rfc:is_literal

This is an old revision of the document!


PHP RFC: Is_Literal

Introduction

Add the function is_literal(), a lightweight and effective way to identify if a string was written by a developer, to remove the risk of a variable containing an Injection Vulnerability.

It's a simple process where a flag is set internally on strings that have been written by a developer (as opposed to a user) and the flag persists through concatenation with other 'literal' strings. The function checks the flag is present and thus no user data is included.

It avoids the “false sense of security” that comes with the flawed “Taint Checking” approach, because escaping is very difficult to get right. It's much safer for developers to use parameterised queries, and well-tested libraries.

is_literal() would be used by libraries, as they require certain sensitive values to only come from the developer; but because it's easy to incorrectly include user values, Injection Vulnerabilities are still introduced by the thousands of developers using these libraries incorrectly. You will notice the linked examples are based on examples found in the Libraries' official documentation, they still “work”, and are typically shorter/easier than doing it correctly (I've found many of them on live websites, and it's why I'm here). A simple Query Builder example being:

$qb->select('u')
   ->from('User', 'u')
   ->where('u.id = ' . $_GET['id']); // INSECURE

(The “Future Scope” section explains why a dedicated type should come later, and how native functions could use the is_literal flag as well.)

Background

The Problem

Injection and Cross-Site Scripting (XSS) vulnerabilities are easy to make, hard to identify, and very common.

We like to think every developer reads the documentation, and would never directly include (inject) user values into their SQL/HTML/CLI - but we all know that's not the case.

It's why these two issues have always been on the OWASP Top 10; a list designed to raise awareness of common issues, ranked on their prevalence, exploitability, detectability, and impact:

Year Injection Position XSS Position
2017 - Latest 1 7
2013 1 3
2010 1 2
2007 2 1
2004 6 4
2003 6 4

Usage Elsewhere

Google are already using this concept with their Go and Java libraries, and it's been very effective.

Christoph Kern (Information Security Engineer at Google) did a talk in 2016 about Preventing Security Bugs through Software Design (also at USENIX Security 2015), pointing out the need for developers to use libraries (like go-safe-html and go-safesql) to do the encoding, where they only accept strings written by the developer (literals). This ensures the thousands of developers using these libraries cannot introduce Injection Vulnerabilities.

It's been so successful Krzysztof Kotowicz (Information Security Engineer at Google, or “Web security ninja”) is now adding it to JavaScript (details below).

Usage in PHP

Libraries would be able to use is_literal() immediately, allowing them to warn developers about Injection Issues as soon as they receive any non-literal values. Some already plan to implement this, for example:

Propel (Mark Scherer): “given that this would help to more safely work with user input, I think this syntax would really help in Propel.”

RedBean (Gabor de Mooij): “You can list RedBeanPHP as a supporter, we will implement this into the core.”

Psalm (Matthew Brown): “I've just added support for a literal-string type to Psalm: https://psalm.dev/r/9440908f39

Proposal

Add the function is_literal().

A string shall pass the is_literal check if it was defined by the programmer in source code, or is the result of a function or instruction whose inputs would all pass the is_literal check.

Concatenation instructions and the following string functions are therefore able to produce literals:

  • str_repeat()
  • str_pad()
  • implode()
  • join()

Namespaces constructed for the programmer by the compiler will also be marked literal for convenience.

is_literal('Example'); // true
 
$a = 'Hello';
$b = 'World';
 
is_literal($a); // true
is_literal($a . $b); // true
is_literal("Hi $b"); // true
 
is_literal($_GET['id']); // false
is_literal(sprintf('Hi %s', $_GET['name'])); // false
is_literal('/bin/rm -rf ' . $_GET['path']); // false
is_literal('<img src=' . htmlentities($_GET['src']) . ' />'); // false
is_literal('WHERE id = ' . $db->real_escape_string($_GET['id'])); // false
 
function example($input) {
  if (!is_literal($input)) {
    throw new Exception('Non-literal value detected!');
  }
  return $input;
}
 
example($a); // OK
example(example($a)); // OK, still the same literal value.
example(strtoupper($a)); // Exception thrown.

Try It

Test it out on 3v4l.org - Currently using the function name is_noble(), while we tested integer support.

How it can be used by libraries - Notice how this example library just raises a warning, to simply let the developer know about the issue, without breaking anything. And it provides an “unsafe_value” value-object to bypass the is_literal() check, but none of the examples need to use it (can be useful as a temporary thing, but there are much safer/better solutions, which developers are/should already be using).

FAQ's

Taint Checking

Taint checking is flawed, isn't this the same?

It is not the same. Taint Checking incorrectly assumes the output of an escaping function is “safe” for a particular context. While it sounds reasonable in theory, the operation of escaping functions, and the context for which their output is safe, is very hard to define and led to a feature that is both complex and unreliable.

$sql = 'SELECT * FROM users WHERE id = ' . $db->real_escape_string($id); // INSECURE
$html = "<img src=" . htmlentities($url) . " alt='' />"; // INSECURE
$html = "<a href='" . htmlentities($url) . "'>..."; // INSECURE

All three examples would be incorrectly considered “safe” (untainted). The first two need the values to be quoted. The third example, htmlentities() does not escape single quotes by default before PHP 8.1 (fixed), and it does not consider the issue of 'javascript:' URLs.

In comparison, is_literal() doesn't have an equivalent of untaint(), or support escaping. Instead PHP will set the is_literal flag, and as soon as the value has been manipulated or includes anything that is not a literal (e.g. user data), the is_literal flag is removed.

This allows libraries to use is_literal() to check the sensitive values they receive from the developer. Then it's up to the library to handle the escaping (if it's even needed). The “Future Scope” section notes how native functions would be able to use the is_literal flag as well.

Education

Why not educate everyone instead?

You can't - developer training simply does not scale, and mistakes still happen.

We cannot expect everyone to have formal training, know everything from day 1, and consider programming a full time job. We want new programmers, with a variety of experiences, ages, and backgrounds. Everyone should be guided to do the right thing, and notified as soon as they make a mistake (we all make mistakes). We also need to acknowledge that many programmers are busy, do copy/paste code, don't necessarily understand what it does, edit it for their needs, then simply move on to their next task.

Static Analysis

Why not use static analysis?

Ultimately it will never be used by most developers.

I still agree with Tyson Andre, you should use Static Analysis, but it's an extra step that most programmers cannot be bothered to do, especially those who are new to programming (its usage tends to be higher among those writing well-tested libraries).

Also, these tools currently focus on other issues (type checking, basic logic flaws, code formatting, etc), rarely attempting to address Injection Vulnerabilities. Those that do are often incomplete, need sinks specified on all library methods (unlikely to happen), and are not enabled by default. For example, Psalm, even in its strictest errorLevel (1), and running --taint-analysis (rarely used), will not notice the missing quote marks in this SQL, and incorrectly assume it's safe:

$db = new mysqli('...');
 
$id = (string) ($_GET['id'] ?? 'id'); // Keep the type checker happy.
 
$db->prepare('SELECT * FROM users WHERE id = ' . $db->real_escape_string($id)); // INSECURE

Performance

What about the performance impact?

Máté Kocsis has created a php benchmark to replicate the old Intel Tests, and the preliminary testing on this implementation has found a 0.124% performance hit for the Laravel Demo app, and 0.161% for Symfony (rounds 4-6, which involved 5000 requests). These tests do not connect to a database, as the variability introduced makes it impossible to measure that low of a difference.

The full stress-test is 3.719% when running this concat test, but this is not representative of a typical PHP script (it's not normal to concatenate 4 strings, 5 million times, with no other actions).

String Concatenation

Is string concatenation supported?

Yes. The is_literal flag is preserved when two literal values are concatenated; this makes it easier to use is_literal(), especially by developers that use concatenation for their SQL/HTML/CLI/etc.

Previously we tried a version that only supported concatenation at compile-time (not run-time), to see if it would reduce the performance impact even further. The idea was to require everyone to use special literal_concat() and literal_implode() functions, which would raise exceptions to highlight where mistakes were made. These two functions can still be implemented by developers themselves (see Support Functions below), as they can be useful; but requiring everyone to use them would have required big changes to existing projects, and exceptions are not a graceful way of handling mistakes.

Performance wise, my simplistic testing found there was still a small impact without run-time concat:

  Laravel Demo App: +0.30% with, vs +0.18% without.
  Symfony Demo App: +0.06% with, vs +0.06% without.
  My Concat Test:   +4.36% with, vs +2.23% without.
  -
  Website with 22 SQL queries: Inconclusive, too variable.
(Under The Hood: This is because concat_function() in “zend_operators.c” uses zend_string_extend() which needs to remove the is_literal flag. Also “zend_vm_def.h” does the same; and supports a quick concat with an empty string (x2), which would need its flag removed as well).

And by supporting both forms of concatenation, it makes it easier for developers to understand (many are not aware of the difference).

String Splitting

Why don't you support string splitting then?

In short, we can't find any real use cases (security features should try to keep the implementation as simple as possible).

Also, the security considerations are different. Concatenation joins known/fixed units together, whereas if you're starting with a literal string, and the program allows the Evil-User to split the string (e.g. setting the length in substr), then they get considerable control over the result (it creates an untrusted modification).

These are unlikely to be written by a programmer, but consider these:

$length = ($_GET['length'] ?? -5);
$url    = substr('https://example.com/js/a.js?v=55', 0, $length);
$html   = substr('<a href="#">#</a>', 0, $length);

If that URL was used in a Content-Security-Policy, then it's necessary to remove the query string, but as more of the string is removed, the more resources can be included (“https:” basically allows resources from anywhere). With the HTML example, moving from the tag content to the attribute can be a problem (technically the HTML Templating Engine should be fine, but unfortunately libraries like Twig are not currently context aware, so you need to change from the default 'html' encoding to explicitly using 'html_attr' encoding).

Or in other words; trying to determine if the is_literal flag should be passed through functions like substr() is complex. Having a security feature be difficult to reason about, gives a much higher chance of mistakes.

Krzysztof Kotowicz has confirmed that, at Google, with “go-safe-html”, splitting is explicitly not supported because it “can cause issues”; for example, “arbitrary split position of a HTML string can change the context”.

WHERE IN

What about an undefined number of parameters, e.g. WHERE id IN (?, ?, ?)?

You can follow the advice from Levi Morrison, PDO Execute, and Drupal Multiple Arguments, and implement as such:

$sql = 'WHERE id IN (' . join(',', array_fill(0, count($ids), '?')) . ')';

Or, you could use concatenation:

$sql = '?';
for ($k = 1; $k < $count; $k++) {
  $sql .= ',?';
}

Non-Parameterised Values

How can this work with Table and Field names in SQL, which cannot use parameters?

They are often in variables written as literal strings anyway (so no changes needed); and if they are dependent on user input, in most cases you can (and should) use literals:

$order_fields = [
    'name',
    'created',
    'admin',
  ];
 
$order_id = array_search(($_GET['sort'] ?? NULL), $order_fields);
 
$sql .= ' ORDER BY ' . $order_fields[$order_id];

By using an allow-list, we ensure the user (attacker) cannot use anything unexpected.

Non-Literal Values

How does this work in cases where you can't use literal values?

For example Dennis Birkholz noted that some Systems/Frameworks currently define some variables (e.g. table name prefixes) without the use of a literal (e.g. ini/json/yaml). And Larry Garfield noted that in Drupal's ORM “the table name itself is user-defined” (not in the PHP script).

