Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-20452

gss: lgss_keyring requires reverse DNS for SK/null

XMLWordPrintable

    • Icon: Improvement Improvement
    • Resolution: Unresolved
    • Icon: Medium Medium
    • None
    • None
    • None
    • 3
    • 9223372036854775807

      On a client with no reverse DNS configured (no PTR records) for the server NIDs it is trying to communicate with, a client SSK mount fails with kernel logs showing errors such as:

      [Mon Jul  6 13:05:22 2026] LustreError: 21776:0:(gss_keyring.c:1822:gss_kt_update()) negotiation: rpc err -13, gss err 0
      ...
      [Mon Jul  6 13:06:25 2026] LustreError: 39256:0:(gss_keyring.c:1822:gss_kt_update()) Skipped 3555 previous similar messages
      [Mon Jul  6 13:06:52 2026] LustreError: 21749:0:(lmv_obd.c:1466:lmv_statfs()) lustre-MDT0000-mdc-ffff9c0fc79ec000: can't stat MDS #0: rc = -1
      [Mon Jul  6 13:06:52 2026] LustreError: 21749:0:(super25.c:183:lustre_fill_super()) llite: Unable to mount <unknown>: rc = -1
      

      lgss_keyring shows huge numbers of the following, repeated errors showing failure to resolve a hostname for each server NID:

      Jul 06 13:05:29 client1 lgss_keyring[21776]: [21776]:ERROR:getaddrcanonname(): failed to get name for 10.0.101.1: Name or service not known
      Jul 06 13:05:29 client1 lgss_keyring[21776]: [21776]:ERROR:ipv4_nid2hostname(): SOCKLND: failed to get canonical name of 0x165000a
      Jul 06 13:05:29 client1 lgss_keyring[21776]: [21776]:ERROR:lgss_get_service_str(): cannot resolve hostname from nid 200650a006501
      Jul 06 13:05:29 client1 lgss_keyring[21776]: [21776]:ERROR:lgssc_kr_negotiate_manual(): key 00d8421e: failed to construct service string
      Jul 06 13:05:29 client1 lgss_keyring[21776]: [21776]:ERROR:error_kernel_key(): unlink key 00d8421e from -3: No such file or directory
      

      Simply adding in host records for the server IPs will fix this:

      [tenant1] [root@client1 ~]# for ip in 10.0.101.1 10.16.101.1 10.0.101.2 10.16.101.2; do printf "%-12s " $ip; getent hosts $ip; done
      10.0.101.1   10.0.101.1      server1-lneteth1.101
      10.16.101.1  10.16.101.1     server1-lneteth2.101
      10.0.101.2   10.0.101.2      server2-lneteth1.101
      10.16.101.2  10.16.101.2     server2-lneteth2.101
      [tenant1] [root@client1 ~]# mount -t lustre -o skpath=/etc/lustre/ssk/tenant1.key 10.0.101.1@tcp101:10.0.101.2@tcp101:/lustre /lustre/tenant1
      [tenant1] [root@client1 ~]# df -t lustre
      Filesystem                                  1K-blocks   Used Available Use% Mounted on
      10.0.101.1@tcp101:10.0.101.2@tcp101:/lustre   3145728 819212   2326516  27% /lustre/tenant1
      

      with zero errors shown in the logs.

      My question is - is it actually architecturally mandatory for pure-SSK deployments to require reverse DNS is configured? I understand it is required for kerberos deployments that will use this, but for SSK-only deployments, can we drop this requirement to simplify?

            mrasobarnett Matt Rásó-Barnett
            mrasobarnett Matt Rásó-Barnett
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated: