-- 좀더 간결한 방법이 있을 것 같지만 생각나는대로 적어봅니다.
WITH T AS (
    SELECT 1 ID, NULL PID, 'A' VAL FROM DUAL UNION ALL
    SELECT 2 ID, 1 PID, 'B' VAL FROM DUAL UNION ALL
    SELECT 3 ID, 1 PID, 'C' VAL FROM DUAL UNION ALL
    SELECT 4 ID, 2 PID, 'D' VAL FROM DUAL UNION ALL
    SELECT 5 ID, NULL PID, 'E' VAL FROM DUAL UNION ALL
    SELECT 6 ID, 5 PID, 'F' VAL FROM DUAL UNION ALL
    SELECT 7 ID, 5 PID, 'G' VAL FROM DUAL
)
SELECT  V2.LV, VAL
FROM    (
            SELECT  V.*, COUNT(DECODE(RNK, 1, 1)) OVER () CNT
            FROM    (
                        SELECT  LEVEL LV, VAL,
                                DENSE_RANK() OVER (ORDER BY CONNECT_BY_ROOT ID) RNK,
                                ROWNUM RN
                        FROM    T
                        START WITH PID IS NULL
                        CONNECT BY PRIOR ID = PID
                        ORDER BY ID
                    ) V
        ) V2
WHERE   RNK = 1 OR RN = CNT + 1;

-- 이런 방법도.

with t as
(
  select 1 rn, 1 lv , 'A' data from dual union all
  select 2 rn, 2 lv , 'B' data from dual union all
  select 3 rn, 2 lv , 'C' data from dual union all
  select 4 rn, 3 lv , 'D' data from dual union all
  select 5 rn, 1 lv , 'E' data from dual union all
  select 6 rn, 2 lv , 'F' data from dual union all
  select 7 rn, 2 lv , 'G' data from dual 
)
select rn,lv,data,grp
from 
(
select rn,lv,data
      ,sum(decode(lv,1,1)) over (order by rn rows between unbounded preceding and 1 preceding) grp
from t
)
where nvl(grp,1) = 1

두 분 답변 감사합니다. 컨트롤 딴에서 처리해버렸는데 답변 달아준 쿼리로 수정해서도 해봐야겠네요. 감사합니다.