Malicious Injection — It's Not Just For SQL Anymore

nywanna writes "When most people think of malicious injection, they think of SQL injection. The fact is, if you are using XML documents or an LDAP directory, you are just as vulnerable to a malicious injection as you would be using SQL. Bryan Sullivan looks at the different types of malicious code injections and examines the very basics of preventing these injections."

Comments

pr0n

[macadamia_harold]

When most people think of malicious injection, they think of SQL injection.

Come on now, considering your audience, you might want to re-think that statement.

Re:pr0n

[spellraiser]

Yeah. Malicious Injection was a pretty good flick. I can't wait for Malicious Injection: The SQL.

Old news

[kill-1]

Shell scripts have been vulnerable to similar "injection" exploits since the invention of CGI.

More old news

[spellraiser]

From TFA:

The only real way to defend against all malicious code injection attacks is to validate every input from every user.

Seems simple enough, but it's amazing how often this is ignored or implemented badly.

Re:More old news

[Vihai]

The only real way to defend against all malicious code injection attacks is to validate every input from every user.

Seems simple enough, but it's amazing how often this is ignored or implemented badly.

...but unfortunately it's mostly WRONG. In many cases, the real way to defend against injections is to ESCAPE values before composing strings, this is mostly the case with SQL (where prepared queries and the help of a good prepare/execute API is very much helpful) but it's not limited to it.

If your parameter is a VALUE, it must remain a VALUE when you compose a command and proper escaping is the correct, reliable way.

Validating input may be helpful as another layer of security, but it's not the "only right way", it's not even the *right* way (in most cases).

Re:More old news

[J0nne]

3) Do not send SQL parameters to your page in GET statements!!!!!! Either use session variables or POST statements, session variables are best.

Using POST instead of GET doesn't make *any* difference. You can fake a POST request just as easily as a GET request. Please stop telling people that a POST is more secure...

Re:More old news

[jrockway]

> Here's my tips for preventing SQL injection.

Here are mine that aren't garbage:

> 1) Use stored procedures!!!!!!!

"EXEC dbo.stored_procedure 'Oops'; DROP DATABASE foo; --'"

> 2) escape your escape characters. i.e. in most statments a "'" is stored as "\'" so escape the \ so its stored as "\\'", it will invalidate the SQL statment because SQL will read it as "\'" instead of just "'"

Not sure what you're talking about, but a literal apostrophe is quoted by doubling it in SQL. ' -> ''. However, don't quote -- you'll get it wrong. Use a proper mechanism instead, like prepared statements.

> 2.5) an alternate to escaping characters is to just strip characters unnecessary to be passed with your stored procedure. i.e. strip all quotes, strip all double quotes, strip all equals signs, bash signs, etc.

That's a great idea, until you need to store unicode or have a customer named "O'Reilly".

> 3) Do not send SQL parameters to your page in GET statements!!!!!! Either use session variables or POST statements, session variables are best.

Right, there's no way anyone can see hidden form fields! They're magical! (Also, session variables aren't "best". If you find the need to store SQL in a variable, your program is terribly designed and you need to rethink it. In this day and age of stored procedures and ORMs, you probably shouldn't have ANY SQL in your code.)

4) You should be secure, but if your not comfortable doing that, then provide additional validation.

Always validate -- it saves work later. If a user types 2-1234 as his phone number, and you store that, you won't be able to call him later, completely defeating the purpose of asking him for the data.

If you're not sure that you can remember to validate everything, use a language that taints incoming data and kills the program when you use it. In perl, turning on taint mode will prevent the common pattern of:

my $value = CGI->param('foo');
$dbh->do("SELECT * FROM foo WHERE bar = $value");

and even:

$sth = $dbh->prepare('SELECT * FROM foo WHERE bar = ?');
$sth->execute($value);

Since you didn't validate $value, you can't use it (correctly or incorrectly).

Hope this helps.

Re:More old news

[TheRealBurKaZoiD]

Wrong. For God's sake, you dweeby little junior programmers need to do some research before you open your damn mouths.

Escaping values only brings up new vulnerabilities. In database servers these are known as SQL truncation, and are the byproduct of buffer overruns in system functions (such as QUOTENAME and PATINDEX in T-SQL). These truncation attacks affect even parameterized queries, and the ubiquitous sp_executesql system stored procedure. I won't go into the details. All the info you need is is BOL, so look it up for yourself.

