Security Considerations for Data
When dealing with data in Windows Communication Foundation (WCF), you must consider a number of threat categories. The following table lists the most important threat classes that relate to data processing. WCF provides tools to mitigate these threats.
- Denial of service
- When receiving untrusted data, the data may cause the receiving side to access a disproportionate amount of various resources, such as memory, threads, available connections, or processor cycles by causing lengthy computations. A denial-of-service attack against a server may cause it to crash and be unable to process messages from other, legitimate clients.
- Malicious code execution
- Incoming untrusted data causes the receiving side to run code it did not intend to.
- Information disclosure
- The remote attacker forces the receiving party to respond to its requests in such a way as to disclose more information than it intends to.
User-Provided Code and Code Access Security
A number of places in the Windows Communication Foundation (WCF) infrastructure run code that is provided by the user. For example, the DataContractSerializer serialization engine may call user-provided property set accessors and get accessors. The WCF channel infrastructure may also call into user-provided derived classes of the Message class.
It is the responsibility of the code author to ensure that no security vulnerabilities exist. For example, if you create a data contract type with a data member property of type integer, and in the set accessor implementation allocate an array based on the property value, you expose the possibility of a denial-of-service attack if a malicious message contains an extremely large value for this data member. In general, avoid any allocations based on incoming data or lengthy processing in user-provided code (especially if lengthy processing can be caused by a small amount of incoming data). When performing security analysis of user-provided code, make sure to also consider all failure cases (that is, all code branches where exceptions are thrown).
The ultimate example of user-provided code is the code inside your service implementation for each operation. The security of your service implementation is your responsibility. It is easy to inadvertently create insecure operation implementations that may result in denial-of-service vulnerabilities. For example, an operation that takes a string and returns the list of customers from a database whose name starts with that string. If you are working with a large database and the string being passed is just a single letter, your code may attempt to create a message larger than all available memory, causing the entire service to fail. (An OutOfMemoryException is not recoverable in the .NET Framework and always results in the termination of your application.)
You should ensure that no malicious code is plugged in to the various extensibility points. This is especially relevant when running under partial trust, dealing with types from partially-trusted assemblies, or creating components usable by partially-trusted code. For more information, see "Partial Trust Threats" in a later section.
Note that when running in partial trust, the data contract serialization infrastructure supports only a limited subset of the data contract programming model - for example, private data members or types using the SerializableAttribute attribute are not supported. For more information, see Partial Trust.
Avoiding Unintentional Information Disclosure
When designing serializable types with security in mind, information disclosure is a possible concern.
Consider the following points:
The DataContractSerializer programming model allows the exposure of private and internal data outside of the type or assembly during serialization. Additionally, the shape of a type can be exposed during schema export. Be sure to understand your type's serialization projection. If you do not want anything exposed, disable serializing it (for example, by not applying the DataMemberAttribute attribute in the case of a data contract).
Be aware that the same type may have multiple serialization projections, depending on the serializer in use. The same type may expose one set of data when used with the DataContractSerializer and another set of data when used with the XmlSerializer. Accidentally using the wrong serializer may lead to information disclosure.
Using the XmlSerializer in legacy remote procedure call (RPC)/encoded mode may unintentionally expose the shape of the object graph on the sending side to the receiving side.
Preventing Denial-of-Service Attacks
Causing the receiving side to allocate a significant amount of memory is a potential denial-of-service attack. While this section concentrates on memory consumption issues arising from large messages, other attacks may occur. For example, messages may use a disproportionate amount of processing time.
Denial-of-service attacks are usually mitigated using quotas. When a quota is exceeded, a QuotaExceededException exception is normally thrown. Without the quota, a malicious message may cause all available memory to be accessed, resulting in an OutOfMemoryException exception, or all available stacks to be accessed, resulting in a StackOverflowException.
The quota exceeded scenario is recoverable; if encountered in a running service, the message currently being processed is discarded and the service keeps running and processes further messages. The out-of-memory and stack overflow scenarios, however, are not recoverable anywhere in the .NET Framework; the service terminates if it encounters such exceptions.
