Kevin Coffman [Wed, 1 Nov 2006 22:40:30 +0000 (22:40 +0000)]
Modify the preath plugin interface so that a plugin's context is
global to all the modules within a plugin. Also, change the
client-side interface so that the preauth plugin context (once
created) lives the lifetime of a krb5_context. This will allow
future changes that can set plugin parameters. The client side
request context lives the lifetime of a call to krb5_get_init_creds().
Make the sample preauth plugins buildable outside the source tree.
Fix minor memory leak in sort_krb5_padata_sequence().
Add a prototype for krb5_do_preauth_tryagain() and change the plugin
interface.
Incorporates fixes from Nalin Dahyabhai <nalin@redhat.com> for leaks
of the function table pointers (rt #4566) and fix KDC crash (rt #4567)
Will Fiveash [Mon, 30 Oct 2006 20:56:57 +0000 (20:56 +0000)]
latest Novell ldap patches and kdb5_util dump support for ldap
I've applied Novell's latest patches for their LDAP KDB plugin. I've
also implemented and tested support for kdb5_util dump using the LDAP
KDB plugin. I also added a Sun copyright on files that I've modified.
Ken Raeburn [Sat, 28 Oct 2006 03:06:30 +0000 (03:06 +0000)]
don't confuse profile iterator in 425 princ conversion
The profile iterator code hangs onto and uses the list of names passed
in. The krb5_425_conv_principal code reuses that array when the
iterator may still be used.
* conv_princ.c (krb5_425_conv_principal): Use separate name arrays for
the iterator and the v4_realm lookup that may be done inside the
iteration loop.
Will Fiveash [Mon, 23 Oct 2006 21:36:46 +0000 (21:36 +0000)]
fix krb5_ldap_iterate to handle NULL match_expr and open_db_and_mkey to use KRB5_KDB_SRV_TYPE_ADMIN
When I ran kdb5_util dump I had two initial problems. First, the LDAP
plugin was not finding the bind DN because open_db_and_mkey() was
passing KRB5_KDB_SRV_TYPE_OTHER to krb5_db_open(). When I change this
to KRB5_KDB_SRV_TYPE_ADMIN then the ldap_kadmind_dn parameter is used
from krb5.conf and a valid bind DN is found. Second,
krb5_ldap_iterate() will core dump when it is called withy a NULL
match_expr arg. This is how dump_db calls krb5_db_iterate(). I updated
krb5_ldap_iterate() to use a default_match_expr of "*" if match_expr ==
NULL.
Kevin Coffman [Mon, 23 Oct 2006 16:15:50 +0000 (16:15 +0000)]
fix invalid access found by valgrind
Valgrind found that we were reading past the end of the
preferred padata string. p is manually updated within
the loop and there is no need for the increment. It was
causing the null terminator to be skipped over, rather
than properly terminating the loop.
Ezra Peisach [Sun, 22 Oct 2006 11:59:02 +0000 (11:59 +0000)]
osf1 -oldstyle_liblookup typo
Not really relevant anymore - as we do not support static linking now.
But in ticket 927 (r16776) - a test was added to determine if gnu ld
was in use and change the linker flags accordingly. The variable in
aclocal.m4 was krb5_cv_prog_gnu_ld and this was testing for
krb5_cv_gnu_ld.
Sam Hartman [Sat, 21 Oct 2006 20:20:30 +0000 (20:20 +0000)]
Delete src/lib/ccapi.
The ccapi shipped in 1.6 will not be based off this code
and will live in src/ccapi.
It will be copied onto the trunk and branch when ready, but this code is being removed before the branch cut.
Will Fiveash [Sat, 21 Oct 2006 00:33:24 +0000 (00:33 +0000)]
enabling LDAP mix-in support for kdb5_util load
I now have mix-in working for the kdb5_util load. If the krbSubTrees
realm attr contains a base DN where non-krb entries live the
load/krb5_ldap_put_principal() code will modify those entries whose
krbPrincipalName attr matches that of the dump princ record being loaded
otherwise a standalone krbprinc entry will be created under the realm
container.
I also fixed a small bug in krb5_ldap_policydn_to_name() for the version
that uses ldap_explode_dn().
Ken Raeburn [Fri, 13 Oct 2006 23:54:24 +0000 (23:54 +0000)]
Use $(VALGRIND) when running programs using $(KRB5_RUN_ENV) or
$(RUN_SETUP). Replaces old hack with MAYBE_VALGRIND added to RUN_ENV
in a way that would break in some of the tests.
Set VALGRIND in site.exp in tests/dejagnu. (Not used yet.)
