If the client installer fails for some reason and --force was not used
then roll back the configuration.
This is needed because we touch /etc/sysconfig/network early in the
configuration and if it fails due to any number of issues (mostly related
to authentication) it will not be reset. We may as well run through the
entire uninstall process to be sure the system has been reset.
https://fedorahosted.org/freeipa/ticket/1704
When adding/removing source hosts if the host isn't found in IPA it is
considered external. The attribute externalhost is used to store
external hosts.
ticket https://fedorahosted.org/freeipa/ticket/1574
The installer and ipactl used two different methods to determine
whether IPA was configured, unify them.
When uninstalling report any thing that looks suspicious and warn
that a re-install may fail. This includes any remaining 389-ds instances
and any state or files that remains after all the module uninstallers
are complete.
Add wrappers for removing files and directories to log failures.
https://fedorahosted.org/freeipa/ticket/1715
When we perform an upgrade 389-ds is set to listen only on its
ldapi port. Theoretically it should be restored to the previous
state regardless of whether the upgrades were successful or not.
To be sure that a subsequent re-install will be successful go ahead
and remove the state for these options. Think of it as wearing a
belt and suspenders. Otherwise a re-install could return an error
message that IPA is already configured.
https://fedorahosted.org/freeipa/ticket/1667
We were orphaning a few files/directories when uninstalling 389-instances
both for IPA and dogtag. This should remove everything but the logs.
ticket https://fedorahosted.org/freeipa/ticket/1700
A Password param always prompted to confirm the entered password.
This doesn't make sense if you want to prompt for a password to another
system like we do with entitlements. This adds a new boolean option to
control the Password prompt parameter.
https://fedorahosted.org/freeipa/ticket/1695
We need an indicator to see if a keytab has been set on host and
service entries. We also need a way to know if a one-time password is
set on a host.
This adds an ACI that grants search on userPassword and
krbPrincipalKey so we can do an existence search on them. This way
we can tell if the attribute is set and create a fake attribute
accordingly.
When a userPassword is set on a host a keytab is generated against
that password so we always set has_keytab to False if a password
exists. This is fine because when keytab gets generated for the
host the password is removed (hence one-time).
This adds has_keytab/has_password to the user, host and service plugins.
ticket https://fedorahosted.org/freeipa/ticket/1538
Ade Lee from the dogtag team looked at the configuration code and
determined that a number of restarts were not needed and recommended
re-arranging other code to reduce the number of restarts to one.
https://fedorahosted.org/freeipa/ticket/1555
This can cause problems if a host is enrolled, unenrolled and a password
set. The password will be marked as expired like all new passwords are.
https://fedorahosted.org/freeipa/ticket/1526
When using the add_indirect helper we create a new map and then add a key
for it all in one step. If adding the key fails for any reason be sure to
remove the map we added.
https://fedorahosted.org/freeipa/ticket/1520
If you had a 64-bit system and installed a 32-bit version of IPA then
ipa-getkeytab probably wouldn't work because yum wouldn't know to pull
in the 32-bit version of cyrus-sasl-gssapi.
https://fedorahosted.org/freeipa/ticket/1499
* Check remote LDAP server to see if it is a V2 server
* Replace numeric return values with alphanumeric constants
* Display the error message from the ipa-enrollment extended op
* Remove generic join failed error message when XML-RPC fails
* Don't display Certificate subject base when enrollment fails
* Return proper error message when LDAP bind fails
https://fedorahosted.org/freeipa/ticket/1417
Allow a long to get as far as the min/max constraints where we can
compare it to min/max int values and reject with a proper error message.
https://fedorahosted.org/freeipa/ticket/1494
We have no visibility into whether an entry has a keytab or not so
krbLastPwdChange is used as a rough guide.
If this value exists during enrollment then it fails because the host
is considered already joined. This was getting set when a OTP was
added to a host that had already been enrolled (e.g. you enroll a host,
unenroll it, set a OTP, then try to re-enroll). The second enrollment
was failing because the enrollment plugin thought it was still
enrolled becaused krbLastPwdChange was set.
https://fedorahosted.org/freeipa/ticket/1357
Add a new command that lets you wait for an attribute to appear in
a value. Using this you can do things like wait for a managed entry
to be created, adding a new objectclass to the parent entry.
This is controlled by a new booleon option, wait_for_attr, defaulting
to False.
https://fedorahosted.org/freeipa/ticket/1144
If the password passed in when creating a NSS certificate database is None
then a random password is generated. If it is empty ('') then an empty
password is set.
Because of this the HTTP instance on replicas were created with an empty
password.
https://fedorahosted.org/freeipa/ticket/1407
The default precedence for plugins is 50 and the run in more or less
alphabetical order (but not guaranteed). This plugin needs to run after
the others have already done their work.
https://fedorahosted.org/freeipa/ticket/1370
The old nickname was 'RA Subsystem' and this may confuse some users
with the dogtag RA subsystem which we do not use.
This will only affect new installs. Existing installations will
continue to work fine.
https://fedorahosted.org/freeipa/ticket/1236
For the most part the existing replication code worked with the
following exceptions:
- Added more port options
- It assumed that initial connections were done to an SSL port. Added
ability to use startTLS
- It assumed that the name of the agreement was the same on both sides.
In dogtag one is marked as master and one as clone. A new option is
added, master, the determines which side we're working on or None
if it isn't a dogtag agreement.
- Don't set the attribute exclude list on dogtag agreements
- dogtag doesn't set a schedule by default (which is actually recommended
by 389-ds). This causes problems when doing a force-sync though so
if one is done we set a schedule to run all the time. Otherwise the
temporary schedule can't be removed (LDAP operations error).
https://fedorahosted.org/freeipa/ticket/1250
There were a few places in the code where certs were loaded from a
PKCS#7 file or a chain in a PEM file. The certificates got very
generic nicknames.
We can instead pull the subject from the certificate and use that as
the nickname.
https://fedorahosted.org/freeipa/ticket/1141
New rules will all be allow type. Existing rules cannot be changed to
deny.
The type attribute now defaults to allow with autofill so it won't be
prompted in interactive mode in the cli.
https://fedorahosted.org/freeipa/ticket/1432
It was my understanding that we would be able to pass in an existing UUID
when registering to connect to an existing registration (for the case where
IPA is re-installed). This is supported in the REST API but not python-rhsm.
https://fedorahosted.org/freeipa/ticket/1216
This fixes a regression.
We don't need to allow enrolledBy to be modified because it gets
written in the ipa_enrollment plugin which does internal operations
so bypasses acis.
https://fedorahosted.org/freeipa/ticket/302
The ipaldap.py code was updated to consider the schema when making
changes and does a REPLACE on single-value attributes. So when you
do an add in an update it will effectively replace the value instead
of ignoring it.
It will only ever return one entry so if more than one are found
then we raise an exception. This is most easily seen in the host
plugin where we search on the server shortname which can be the
same across sub-domains (e.g. foo.example.com &
foo.lab.example.com).
https://fedorahosted.org/freeipa/ticket/1388
This started as a problem in allowing leading/trailing whitespaces
on primary keys. In nearly every command other than add query is True
so all rules were ignored on the primary key. This meant that to
enforce whitespace we would need to define a validator for each one.
I decided instead to set self.all_rules to just the class rules if
query == True. So the minimum set of validators will be executed
against each type but param-specific validators will only run on add.
https://fedorahosted.org/freeipa/ticket/1285https://fedorahosted.org/freeipa/ticket/1286https://fedorahosted.org/freeipa/ticket/1287
If we set the callback before calling connect() then if the connection
tries a network family type and fails, it will try other family types.
If this happens then the callback set on the first socket will be lost
when a new socket is created. There is no way to query for the callback
in an existing socket.
https://fedorahosted.org/freeipa/ticket/1349
A dogtag replica file is created as usual. When the replica is installed
dogtag is optional and not installed by default. Adding the --setup-ca
option will configure it when the replica is installed.
A new tool ipa-ca-install will configure dogtag if it wasn't configured
when the replica was initially installed.
This moves a fair bit of code out of ipa-replica-install into
installutils and cainstance to avoid duplication.
https://fedorahosted.org/freeipa/ticket/1251
The hostname is passed in during the server installation. We should use
this hostname for the resulting server as well. It was being discarded
and we always used the system hostname value.
Important changes:
- configure ipa_hostname in sssd on masters
- set PKI_HOSTNAME so the hostname is passed to dogtag installer
- set the hostname when doing ldapi binds
This also reorders some things in the dogtag installer to eliminate an
unnecessary restart. We were restarting the service twice in a row with
very little time in between and this could result in a slew of reported
errors, though the server installed ok.
ticket 1052
For the most part certificates will be treated as being in DER format.
When we load a certificate we will generally accept it in any format but
will convert it to DER before proceeding in normalize_certificate().
This also re-arranges a bit of code to pull some certificate-specific
functions out of ipalib/plugins/service.py into ipalib/x509.py.
This also tries to use variable names to indicate what format the certificate
is in at any given point:
dercert: DER
cert: PEM
nsscert: a python-nss Certificate object
rawcert: unknown format
ticket 32
Otherwise it is possible for sssd to pick a different master to
communicate with via the DNS SRV records and if the remote master
goes down the local one will have problems as well.
ticket https://fedorahosted.org/freeipa/ticket/1187
The goal is to not import foreign certificates.
