Fix cinder's cephx keyring file permissions

This patch updates cinder's kolla permissions so that cinder can
access any cephx keyring associated with CephExternalMultiConfig
ceph clusters. The new approach parses the cluster names out of the
CephExternalMultiConfig array, and uses a wildcard to grant access
to all keys (regardless of the key name) defined for each cluster.
There is no risk of the wildcard granting improper access to a
privileged key (e.g. the admin key), because CephExternalMultiConfig
doesn't include privileged keys.

This patch replaces similar (but more restrictive) code added in
I73af5b868de629870a35d38f8436e7025aae791e. That patch allowed cinder
to access cephx keyrings associated with a new CinderRbdMultiConfig
parameter, but it didn't cover all potential use cases. For example,
in a DCN/Edge deployment, cinder services running at the edge need
access to the central site's client key in order to perform operations
like offline volume migration.

Closes-Bug: #1930620
Resolves: rhbz#1962304
Change-Id: I4423fcbd62b09ef323590fc740dd29e1a17777f5
This commit is contained in:
Alan Bishop 2021-06-02 12:52:48 -07:00
parent 815143179f
commit f1cd8006fe
1 changed files with 30 additions and 20 deletions

View File

@ -87,23 +87,34 @@ parameters:
description: >
The Ceph cluster name must be at least 1 character and contain only
letters and numbers.
CinderRbdMultiConfig:
CephExternalMultiConfig:
type: json
default: {}
hidden: true
description: |
Dictionary of settings when configuring multiple RBD backends. The
hash key is the backend name, and the value is a dictionary of parameter
values unique to that backend. The following parameters are required,
and must match the corresponding value defined in CephExternalMultiConfig.
CephClusterName (must match the CephExternalMultiConfig entry's 'cluster')
CephClusterFSID (must match the CephExternalMultiConfig entry's 'fsid')
The following parameters are optional, and override the corresponding
parameter's default value.
CephClientUserName
CinderRbdPoolName
CinderRbdExtraPools
CinderRbdAvailabilityZone
CinderRbdFlattenVolumeFromSnapshot
List of maps describing extra overrides which will be applied when configuring
extra external Ceph clusters. If this list is non-empty, ceph-ansible will run
an extra count(list) times using the same parameters as the first run except
each parameter within each map will override the defaults. If the following
were used, the second run would configure the overcloud to also use the ceph2
cluster with all the previous parameters except /etc/ceph/ceph2.conf would have
a mon_host entry containing the value of external_cluster_mon_ips below, and
not the default CephExternalMonHost. Subsequent ceph-ansible runs are restricted
to just ceph clients. CephExternalMultiConfig may not be used to deploy additional
internal Ceph clusters within one Heat stack. The map for each list should contain
not tripleo-heat-template parameters but ceph-ansible parameters.
- cluster: 'ceph2'
fsid: 'e2cba068-5f14-4b0f-b047-acf375c0004a'
external_cluster_mon_ips: '172.18.0.5,172.18.0.6,172.18.0.7'
keys:
- name: "client.openstack"
caps:
mgr: "allow *"
mon: "profile rbd"
osd: "osd: profile rbd pool=volumes, profile rbd pool=backups, profile rbd pool=vms, profile rbd pool=images"
key: "AQCwmeRcAAAAABAA6SQU/bGqFjlfLro5KxrB1Q=="
mode: "0600"
dashboard_enabled: false
default: []
conditions:
cvol_active_active_tls_enabled:
@ -194,16 +205,15 @@ outputs:
owner: cinder:cinder
- repeat:
template:
path: /etc/ceph/<%keyring%>
path: /etc/ceph/<%cluster%>.client.*.keyring
owner: cinder:cinder
perm: '0600'
for_each:
<%keyring%>:
<%cluster%>:
yaql:
expression: let(u => $.data.default_user) -> $.data.multiconfig.values().select("{0}.client.{1}.keyring".format($.CephClusterName, $.get("CephClientUserName", $u)))
expression: $.data.multiconfig.select($.cluster)
data:
default_user: {get_param: CephClientUserName}
multiconfig: {get_param: CinderRbdMultiConfig}
multiconfig: {get_param: CephExternalMultiConfig}
cinder_volume_host_prep_tasks:
description: Host prep tasks for the cinder-volume service (HA or non-HA)