Quotas in WCF do not involve any pre-allocation. For example, if the MaxReceivedMessageSize quota (found on various classes) is set to 128 KB, it does not mean that 128 KB is automatically allocated for each message. The actual amount allocated depends on the actual incoming message size.
Many quotas are available at the transport layer. These are quotas enforced by the specific transport channel in use (HTTP, TCP, and so on). While this topic discusses some of these quotas, these quotas are described in detail in Transport Quotas.
Limiting Memory Consumption Without Streaming
The security model around large messages depends on whether streaming is in use. In the basic, non-streamed case, messages are buffered into memory. In this case, use the MaxReceivedMessageSize quota on the TransportBindingElement or on the system-provided bindings to protect against large messages by limiting the maximum message size to access. Note that a service may be processing multiple messages at the same time, in which case they are all in memory. Use the throttling feature to mitigate this threat.
Also note that MaxReceivedMessageSize does not place an upper bound on per-message memory consumption, but limits it to within a constant factor. For example, if the MaxReceivedMessageSize is 1 MB and a 1-MB message is received and then deserialized, additional memory is required to contain the deserialized object graph, resulting in total memory consumption well over 1 MB. For this reason, avoid creating serializable types that could result in significant memory consumption without much incoming data. For example, a data contract "MyContract" with 50 optional data member fields and an additional 100 private fields could be instantiated with the XML construction "<MyContract/>". This XML results in memory being accessed for 150 fields. Note that data members are optional by default. The problem is compounded when such a type is part of an array.
MaxReceivedMessageSize alone is not enough to prevent all denial-of-service attacks. For example, the deserializer may be forced to deserialize a deeply-nested object graph (an object that contains another object that contains yet another one, and so on) by an incoming message. Both the DataContractSerializer and the XmlSerializer call methods in a nested way to deserialize such graphs. Deep nesting of method calls may result in an unrecoverable StackOverflowException. This threat is mitigated by setting the MaxDepth quota to limit the level of XML nesting, as discussed in the "Using XML Safely" section later in the topic.
Setting additional quotas to MaxReceivedMessageSize is especially important when using binary XML encoding. Using binary encoding is somewhat equivalent to compression: a small group of bytes in the incoming message may represent a lot of data. Thus, even a message fitting into the MaxReceivedMessageSize limit may take up much more memory in fully expanded form. To mitigate such XML-specific threats, all of the XML reader quotas must be set correctly, as discussed in the "Using XML Safely" section later in this topic.
Limiting Memory Consumption with Streaming
When streaming, you may use a small MaxReceivedMessageSize setting to protect against denial-of-service attacks. However, more complicated scenarios are possible with streaming. For example, a file upload service accepts files larger than all available memory. In this case, set the MaxReceivedMessageSize to an extremely large value, expecting that almost no data is buffered in memory and the message streams directly to disk. If a malicious message can somehow force WCF to buffer data instead of streaming it in this case, MaxReceivedMessageSize no longer protects against the message accessing all available memory.
To mitigate this threat, specific quota settings exist on various WCF data-processing components that limit buffering. The most important of these is the MaxBufferSize property on various transport binding elements and standard bindings. When streaming, this quota should be set taking into account the maximum amount of memory you are willing to allocate per message. As with MaxReceivedMessageSize, the setting does not put an absolute maximum on memory consumption but only limits it to within a constant factor. Also, as with MaxReceivedMessageSize, be aware of the possibility of multiple messages being processed simultaneously.
The MaxBufferSize property limits any bulk buffering WCF does. For example, WCF always buffers SOAP headers and SOAP faults, as well as any MIME parts found to be not in the natural reading order in an Message Transmission Optimization Mechanism (MTOM) message. This setting limits the amount of buffering in all these cases.
