I have a user table with and id field and 10 other fields storing user details of various types that the user can change via various web forms. I want to have a PHP script that gets POSTed changed values for some subset of these fields, and UPDATEs only those fields that are received in the POST data. I'm finding this surprisingly difficult to do in a way that doesn't suck. We use mysqli for all database interaction in the rest of this application so mysqli-based solutions are strongly preferred.
The options I've considered and dismissed so far:
1) Run a separate UPDATE query for every field provided in the POST data - yuck, I don't want to hit the database up to 10 times for something that could be done in one query.
2) Have a dictionary mapping field names to the fields' data types, and iteratively construct the query by looping through the provided fields, checking whether they are text fields or not, calling mysqli_real_escape_string on the string fields and otherwise sanitizing the others (e.g. by type checking or sprintf with '%i' placeholders). - Yuck! I could probably safely do things this way if I was careful, but I don't want to make a habit of using this kind of approach because if I'm careless I'll leave myself open to SQL injection. Parameterized queries don't give me the potential to screw up dangerously, but this approach does. My ideal is to never concatenate any data into an SQL query manually and always rely upon parameterized queries; the database libraries of other languages, like Python, let me easily do this.
3) Use a parameterized query - this is my ideal for everything, since as long as I insert all externally-provided data into my query via the bind_param method of a mysqli statement object, I'm immune to SQL injection and don't have to worry about sanitization, but using parameterized queries seems to be impossible here. The trouble is that bind_param requires that the data be passed as variables, since all arguments after the first are passed by reference. I can reasonably elegantly iteratively construct a query with ? placeholders, and while I'm at it construct the string of types that gets passed as the first argument to bind_param ('ssiddsi' etc.), but then there's no way I can see to choose at runtime which of my 10 fields I pass to bind_params (unless I have a switch statement with 10^2 cases).
Is there some PHP language construct I'm missing (something similar to array unpacking) that will allow me to choose at runtime which variables to pass as arguments to bind_param? Or is there some other approach I haven't considered that will let me solve this simple problem cleanly and safely?
You can easily combine 2 and 3 by means of my SafeMySQL library.
The code will look like
$allowed = array('title','url','body','rating','term','type');
$data = $db->filterArray($_POST,$allowed);
$sql = "UPDATE table SET ?u WHERE id=?i";
$db->query($sql, $data, $_POST['id']);
note that $allowed array doesn't make all these fields necessarily updated - it just filters POST fields out. So, even $_POST with only id and url would be correctly updated.
Nevertheless, using prepared statements, although toilsome, also quite possible.
See the code below
public function update($data, $table, $where) {
$data_str = '' ;
foreach ($data as $column => $value) {
//append comma each time after first item
if (!empty($data_str)) $data_str .= ', ' ;
$data_str .= "$column = $value" ;
}
$sql = "UPDATE $table SET $data_str WHERE $where";
mysqli_query($sql) or die(mysqli_error());
return true;
}
$data is an array, in your case it's $_POST.
If you want to be more specific about the data to be saved from $_POST array, you can define an array of allowed columns. For example,
$allowed = array('id', 'username', 'email', 'password');
By doing this, you can filter your $_POST array and pass it to update() function.
Related
I have a loop that allocates values to all params set within a bind param. My problem is I'm not quite sure how to dynamically set the param type when I bind them.
For instance, here is the loop:
foreach(array_combine($cols,$vals) as $col => $val) {
$f->bindParam(':'.$col,$val,PDO::PARAM_INT); // loop through columns and values array and bind those params
}
however - if $vals contains a string for one of the values in the array, this breaks (obviously).
How do I set the PARAM_INT dynamically depending on the type of value in $vals?
I've considered using gettype() but that seemed like it would be messy and ugly... :)
Don't bother with binding at all - just send everything right into execute:
$pdo->prepare($sql)->execute(array_combine($cols,$vals));
Don't forget to turn emulation mode off though, as otherwise this method will choke on LIMIT clause parameters.
however, keep in mind that such a dynamical query building you are trying to implement is prone to SQL injection and apply the the appropriate protection.
I want to make a "dynamic" WHERE clause in my query based on a array of strings. And I want to run the created query using Mysqi's prepared statements.
My code so far, PHP:
$searchArray = explode(' ', $search);
$searchNumber = count($searchArray);
$searchStr = "tags.tag LIKE ? ";
for($i=1; $i<=$searchNumber-1 ;$i++){
$searchStr .= "OR tags.tag LIKE ? ";
}
My query:
SELECT tag FROM tags WHERE $searchStr;
More PHP:
$stmt -> bind_param(str_repeat('s', count($searchArray)));
Now this obviously gives me an error since the bind_param part only contains half the details it need.
How should I proceed?
Are there any other (better) way of doing this?
Is it secure?
Regarding the security part of the question, prepared statements with placeholders are as secure as the validation mechanism involved in filling these placeholders with values up. In the case of mysqli prepared statements, the documentation says:
The markers are legal only in certain places in SQL statements. For example, they are allowed in the VALUES() list of an INSERT statement (to specify column values for a row), or in a comparison with a column in a WHERE clause to specify a comparison value.
However, they are not allowed for identifiers (such as table or column names), in the select list that names the columns to be returned by a SELECT statement, or to specify both operands of a binary operator such as the = equal sign. The latter restriction is necessary because it would be impossible to determine the parameter type. It's not allowed to compare marker with NULL by ? IS NULL too. In general, parameters are legal only in Data Manipulation Language (DML) statements, and not in Data Definition Language (DDL) statements.
This clearly excludes any possibility of modifying the general semantic of the query, which makes it much harder (but not impossible) to divert it from its original intent.
Regarding the dynamic part of your query, you could use str_repeat in the query condition building part, instead of doing a loop:
$searchStr = 'WHERE tags.tag LIKE ?' .
str_repeat($searchNumber - 1, ' OR tags.tag LIKE ?');
For the bind_param call, you should use call_user_func_array like so:
$bindArray[0] = str_repeat('s', $searchNumber);
array_walk($searchArray,function($k,&$v) use (&$bindArray) {$bindArray[] = &$v;});
call_user_func_array(array($stmt,'bind_param'), $bindArray);
Hopefully the above snippet should bind every value of the $bindArray with its corresponding placeholder in the query.
Addenum:
However, you should be wary of two things:
call_user_func_array expects an integer indexed array for its second parameter. I am not sure how it would behave with a dictionary.
mysqli_stmt_bind_param requires its parameters to be passed by reference.
For the first point, you only need to make sure that $bindArray uses integer indices, which is the case in the code above (or alternatively check that call_user_func_array doesn't choke on the array you're providing it).
For the second point, it will only be a problem if you intend to modify the data within $bindArray after calling bind_param (ie. through the call_user_func_array function), and before executing the query.
If you wish to do so - for instance by running the same query several times with different parameters' values in the same script, then you will have to use the same array ( $bindArray) for the following query execution, and update the array entries using the same keys. Copying another array over won't work, unless done by hand:
foreach($bindArray as $k => $v)
$bindArray[$k] = some_new_value();
or
foreach($bindArray as &$v)
$v = some_new_value();
The above would work because it would not break the references on the array entries that bind_param bound with the statement when it was called earlier. Likewise, the following should work because it does not change the references which have been set earlier up.
array_walk($bindArray, function($k,&$v){$v = some_new_value();});
A prepared statement needs to have a well-defined number of arguments; it can't have any element of dynamic functionality. That means you'll have to generate the specific statement that you need and prepare just that.
What you can do – in case your code actually gets called multiple times during the existence of the database connection - is make cache of those prepared statements, and index them by the number of arguments that you're taking. This would mean that the second time you call the function with three arguments, you already have the statement done. But as prepared statements don't survive the disconnect anyway, this really only makes sense if you do multiple queries in the same script run. (I'm deliberately leaving out persistent connections, because that opens up an entirely different can of worms.)
By the way, I'm not an MySQL expert, but would it not make a difference to not have the where conditions joined,but rather writing WHERE tags in (tag1, tag2, tag3, tag4)?
Solved it by the help of an answer found here.
$query = "SELECT * FROM tags WHERE tags.tag LIKE CONCAT('%',?,'%')" . str_repeat(" OR tags.tag LIKE CONCAT('%',?,'%')", $searchNumber - 1)
$stmt = $mysqli -> prepare($query);
$bind_names[] = str_repeat('s', $searchNumber);
for ($i = 0; $i < count($searchArray); $i++){
$bind_name = 'bind'.$i; //generate a name for variable bind1, bind2, bind3...
$$bind_name = $searchArray[$i]; //create a variable with this name and put value in it
$bind_names[] = & $$bind_name; //put a link to this variable in array
}
call_user_func_array(array($stmt, 'bind_param'), &$bind_names);
$stmt -> execute();
I'm learning PDO, and finding it tricky to make sure my statements work correctly. I have a PHP function which is updating my database by simply adding the number 1 to the total.
function add_rating($place_id,$rating_id) {
//make $db accessible inside the function
global $db;
// query v1
$sql = "UPDATE places_ratings SET ? +1 WHERE place_id=?";
$q = $db->prepare($sql);
$q->execute(array($rating_id,$place_id));
}
I tried variations of this, none of which I could get to work. I don't know if I was using question marks wrong. I was following this guide and also a previous SO question. In the end I tried a different method which worked first time, so I am tempted to re-use it as it also seems a lot simpler.
function add_rating($place_id,$rating_id) {
//make $db accessible inside the function
global $db;
// query v2
$query = "UPDATE places_ratings SET $rating_id = ($rating_id +1) WHERE place_id = $place_id";
$update = $db->query($query);
}
My question is: which statement is better/safer? And secondly, what am I doing wrong with the first version with question marks? Thanks...
In general prepared statements as in your first example are safer because they are immune to SQL injection.
Your example doesn't work because you can't specify field names using a ? parameter in a prepared statement. Even if you could your SQL still would be wrong, this would expand to
UPDATE places_ratings SET whatever +1 WHERE place_id=?
which is not valid.
If your $rating_id is generated in code and not taken from user input you could combine both approaches.
Prepared statements are not simply like copy'n'pasting variables into a piece of text. Prepared statements separate between the query logic and the values the query should work on. They're there so you're able to tell your database "You're supposed to do this", let the database understand it, then give it the values it's supposed to do that something with. The logic itself cannot be variable, it needs to be complete the first time.
Therefore, you can only use placeholders for values. Your query needs to read UPDATE ... SET FIELD = VALUE WHERE FIELD = VALUE. The FIELD parts need to be in the statement, the VALUE parts you can use placeholders for. It looks like your $rating_id variable is a variable field name. First of all, that's a bad idea. You should not make field names variable if possible. But if you have to, you cannot use prepared statement placeholders for them. Instead, you'll have to do it like this:
$rating_id = 'field_name';
$query = "UPDATE places_ratings SET `$rating_id` = `$rating_id` + 1 WHERE `place_id` = ?";
$stmt = $db->prepare($query);
$stmt->execute(array($place_id));
It's up to you to make sure $rating_id is safe and contains known, whitelisted values. Don't let the user supply the value for it in any way.
Please, go an learn what prepared statements are. And you could also use a tutorial, that does not promote bad practices and vulnerable code.
A correctly created and used prepared statement will always be more secure then concatenated query string, because prepared statements send query logic and data separately.
Also , if you are using PDO, then quite often the use of bindParam() method should be preferred over passing the values directly in the execute() method as an array. This is because, when passing values in execute(), the values are bound as PDO::PARAM_STR, even if DB column expects and integer.
P.S. Stop using global in your code !!
Where and when do you use the quote method in PDO? I'm asking this in the light of the fact that in PDO, all quoting is done by the PDO object therefore no user input should be escaped/quoted etc. This makes one wonder why worry about a quote method if it's not gonna get used in a prepared statement anyway?
When using Prepared Statements with PDO::prepare() and PDOStatement::execute(), you don't have any quoting to do : this will be done automatically.
But, sometimes, you will not (or cannot) use prepared statements, and will have to write full SQL queries and execute them with PDO::exec() ; in those cases, you will have to make sure strings are quoted properly -- this is when the PDO::quote() method is useful.
While this may not be the only use-case it's the only one I've needed quote for. You can only pass values using PDO_Stmt::execute, so for example this query wouldn't work:
SELECT * FROM tbl WHERE :field = :value
quote comes in so that you can do this:
// Example: filter by a specific column
$columns = array("name", "location");
$column = isset($columns[$_GET["col"]]) ? $columns[$_GET["col"]] : $defaultCol;
$stmt = $pdo->prepare("SELECT * FROM tbl WHERE " . $pdo->quote($column) . " = :value");
$stmt->execute(array(":value" => $value));
$stmt = $pdo->prepare("SELECT * FROM tbl ORDER BY " . $pdo->quote($column) . " ASC");
and still expect $column to be filtered safely in the query.
The PDO system does not have (as far as I can find) any mechanism to bind an array variable in PHP into a set in SQL. That's a limitation of SQL prepared statements as well... thus you are left with the task of stitching together your own function for this purpose. For example, you have this:
$a = array(123, 'xyz', 789);
You want to end up with this:
$sql = "SELECT * FROM mytable WHERE item IN (123, 'xyz', 789)";
Using PDO::prepare() does not work because there's no method to bind the array variable $a into the set. You end up needing a loop where you individually quote each item in the array, then glue them together. In which case PDO::quote() is probably better than nothing, at least you get the character set details right.
Would be excellent if PDO supported a cleaner way to handle this. Don't forget, the empty set in SQL is a disgusting special case... which means any function you build for this purpose becomes more complex than you want it to be. Something like PDO::PARAM_SET as an option on the binding, with the individual driver deciding how to handle the empty set. Of course, that's no longer compatible with SQL prepared statements.
Happy if someone knows a way to avoid this difficulty.
A bit late anwser, but one situation where its useful is if you get a load of data out of your table which you're going to put back in later.
for example, i have a function which gets a load of text out of a table and writes it to a file. that text might later be inserted into another table. the quote() method makes all the quotes safe.
it's real easy:
$safeTextToFile = $DBH->quote($textFromDataBase);
I know that mysql_real_escape_string()
prepends backslashes to the following characters: \x00, \n, \r, \, ', " and \x1a
I know how this protects a query from injection into something like a variable in a where clause. But here's a scenario I am unsure of:
$query = "SELECT * FROM $db WHERE 1";
If $db is taken from a user input, then the user could insert something like:
$db = 'RealDatabase WHERE 1; DELETE FROM RealDatabase WHERE 1; SELECT FROM RealDatabase';
From my understanding, mysql_real_escape_string() would not affect this string,
making the final query:
$query = "SELECT * FROM RealDatabase WHERE 1; DELETE FROM RealDatabase WHERE 1; SELECT FROM RealDatabase WHERE 1";
which would delete the database. Is there another level of protection I am unaware of?
The level of protection you are looking for is supplied by backticks:
"SELECT * FROM `$db` WHERE 1";
Backticks are used to qualify identifiers that could otherwise be ambiguous (ie. MySQL reserved words), and if you are accepting user input or have variably-named columns or databases, you absolutely should use backticks, or I can promise that you will run into trouble in the future. For example, what if you had a system where a temporary field name was created with some user input, only it turned out the field ended up being named update?
"SELECT field1,field2,update FROM table;"
It fails miserably. However:
"SELECT `field`,`field2`,`update` FROM table"
works just fine. (This is actually a real example from a system I worked on a few years ago that had this problem).
This solves your problem in terms of putting in bad SQL. For instance, the following query will simply return an "unknown column" error, where test; DROP TABLE test is the injected attack code:
"SELECT * FROM `test; DROP TABLE test`;"
Be careful though: SQL Injection is still possible with backticks!
For instance, if your $db variable contained data that had a backtick in it, you could still inject some SQL in the normal way. If you're using variable data for database and field names, you should strip it of all backticks before putting it into your statement, and then qualifying it with backticks once inside.
$db = str_replace('`','',$db);
$sql = "SELECT * FROM `$db` WHERE 1";
I utilize a database wrapper which has separate functions for sanitizing data and sanitizing database identifiers, and this is what the latter does :)
You should really look into binding your SQL queries.
This will protect you from basically all SQL injection. It boils down to this:
(taken from PHP.net)
$stmt = mssql_init('NewUserRecord');
// Bind the field names
mssql_bind($stmt, '#username', 'Kalle', SQLVARCHAR, false, false, 60);
// Execute
mssql_execute($stmt);
And PHP has support for binded queries on basically all databases. Oh and of course you should still sanitize all input & output(display).
More info:
- http://php.net/manual/en/function.mssql-bind.php
No, mysql_real_escape_string isn't going to help you here. The function is not context-sensitive (it can't be, because it doesn't HAVE any context), and this is a completely different threat model.
You need to go and validate that the table exists, without sending the user-inputted table name directly to the server. The best solution is to use a server-side array/look-up table containing the table names they are allowed to use. If they try to use something that's not in there, then don't let them.
If you really need ALL of the tables, then you can just ask the server "what tables do you have?" and run through it's output (optionally caching it for some period of time to prevent asking the server every time) - but chances are, eventually you'll have a table that you don't want then to poke around in, and then you need to use the array thing anyway, so just go ahead and do that.
Instead of inserting the database name in the get query you can make a separate table of database names and ids. Then append only the id to the query. Then you can look up the corresponding database name for that id and use that. You can then make sure that the id received is numeric (is_numeric) and you can also be certain that the user can only choose from the databases that are in your list.
(Additionally this will prevent users from finding out names of databases and possibly use them elsewhere in an SQL injection on your site.)
Using the first method you parse the database name before using it in your query and make sure it contains no spaces.
Since table names do not accept whitespace characters, just strip them out. That would make the above $DB RealDatabaseWHERE1;DELETEFROMRealDatabase..... Such would invalidate the query, but prevent the flaw.
If you want to prevent this kind of 'hackish' things, just do explode(' ', $db) then get the result array's [0]. That would get the first part (RealDatabase) and nothing else.
Its just best to use it any time that you have questionable data being used. If you are specifying the table yourself and there's no room for tampering, there's no need to escape it. If your users are deciding anything that could potentially get run as a query, escape it.
If you really really must use a get from the user (bad bad bad) for your database then use the following style of coding...
$realname = '';
switch ($_GET['dbname']){
case 'sometoken' : $realname = 'real_name'; break;
case 'sometoken1' : $realname = 'real_name1'; break;
case 'sometoken2' : $realname = 'real_name2'; break;
case 'sometoken3' : $realname = 'real_name3'; break;
case 'sometoken4' : $realname = 'real_name4'; break;
case default : die ('Cheeky!!!');
}
$query = "SELECT * FROM `{$realname}` WHERE 1";
or alternatively ...
$realname = $tablenames[$_GET['dbname']];
if (!$realname)
die ('Cheeky!!!');
Using these 2 ways or some similar coding will protect your input from unexpected values.
It also means the user never gets to see the real table or database names which they may be able to infer information from.
Make sure you check the content of $_GET['dbname'] to make sure it's valid first otherwise warnings will be issued.
I still say this is a very bad design, it is reminiscent of allowing users to provide a filename and passing that through to I/O functions without a check. It simply too unsafe to consider.
Security is too important to let laziness rule.