Handling Oracle Block Corruptions

If you cannot recover data by yourself, ask Parnassusdata, the professional ORACLE database recovery team for help.

Parnassusdata Software Database Recovery Team

Service Hotline:  +86 13764045638 E-mail: service@parnassusdata.com

Introduction

This article discusses how to handle one or more block corruptions on an Oracle datafile and describes the main actions to take to deal with them. Please read the complete article before taking any action. 
This note does not cover memory corruption issues (typically ORA-600 [17xxx] type errors).

Note: If the problem is an ORA-1578 on STARTUP then please contact Oracle Support Services for advice. 

This article can be referred from many places for many forms of error - it is important to have the following information for each corrupt block:

• An absolute FILE NUMBER of the file containing the corrupt block.
  Referred to as "&AFN" in this article.
• The file name of the file containing the corrupt block.
  Referred to as "&FILENAME" in this article.
  If the FILE NUMBER is known but not its name then V$DATAFILE can be used to get the file name:
  SELECT name FROM v$datafile WHERE file#=&AFN;

   If the file number does not appear in V$DATAFILE in Oracle8i
  AND &AFN is greater than the DB_FILES parameter value then it is probably a TEMPFILE. In this case the filename can be found using:

  SELECT name FROM v$tempfile WHERE file#=(&AFN - &DB_FILES_value); 
• The BLOCK NUMBER of the corrupt block in that file.
  Referred to as "&BL" in this article.
• The tablespace number and name containing the affected block.
  Referred to as "&TSN" (tablespace number) and "&TABLESPACE_NAME" in this article.
   If they are not known then they can be found using:
  SELECT ts# "TSN" FROM v$datafile WHERE file#=&AFN;
  SELECT tablespace_name FROM dba_data_files WHERE file_id=&AFN;
• The block size of the tablespace where the corruption lies.
  Referred to as "&TS_BLOCK_SIZE" in this article.
  SELECT block_size FROM dba_tablespaces 
  WHERE tablespace_name = 
  (SELECT tablespace_name FROM dba_data_files WHERE file_id=&AFN);

For Oracle 7, 8.0 and 8.1 every tablespace in the database has the same block size.  For these versions, execute "SHOW PARAMETER DB_BLOCK_SIZE" and use this value as the &TS_BLOCK_SIZE.

Example: For the ORA-1578 error:

ORA-01578: ORACLE data block corrupted (file # 7, block # 12698)
ORA-01110: data file 22: '/oracle1/oradata/V816/oradata/V816/users01.dbf'
then:

   For other errors (ORA-600 , ORA-1498 etc...) the above values should either be given by Oracle Support, or be given from the article which covers the relevant error.

   Some errors, such as ORA-1410 "invalid ROWID" , ORA-12899 "value too large for column" etc.., do not give details of the corrupt file / block. For such casesNote:869305.1 may help in locating the corrupt row.

{go to content}

Overview of Steps to handle a Corruption

There are many possible causes of a block corruption including: 

• • Bad IO hardware / firmware
  • OS problems
  • Oracle problems
  • Recovering through "UNRECOVERABLE" or "NOLOGGING" database actions
    (in which case ORA-1578 is expected behaviour - see below)

The point in time when an Oracle error is raised may be much later than when any corruption initially occurred.

As the root cause is not usually known at the time the corruption is encountered, and as in most cases the key requirement is to get up and running again, then the steps used tackle corruption problems in this article are:

1. Determine the extent of the corruption problems and also determine if the problems are permanent or transient.
   If the problem is widespread or the errors move about then focus on identifying the cause first (check hardware etc..). This is important as there is no point recovering a system if the underlying hardware is faulty.
2. Replace or move away from any faulty or suspect hardware.
3. Determine which database objects are affected.
4. Choose the most appropriate database recovery / data salvage option.

For all steps above it is sensible to collect evidence and document exactly what actions are being taken. The 'Evidence>>' tags in this article list the information which should be collected to assist with identifying the root cause of the problem.

Corruption due to NOLOGGING or UNRECOVERABLE

If a NOLOGGING (or UNRECOVERABLE) operation is performed on an object and the datafile containing that object is subsequently recovered then the data blocks affected by the NOLOGGING operation are marked as corrupt and will signal an ORA-1578 error when accessed.  
In Oracle8i and greater an ORA-26040 is also signalled ("ORA-26040: Data block was loaded using the NOLOGGING option" ) which makes the cause fairly obvious, but earlier releases have no additional error message. If a block is corrupt due to recovery through a NOLOGGING operation then use this article from  Section 3 "Information to Record for Each Corruption" onwards but note that:

1. 1. Recovery cannot retrieve the NOLOGGING data
   2. No data is salvagable from inside the block
2.  

3. Solution

4. For Solution reference Note:794505.1

(1) Determine the Extent of the Corruption Problem

Reference Note 836658.1 for identifying the corruption extension.  It is a good idea to scan affected files (and any important files) with RMAN or DBVERIFY to check for other corruptions in order to determine the extent of the problem. For details of using DBVERIFY see Note:35512.1

Whenever a corruption error occurs note down the FULL error message/s and look in the instance's alert log and trace files for any associated errors. It is important to do this first to assess whether this is a single block corruption, an error due to an UNRECOVERABLE operation or a more severe issue. 
 

Once that a list of corrupt file/block combinations has been determined then the steps below can be used to help in what action can be taken.

(2) Replace or Move Away from Suspect Hardware

The vast majority of corruption problems are caused by faulty hardware.
If there is a hardware fault or a suspect component then it is sensible to either repair the problem, or make disk space available on a separate disk sub-system prior to proceeding with a recovery option.

One option is to move the datafiles using the following steps:

• • Make sure the file to be relocated is either OFFLINE or the instance is in the MOUNT state (not open)
  • Physically restore (or copy) the datafile to its new location 
    example: /newlocation/myfile.dbf
  • Tell Oracle the new location of the file.
    example: ALTER DATABASE RENAME FILE '/oldlocation/myfile.dbf' TO '/newlocation/myfile.dbf';
        (Note that a TEMPFILE cannot be renamed - TEMPFILEs should be dropped and recreated at the new location)

 If using any special IO options such as direct IO, async IO or similar it may be worth disabling them in order to eliminate such options as a potential source of problems.

(3) Information to record for each Corruption

It is best to determine which objects are affected BEFORE making any decisions about how to recover - this is because the corruption/s may be on object/s which can easily be re-created.   
Example: For a corruption on a 5 row lookup table it may be far quicker to drop and recreate the table than to perform a recovery.

For each corruption collect the information in the following table. 
The steps to do this are explained below.

 The notes below will help to fill in this table for each corruption.

1.  "Original Error"This is the error as initially reported. Eg: ORA-1578 / ORA-1110 , ORA-600 with all arguments etc..
2. "Absolute File#", "Relative File#" and "Block#"The File# and Block# should have been given either by the error, by Oracle Support, or by the steps in an error article which directed to this article.
   In Oracle8/8i/9i/10g: The absolute and relative file numbers are often the same but can differ (especially if the database has been migrated from Oracle7). It is important to get the correct numbers for &AFN and &RFN to avoid salvaging the wrong object
   An ORA-1578 reports the RELATIVE file number, with the ABSOLUTE file number given in the accompanying ORA-1110 error. ORA-600 errors normally provide the absolute file number.
   The following query will show the absolute and relative file numbers for datafiles in the database:
   SELECT tablespace_name, file_id "AFN", relative_fno "RFN" FROM dba_data_files;

   In Oracle8i/9i/10g:
   In addition to the notes above about Oracle8, Oracle8i onwards can have TEMPFILES.  The following query will show the absolute and relative file numbers for tempfiles in the database:

   SELECT tablespace_name, file_id+value "AFN", relative_fno "RFN" FROM dba_temp_files, v$parameter WHERE name='db_files';

   In Oracle 7: Use the same file number for both the "Absolute File#" and the "Relative File#"

3. "Segment Type", "Owner", "Name" and "Tablespace"
   The following query provides the object TYPE , OWNER and NAME of a segment given the absolute file number "&AFN" and block number "&BL" of the corrupt block - the database must be open in order to use this query:
     
   SELECT *
   FROM dba_extents
   WHERE file_id = &AFN
   and &BL between block_id AND block_id + blocks - 1;

     
   If the block is in a TEMPFILE the above query will return no data.
   For TEMPFILES the "Segment Type" will be "TEMPORARY".

   If the above query does not return rows, it can also be that the corrupted block is a segment header in a Locally Managed Tablespace (LMT).  When the corrupted block is a segment header block in a LMT,  the above query produces a corruption message in the alert.log  but the query does not not fail.  In that case use this query:

   SELECT owner, segment_name, segment_type, partition_name 
   FROM dba_segments
   WHERE header_file = &AFN
   and header_block = &BL;

{go to content}

(4) Which Object is affected and possible Recovery options:

The related objects and recovery options which can be used depend on the SEGMENT_TYPE. The additional queries and possible recovery options are listed below for each of the most common segment types

CACHE        
CLUSTER     
INDEX PARTITION     
INDEX         
LOBINDEX         
LOBSEGMENT
ROLLBACK             
TABLE PARTITION        
TABLE
TEMPORARY
IOT
TYPE2 UNDO     
Some other Segment Type
"no rows" from the query

CACHE

    If the segment type is CACHE, recheck what have been entered and if the SQL / parameters are correct.
          