WCF accomplishes this by passing the MaxBufferSize value to the various components that may buffer. For example, some CreateMessage overloads of the Message class take a maxSizeOfHeaders parameter. WCF passes the MaxBufferSize value to this parameter to limit the amount of SOAP header buffering. It is important to set this parameter when using the Message class directly. In general, when using a component in WCF that takes quota parameters, it is important to understand the security implications of these parameters and set them correctly.
The MTOM message encoder also has a MaxBufferSize setting. When using standard bindings, this is set automatically to the transport-level MaxBufferSize value. However, when using the MTOM message encoder binding element to construct a custom binding, it is important to set the MaxBufferSize property to a safe value when streaming is used.
XML-Based Streaming Attacks
MaxBufferSize alone is not enough to ensure that WCF cannot be forced into buffering when streaming is expected. For example, the WCF XML readers always buffer the entire XML element start tag when starting to read a new element. This is done so that namespaces and attributes are properly processed. If MaxReceivedMessageSize is configured to be large (for example, to enable a direct-to-disk large file streaming scenario), a malicious message may be constructed where the entire message body is a large XML element start tag. An attempt to read it results in an OutOfMemoryException. This is one of many possible XML-based denial-of-service attacks that can all be mitigated using XML reader quotas, discussed in the “Using XML Safely” section later in this topic. When streaming, it is especially important to set all of these quotas.
Mixing Streaming and Buffering Programming Models
Many possible attacks arise from mixing streaming and non-streaming programming models in the same service. Suppose there is a service contract with two operations: one takes a Stream and another takes an array of some custom type. Suppose also that MaxReceivedMessageSize is set to a large value to enable the first operation to process large streams. Unfortunately, this means that large messages can now be sent to the second operation as well, and the deserializer buffers data in memory as an array before the operation is called. This is a potential denial-of-service attack: the MaxBufferSize quota does not limit the size of the message body, which is what the deserializer works with.
For this reason, avoid mixing stream-based and non-streamed operations in the same contract. If you absolutely must mix the two programming models, use the following precautions:
Turn off the IExtensibleDataObject feature by setting the IgnoreExtensionDataObject property of the ServiceBehaviorAttribute to true. This ensures that only members that are a part of the contract are deserialized.
Set the MaxItemsInObjectGraph property of the DataContractSerializer to a safe value. This quota is also available on the ServiceBehaviorAttribute attribute or through configuration. This quota limits the number of objects that are deserialized in one deserialization episode. Normally, each operation parameter or message body part in a message contract is deserialized in one episode. When deserializing arrays, each array entry is counted as a separate object.
Set all of the XML reader quotas to safe values. Pay attention to MaxDepth, MaxStringContentLength, and MaxArrayLength and avoid strings in non-streaming operations.
Review the list of known types, keeping in mind that any one of them can be instantiated at any time (see the "Preventing Unintended Types from Being Loaded" section later in this topic).
Do not use any types that implement the IXmlSerializable interface that buffer a lot of data. Do not add such types to the list of known types.
Do not use the XmlElement, XmlNode arrays, Byte arrays, or types that implement ISerializable in a contract.
Do not use the XmlElement, XmlNode arrays, Byte arrays, or types that implement ISerializable in the list of known types.
The preceding precautions apply when the non-streamed operation uses the DataContractSerializer. Never mix streaming and non-streaming programming models on the same service if you are using the XmlSerializer, because it does not have the protection of the MaxItemsInObjectGraph quota.
Slow Stream Attacks
A class of streaming denial-of-service attacks does not involve memory consumption. Instead, the attack involves a slow sender or receiver of data. While waiting for the data to be sent or received, resources such as threads and available connections are exhausted. This situation could arise either as a result of a malicious attack or from a legitimate sender/receiver on a slow network connection.
To mitigate these attacks, set the transport time-outs correctly. For more information, see Transport Quotas. Secondly, never use synchronous Read or Write operations when working with streams in WCF.
