DCE DCEECO1021 DCE V2.1 for DIGITAL UNIX ECO Summary

TITLE: DCE DCEECO1021 DCE V2.1 for DIGITAL UNIX ECO Summary
 
Modification Date:  20-NOV-98
Modification Type:  New Kit

Copyright (c) Compaq Computer Corporation 1998.  All rights reserved.

PRODUCT:     Distributed Computing Environment (DCE) for DIGITAL UNIX

COMPONENTS:  RPC
             Security
             Kerberos 5 Support
             SIA
             Distributed Time Service (DTS)
             CDS
             DCED
             DFS

OP/SYS:      DIGITAL UNIX [R]

SOURCE:      Compaq Computer Corporation

ECO INFORMATION:

     ECO Kit Name:  DCEECO1021
     ECO Kits Superseded by This ECO Kit:  None
     ECO Kit Approximate Size:  67340 Blocks
                                34478080 Bytes

     Kit Applies To:  DCE V2.1 on DIGITAL UNIX V4.0 - V4.0E
     System/Cluster Reboot Necessary:  No
     Rolling Re-boot Supported:  Information Not Available
     Installation Rating:  INSTALL_UNKNOWN

     Kit Dependencies:

       The following remedial kit(s) must be installed BEFORE
       installation of this kit:

         None 

       In order to receive all the corrections listed in this
       kit, the following remedial kits should also be installed:

         None 


ECO KIT SUMMARY:

An ECO kit exists for DCE V2.1 on DIGITAL UNIX V4.0 through V4.0E.  This
kit addresses the following problems: 

Corrections to RPC:

  o  Previously, the timer event was being placed on the queue with a
     stale timestamp. The problem was fixed by making a call to
     rpc__clock_update() in rpc__timer_set_int() to get an accurate
     timestamp for the event trigger. In addition, a test was removed
     that determined whether to signal the timer loop in
     rpc__timer_set_int() when the timer queue was empty. 

  o  A correction was incorporated to allow the use of cluster service
     addresses. 

  o  This release adds a check_unsupported_ifs() function to check for
     interfaces in RPC_UNSUPPORTED_NETIFS that should be avoided when
     initializing DLI. The ifs are stored in the static variable
     unsupported_if_list, and the number of unsupported ifs is stored in
     the static variable int num_unsupported_ifs. A correction was
     incorporated to allow the use of cluster service addresses. 

  o  An array indexing problem in rpc_object_reference::fast_client_ping()
     has been fixed. Under certain circumstances, the problem caused
     memory corruptions in seemingly unrelated areas. Zeroing was also
     added to enhance rpc_object_reference::init_client_ping_list(). 

  o  RPC runtime corrections made in this release allow an RPC
     application to reject an unsupported authentication service.  Prior
     to this fix, a request for an unsupported authentication service
     caused a core dump of the server. 

Corrections to Security:

  o  A leak that affected secd was fixed by changes to dce_aud_commit.

  o  The component of the security server that handles invalid logins was
     built using the assumption that pointers are 4 bytes long, which is
     not true for Alpha systems. As a result, secd crashed on Alpha
     systems when invalid login handling was enabled for a principal.
     This problem was corrected. 

  o  The secd lock manager was enhanced to ensure fair access for both
     readers and writers on a heavily used system. The changes ensure
     that a succession of readers does not indefinitely delay writers. 

  o  Several memory leaks were fixed in the processing of DCE third-party
     preauthentication data by a DCE security server. Prior to this fix,
     secd leaked memory with every successful authentication. 

Corrections to Kerberos 5 Support:

  o  A problem was fixed that was causing credentials acquisition during
     an intercell dfs operation to take too long. Previously, in the
     function krb5_get_cred_from_kdc, a pointer array was not initialized
     correctly. 

  o  A problem was fixed that had blocked the acquisition of a valid
     ticket from the KDC if an expired matching ticket was present in the
     cache. On the client, the DCE Kerberos ticket acquisition code
     checks a cache of previously acquired tickets before requesting a
     ticket from the KDC. If a matching ticket is found in the cache, it
     is reused, and the KDC is not contacted. The matching process
     previously failed to check whether a matching ticket in the cache
     had expired. The presence of an expired matching ticket in the cache
     could thus prevent the client from consulting the KDC for a valid,
     non-expired ticket. 

