Escaping of key-strings in GPC-arrays behave different to the escaping of their values.
First I expected that keys in submitted gpc-arrays are never escaped.
Anyway. After I saw escaped keys, I assumed they're escaped according to the settings of magic quotes.
... it's even worse...
It took me over 2 days of testing to figure out the exact behavior and creating two functions (one for each php-version) that strips slashes reliably from any array submitted to a script. Hope this saves someones time and nerves.
The following is true for $_GET- and $_POST-arrays. I hope other arrays affected by magic quotes behave equally.
I did not test the behavior for cases where magic_quotes_sybase is set.
== legend for possible case combinations ==
Px = php version we're using
P4 = php 4.3.9
P5 = php 5.0.2
MQ = MagicQuotes GPC
+MQ = magic quotes enabled
-MQ = magic quotes disabled
TL = TopLevel key
+TL = key is on top level (i.e. $_GET['myKey'])
-TL = key is nested within another array (i.e. $_GET['myList']['myKey'])
AK = ArrayKey
+AK = the value of the key is another array (i.e. is_array($_GET['myKey']) == true)
-AK = the value is a normal string (i.e. is_string($_GET['myKey']) == true)
== legend for possible results ==
KE = KeyEscaping
+KE = control chars are prefixed with a backslash
-KE = key is returned as submitted and needn't to be stripped
VE = ValueEscaping (doesn't apply for array as value)
+VE = control chars are prefixed with a backslash
-VE = value is returned as submitted and needn't to be stripped
== here we go - the following rules apply ==
1) P4 +MQ +AK +TL --> -KE
2) P4 +MQ +AK -TL --> +KE
3) P4 +MQ -AK +TL --> -KE +VE
4) P4 +MQ -AK -TL --> +KE +VE
5) P4 -MQ +AK +TL --> -KE
6) P4 -MQ +AK -TL --> -KE
7) P4 -MQ -AK +TL --> -KE -VE
8) P4 -MQ -AK -TL --> -KE -VE
9) P5 +MQ +AK +TL --> -KE
10) P5 +MQ +AK -TL --> +KE
11) P5 +MQ -AK +TL --> +KE +VE
12) P5 +MQ -AK -TL --> +KE +VE
13) P5 -MQ +AK +TL --> -KE
14) P5 -MQ +AK -TL --> -KE
15) P5 -MQ -AK +TL --> +KE -VE
16) P5 -MQ -AK -TL --> +KE -VE
17) The chars '.', ' ' are always replaced by '_' when used in keys.
Example (rule 15):
When running under php 5.0.2 having magic quotes disabled, gpc-keys on top level containing strings are escaped while their associated values are not.
== The following function will strip GPC-arrays for php 4.3.9 ==
<?php
function transcribe($aList, $aIsTopLevel = true) {
$gpcList = array();
$isMagic = get_magic_quotes_gpc();
foreach ($aList as $key => $value) {
$decodedKey = ($isMagic && !$aIsTopLevel)?stripslashes($key):$key;
if (is_array($value)) {
$decodedValue = transcribe($value, false);
} else {
$decodedValue = ($isMagic)?stripslashes($value):$value;
}
$gpcList[$decodedKey] = $decodedValue;
}
return $gpcList;
}
?>
== The following function will strip GPC-arrays for php 5.0.2 ==
<?php
function transcribe($aList, $aIsTopLevel = true) {
$gpcList = array();
$isMagic = get_magic_quotes_gpc();
foreach ($aList as $key => $value) {
if (is_array($value)) {
$decodedKey = ($isMagic && !$aIsTopLevel)?stripslashes($key):$key;
$decodedValue = transcribe($value, false);
} else {
$decodedKey = stripslashes($key);
$decodedValue = ($isMagic)?stripslashes($value):$value;
}
$gpcList[$decodedKey] = $decodedValue;
}
return $gpcList;
}
?>
Usage:
<?php
$unstrippedGET = transcribe($_GET);
$unstrippedPOST = transcribe($_POST);
?>
Maybe someone is willing to test those combinations for other php-versions and with magic_quotes_sybase set to 'on' - let me know.
Sorry for this huge amount of text, but it's complete. I was unable to compress the decision table more than this.
