[HDFS-12121] Audit logging of unsuccessful namesystem calls should not be done while holding fslock.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.7.0
• Target Version/s: 2.9.0, 3.0.0-beta1
• Assignee: Rushabh Shah
• Reporter: Rushabh Shah
• Created: 11/Jul/17 15:36
• Updated: 11/Jul/17 17:59
• Resolved: 11/Jul/17 17:59

Description

Currently we are audit logging the unsuccessful namesystem calls while holding fslock.
Following is just one example.

FSNamesystem.java

HdfsFileStatus getFileInfo(final String src, boolean resolveLink)
    throws IOException {
    final String operationName = "getfileinfo";
    checkOperation(OperationCategory.READ);
    HdfsFileStatus stat = null;
    readLock();
    try {
      checkOperation(OperationCategory.READ);
      stat = FSDirStatAndListingOp.getFileInfo(dir, src, resolveLink);
    } catch (AccessControlException e) {
      logAuditEvent(false, operationName, src); //<---- audit logging while holding fs read lock.
      throw e;
    } finally {
      readUnlock(operationName);
    }
    logAuditEvent(true, operationName, src);
    return stat;
  }

We should avoid doing this.