Runs some shell scripts under valgrind, rather than changing them to
run only the executables under valgrind; this is mostly okay, just
creates lots of extra log data, and requires --trace-children=yes.
This should work for any instrumentation program invocation that gets
followed immediately by the name and argument list for the program
being instrumented. For example, VALGRIND="env LD_PRELOAD=..." should
work, though I haven't tested it.
Ken Raeburn [Fri, 13 Oct 2006 23:19:57 +0000 (23:19 +0000)]
Always zap newly allocated pages, intead of conditional on PURIFY.
Minor performance penalty; c'est la vie. Better to be able to run
purify or valgrind or whatever on the binaries we actually use.
Ken Raeburn [Fri, 13 Oct 2006 20:58:14 +0000 (20:58 +0000)]
Eliminate some warnings: missing "const", unused variables, success
case in load_preauth_plugins was missing a return value (which isn't
checked anyways).
Sam Hartman [Fri, 13 Oct 2006 19:15:17 +0000 (19:15 +0000)]
* preauth_plugin.h: Don't use an enum in a public api
* kdc_preauth.c (get_etype_info): Rename get_entry_data to avoid shadowing
(get_etype_info2): likewise
(return_etype_info): likewise
(return_pw_salt): likewise
(get_entry_data): Update prototype not to use enum in a API
Ken Raeburn [Thu, 12 Oct 2006 02:39:14 +0000 (02:39 +0000)]
Another patch from Will Fiveash, to make "kdb5_util load <dumpfile>"
work with the LDAP KDB back end, in simple cases (all entries stored
under the krbcontainer entry).
Ken Raeburn [Thu, 12 Oct 2006 00:33:12 +0000 (00:33 +0000)]
(add_to_transited): Change the current logic to keep all array references
in bounds, assuming that what would've been next[-1] would not be '.'. I
haven't fully reexamined the logic, but this seems consistent with the
actual current behavior, and the existing test cases.
Also, factored out code for copying a string from a krb5_data to a char*.
Kevin Coffman [Tue, 10 Oct 2006 20:52:34 +0000 (20:52 +0000)]
Don't segfault if a preauth plugin module fails to load
Move the zeroing of the errinfo struct before the call to
krb5int_open_plugin_dirs() to prevent segfault in the case
where a plugin fails to load. (For example if there are
unresolved symbols.)
Jeffrey Altman [Mon, 9 Oct 2006 18:08:10 +0000 (18:08 +0000)]
final commits for KFW 3.1 Beta 2
krb5cred.dll (1.1.2.0)
- Fix the control logic so that if the password is expired for an
identity, the krb5 credentials provider will initiate a change
password request. Once the password is successfully changed, the
new password will be used to obtain new credentials.
- Fix an incorrect condition which caused the new credentials dialog
to refresh custom prompts unnecessarily.
- Removing an identity from the list of NetIDMgr identities now causes
the corresponding principal to be removed from the LRU principals
list.
- Properly handle KMSG_CRED_PROCESS message when the user is
cancelling out.
- Add more debug output
- Do not renew Kerberos tickets which are not initial tickets.
- Fix whitespace in source code.
- When providing identity selection controls, disable the realm
selector when the user specifies the realm in the username control.
- k5_ident_valiate_name() will refuse principal names with empty or
unspecified realms.
- When updating identity properties, the identity provider will
correctly set the properties for identities that were destroyed.
This fixes a problem where the values may be incorrect if an
identity has two or more credential caches and one of them is
destroyed.
nidmgr32.dll (1.1.2.0)
- Send out a separate notification if the configuration information
associated with an identity is removed.
- If an identity is being removed from the NetIDMgr identity list in
the configuration panel, do not send out APPLY notifications to the
subpanels after the configuration information has been removed.
Otherwise this causes the configuration information to be reinstated
and prevent the identity from being removed.
- Properly initialize the new credentials blob including the UI
context structure.
netidmgr.exe (1.1.2.0)
- When suppressing error messages, make sure that the final
KMSG_CRED_END notification is sent. Otherwise the new credentials
acquisition operation will not be cleaned up.
- Autoinit option now checks to see if there are identity credentials
for the default identity and triggers the new credentials dialog if
there aren't any.
- Properly synchronize the configuration node list when applying
changes (e.g.: when removing or adding an identity).
- Fix a handle leak when removing an identity from the NetIDMgr
identity list.
- Refresh the properties for the active identities before calculating
the renewal and expiration timers. Otherwise the timestamps being
used might be incorrect.
- Add Identity dialog (in the configuration panel) now uses the
identity selection controls provided by the identity provider.
