Commit efed5633 (r24759) prefixed some symbols with rk_, but
introduced 6 duplicate symbols in the version script (because the
rk_-prefixed versions of the symbols were already present).
When the master's log does not contain the complete history, slaves
that bootstrap from scratch encountered a loop, because the master
falsely assumed a race with log truncation.
The function, found in lib/krb5/mit_glue.c, is currently using
krb5_generate_random_keyblock(). This compiles because warning-level is
not high enough, but does not work. At runtime the
krb5_generate_random_keyblock() interprets the second argument as the
krb5_enctype (rather than a length of anything) and tries to verify it.
When the length does not match any known enctype, as usually happens, the
function fails and returns an error. If the length happened to correspond
to an enctype, the function would likely crash due to misinterpreting its
third argument as a valid krb5_keyblock.
The change uses krb5_generate_random_block() instead. This function does
not return anything -- upon detecting failure it will cause the entire
application to exist instead...
Change-Id: I865a360037a513ce91abc7abba1dc554f844b464
Follow Apple's lead and do not fail krb5_set_config_files() simply
because one of the files in the profile list fails to parse correctly.
Doing so can lead to hard to find failures and could lead to an end
user shooting themselves in the foot and no longer be able to login
to their system to fix it.
Parse as many of the files as we can. Only fail krb5_set_config_files()
if init_context_from_config_file() fails.
Change-Id: I122664c6d707a5f926643808ba414bf4f681f8b8
When flushing the uber record, retain the current log version. When the
uber record is the only (thus last) record in the log, return its nominal
version as the last version, not 0. Upgrade the log if the current uber
record version number is not 0.
Otherwise we risk storing a name with the referral (empty) realm name,
which will then cause various knock-on effects, such as thinking that
the start_realm is "", and failing to find matching credentials in the
ccache.
When storing credentials whose ticket principal is not equal to the
requested principal, store the ticket under both names not only when
the original realm is the referral realm, but more generally for any
difference at all. This matches MIT behaviour.
Allow explicit name_canon rules to specify a realm to go with the
canonicalized hostname, if that realm is empty the effect is the
same "use-referrals"
Also fix segfault when no creds and debugging
On OS X anyways, since Heimdal cannot be built to replace the system
Kerberos implementation in OS X (even though it's based on Heimdal).
Heimdal plugins other than the CCAPI plugins have private ABIs with
strong coupling to the internals of the Heimdal libraries, thus using
system plugins in a non-system Heimdal is likely to end in tears (e.g.,
segfaults).
This means, for example, that OS X's plugins for PAC creation and
verification cannot be used by Heimdal.
When the master's log has all entries from version 1 to now, and no
uber entry (legacy master), then new slaves will not pull version 1,
because their uber record has version 1.
The fix is to force the uber version to 0 always, and avoid adding a
truncate nop when doing a full prop. The uber record now records the
database version even in the absence of any other log entries so that
we know what to pull going forward.
For consistency make "-I" part of the macro value set by autoconf.
For now, don't attempt to handle OpenSSL rpath in cf/crypto.m4.
That's much easier by just setting LDFLAGS when running configure.
Otherwise too many Makefiles to edit and libtool and automake do
their best to undo the rpath.
We added some new OID symbols in libasn1, make them public.
When an older Heimdal is already installed and its libraries don't
have some newly created symbols we run into build or test problems,
if libtool decides to use installed rather than just-built libraries.
This was happening with a few of test programs in libhx509. Fixed.
When building evp-openssl.c without support for OpenSSL it is necessary
to include evp-hcrypto.h to define the HCRYPTO_FALLBACK functions.
Change-Id: Ifd51f9fcd2b1805a534a9f88992162818afffe7d
Move roken.h to the top of the header after config.h and remove the
includes that are duplicated by roken.h.
Change-Id: I33e29736519177f17c36e5c5948d7022011ab2b2
All source files must start with
include <config.h>
include <roken.h>
when krb5_locl.h or kdc_locl.h are includes, they must come before
other Heimdal include files.
Do not include stdint.h when roken.h is included.
Do not include config.h more than once.
Change-Id: I0baecb5d48317996f48b1a6c41b051f42f2fde61
Coverity complains about the leakage of 'handle' when the identifier
goes out of scope. Change handle into a static global to hold the
value instead of a stack variable.
Change-Id: I040707ac731558f7d523f128a006a80b98d45b79
The Windows and PKCS#11 backends do this. The Common Crypto (OS X)
backend does not. Ideally this should be a ./configure option, and that
might be the next step, but right now we need this fallback in order to
get tests passing in Travis-CI.
The resolver plugin doesn't have a way to say "and stop here", so the
lookup done in lib/krb5/test_plugin.c can produce more results than used
to be expected, and indeed nowadays it does. The fix is to ensure that
we have the desired results and ignore the others.
