https://bugzilla.redhat.com/show_bug.cgi?id=1868696
Bug ID: 1868696 Summary: sssd_kcm loops on NFS server occasionally after an NFSv4.0 mount using Kerberos Product: Fedora Version: 32 Hardware: All OS: Linux Status: NEW Component: sssd Severity: high Assignee: sssd-maintainers@lists.fedoraproject.org Reporter: chuck.lever@oracle.com QA Contact: extras-qa@fedoraproject.org CC: abokovoy@redhat.com, atikhono@redhat.com, jhrozek@redhat.com, lslebodn@redhat.com, mzidek@redhat.com, pbrezina@redhat.com, rharwood@redhat.com, sbose@redhat.com, ssorce@redhat.com, sssd-maintainers@lists.fedoraproject.org Target Milestone: --- Classification: Fedora
Description of problem: Multi-homed Fedora 32 NFS client and server, both with keytabs. Sometimes after performing an NFSv4.0 mount, the sssd_kcm daemon on the server runs at 100% and fills up the sssd_kcm.log file (and eventually, the root partition).
(2020-08-10 15:03:04:680265): [kcm] [kcm_input_parse] (0x1000): Received message with length 0 (2020-08-10 15:03:04:680284): [kcm] [kcm_input_parse] (0x0020): Illegal zero-length message (2020-08-10 15:03:04:680302): [kcm] [kcm_recv] (0x0010): Failed to parse data (74, Bad message), aborting client (2020-08-10 15:03:04:680319): [kcm] [kcm_reply_error] (0x0040): KCM operation returs failure [74]: Bad message (2020-08-10 15:03:04:680353): [kcm] [kcm_failbuf_construct] (0x1000): Sent reply with error -1765328188
Version-Release number of selected component (if applicable): sssd-kcm-2.3.1-2.fc32.x86_64
How reproducible: Happens every second or third mount operation.
Steps to Reproduce: 1. Set up NFS client and server with keytabs 2. Repeat: "mount -o vers=4.0,sec=sys" / do some operations / umount 3. Watch the server with "top"
Actual results: Sometimes sssd_kcm goes to 100% of a CPU and must be killed.
Expected results: No looping.
Additional info: The NFSv4.0 backchannel, in this case, is secured with GSS krb5i. It appears that gssd on the server is accessing the Kerberos ticket cache while setting up the backchannel, and sometimes this triggers the kcm loop.