Wednesday, 3 November 2010

Beware of the Byte

Recently our test department raised a bug against one our applications that occurred when trying to insert a record into a table.

The error message encountered was a fairly innocuous "ORA-01704: string literal too long".

Following the test case to the letter, I successfully generated the same error and located the table that the APEX form was inserting into. A quick check of the Data Dictionary confirmed that the column in question was of type VARCHAR2(10). At this stage, I though the obvious cause was that there was no limit on the APEX form item (a Text Area) of 10 characters. Having checked the item in question, not only was there a “maxWidth” value of 10, the text area had been created with a “Character Counter”. Strange then how a form item accepting 10 characters was erroring whilst inserting into a column of VARCHAR2(10).

A little while later...... (after some head scratching and several discussions with our DBA’s and a colleague) the problem was all too clear. Somewhere between Database Character Sets, VARCHAR2 column definitions and non ASCII characters lay the answer.

Please forgive the rather verbose narrative but allow me to delve a little deeper.

Firstly the facts:

1. The character set of our database is set to AL32UTF8

FROM v$nls_parameters

2. Definition of table causing the error:

col1 VARCHAR2(10 BYTE)

3. SQL Statement causing the error:

INSERT INTO nls_test VALUES ('““““““““““');

NB: 10 individual characters.

The character used in this insert is typical of a double quote produced by MS Word (it was in fact a copy and paste from a Word document into our Text Area that caused our error).


The reason we encountered the error was all to do with the attempt to insert a 'Multi Byte' character (a double quote from word in our case) into our table as opposed to typical single byte characters (A-Z, 0-9 etc).

Performing a simple LENGTHB to return the number of bytes this character uses demonstrated this perfectly:

SELECT lengthb('“') from dual;


Because our column definition is of type VARCHAR2(10 BYTE), we are only permitted to store values that do not exceed 10 bytes in length.

Beware, 1 character in our case definitely does not = 1 byte. As already proved, our single character (a Word double quote) occupies 3 bytes so the maximum number of this Multi Byte Character we could possible insert according to our table definition is worked out simply as:

10 bytes (Column Definition) / 3 (length in bytes of our character) = 3

So whilst the APEX form item does not distinguish between single and multi byte characters and will allow you to input the full 10 characters, Oracle Database will bounce it back in our case as the total bytes in our string is 30 hence the error.

One solution suggested was to alter the Data Type to be of type VARCHAR2(10 CHAR) instead of BYTE. This in theory would force the database to respect the actual number of characters entered and not worry too much about single vs. multi byte occupancy. This would allow us resolve our immediate issue of 10 multi byte characters inserting into our table however there are further considerations.

As it turns out, even when you define your columns to use CHAR over BYTE, Oracle still enforces a hard limit of up to 4000 BYTES (given a mixed string of single and multi byte characters, it implicitly works out the total bytes of the string).

So beware that even if your column for example accepts only 3000 CHAR and you supply 2001 multi byte characters in an insert statement, it may still fail as it will convert your sting into BYTES enforcing an upper limit of 4000 BYTES.

Sorry for the really long post but it was a much for my own documentation as anything else.

Tuesday, 23 February 2010

APEX - Identify Report Columns Vulnerable to XSS

The following query is a very simple way of identifying all report columns within your APEX application that may be exposed by Cross Site Scripting (XSS).

XSS allows an attacker to inject web script (JavaScript) into an application and when this is rendered in the report, the script is interpreted rather than rendered as text.

To safe guard against this attack, APEX provides a "Display as Text (escape special characters)" report column attribute that can be applied to classic and Interactive Reports. This causes the script text to be displayed as text rather than interpreted by the browser. If you have any markup (HTML) within your query that the report is based on, this markup will also be displayed as text and not interpreted. I personally think this is a good by product as you should not really be coding look and feel into your raw SQL.

Anyway I digress. Here is the query that will identify all vulnerable report columns within your APEX application:

SELECT application_id,
FROM apex_application_page_rpt_cols
WHERE display_as NOT IN
('Display as Text (escape special characters, does not save state)',
'Text Field')
AND workspace != 'INTERNAL'
AND application_id = :APP_ID
ORDER BY 1, 2, 3;


Friday, 19 February 2010

Oracle SQL Developer on OS X Snow Leopard

I have been using Oracle SQL Developer Data Modeller for a while now within a Windows XP environment. It seems pretty good (albeit a little slow but hey show some an Oracle Java client application that is quick. Oracle Directory Manager?, OWB Design Centre? I shall labour this point no more) and I was looking forward to trying it out on my new 27" iMac.

I promptley downloaded the software from OTN and a quick read of the instructions suggested I need to do no more other than run the shell script since I already had Java SE 6 installed.

As it turns out, the script in the root location does little more than call another script called found in the /datamodeler/bin directory which is the once you actually need to execute to fire up SQL Data Modeler

When this script runs, it prompts you for a the full J2SE file path (which I had no idea where it was) before it will run. After a quick look around google and I came across the command: java_home which when executed like:

cd /usr/libexec

prints the full path value that you need to open SQL Data Modeler

e.g. /System/Library/Frameworks/JavaVM.framework/Versions/1.6.0/Home

Now we are armed with the full path needed, opening up SQL Data Modeller from a virgin command window goes like this:

cd Desktop/datamodeler/bin
. ./

Oracle SQL Developer Data Modeler
Copyright (c) 1997, 2009, Oracle and/or its affiliates.All rights reserved.

Type the full pathname of a J2SE installation (or Ctrl-C to quit), the path will be stored in ~/jdk.conf

and hey presto, SQL Data Modeller is up and running.

Once you have pointed the shell script at your J2SE installation, you wont have to do it again.

Now I can finally use Data Modeler on my 27" Screen :)