Corrections to SIA:

  o  A problem was corrected that caused the file matrix.conf to become
     corrupted upon DCE reboot. In matrix.conf, the path was not
     specified for the libsecurity.so entry. The full path is now
     specified as: /usr/shlib/libsecurity.so. 

  o  When a DCE group contains many members, a call to the getgrent
     routine no longer results in a core dump of the calling program (for
     example, ls -l). 

  o  When DCE SIA is enabled, mailx no longer dumps core.

  o  Users will no longer experience inordinate delays at login when DCE
     SIA is enabled. Previously, such delays occurred whenever the DCE
     Registry contained many groups. The delays were caused by making one
     remote procedure call per group to the security server to compute a
     user's group memberships. The same information is now obtained by a
     single remote procedure call to the security server. 

  o  When DCE SIA is enabled, the login program performs a DCE
     authentication. If the authentication succeeds, the environment
     variable KRB5CCNAME should be set so that programs running within
     the login environment can inherit the authenticated user's DCE
     credentials. Previously, the value of KRB5CCNAME was not preserved
     in the login environment. A fix in this release ensures that
     KRB5CCNAME is preserved. 

  o  Previously, when DCE SIA was enabled, a user with an entry in the
     passwd_override file was incorrectly prevented from logging in to
     the local system (in addition to being correctly denied DCE
     credentials). This problem has been fixed. 

  o  The DCE SIA group information server (proxied by dced) now employs
     the group override facility to localize group information obtained
     from the DCE registry. Prior to this fix, group overrides were not
     considered when a user's group memberships were returned from the
     registry. As a result of this change, the group override facility
     can be used to constrain or modify a user's DCE group memberships to
     satisfy local machine security policy. 

     Group override processing by the DCE SIA group information server
     had been inadvertently disabled as the result of a patch distributed
     to speed up integrated logins where the registry contains many
     groups. The fix in this ECO preserves the performance improvements
     provided by the patch while restoring and improving group
     processing. Group password and gid overrides are now handled
     correctly. 

Corrections to Distributed Time Service:

  o  Minor changes were applied to the following files to improve memory
     cleanup: time/service/mgtrpc.c
              time/service/time_request.c
              time/service/transport_rpc.c
              time/service/dtss_service_main.c
              time/service_dtss_service_global_set.c
              time/service/dtss_service_state.c

  o  Previously, a few DTS functions handled NULL parameters incorrectly.
     The problem was corrected. 

  o  Several changes were made to ParseTime to comply with Year 2000
     requirements and to allow for correct leap year calculation in the
     Year 2000. 

Corrections to CDS:

  o  A problem was corrected in the dcecp directory synchronize command
     that was causing directory synchronization to fail. An error was
     found in the syntax used by dcecp when it employed the cdscp set dir
     to new epoch command in the directory synchronization process. 

  o  A change made to CDS allows it to handle arbitrarily large output
     results. The maximum size of the output buffer passed to readentry
     is now reduced by the size of the area reserved for the progress
     record. This change eliminates problems such as limitations on the
     number of member names an nsi group could contain. 

  o  The dcecp clearinghouse repair command was fixed. The command no
     longer returns the error, "Clearinghouse exists but it not
     available," after a successful clearinghouse repair. 

Corrections to DCED:

  o  A change was incorporated to prevent the loss of diagnostic
     information when DCE is restarted. Previously, whenever dced was
     restarted, it recreated the log file used to record its error
     messages. Dced now creates a log file only if one does not exist.
     When a log file exists, dced appends all new output to the existing
     file. 

  o  The dced concurrent lock manager was modified so that it is no
     longer vulnerable to thread cancellation.  Previously, if a dced
     thread was canceled while in the process of acquiring or releasing a
     lock, the lock manager would deadlock on any subsequent lock
     operations, resulting in a hang of the calling thread, and
     ultimately of the dced process. 

