Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-361

Lustre Client crashes due to ASSERTION(!request->rq_replay) failed

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Duplicate
    • Icon: Major Major
    • Lustre 2.1.0
    • Lustre 2.0.0
    • None
    • 3
    • 8536

      At CEA they face Lustre-Client crashes due to ASSERTION(!request->rq_replay) failed.
      As a consequence the application is lost and the node needs to be rebooted.

      The crashing task stack always looks like :
      ===========================================
      crash_kexec()
      panic()
      lbug_with_loc()
      __ptlrpc_req_finished()
      ptlrpc_req_finished()
      ll_intent_release()
      ll_create_nd()
      vfs_create()
      do_filp_open()
      do_sys_open()
      sys_open()
      system_call_fastpath()
      ===========================================

      It is possible to reproduce this issue very simply, by running on a Lustre client a short program that do an 'open' with the O_CREAT flag, but without specifying the third, mode parameter.
      I know that the third parameter is mandatory when using the O_CREAT flag. If it is not given, then arbitrary permissions will be set, depending on the content of the stack.
      But in any case it should lead to a client crash. When running the same test on ext4 for instance, the permissions set are non-sense but no crash occurs.

      Sebastien.

            johann Johann Lombardi (Inactive)
            sebastien.buisson Sebastien Buisson (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

              Created:
              Updated:
              Resolved: