Pièges liés à l'autorisation : que dissimule Keycloak ?

Auteur : Valerii Filatov

L'autorisation et l'enregistrement des utilisateurs sont des éléments importants de toute application, non seulement pour les utilisateurs mais aussi pour la sécurité. Quels pièges cache le code source d’une solution open source populaire de gestion des identités ? Comment affectent-ils l’application ?

Si vous avez déjà mis en place une autorisation pour des applications Web, vous connaissez probablement tous les problèmes frustrants qui peuvent survenir. Je ne fais pas exception non plus.

Une fois, j'ai implémenté l'autorisation basée sur Messenger dans la partie frontend d'un projet. Cela semblait être la tâche la plus simple au monde, mais s'est avéré être le contraire : les délais approchaient, le code trébuchait sur les API de messagerie et les gens autour de moi criaient.

Après ce cas, mon collègue m'a montré un outil sympa qui rationaliserait la mise en œuvre de l'autorisation dans nos futurs projets. Ce projet était Keycloak, une solution Java open source permettant l'authentification unique (SSO) avec gestion des identités et des accès destinée aux applications et services modernes.

Comme j'utilise moi-même la solution, je trouve intéressant de rentrer dans le code source et d'utiliser l'analyseur statique PVS-Studio pour rechercher les bugs qui s'y cachent.

He-he, NullPointerException classique

Quelqu’un a frappé à la porte. Le Junior Dev a tenté d'ouvrir une porte et s'est écrasé. « NullPointerException ! » a conclu le développeur senior.

Des erreurs liées à la vérification de null sont rencontrées dans presque tous les projets que nous avons vérifiés auparavant. Alors, commençons par cette vieille mais précieuse erreur.

Fragment N1

private void checkRevocationUsingOCSP(X509Certificate[] certs) 
  throws GeneralSecurityException {
    ....    
    if (rs == null) {
      if (_ocspFailOpen)
        logger.warnf(....);
      else
        throw new GeneralSecurityException(....);
    }

    if (rs.getRevocationStatus() == 
      OCSPProvider.RevocationStatus.UNKNOWN) { // <=
        if (_ocspFailOpen)
            logger.warnf(....);
        else
            throw new GeneralSecurityException(....);
    ....
}

Attention :

V6008 Déréférencement nul potentiel de 'rs'.

Cet extrait de code a une vérification null, mais nous devons comprendre ce qui se passe à l'intérieur. Si la variable _oscpFailOpen est vraie, le programme ne lancera pas d'exception. Il enregistrera uniquement le journal à ce sujet et poursuivra l'exécution. Il recevra une NullPointerException dans le if suivant, puisque la variable rs a déjà été utilisée dans si.

Il peut sembler que le programme lancerait simplement une autre exception s'il n'y avait pas la NullPointerException. Mais ce n'est pas le cas car la variable _oscpFailOpen est vraie et le programme doit continuer son exécution. Pas de destin, il a trébuché sur le pointeur null et est tombé dans l'exception.

Fragment N2

public void writeDateAttributeValue(XMLGregorianCalendar attributeValue) 
  throws ProcessingException {
    ....
    StaxUtil.writeAttribute(
      writer, 
      "xsi", 
      JBossSAMLURIConstants.XSI_NSURI.get(), 
      "type", 
      "xs:" + attributeValue.getXMLSchemaType().getLocalPart()   // <=
    ); 

    if (attributeValue == null) {
      StaxUtil.writeAttribute(
        writer, 
        "xsi", 
        JBossSAMLURIConstants.XSI_NSURI.get(), 
        "nil", 
        "true"
      );
    ....
}

Attention* :*

V6060 La référence 'attributeValue' a été utilisée avant d'être vérifiée par rapport à null.

"Mieux vaut tard que jamais", cette phrase est suffisante dans de nombreux cas, mais malheureusement, pas pour le contrôle null. Dans l'extrait de code ci-dessus, l'objet attributeValue est utilisé avant que son existence ne soit vérifiée, ce qui conduit à l'NullPointerException.

Remarque. Si vous souhaitez découvrir d'autres exemples de tels bugs, nous avons dressé une liste pour vous !

Des disputes ?

