should already just work:
 
<ol>
 
<li> PAC handling; </li>
 
<li> AD-style name canonicalization; </li>
 
<li> NT-ENTERPRISE names, which carry two realms; </li>
 
<li> CHECK_POLICY/AUDIT methods; </li>
 
<li> DCE_STYLE; </li>
 
<li> Accept legacy Samba3 clients' bad GSSAPI checksums; </li>
 
<li> Principal-manipulation functions; </li>
 
</ol>
 
 
----
 
 
=== MIT KDC changes ===
 
<ol>
 
<li> Add HBAC to the KDC's TGT-issuance, so that Samba4 can refuse TGTs
 
to kinit, based on time-of-day & IP-addr constraints;
 
<ol>
 
<li> LH: "use KRB5_KDB_METHOD_CHECK_POLICY_TGS method. We have access
 
to the complete request. See against_local_policy_tgs() in
 
policy.c .</li>
 
</ol>
 
 
</li>
 
</li>
<li> Add a heuristic for failed-kinit counts, to support AD-style
 
  +
<li> A Windows client's tickets will carry PACs, as usual for AD;
unified account-lockouts across all authentication methods
 
  +
</li>
(Krb, NTLM, LDAP simple bind, etc).
 
  +
<li> The UNIX clients should be able to access the KDC as a traditional
<ol>
 
  +
non-AD-style KDC, so as to access UNIX services securely;
<li> LH: "Use a KRB5_KDB_METHOD_AUDIT_AS method for this." </li>
 
  +
<li> A UNIX client's ticket will ''not'' carry a PAC, except when
</ol>
 
  +
the UNIX client accesses a Windows server
  +
([http://k5wiki.kerberos.org/wiki/Samba4:_Optional_PACs_for_Unix_clients '''Rationale'''])
  +
.
 
</li>
 
</li>
 
</ol>
 
</ol>
   
----
 
  +
The Samba4 team, the MIT Krb Consortium, RedHat, Ubuntu, and Sun all have
 
  +
shown some interest in this Samba4 Port project.
=== Controversial proposed changes for the port ===
 
  +
[http://k5wiki.kerberos.org/wiki/Supported_platforms_for_Samba4_port '''Here''']
 
  +
is a table showing which OS platforms are supported by Samba4, Heimdal, and MIT kerberos.
==== Maybe: Improve or replace MIT's DAL ====
 
  +
Summary: MIT-krb5 & Samba4 both run on Mac OS X, NetBSD, Debian, RedHat, Ubuntu, & Solaris.
Rewrite the MIT KDC's Data-Abstraction Layer (DAL),
 
mostly because the MIT KDC needs to see & manipulate
 
more LDAP detail, on Samba4's behalf;
 
<ol>
 
<li>
 
</ol>
 
   
 
----
 
----
   
==== Maybe not: Add a KDC-as-library API ====
 
  +
== Concise to-do list ==
Samba4 currently runs as a single process, and Samba4 invokes the Heimdal
 
KDC via a libkdc interface (KDC as library).
 
Andrew Bartlett says this libkdc interface is "nice to have",
 
not essential.
 
Tom Yu says adding a libkdc interface to MIT's code would be a lot
 
of work, but would tie naturally into code-cleanup work that MIT wants
 
to do, anyway.
 
   
----
 
  +
This is a condensed version of the
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_%28Andrew_Bartlett%29#Data-Abstraction_Layer_.28DAL.29 '''task-list'''] offered by Samba4's Andrew Bartlett,
  +
containing only what hasn't yet been done already by MIT.
   
== Andrew Bartlett's Task-List (June '09) ==
 
  +
The two big chunks of work are
<pre>
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#LDAP_driver '''LDAP Driver'''] and
Copyright Andrew Bartlett <abartlet@samba.org> 2005-2009
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Data-Abstraction_Layer_.28DAL.29 '''Replacing / improving MIT's DAL'''],
Copyright Donald T. Davis <don@mit.edu> 2009
 
  +
but the DAL work may not be needed.
   
Released under the GPLv3
 
  +
=== Replace the MIT KDC's LDAP driver ===
</pre>
 
   
Title: Porting Samba4 to MIT-Krb
 
  +
Samba4's LDAP driver for the MIT KDB needs to know how to do
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#LDAP_driver '''AD's intricate naming''']:
IPA v3 will use a version of Samba4 built on top of MIT's kerberos implementation, instead of Heimdal's version of kerberos.
 
 
Task list summary for porting changes needed, from andrew Bartlett:
 
* Rewrite or extend the LDAP driver that MIT-KDC will use.
 
* MIT KDC changes: rewrite DAL, add TGS-KBAC, enable PACs,...
 
* Full thread-safety for MIT's library code,
 
* Many small changes
 
 
=== Task list, without explanations: ===
 
 
Porting Samba4 to MIT-krb comprises four main chunks of work:
 
 
<ol>
 
<ol>
<li> Rewrite or extend the LDAP driver that MIT-KDC will use:
 
  +
<li> Canonicalization of server names,
<ol>
 
  +
user-names, and realm names. MIT 1.7 already
<li> Our LDAP driver for the KDB needs to know how to do
 
  +
[[#Use_1.7.27s_AD-support_features | '''supports canonicalization''']].
Samba4's intricate canonicalization of server names,
 
  +
</li>
user-names, and realm names. </li>
 
  +
<li> AD-style aliases for
<li> AD-style aliases for HOST/ service names. </li>
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Keytab-sharing_amongst_HOST.2F_service_names '''HOST/ service names'''].
<li> Implicit names for Win2k accounts. </li>
 
<li> Principal "types": client / server / krbtgs </li>
 
<li> Most or all of this code is in 3 source files,
 
~1000 lines in all. </li>
 
</ol>
 
 
</li>
 
</li>
<li> MIT KDC changes
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Implicit_names_for_Win2000_Accounts '''Implicit names''']
<ol>
 
  +
for Win2k accounts.
<li> Rewrite the MIT KDC's Data-Abstraction Layer (DAL),
 
mostly because he MIT KDC needs to see & manipulate
 
more LDAP detail, on Samba4's behalf; </li>
 
<li> Add HBAC to the KDC's TGT-issuance, so that Samba4
 
can refuse TGTs to kinit, based on time-of-day &
 
IP-addr constraints; </li>
 
<li> Turn on MIT-krb 1.7's PAC handling </li>
 
<li> Add bad-password counts, for unified account-lockouts
 
across all authT methods (Krb, NTLM, LDAP simple bind,
 
etc) </li>
 
</ol>
 
 
</li>
 
</li>
<li> Make sure MIT's library code is more fully thread-safe,
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Principal_.22types.22 '''Principal "types":'''] client / server / krbtgs
by replacing all global and static variables with context
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Flexible_server-naming '''Flexible server-naming''']
parameters for the library routines. This may already be
 
done.
 
 
</li>
 
</li>
<li> Many small changes (~15)
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A_Keytabs_.26_Name-canonicalization '''Keytabs & name-canonicalization''']
<ol>
 
<li> Some extensions to MIT's libkrb5 & GSSAPI libraries,
 
including GSSAPI ticket-forwarding </li>
 
<li> Some refitting in Samba4's use of the MIT libraries; </li>
 
<li> Make sure Samba4's portable socket API works,
 
including "packet too large" errors; </li>
 
<li> MIT's GSSAPI code should support some legacy Samba3
 
clients that present incorrectly-calculated checksums;</li>
 
<li> Samba4 app-server-host holds a [[UTF-16 PW]], plus a
 
key bitstring; </li>
 
<li> In-memory-only credentials cache; </li>
 
<li> In-memory-only keytab (nice to have); </li>
 
<li> Get OSS NTLM authT library (Likewise Software?); </li>
 
<li> Special Heimdal-specific functions; </li>
 
<li> Principal-manipulation functions; </li>
 
<li> Special check for misconfigured Samba4 hostnames; </li>
 
<li> Improved krb error-messages; </li>
 
<li> improved krb logging </li>
 
<li> MS GSSMonger test-suite </li>
 
<li> Testsuite for kpasswd daemon </li>
 
</ol>
 
 
</li>
 
</li>
 
</ol>
 
</ol>
 
  +
Most or all of Heimdal's LDAP driver code is in
=== Introduction: ===
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#LDAP_driver '''three Samba4 source files'''],
This document should be read alongside
 
  +
~1000 lines in all.
the Samba4 source code, as follows:
 
<ol>
 
<li> For [[DAL and KDC requirements]], please see Samba4's
 
'''source4/kdc/hdb-samba4.c''' in particular. This file
 
is an implementation against Heimdal's HDB abstraction
 
layer, and is the biggest part of the samba-to-krb
 
glue layer, so the main part of the port to MIT is
 
to replace hdb-samba4 with a similar glue layer
 
that's designed for MIT's code. </li>
 
<li> Samba4's [[PAC requirements]] are implemeneted in
 
'''source4/kdc/pac-glue.c''' </li>
 
<li> Both of the above two layers are [[Heimdal plugins]], and
 
both get loaded in '''source4/kdc/kdc.c''' </li>
 
<li> For [[GSSAPI requirements]], see auth/gensec/gensec_gssapi.c
 
(the consumer of GSSAPI in Samba4)</li>
 
<li> For [[Kerberos library requirements]], see
 
'''auth/kerberos/krb5_init_context.c''' .</li>
 
<li> [[Samba's credentials system]] wraps GSS creds,
 
just as GSS creds wrap around krb5 creds. For the
 
interaction between Samba4 credential system and GSSAPI
 
and Kerberos, see '''auth/credentials/credentials_krb5''' . </li>
 
</ol>
 
   
 
----
 
----
   
=== LDAP driver ===
+
=== Small changes ===
Rewrite or extend the LDAP driver that MIT-KDC will use.
+
Of the things on this list, only NTLM support (bullet 2)
Most or all of the Heimdal LDAP driver code is in three source
+
is needed for the Samba4 KDC port.
files, ~1000 lines in all. These files are in '''samba4/kdc''' :
+
The other tasks are all application-library stuff,
* '''hdb-samba4.c''' (samba4-to-kdb glue-layer plugin)
+
and arguably aren't needed at all, because Samba3
* '''pac-glue.c''' (samba4's pac glue-layer plugin)
+
already works well with MIT application libraries.
* '''kdc.c''' (loads the above two plugins).
+
<ol>
+
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#MIT_libraries '''MIT library changes''']
==== Name Canonicalization ====
+
</li>
The LDAP driver for the Samba4 KDB needs to know how to do
+
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#NTLM_support '''Samba4/AD libraries: NTLM support''']. See also
Samba4's intricate canonicalization of server names,
+
[http://k5wiki.kerberos.org/wiki/Samba4_Port:_NTLM_thread '''this Sept-2009 NTLM thread'''] (this implies to me that a GSS NTLM mech is not an immediate requirement - LH)
user-names, and realm names.
+
</li>
<ol>
+
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Key-handling_changes '''Key-handling changes''']]
<li> [[For hostnames & usernames,]] alternate names appear in
+
</li>
LDAP as extra values in the multivalued "principal name"
+
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A_Extra_krb_library_functions '''Extra Krb library functions''']
attributes:
+
</li>
<ol>
+
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Error-handling.2C_logging.2C_testing '''Error-handling, logging, testing''']
<li> For a hostname, the alternate names (other than
 
the short name, implied from the CN), are stored in
 
the servicePrincipalName. </li>
 
<li> For a username, the alternate names are stored in
 
the userPrincipalName attribute, and can be long
 
email-address-like names, such as joe@microsoft.com
 
(see "Type 10 names," below). </li>
 
</ol></li>
 
<li> [[GSSAPI layer requirements]]: The MIT Krb5 libs (including
 
GSSAPI) do not enable the AS to send kinit a TGT containing
 
a different realm-name than what the client asked for,
 
even in U/L case differences. Heimdal has the same problem,
 
and this applies to the krb5 layer too, not just GSSAPI.
 
There are two kinds of name-canonicalization that can
 
occur on Windows:
 
<ol>
 
<li> Lower-to-upper case conversion, because Windows domain
 
names are usually in upper case; </li>
 
<li> An unrecognizable subsitution of names, such as might
 
happen when a user requests a ticket for a NetBIOS domain
 
name, but gets back a ticket for the corresponging FQDN.</li>
 
</ol>
 
As developers, we should test if the AD KDC's name-canonicalisation can be turned off with the KDCOption flags in the AS-REQ or TGS-REQ; Windows clients always send the Canonicalize flags as KDCOption values. </li>
 
</li>
 
<li> [[Principal Names, long and short names:]]
 
AD's KDC does not canonicalize servicePrincipalNames, except
 
for the realm in the KDC reply. That is, the client gets
 
back the principal it asked for, with the realm portion
 
'fixed' to uppercase, long form.
 
 
Samba4 does some canonicalization, though Heimdal doesn't
 
canonicalize names itself: For hostnames and usernames,
 
Samba4 canonicalizes the requested name only for the LDAP
 
principal-lookup, but then Samba4 returns the retrieved LDAP
 
record with the request's original, uncanonicalized hostname
 
replacing the canonicalized name that actually was found. </li>
 
<li> [[Usernames]]: AndrewB says that Samba4 used to return
 
the canonicalized username exactly as retrieved from LDAP.
 
The reason Samba4 treated usernames differently was that
 
the user needs to present his own canonicalized username
 
to servers, for ACL-matching. For hostnames this isn't
 
necessary. </li>
 
<li> [[Realm-names]]: AD seems to accept a realm's short name
 
in krb-requests, at least for AS_REQ operations, but the
 
AD KDC always performs realm-canonicalisation, which
 
converts the short realm-name to the canonical long form.
 
So, this causes pain for current krb client libraries.
 
[[Punchline]]: For bug-compatibility, we may need to
 
selectively or optionally disable the MIT-KDC's name-
 
canonicalization. </li>
 
<li> [[Application-code:]]
 
Name-canonicalisation matters not only for the KDC, but
 
also for app-server-code that has to deal with keytabs.
 
Further, with credential-caches, canonicalization can
 
lead to cache-misses, but then the client just asks for
 
new credentials for the variant server-name. This could
 
happen, for example, if the user asks to access the
 
server twice, using different variants of the server-name.</li>
 
<li> [[Doubled realm-names]]: We also need to handle type 10
 
names (NT-ENTERPRISE), which have a full principal name
 
in the principal field, unrelated to the realm. Thus,
 
the principal field contains both principal & realm names,
 
while the realm field contains a realm name, too, possibly
 
different.
 
This is used for AD's "email-address-like login-names."
 
For example, an NT-ENTERPRISE principal name might look like:
 
<pre>
 
joeblow@microsoft.com@NTDEV.MICROSOFT.COM ,
 
<--principal field-->|<----realm name--->|
 
</pre>
 
Where joe@microsoft.com is the leading portion, and
 
NTDEV.MICROSOFT.COM is the realm.
 
 
</li>
 
</li>
 
</ol>
 
</ol>
 
==== Keytab-sharing amongst HOST/ service names ====
 
AD keeps a list of service-prefixed aliases for the host's
 
principal name. The AD KDC reads & parses this list, so
 
as to allow the aliased services to share the HOST/ key.
 
This means that every ticket-request for a service-alias
 
gets a service-ticket encrypted in the HOST/ key.
 
 
For example, this is how HTTP/ and CIFS/ can use the
 
HOST/ AD-LDAP entry, without any explicitly CIFS-prefixed
 
entry in the host's servicePrincipalName attribute. In the
 
app-server host's AD record, the servicePrincipalName says
 
only HOST/my.computer@MY.REALM , but the client asks
 
for CIFS/my.omputer@MY.REALM tickets. So, AD looks in
 
LDAP for both name-variants, and finds the HOST/ version,
 
In AD's reply, AD replaces the HOST/ prefix with CIFS/ .
 
We implement this in '''hdb-ldb'''.
 
