上周在客户那里遇到了一例由Oracle Bug引发的表数据块逻辑讹误触发ORA-00600:[13013], [5001]的问题，这里为了更好地说明该问题，于是萌发了手工模拟该数据块逻辑讹误的想法。 基础知识 Oracle中表的数据块由块头、事务槽、行字典、行数据等多种结构组成。 行数据(rowdata)实际是由许多row piece 行片组成的，每一条row piece的头部都有flag、locks、cols(cc)三个标志位。 其中flag标记了该row piece的类型，该flag位占用一个字节，其不同的bit位代表不同的含义，见下表：
ROW_CLUSTER_KEY = 0x80;              KDRHFK
ROW_CTABLE_NUMBER = 0x40;            KDRHFC
ROW_HEAD_PIECE = 0x20;               KDRHFH
ROW_DELETED_ROW = 0x10;              KDRHFD
ROW_FIRST_PIECE = 0x08;              KDRHFF
ROW_LAST_PIECE = 0x04;               KDRHFL
ROW_FROM_PREVIOUS = 0x02;            KDRHFP
ROW_CONTINUE_NEXT = 0x01;            KDRHFN
一般来说最普通的一条row piece是普通堆表(heap table)的未被删除的且无行迁移/链接的，其flag位应为
普通row的flag一般为

Single Row =
ROW_HEAD_PIECE + ROW_FIRST_PIECE + ROW_LAST_PIECE= 0x20 + 0x08 + 0x04= 0x2c

===================================================================================

cluster key的flag一般为

Cluster Key =
ROW_CLUSTER_KEY + ROW_HEAD_PIECE + ROW_FIRST_PIECE + ROW_LAST_PIECE=
KDRHFL, KDRHFF, KDRHFH, KDRHFK =0x80 + 0x2c =  0xac

BBED> x /rn
rowdata[68]                                 @8166
-----------
flag@8166: 0xac (KDRHFL, KDRHFF, KDRHFH, KDRHFK)
lock@8167: 0x00
cols@8168:    1
kref@8169:    1
mref@8171:    1
hrid@8173:0x01800014.0
nrid@8179:0x01800014.0

col    0[2] @8185: 10

===================================================================================

Cluster Row =
ROW_CTABLE_NUMBER + ROW_HEAD_PIECE + ROW_FIRST_PIECE + ROW_LAST_PIECE =
(KDRHFL, KDRHFF, KDRHFH, KDRHFC) = 0x6c

BBED> x /rncc
rowdata[0]                                  @8098
----------
flag@8098: 0x6c (KDRHFL, KDRHFF, KDRHFH, KDRHFC)
lock@8099: 0x00
cols@8100:   10

col    0[2] @8102: 200
col    1[8] @8105: Jennifer
col    2[6] @8114: Whalen
col    3[7] @8121: JWHALEN
col   4[12] @8129: 515.123.4444
col    5[7] @8142: w....
col    6[7] @8150: AD_ASST
col    7[2] @8158:

                    col    8[0] @8161: *NULL*
col    9[3] @8162: .

出现ORA-00600:[13013], [5001]且Arg [f] Code =3