forensically deleting/updating data

Question

I have a need to forensically remove data from oracle. If I just delete it, my understanding is the data will still actually be in the data file until that space is reused. I'm not concerned about the redo/archive/undo space, those will age out reasonable quickly.

Is there any methods for ensuring data is actually removed from a data file?

Answer 1

This is an interesting question: When does Oracle really delete data physically ?

The unit of data in Oracle is a block. Let's see what happens when we delete a row.

Here's an example with a simple table on 11gR2 (see "How to dump Oracle Data Block?"):

CREATE TABLE test_delete_data(id NUMBER,data VARCHAR2(100));
INSERT INTO test_delete_data VALUES (1, rpad('1', 100, '1'));
INSERT INTO test_delete_data VALUES (2, rpad('2', 100, '2'));
INSERT INTO test_delete_data VALUES (3, rpad('3', 100, '3'));
COMMIT;

SELECT dbms_rowid.rowid_to_absolute_fno(rowid, user, 'TEST_DELETE_DATA') fileno,
       dbms_rowid.rowid_block_number(rowid) blockno
  FROM test_delete_data;

-- replace with values from query
alter system dump datafile 4 block 16573;

You should get something like this at the end of the file created in your user_dump_dest directory:

data_block_dump,data header at 0x8b02264
===============
[...]
block_row_dump:
tab 0, row 0, @0x1f2d
tl: 107 fb: --H-FL-- lb: 0x1  cc: 2
col  0: [ 2]  c1 02
col  1: [100]
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
tab 0, row 1, @0x1ec2
tl: 107 fb: --H-FL-- lb: 0x1  cc: 2
col  0: [ 2]  c1 03
col  1: [100]
 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32
 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32
 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32
 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32 32
tab 0, row 2, @0x1e57
tl: 107 fb: --H-FL-- lb: 0x1  cc: 2
col  0: [ 2]  c1 04
col  1: [100]
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
end_of_block_dump

If I delete the second row, commit and dump the same block, I will get something like this:

block_row_dump:
tab 0, row 0, @0x1f2d
tl: 107 fb: --H-FL-- lb: 0x0  cc: 2
col  0: [ 2]  c1 02
col  1: [100]
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
tab 0, row 1, @0x1ec2
tl: 2 fb: --HDFL-- lb: 0x2 
tab 0, row 2, @0x1e57
tl: 107 fb: --H-FL-- lb: 0x0  cc: 2
col  0: [ 2]  c1 04
col  1: [100]
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33 33
end_of_block_dump

The record is still there (with a D flag set). If we look at the actual binary data (just before the block header dump section, we see that the data has not yet been overwritten:

8B040C0 33336404 33333333 33333333 33333333  [.d33333333333333]
8B040D0 33333333 33333333 33333333 33333333  [3333333333333333]
        Repeat 4 times
8B04120 33333333 023C3333 03C10202 32323264  [333333<.....d222]
8B04130 32323232 32323232 32323232 32323232  [2222222222222222]
        Repeat 5 times
8B04190 02002C32 6402C102 31313131 31313131  [2,.....d11111111]
8B041A0 31313131 31313131 31313131 31313131  [1111111111111111]
        Repeat 4 times
8B041F0 31313131 31313131 31313131 30A30602  [111111111111...0]

One way to force data to actually be overwritten would be to update it to a meaningless value before deleting the row. This wouldn't work with indexes since updates are translated to delete+insert in a b*tree index.

Answer 2

I don't think data persists after a delete but you're correct that the space will be kept in use by that table until it has been refilled. The top space usage in a table is known as the high water mark. Tom Kyte has a really good (unsurprisingly) post about it.

You reduce the high water mark by rebuilding the table:

alter table my_table_name move

or by truncating it; though in an active table this is obviously not an option.