Solution:

Repair the Block with RMAN Block Media Recovery

    {Continue}  {Back to Segment List} 

CLUSTER

    If the segment is a CLUSTER determine which tables it contains.    

Solution:

Repair the Block with RMAN Block Media Recovery

OR 

Salvage data from all tables in the CLUSTER then recreate the CLUSTER and all its TABLES. If the OWNER is "SYS" then contact Oracle support with all details.

      As the CLUSTER may contain a number of TABLES, it is best to collect information for each table in the cluster before making a decision.

    {Collect TABLE information}  {Back to Segment List}

INDEX PARTITION

    If the segment is an INDEX PARTITION note the PARTITION NAME and OWNER and then determine which partition is affected, example:   

Solution:

Repair the Block with RMAN Block Media Recovery

OR

Index partitions can be rebuilt using: 
ALTER INDEX xxx REBUILD PARTITION ppp;    
     (take care with the REBUILD option as described in "Recreating Indexes" below)

    {Continue}  {Back to Segment List}

INDEX

    Find out which table the INDEX is on:

 Possible values for CONSTRAINT_TYPE are:
    -  P    The index supports a primary key constraint.
    -  U    The index supports a unique constraint.
    If the INDEX supports a PRIMARY KEY constraint (type "P") then check if the primary key is referenced by any foreign key constraints:

Solution:    

Repair the Block with RMAN Block Media Recovery

OR 

Recreate the Index (with any associated constraint disables/enables). If the OWNER is "SYS" then contact Oracle support with all details.
     

    {Continue}  {Back to Segment List}

ROLLBACK

If the segment is a ROLLBACK segment contact Oracle support as rollback segment corruptions require special handling.

Solution:

Repair the Block with RMAN Block Media Recovery

    {Continue}  {Back to Segment List}

TYPE2 UNDO

TYPE2 UNDO is a system managed undo segment which is a special form of rollback segment. Corruptions in these segments require special handling.

Solution:

Repair the Block with RMAN Block Media Recovery

    {Continue}  {Back to Segment List}

TABLE PARTITION

If the segment is a TABLE PARTITION note the NAME and OWNER and then determine which partition is affected thus:

then continue below as if the segment was a TABLE segment. 

Solution:

Repair the Block with RMAN Block Media Recovery

OR

If all corruptions are in the same partition then one option at this point is to EXCHANGE the corrupt partition with an empty TABLE - this can allow the application to continue (without access to the data in the corrupt partition) whilst any good data can then be extracted from the table.
For other options see the TABLE options below.

    {Continue}  {Back to Segment List}

TABLE

If the OWNER is "SYS" then contact Oracle support with all details.
The database is likely to require recovery.

For a non-dictionary TABLE or TABLE PARTITIONs find out which iNDEXES exist on the TABLE:

and determine if there is any PRIMARY key on the table: 

If there is a primary key then check if this is referenced by any foreign key constraints:

Solution:

Repair the Block with RMAN Block Media Recovery

OR

Salvage data from the TABLE (or PARTITION).  If the OWNER is "SYS" then contact Oracle support with all details.

THEN Recreate the table (or partition)
OR  Leave the corruption in place
(example: Use DBMS_REPAIR to mark the problem blocks to be skipped)

    {Continue}  {Back to Segment List}

IOT (INDEX ORGANIZED TABLE)

The corruption in IOT table should be handled in the same way as in a heap or partitioned table.
The only exception is if the PK (Primary Key) is corrupted.
PK of an IOT table is the table itself and cannot be dropped and recreated.

Solution:

Repair the Block with RMAN Block Media Recovery

OR

Salvage data from the TABLE (or PARTITION).  If the OWNER is "SYS" then contact Oracle support with all details.

THEN
Recreate the table (or partition)
OR  
Leave the corruption in place

(DBMS_REPAIR cannot be used with IOTs)

  {Continue}  {Back to Segment List}

LOBINDEX

FInd out which table the LOB belongs to:  

It is not possible to rebuild LOB indexes; treat the problem as a corruption on the LOB column of the affected table.
Get index and constraint information for the table which has the corrupt LOB index using the SQL in the TABLE section, then return here.

Solution:

Repair the Block with RMAN Block Media Recovery

OR

MOVE the LOB Segment which rebuilds the LOB index:

If the OWNER is "SYS" then contact Oracle support with all details.

    {Continue}  {Back to Segment List}

LOBSEGMENT

Find out which table the LOB belongs to:

For non-dictionary tables (Owner != SYS) ... 

