The correct way to avoid SQL injection attacks, no matter which database you use, is to separate the data from SQL, so that data stays data and will never be interpreted as commands by the SQL parser. It is possible to create SQL statement with correctly formatted data parts, but if you don't fully understand the details, you should always use prepared statements and parameterized queries. These are SQL statements that are sent to and parsed by the database server separately from any parameters. This way it is impossible for an attacker to inject malicious SQL.
You basically have two options to achieve this:
Using PDO (for any supported database driver):
$stmt = $pdo->prepare('SELECT * FROM employees WHERE name = :name');
$stmt->execute([ 'name' => $name ]);
foreach ($stmt as $row) {
// Do something with $row
}
Using MySQLi (for MySQL):
$stmt = $dbConnection->prepare('SELECT * FROM employees WHERE name = ?');
$stmt->bind_param('s', $name); // 's' specifies the variable type => 'string'
$stmt->execute();
$result = $stmt->get_result();
while ($row = $result->fetch_assoc()) {
// Do something with $row
}
If you're connecting to a database other than MySQL, there is a driver-specific second option that you can refer to (for example, pg_prepare()
and pg_execute()
for PostgreSQL). PDO is the universal option.
Correctly setting up the connection
Note that when using PDO to access a MySQL database real prepared statements are not used by default. To fix this you have to disable the emulation of prepared statements. An example of creating a connection using PDO is:
$dbConnection = new PDO('mysql:dbname=dbtest;host=127.0.0.1;charset=utf8', 'user', 'password');
$dbConnection->setAttribute(PDO::ATTR_EMULATE_PREPARES, false);
$dbConnection->setAttribute(PDO::ATTR_ERRMODE, PDO::ERRMODE_EXCEPTION);
In the above example the error mode isn't strictly necessary, but it is advised to add it. This way the script will not stop with a Fatal Error
when something goes wrong. And it gives the developer the chance to catch
any error(s) which are throw
n as PDOException
s.
What is mandatory, however, is the first setAttribute()
line, which tells PDO to disable emulated prepared statements and use real prepared statements. This makes sure the statement and the values aren't parsed by PHP before sending it to the MySQL server (giving a possible attacker no chance to inject malicious SQL).
Although you can set the charset
in the options of the constructor, it's important to note that 'older' versions of PHP (before 5.3.6) silently ignored the charset parameter in the DSN.
Explanation
The SQL statement you pass to prepare
is parsed and compiled by the database server. By specifying parameters (either a ?
or a named parameter like :name
in the example above) you tell the database engine where you want to filter on. Then when you call execute
, the prepared statement is combined with the parameter values you specify.
The important thing here is that the parameter values are combined with the compiled statement, not an SQL string. SQL injection works by tricking the script into including malicious strings when it creates SQL to send to the database. So by sending the actual SQL separately from the parameters, you limit the risk of ending up with something you didn't intend.
Any parameters you send when using a prepared statement will just be treated as strings (although the database engine may do some optimization so parameters may end up as numbers too, of course). In the example above, if the $name
variable contains 'Sarah'; DELETE FROM employees
the result would simply be a search for the string "'Sarah'; DELETE FROM employees"
, and you will not end up with an empty table.
Another benefit of using prepared statements is that if you execute the same statement many times in the same session it will only be parsed and compiled once, giving you some speed gains.
Oh, and since you asked about how to do it for an insert, here's an example (using PDO):
$preparedStatement = $db->prepare('INSERT INTO table (column) VALUES (:column)');
$preparedStatement->execute([ 'column' => $unsafeValue ]);
Can prepared statements be used for dynamic queries?
While you can still use prepared statements for the query parameters, the structure of the dynamic query itself cannot be parametrized and certain query features cannot be parametrized.
For these specific scenarios, the best thing to do is use a whitelist filter that restricts the possible values.
// Value whitelist
// $dir can only be 'DESC', otherwise it will be 'ASC'
if (empty($dir) || $dir !== 'DESC') {
$dir = 'ASC';
}
It's a common misconception that user input can be filtered. PHP even has a (now deprecated) "feature", called magic-quotes, that builds on this idea. It's nonsense. Forget about filtering (or cleaning, or whatever people call it).
What you should do, to avoid problems, is quite simple: whenever you embed a a piece of data within a foreign code, you must treat it according to the formatting rules of that code. But you must understand that such rules could be too complicated to try to follow them all manually. For example, in SQL, rules for strings, numbers and identifiers are all different. For your convenience, in most cases there is a dedicated tool for such an embedding. For example, when you need to use a PHP variable in the SQL query, you have to use a prepared statement, that will take care of all the proper formatting/treatment.
Another example is HTML: If you embed strings within HTML markup, you must escape it with htmlspecialchars
. This means that every single echo
or print
statement should use htmlspecialchars
.
A third example could be shell commands: If you are going to embed strings (such as arguments) to external commands, and call them with exec
, then you must use escapeshellcmd
and escapeshellarg
.
Also, a very compelling example is JSON. The rules are so numerous and complicated that you would never be able to follow them all manually. That's why you should never ever create a JSON string manually, but always use a dedicated function, json_encode()
that will correctly format every bit of data.
And so on and so forth ...
The only case where you need to actively filter data, is if you're accepting preformatted input. For example, if you let your users post HTML markup, that you plan to display on the site. However, you should be wise to avoid this at all cost, since no matter how well you filter it, it will always be a potential security hole.
Best Answer
The article mentioned by sgbj in the comments written by Google's Paul Turner explains the following in much more detail, but I'll give it a shot:
As far as I can piece this together from the limited information at the moment, a retpoline is a return trampoline that uses an infinite loop that is never executed to prevent the CPU from speculating on the target of an indirect jump.
The basic approach can be seen in Andi Kleen's kernel branch addressing this issue:
It introduces the new
__x86.indirect_thunk
call that loads the call target whose memory address (which I'll callADDR
) is stored on top of the stack and executes the jump using a theRET
instruction. The thunk itself is then called using the NOSPEC_JMP/CALL macro, which was used to replace many (if not all) indirect calls and jumps. The macro simply places the call target on the stack and sets the return address correctly, if necessary (note the non-linear control flow):The placement of
call
in the end is necessary so that when the indirect call is finished, the control flow continues behind the use of theNOSPEC_CALL
macro, so it can be used in place of a regularcall
The thunk itself looks as follows:
The control flow can get a bit confusing here, so let me clarify:
call
pushes the current instruction pointer (label 2) to the stack.lea
adds 8 to the stack pointer, effectively discarding the most recently pushed quadword, which is the last return address (to label 2). After this, the top of the stack points at the real return address ADDR again.ret
jumps to*ADDR
and resets the stack pointer to the beginning of the call stack.In the end, this whole behaviour is practically equivalent to jumping directly to
*ADDR
. The one benefit we get is that the branch predictor used for return statements (Return Stack Buffer, RSB), when executing thecall
instruction, assumes that the correspondingret
statement will jump to the label 2.The part after the label 2 actually never gets executed, it's simply an infinite loop that would in theory fill the instruction pipeline with
JMP
instructions. By usingLFENCE
,PAUSE
or more generally an instruction causing the instruction pipeline to be stall stops the CPU from wasting any power and time on this speculative execution. This is because in case the call to retpoline_call_target would return normally, theLFENCE
would be the next instruction to be executed. This is also what the branch predictor will predict based on the original return address (the label 2)To quote from Intel's architecture manual:
Note however that the specification never mentions that LFENCE and PAUSE cause the pipeline to stall, so I'm reading a bit between the lines here.
Now back to your original question: The kernel memory information disclosure is possible because of the combination of two ideas:
Even though speculative execution should be side-effect free when the speculation was wrong, speculative execution still affects the cache hierarchy. This means that when a memory load is executed speculatively, it may still have caused a cache line to be evicted. This change in the cache hierarchy can be identified by carefully measuring the access time to memory that is mapped onto the same cache set.
You can even leak some bits of arbitrary memory when the source address of the memory read was itself read from kernel memory.
The indirect branch predictor of Intel CPUs only uses the lowermost 12 bits of the source instruction, thus it is easy to poison all 2^12 possible prediction histories with user-controlled memory addresses. These can then, when the indirect jump is predicted within the kernel, be speculatively executed with kernel privileges. Using the cache-timing side-channel, you can thus leak arbitrary kernel memory.
UPDATE: On the kernel mailing list, there is an ongoing discussion that leads me to believe retpolines don't fully mitigate the branch prediction issues, as when the Return Stack Buffer (RSB) runs empty, more recent Intel architectures (Skylake+) fall back to the vulnerable Branch Target Buffer (BTB):