Greg Hudson [Sun, 4 Dec 2011 22:38:36 +0000 (22:38 +0000)]
Set a default enctype for optimistic preauth
When the client application requests optimistic preauth for a preauth
type which uses the password, we don't have an etype-info2 to
interpret since we haven't talked to the KDC. So we need to guess an
enctype, salt, and s2k parameters. In 1.9 and prior, encrypted
timestamp contained code to use the first requested enctype in this
case, but encrypted challenge did not. In 1.10 prior to this change,
neither mechanism uses a reasonable default.
Set a default enctype in krb5_init_creds_init so that all
password-based preauth mechanisms will use a reasonable default in the
optimistic preauth case. The default salt and s2k parameters for this
case will be the principal-based default salt and the enctype-based
default parameters.
Sam Hartman [Tue, 29 Nov 2011 21:22:21 +0000 (21:22 +0000)]
AC_CHECK_LIB should put -lcrypto in PKINIT_CRYPTO_IMPL_LIBS not LIBS
for pkinit. A similar problem exists for crypto_impl and is not
addressed by this patch.
ticket: new Subject: LIBS should not include PKINIT_CRYPTO_IMPL_LIBS tags: pullup target_version: 1.10
Sam Hartman [Wed, 23 Nov 2011 01:00:36 +0000 (01:00 +0000)]
FAST: error handling and const keyblock
krb5int_fast_process_error: Allow out_padata and retry to be null for
TGS case. Refactor function to do more frees in the exit handling and
to declare variables at the top.
krb5int_fast_reply_key: input keyblock arguments should be const
Sam Hartman [Wed, 23 Nov 2011 01:00:27 +0000 (01:00 +0000)]
ticket: new
subject: FAST PKINIT
target_version: 1.10
tags: pullup
Per RFC 6113 fast should use the inner request body for the pkinit
checksum. We did that on the KDC; now do so on the client. Remove
code that explicitly blocked pkinit under FAST.
Also, use the reply key *before* the strengthen key is applied when
verifying the PADATA_PKINIT_KX.
Greg Hudson [Tue, 22 Nov 2011 17:52:55 +0000 (17:52 +0000)]
Use POSIX constants instead of S_IREAD/S_IWRITE
S_IREAD and S_IWRITE are ancient names for S_IRUSR and S_IWUSR, and
are not defined on some modern platforms (such as Android). Use the
POSIX names instead.
Greg Hudson [Mon, 21 Nov 2011 21:14:39 +0000 (21:14 +0000)]
Clean up client-side preauth error data handling
Change the clpreauth tryagain method to accept a list of pa-data,
taken either from the FAST response or from decoding the e_data as
either pa-data or typed-data. Also change the in_padata argument to
contain just the type of the request padata rather than the whole
element, since modules generally shouldn't care about the contents of
their request padata (or they can remember it).
In krb5int_fast_process_error, no longer re-encode FAST pa-data as
typed-data for the inner error e_data, but decode traditional error
e_data for all error types, and try both pa-data and typed-data
encoding.
In PKINIT, try all elements of the new pa-data list, since it may
contain FAST elements as well as the actual PKINIT array. (Fixes an
outstanding bug in FAST PKINIT.)
Greg Hudson [Mon, 21 Nov 2011 17:30:41 +0000 (17:30 +0000)]
Recognize IAKERB mech in krb5_gss_display_status
Minor status codes were not displaying properly when originated from
the IAKERB mech, because of a safety check on mech_type. From Ralf
Haferkamp <rhafer@suse.de>.
Greg Hudson [Sun, 20 Nov 2011 05:19:45 +0000 (05:19 +0000)]
Fix failure interval of 0 in LDAP lockout code
A failure count interval of 0 caused krb5_ldap_lockout_check_policy to
pass the lockout check (but didn't cause a reset of the failure count
in krb5_ldap_lockout_audit). It should be treated as forever, as in
the DB2 back end.
This bug is the previously unknown cause of the assertion failure
fixed in CVE-2011-1528.
Greg Hudson [Tue, 15 Nov 2011 02:42:58 +0000 (02:42 +0000)]
Make verto context available to kdcpreauth modules
Add an event_context callback to kdcpreauth. Adjust the internal KDC
and main loop interfaces to pass around the event context, and expose
it to kdcpreauth modules via the rock.
Greg Hudson [Tue, 15 Nov 2011 01:59:01 +0000 (01:59 +0000)]
Update verto to 0.2.2 release
Update verto sources to 0.2.2 release versions. verto_reinitialize()
now has a return value; check it in kdc/main.c. Store verto-libev.c
alongside verto-k5ev.c to make it easy to diff corresponding versions
when updating.
Greg Hudson [Mon, 14 Nov 2011 21:45:33 +0000 (21:45 +0000)]
Simplify and fix kdcpreauth request_body callback
Alter the contract for the kdcpreauth request_body callback so that it
returns an alias to the encoded body instead of a fresh copy. At the
beginning of AS request processing, save a copy of the encoded request
body, or the encoded inner request body for FAST requests. Previously
the request_body callback would re-encode the request structure, which
in some cases has been modified by the AS request code.
No kdcpreauth modules currently use the request_body callback, but
PKINIT will need to start using it in order to handle FAST requests
correctly.
Greg Hudson [Mon, 14 Nov 2011 18:02:52 +0000 (18:02 +0000)]
Handle TGS referrals to the same realm
krb5 1.6 through 1.8 contained a workaround for the Active Directory
behavior of returning a TGS referral to the same realm as the request.
1.9 responds to this behavior by caching the returned TGT, trying
again, and detecting a referral loop. This is a partial regression of
ticket #4955. Detect this case and fall back to a non-referreal
request.
Greg Hudson [Fri, 11 Nov 2011 17:01:12 +0000 (17:01 +0000)]
Avoid looping when preauth can't be generated
If we receive a PREAUTH_REQUIRED error and fail to generate any real
preauthentication, error out immediately instead of continuing to
generate non-preauthenticated requests until we hit the loop count.
There is a lot of room to generate a more meaningful error about why
we failed to generate preauth (although in many cases the answer may
be too complicated to explain in an error message), but that requires
more radical restructuring of the preauth framework.
Greg Hudson [Fri, 11 Nov 2011 04:04:58 +0000 (04:04 +0000)]
Fix com_err.h dependencies in gss-kernel-lib
make check was failing in util/gss-kernel-lib due to dependencies
when the build is configured with --with-system-et, because depfix.pl
wasn't smart enough to substitute the dependency on com_err.h in the
current directory. Make depfix.pl smarter, and adjust COM_ERR_DEPS
to be com_err.h in gss-kernel-lib when building with the bundled
com_err.
Greg Hudson [Fri, 11 Nov 2011 02:14:56 +0000 (02:14 +0000)]
Remove -v from the preset valgrind flags
Verbose output from valgrind is rarely needed and makes it harder to
find errors and leaks in the output. When it is needed, it's easier
to add it in than to take it out.
Greg Hudson [Sun, 6 Nov 2011 04:32:34 +0000 (04:32 +0000)]
Fix month/year units in getdate
getdate strings like "1 month" or "next year" would fail some of the
time, depending on the value of stack garbage, because DSTcorrect()
doesn't set *error on success and RelativeMonth() doesn't initialize
error. Make DSTcorrect() responsible for setting *error in all cases.
Greg Hudson [Sat, 5 Nov 2011 19:55:34 +0000 (19:55 +0000)]
Improve verto and libev documentation
NOTICE was missing the copyright statement for verto (it's not quite
the same as other Red Hat licenses). util/verto had no README file,
and neither the verto nor k5ev README contained pointers to the
upstream project pages.
