If the OSD directory is using symlinks for referencing devices (like
block, db, wal for bluestore and journal for filestore) then the chown
command could fail to change the owner:group on some system.
$ ls -hl /var/lib/ceph/osd/ceph-0/
total 28K
lrwxrwxrwx 1 ceph ceph 92 Sep 15 01:53 block -> /dev/ceph-
45113532-95ca-471b-bd75-
51de46f1339c/osd-data-
570a1aee-60c0-44c9-8036-
ffed7d67a4e6
-rw------- 1 ceph ceph 37 Sep 15 01:53 ceph_fsid
-rw------- 1 ceph ceph 37 Sep 15 01:53 fsid
-rw------- 1 ceph ceph 55 Sep 15 01:53 keyring
-rw------- 1 ceph ceph 6 Sep 15 01:53 ready
-rw------- 1 ceph ceph 3 Sep 15 02:00 require_osd_release
-rw------- 1 ceph ceph 10 Sep 15 01:53 type
-rw------- 1 ceph ceph 2 Sep 15 01:53 whoami
$ find /var/lib/ceph/osd/ceph-0 -not -user 167 -execdir chown 167:167 {} +
chown: cannot dereference './block': Permission denied
$ find /var/lib/ceph/osd/ceph-0 -not -user 167
/var/lib/ceph/osd/ceph-0/block
Signed-off-by: Dimitri Savineau <dsavinea@redhat.com>
(cherry picked from commit
da4280e243f50114e1ae6455a46360012feb8f3d)
# The file module has to run checks on current ownership of all directories and files. This is unnecessary
# as in this case we know we want all owned by ceph user
- name: set proper ownership on ceph directories
- command: "find /var/lib/ceph /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
+ command: "find /var/lib/ceph /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown -h {{ ceph_uid }}:{{ ceph_uid }} {} +"
changed_when: false
- name: check for existing old leveldb file extension (ldb)