views:

1755

answers:

1

Sorry for the long question but this contains all the SQL I've used to test the scenario to hopefully make it clear as to what I'm doing.

I'm build up some dynamic SQL to produce a PIVOT table in SQL Server 2005.

Below is code to do this. With various selects showing the raw data the values using GROUP BY and the values in a PIVOT as I want them.

BEGIN TRAN
--Create the table
CREATE TABLE #PivotTest
(
 ColumnA nvarchar(500),
 ColumnB nvarchar(500),
 ColumnC int
)

--Populate the data
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'X', 1)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'Y', 2)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'Z', 3)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'X', 4)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'Y', 5)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('B', 'Z', 6)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('B', 'X', 7)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('B', 'Y', 8)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('B', 'Z', 9)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('C', 'X', 10)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('C', 'Y', 11)
INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('C', 'Z', 12)

--The data
SELECT * FROM #PivotTest

--Group BY
SELECT
 ColumnA,
 ColumnB,
 SUM(ColumnC)
FROM
 #PivotTest
GROUP BY
 ColumnA,
 ColumnB

--Manual PIVOT
SELECT
 *
FROM
 (
  SELECT
   ColumnA,
   ColumnB,
   ColumnC
  FROM
   #PivotTest
 ) DATA
 PIVOT
 (
  SUM(DATA.ColumnC)
 FOR
  ColumnB
  IN
  (
   [X],[Y],[Z]
  )
 ) PVT

--Dynamic PIVOT
DECLARE @columns nvarchar(max)

SELECT
 @columns = 
 STUFF
 (
  (
   SELECT DISTINCT
    ', [' + ColumnB + ']'
   FROM
    #PivotTest
   FOR XML PATH('')
  ), 1, 1, ''
 )

EXEC
('
 SELECT
  *
 FROM
  (
   SELECT
    ColumnA,
    ColumnB,
    ColumnC
   FROM
    #PivotTest
  ) DATA
  PIVOT
  (
   SUM(DATA.ColumnC)
  FOR
   ColumnB
   IN
   (
    ' + @columns + '
   )
  ) PVT
')

--The data again
SELECT * FROM #PivotTest

ROLLBACK

Anytime that I produce any dynamic SQL I'm always aware of SQL Injection attacks. Therefore I've added the following line with the other INSERT statements.

INSERT INTO #PivotTest (ColumnA, ColumnB, ColumnC) VALUES('A', 'FOO])) PVT; DROP TABLE #PivotTest;SELECT ((GETDATE()--', 1)

When I now run the SQL, low and behold, the EXEC part drops the #PivotTest table thus making the last SELECT fail.

So my question is, does anyone know of a way to perform a dynamic PIVOT without risking SQL Injection attacks?

+4  A: 

We've done a lot of work similar to your example. We haven't worried about SQL injenction, in part because we have complete and total control over the data being pivoted--there's just no way malicious code could get through ETL into our data warehouse.

Some thoughts and advice:

  • Are you required to pivot with nvarcahr(500) columns? Ours are varchar(25) or numerics, and it would be pretty hard to sneak damaging code in through there.
  • How about data checking? Seems like if one of those strings contained a "]" character, it's either a hack attempt or data that will blow up on you anyway.
  • How robust is your security? Is the system locked down such that Malorey can't sneak his hacks into your database (either directly or through your application)?

Hah. It took writing all that to remember function QUOTENAME(). A quick test would seem to indicate that adding it to your code like so would work (You'll get an error, not a dropped temp table):

SELECT
        @columns = 
        STUFF
        (
                (
                        SELECT DISTINCT
                                ', [' + quotename(ColumnB, ']') + ']'
                        FROM
                                #PivotTest
                        FOR XML PATH('')
                ), 1, 1, ''
        )

This should work for pivot (and unpivot) situations, since you almost always have to [bracket] your values.

Philip Kelley
1) My test sample is a simple one. The actual columns are nvarchar(max). We have no data that size at present and the data that would be used for the PIVOT would rarely be as much as 100 so I may perform a forced truncate in this instance! Great Idea.2) I was thinking about the '[' and ']'. I'm tempted to strip all square brackets from the data and just have that as a limitation of this functionality.3) The only people that can add this data are so called "Super Users", however, this is not enough to give me peace of mind.
Robin Day
QUOTENAME! First time I've seen it! Perfect! This completely solves the problem. I was adding the QUOTES manually. If I remove this and do it using QUOTENAME it will disable any SQL within that field! THANKYOU!
Robin Day