Functional changes:
(1) do not restrict ourselves to DES-CBC-CRC instead support any
ticket with an enctype we support. as of this date (rev 1.3)
this includes all but RC4-MD4.
(2) do not accept invalid tickets
(3) when attempting to retrieve tickets do not specify either the
enctype or cache options (if possible). doing so will force a
TGS request and prevent the results from being stored into the
cache.
(4) when the LSA cache contains a TGT which has expired Microsoft will
not perform a new TGS request until the cache has been purged.
Instead the expired ticket continues to be used along with its
embedded authorization data. When PURGE_ENABLED is defined, if the
tickets are expired, the cache will be purged before requesting
new tickets, else we ignore the contents of the cache and force
a new TGS request.
(5) when the LSA cache is empty do not abort. On XP or 2003, use
the SecurityLogonSessionData to determine the Realm (UserDnsDomain
in MS-speak) and request an appropriate TGT. On 2000, check the
Registry for the HKCU\"Volatile Environment":"USERDNSDOMAIN"
instead. This will allow ms2mit to be used to repopulate the
LSA cache. If the current session is not Kerberos authenticated
an appropriate error message will be generated.
Code changes:
(1) several memory leaks plugged
(2) several support functions copied from the Leashw32.dll sources
(3) get_STRING_from_registry() uses the ANSI versions of the Registry
functions and should at a later date be converted to use the
Unicode versions.
Notes: an ms2mit.exe based on the Leash_import() function
should be considered. Leash_import() not only imports the TGT from
the LSA but also performs the krb524 conversion and AFS token retrieval.
Of course, that version of ms2mit.exe could not exist within the krb5
source tree.
Tom Yu [Thu, 17 Jul 2003 20:50:36 +0000 (20:50 +0000)]
Remove kg_release_defcred and caching of default credential. Rewrite
krb5_gss_init_sec_context() while we're at it to make defcred-related
changes easier, and as a side effect, fix some error condition memory
leaks.
RealmsConfig-glue.c: Don't fail when krb5.conf is valid and krb.conf isn't. Also, don't assert v4 realm is in profile unless that realm is a valid v4 realm
Rewrote krb_get_lrealm to correctly handle the v4 realms section in the profile and to return KFAILURE in the absence of any config (instead of returning ATHENA.MIT.EDU)
Tom Yu [Thu, 10 Jul 2003 22:53:02 +0000 (22:53 +0000)]
* acquire_cred.c (acquire_init_cred): Close the ccache if
krb5_cc_set_flags() fails, as krb5int_cc_default succeeds even if
the file is not there, but krb5_cc_set_flags will fail in turning
off OPENCLOSE mode if the file can't be opened. Thanks to Kent Wu.
krb_prof_get_nth() no longer assumes that its retlen argument is correct (call strcpy instead of strncpy) because this argument is a guess for some callers (eg: krb_get_admhst())
Ken Raeburn [Fri, 4 Jul 2003 20:49:06 +0000 (20:49 +0000)]
darwin build can't compile util/et test program
The autoconf tests correctly determine that sys_nerr exists, and that
no declaration is needed. The test_et.c code checks the wrong condition,
though, and the system header declaration of sys_nerr conflicts with the
one in that source file. (The native one is const.)
* test_et.c: Conditionalize sys_nerr declaration on NEED_SYS_ERRLIST, not
HAVE_SYS_ERRLIST.
Ken Raeburn [Fri, 27 Jun 2003 20:36:24 +0000 (20:36 +0000)]
This change fixes the problem in the ss directory. It probably won't
get into our 1.3 release, but perhaps in a patch release afterwards.
I'll look and see where else things are getting needlessly rebuilt....
* Makefile.in (HDRS): Remove mit-sipb-copyright.h.
(includes): Depend on copied version of the headers, not local versions. Don't
do any copying; instead, move the copying commands to new targets for each
header.
($(HDRDIR)/timestamp): New target; create the directory here if needed.
(clean-unix): Remove the timestamp file.
Tom Yu [Wed, 25 Jun 2003 21:10:48 +0000 (21:10 +0000)]
clean up ftp client gss memory management
* ftp.c (do_auth): Call gss_release_buffer() on send_tok
immediately after sending it. Call gss_release_name() on
target_name at outer_loop instead of inside the loop.
Tom Yu [Tue, 24 Jun 2003 02:29:47 +0000 (02:29 +0000)]
apply patch from jaltman:
* ms2mit.c: Windows Credentials are addressless. Do not store the
credentials in the MIT cache with addresses since they do not
contain addresses in the encrypted portion of the credential.
Instead generate a valid empty address list.
Tom Yu [Fri, 20 Jun 2003 03:24:07 +0000 (03:24 +0000)]
Delete references to --with-system-db from README
* build.texinfo (HPUX): Fix typo.
(Options to Configure): Note that --with-system-db is unsupported,
concerning possible lossage with loading dumpfiles.
Tom Yu [Tue, 17 Jun 2003 01:00:45 +0000 (01:00 +0000)]
Update for krb5-1.3-beta4. Fix note on [999]. Move notes re
addressless tickets and NAT-friendliness to "major changes". Still
need to fill out the TODO for IPv6.
