Red Hat Bugzilla – Bug 1479834
Multiple ptp4l use same /var/run/ptp4l
Last modified: 2018-05-17 05:09:53 EDT
Description of problem: ptp4l uses the socket /var/run/ptp4l to communicate with the management tool pmc and to deliver an up to date interface list to phc2sys. Multiple ptp4l by intention or by mistake will use the same socket. Killing one of those ptp4l processes would remove the socket, phc2sys won't be able to synchronize any longer due to the lack of communication with ptp4l. Version-Release number of selected component (if applicable): Tested on linuxptp-1.4-3.20140718gitbdb6a3.el7.x86_64 linuxptp-1.7.1.fc25.x86_64 linuxptp-1.8-3.el7.x86_64 How reproducible: always Steps to Reproduce: 1. Install linuxptp 2. Run "ptp4l -i enp0s25 -m" 3. Run "phc2sys -a -m" 4. Run "ptp4l -i enp0s25 -m" 5. Stop second instance of ptp4l Actual results: phc2sys will stop synchronizing time, complaining... phc2sys[3829328.174]: uds: sendto failed: No such file or directory phc2sys[3829328.174]: uds: sendto failed: No such file or directory phc2sys[3829329.174]: uds: sendto failed: No such file or directory Expected results: phc2sys will continue synchronizing time Additional info: phc2sys requires bare metal system with an PTP capable NIC. It's unclear to me if and when how we could change the current behavior. In multiple ptp4l setups the phc2sys process is usually configured manually, still we have an issue in cases where a second ptp4l was started "to check the functionality" and killed afterwards. Leaving the socket in place might be an option, but requires further testing.
I think one way to solve this would be to modify ptp4l to not try to remove the socket on start. If binding to the socket fails, ptp4l will work without the UDS transport, or perhaps it could be handled as a fatal error. The question is if it would be acceptable behavior in case the socket was left there due to ptp4l terminating in an abnormal way (e.g. crash). The user would have to manually remove the socket in order to get phc2sys working. It will change the user experience.
FWIW, an RFC patch was posted for this on the upstream list https://www.mail-archive.com/linuxptp-devel@lists.sourceforge.net/msg02103.html
As this would change the behavior of ptp4l in case of an existing socket (from existing or crashed ptp4l instance), I think it would rather be a candidate for the next major RHEL release instead of minor release.
Development Management has reviewed and declined this request. You may appeal this decision by reopening this request.