Perangkap keizinan: apakah yang digunakan oleh Keycloak?

Pengarang: Valerii Filatov

Keizinan dan pendaftaran pengguna adalah bahagian penting dalam mana-mana aplikasi, bukan sahaja untuk pengguna tetapi juga untuk keselamatan. Apakah perangkap yang disembunyikan oleh kod sumber penyelesaian pengurusan identiti sumber terbuka yang popular? Bagaimanakah ia mempengaruhi aplikasi?

Jika anda pernah melaksanakan keizinan untuk apl web, anda mungkin tahu semua isu mengecewakan yang boleh timbul. Saya juga tidak terkecuali.

Pernah sekali, saya melaksanakan kebenaran berasaskan messenger di bahagian hadapan satu projek. Ia kelihatan seperti tugas paling mudah di dunia tetapi ternyata sebaliknya: tarikh akhir semakin hampir, kod tersandung pada API messenger dan orang di sekeliling saya menjerit.

Selepas kes ini, rakan sekerja saya menunjukkan kepada saya alat hebat yang akan menyelaraskan pelaksanaan kebenaran dalam projek masa hadapan kami. Projek ini ialah Keycloak, penyelesaian Java sumber terbuka untuk membolehkan log masuk tunggal (SSO) dengan pengurusan identiti dan akses yang bertujuan untuk apl dan perkhidmatan moden.

Ketika saya menggunakan penyelesaian itu sendiri, saya mendapati menarik untuk masuk ke dalam kod sumber dan menggunakan penganalisis statik PVS-Studio untuk mencari pepijat yang tersembunyi di sana.

He-he, NullPointerException klasik

Seseorang mengetuk pintu. Junior Dev cuba membuka pintu dan terhempas. "NullPointerException!" menyimpulkan Dev Kanan.

Ralat yang berkaitan dengan menyemak null ditemui dalam hampir setiap projek yang telah kami semak sebelum ini. Jadi, mari kita mulakan dengan kesilapan lama tetapi emas ini.

Serpihan 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(....);
    ....
}

Amaran:

V6008 Potensi pembatalan rujukan 'rs'.

Coretan kod ini mempunyai semakan null, tetapi kita perlu memahami perkara yang berlaku di dalam. Jika pembolehubah _oscpFailOpen adalah benar, program tidak akan membuang pengecualian. Ia hanya akan menyimpan log mengenainya dan meneruskan pelaksanaan—ia menerima NullPointerException dalam if berikut, kerana pembolehubah rs telah digunakan di dalam jika.

Nampaknya program ini hanya akan membuang pengecualian lain jika tiada NullPointerException. Tetapi itu tidak berlaku kerana pembolehubah _oscpFailOpen adalah benar dan program harus meneruskan pelaksanaan. Tiada nasib, ia tersandung pada null penunjuk dan termasuk dalam pengecualian.

Serpihan 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"
      );
    ....
}

Amaran*:*

V6060 Rujukan 'attributeValue' telah digunakan sebelum ia disahkan terhadap null.

"Lebih baik lewat daripada tidak sama sekali," frasa ini cukup baik untuk banyak kes, tetapi malangnya, bukan untuk semakan null. Dalam coretan kod di atas, objek attributeValue digunakan sebelum ia disemak untuk kewujudan, yang membawa kepada NullPointerException.

Nota. Jika anda ingin melihat contoh lain pepijat sedemikian, kami telah menyusun senarai untuk anda!

Hujah?

Saya tidak tahu bagaimana ia berlaku, tetapi Keycloak mempunyai ralat yang berkaitan dengan bilangan argumen dalam fungsi rentetan format. Ia bukan statistik ilustrasi—hanya fakta yang menyeronokkan.

Serpihan 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);                // <=
      ....
    }
  }
}

Amaran:

Format V6046 salah. Bilangan item format yang berbeza dijangka. Hujah tidak digunakan: 1.

Dalam serpihan di atas, apabila fungsi printf dipanggil, kami menghantar rentetan format dan nilai untuk digantikan ke dalam rentetan. Hanya ada satu masalah: tiada tempat dalam rentetan untuk menggantikan hujah.

Agak menarik bahawa pembangun bukan sahaja menyalin dan menampal kedua-dua serpihan kod daripada jika kepada jika lain, tetapi pembangun juga menyalin dan menampal ralat di dalam lain jika.

Dalam coretan seterusnya, kita mempunyai kes yang bertentangan: pembangun telah melepaskan hujah.

Serpihan N4

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

Amaran:

Format V6046 salah. Bilangan item format yang berbeza dijangka. Tiada hujah: 4.

Walaupun devs menghantar argumen authSessionId, clientUUID dan authNotesFragment kepada fungsi, argumen tabld keempat adalah sedikit terlepas.

Dalam kes ini, kaedah String.format akan membuang IllegalFormatException, yang boleh menjadi satu kejutan yang tidak menyenangkan.

Kod yang tidak bersih

Amaran PVS-Studio berikut tidak berkaitan dengan ralat kesihatan kod. Amaran ini lebih kepada tahap kualiti tinggi kod tersebut. Saya cuma nyatakan bahawa ini bukan ralat yang jelas.

"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.

Mengapa pembangun tidak membetulkan serpihan kod ini jika pepijat itu sangat berbahaya? Kesilapan itu bukan sahaja berbahaya tetapi juga licik. Semuanya berfungsi sebagaimana mestinya pada kebanyakan masa. Terdapat banyak sebab berbeza mengapa tingkah laku buruk boleh berlaku, disebabkan oleh, katakan, JVM yang digunakan atau operasi penjadual benang. Jadi, agak sukar untuk menghasilkan semula keadaan ralat.

Anda boleh membaca lebih lanjut tentang ralat tersebut dan sebabnya dalam artikel.

Kesimpulan

Pada penghujung artikel ini, saya ingin menyatakan bahawa saya menggunakan penganalisis agak tidak teratur. Saya hanya mahu menghiburkan anda dan menunjukkan kepada anda pepijat dalam projek itu. Walau bagaimanapun, untuk membetulkan ralat dan menghalangnya, adalah lebih baik untuk menggunakan penganalisis statik secara kerap semasa menulis kod. Anda boleh membaca lebih lanjut mengenainya di sini.

Walau bagaimanapun, penganalisis membantu kami mengesan pelbagai ralat yang berkaitan dengan pengendalian program dan kebersihan kod yang tidak mencukupi (saya masih fikir ia penting, dan anda tidak akan mengubah fikiran saya). Ralat bukanlah akhir dunia jika anda melihat dan membetulkannya tepat pada masanya. Analisis statik membantu dengan ini. Cuba PVS-Studio dan gunakannya untuk menyemak projek anda secara percuma.

Atas ialah kandungan terperinci Perangkap keizinan: apakah yang digunakan oleh Keycloak?. Untuk maklumat lanjut, sila ikut artikel berkaitan lain di laman web China PHP!