Using XML Safely
Secure XML Readers
The XML Infoset forms the basis of all message processing in WCF. When accepting XML data from an untrusted source, a number of denial-of-service attack possibilities exist that must be mitigated. WCF provides special, secure XML readers. These readers are created automatically when using one of the standard encodings in WCF (text, binary, or MTOM).
Some of the security features on these readers are always active. For example, the readers never process document type definitions (DTDs), which are a potential source of denial-of-service attacks and should never appear in legitimate SOAP messages. Other security features include reader quotas that must be configured, which are described in the following section.
When working directly with XML readers (such as when writing your own custom encoder or when working directly with the Message class), always use the WCF secure readers when there is a chance of working with untrusted data. Create the secure readers by calling one of the static factory method overloads of CreateTextReader, CreateBinaryReader, or CreateMtomReader on the XmlDictionaryReader class. When creating a reader, pass in secure quota values. Do not call the Create method overloads. These do not create a WCF reader. Instead, a reader is created that is not protected by the security features described in this section.
The secure XML readers have five configurable quotas. These are normally configured using the ReaderQuotas property on the encoding binding elements or standard bindings, or by using an XmlDictionaryReaderQuotas object passed when creating a reader.
This quota limits the number of bytes that are read in a single Read operation when reading the element start tag and its attributes. (In non-streamed cases, the element name itself is not counted against the quota.) MaxBytesPerRead is important for the following reasons:
The element name and its attributes are always buffered in memory when they are being read. Therefore, it is important to set this quota correctly in streaming mode to prevent excessive buffering when streaming is expected. See the MaxDepth quota section for information about the actual amount of buffering that takes place.
Having too many XML attributes may use up disproportionate processing time because attribute names have to be checked for uniqueness. MaxBytesPerRead mitigates this threat.
This quota limits the maximum nesting depth of XML elements. For example, the document “<A><B><C/></B></A>” has a nesting depth of three. MaxDepth is important for the following reasons:
MaxDepth interacts with MaxBytesPerRead: the reader always keeps data in memory for the current element and all of its ancestors, so the maximum memory consumption of the reader is proportional to the product of these two settings.
When deserializing a deeply-nested object graph, the deserializer is forced to access the entire stack and throw an unrecoverable StackOverflowException. A direct correlation exists between XML nesting and object nesting for both the DataContractSerializer and the XmlSerializer. Use MaxDepth to mitigate this threat.
This quota limits the size of the reader’s nametable. The nametable contains certain strings (such as namespaces and prefixes) that are encountered when processing an XML document. As these strings are buffered in memory, set this quota to prevent excessive buffering when streaming is expected.
This quota limits the maximum string size that the XML reader returns. This quota does not limit memory consumption in the XML reader itself, but in the component that is using the reader. For example, when the DataContractSerializer uses a reader secured with MaxStringContentLength, it does not deserialize strings larger than this quota. When using the XmlDictionaryReader class directly, not all methods respect this quota, but only the methods that are specifically designed to read strings, such as the ReadContentAsString method. The Value property on the reader is not affected by this quota, and thus should not be used when the protection this quota provides is necessary.
This quota limits the maximum size of an array of primitives that the XML reader returns, including byte arrays. This quota does not limit memory consumption in the XML reader itself, but in whatever component that is using the reader. For example, when the DataContractSerializer uses a reader secured with MaxArrayLength, it does not deserialize byte arrays larger than this quota. It is important to set this quota when attempting to mix streaming and buffered programming models in a single contract. Keep in mind that when using the XmlDictionaryReader class directly, only the methods that are specifically designed to read arrays of arbitrary size of certain primitive types, such as ReadInt32Array, respect this quota.
Threats Specific to the Binary Encoding
The binary XML encoding WCF supports includes a dictionary strings feature. A large string may be encoded using only a few bytes. This enables significant performance gains, but introduces new denial-of-service threats that must be mitigated.