The real problem (again, in database servers) is dynamic SQL. That, and incorrect security permissions (some dbas need to be beaten with a stick), but I'm not here to teach a SQL Administration class. Many, many dynamic SQL statements, because they are only filtering the data set, can be re-written as a non-dynamic SQL stored procedure, but still afforded their dynamic nature. For example, this vulnerable stored procedure:

create proc sp_get_product
@prod varchar(255) = null
as
declare @sql varchar(1000)
set @sql = 'SELECT PRODUCTID, PRODUCTNAME, CATEGORY, PRICE FROM PRODUCTION'
if @prod is not null set @sql = @sql + ' PRODUCTNAME LIKE ''' + @PRODNAME + '''
exec(@sql)

can be re-written to:

create proc sp_get_product
@prod varchar(255) = null
as
select productid, productname, category, price
from product
where productname like coalesce(@prod, productname)

Coalesce returns the first non-null argument in it's argument list. If @prod has a value, the resulting dataset is filtered on it. If @prod is null, then the entire dataset will be returned because productname will filter on it's own value. Only near-infinite where clauses can't be done this way.

Also, production database server accounts should have no access to any database objects. They should only be able to run stored procedures, and that's it.

You're advice about "escaping" values is just as bad as those computer science professors in college that say "re-writing your conditional statements to automatically exclude bad data will eliminate the need to validate the data that makes it through."

Goddammit. Comments (and articles) like this make me so fucking mad.

My rock-solid solution to the injection problem...

[Anonymous Coward]

...is to replace database storage, xml, and ldap with comma-delimited text files on anonymous ftp. In fact, my last job fired me for gross incompetence because the other programmers were jealous of the simplicity of my solution.

XML Logic Is Flawed

[CowboyBob500]

In his XML example with XPath injection he states that running a certain query can return the entire order history of all customers. That may be true, but if the application is returning an XML document containing the entire order history of all customers for each customer request before running an XPath query, then I think the application has more problems than being vulnerable to XPath injection.

Bob

Re:XML Logic Is Flawed

[Ant P.]

Yeah, but XPath can be done server-side just like SQL.

Re:XML Logic Is Flawed

[CowboyBob500]

But not over an XML representation of the entire damn customer orders table. That's insane.

Bob

Validate this

[gigne]

FTA
RE: validating input fields...

To be completely thorough, a developer should set up both white- and blacklists in order to cover all bases.

I can't help but feel that most developers have at least a little common sense and do something along those lines anyway.
I often write little validate_input(char *string, char *format) that checks all input string from a user are simple, but more often than not very effective. How is this any different from using white and black lists. Any coder worth their salt would do something to stop malicious input, but no one in completely infallible.

Security of anything in this world is near on impossible. Hackers will get around anything given time.

Re:Validate this

[thsths]

>> To be completely thorough, a developer should set up both white- and blacklists in order to cover all bases.
> I can't help but feel that most developers have at least a little common sense and do something along those lines anyway.

I hope that most developers have the common sense to take the correct approach: avoid injection problems by proper quoting! There is no need to validate the data, you just have to make sure that it stays data when you parse it on. Just use the proper library functions, and you will be fine (especially if you use hex encoding :-)).

White lists are a good idea if you don't trust you quoting, or if you need to verify the input for another reason. Black lists are most certainly not a good idea. Just imagine that the web shop tries to sell a product called "Selecta[tm]", but you block all attempts to buy it because it matches your black word "SELECT" :-(

P.S.: Anybody with an apostrophe in their name naturally develops an unsatisfiable urge to kill web programmers.

It may sound trite, but...

[PHAEDRU5]

I blame Microsoft for a lot of these vulnerabilities.

I recently attended a Microsoft-sponsored seminar on web site security at the DeVry Institute in Decatur, GA.

One of the speakers was a man from SPI Dynamics (sorry, forgot his name). He demonstrated how Microsoft's tools make it very easy to expose a database to the web, but how the same tools make exploiting the database very easy. He demonstrated an application that used SQL injection to first reverse-engineer the schema of an exposed database, and then the data in the database. It was quite a revelation.

Re:It may sound trite, but...

[bdigit]

Um you can just as easily reverse engineer a mysql or postgresql database through sql injection attacks also. What's your point?

If you only knew the POWER of languages

[Sloppy]

Heh, remember when we had binary file formats and protocols, fixed-length fields (didn't need delimiters), and there was no parsing or worrying about "escaping" data? We didn't have these problems.

Anyway, I like this article because it admits that whitelists are better than blacklists. You have to validate data: make sure it is known to be non-harmful, rather than looking for whatever problems that you have imagined so far. You'll never guess all the things that can go wrong; you just know what is right.

Phishers like frame injections

[miller60]

Phishers have been known to use frame injections to insert their content into framesets, allowing them to grab login info from within the bank's own web site. It's not nearly as fancy as an SQL injection, but it's sure malicious and quite difficult for victims to recognize.

Defensive Programming

[davidwr]

"It's not just for breakfast anymore."

Email header injection attack

[DeadSea]

Another example of an injection attack allow an attacker to send spam through a contact form that doesn't normally allow the recipient to be specified by the user.

A webmaster hosts a contact form on his website that allows users to fill out a form to contact him. He allows the user to specify a subject and a message but the recipient is hard coded to webmaster@example.com.

The message ends up looking like this:

To: webmaster@example.com
From: thewebserver@example.com
Subject: $subject

$message

Where $subject and $message are captured from the user on the website.

If the $subject is not properly sanitized, a bot could submit it with a new line in it and be able to start a new line in the headers of the email. That new line could be, for example, a large CC list of people to spam with his message:

Buy my weight loss pills!
CC: spammee1@example.com, spammee2@example.com

Which is why I would suggest using a contact form such as the one that I have written that has already thought about this sort of thing.

Wash it before you eat it

[Jhan]

It's a simple matter of hygiene:

Wash it before you eat it.

All data read from external sources must be validated before being used. In some languages/frameworks this is as hard as nails (ie. I programmed a pretty large web application with only straight CGI programs written in pure Unix/C), in some you have help (Perl with taint), in some it's kinda-sorta-almost not an issue (PHP with Agavi and Creole).

If I had to choose, I would have to say that the middle way, the Perl way, is the best. It does not pretend to solve all your problems for you, even when it can't really. Rather it brings the problems at hand to your attention. Problems surface, fix problems, code gets better.

Re:Ignorance

[bluebox_rob]

I think you're right - as long as you are sure that you know what's going to be done with the data after its been written away to your database. You might have your escaping/quoting routine solidly implemented for all inputs to your system, but the trainee down the hall who writes the reporting application that parses the table once a month might not be so savvy - the cunningly crafted SQL injection attack that your quoting has preserved and saved away into the db could wreak havoc when it gets read out again at the other end. The same goes for any HTML/XML that has been saved away, and then gets blindly written out by a web developer on the Order Summary page, or merged into some larger XML document without proper checks.

I suppose an apt analogy would be saying that it's ok to allow infectious material into a building as long as it is first correctly sealed in a bio-safe container - well that's true as long as you're sure the janitor isn't going to open it up later that evening and use it for a cookie jar.

Comment removed

[account_deleted]

Comment removed based on user account deletion

How I'd do stuff

[TheLink]

So far everyone seems to be focusing on "input" and forgetting about "output", or even mixing the two.

Anyway, my suggestion has always been to do something like the following:

Inputs to your program
|||
Corresponding Input filters
|||
Your program
|||
Corresponding Output filters
|||
Outputs from your program
|||
Stuff receiving the outputs

You have a different "input filter" for each class of input so that your program can handle those inputs correctly.

Then you have a different output filter (e.g. SQL bind vars, HTML, XML) so that the stuff receiving your outputs (browser, database, viewer, etc) will handle them correctly.

NEVER do stuff like magic quotes (PHP is one of the worst and most braindead language in popular use) - mixing input and output filtering is so wrong it isn't funny (there are so many other things PHP does wrong that it's almost criminal).

Depending on the circumstances your program could output a single quote ' differently e.g. %27 for a cgi parameter, '' for Oracle data and \' for MySQL data (BTW MySQL is the PHP of databases). So it should be obvious that "one size fits all" doesn't work.

By filtering I mean quoting/encoding sanity checking etc - whatever it takes to get the data in a suitable form (with hopefully minimal data loss/corruption).