假设我们的数据库遇到以下情况：
第一, 没有备份;
第二, 常规方法无法恢复;
第三, 数据很重要, 但又无法或成本太高而进行重新输入. 如丢失了Oracle的System表空间, System表空间损坏到无法启动的地步, 意外删除表空间或表, 意外截断(Truncate)表等,
在这3中情况下， 最后的方法就是通过工具直接读取数据文件里的数据，将我们的数据找回来。并且工具不需要Oracle 环境的支持。
据我目前的了解，有3种工具：
（1）Oracle 的内部工具是DUL（Data UnLoader）。 这个需要Oracle 的支持。
（2）老熊写的ODU。 网址：http://www.oracleodu.com/en/
（3）d.c.b.a （支付宝 楼方鑫）写的AUL. 网址：http://www.anysql.net/download
ODU 之前是免费的， 现在老熊和dbsnake 在维护ODU，需要购买才能使用。
d.c.b.a的AUL 是用C 语言写的， 免费版本最大只支持2个，最大256M的datafile。 如果是更大的datafile，也是需要购买授权。
能写出这样的软件都是牛人了，需要了解各个版本Oracle block的详细信息。 AUL 和 ODU的操作和Oracle DUL 类似。
关于DUL,MOS 有说明: MOS Note 72554.1
Using DUL to Recover fromDatabase Corruption
Table Of Contents
~~~~~~~~~~~~~~~~~
1. Introduction
2. Using DUL
2.1 Create an appropriate init.dul file
2.2 Create the control.dul file
2.3 Unload the object information
2.4 Invoke DUL
2.5 Rebuild the database
3. How to rebuild object definitions that are stored in the data dictionary ?
4. How to unload data when the segment header block is corrupted ?
5. How to unload data when the file header block is corrupted ?
6. How to unload data without the system tablespace ?
7. Appendix A : Where to find the executables ?
8. References
1. Introduction
~~~~~~~~~~~~~~~
This document is to explain how to use DUL rather than to give a full
explanation of Bernard's Data UnLoader capabilities.
This document is for internal use only and should not be given to customers at
any time, Dul should always be used by or under the supervision of a support
analyst.
DUL (Data UnLoader) is intended to retrieve data from the Oracle Database that
cannot be retrieved otherwise. This is not an alternative for the export
utility or SQL*Loader. The database may be corrupted but an individual data
block used must be 100% correct. During all unloading checks are made to make
sure that blocks are not corrupted and belong to the correct segment. If a
corrupted block is detected by DUL, an error message is printed in the loader
file and to the standard output, but this will not terminate the unloading of
the next row or block.
2. Using DUL
~~~~~~~~~~~~
First you must retrieve the necessary information about the objects that exists
in the database, these statistics will be loaded into the DUL dictionary to
unload the database objects.
This information is retrieved from the USER$, OBJ$, TAB$ and COL$ tables that
were created at database creation time, they can be unloaded based on the fact
that object numbers are fixed for these tables due to the rigid nature of sql.
bsq. DUL can find the information in the system tablespace, therefor the system
tablespace datafile(s) must be included in the control file, if this datafile(s)
is not present see chapter 6.
2.1 Create an appropriate "init.dul" file
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
REM Platform specific parameters (NT)
REM A List of parameters for the most common platforms can be obtained from
REM http://www.nl.oracle.com/support/dul/index.html
osd_big_endian_flag=false
osd_dba_file_bits=10
osd_c_struct_alignment=32
osd_file_leader_size=1
osd_word_size = 32
REM Sizes of dul dictionary caches. If one of these is too low startup will
REM fail.
dc_columns=2000000
dc_tables=10000
dc_objects=1000000
dc_users=400
dc_segments=100000
REM Location and filename of the control file, default value is control.dul
REM in the current directory
control_file = D:\Dul\control_orcl.dul
REM Database blocksize, can be found in the init<SID>.ora file or can be
REM retrieved by doing "show parameter %db_block_size%" in server manager
REM (svrmgr23/30/l) changes this parameter to whatever the block size is of
REM the crashed database.
db_block_size=4096
REM Can/must be specified when data is needed into export/import format.
REM this will create a file suitable to use by the oracle import utility,
REM although the generated file is completely different from a table mode
REM export generated by the EXP utility. It is a single table dump file
REM with only a create table structure statement and the table data.