- Improve type safety when handling timer refreshes.
- When getting the expiration times and issue times for an identity,
the timer refresh code may fail over to the expiration and issue
times for the credential it is currently looking at. Now the code
makes sure that both the issue and expiration times come from the
identity or the credential but not mixed.
- Not being able to get the time of issue of a credential now does not
result in the credential being skipped from the timer refresh pass.
However, not having a time of issue will result in the half-life
algorithm not being applied for the renew timer.
- Fix a bug which caused a credential to be abandoned from the timer
refresh pass if the reamining lifetime of the credential is less
than the renewal threshold.
- Fix a bug where the vertical scroll bars for the hypertext window
would not appear when the contents of the window changed.
- Trigger a refresh of the configuration nodes when adding or removing
an identity.
source for (1.1.2.0)
- Explicitly include <prsht.h> so that the SDK can be used in build
environments that define WIN32_LEAN_AND_MEAN.
Russ Allbery [Sun, 8 Oct 2006 03:22:48 +0000 (03:22 +0000)]
Initialize buffer before calling res_ninit
Per Paul Vixie: It is necessary to zero out the statbuf before calling
res_ninit(), or else res_vinit() will call res_nclose() and res_ndestroy()
with stack trash as a statbuf, and they will call free() with stack trash,
and programs will dump core.
Ticket: new
Component: krb5-libs
Version_Reported: 1.5.1
Target_Version: 1.5.2
Tags: pullup
Ken Raeburn [Sat, 7 Oct 2006 06:10:27 +0000 (06:10 +0000)]
Check for ldap_initialize and other functions that Solaris (Mozilla-based)
LDAP does not provide, and define versions a couple of them if needed.
Based on patches from and discussions with Will Fiveash.
Ken Raeburn [Sat, 7 Oct 2006 01:02:08 +0000 (01:02 +0000)]
Eliminate some warnings and non-gcc build problems:
- nested function
- bogus pointer casts
- C++-style comments
- unused variables
- variables of same name in nested scopes
- if condition syntax
- unused function
- use of GNU-only strndup() function
Tom Yu [Thu, 5 Oct 2006 22:58:41 +0000 (22:58 +0000)]
cursor for iterating over ccaches
Some ccache back ends need per-type cursors implemented.
* src/include/k5-int.h: Declare krb5_cc_ptcursor. Update
krb5_cc_ops vector to include functions for ptcursor and some
not-yet-implemented functionality.
Sam Hartman [Thu, 5 Oct 2006 21:28:58 +0000 (21:28 +0000)]
* kdc_preauth.c (return_padata): Allocate a padata context if not
already allocated. In the preauth_required path check will not be
called to set up the context first.
Ken Raeburn [Thu, 5 Oct 2006 19:45:52 +0000 (19:45 +0000)]
Patch from Kevin Coffman:
- adds a function to get ccache keycount
- uses it in two places
- fixes free problem if next_cred fails
- simplifies the clearcache function by using keyctl_clear
Ken Raeburn [Thu, 5 Oct 2006 13:17:59 +0000 (13:17 +0000)]
Add decode_tagged_unsigned_integer, and try to fix signed/unsigned and
long/int/int32 mixups in ASN.1 decoding. Add comments describing encoding
of key data. Don't always parenthesize safe_syncbuf arguments.
Sam Hartman [Tue, 3 Oct 2006 19:07:17 +0000 (19:07 +0000)]
Preauthentication Plugin Framework
Patch from Nalin Dahyabhai at Redhat to implement a preauthentication
framework based on the plugin architecture. Currently. the API is
considered internal and the header is not installed.
See src/include/krb5/preauth_plugin.h for the interface.
Ken Raeburn [Mon, 2 Oct 2006 22:50:10 +0000 (22:50 +0000)]
Merge Kevin Coffman's keyring ccache branch for Linux, with some modifications:
aclocal.m4: Enable keyring ccache if the header and library are available; no
configure-time option. No error if it's not found.
ccdefname.c: Keep old default of FILE: cache, at least for now.
libkrb5.exports: Don't export krb5_krcc_ops.
ccbase.c: Only initialize krb5int_krcc_mutex if USE_KEYRING_CCACHE; destroy it
in finalization. Define INITIAL_TYPEHEAD macro (for file vs keyring), and use
it for initialization and in krb5int_cc_finalize. Re-enable freeing of
additional registered-type structures.
cc_keyring.c: Avoid calls to com_err from within library.
cc_file.c: Punt change; generate_new is badly broken, and we expect to replace
it with a new API anyways.
projects / thirdparty / krb5.git / log