While most systems can use literal values entirely, these special non-literal values should still be handled separately (and carefully). This approach allows the library to ensure the majority of the input (SQL) is a literal, and then it can consistently check/escape those special values (e.g. does it match a valid table/field name, which can be included safely).

How this can be done with aliases, or the example Query Builder.

Faking It

What if I really really need to mark a value as a literal?

This implementation does not provide a way for a developer to mark anything they want as a literal. This is on purpose. We do not want to recreate the biggest flaw of Taint Checking. It would be very easy for a naive developer to mark all escaped values as a literal (seeing it as a safe value, which is wrong).

That said, we do not pretend there aren't ways around this (e.g. using var_export), but doing so is clearly the developer doing something wrong. We want to provide safety rails, but there is nothing stopping the developer from jumping over them if that's their choice.

Usage by Libraries

How can libraries use is_literal()?

The main focus is on values that developers provide to the library, this example library shows how certain sensitive values are checked as they are received, where it just uses basic warnings by default, could raise exceptions, or have the checks turned off on a per query basis (or entirely). Libraries could choose to only run these checks in development mode (and turned off in production), or do additional checks to see if the value is likely to be an issue (e.g. value matches a field name), or write to a log, or report via an API/email, etc.

They could also use additional is_literal() checks later in the process (internally), to ensure the library hasn't introduced a vulnerability either; but this isn't a priority, simply because libraries are rarely the source of Injection Vulnerabilities.

Integer Values

We wanted to flag integers defined in the source code, in the same way we are doing with strings. Unfortunately it would require a big change to add a literal flag on integers. Changing how integers work internally would have made a big performance impact, and potentially affected every part of PHP (including extensions).

Due to this limitation, we considered an approach to trust all integers. It was noted that existing code and tutorials already use integers directly. While this is not as philosophically pure, we continued to explore this possibility because we could not find any way that an Injection Vulnerability could be introduced with integers in SQL, HTML, CLI; and other contexts as well (e.g. preg, mail additional_params, XPath query, and even eval).

We could not find any character encoding issues either (The closest we could find was EBCDIC, an old IBM character encoding, which encodes the 0-9 characters differently; which anyone using it would need to re-encode either way, and EBCDIC is not supported by PHP). And we could not find any issue with a 64bit PHP server sending a large number to a 32bit database, because the number is being encoded as characters in a string, so that's also fine.

However, the feedback received on the Internals mailing list was that while safe from Injection Vulnerabilities it might cause developers to assume them to be safe from developer/logic errors, and ultimately the preference was the simpler approach, that did not allow integers from any source.

Other Values

Why don't you support Boolean/Float values?

It's a very low-value feature, and we cannot be sure of the security implications.

For example, the value you put in is not always the same as what you get out:

var_dump((string) true);  // "1"
var_dump((string) false); // ""
var_dump(2.3 * 100);      // 229.99999999999997
 
setlocale(LC_ALL, 'de_DE.UTF-8');
var_dump(sprintf('%.3f', 1.23)); // "1,230"
 // Note the comma, which can be bad for SQL.
 // Pre 8.0 this also happened with string casting.

Naming

Why is it called is_literal()?

A “Literal String” is the standard name for strings in source code. See Google.

A string literal is the notation for representing a string value within the text of a computer program. In PHP, strings can be created with single quotes, double quotes or using the heredoc or the nowdoc syntax.

We also need to keep to a single word name (to support a dedicated type in the future).

Support Functions

What about other support functions?

We did consider literal_concat() and literal_implode() functions (see String Concatenation above), but these can be userland functions:

function literal_implode($separator, $array) {
  $return = implode($separator, $array);
  if (!is_literal($return)) {
      // You will probably only want to raise
      // an exception on your development server.
    throw new Exception('Non-literal value detected!');
  }
  return $return;
}
 
function literal_concat(...$a) {
  return literal_implode('', $a);
}

Developers can use these to help identify exactly where they made a mistake, for example:

$sortOrder = 'ASC';
 
// 300 lines of code, or multiple function calls
 
$sql .= ' ORDER BY name ' . $sortOrder;
 
// 300 lines of code, or multiple function calls
 
$db->query($sql);

If a developer changed the literal 'ASC' to $_GET['order'], the error would be noticed by $db->query(), but it's not clear where the non-literal value was introduced. Whereas, if they used literal_concat(), that would raise an exception much earlier, and highlight exactly where the mistake happened:

$sql = literal_concat($sql, ' ORDER BY name ', $sortOrder);

Other Functions

Why not support other string functions?