This caused a bunch of tests to fail because we had a hardcoded server
certificate. Instead a developer will need to run make-testcert to
create a server certificate generated by the local CA to test against.
ticket 1134
To support multiple direct maps we added description to the DN of
automount key entries. The downside of this is that to display a key
you had to know the information as well, which was rather pointless if
that is what you were trying to get.
So now both modes are supported. It will first look for just a key
in the description and fall back to including automountinformation
if it needs to.
Multiple direct maps are still supported and for those the info is
always required.
ticket 1229
Indirect membership is calculated by looking at each member and pulling
all the memberof out of it. What was missing was doing nested searches
on any members in that member group.
So if group2 was a member of group1 and group3 was a member of group2
we would miss group3 as being an indirect member of group1.
I updated the nesting test to do deeper nested testing. I confirmed
that this test failed with the old code and works with the new.
This also prevents duplicate indirect users and looping on circular
membership.
ticket https://fedorahosted.org/freeipa/ticket/1273
The conditional used to determine if thd CA 389-ds instance was already
configured was rather poor so it was possible to pass command-line
arguments in to confuse it. This would cause it to not be installed at
all causing the dogtag installation to fail in a strange way.
https://fedorahosted.org/freeipa/ticket/1244
The last step of a replication re-initiailization is to run the
memberof task. The current function would only authenticate using simple
auth to monitor the task but we may be doing this using admin GSSAPI
credentials so support that type of bind as well.
In short this fixes:
# kinit admin
# ipa-replica-manage re-initialize --from=master.example.com
https://fedorahosted.org/freeipa/ticket/1248
There was no point in limiting autobind root to just search cn=config since
it could always just modify its way out of the box, so remove the
restriction.
The upgrade log wasn't being created. Clearing all other loggers before
we calling logging.basicConfig() fixes this.
Add a global exception when performing updates so we can gracefully catch
and log problems without leaving the server in a bad state.
https://fedorahosted.org/freeipa/ticket/1243https://fedorahosted.org/freeipa/ticket/1254
When re-creating the CADS instance it needs to be more fully-populated
so we have enough information to create an SSL certificate and move
the principal to a real entry.
https://fedorahosted.org/freeipa/ticket/1245
Attempt to retrieve the schema the first time it is needed rather than
when Apache is started. A global copy is cached for future requests
for performance reasons.
The schema will be retrieved once per Apache child process.
ticket 583
Kerberos ticket policy can update policy in a user entry. This allowed
set/addattr to be used to modify attributes outside of the ticket policy
perview, also bypassing all validation/normalization. Likewise the
ticket policy was updatable by the user plugin bypassing all validation.
Add two new LDAPObject values to control this behavior:
limit_object_classes: only attributes in these are allowed
disallow_object_classes: attributes in these are disallowed
By default both of these lists are empty so are skipped.
ticket 744
Migration from a v2 server would fail because of our fake memberofindirect
attribute. This isn't in any objectclass so would cause entries to fail
to migrate. We can safely just remove it.
Also remove any limits on time/size when searching for entries on the
remote server. Otherwise only the number of entries configured in the
local IPA server can be migrated.
ticket 1124
If you install a replica, delete the replica, then re-add it and then
try to re-initialize the agreement it will fail because the remote master
has the old service principals cached. It needs to be restarted to work.
ticket 1077
The Managed Entries plugin configurations weren't being created on
replica installs. The templates were there but the cn=config
portions were not.
This patch adds them as updates. The template portion will be added
in the initial replication.
ticket 1222
Enforce that the remote hostname matches the remote SSL server certificate
when 389-ds operates as an SSL client.
Also add an update file to turn this off for existing installations.
This also changes the way the ldapupdater modlist is generated to be more
like the framework. Single-value attributes are done as replacements
and there is a list of force-replacement attributes.
ticket 1069
Even with --no-sssd authconfig was setting nsswitch.conf to use sssd
for users, groups, shadow and netgroups. We need to pass in the
--enableforcelegacy option hwen configuring nss_ldap.
Also always back up and restore sssd.conf. It still gets configured for
kerberos.
ticket 1142
We should more gracefully handle if the TGT has not been forwarded
than returning a 500 error.
Also catch and display KerberosErrors from ping() in the client better.
ticket 1101
There are times where one side or the other is missing its peers
krbprincipalname when converting from simple to GSSAPI replication. Ticket
1188 should address the cause of this.
This patch provides better information and handling should either side
be missing.
ticket 1044
This change means the UI can stop using the --all option and have to
retrieve significantly less information from the server. It also
speeds up user-find as it doesn't have to calculate membership.
This adds a new baseclass parameter, search_display_attributes, which
can provide a separate list from default_attributes just for find
commands.
The UI will need to be changed to switch from using cn to using
givenname and sn.
ticket 1136
Rob Crittenden