Greg Hudson [Fri, 4 Nov 2011 18:39:35 +0000 (18:39 +0000)]
Get rid of fake-install
Instead, use $(BUILDTOP)/plugins as the plugin base for tests. For
each real plugin module, create a link in the parent directory if
we're doing a shared-library build--so built KDB modules can be found
in plugins/kdb, preauth modules in plugins/preauth, etc..
Greg Hudson [Fri, 4 Nov 2011 05:53:23 +0000 (05:53 +0000)]
Exit on error in kadmind kprop child
When we fork from kadmind to dump the database and kprop to an iprop
slave, if we encounter an error in the child process we should exit
rather than returning to the main loop.
Greg Hudson [Thu, 3 Nov 2011 17:27:59 +0000 (17:27 +0000)]
Add cross-realm tests to python test framework
Add a cross_realms function to k5test.py to generate several linked
realms. Add a test script t_crossrealm.py to exercise six different
cross-realm scenarios.
Greg Hudson [Tue, 1 Nov 2011 03:49:16 +0000 (03:49 +0000)]
Conditionalize po subdir on msgfmt, not dgetext
The presence of dgettext in libc or libintl doesn't imply that msgfmt
is installed, so conditionalize building the po subdir on whether
msgfmt is installed.
Greg Hudson [Mon, 31 Oct 2011 16:43:40 +0000 (16:43 +0000)]
Make krb5_check_clockskew public
Rename krb5int_check_clockskew to krb5_check_clockskew and make it
public, in order to give kdcpreauth plugins a way to check timestamps
against the configured clock skew.
Greg Hudson [Fri, 28 Oct 2011 16:18:45 +0000 (16:18 +0000)]
Use zero-filled states for all async ops in KDC
There have been a couple of uninitialized field bugs in the
restructured KDC code, partly because compilers can't find these bugs
as easily as they can find uninitialized local variable bugs. Use
zero-filled state structures to make this type of bug less likely.
Greg Hudson [Fri, 28 Oct 2011 15:53:50 +0000 (15:53 +0000)]
Initialize typed_e_data in as_req_state
The typed_e_data field in struct as_req_state was not properly
initialized, causing the KDC to sometimes respond with typed-data
e_data for a preauth-required error when the client sends no padata.
This bug was masked with recent clients, which send a
KRB5_ENCPADATA_REQ_ENC_PA_REP padata.
Greg Hudson [Fri, 28 Oct 2011 15:45:03 +0000 (15:45 +0000)]
Fix intermediate key length in hmac-md5 checksum
When using hmac-md5, the intermediate key length is the output of the
hash function (128 bits), not the input key length. Relevant if the
input key is not an RC4 key.
Greg Hudson [Thu, 20 Oct 2011 15:16:03 +0000 (15:16 +0000)]
Fix handling of null edata method in KDC preauth
Correctly include an empty padata value if a KDC preauth system has no
get_edata method. This bug prevented the KDC from indicating FAST
support in preauth-required errors.
Greg Hudson [Thu, 20 Oct 2011 03:45:12 +0000 (03:45 +0000)]
Fix krb5_cc_set_config
krb5_cc_set_config has been non-functional since r24753 on cache types
which don't support removal of credential entries. Fix it by only
calling krb5_cc_remove_cred if data is NULL, since krb5_cc_store_cred
will do it anyway in the positive case.
Also fix an old memory leak in an uncommon error case.
Greg Hudson [Mon, 17 Oct 2011 17:15:31 +0000 (17:15 +0000)]
Add AC_LANG_SOURCE to PKINIT NSS version check
The configure.in code for the PKINIT NSS back end version check was
copied from the k5crypto NSS back end version check, but from before
r25181 which added AC_LANG_SOURCE wrappers.