[[(TBD: Andrew, is this correct?:)]]
 
 
List of HOST/ aliases: Samba4 currently uses only a small
 
set of HOST/ aliases, called "sPNMappings:"
 
 
<pre>
 
host=ldap,dns,cifs,http .
 
</pre>
 
BTW, dns's presence in this short list is a bug, somehow.
 
 
AD's real sPNMappings list has 52 entries:
 
<pre>
 
host =
 
alerter eventsystem netlogon rpclocator tapisrv
 
appmgmt fax netman rpcss time
 
browser http nmagent rsvp trksvr
 
cifs ias oakley samss trkwks
 
cisvc iisadmin plugplay scardsvr ups
 
dcom mcsvc policyagent scesrv w3svc
 
dhcp messenger protectedsto schedule wins
 
dmserver msdtc rasman scm www
 
dns msiserver remoteaccess seclogon
 
dnscache netdde replicator snmp
 
eventlog netddedsm rpc spooler
 
</pre>
 
Domain members that expect the longer list will break in
 
Samba4, as of 6/09. AB says he'll try to fix this right
 
away. There is another post somewhere (ref lost for the
 
moment) that details where active directory stores this long
 
list of stored aliases for HOST/.
 
 
==== Implicit names for Win2000 Accounts ====
 
AD keys its
 
server-records by CN or by servicePrincipalName, but a
 
win2k box's server-entry in LDAP doesn't include the
 
servicePrincipalName attribute, So, win2k server-accounts
 
are keyed by the CN attribute instead. Because AD's LDAP
 
doesn't have a servicePrincipalName for win2k servers'
 
entries, Samba4 has to have an implicit mapping from
 
host/computer.full.name and from host/computer, to the
 
computer's CN-keyed entry in the AD LDAP database, so as to
 
be able to find the win2k server's host name in the KDB.
 
 
==== Principal "types" ====
 
Samba4 has modified Heimdal's 'hdb' interface to specify
 
the 'class' of Principal being requested. This allows
 
us to correctly behave with the different 'classes' of
 
Principal name. This is necessary because of AD's LDAP
 
structure, which uses very different record-structures
 
for user-principals, trust principals & server-principals.
 
We currently define 3 classes:
 
* client (kinit)
 
* server (tgt)
 
* krbtgt the TGS's own ldap record
 
Samba4 also now specifies the kerberos principal as an
 
explicit parameter to LDB_fetch(), not an in/out value
 
on the struct hdb_entry parameter itself.
 
   
 
----
 
----
   
=== MIT KDC changes ===
+
=== Use 1.7's AD-support features ===
+
This stuff should already just work:
==== Data-Abstraction Layer (DAL) ====
 
It would be good to
 
rewrite or circumvent the MIT KDC's DAL, mostly because
 
the MIT KDC needs to see & manipulate more LDAP detail,
 
on Samba4's behalf. AB says the MIT DAL may serve well-
 
enough, though, mostly as is. AB says Samba4 will need
 
the private pointer part of the KDC plugin API, though,
 
or the PAC generation won't work (see sec.2.c, below).
 
 
<ol>
 
<ol>
<li> MIT's DAL calls lack context parameters (as of 2006),
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A.2A_Turn_on_MIT-krb_1.7.27s_PAC_handling '''PAC handling''']; </li>
so presumably they rely instead on global storage, and
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Name_Canonicalization '''AD-style name canonicalization''']; </li>
aren't fully thread-safe.
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Doubled_realm-names '''NT-ENTERPRISE names'''],
</li>
 
  +
which carry two realm-suffixes; </li>
<li> In Novell's pure DAL approach, the DAL only read in the
 
  +
<li> CHECK_POLICY/AUDIT methods (needed for
principalName as the key, so it had trouble performing
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A.2A.2A_Add_access-control_to_the_TGS '''TGS access-control''']); </li>
access-control decisions on things other than the user's
 
  +
<li> DCE_STYLE Challenge/Response handshakes: see
name (like the addresses).
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A_Krb5_lib_.26_GSSAPI '''Krb lib & GSSAPI''']. </li>
</li>
 
  +
<li> Accept legacy Samba3 clients'
<li> Here's why Samba4 needs more entry detail than the DAL
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A.2A_Legacy_Samba3_clients_.26_GSSAPI '''bad GSSAPI checksums''']; </li>
provides: The AS needs to have ACL rules that will allow
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A_Extra_krb_library_functions '''Principal-manipulation functions''']; </li>
a TGT to a user only when the user logs in from the
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A.2A_State-machine_safety_for_krb_libraries '''State-machine safety''']; </li>
right desktop addresses, and at the right times of day.
 
This coarse-granularity access-control could be enforced
 
directly by the KDC's LDAP driver, without Samba having
 
to see the entry's pertinent authZ attributes. But,
 
there's a notable exception: a user whose TGT has
 
expired, and who wants to change his password, should
 
be allowed a restricted-use TGT that gives him access
 
to the kpasswd service. This ACL-logic could be buried
 
in the LDAP driver, in the same way as the TGS ACL could
 
be enforced down there, but to do so would just be even
 
uglier than it was to put the TGS's ACL-logic in the driver.
 
</li>
 
<li> Yet another complaint is that the DAL always pulls an
 
entire LDAP entry, non-selectively. The current DAL
 
is OK for Samba4's purposes, because Samba4 only reads,
 
and doesn't write, the KDB. But this all-or-nothing
 
retrieval hurts the KDC's performance, and would do so
 
even more, if Samba had to use the DAL to change KDB
 
entries.</li>
 
 
</ol>
 
</ol>
 
==== Add HBAC to the TGS ====
 
For AD compatibility, Samba4 has to be able to
 
refuse TGTs to kinit, based on time-of-day &
 
IP-address constraints. Samba4 doesn't yet have
 
these access control hooks in the Heimdal KDC.
 
We need to lockout accounts (eg, after 10 failed PW-
 
attempts), and to perform other controls. This is
 
standard AD behavior, that Samba4 needs to get right,
 
whether Heimdal or MIT-krb is doing the ticket work.
 
<ol>
 
<li> If PADL doesn't publish their patch for this,
 
we'll need to write our own.
 
</li>
 
<li> The '''windc''' plugin proivides a function for the main
 
access control routines. A new '''windc''' plugin function
 
should be added to increment the bad password counter
 
on failure.
 
</li>
 
<li> Samba4 doesn't yet handle bad password counts (or good
 
password notification), so that a single policy can be
 
applied against all means of checking a password (NTLM,
 
Kerberos, LDAP Simple Bind, etc). Novell's original DAL
 
did not provide a way to update the PW counts information.
 
</li>
 
<li> Tom Yu correctly points out that strictly speaking,
 
no-one can tell how many failed PW-attempts the user
 
makes against kinit's PW-encrypted tickets. Maybe we
 
can count rapidly-repeated kinit requests, instead?
 
</li>
 
<li> Nevertheless, we know that this is very much required in
 
AD, because Samba3 + eDirectory goes to great lengths to
 
update this information. This may have been addressed in
 
Simo's subsequent IPA-KDC design).
 
</li>
 
<li>
 
[[AllowedWorkstationNames and Krb5]]: Microsoft uses the
 
clientAddresses *multiple value* field in the krb5
 
protocol (particularly the AS_REQ) to communicate the
 
client's netbios name (legacy undotted name, <14 chars)
 
AB guesses that this is to support the userWorkstations
 
field (in user's AD record). The idea is to support
 
client-address HBAC restrictions, as was standard in NT:
 
The AD authentication server probably checks the netbios
 
address against this userWorkstations value (BTW, the
 
NetLogon server does this, too).</li>
 
</ol>
 
 
AB asks, "Is a DAL the layer we need for Samba4 HBAC?"
 
Looking at what we need to pass around, AB doesn't think
 
the DAL is the right layer; what we really want instead
 
is to create an account-authorization abstraction layer
 
(e.g., is this account permitted to login to this computer,
 
at this time?). Samba4 ended up doing account-authorization
 
inside Heimdal, via a specialized KDC plugin. For a summary
 
description of this plugin API, see Appendix 2.
 