     The use of concurrent locks by dced to serialize reading of the
     password and override files was made cancel and exception-safe.
     Prior to this fix, a thread that was canceled or that encountered an
     exception while holding an override lock would neglect to free the
     lock for use by other threads. Once an override lock was lost,
     requests by clients for override service from dced would hang at the
     server (dced), waiting for a lock. This was especially problematic
     with the DCE SIA mechanism configured for integrated login, as it
     could result in an inability to log in to the machine or perform a
     certified DCE authentication. 

Corrections to DFS:

  o  This ECO1 release includes a fix for a dfsbind core dump.

NOTE:  All of the following DFS corrections require rebuilding a kernel
       and rebooting. 

  o  A fix was completed that restricts the range of UDP ports used by
     DFS. Part one of the implementation was included in the Version 2.1
     release: dfsbind reads an environment variable, RPC_RESTRICTED_PORTS,
     and passes the restriction down to the kernel. This ECO1 release
     includes part two of the fix: the kernel allocates ports in
     accordance with the restriction. This fix affects dcedfs.mod. 

  o  This release includes a fix for the premature umask application to
     the mode bits before they are passed to the server. This fix works
     in conjunction with changes incorporated into DIGITAL UNIX V4.0D. It
     affects dcedfs.mod. 

     If you are running a version of DIGITAL UNIX earlier than 4.0D, do
     not attempt to apply this fix. An appropriate patch for versions
     earlier than 4.0D is under development. 

     If you do not need the umask fix, no action is required.

     If you are running V4.0D of DIGITAL UNIX or higher and need the
     umask fix, apply it using the following procedure: 

       1. Enter:

          dbx -k /vmunix
          patch dfs_umask_rawmode_fix_present = 1
          quit

       2. Verify that the change has been made:

          dbx -k /vmunix
          print dfs_umask_rawmode_fix_present
          {THIS SHOULD PRINT 1}
          quit

       3. Reboot.

  o  The Cache Manager now passes through the setuid/setgid mode bits of
     directories without changing the bits. These bits still get turned
     off on regular files unless the client explicitly enables the
     capability to leave the bits unchanged. This capability can be set
     on a fileset with the cm setsetuid command. This change affects
     dcedfs.mod. 

  o  This release includes a change to the token expiration time of
     freely given tokens. The change was made in response to a problem
     that caused clients to hang occasionally for approximately 4
     minutes, and then return a communications failure. The change
     affects dcedfs.mod. 

  o   Because of insufficient locking, it was previously possible for
      requests from the kernel to the dfsbind process to be lost.
      Eventually the kernel would run out of request space and hang. Two
      fixes were made: 

       1.  Sufficient locking was added to prevent the loss of requests.

       2.  The request queue is periodically pruned of old requests. This
           enables the administrator to restart dfsbind only, instead of
           all of DFS, in the event of a problem, and to reclaim the
           resources that the kernel was using. 


INSTALLATION NOTES:

Follow these steps to install the DCE for DIGITAL UNIX V2.1 ECO1 kit. 

  1. Verify a successful installation of DCE for DIGITAL UNIX V2.1 before
     installing the ECO1 kit. 

  2. Untar the ECO1 kit into a local directory, using the following command:

          % tar xvf /DCEECO1021.tar

  3. Use the setld procedure to start the installation procedure:

          %  setld -l ./output

  4. Select the subsets to install from the following choices:

          DCERTSECO1021
          DCEADKECO1021
          DCEDFSECO1021
          DCESECECO1021
          DCECDSECO1021
          DCEDFSBINECO1021

     To install multiple subsets, enter the number of each subset
     followed by a space.  Use a hyphen between numbers to indicate a
     range of subsets. 

     An example of the installation screen is in the ECO kit Cover Letter.

  5. After the installation completes successfully, restart DCE by
     entering the following command:                                   
        
          %  /usr/sbin/dcesetup restart


[R] UNIX is a registered trademark in the United States and other         
    countries, licensed exclusively through X/Open Company Ltd.  

    All other trademarks are the property of their respective owners.

dceeco1021.README
dceeco1021.CHKSUM
dceeco1021.CVRLET_TXT
dceeco1021.tar