http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c4 Markos Chandras changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |fbui@suse.com --- Comment #4 from Markos Chandras --- So I had a look but I am not sure what's going on. The ovsdb-server service file is the following one https://github.com/openvswitch/ovs/blob/master/rhel/usr_lib_systemd_system_o... As you can see, it contains RuntimeDirectory=openvswitch RuntimeDirectoryMode=0755 which means that /run/openvswitch will be created by systemd. Then we have these lines ExecStartPre=/usr/bin/chown ${OVS_USER_ID} /var/run/openvswitch ExecStart=/usr/share/openvswitch/scripts/ovs-ctl \ --no-ovs-vswitchd --no-monitor --system-id=random \ --ovs-user=${OVS_USER_ID} \ start $OPTIONS and as you can see the ExecStartPre line will chown that directory to the correct user. However, this doesn't seem to happen anymore (again, nothing changed in the package so this _used to_ work fine. Is there any chance for systemd to have changed its behavior and now create RuntimeDirectory before ExecStart is executed instead of ExecStartPre? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c6 Franck Bui changed: What |Removed |Added ---------------------------------------------------------------------------- Flags|needinfo?(fbui@suse.com) | --- Comment #6 from Franck Bui --- Indeed there's a slight behavior change in systemd introduced by commit 3536f49e8fa281539798a7bc5004d73302f39673 (v235). Basically RuntimeDirectory= is supposed to be used by unprivileged services (the ones using User= option) since they lack privileges to create directories in /run. systemd will create, set the permissions and change the owner of those directories before spawning every commands defined by ExecStart* options. Admittedly I don't know the reason why it does that for every spawned commands... Before the mentioned commit, there was a special case though: if User= wasn't specified, the runtime dirs were created and owned by root and that was done only once before the first command was spawned. Now the special case is gone and systemd does that for every commands. As said previously I don't know if it's intended to process the runtime dirs each time a command specified by ExecStart*= option is spawned. It would be interesting to create a bug on systemd upstream. However what does openvswitch is pretty weird since it wants systemd to manage /run/openvswitch but in the meantime it manually set the owner of the directory. Is there any reason why the runtime dir isn't directly created by ovs-ctl or the daemon ovs-ctl starts ? That should be the case already otherwise if one starts the daemon directly (without systemd involved at all) it would fail, no ? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 Markos Chandras changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[Build 20180318] |openvswitch: Permission |openvswitch service fails |denied when starting the |to start |ovsdb-server service -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c8 Markos Chandras changed: What |Removed |Added ---------------------------------------------------------------------------- Flags| |needinfo?(fbui@suse.com) --- Comment #8 from Markos Chandras --- systemd restored its original behavior in v238 with https://github.com/systemd/systemd/commit/30c81ce2ce so ovs upstream said that my patch is probably not needed provided we backport the fix in our v237 or move to v238 in TW. My preference would be to patch systemd with the upstream fix rather than patching ovs with an out-of-tree patch to temporary workaround the systemd behavior. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c10 Franck Bui changed: What |Removed |Added ---------------------------------------------------------------------------- Status|IN_PROGRESS |RESOLVED Resolution|--- |FIXED --- Comment #10 from Franck Bui --- Fix backported into SUSE/v237 branch (the other branches don't suffer from this, see comment #6). It will be part of the next update of Factory/TW. Closing. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c15 Pavol Cupka changed: What |Removed |Added ---------------------------------------------------------------------------- Status|VERIFIED |REOPENED CC| |palica@liguros.net Resolution|FIXED |--- --- Comment #15 from Pavol Cupka --- I would like to reopen this bug as I am hitting the error again with TW (20190525) May 28 15:34:29 borvo ovsdb-server[8015]: ovs|00002|daemon_unix|EMER|/run/openvswitch/ovsdb-server.pid.tmp: create failed (Permission denied) May 28 15:34:29 borvo ovs-ctl[7972]: ovsdb-server: /run/openvswitch/ovsdb-server.pid.tmp: create failed (Permission denied) May 28 15:34:29 borvo ovs-ctl[7972]: Starting ovsdb-server ... failed! May 28 15:34:29 borvo systemd[1]: ovsdb-server.service: Control process exited, code=exited, status=1/FAILURE -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c17 --- Comment #17 from Pavol Cupka --- also failing again in QA https://openqa.opensuse.org/tests/942903#step/openvswitch/6 -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1085971 http://bugzilla.opensuse.org/show_bug.cgi?id=1085971#c18 --- Comment #18 from Franck Bui --- The behavior in systemd changed again. I opened an issue against upstream [1] so hopefully a clear description of the attended behavior will be provided. Then we could decide whether the fix will leave in systemd or openvswitch will need some reworks. [1] https://github.com/systemd/systemd/issues/12722 -- You are receiving this mail because: You are on the CC list for the bug.