Like String Splitting, we can't find any real use cases, and don't want to make this complicated. For example strtoupper() might be reasonable, but we would need to consider how it would be used, and check for any oddities (e.g. output varying based on the current locale). Also, functions like str_shuffle() create unpredictable results.

Limitations

Does this mean the value is completely safe?

While these values are not at risk of containing an Injection Vulnerability, obviously they cannot be completely safe from every kind of developer/logic issue, For example:

$cli = 'rm -rf ?'; // RISKY
$sql = 'DELETE FROM my_table WHERE my_date >= ?'; // RISKY

The parameters could be set to “/” or “0000-00-00”, which can result in deleting a lot more data than expected.

There's no single RFC that can completely solve all developer errors, but this takes one of the biggest ones off the table.

Extensions

Extensions create and manipulate strings, won't this break the flag on strings?

Strings have multiple flags already that are off by default - this is the correct behaviour when extensions create their own strings (should not be flagged as a literal). If an extension is found to be already using the flag we're using for is_literal (unlikely), that's the same as any new flag being introduced into PHP, and will need to be updated in the same way.

Reflection API

Why don't you use the Reflection API?

This allows you to “introspect classes, interfaces, functions, methods and extensions”; it's not currently set up for object methods to inspect the code calling it. Even if that was to be added (unlikely), it could only check if the literal value was defined there, it couldn't handle variables (tracking back to their source), nor could it provide any future scope for a dedicated type, nor could native functions work with this (see “Future Scope”).

Previous Examples

Go programs can use “ScriptFromConstant” to express the concept of a “compile time constant” (more details).

Java can use Error Prone with @CompileTimeConstant to ensure method parameters can only use “compile-time constant expressions”.

JavaScript is getting isTemplateObject, for “Distinguishing strings from a trusted developer from strings that may be attacker controlled” (intended to be used with Trusted Types).

Perl has a Taint Mode, via the -T flag, where all input is marked as “tainted”, and cannot be used by some methods (like commands that modify files), unless you use a regular expression to match and return known-good values (where regular expressions are easy to get wrong).

There is a Taint extension for PHP by Xinchen Hui, and a previous RFC proposing it be added to the language by Wietse Venema.

And there is the Automatic SQL Injection Protection RFC by Matt Tait (this RFC uses a similar concept of the SafeConst). When Matt's RFC was being discussed, it was noted:

  • “unfiltered input can affect way more than only SQL” (Pierre Joye);
  • this amount of work isn't ideal for “just for one use case” (Julien Pauli);
  • It would have effected every SQL function, such as mysqli_query(), $pdo->query(), odbc_exec(), etc (concerns raised by Lester Caine and Anthony Ferrara);
  • Each of those functions would need a bypass for cases where unsafe SQL was intentionally being used (e.g. phpMyAdmin taking SQL from POST data) because some applications intentionally “pass raw, user submitted, SQL” (Ronald Chmara 1/2).

All of these concerns have been addressed by is_literal().

I also agree with Scott Arciszewski, “SQL injection is almost a solved problem [by using] prepared statements”, where is_literal() is essential for identifying the mistakes developers are still making.

Backward Incompatible Changes

No known BC breaks, except for code-bases that already contain the userland function is_literal() which is unlikely.

Proposed PHP Version(s)

PHP 8.1

RFC Impact

To SAPIs

None known

To Existing Extensions

None known

To Opcache

None known

Open Issues

Inconsistencies

While compile-time and run-time concatenation with literal strings is consistent and works, when it comes to some developer-defined values (like integers, which we cannot flag), the compiling process can optimise a value so it's seen as a literal.