Je ne sais pas comment cela s'est produit, mais Keycloak a des erreurs liées au nombre d'arguments dans les fonctions de chaîne de format. Il ne s’agit pas d’une statistique illustrative, mais juste d’un fait amusant.

Fragment N3

protected void process() {
  ....
  if (f == null) {
    ....
  } else {
    ....
    if (isListType(f.getType()) && t instanceof ParameterizedType) {
      t = ((ParameterizedType) t).getActualTypeArguments()[0];
      if (!isBasicType(t) && t instanceof Class) {
        ....
        out.printf(", where value is:\n", ts);              // <=
        ....
      }
    } else if (isMapType(f.getType()) && t instanceof ParameterizedType) {
      ....
      out.printf(", where value is:\n", ts);                // <=
      ....
    }
  }
}

Attention :

V6046 Format incorrect. Un nombre différent d'éléments de format est attendu. Arguments non utilisés : 1.

Dans le fragment ci-dessus, lorsque la fonction printf est appelée, nous passons une chaîne de format et une valeur à substituer dans la chaîne. Il n'y a qu'un seul problème : il n'y a tout simplement pas de place dans la chaîne pour substituer des arguments.

Il est assez intéressant qu'un développeur ait non seulement copié et collé le fragment de code de if à else if, mais qu'un développeur ait également copié et collé l'erreur à l'intérieur de else si.

Dans l'extrait suivant, nous avons le cas inverse : les développeurs ont épargné les arguments.

Fragment N4

public String toString() {
  return String.format(
    "AuthenticationSessionAuthNoteUpdateEvent 
      [ authSessionId=%s,
        tabId=%s,
        clientUUID=%s,
        authNotesFragment=%s ]", 
    authSessionId, 
    clientUUID, 
    authNotesFragment);      // <=
}

Attention :

V6046 Format incorrect. Un nombre différent d'éléments de format est attendu. Arguments manquants : 4.

Bien que les développeurs aient transmis les arguments authSessionId, clientUUID et authNotesFragment à la fonction, le quatrième argument tabld est un peu manqué.

Dans ce cas, la méthode String.format lancera une IllegalFormatException, ce qui peut être une mauvaise surprise.

Code impur

Les avertissements PVS-Studio suivants ne sont pas liés aux erreurs d'intégrité du code. Ces avertissements concernent davantage la qualité du code. Je viens de souligner que ce ne sont pas des erreurs fermes.

"What's the point of looking at them?" you may think. First, I believe that code should be clean and neat. It's not a matter of tastes: clean code is not only about the visual experience but also about code readability. Imho, it's important for any project, especially Open Source. Secondly, I'd like to show that the PVS-Studio static analyzer helps not only fix errors in code but also makes it beautiful and clean.

Not enough variables, My Lord!

For some reason, the following code fragment looks in my mind like an evil plan of someone who has a bad attitude to use variables: "Let's adopt variables and leave them waiting in horror for the garbage collector to gobble them up..."

Fragment N5

 private void onUserRemoved(RealmModel realm, String userId) {
    int num = em.createNamedQuery("deleteClientSessionsByUser")
      .setParameter("userId", userId).executeUpdate();             // <=
    num = em.createNamedQuery("deleteUserSessionsByUser")
      .setParameter("userId", userId).executeUpdate();
}

Warning:

V6021 The value is assigned to the 'num' variable but is not used.

From the point of program operation, nothing terrible happens in this code snippet. But it's still not clear why developers wrote that.

The num variable contains the number of set parameters that the executeUpdate method returns. So, I thought that the method might have had a check for changes. Even if I rewind in time, I'll only find that calls to the method aren't written to a variable but accept the current state a little later.

The result is a useless assignment to the variable—just like in the next fragment.

Fragment N6

private void verifyCodeVerifier(String codeVerifier, String codeChallenge, 
  String codeChallengeMethod) throws ClientPolicyException {
    ....
    String codeVerifierEncoded = codeVerifier;

    try {
      if (codeChallengeMethod != null &&
        codeChallengeMethod.equals(OAuth2Constants.PKCE_METHOD_S256)) {

        codeVerifierEncoded = generateS256CodeChallenge(codeVerifier);

        } else {
          codeVerifierEncoded = codeVerifier; // <=
        }
    } catch (Exception nae) {
      ....
    }
}

Warning:

V6026 This value is already assigned to the 'codeVerifierEncoded' variable.

If you look at the code, you can see that before this, the codeVerifierEncoded variable has already assigned the same value in the else branch. A developer just did redundant action: and useless, and overcomplicating.

The next code fragment just amuses me.

Fragment N7

private static Type[] extractTypeVariables(Map<String, Type> typeVarMap, 
  Type[] types){
    for (int j = 0; j < types.length; j++){
      if (types[j] instanceof TypeVariable){
        TypeVariable tv = (TypeVariable) types[j];
        types[j] = typeVarMap.get(tv.getName());
      } else {
        types[j] = types[j];     // <=
      }
    }
    return types;
}

Warning:

V6005 The variable 'types[j]' is assigned to itself.

It looks just like the previous snippet, but honestly, I'm totally lost on what this code is trying to do.

At first, I thought we were facing a nested loop here, and the author had just mixed up the variables i and j. But eventually I realized that there's only one loop here.

I also thought the assignment appeared when developers were refactoring the code, which might have been even more complicated before. In the end, I found that the function was originally created this way (commit).

So sweet copy-paste

Copy-paste errors are quite common. I'm sure you've seen them even in your own code.

Keycloak has some traces of the copy-paste use, too.

Fragment 8

public class IDFedLSInputResolver implements LSResourceResolver {
  ....

  static {
    ....
    schemaLocations.put("saml-schema-metadata-2.0.xsd", 
      "schema/saml/v2/saml-schema-metadata-2.0.xsd");
    schemaLocations.put("saml-schema-x500-2.0.xsd", 
      "schema/saml/v2/saml-schema-x500-2.0.xsd");
    schemaLocations.put("saml-schema-xacml-2.0.xsd", 
      "schema/saml/v2/saml-schema-xacml-2.0.xsd");

    schemaLocations.put("saml-schema-xacml-2.0.xsd", 
      "schema/saml/v2/saml-schema-xacml-2.0.xsd");          // <=

    schemaLocations.put("saml-schema-authn-context-2.0.xsd", 
      "schema/saml/v2/saml-schema-authn-context-2.0.xsd");
    ....
  }
  ....
}

Warning*:*

V6033 An item with the same key '"saml-schema-xacml-2.0.xsd"' has already been added.

Honestly, even though I knew there was a typo in the source code, I had a hard time finding it right away in the code.

If you notice, in the schemaLocations.put method calls, the passed arguments are quite similar. So, I assume that the dev who wrote this code simply copied a string as a template and then just changed values. The problem is that during copy-pasting, one line that repeats the previous one has crept into the project.

Such "typos" can either lead to serious consequences or have no effect at all. This copy-pasting example has been in the project since November 21, 2017 (commit), and I don't think it causes any serious problems.

We're including this error in the article to remind developers to be careful when copying and pasting code fragments and to keep an eye on any changes in code. Want to read more about it? Here's the article about the copy-paste typos.

Unreachable code

The headline gives us a little clue as to what kind of warning awaits us in the following snippet. I suggest you use your detective skills to spot the flaw yourself.

Fragment N9

public void executeOnEvent(ClientPolicyContext context) 
  throws ClientPolicyException {
    switch (context.getEvent()) {
      case REGISTER:
      case UPDATE:
        ....
      case RESOURCE_OWNER_PASSWORD_CREDENTIALS_REQUEST:
        ....
        executeOnAuthorizationRequest(ropcContext.getParams());
        return;
      default:
        return;
    }
}

It's not that easy to detect, isn't it? I'm not gloating over you, I just give you a chance to roughly access the situation. I show it because a small flaw makes it harder for the developer to find the error without examining the rest of the code. To find out what error is lurking in this code snippet, we need to look at what is cloaked in the executeOnAuthorizationRequest method:

private void executeOnAuthorizationRequest(MultivaluedMap<String, 
  String> params) throws ClientPolicyException {
    ....
    throw new ClientPolicyException(....);
}

Yes, this method throws an exception. That is, all the code written after calling this method will be unreachable—the PVS-Studio analyzer detected it.

public void executeOnEvent(ClientPolicyContext context) 
  throws ClientPolicyException {
    switch (context.getEvent()) {
      case REGISTER:
      case UPDATE:
        ....
      case RESOURCE_OWNER_PASSWORD_CREDENTIALS_REQUEST:
        ....
        executeOnAuthorizationRequest(ropcContext.getParams());
        return;       // <=
      default:
        return;
    }
}

Warning:

V6019 Unreachable code detected. It is possible that an error is present.