REM Grants, storage clauses, triggers are not included into this dump file !
export_mode=true
REM Compatible parameter must be specified an can be either 6, 7 or 8
compatible=8
REM This parameter is optional and can be specified on platforms that do
REM not support long file names (e.g. 8.3 DOS) or when the file format that
REM DUL uses "owner_name.table_name.ext" is not acceptable. The dump files
REM will be something like dump001.ext, dump002.ext, etc in this case.
file = dump
A complete list can be obtained at http://www.nl.oracle.com/support/DUL/ucg8.
html section "DUL Parameters" although this init.dul file will work in most
cases and contains all accurate parameters to succesfully complete the
unloading.
2.2 Create the "control.dul" file
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
A good knowledge about the logical tablespace and physical datafile
structure is needed or you can do the following queries when the database
is mounted :
Oracle 6, 7
-----------
> connect internal
> spool control.DUL
> select * from v$dbfile;
> spool off
Oracle 8
--------
> connect internal
> spool control.DUL
> select ts#, rfile#, name from v$datafile;
> spool off
Edit the spool file and change, if needed, the datafile location and stripe
out unnecessary information like table headers, feedback line, etc...
A sample control file looks something like this :
REM Oracle7 control file
1 D:\DUL\DATAFILE\SYS1ORCL.DBF
3 D:\DUL\DATAFILE\DAT1ORCL.DBF
7 D:\DUL\DATAFILE\USR1ORCL.DBF
REM Oracle8 control file
0 1 D:\DUL\DATAFILE\SYS1ORCL.DBF
1 2 D:\DUL\DATAFILE\USR1ORCL.DBF
1 3 D:\DUL\DATAFILE\USR2ORCL.DBF
2 4 D:\DUL\DATAFILE\DAT1ORCL.DBF
Note : Each entry can contain a part of a datafile, this can be useful when
you need to split datafiles that are too big for DUL, so that each
part is smaller than for example 2GB. For example :
REM Oracle8 with a datafile split into multiple parts, each part is
REM smaller than 1GB !
0 1 D:\DUL\DATAFILE\SYS1ORCL.DBF
1 2 D:\DUL\DATAFILE\USR1ORCL.DBF startblock 1 endblock 1000000
1 2 D:\DUL\DATAFILE\USR1ORCL.DBF startblock 1000001 endblock 2000000
1 2 D:\DUL\DATAFILE\USR1ORCL.DBF startblock 2000001 endblock 2550000
2.3 Unload the object information
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Start the DUL utility with the appropriate ddl (Dul Description Language)
script. There are 3 scripts available to unload the USER$, OBJ$, TAB$ and
COL$ tables according to the database version.
Oracle6 :> dul8.exe dictv6.ddl
Oracle7 :> dul8.exe dictv7.ddl
Oracle8 :> dul8.exe dictv8.ddl
Data UnLoader: Release 8.0.5.3.0 - Internal Use Only - on Tue Jun 22 22:19:
Copyright (c) 1994/1999 Bernard van Duijnen All rights reserved.
Parameter altered
Session altered.
Parameter altered
Session altered.
Parameter altered
Session altered.
Parameter altered
Session altered.
. unloading table OBJ$ 2271 rows unloaded
. unloading table TAB$ 245 rows unloaded
. unloading table COL$ 10489 rows unloaded
. unloading table USER$ 22 rows unloaded
. unloading table TABPART$ 0 rows unloaded
. unloading table IND$ 274 rows unloaded
. unloading table ICOL$ 514 rows unloaded
. unloading table LOB$ 13 rows unloaded
Life is DUL without it
This will unload the data of the USER$, OBJ$, TAB$ and COl$ data dictionary
tables into SQL*Loader files , this can not be manipulated into dump files
of the import format. The parameter export_mode = false is hardcoded into
the ddl scripts and can not be changed to the value "true" since this will
cause DUL to fail with the error:
. unloading table OBJ$
DUL: Error: Column "DATAOBJ#" actual size(2) greater than length in column
definition(1)
.............etc...............
2.4 Invoke DUL
~~~~~~~~~~~~~~
Start DUL in interactive mode or you can prepare a scripts that contains all
the ddl commands to unload the necessary data from the database. I will
describe in this document the most used commands, this is not a complete list
of possible parameters that can be specified. A complete list can be found at
http://www.nl.oracle.com/support/DUL/ucg8.html section "DDL Description".
DUL> unload database;
=> this will unload the entire database tables(includes sys'tables as well)
DUL> unload user <username>;
=> this will unload all the tables owned by that particullarly user.
DUL> unload table <username.table_name>;
=> this will unload the specified table owned by that username
DUL> describe <owner_name.table_name>;
=> will represent the table columns with there relative pointers to the
datafile(s) owned by the specified user.
DUL> scan database;
=> Scans all blocks of all data files.