Ken Raeburn [Mon, 16 Jun 2003 19:02:01 +0000 (19:02 +0000)]
* ftp.c (recvrequest): Add new argument indicating whether "-" and "|..."
special treatment should be disabled.
* ftp_var.h (recvrequest): Update declaration.
* cmds.c (remglob, ls, mls): Pass 0 as the extra argument.
(mget): Pass 1.
(getit): Pass 1 iff only one filename was supplied.
Tom Yu [Fri, 13 Jun 2003 22:30:59 +0000 (22:30 +0000)]
* server_kdb.c (kdb_init_hist): Force history principal's key to
be of the same enctype as the master key, as searches for it later
on explicitly specify the enctype.
Tom Yu [Fri, 13 Jun 2003 21:43:07 +0000 (21:43 +0000)]
libgss leaks, UMRs
* init_sec_context.c (krb5_gss_init_sec_context): Free
default_enctypes to avoid leaking returned value from
krb5_get_tgs_ktypes.
* k5unseal.c (kg_unseal_v1): Explicitly set token.value to NULL if
token.length == 0, to avoid spurious uninitialized memory
references when calling memcpy() with a zero length.
ticket: new
target_version: 1.3
tags: pullup
component: krb5-libs
cc: Kent_Wu@trendmicro.com
* error_table.h, et_c.awk, et_c.pl, et_h.awk, et_c.awk: Removed Mac OS support because it prevents darwin builds from getting com error strings via the initialize_*_error_table function
Ken Raeburn [Sat, 7 Jun 2003 00:45:41 +0000 (00:45 +0000)]
Checkpoint some working code.
A better solution is in the works, but may or may not make the deadline for
the next beta...
* RealmsConfig-glue.c: Include k5-int.h.
(dnscache): New variable.
(DNS_CACHE_TIMEOUT): New macro.
(krb_get_krbhst) [KRB5_DNS_LOOKUP]: If no krb.conf info is found, try DNS SRV
records for "kerberos-iv". Cache results in case they're immediately requested
again.
Ken Raeburn [Fri, 6 Jun 2003 23:07:55 +0000 (23:07 +0000)]
* locate_kdc.c (struct srv_dns_entry): Moved to k5-int.h.
(krb5int_make_srv_query_realm): Renamed from make_srv_query_realm.
(krb5int_free_srv_dns_data): New function.
(krb5_locate_srv_dns_1): Use it.
* accessor.c (krb5int_accessor): Fill in make_srv_query_realm and
free_srv_dns_data fields.
Tom Yu [Fri, 6 Jun 2003 14:44:33 +0000 (14:44 +0000)]
* g_cnffile.c (krb__get_srvtabname): Make retname be a static
array rather than a static pointer, to avoid callers' possible
retention of free()d pointers. Yes, this may cause difficulty
with making this function thread-safe.
Tom Yu [Thu, 5 Jun 2003 21:39:59 +0000 (21:39 +0000)]
document team procedures
* procedures.txt: New file. Draft of current procedures.
Currently mostly contains a few email messages from Sam. More
clarification should follow later.
Ken Raeburn [Thu, 5 Jun 2003 20:05:37 +0000 (20:05 +0000)]
* locate_kdc.c (make_srv_query_realm): Punt if strdup fails. Always return
what data we can, even if memory allocation or other problems prevent us from
returning more.
(krb5_locate_srv_dns_1): Always return what data we can. Fix memory leak.
Free up temporary storage as quickly as possible, while building up address
list to return.
Ken Raeburn [Wed, 4 Jun 2003 02:57:11 +0000 (02:57 +0000)]
Handle SRV RR target of "." by returning a new error code.
May have memory leaks -- needs checking before pullup.
* locate_kdc.c (struct srv_dns_entry): Move to top level.
(make_srv_query_realm): Separate from krb5_locate_srv_dns_1; just do query and
return results.
(krb5_locate_srv_dns_1): Call it, and build addlist entries. Check for one RR
with a target of ".", and return an error.
(krb5_locate_srv_dns): Deleted.
Tom Yu [Tue, 3 Jun 2003 03:18:41 +0000 (03:18 +0000)]
* change_password.c (krb_change_password): Explicitly zero the
session key. Zero the key derived from the new password.
* mk_req.c (krb_mk_req): Explicitly zero the session key.
(krb_mk_req_creds_prealm): Don't zero the session key, in case the
caller wants to make use of it.
ticket: 1546
status: open
target_version: 1.3
tags: pullup
Ken Raeburn [Sun, 1 Jun 2003 20:24:02 +0000 (20:24 +0000)]
* default.exp: Default RLOGIN_FLAGS to "-x".
(start_kerberos_daemons): Watch for "Cannot bind server socket" and log it.
Watch for "no sockets set up" and report an error.
(setup_root_shell): Watch for "Cannot assign requested address", log it and
give up.
Ken Raeburn [Sat, 31 May 2003 03:33:22 +0000 (03:33 +0000)]
* krb5.conf: Delete Athena KDC specifications. Delete Cygnus realm info.
Replace CLUB.CC.CMU.EDU info with ANDREW.CMU.EDU, which has SRV records and
thus doesn't need KDC specs. Provide a commented-out example of a [logging]
spec.
projects / thirdparty / krb5.git / log
