[LU-11756] kib_conn leak

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.13.0, Lustre 2.12.3
• Labels: patch
• Severity: 3
• Assignee: Andriy Skulysh
• Reporter: Andriy Skulysh

Description

crash>kib_conn_t.ibc_peer,ibc_refcount,ibc_list,ibc_state,ibc_comms_error,ibc_tx_queue_nocred 0xffff8808c0612e00
  ibc_peer = 0xffff880f0e2a2780
  ibc_refcount = {
    counter = 1
  }
  ibc_list = {
    next = 0xdead000000100100, 
    prev = 0xdead000000200200
  }
  ibc_state = 5
  ibc_comms_error = -12
  ibc_tx_queue_nocred = {
    next = 0xffffc900200de3e0, 
    prev = 0xffffc900200de3e0
  }

A tx was queued by a race into ibc_tx_queue_nocred while disconnecting connection. So the connection stays in unused state but it can't be destroyed because it is referenced by the tx.
It results in 

[18891797.073780] Lustre: 1572:0:(niobuf.c:292:ptlrpc_abort_bulk()) Unexpectedly long timeout: desc ffff880a4eaf1c00