str_replace() on multibyte strings dangerous?

Question

Given certain multibyte character sets, am I correct in assuming that the following doesn't do what it was intended to do?

$string = str_replace('"', '\\"', $string);

In particular, if the input was in a character set that might have a valid character like 0xbf5c, so an attacker can inject 0xbf22 to get 0xbf5c22, leaving a valid character followed by an unquoted double quote (").

Is there an easy way to mitigate this problem, or am I misunderstanding the issue in the first place?

(In my case, the string is going into the value attribute of an HTML input tag: echo 'input type="text" value="' . $string . '">';)

EDIT: For that matter, what about a function like preg_quote()? There's no charset argument for it, so it seems totally useless in this scenario. When you DON'T have the option of limiting charset to UTF-8 (yes, that'd be nice), it seems like you are really handicapped. What replace and quoting functions are available in that case?

Answer 1

No, you’re right: Using a singlebyte string function on a multibyte string can cause an unexpected result. Use the multibyte string functions instead, for example mb_ereg_replace or mb_split:

$string = mb_ereg_replace('"', '\\"', $string);
$string = implode('\\"', mb_split('"', $string));

---

Edit    Here’s a mb_replace implementation using the split-join variant:

function mb_replace($search, $replace, $subject, &$count=0) {
    if (!is_array($search) && is_array($replace)) {
        return false;
    }
    if (is_array($subject)) {
        // call mb_replace for each single string in $subject
        foreach ($subject as &$string) {
            $string = &mb_replace($search, $replace, $string, $c);
            $count += $c;
        }
    } elseif (is_array($search)) {
        if (!is_array($replace)) {
            foreach ($search as &$string) {
                $subject = mb_replace($string, $replace, $subject, $c);
                $count += $c;
            }
        } else {
            $n = max(count($search), count($replace));
            while ($n--) {
                $subject = mb_replace(current($search), current($replace), $subject, $c);
                $count += $c;
                next($search);
                next($replace);
            }
        }
    } else {
        $parts = mb_split(preg_quote($search), $subject);
        $count = count($parts)-1;
        $subject = implode($replace, $parts);
    }
    return $subject;
}

As regards the combination of parameters, this function should behave like the singlebyte str_replace.

Answer 2

The code is perfectly safe with sane multibyte-encodings like UTF-8 and EUC-TW, but dangerous with broken ones like Shift_JIS, GB*, etc. Rather than going through all the headache and overhead to be safe with these legacy encodings, I would recommend just supporting only UTF-8.

Answer 3

You could use either mb_ereg_replace by first specifying the charset with mb_regex_encoding(). Alternatively if you use UTF-8, you can use preg_replace with the u modifier.