There are two kinds of dictionaries: static and dynamic. The static dictionary is a built-in list of long strings that may be represented using a short code in the binary encoding. This list of strings is fixed when the reader is created and cannot be modified. None of the strings in the static dictionary that WCF uses by default are sufficiently large to pose a serious denial-of-service threat, although they may still be used in a dictionary expansion attack. In advanced scenarios where you supply your own static dictionary, be careful when introducing large dictionary strings.
The dynamic dictionaries feature allows messages to define their own strings and associate them with short codes. These string-to-code mappings are kept in memory during the entire communication session, such that subsequent messages do not have to resend the strings and can utilize codes that are already defined. These strings may be of arbitrary length and thus pose a more serious threat than those in the static dictionary.
The first threat that must be mitigated is the possibility of the dynamic dictionary (the string-to-code mapping table) becoming too large. This dictionary may be expanded over the course of several messages, and so the MaxReceivedMessageSize quota offers no protection because it applies only to each message separately. Therefore, a separate MaxSessionSize property exists on the BinaryMessageEncodingBindingElement that limits the size of the dictionary.
Unlike most other quotas, this quota also applies when writing messages. If it is exceeded when reading a message, the QuotaExceededException is thrown as usual. If it is exceeded when writing a message, any strings that cause the quota to be exceeded are written as-is, without using the dynamic dictionaries feature.
Dictionary Expansion Threats
A significant class of binary-specific attacks arises from dictionary expansion. A small message in binary form may turn into a very large message in fully expanded textual form if it makes extensive use of the string dictionaries feature. The expansion factor for dynamic dictionary strings is limited by the MaxSessionSize quota, because no dynamic dictionary string exceeds the maximum size of the entire dictionary.
The MaxNameTableCharCount, MaxStringContentLength, and MaxArrayLength properties only limit memory consumption. They are normally not required to mitigate any threats in the non-streamed usage because memory usage is already limited by MaxReceivedMessageSize. However, MaxReceivedMessageSize counts pre-expansion bytes. When binary encoding is in use, memory consumption could potentially go beyond MaxReceivedMessageSize, limited only by a factor of MaxSessionSize. For this reason, it is important to always set all of the reader quotas (especially MaxStringContentLength) when using the binary encoding.
When using binary encoding together with the DataContractSerializer, the IExtensibleDataObject interface can be misused to mount a dictionary expansion attack. This interface essentially provides unlimited storage for arbitrary data that is not a part of the contract. If quotas cannot be set low enough such that MaxSessionSize multiplied by MaxReceivedMessageSize does not pose a problem, disable the IExtensibleDataObject feature when using the binary encoding. Set the IgnoreExtensionDataObject property to true on the ServiceBehaviorAttribute attribute. Alternatively, do not implement the IExtensibleDataObject interface. For more information, see Forward-Compatible Data Contracts.
The following table summarizes the guidance about quotas.
|Condition||Important quotas to set|
No streaming or streaming small messages, text, or MTOM encoding
MaxReceivedMessageSize, MaxBytesPerRead, and MaxDepth
No streaming or streaming small messages, binary encoding
MaxReceivedMessageSize, MaxSessionSize, and all ReaderQuotas
Streaming large messages, text, or MTOM encoding
MaxBufferSize and all ReaderQuotas
Streaming large messages, binary encoding
MaxBufferSize, MaxSessionSize, and all ReaderQuotas
Transport-level time-outs must always be set and never use synchronous reads/writes when streaming is in use, regardless of whether you are streaming large or small messages.
When in doubt about a quota, set it to a safe value rather than leaving it open.
Preventing Malicious Code Execution
The following general classes of threats can execute code and have unintended effects:
The deserializer loads a malicious, unsafe, or security-sensitive type.
An incoming message causes the deserializer to construct an instance of a normally safe type in such a way that it has unintended consequences.
The following sections discuss these classes of threats further.
(For security information on the XmlSerializer, see the relevant documentation.) The security model for the XmlSerializer is similar to that of the DataContractSerializer, and differs mostly in details. For example, the XmlIncludeAttribute attribute is used for type inclusion instead of the KnownTypeAttribute attribute. However, some threats unique to the XmlSerializer are discussed later in this topic.