Even if this flaw is quite small, a similar case could lead to more serious consequences. I can only note here that a static analyzer will help you avoid such unpleasant things.

I dictate CONDITION

Now, let's look at conditional statements and cases when they execute their code.

Fragment N10

public boolean validatePassword(AuthenticationFlowContext context, 
  UserModel user, MultivaluedMap<String, String> inputData, 
  boolean clearUser) {

    ....

    if (password == null || password.isEmpty()) {
      return badPasswordHandler(context, user, clearUser,true);
    }

    ....

    if (password != null && !password.isEmpty() &&    // <=
      user.credentialManager()
        .isValid(UserCredentialModel.password(password))) { 
      ....
    }
}

Warning:

V6007 Expression 'password != null' is always true.

V6007 Expression '!password.isEmpty()' is always true.

Here are two warnings in one line! What does the analyzer warn us about? The first conditional statement checks that the password is non-null and non-empty. If the opposite occurs, the function is no longer executed. In the line the analyzer highlighted, both of these checks are repeated, so the conditions are always true.

On the one hand, it's better to check than not to check. On the other, such duplicates may lead to missing the part of the condition that may be equal to false—it can really affect the program operation.

Let's repeat the exercise.

Fragment N11

public KeycloakUriBuilder schemeSpecificPart(String ssp)
  throws IllegalArgumentException {
    if (ssp == null) 
      throw new IllegalArgumentException(....);
    ....
    if (ssp != null) // <=
      sb.append(ssp);
    ....
}

Warning:

V6007 Expression 'ssp != null' is always true.

In general, the case is the same. If the ssp variable is null, the program throws an exception. So, the condition below isn't only true all the time but is also redundant because the corresponding code block will always be executed.

The condition in the next fragment is also redundant.

Fragment N12

protected String changeSessionId(HttpScope session) {
  if (!deployment.turnOffChangeSessionIdOnLogin()) 
    return session.getID();     // <=
  else return session.getID();
}

Warning:

V6004 The 'then' statement is equivalent to the 'else' statement.

In this method, the same code is executed under different seasons, the moon phases, and, most importantly, under different conditions. So, again, the condition is redundant here.

Digging into the code, I found the code snippet that is like two drops of water similar to the one above:

protected String changeSessionId(HttpSession session) {
  if (!deployment.turnOffChangeSessionIdOnLogin()) 
    return ChangeSessionId.changeSessionId(exchange, false);
  else return session.getId();
}

As you can see, when the condition is executed, the method that changes the session ID is called.

So, we can make two guesses: either devs just copied the code and changed the condition result, or the first condition still should have updated the session, and this error goes way beyond the "sloppy" code.

But we mustn't live by redundant conditions alone*!*

Fragment N13

static String getParameter(String source, String messageIfNotFound) {
  Matcher matcher = PLACEHOLDER_PARAM_PATTERN.matcher(source);

  while (matcher.find()) {
    return matcher.group(1).replaceAll("'", ""); // <=
  }

  if (messageIfNotFound != null) {
    throw new RuntimeException(messageIfNotFound);
  }

  return null;
}

Warning:

V6037 An unconditional 'return' within a loop.

I have a feeling this while *was raised by *ifs. This code may have some hidden intentions, but the analyzer and I see the same thing here: a loop that always performs one iteration.

The code author might have wanted a different behavioral outcome. Even if they don't want, we'll find this code a bit harder to understand than if there is just a conditional operator.

String comparison

In the following snippet, a developer suggests everything is so easy. But it turns out it's not.

