- check more thorougly for errors when resolving KEYRING ccache names of type
"persistent", which should only have a numeric UID as the next part of the
name (#1029110)
- drop patch to add additional access() checks to ksu - they add to breakage
when non-FILE: caches are in use (#1026099), shouldn't be resulting in any
benefit, and clash with proposed changes to fix its cache handling
Rebuild against a keyutils which tags the new symbols we're using with a
newer symbol version, so that RPM can tell the difference between
versions of the package which contain a shared library that doesn't
include them and versions of the package which contain a shared library
which does.
- switch to the version of persistent-keyring that was just merged to
master (RT#7711), along with related changes to kinit (RT#7689)
- go back to setting default_ccache_name to a KEYRING type
- pull up fix for not calling a kdb plugin's check-transited-path
method before calling the library's default version, which only knows
how to read what's in the configuration file (RT#7709, #1013664)
- configure --without-krb5-config so that we don't pull in the old default
ccache name when we want to stop setting a default ccache name at configure-
time
- back out setting default_ccache_name to the new default for now, resetting
it to the old default while the kernel/keyutils bits get sorted (sgallagh)
- don't let comments intended for one scriptlet become part of the "script"
that gets passed to ldconfig as part of another one (Mattias Ellert, #1005675)
- on releases where we expect krb5.conf to be configured with a
default_ccache_name, add it whenever we upgrade from an older version of
the package that wouldn't have included it in its default configuration
file (#991148)
- drop a patch we're not applying
- wrap kadmind and kpropd in scripts which check for the presence/absence
of files which dictate particular exit codes before exec'ing the actual
binaries, instead of trying to use ConditionPathExists in the unit files
to accomplish that, so that we exit with failure properly when what we
expect isn't actually in effect on the system (#800343)
- tweak configuration files used during tests to try to reduce the number
of conflicts encountered when builds for multiple arches land on the same
builder
Use nss_wrapper (from cwrap.org) to be able to run more of the
self-tests during %%check. Help it along a little bit by being
more emphatic about cutting off access to DNS.
This reverts commit 8a5a8d492c.
Special-case /run/user/0, attempting to create it when resolving a
directory cache below it fails due to ENOENT and we find that it doesn't
already exist, either, before attempting to create the directory cache
(maybe helping, maybe just making things more confusing for #961235).
- update to 1.11.3
- drop patch for RT#7605, fixed in this release
- drop patch for CVE-2002-2443, fixed in this release
- drop patch for RT#7369, fixed in this release
- pull upstream fix for breaking t_skew.py by adding the patch for #961221
It's not a complete fix, and it may only muddy things further on systems
that are having the kind of trouble it's trying to avoid, so hold off.
For now, at least.
- pull in proposed fix for attempts to get initial creds, which end up
following referrals, incorrectly trying to always use master KDCs if
they talked to a master at any point (should fix RT#7650)
Add a patch to create /run/user/0 if we're trying to resolve a
DIR: ccache somewhere below it and neither the target location
nor /run/user/0 exist yet.
The better workaround is to set the location's owner to "linger"
via logind, since even after we do what we're doing here, if
the user logs in and logs back out, our location is still removed.