 
==== Turn on MIT-krb 1.7's [[PAC handling]] ====
 
In addition, I have added a new interface '''hdb_fetch_ex()''',
 
which returns a structure including a private data-pointer,
 
which may be used by the '''windc''' plugin inferface functions.
 
The '''windc''' plugin provides the hook for the PAC.
 
   
 
----
 
----
   
=== State-machine safety for krb libraries ===
+
=== Controversial proposed changes for the port ===
Samba's client-side & app-server-side libraries are built
 
on a giant state machine, and as such have very different
 
requirements to those traditionally expressed for kerberos
 
and GSSAPI libraries.
 
<ol>
 
<li>
 
Samba requires all of the libraries it uses to be "state
 
machine safe" in their use of internal data. This does not
 
necessarily mean "thread safe," and an application could be
 
thread safe, but not state machine safe (if it instead used
 
thread-local variables). so, if MIT's libraries were made
 
thread-safe only by inserting spinlock() code, then the MIT
 
libraries aren't yet "state machine safe."
 
</li>
 
<li>
 
So, what does it mean for a library to be state machine safe?
 
This is mostly a question of context, and how the library manages
 
whatever internal state machines it has. If the library uses a
 
context variable, passed in by the caller, which contains all
 
the information about the current state of the library, then it
 
is safe. An example of this state is the sequence number and
 
session keys for an ongoing encrypted session).
 
</li>
 
<li>
 
The other issue affecting state machines is 'blocking' (waiting for a
 
read on a network socket). Samba's non-blocking I/O doesn't like
 
waiting for libkrb5 to go away for awhile to talk to the KDC.
 
</li>
 
<li>
 
Samba4 provides a hook 'send_to_kdc', that allows Samba4 to take over the
 
IO handling, and run other events in the meantime. This uses a
 
'nested event context' (which presents the challenges that the kerberos
 
library might be called again, while still in the send_to_kdc hook).
 
</li>
 
<li>
 
Heimdal has this 'state machine safety' in parts, and we have modified
 
Samba4's lorikeet branch to improve this behaviour, when using a new,
 
non-standard API to tunnelling a ccache (containing a set of tickets)
 
through the gssapi, by temporarily casting the ccache pointer to a
 
gss credential pointer. This new API is Heimdal's samba4-requested
 
gss_krb5_import_cred() fcn; this will have to be rewritten or ported
 
in the MIT port.
 
</li>
 
<li>
 
This tunnelling trick replaces an older scheme using the KRB5_CCACHE
 
environment variable to get the same job done. The tunnelling trick
 
enables a command-line app-client to run kinit tacitly, before running
 
GSSAPI for service-authentication. The tunnelling trick avoids the
 
more usual approach of keeping the ccache pointer in a global variable.
 
</li>
 
<li>
 
[Heimdal uses a per-context variable for the 'krb5_auth_context',
 
which controls the ongoing encrypted connection, but does use global
 
variables for the ubiquitous krb5_context parameter. (No longer true,
 
because the krb5_context global is gone now.)]
 
</li>
 
<li>
 
The modification that has added most to 'state machine safety' of
 
GSSAPI is the addition of the gss_krb5_acquire_creds() function.
 
This allows the caller to specify a keytab and ccache, for use by
 
the GSSAPI code. Therefore there is no need to use global variables
 
to communicate this information about keytab & ccache.
 
</li>
 
<li>
 
At a more theoretical level (simply counting static and global
 
variables) Heimdal is not state machine safe for the GSSAPI layer.
 
(But Heimdal is now (6/09) much more nearly free of globals.)
 
The Krb5 layer alone is much closer, as far as I can tell, blocking
 
excepted. .
 
</li>
 
<li>
 
As an alternate to fixing MIT Kerberos for better safety in this area,
 
a new design might be implemented in Samba, where blocking read/write
 
is made to the KDC in another (fork()ed) child process, and the results
 
passed back to the parent process for use in other non-blocking
 
operations.
 
</li>
 
<li>
 
To deal with blocking, we could have a fork()ed child per context,
 
using the 'GSSAPI export context' function to transfer
 
the GSSAPI state back into the main code for the wrap()/unwrap() part
 
of the operation. This will still hit issues of static storage (one
 
gss_krb5_context per process, and multiple GSSAPI encrypted sessions
 
at a time) but these may not matter in practice.
 
</li>
 
<li>
 
This approach has long been controversial in the Samba team.
 
An alternate way would be to be implement E_AGAIN in libkrb5: similar
 
to the way to way read() works with incomplete operations. to do this
 
in libkrb5 would be difficult, but valuable.
 
</li>
 
<li>
 
In the short-term, we deal with blocking by taking over the network
 
send() and recv() functions, therefore making them 'semi-async'. This
 
doens't apply to DNS yet.These thread-safety context-variables will
 
probably present porting problems, during the MIT port. This will
 
probably be most of the work in the port to MIT.
 
This may require more thorough thread-safe-ing work on the MIT libraries.
 
</li>
 
</ol>
 
   
----
 
   
=== Many small library changes (~15) ===
 
  +
==== Maybe: Improve or replace MIT's DAL ====
==== MIT libraries ====
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Data-Abstraction_Layer_.28DAL.29 '''Rewrite the MIT KDC's Data-Abstraction Layer (DAL)'''],
===== Krb5 lib & GSSAPI =====
 
  +
mostly because the MIT KDC needs to see & manipulate
Some extensions to MIT's [[libkrb5 & GSSAPI libraries]], including
 
  +
more LDAP detail, on Samba4's behalf;
GSSAPI ticket-forwarding: This is a general list of the other
 
extensions Samba4 has made to / need from the kerberos libraries
 
<ol>
 
</li>
 
<li> DCE_STYLE : Microsoft's hard-coded 3-msg Challenge/Response handshake
 
emulates DCE's preference for C/R. Microsoft calls this DCE_STYLE.
 
MIT already has this nowadays (6/09).
 
</li>
 
<li> gsskrb5_get_initiator_subkey() (return the exact key that Samba3
 
has always asked for. gsskrb5_get_subkey() might do what we need
 
anyway). This routine is necessary, because in some spots,
 
Microsoft uses raw Kerberos keys, outside the Kerberos protocols,
 
as a direct input to MD5 and ARCFOUR, without using the make_priv()
 
or make_safe() calls, and without GSSAPI wrappings etc.
 
</li>
 
<li> gsskrb5_acquire_creds() (takes keytab and/or ccache as input
 
parameters, see keytab and state machine discussion in prev section)
 
</li>
 
<li> The new function to handle the PAC fully
 
gsskrb5_extract_authz_data_from_sec_context()
 
need to test that MIT's PAC-handling code checks the PAC's signature.
 
</li>
 
<li> gsskrb5_wrap_size (Samba still needs this one, for finding out how
 
big the wrapped packet will be, given input length).
 
</ol>
 
===== Some refitting in Samba4's use of the MIT libs =====
 
===== Legacy Samba3 clients & GSSAPI =====
 
MIT's GSSAPI code should support some legacy Samba3
 
clients that present incorrectly-calculated checksums.
 
(Luke H: this is already in MIT's v1.7 .)
 
<ol>
 
<li> Old Clients (samba3 and HPUX clients) use 'selfmade'
 
gssapi/krb5 tokens for use in the CIFS session setup.
 
These hand-crafted ASN.1 packets don't follow rfc1964
 
(GSSAPI) perfectly, so server-side krblib code has to
 
be flexible enough to accept these bent tokens.
 
</li>
 
<li> It turns out that Windows' GSSAPI server-side code is
 
sloppy about checking some GSSAPI tokens' checksums.
 
During initial work to implement an AD client, it was
 
easier to make an acceptable solution (acceptable to
 
Windows servers) than to correctly implement the
 
GSSAPI specification, particularly on top of the
 
(inflexible) MIT Kerberos API. It did not seem
 
possible to write a correct, separate GSSAPI
 
implementation on top of MIT Kerberos's public
 
krb5lib API, and at the time, the effort did not
 
need to extend beyond what Windows would require.
 