Fragment N14

public boolean equals(Object o) {
  if (this == o) return true;
  if (o == null || getClass() != o.getClass()) return false;

  Key key = (Key) o;

  if (action != key.action) return false;         // <=
  ....
}

Warning:

V6013 Strings 'action' and 'key.action' are compared by reference. Possibly an equality comparison was intended.

Comparing strings implies that we compare their contents. We actually compare object references. This also applies to arrays and collections, not only strings. I think it's clear that in certain cases, code operations can lead to unexpected consequences. Most importantly, it's pretty easy to fix such an error:

public boolean equals(Object o) {
  ....
  if (!action.equals(key.action)) return false;
  ....
}

The equals method returns a comparison exactly to the contents of two strings. Victory!

I'll draw your attention to the fact that the static analyzer has detected the error, which a developer would most likely have missed when reviewing the code. You can read about this and other reasons for using static analysis in this article.

Double-checked locking

Double-checked locking is a parallel design pattern used to reduce the overhead of acquiring a lock. We first check the lock condition without synchronization. If it's encountered, the thread tries to acquire the lock.

If we streamline it, this pattern helps get the lock only when it's actually needed.

I think you've already guessed that there are bugs in the implementation of this template as I've started talking about it. Actually, they are.

Fragment N15

public class WelcomeResource {
  private AtomicBoolean shouldBootstrap;

  ....

  private boolean shouldBootstrap() {
    if (shouldBootstrap == null) {
      synchronized (this) {
        if (shouldBootstrap == null) {
          shouldBootstrap = new AtomicBoolean(....);
        }
      }
    }
    return shouldBootstrap.get();
  }

Warning:

V6082 Unsafe double-checked locking. The field should be declared as volatile.

The analyzer warns that the shouldBootstrap field doesn't have the volatile modifier. What does it affect? In such code, it's likely that different threads use an object until they're fully initialized.

This fact doesn't seem to be that significant, does it? In the next example, the compiler may change the action order with non-volatile fields.

Fragment N16

public class DefaultFreeMarkerProviderFactory 
  implements FreeMarkerProviderFactory {

    private DefaultFreeMarkerProvider provider;   // <=

    ....

    public DefaultFreeMarkerProvider create(KeycloakSession session) {
      if (provider == null) {
        synchronized (this) {
          if (provider == null) {
            if (Config.scope("theme").getBoolean("cacheTemplates", true)) {
              cache = new ConcurrentHashMap<>();
            }
            kcSanitizeMethod = new KeycloakSanitizerMethod();
            provider = new DefaultFreeMarkerProvider(cache, kcSanitizeMethod);
          }
        }
      }
      return provider;
    }
}

Warning:

V6082 Unsafe double-checked locking. The field should be declared as volatile.

Pourquoi les développeurs ne corrigent-ils pas ces fragments de code si le bug est si dangereux ? L’erreur est non seulement dangereuse mais aussi sournoise. Tout fonctionne comme il se doit la plupart du temps. Il existe de nombreuses raisons différentes pour lesquelles un mauvais comportement peut se produire, en raison, par exemple, de l'utilisation de la JVM ou des opérations du planificateur de threads. Il peut donc être assez difficile de reproduire les conditions d'erreur.

Vous pouvez en savoir plus sur ces erreurs et leurs raisons dans l'article.

Conclusion

À la fin de cet article, je tiens à préciser que j'ai utilisé l'analyseur un peu dans le désordre. Je voulais juste vous divertir et vous montrer les bugs du projet. Cependant, pour corriger les erreurs et les éviter, il est préférable d'utiliser régulièrement l'analyseur statique lors de l'écriture du code. Vous pouvez en savoir plus ici.

Cependant, l'analyseur nous a aidé à repérer diverses erreurs liées à la fois au fonctionnement du programme et à une propreté insuffisante du code (je pense toujours que c'est important, et vous ne changerez guère d'avis). Les erreurs ne sont pas la fin du monde si vous les repérez et les corrigez à temps. L'analyse statique y contribue. Essayez PVS-Studio et utilisez-le pour vérifier votre projet gratuitement.

Ce qui précède est le contenu détaillé de. pour plus d'informations, suivez d'autres articles connexes sur le site Web de PHP en chinois!