Get index and constraint information for the table which has the corrupt LOB data using the SQL in the TABLE section, then return here to find details of the exact rows affected.

Finding the exact row which references the corrupt LOB block can be a challenge as the errors reported do not show any detail about which table row owns the lob entry which is corrupt.

Typically one can refer to application logs or any SQL_TRACE or 10046 trace of a session hitting the error (if available) or see if having 

 set in the session helps identify the current SQL/binds/row.

 Then wait for the error to be hit by the application and find the trace file. 

 If there are no clues then there is the option to construct a PLSQL block to scan the problem table row by row extracting the LOB column data which loops until it hits an error. Such a technique may take a while but it should be possible to get a primary key or rowid of any row which references a corrupt LOB block.

 another script more generic:

Solution:

Repair the Block with RMAN Block Media Recovery

OR

Update the lob column with empty_clob/empty_blob to avoid ORA-1578; this is cleaning the LOB column in the table - note that the LOB is already corrupt and cannot be accessed:

Setting the corrupt lob to empty lob will add the blocks formerly mapped to this lob to the freelist.  Eventually when PCTVERSION or RETENTION criteria causes the space to be salvaged and reused for new data, error ORA-1578  can be seen again in the same LOB blocks.  In that case and after applying the above procedure the lob segment can be moved to a new segment:

OR   Salvage data from the TABLE (and its LOB column/s)
THEN Recreate the table 
OR  Leave the corruption in place 
(It is not possible to use DBMS_REPAIR on LOB segments)

{Continue}  {Back to Segment List}

TEMPORARY

 If the segment type is TEMPORARY then the corruption does not affect a permanent object. Check if the tablespace where the problem occurred is being used as a TEMPORARY tablespace thus:

Solution:

Use Note 1332088.1

No restore is normally required, although if the disk is suspect and the tablespace contains useful data then a database recovery of the affected file/s may be wise.

{Continue}  {Back to Segment List}

Some Other SEGMENT_TYPE 

If the segment type returned is not covered above then contact Oracle support for advice with all information collected so far.

{Continue}  {Back to Segment List}

"NO ROWS SELECTED" 

If there appears to be no extent containing the corrupt block then first double check the figures used in the query. If sure that the file and block are correct and do not appear as belonging to an object in DBA_EXTENTS then:

• Double check if the file involved is a TEMPFILE.
  Note that TEMPFILE file numbers depend on the init.ora parameter DB_FILES so any changes to this parameter change the absolute file number reported in errors.
• DBA_EXTENTS does not include blocks which are used for local space management in locally managed tablespaces.
• If the database being queried is from a different point in time to the datafile with the error then the problem object may have been dropped and so queries against DBA_EXTENTS may show no rows.
• If the error being investigated was reported by DBVERIFY then DBV checks all blocks regardless of whether they belong to an object or not. It is possible for a corrupt block to exist in the datafile but in a block not in use by any segment.

Options:

An error on an UNUSED Oracle block can be ignored as Oracle will create a new block image should the block need to be used so any existing problem on the block will never get read.

If suspected that the block may be a space management block then use DBMS_SPACE_ADMIN to help check this by running: 

 This should write inconsistencies to the trace file but if it encounters a fatally corrupt block it will report an error like:

{Continue}  {Back to Segment List}

{go to content}

EVIDENCE

For each corrupt block it is also a good idea to collect the following physical evidence if there is a need to try and identify the actual cause of the corruption:

1. An operating system HEX dump of the bad block and the block either side of it.

   On UNIX: 

   dd if=&FILENAME bs=&TS_BLOCK_SIZE skip=&BL-1 count=3 of=BL.dd
      ^^^^^^^^      ^^^^^^^^^^^^^^         ^^^ 

   Example: For BL=1224:

   dd if=ts11.dbf bs=4k skip=1223 count=3 of=1223_1225.dd 

   On VMS:

   DUMP/BLOCKS=(start:XXXX,end:YYYY)/out=dump.out &FILENAME

   Where XXXX=Operating system block number (in 512 byte chunks)
   To calculate this multiply the block number reported by '&TS_BLOCK_SIZE/512'

2. If the database is in ARCHIVELOG mode, make a safe copy of the archived log files around the time of the error, and preferably for a few hours before the error was reported. Also secure any backup/s of the problem datafile from before the errors as the before image PLUS redo can help point towards a cause.
   (DBV can often be used to check if the problem exists in a backup copy of a file). The ideal scenario is to have a datafile backup image which does not have any corruption and all the redo from that point in time up to and just past the time when the corruption is first reported.
    