Two files are generated:
1: seg.dat information of found segment headers (index/cluster/table)
(object id, file number, and block number).
2: ext.dat information of contiguous table/cluster data blocks.
(object id(V7), file and block number of segment header (V6),
file number and block number of first block,
number of blocks, number of tables)
DUL> scan tables;
=> Uses seg.dat and ext.dat as input.
Scans all tables in all data segments (a header block and at least one
matching extent with at least 1 table).
2.5 Rebuild the database
~~~~~~~~~~~~~~~~~~~~~~~~
Create the new database and use import or SQL*Loader to restore the data
retrieved by DUL. Note that when you only unloaded the data that table
structures, indexation, grants, PL/SQL and triggers will no longer exist in
the new database. To obtain an exactly same copy of the database as before
you will need to rerun your creation scripts for the tables, indexes, PL/SQL,
etc.
If you don't have these scripts then you will need to perform the steps
described in section 3 of this document.
3. How to rebuild object definitions that are stored in the data dictionary
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
You want to rebuild PL/SQL(packages, procedures, functions or triggers), grants,
indexes, constraints or storage clauses(old table structure) with DUL. This can
be done but is a little bit tricky. You need to unload the relevant data
dictionary tables using DUL and then load these tables into a healthy database,
be sure to use a different user than sys or (system). Loading the data
dictionary tables of the crashed database into the healthy database dictionary
could corrupt the healthy database as well.
Detailed explanation to retrieve for example pl/sql packages / procedures /
functions from a corrupted database :
1) Follow the steps explained in the "Using DUL" section and unload the data
dictionary table "source$"
2) Create a new user into a healthy database and specify the desired default
and temporary tablespace.
3) Grant connect, resource, imp_full_database to the new user.
4) Import/load the table "source$" into the new created schema:
e.g.: imp80 userid=newuser/passw file=d:\dul\scott_emp.dmp
log=d:\dul\impemp.txt full=y
5) You can now query from the table <newuser.source$> to rebuild the pl/sql
procedures/functions from the corrupted database. Scripts can be found on
WebIv to generate such PL/SQL creation scripts.
The same steps can be followed to recreate indexes, constraints, and storage
parameters or to regrant privileges to the appropiate users. Please notice that
you always need to use a script of some kind that can recreate the objects and
include all the features of the crashed database version. For example : when
the crashed database is of version 7.3.4 and you have several bitmap indexes,
if you would use a script that supports version 7.3.2 or prior, then you won't
be able to recreate the bitmap indexes succesful !
4. How to unload data when the segment header block is corrupted
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When DUL can't retrieve data block information on the normal way, it can scan
the database to create its own segment/extent map. The procedure of scanning
the database is necessary to unload the data from the datafiles.
(to illustrate this example I copied an empty block ontop of the segment header
block)
1) Create an appropiate "init.dul" (see 2.1) and "control.dul" (see 2.2) file.
2) Unload the table. This will fail and indicate that there is a corruption in
the segment header block:
DUL> unload table scott.emp;
. unloading table EMP
DUL: Warning: Block is never used, block type is zero
DUL: Error: While checking tablespace 6 file 10 block 2
DUL: Error: While processing block ts#=6, file#=10, block#=2
DUL: Error: Could not read/parse segment header
0 rows unloaded
3) run the scan database command :
DUL> scan database;
tablespace 0, data file 1: 10239 blocks scanned
tablespace 6, data file 10: 2559 blocks scanned
4) Indicate to DUL that it should use its own generated extent map rather than
the segment header information.
DUL> alter session set use_scanned_extent_map = true;
Parameter altered
Session altered.
DUL> unload table scott.emp;
. unloading table EMP 14 rows unloaded
5. How to unload data when the datafile header block is corrupted
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
A corruption in the datafile header block is always listed at the moment you
open the database this is not like a header segment block corruption (see point
4) where the database can be succesfully openend and the corruption is listed
at the moment you do a query of a table. Dul has no problems with recovering
from such situations although there are other alternatives of recovering from
this situation like patching the datafile header block.
The error you will receive looks something like :
ORACLE instance started.
Total System Global Area 11739136 bytes
Fixed Size 49152 bytes
Variable Size 7421952 bytes
Database Buffers 4194304 bytes
Redo Buffers 73728 bytes
Database mounted.
ORA-01122: database file 10 failed verification check
ORA-01110: data file 10: 'D:\DATA\TRGT\DATAFILES\JUR1TRGT.DBF'
ORA-01251: Unknown File Header Version read for file number 10
6. How to unload data without the system tablespace
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If datafiles are not available for the system tablespace the unload can still
continue but the object information can't be retrieved from the data dictionary
tables USER$, OBJ$, TAB$ and COL$. So ownername, tablename and columnnames will
not be loaded into the DUL dictionary. Identifying the tables can be an
overwhelming task and a good knowledge of the RDBMS internals are needed here.