Sam Hartman [Mon, 17 Oct 2011 00:45:30 +0000 (00:45 +0000)]
gssalloc-related fixes to naming_exts.c
renamed kg_data_list_to_buffer_set_nocopy to data_list_buffer_set
(since nocopy is no longer guaranteed).
removed extra indirection to input krb5_data list.
ensured input krb5_data list is always completely freed.
no longer returns EINVAL when output buffer set is NULL.
fixed krb5_gss_get_name_attribute to use data_to_gss.
Signed-off-by: Kevin Wasserman <kevin.wasserman@painless-security.com>
git-svn-id: svn://anonsvn.mit.edu/krb5/trunk@25358 dc483132-0cff-0310-8789-dd5450dbe970
Sam Hartman [Mon, 17 Oct 2011 00:45:23 +0000 (00:45 +0000)]
RFC 4120 says that we should not canonicalize using DNS. We cannot get
that far today, but there's no reason we should fail to use a
perfectly good principal name just because DNS is failing. For some
services there isn't even a requirement they be in DNS. With
AI_ADDRCONFIG there's no reason that Kerberos canonicalization should
fail simply because a v6 address is not present, for example. So, if
getaddrinfo fails in krb5_sname_to_principal simply use the input
hostname uncanonicalized.
Greg Hudson [Sat, 15 Oct 2011 16:56:30 +0000 (16:56 +0000)]
Allow password changes over NATs
In the kpasswd server code, don't set a remote address in the auth
context before calling krb5_rd_priv, since the kpasswd protocol is
well-protected against reflection attacks. This allows password
changes to work in cases where a NAT has changed the client IP address
as it is seen by the server.
Greg Hudson [Sat, 15 Oct 2011 16:56:26 +0000 (16:56 +0000)]
Allow rd_priv/rd_safe without remote address
Allow krb5_rd_priv and krb5_rd_safe to work when there is no remote
address set in the auth context, unless the KRB5_AUTH_CONTEXT_DO_TIMES
flag is set (in which case we need the remote address for the replay
cache name). Note that failing to set the remote address can create a
vulnerability to reflection attacks in some protocols, although it is
fairly easy to defend against--either use sequence numbers, or make
sure that requests don't look like replies, or both.
Greg Hudson [Sat, 15 Oct 2011 16:29:28 +0000 (16:29 +0000)]
Install krb5/preauth_plugin.h
The clpreauth and kdcpreauth pluggable interfaces are public as of
krb5 1.10. Install the header so that preauth modules can be built
outside of the krb5 source tree.
Greg Hudson [Sat, 15 Oct 2011 16:26:27 +0000 (16:26 +0000)]
Rename PAC type constants to avoid conflicts
Since the PAC type constants are now exposed in krb5.h, give them a
KRB5_ prefix so they don't conflict with similar PAC type constants
in other packages, like Samba.
Greg Hudson [Sat, 15 Oct 2011 16:06:03 +0000 (16:06 +0000)]
Hide gak_fct interface and arguments in clpreauth
Remove the gak_fct, gak_data, salt, s2kparams, and as_key arguments
of krb5_clpreauth_process_fn and krb5_clpreauth_tryagain_fn. To
replace them, add two callbacks: one which gets the AS key using the
previously selected etype-info2 information, and a second which lets
the module replace the AS key with one it has computed.
This changes limits module flexibility in a few ways. Modules cannot
check whether the AS key was already obtained before asking for it,
and they cannot use the etype-info2 salt and s2kparams for purposes
other than getting the password-based AS key. It is believed that
of existing preauth mechanisms, only SAM-2 preauth needs more
flexibility than the new interfaces provide, and as an internal legacy
mechanism it can cheat. Future mechanisms should be okay since the
current IETF philosophy is that etype-info2 information should not be
used for other purposes.
Greg Hudson [Sat, 15 Oct 2011 15:35:46 +0000 (15:35 +0000)]
Drop retransmits while processing requests
Supporting asynchronous preauth modules means that the KDC can receive
a retransmitted request before it finishes processing the initial
request. Ignore those retransmits instead of processing them.
projects / thirdparty / krb5.git / log