</li>
 
<li> The upshot is that old Samba3 clients send GSSAPI
 
tokens bearing incorrect checksums, which AD's
 
GSSAPI library cheerfully accepts (but accepts
 
the good checksums, too). Similarly, Samba4's
 
Heimdal krb5lib accepts these incorrect checksums.
 
Accordingly, if MIT's krb5lib wants to interoperate
 
with the old Samba3 clients, then MIT's library will
 
have to do the same.
 
</li>
 
<li> Because these old clients use krb5_mk_req()
 
the app-servers get a chksum field depending on the
 
encryption type, but that's wrong for GSSAPI (see
 
rfc 1964 section 1.1.1). The Checksum type 8003
 
should be used in the Authenticator of the AP-REQ!
 
That (correct use of the 8003 type) would allow
 
the channel bindings, the GCC_C_* req_flags and
 
optional delegation tickets to be passed from the
 
client to the server. However windows doesn't seem
 
to care whether the checksum is of the wrong type,
 
and for CIFS SessionSetups, it seems that the
 
req_flags are just set to 0. This deviant checksum
 
can't work for LDAP connections with sign or seal,
 
or for any DCERPC connection, because those
 
connections do not require the negotiation of
 
GSS-Wrap paraemters (signing or sealing of whole
 
payloads). Note: CIFS has an independent SMB
 
signing mechanism, using the Kerberos key.
 
</li>
 
<li> For the code that handles the incorrect & correct
 
checksums, see heimdal/lib/gssapi/krb5/accept_sec_context.c,
 
lines 390-450 or so.
 
</li>
 
<li> This bug-compatibility is likely to be controversial
 
in the kerberos community, but a similar need for bug-
 
compatibility arose around MIT's & Heimdal's both
 
failing to support TGS_SUBKEYs correctly, and there
 
are numerous other cases.
 
see https://lists.anl.gov/pipermail/ietf-krb-wg/2009-May/007630.html
 
</li>
 
<li> So, MIT's krb5lib needs to also support old clients!
 
</li>
 
</ol>
 
==== Samba4 / AD libraries ====
 
===== NTLM support =====
 
For AD compatibility, Samba4 has to support NTLM authentication,
 
so the NTLM library has to be able to access the MIT KDB.
 
AB says we should be able to get an OSS NTLM authentication
 
library: AB says Likewise software probably will give us their
 
OSS "NTLM for MIT-krb" implementation.
 
   
===== Samba4's portable socket API =====
+
==== Maybe, or not: Add a KDC-as-library API ====
(This seems to assume that we'll build a libkdc, which seems unlikely.)
+
Samba4 currently runs as a single process, and Samba4's smbd invokes the Heimdal KDC via a
Make sure Samba4's portable socket API works:
+
[http://k5wiki.kerberos.org/wiki/Samba4_port:_libkdc_Interface#krb5_kdc_update_time.28.29 '''libkdc interface'''] (KDC as library).
 
<ol>
 
<ol>
<li> An important detail in the use of libkdc is that samba4 uses its
 
  +
<li> Rationale:
own socket lib. This allows the KDC code to be as portable as
 
  +
# smbd uses the libkdc interface to configure the KDC, both at startup & during runtime.
the rest of samba, but more importantly it ensures consistancy
 
  +
# Samba4's build/test environment uses libkdc's socket-passing, to simulate network traffic.
in the handling of requests, binding to sockets etc.
 
 
</li>
 
</li>
<li> To handle TCP, we use of our socket layer in much the same way as
 
  +
<li> Andrew Bartlett says this libkdc interface is
we deal with TCP for CIFS. Tridge created a generic packet handling
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#libkdc '''"nice to have"'''],
layer for this.
 
  +
but not essential for getting the port to work.
 
</li>
 
</li>
<li> For the client, samba4 likewise must take over the socket functions,
 
  +
<li> Tom Yu says adding a libkdc interface to MIT's code would be a lot
so that our single thread smbd will not lock up talking to itself.
 
  +
of work, but would tie naturally into code-cleanup work that MIT wants
(We allow processing while waiting for packets in our socket routines).
 
  +
to do, anyway.
send_to_kdc() presents to its caller the samba-style socket interface,
 
but the MIT port will reimplement send_to_kdc(), and this routine will
 
use internally the same socket library that MIT-krb uses.
 
 
</li>
 
</li>
<li> The interface we have defined for libkdc allows for packet injection
 
  +
<li> Sam Hartman says he needs the libkdc interface, too, for his work on PK-U2U (but not immediately).
into the post-socket layer, with a defined krb5_context and
 
kdb5_kdc_configuration structure. These effectively redirect the
 
kerberos warnings, logging and database calls as we require.
 
 
</li>
 
</li>
<li> Samba4 socket-library's current TCP support does not send back
 
  +
<li>
'too large' error messages if the high bit is set. This is
 
  +
Another way, which Simo dismisses on Samba4's behalf:
needed for a proposed extension mechanism (SSL-armored kinit,
 
  +
Samba can use
by Leif Johansson <leifj@it.su.se>), but is currently unsupported
 
  +
[http://k5wiki.kerberos.org/wiki/Samba4_Port:_iptables_Remapping '''iptables remapping'''],
in both Heimdal and MIT.
 
  +
but only for kdc packets, so that Samba acts as a router between the AD client and the KDC.
  +
This would work for MIT-krb & for Heimdal.
 
</li>
 
</li>
</ol>
 
  +
<li> If we do have to build a libkdc interface for MIT's KDC,
==== Key-handling changes ====
 
  +
Samba4 will need the KDC to use
<ol>
 
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#.2A.2A_Samba4.27s_portable_socket_API '''Samba's socket library''']
<li> Samba4 app-server-host holds a [[UTF-16 PW]], plus a key bitstring;
 
  +
correctly.
See Appendix 1, "Keytab Requirements."
 
</li>
 
<li> In-memory-only credentials cache for forwarded tickets
 
Samba4 extracts forwarded tickets from the GSSAPI layer,
 
and puts them into the memory-based credentials cache.
 
We can then use them for proxy work. This needs to be
 
ported, if the MIT library doesn't do it yet.
 
</li>
 
<li> [[In-memory-only keytab]] (nice to have):
 
Heimdal used to offer "in-memory keytabs" for servers that use
 
passwords. These server-side passwords were held in a Samba LDB
 
database called secrets.ldb . The heimdal library would fetch
 
the server's password from the ldb file and would construct an
 
in-memory keytab struct containing the password, somewhat as if
 
the library had read an MIT-style keytab file. Unfortunately,
 
only later, at recv_auth() time, would the Heimdal library convert
 
the server-PW into a salted-&-hashed AES key, by hashing 10,000
 
times with SHA-1. Naturally, this is really too slow for recv_auth(),
 
which runs when an app-server authenticates a client's app-service-
 
request. So, nowadays, this password-based in-memory keytab is
 
falling into disuse.
 
</li>
 
</ol>
 
==== Extra krb library functions ====
 
<ol>
 
<li> Special Heimdal-specific functions; These functions didn't
 
exist in the MIT code, years ago, when Samba started.
 
* krb5_free_keyblock_contents()
 
AB will try to build a final list of these functions:
 
</li>
 
<li> Principal-manipulation functions: Samba makes extensive
 
use of the principal manipulation functions in Heimdal,
 
including the known structure behind '''krb5_principal''' and
 
'''krb5_realm''' (a char *). For example,
 
<ol>
 
<li> '''krb5_parse_name_flags'''(smb_krb5_context->krb5_context, name,
 
KRB5_PRINCIPAL_PARSE_REQUIRE_REALM, &principal);
 
</li>
 
<li> '''krb5_unparse_name_flags'''(smb_krb5_context->krb5_context, principal,
 
KRB5_PRINCIPAL_UNPARSE_NO_REALM, &new_princ);
 
</li>
 
<li> '''krb5_principal_get_realm'''()
 
</li>
 
<li> ''' krb5_principal_set_realm'''()
 
</ol>
 
These are needed for juggling the AD
 
variant-structures for server names.
 
(Luke says these are all in MIT's code now.)
 
 
</li>
 
</li>
 
</ol>
 
</ol>
   
==== Error-handling, logging, testing ====
 
  +
==== [[Later: TGS access-control]] ====
  +
MIT krb will need to support these AD features, once Samba4 does.
  +
Alternatively, this could be seen as an opportunity for MIT-based
  +
Samba4 to surpass Heimdal-based Samba.
 
<ol>
 
<ol>
<li> Special [[Short name rules]] check for misconfigured Samba4
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#HBAC_for_the_TGS '''Add HBAC to the TGS'''],
hostnames; Samba is highly likely to be misconfigured, in
 
  +
so that Samba4 can refuse TGTs to kinit,
many weird and interesting ways. So, we have a patch for
 
  +
based on time-of-day & IP-addr constraints;
Heimdal that avoids DNS lookups on names without a "." in
 
  +
<ol>
them. This should avoid some delay and root server load.
 
  +
<li> DTD: This is natural; the TGS should enforce its own
(These errors need to be caught in MIT's library.)
 
  +
access-control, as all other services do.
  +
</li>
  +
<li> TGS-HBAC is part of the rationale for
  +
[http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Data-Abstraction_Layer_.28DAL.29 '''rewriting the DAL'''].
  +
</li>
  +
</ol>
 
</li>
 
</li>
<li> Improved krb error-messages;
 
  +
<li> [http://k5wiki.kerberos.org/wiki/Task-List_for_Samba4_Port_(Andrew_Bartlett)#Failed_PW_lockouts '''Failed-kinit counts''']:
</li>
 
  +
Add a KDC heuristic for tracking intervals between kinits,
<li> Improved Kerberos logging support:
 
  +
so that Samba4 can enforce AD's unified account-lockout on kinit.
<ol>
 
  +
Samba4 already does lockouts for other PW-based authentication methods
<li> '''krb5_log_facility()''': Samba4 now uses this Heimdal function,
 
  +
(NTLM, LDAP simple bind, etc).
which allows us to redirect the warnings and status from
 
the KDC (and client/server Kerberos code) to Samba's DEBUG()
 
system. Samba uses this logging routine optionally in the
 
main code, but it's required for KDC errors.
 
</li>
 
<li> '''krb5_get_error_string'''(): This Heimdal-specific function
 
does a lot to reduce the 'administrator pain' level, by
 
providing specific, English text-string error messages
 
instead of just error code translations. (This isn't
 
necessary for the port, but it's more useful than MIT's
 
default err-handling; Make sure this works for MIT-krb)
 
</li>
 
</ol>
 
</li>
 
<li> MS GSSMonger test-suite: Microsoft has released a krb-specific
 
testsuite called gssmonger, which tests interoperability. We
 
should compile it against lorikeet-heimdal & MIT and see if we
 
can build a 'Samba4' server for it. GSSMonger wasn't intended
 
to be Windows-specific.
 
</li>
 
<li> Testsuite for kpasswd daemon: I have a partial kpasswd server
 
which needs finishing, and Samba4 needs a client testsuite
 
written, either via the krb5 API or directly against GENSEC and
 
the ASN.1 routines. Samba4 likes to test failure-modes, not
 
just successful behavior. Currently Samba4's kpasswd only works
 
for Heimdal, not MIT clients. This may be due to call-ordering
 
constraints.
 
 
</li>
 
</li>
 
</ol>
 
</ol>
Line 629: Line 183:
 
----
 
----
   
=== Appendix 1: Keytab Requirements ===
 
  +
== Samba's use of Heimdal symbols, with MIT differences ==
   
Traditional 'MIT' keytab operation is very different from AD's
 
  +
Samba4 uses around
account-handling for application-servers:
 
  +
[http://k5wiki.kerberos.org/wiki/Samba%27s_use_of_Heimdal_symbols%2C_with_MIT_differences '''265 Heimdal symbols:''']
==== Host PWs vs service-keys ====
 
  +
# 150 functions,
<ol>
 
  +
# 45 structs & typedefs, and
<li> Traditional 'MIT' behaviour is for the app-server to use a keytab
 
  +
# 70 macros & enums.
containing several named random-bitstring service-keys, created
 
by the KDC. An MIT-style keytab holds a different service-key
 
for every kerberized application-service that the server offers
 
to clients. Heimdal also implements this behaviour. MIT's model
 
doesn't use AD's UTF-16 'service password', and no salting is
 
necessary for service-keys, because each service-key is random
 
enough to withstand an exhaustive key-search attack.
 
</li>
 
<li> In the Windows model, the server key's construction is very
 
different: The app-server itself, not the KDC, generates a
 
random UTF-16 pseudo-textual password, and sends this password
 
to the KDC using SAMR, a DCE-RPC "domain-joining" protocol (but
 
for windows 7, see below). Then, the KDC shares this server-
 
password with every application service on the whole machine.
 
</li>
 
<li> Only when the app-server uses kerberos does the password get
 
salted by the member server (ie, an AD server-host). (That
 
is, no salt information appears to be conveyed from the AD KDC
 
to the member server, and the member server must use the rules
 
described in Luke's mail, in Appendix 3, below). The salted-
 
and-hashed version of the server-host's PW gets stored in the
 
server-host's keytab.
 
</li>
 
<li> Samba file-servers can have many server-names simultaneously
 
(kind of like web servers' software-virtual-hosting), but since
 
these servers are running in AD, these names can be set up to
 
all share the same secret key. In AD, co-located server names
 
almost always share a secret key like this. In samba3, this
 
key-sharing was optional, so some samba3 hosts' keytabs did
 
hold multiple keys. Samba4 abandons this traditional "old MIT"
 
style of keytab, and only supports one key per keytab, and
 
multiple server-names can use that keytab key in common. In
 
dealing with this model, Samba4 uses both the traditional file
 
keytab and an in-MEMORY keytabs.
 
</li>
 
<li> Pre-Windows7 AD and samba3/4 both use SAMR, an older protocol,
 
to jumpstart the member server's PW-sharing with AD (the "windows
 
domain-join process"). This PW-sharing transfers only the PW's
 
UTF-16 text, without any salting or hashing, so that non-krb
 
security mechanisms can use the same utf-16 text PW. For
 
Windows 7, this domain-joining uses LDAP for PW-setting.
 
</li>
 
</ol>
 
==== Flexible server-naming ====
 
   
The other big difference between AD's keytabs and MIT's is that
 
  +
Of these, roughly half present problems for the port:
Windows offers a lot more flexibility about service-principals'
 
  +
# 25 symbols have different definitions in the MIT & Heimdal trees.
names. When the kerberos server-side library receives Windows-style
 
  +
# 110 symbols are missing from MIT's krb5 tree.
tickets from an app-client, MIT's krb library (or GSSAPI) must
 
accommodate Windows' flexibility about case-sensitivity and
 
canonicalization.
 
This means that an incoming application-request to a member server
 
may use a wide variety of service-principal names. These include:
 
<pre>
 
machine$@REALM (samba clients)
 
HOST/foo.bar@realm (win2k clients)
 
cifs/foo.bar@realm (winxp clients)
 
HOST/foo@realm (win2k clients, using netbios)
 
cifs/foo@realm (winxp clients, using netbios),
 
</pre>
 
as well as all upper/lower-case variations on the above.
 
</ol>
 
==== Keytabs & Name-canonicalization ====
 
<ol>
 
<li> Heimdal's GSSAPI expects to to be called with a
 
principal-name & a keytab, possibly containing
 
multiple principals' different keys. However,
 
AD has a different problem to solve, which is
 
that the client may know the member-server by a
 
non-canonicalized principal name, yet AD knows
 
the keytab contains exactly one key, indexed by
 
the canonical name. So, GSSAPI is unprepared
 
to canonicalize the server-name that the cliet
 
requested, and is also overprepared to do an
 
unnecessary search through the keytab by
 
principal-name. So Samba's server-side GSSAPI
 
calls have to "game" the GSSAPI, by supplying
 
the server's known canonical name, with the one-key
 
keytab. This doesn't really affect IPA's port of
 
Samba4 to MIT-krb.
 
</li>
 
<li> Because the number of U/L case combinations got
 
'too hard' to put into a keytab in the traditional
 
way (with the client to specify the name), we
 
either pre-compute the keys into a traditional
 
keytab or make an in-MEMORY keytab at run time.
 
In both cases we specifiy the principal name to
 
GSSAPI, which avoids the need to store duplicate
 
principals.
 
</li>
 
<li> We use a 'private' keytab in our private dir,
 
referenced from the secrets.ldb by default.
 
</li>
 
</ol>
 
   
 
----
 
----
   
=== Appendix 2: windc KDC Plugin for Account-AuthZ ===
 
  +
== Samba4 Interfaces with Heimdal ==
   
Here is how Samba4 ended up doing account-authorization in
 
  +
<ol>
Heimdal, via a specialized KDC plugin. This '''windc'''
 
  +
<li> Samba4's
plugin helps bridge an important gap: The user's AD
 
  +
[http://k5wiki.kerberos.org/wiki/Samba4_Port:_hdb_%26_ldb_Interfaces '''Database Interfaces''']
record is much richer than the Heimdal HDB format allows,
 
  +
enable Heimdal to use Samba4's directory data,
so we do AD-specific access-control checks in the plugin's
 
  +
whether the directory is stored in LDAP or in local disk files.
AD-specific layer, not in the DB-agnostic KDC server:
 
<ol>
 
<li> We created a separate KDC plugin, with this API:
 
<pre>
 
typedef struct
 
hdb_entry_ex { void *ctx;
 
hdb_entry entry;
 
void (*free_entry)(krb5_context, struct hdb_entry_ex *);
 
} hdb_entry_ex;
 
</pre>
 
The void *ctx is a "private pointer," provided by the
 
'get' method's hdb_entry_ex retval. The APIs below use
 
the void *ctx so as to find additional information about
 
the user, not contained in the hdb_entry structure.
 
Both the provider and the APIs below understand how to
 
cast the private void *ctx pointer.
 
<pre>
 
typedef krb5_error_code
 
(*krb5plugin_windc_pac_generate)(void * krb5_context,
 
struct hdb_entry_ex *,
 
krb5_pac*);
 
typedef krb5_error_code
 
(*krb5plugin_windc_pac_verify)(void * krb5_context,
 
const krb5_principal,
 
struct hdb_entry_ex *,
 
struct hdb_entry_ex *,
 
krb5_pac *);
 
typedef krb5_error_code
 
(*krb5plugin_windc_client_access)(void * krb5_context,
 
struct hdb_entry_ex *,
 
KDC_REQ *,
 
krb5_data *);
 
</pre>
 
The krb5_data* here is critical, so that samba's KDC can return
 
the right NTSTATUS code in the 'error string' returned to the
 
client. Otherwise, the windows client won't get the right error
 
message to the user (such as 'password expired' etc). The pure
 
Kerberos error is not enough.
 
<pre>
 
typedef struct
 
krb5plugin_windc_ftable { int minor_version;
 
krb5_error_code (*init)(krb5_context, void **);
 
void (*fini)(void *);
 
krb5plugin_windc_pac_generate pac_generate;
 
krb5plugin_windc_pac_verify pac_verify;
 
krb5plugin_windc_client_access client_access;
 
} krb5plugin_windc_ftable;
 
</pre>
 
This API has some Heimdal-specific stuff, that'll
 
have to change when we port this KDC plugin to MIT krb.
 
 
</li>
 
</li>
<li> '''pac_generate()''' creates an initial PAC from the user's AD record.
 
  +
<li> Heimdal's
</li>
 
  +
[http://k5wiki.kerberos.org/wiki/Samba4_port:_libkdc_Interface '''libkdc Interface''']
<li> '''pac_verify()''' checks that a AC is correctly signed,
 
  +
gives Samba4 a direct subroutine interface to the Heimdal KDC,
adds additional groups (for cross-realm tickets)
 
  +
with the KDC running as part of the Samba4 process.
and re-signs with the key of the target kerberos
 
service's account
 
</li>
 
<li> '''client_access()''' performs additional access checks, such as
 
allowedWorkstations and account expiry.
 
</li>
 
<li> For example, to register this plugin, use the kdc's standard
 
plugin-system at Samba4's initialisation:
 
<pre>
 
/* first, setup the table of callback pointers */
 
/* Registar WinDC hooks */
 
ret = krb5_plugin_register(krb5_context, PLUGIN_TYPE_DATA,
 
"windc", &windc_plugin_table);
 
/* once registered, the KDC will invoke the callbacks */
 
/* while preparing each new ticket (TGT or app-tkt) */
 
</pre>
 
</li>
 
<li> An alternative way to register the plugin is with a
 
config-file that names a DSO (Dynamically Shared Object).
 
 
</li>
 
</li>
  +
</ol>
  +
 
----
 
----
 
=== Appendix 3: Samba4 stuff that doesn't need to get ported. ===
 
<pre>
 
Heimdal oddities
 
* Heimdal is built such that it should be able to serve multiple realms
 
at the same time. This isn't relevant for Samba's use, but it shows
 
up in a lot of generalisations throughout the code.
 
* Samba4's code originally tried internally to make it possible to use
 
Heimdal's multi-realms-per-KDC ability, but this was ill-conceived,
 
and AB has recently (6/09) ripped the last of that multi-realms
 
stuff out of samba4. AB says that in AD, it's not really possible
 
to make this work; several AD components structurally assume that
 
there's one realm per KDC. However, we do use this to support
 
canonicalization of realm-names: case variations, plus long-vs-short
 
variants of realm-names. No MIT porting task here, as long as MIT kdc
 
doesn't refuse to do some LDAP lookups (eg, alias' realm-name looks
 
wrong).
 
* Heimdal supports multiple passwords on a client account: Samba4
 
seems to call hdb_next_enctype2key() in the pre-authentication
 
routines, to allow multiple passwords per account in krb5.
 
(I think this was intended to allow multiple salts). AD doesn't
 
support this, so the MIT port shouldn't bother with this.
 
Not needed anymore, because MIT's code now handles PACs fully:
 
* gss_krb5_copy_service_keyblock() (get the key used to actually
 
encrypt the ticket to the server, because the same key is used for
 
the PAC validation).
 
* gsskrb5_extract_authtime_from_sec_context (get authtime from
 
kerberos ticket)
 
* gsskrb5_extract_authz_data_from_sec_context (get authdata from
 
ticket, ie the PAC. Must unwrap the data if in an AD-IFRELEVANT)]
 
Authz data extraction
 
* We use krb5_ticket_get_authorization_data_type(), and expect
 
it to return the correct authz data, even if wrapped in an
 
AD-IFRELEVANT container. This doesn't need to be ported to MIT.
 
This should be obsoleted by MIT's new PAC code.
 
libkdc
 
* Samba4 needs to be built as a single binary (design requirement),
 
and this should include the KDC. Samba also (and perhaps more
 
importantly) needs to control the configuration environment of
 
the KDC.
 
* But, libkdc doesn't matter for IPA; Samba invokes the Heimdal kdc
 
as a library call, but this is just a convenience, and the MIT
 
port can do otherwise w/o trouble.)
 
Returned Salt for PreAuthentication
 
When the AD-KDC replies to pre-authentication, it returns the
 
salt, which may be in the form of a principalName that is in no
 
way connected with the current names. (ie, even if the
 
userPrincipalName and samAccountName are renamed, the old salt
 
is returned).
 
This is the kerberos standard salt, kept in the 'Key'. The
 
AD generation rules are found in a Mail from Luke Howard dated
 
10 Nov 2004. The MIT glue layer doesn't really need to care about
 
these salt-handling details; the samba4 code & the LDAP backend
 
will conspire to make sure that MIT's KDC gets correct salts.
 