First of all you need a good knowledge of your application and it's tables.
Column types can be guessed by DUL, but table and column names will be lost.
Any old system tablespace from the same database (may be weeks old) can be a
great help !
1) Create the "init.dul" file and the "control.dul" file as explained in above
steps 1 and 2. In this case the control file will contain all the datafiles
from which you want to restore but it doesn't require the system tablespace
information.
2) Then You invoke dul and type the following command :
DUL> scan database;
data file 6 1280 blocks scanned
This will build the extent and segment map. Probably the dul command
interpreter will be terminated as well.
3) reinvoke the dul command interpreter and do the following :
Data UnLoader: Release 8.0.5.3.0 - Internal Use Only - on Tue Aug 03 13:33:
Copyright (c) 1994/1999 Oracle Corporation, The Netherlands. All rights res
Loaded 4 segments
Loaded 2 extents
Extent map sorted
DUL> alter session set use_scanned_extent_map = true;
DUL> scan tables; (or scan extents;)
Scanning tables with segment header
Oid 1078 fno 6 bno 2 table number 0
UNLOAD TABLE T_O1078 ( C1 NUMBER, C2 UNKNOWN, C3 UNKNOWN )
STORAGE ( TABNO 0 EXTENTS( FILE 6 BLOCK 2));
Colno Seen MaxIntSz Null% C75% C100 Num% NiNu% Dat% Rid%
1 4 2 0% 0% 0% 100% 100% 0% 0%
2 4 10 0% 100% 100% 100% 0% 0% 0%
3 4 8 0% 100% 100% 100% 0% 0% 50%
"10" "ACCOUNTING" "NEW YORK"
"20" "RESEARCH" "DALLAS"
"30" "SALES" "CHICAGO"
"40" "OPERATIONS" "BOSTON"
Oid 1080 fno 6 bno 12 table number 0
UNLOAD TABLE T_O1080 ( C1 NUMBER, C2 UNKNOWN, C3 UNKNOWN, C4 NUMBER,
C5 DATE, C6 NUMBER, C7 NUMBER, C8 NUMBER )
STORAGE ( TABNO 0 EXTENTS( FILE 6 BLOCK 12));
Colno Seen MaxIntSz Null% C75% C100 Num% NiNu% Dat% Rid%
1 14 3 0% 0% 0% 100% 100% 0% 0%
2 14 6 0% 100% 100% 100% 0% 0% 21%
3 14 9 0% 100% 100% 100% 0% 0% 0%
4 14 3 7% 0% 0% 100% 100% 0% 0%
5 14 7 0% 0% 0% 0% 0% 100% 0%
6 14 3 0% 0% 0% 100% 100% 0% 0%
7 14 2 71% 0% 0% 100% 100% 0% 0%
8 14 2 0% 0% 0% 100% 100% 0% 0%
"7369" "SMITH" "CLERK" "7902" "17-DEC-1980 AD 00:00:00" "800" "" "20"
"7499" "ALLEN" "SALESMAN" "7698" "20-FEB-1981 AD 00:00:00" "1600" "300" "30"
"7521" "WARD" "SALESMAN" "7698" "22-FEB-1981 AD 00:00:00" "1250" "500" "30"
"7566" "JONES" "MANAGER" "7839" "02-APR-1981 AD 00:00:00" "2975" "" "20"
"7654" "MARTIN" "SALESMAN" "7698" "28-SEP-1981 AD 00:00:00" "1250" "1400" "30"
Note : it might be best that you redirect the output to a logfile since
commands like the "scan tables" can produce a lot of output.
On Windows NT you can do the following command :
C:\> dul8 > c:\temp\scan_tables.txt
scan tables;
exit;
4) Identify the lost tables from the output of step 3; if you look carefully to
the output above then you will notice that the unload syntax is already given
but that the table name will be of the format t_0<objectno> and the column
names will be of the format C<no>; datatypes will not be an exact match of
the datatype as it was before.
Look especially for strings like "Oid 1078 fno 6 bno 2 table number 0" where:
oid = object id, will be used to unload the object
fno = (data)file number
bno = block number
5) Unload the identified tables with the "unload table" command :
DUL> unload table dept (deptno number(2), dname varchar2(14),
loc varchar2(13)) storage (OBJNO 1078)
Unloading extent(s) of table DEPT 4 rows