3. Obtain an Oracle dump of the problem block/s:
   ALTER SYSTEM DUMP DATAFILE '&FILENAME' BLOCK &BL;

   (The output will go to a tracefile in the USER_DUMP_DEST).

{Continue}  {Back to Segment List}

{go to content}

(5) Choosing A Recovery Option.

The best recovery option now depends on the objects affected. The notes in Section (3) above should have highlighted the main options available for each affected object. The actual recovery method chosen may include a mix or one or more methods thus:

Recovery Required ?

If the error is in a TEMPORARY tablespace, or is in a block which is no longer part of any database object then no action is required, although it may be wise to relocate the problem tablespace to a different storage device.

See Warnings

Is Complete Recovery an option ?

In order for complete recovery to be an option the following must be true:

• The database is in ARCHIVELOG mode 

  (The "ARCHIVE LOG LIST" command shows Archivelog Mode)

• There is a good backup of the affected files.  Note that in some cases, the corruption may have been present, but undetected, for a long period of time.  If the most recent datafile backup still contains the corruption, try an earlier backup as long as all the necessary ARCHIVELOGS are available. 
  (DBV with the START / END options can be used to check if specific block/s in a restored copy of a backup file are corrupt)

• All ARCHIVELOGS are available from the time of the backup to the current point in time

• The current online log/s are available and intact

• The errors are NOT due to recovery through a NOLOGGING operation 

When the above criteria are satisfied then complete recovery is usually the preferred option 

*BUT NOTE*

1. If the rollback of a transaction has seen a corrupt block on an object other than the rollback segment itself then UNDO may have been discarded. In this case the indexes may need to be rebuilt; check data integrity AFTER the recovery completes. 
    
2. If the files to be recovered contain data from NOLOGGING operations performed since the last backup then those blocks will be marked corrupt if datafile or database recovery is used. In some cases this can put the problem in a worse scenario than the current position.

If database recovery has already been performed and the corruption is still there then either all the backups contain the corruption, the underlying fault is still present or the problem is replaying through redo.  In these cases choose some other recovery option.

 See "(4A) Complete Recovery" for complete recovery steps.

{go to content}

Can the object be Dropped or Re-created without needing to extract any data from the object itself?

It may be possible to lose the object, or to recreate it from a script / recent export.  Once an object is dropped then blocks in that object are marked as "free" and will be re-formatted when the block gets allocated to a new object. It is advisable to RENAME rather than DROP a table unless absolutely sure that the data in it is not needed. 

In the case of a table partition then only the affected partition needs to be dropped.

If the corruption affects the partition segment header, or the file containing the partition header is offline, then DROP PARTITION may fail. In this case it may still be possible to drop the partition by first exchanging it with a table of the same definition. 

The most common object which can be re-created is an index. 
Always address TABLE corruptions before INDEX problems on a table. 
See "(4B) Recreating Indexes" for more details.

For any segment, a quick way to extract the DDL for an object, when the absolute file number and block number of the corrupt block are known, is:

{go to content}

Is it required to salvage data before recreating the object ?

If the problem is on a critical application table which is regularly updated then it may be required to salvage as much data from the table as possible, then recreate the table.

See "(5C) Salvaging Data from Tables" for more details.

{go to content}

Is it acceptable to leave the corruption in place for the moment?

In some cases the best immediate option may be to leave the corruption in place and isolate it from application access.

See "(5D) Leaving the Corruption In Place" for more details. 

{go to content}

Last Options

Are any of the following possible ?

Recovery to an old point-in-time (via point in time recovery)of either the database or tablespace point in time recovery

OR Restore of a COLD backup from before the corruption

OR Use of an existing export file

See "(5E) Last Options" for more details.

{go to content}

 (5A) Complete Recovery

If the database is in ARCHIVELOG mode and there is a good backup of the affected files then recovery is usually the preferred option.
This is not GUARANTEED to clear a problem, but is effective for the majority of corruption issues. If recovery re-introduces the problem then return to the list of options above and choose another method.

If using Oracle9i (or higher) then it may be possible to perform block level recovery using the RMAN BLOCKRECOVER command. 
If using an earlier Oracle release then either perform a Datafile Recovery (which can be done while the rest of the database is still up and running), or Database Recovery(which requires the database to be taken down) . 

If using Oracle 11g (or higher) then it may be possible to use Data Recovery Advisor.

RMAN Block Media Recovery 

As of Oracle9i RMAN allows individual blocks to be recovered whilst the rest of the database (including other blocks in the datafile) are available for normal access. Note that block level recovery can only be used to recover a block fully to the current point in time.

Example:

Consider having an ORA-1578 on file #6 block #30 which is likely due to a media corruption problem and there is a good cold backup image of that file which has been restored to '.../RESTORE/filename.dbf'.
Provided all archivelogs exist (in the default location) then RMAN can be used to perform a Block Media Recovery (BMR) using a command sequence like:

 This uses the registered datafile backup image and any required archivelogs to perform block recovery of just the one problem block to current point in time.

See the documentation for full details of the RMAN BLOCKRECOVER command and limitations. See Note 144911.1 for an example and Note 342972.1 to perform Block Media Recovery when backups are not taken by RMAN.

 {go to content}

Datafile Recovery

Datafile recovery of a file involves the following steps. If there are several files repeat the steps for each file or see "Database Recovery" below.  These steps can be used if the database is either OPEN or MOUNTED.

OFFLINE the affected data file:

Copy it to a safe location (in case the backup is bad)

Restore the latest backup of the file onto a GOOD disk

Check the restored file for obvious corruptions with DBVERIFY
    For details of using DBVERIFY see Note:35512.1

Assuming the restored file is OK, then RENAME the datafile to the NEW location (if different from the old location)

 Recover the datafile:

 Online the file/s

 {Continue}

 {go to content}

Database Recovery

Database recovery generally involves the following steps:

Copy the current copy of all files to be recovered to a safe location

Restore the backup files to a GOOD disk location
DO NOT RESTORE THE CONTROL FILES or ONLINE REDO LOG FILES

Check restored files with DBVERIFY. For details of using DBVERIFY see Note:35512.1

 Rename any relocated files:

Ensure all required files are online: 

 Recover the database:

 Open the database:

{go to content}

After a Complete Recovery

Once a complete recovery has been performed it is advisable to check the database before allowing it to be used:

Run: 

against each problem object to check for table/index mis-matches.

If there has been any UNDO discarded this may show a mismatch requiring indexes to be re-created.

Check the logical integrity of data in the table at the application level.

{go to content}

(5B) Recreating Indexes

If the corrupt object is a user INDEX, simply drop and re-create it PROVIDED the underlying table is not also corrupt.  
If the underlying table is also corrupt it is advisable to sort out the TABLE before recreating any indexes.

If the information collected shows that the index has dependent FOREIGN KEY constraints then something like this may be needed: 

•  for each foreign key
  ALTER TABLE &table_name DISABLE CONSTRAINT &pk_constraint;
• Rebuild the primary key using
  ALTER TABLE &table_name DISABLE CONSTRAINT &pk_constraint;
  DROP INDEX &index_name;
  CREATE INDEX &index_name .. with appropriate storage clause
  ALTER TABLE &table_name ENABLE CONSTRAINT &pk_constraint; 
• Enable the foreign key constraints
  ALTER TABLE &child_table ENABLE CONSTRAINT &fk_constraint;

For an index partition:

Notes:

1. It is important not to REBUILD a non-partitioned corrupt index using an "ALTER INDEX ..  REBUILD" command as this will usually try to build the new index from the existing index segment, which contains a corrupt block. 
   "ALTER INDEX ... REBUILD ONLINE" and "ALTER INDEX ... REBUILD PARTITION ..." 

   do not build the new index from the old index segment and so can be used.

2. Create INDEX can use the data from an existing index if the new index is a sub-set of the columns in the existing index. Hence if there are 2 corrupt indexes drop them BOTH before re-creating them.
3. Be sure to use the correct storage details when recreating indexes.

{go to content}

(5C) Salvaging Data from Tables

If the corrupt object is a TABLE or CLUSTER or LOBSEGMENT then it must be understood that the data within the corrupt block is lost. 
Some of the data may be salvageable from a HEX dump of the block, or from columns covered by indexes.

There are many ways to get data out of a table which contains a corrupt block. Choose the most appropriate method as detailed below. The aim of these methods is to extract as much data as  possible from the table blocks which can be accessed. It is  usually a good idea to RENAME the corrupt table so that the new object can be created with the correct name. 

Methods of extracting data from a corrupt table AROUND a corrupt block

1. From Oracle 7.2 onwards, it is possible to SKIP over corrupt blocks in a table.
   This is the simplest option to extract table data and is discussed in:

   Extracting data using DBMS_REPAIR.SKIP_CORRUPT_BLOCKS or Event 10231  Note:33405.1
   If the corruption is in an IOT overflow segment then the same method should be followed, but using event 10233 together with a full index scan or in 10.2.0.4 / 10.2.0.5 use Event:43810; in 11g onwards use parameter _index_scan_check_skip_corrupt=TRUE. 

   Note that this method can only be used if the block "wrapper" is marked corrupt. Example: If the block reports ORA-1578.
   If the problem is an ORA-600 or other error which does not report and ORA-1578 error then it is often possible to use DBMS_REPAIR to mark the problem blocks in a table as "soft corrupt" such that they will then signal ORA-1578 when accessed which then allows to use DBMS_REPAIR.SKIP_CORRUPT_BLOCKS

   Note: Any blocks which are marked corrupt by the "FIX_CORRUPT_BLOCKS" procedure will also be marked corrupt following any restore / recover operation through the time of the FIX_CORRUPT_BLOCKS. 

   Full details of using DBMS_REPAIR for this can be found in the documentation but in summary the steps are:

   - Use DBMS_REPAIR.ADMIN_TABLES to create the admin tables
   - Use DBMS_REPAIR.CHECK_OBJECT to find problem blocks
   - Get any good data out of problem blocks before corrupting them.
   - Use DBMS_REPAIR.FIX_CORRUPT_BLOCKS to mark the found problem
   blocks as corrupt so that they will then signal ORA-1578
   - If required use DBMS_REPAIR.SKIP_CORRUPT_BLOCKS to skip corrupt
   blocks on the table.

2. From Oracle 7.1 onwards use a ROWID range scan.
   The syntax for this is a little tricky but it is possible to select around a corrupt block using a ROWID hint.
   As the format of ROWIDs changed between Oracle7 and Oracle8 there are 2 articles which discuss this:

   Using ROWID Range Scans to extract data in Oracle8 and higher Note:61685.1
   Using ROWID Range Scans to extract data in Oracle7 Note:34371.1
    

3. If there is a primary key select table data via this index.
   It may also be possible to select some of data via any other index. 
   This can be slow and time consuming and is only normally needed for Oracle 7.0 releases. This method is described in Note:34371.1 (which also describes the ROWID range scans)
4. There are various salvage programs / PLSQL scripts which can be used to salvage data from a table. These can take longer to set up and use than the above methods but can often cope with various kinds of corruption besides an ORA-1578.
   As these methods typically require much hand-holding from support then some of these articles may not be visible to customers.

   These require Pro*C to be available and an understanding of how to build Pro*C executables:

   SALVAGE.PC for Oracle8.1 Note:97357.1
   SALVAGE.PC for Oracle7  Note:2077307.6

   These requires manual interaction:

   SALVAGE.SQL for Oracle7/8 Note:2064553.4
   SALVAGE.SQL for Oracle7/8 Note:28308.1
   The following is only possible in Oracle8i and 9i. The aim is to mark the block as corrupt and then use the SKIP_CORRUPT table attribute to extract the table data: 

   Use DBMS_REPAIR to mark the block corrupt Note:68013.1 
   For corruption in a LONG column: Recreating a Table with a corruption in a LONG Note:876493.1

{go to content}

Methods of extracting data from a table with a corrupt LOBSEGMENT block

It is not possible to used DBMS_REPAIR on LOB segments.

If the corrupt LOB block is NOT referenced by any row in the table then it should be possible to CREATE TABLE as SELECT (CTAS) or export / drop / import the table as is.

If the corrupt LOB block is referenced by a row then it should be possible to select or export with a WHERE predicate that excludes the problem row/s.

Extracting data from the corrupt block itself

As the corrupt block itself is "corrupt" then any data extracted from the block should be treated as suspect.  The main methods of getting the rows from the corrupt block itself are:

• For TABLE blocks Oracle Support can use a tool which attempts to interpret the block contents. 
  "Convert HEX or BLOCKDUMP to Readable form" Note:47419.1

• Use any existing indexes on the table to extract data for columns covered by the index where the ROWID falls inside the corrupt block. This is described towards the end of the ROWID range scan articles mentioned above:
  For Oracle8/8i see Note:61685.1
  For Oracle7 see Note:34371.1

• It may be possible to use LogMiner on the redo stream to find the original inserts/updates which loaded the data to the problem block. The main factor here is WHEN the data was actually put in the block.
  example: row 2 may have been inserted yesterday but row 1 may have been inserted 5 years ago.

 {go to content}

(5D) Leaving A Corruption In Place

It is possible to leave a corruption in place and just accept the errors reported, or prevent access to the problem rows at an application level.

Example: If the problem block / row is in a child table then it may be possible at application level to prevent access via the parent row/s such that the child rows are never accessed. (Be wary of cascade type constraints though)

This may not help with reports and other jobs which access data in bulk so it may also be desirable to use the DBMS_REPAIR options shown in 4C above to prevent the block/s erroring when accessed. Marking a corruption like this and leaving it around may  give a short term solution allowing full data salvage and/or recovery  to be attempted at scheduled outage, or allowing time to check other recovery options on a second (clone) database. Note though that marking a block corrupt withDBMS_REPAIR.FIX_CORRUPT_BLOCKS will cause the marked block/s to also be corrupt after recovery through the  time that FIX_CORRUPT_BLOCKS was executed.