</pre>
 
 
<pre>
 
>
 
> From: Luke Howard <lukeh@padl.com>
 
> Organization: PADL Software Pty Ltd
 
> To: lukeh@padl.com
 
> Date: Wed, 10 Nov 2004 13:31:21 +1100
 
> Cc: huaraz@moeller.plus.com, samba-technical@lists.samba.org
 
> Subject: Re: Samba-3.0.7-1.3E Active Directory Issues
 
> -------
 
>
 
> Did some more testing, it appears the behaviour has another
 
> explanation. It appears that the standard Kerberos password salt
 
> algorithm is applied in Windows 2003, just that the source principal
 
> name is different.
 
>
 
> Here is what I've been able to deduce from creating a bunch of
 
> different accounts:
 
> [SAM name in this mail means the AD attribute samAccountName .
 
> E.g., jbob for a user and jbcomputer$ for a computer.]
 
>
 
> [UPN is the AD userPrincipalName attribute. For example, jbob@mydomain.com]
 
> Type of account Principal for Salting
 
> ========================================================================
 
> Computer Account host/<SAM-Name-Without-$>.realm@REALM
 
> User Account Without UPN <SAM-Name>@REALM
 
> User Account With UPN <LHS-Of-UPN>@REALM
 
>
 
> Note that if the computer account's SAM account name does not include
 
> the trailing '$', then the entire SAM account name is used as input to
 
> the salting principal. Setting a UPN for a computer account has no
 
> effect.
 
>
 
> It seems to me odd that the RHS of the UPN is not used in the salting
 
> principal. For example, a user with UPN foo@mydomain.com in the realm
 
> MYREALM.COM would have a salt of MYREALM.COMfoo. Perhaps this is to
 
> allow a user's UPN suffix to be changed without changing the salt. And
 
> perhaps using the UPN for salting signifies a move away SAM names and
 
> their associated constraints.
 
>
 
> For more information on how UPNs relate to the Kerberos protocol,
 
> see:
 
>
 
> http://www.ietf.org/proceedings/01dec/I-D/draft-ietf-krb-wg-kerberos-referrals-02.txt
 
>
 
> -- Luke
 
</pre>
 

Latest revision as of 08:40, 18 September 2009

This is an early stage project for MIT Kerberos. It is being fleshed out by its proponents. Feel free to help flesh out the details of this project. After the project is ready, it will be presented for review and approval.


Introduction

Samba4 aims to provide a complete OSS replacement for Active Directory. Samba4, like earlier versions of Samba, uses Heimdal Kerberos. The Samba4 Port project proposes to enable Samba4 to use MIT kerberos as an alternative. The near-term goal is that mixed krb5+AD deployments could use Samba4 to provide better interoperation between AD realms and MIT-krb5 realms.

Use case: For example, suppose a kerberos customer is deploying a network with mixed operating systems using kerberos and would want to use one KDC for all of them. In this case, a single MIT Kerberos deployment should be able to support both traditonal UNIX clients and servers, intermixed with Windows clients and Samba servers:

  1. The Windows clients should be able to use the MIT KDC(s) as AD servers, so as to authenticate themselves to Samba file-servers and to Windows servers;
  2. A Windows client's tickets will carry PACs, as usual for AD;
  3. The UNIX clients should be able to access the KDC as a traditional non-AD-style KDC, so as to access UNIX services securely;
  4. A UNIX client's ticket will not carry a PAC, except when the UNIX client accesses a Windows server (Rationale) .

The Samba4 team, the MIT Krb Consortium, RedHat, Ubuntu, and Sun all have shown some interest in this Samba4 Port project. Here is a table showing which OS platforms are supported by Samba4, Heimdal, and MIT kerberos. Summary: MIT-krb5 & Samba4 both run on Mac OS X, NetBSD, Debian, RedHat, Ubuntu, & Solaris.


Concise to-do list

This is a condensed version of the task-list offered by Samba4's Andrew Bartlett, containing only what hasn't yet been done already by MIT.

The two big chunks of work are LDAP Driver and Replacing / improving MIT's DAL, but the DAL work may not be needed.

Replace the MIT KDC's LDAP driver

Samba4's LDAP driver for the MIT KDB needs to know how to do AD's intricate naming:

  1. Canonicalization of server names, user-names, and realm names. MIT 1.7 already supports canonicalization.
  2. AD-style aliases for HOST/ service names.
  3. Implicit names for Win2k accounts.
  4. Principal "types": client / server / krbtgs
  5. Flexible server-naming
  6. Keytabs & name-canonicalization

Most or all of Heimdal's LDAP driver code is in three Samba4 source files, ~1000 lines in all.


Small changes

Of the things on this list, only NTLM support (bullet 2) is needed for the Samba4 KDC port. The other tasks are all application-library stuff, and arguably aren't needed at all, because Samba3 already works well with MIT application libraries.

  1. MIT library changes
  2. Samba4/AD libraries: NTLM support. See also this Sept-2009 NTLM thread (this implies to me that a GSS NTLM mech is not an immediate requirement - LH)
  3. Key-handling changes]
  4. Extra Krb library functions
  5. Error-handling, logging, testing

Use 1.7's AD-support features

This stuff should already just work:

  1. PAC handling;
  2. AD-style name canonicalization;
  3. NT-ENTERPRISE names, which carry two realm-suffixes;
  4. CHECK_POLICY/AUDIT methods (needed for TGS access-control);
  5. DCE_STYLE Challenge/Response handshakes: see Krb lib & GSSAPI.
  6. Accept legacy Samba3 clients' bad GSSAPI checksums;
  7. Principal-manipulation functions;
  8. State-machine safety;

Controversial proposed changes for the port

Maybe: Improve or replace MIT's DAL

Rewrite the MIT KDC's Data-Abstraction Layer (DAL), mostly because the MIT KDC needs to see & manipulate more LDAP detail, on Samba4's behalf;

Maybe, or not: Add a KDC-as-library API

Samba4 currently runs as a single process, and Samba4's smbd invokes the Heimdal KDC via a libkdc interface (KDC as library).

  1. Rationale:
    1. smbd uses the libkdc interface to configure the KDC, both at startup & during runtime.
    2. Samba4's build/test environment uses libkdc's socket-passing, to simulate network traffic.
  2. Andrew Bartlett says this libkdc interface is "nice to have", but not essential for getting the port to work.
  3. Tom Yu says adding a libkdc interface to MIT's code would be a lot of work, but would tie naturally into code-cleanup work that MIT wants to do, anyway.
  4. Sam Hartman says he needs the libkdc interface, too, for his work on PK-U2U (but not immediately).
  5. Another way, which Simo dismisses on Samba4's behalf: Samba can use iptables remapping, but only for kdc packets, so that Samba acts as a router between the AD client and the KDC. This would work for MIT-krb & for Heimdal.
  6. If we do have to build a libkdc interface for MIT's KDC, Samba4 will need the KDC to use Samba's socket library correctly.

Later: TGS access-control

MIT krb will need to support these AD features, once Samba4 does. Alternatively, this could be seen as an opportunity for MIT-based Samba4 to surpass Heimdal-based Samba.

  1. Add HBAC to the TGS, so that Samba4 can refuse TGTs to kinit, based on time-of-day & IP-addr constraints;
    1. DTD: This is natural; the TGS should enforce its own access-control, as all other services do.
    2. TGS-HBAC is part of the rationale for rewriting the DAL.
  2. Failed-kinit counts: Add a KDC heuristic for tracking intervals between kinits, so that Samba4 can enforce AD's unified account-lockout on kinit. Samba4 already does lockouts for other PW-based authentication methods (NTLM, LDAP simple bind, etc).

Samba's use of Heimdal symbols, with MIT differences

Samba4 uses around 265 Heimdal symbols:

  1. 150 functions,
  2. 45 structs & typedefs, and
  3. 70 macros & enums.

Of these, roughly half present problems for the port:

  1. 25 symbols have different definitions in the MIT & Heimdal trees.
  2. 110 symbols are missing from MIT's krb5 tree.

Samba4 Interfaces with Heimdal

  1. Samba4's Database Interfaces enable Heimdal to use Samba4's directory data, whether the directory is stored in LDAP or in local disk files.
  2. Heimdal's libkdc Interface gives Samba4 a direct subroutine interface to the Heimdal KDC, with the KDC running as part of the Samba4 process.