As these optimisations happen during the compilation process, they cannot contain user data; but developers may wonder why a developer defined integer can sometimes be seen as a literal, e.g.

$one = 1;
is_literal('A' . $one); // false, flag removed with runtime concatenation of an integer.
is_literal('A' . 1); // true, compiler optimises this to the literal 'A1'.
 
$a = "Hello ";
$b = $a . 2; // The 2 is coerced to the string '2' to optimise the concatenation.
is_literal($b); // true
 
$a = implode("-", [1, 2, 3]);
is_literal($a); // Normally false, but OPcache can optimise to the literal '1-2-3'

Interned Strings

Output from chr() appears to be a literal. This was noticed by Claude Pache, and on a technical level is due to the use of Interned Strings, an optimisation used by RETURN_CHAR that re-uses single character values. While this could be used as a way to intentionally fake a literal string, it's unlikely to be used to create sensitive strings.

Future Scope

1) As noted by someniatko and Matthew Brown, having a dedicated type would be useful in the future, as “it would serve clearer intent”, which can be used by IDEs, Static Analysis, etc. It was agreed we would add this type later, via a separate RFC, so this RFC can focus on the is_literal flag, and provide libraries a simple backwards-compatible function, where they can decide how to handle non-literal values.

2) As noted by MarkR, the biggest benefit will come when this flag can be used by PDO and similar functions (mysqli_query, preg_match, exec, etc).

However, first we need libraries to start using is_literal() to check their inputs. The library can then do their thing, and apply the appropriate escaping, which can result in a value that no longer has the is_literal flag set, but is perfectly safe for the native functions.

With a future RFC, we could potentially introduce checks for the native functions. For example, if we use the Trusted Types concept from JavaScript (which protects 60+ Injection Sinks, like innerHTML), the libraries create a stringable object as their output. These objects can be added to a list of safe objects for the relevant native functions. The native functions could then warn developers when they do not receive a value with the is_literal flag, or one of the safe objects. These warnings would not break anything, they just make developers aware of the mistakes they have made, and we will always need a way of switching them off entirely (e.g. phpMyAdmin).

Proposed Voting Choices

Accept the RFC. Yes/No

Implementation

References

N/A

Rejected Features

N/A

Thanks

  1. Joe Watkins, krakjoe, for writing the full implementation, including support for concatenation and integers, and helping me though the RFC process.
  2. Máté Kocsis, mate-kocsis, for setting up and doing the performance testing.
  3. Scott Arciszewski, CiPHPerCoder, for checking over the RFC, and provided text on how we could implement integer support under a is_noble() name.
  4. Dan Ackroyd, DanAck, for starting the first implementation, which made this a reality, providing literal_concat() and literal_implode(), and followup on how it should work.
  5. Xinchen Hui, who created the Taint Extension, allowing me to test the idea; and noting how Taint in PHP5 was complex, but “with PHP7's new zend_string, and string flags, the implementation will become easier” source.
  6. Rowan Francis, for proof-reading, and helping me make an RFC that contains readable English.
  7. Rowan Tommins, IMSoP, for re-writing this RFC to focus on the key features, and putting it in context of how it can be used by libraries.
  8. Nikita Popov, NikiC, for suggesting where the flag could be stored. Initially this was going to be the “GC_PROTECTED flag for strings”, which allowed Dan to start the first implementation.
  9. Mark Randall, MarkR, for suggestions, and noting that “interned strings in PHP have a flag”, which started the conversation on how this could be implemented.
  10. Sara Golemon, SaraMG, for noting how this RFC had to explain how is_literal() is different to the flawed Taint Checking approach, so we don't get “a false sense of security or require far too much escape hatching”.
rfc/is_literal.1625416479.txt.gz · Last modified: 2021/07/04 16:34 by krakjoe