Leaving a corruption may be sensible for data which rapidly ages and is subsequently purged (example: In a date partitioned table where older partitions are dropped at some point).

Leaving Corruptions in LOB segments

At application level it can be possible to leave a corrupt LOB column in place until such time as the table can be rebuilt.

One way to ensure to not hit the "WARNING" scenario above is to ensure that the table is only ever accessed via a view which includes a WHERE predicate to prevent the problem row/s from being seen.
example: Consider table MYTAB( a number primary key, b clob ) has one or more rows pointing at corrupt LOB data.

Set BAD='Y' for any problem row/s

If MYTAB is only accessed via MYVIEW, the row will never be visible and cannot be updated; keeping the corrupt entry isolated until it can be dealt with.

Clearly this example is more of a design time solution but some applications may already have similar mechanisms and may only access data via a view (or via an RLS policy) giving some option/s to hide the problem rows.

{go to content}

Warnings when Leaving a Corruption in Place

Whilst it is possible to leave a corruption in place it should be noted that the corrupt blocks will still show up in runs of DBVERIFY, in RMAN backup warnings / errors etc.. 

It is important to make a careful record of any corruption expected to be seen from these tools, particularly any blocks expected to be skipped with RMAN (having MAX_CORRUPT set) and be sure to remove any "acceptance" of the errors once the corruptions have been cleared.

Example: Consider that a corrupt block has been handled by leaving the corruption in place and avoiding the problem row/s at application level.
RMAN may be configured to allow the corruptions during backup.
The table is then recreated at a later date during some table reorganisation.
If RMAN is not updated to reflect that no errors should now be expected then RMAN may ignore some other corruption which occurs at a later time.

It is also important to note that leaving corrupt blocks around in table segments can lead to mismatched results from queries; different results can occur for tables with SKIP_CORRUPT set depending on whether an index scan or table access occurs. Other reports may just error .

Note that leaving a corruption in place but marking the block with DBMS_REPAIR.FIX_CORRUPT_BLOCKS writes redo to corrupt the block which may limit subsequent recovery options.

{go to content}

(5E) Last Options

If there is a STANDBY setup (physical or logical) then check that first.
Whatever sort of block the problem occurred on, one possible option is to recover the database, or problem tablespace, to a point in time BEFORE the corruption appeared. The difficulty with this option is that it is not always possible to know when the problem first appeared.

DBVERIFY can be often be used to check a restored file for corruptions.
For details of using DBVERIFY see Note:35512.1 . In particular the START/END DBV options can be used to give a quick first test of whether the problem block itself is bad on a restored backup image.

This section outlines some final options available for recovering.
If arrived here then one or more of the following have happened:

• A "vital" datafile has been lost (or have a corruption on it) and have no good backup of the problem file/s (without the corruption)
• The database is either not in ARCHIVELOG mode OR do not have all archivelogs since the file was first created
• Complete recovery keeps reintroducing the problem

Last chance:

Please note that if all copies of a datafile have been lost but DO still have the ARCHIVE logs from when the file was first created it is still possible to recover the file.

If currently in this scenario try to recover the datafile using these steps before proceeding below.

If reached this line there are no options left to recover to the current point in time.  It is advisable to shutdown the instance and take a BACKUP of the current database NOW in order to provide a fall-back  position if the chosen course of action fails. (Example: if identified that the backup is bad). 

Some outline options available are:

Revert to an old COLD backup
- example: If in NOARCHIVELOG mode

Set up a clone database from a COLD backup and extract (export) the problem table/s or transport the problem tablespace

Point in time recovery to an older point in time that is consistent:

• requires a good backup and any necessary archive logs
• ALL files have to be restored and the whole DB rolled forward to a suitable point in time.
• It may be possible to do the point in time recovery in a clone database and then transport the problem tablespace to the problem database, or export / import the problem table from the clone to the problem database . 

Tablespace point in time recovery
- It may be possible to perform a point in time recovery of the affected tablespace only. There are many notes describing tablespace point in time recovery such asNote:223543.1. 

Rebuild of DB from some logical export / copy
- Requires there to already be a good logical backup of the database
- NB: The database has to be RECREATED for this option.
- As with other options the rebuild could be in a clone database just to get a good image of the problem table/s.

If a good backup is available then rolling forwards with DB_BLOCK_CHECKING=TRUE can help find the first point in time where something started to go wrong. 
It is not generally necessary to take the problem database down while investigating the recovery options.