Preventing Unintended Types from Being Loaded
Loading unintended types may have significant consequences, whether the type is malicious or just has security-sensitive side effects. A type may contain exploitable security vulnerability, perform security-sensitive actions in its constructor or class constructor, have a large memory footprint that facilitates denial-of-service attacks, or may throw non-recoverable exceptions. Types may have class constructors that run as soon as the type is loaded and before any instances are created. For these reasons, it is important to control the set of types that the deserializer may load.
The DataContractSerializer deserializes in a loosely coupled way. It never reads common language runtime (CLR) type and assembly names from the incoming data. This is similar to the behavior of the XmlSerializer, but differs from the behavior of the NetDataContractSerializer, BinaryFormatter, and the SoapFormatter. Loose coupling introduces a degree of safety, because the remote attacker cannot indicate an arbitrary type to load just by naming that type in the message.
The DataContractSerializer is always allowed to load a type that is currently expected according to the contract. For example, if a data contract has a data member of type
Customer, the DataContractSerializer is allowed to load the
Customer type when it deserializes this data member.
Additionally, the DataContractSerializer supports polymorphism. A data member may be declared as Object, but the incoming data may contain a
Customer instance. This is possible only if the
Customer type has been made "known" to the deserializer through one of these mechanisms:
KnownTypeAttribute attribute applied to a type.
KnownTypeAttribute attribute specifying a method that returns a list of types.
The KnownTypes configuration section.
A list of known types explicitly passed to the DataContractSerializer during construction, if using the serializer directly.
Each of these mechanisms increases the surface area by introducing more types that the deserializer can load. Control each of these mechanisms to ensure no malicious or unintended types are added to the known types list.
Once a known type is in scope, it can be loaded at any time, and instances of the type can be created, even if the contract forbids actually using it. For example, suppose the type "MyDangerousType" is added to the known types list using one of the mechanisms above. This means that:
MyDangerousTypeis loaded and its class constructor runs.
Even when deserializing a data contract with a string data member, a malicious message may still cause an instance of
MyDangerousTypeto create. Code in
MyDangerousType, such as property setters, may run. After this is done, the deserializer tries to assign this instance to the string data member and fail with an exception.
When writing a method that returns a list of known types, or when passing a list directly to the DataContractSerializer constructor, ensure that the code that prepares the list is secure and operates only on trusted data.
If specifying known types in configuration, ensure that the configuration file is secure. Always use strong names in configuration (by specifying the public key of the signed assembly where the type resides), but do not specify the version of the type to load. The type loader automatically picks the latest version, if possible. If you specify a particular version in configuration, you run the following risk: A type may have a security vulnerability that may be fixed in a future version, but the vulnerable version still loads because it is explicitly specified in configuration.
Having too many known types has another consequence: The DataContractSerializer creates a cache of serialization/deserialization code in the application domain, with an entry for each type it must serialize and deserialize. This cache is never cleared as long as the application domain is running. Therefore, an attacker who is aware that an application uses many known types can cause the deserialization of all these types, causing the cache to consume a disproportionately large amount of memory.
Preventing Types from Being in an Unintended State
A type may have internal consistency constraints that must be enforced. Care must be taken to avoid breaking these constraints during deserialization.
The following example of a type represents the state of an airlock on a spacecraft, and enforces the constraint that both the inner and the outer doors cannot be open at the same time.
<DataContract()> _ Public Class SpaceStationAirlock <DataMember()> Private innerDoorOpenValue As Boolean = False <DataMember()> Private outerDoorOpenValue As Boolean = False Public Property InnerDoorOpen() As Boolean Get Return innerDoorOpenValue End Get Set(ByVal value As Boolean) If (value & outerDoorOpenValue) Then Throw New Exception("Cannot open both doors!") Else innerDoorOpenValue = value End If End Set End Property Public Property OuterDoorOpen() As Boolean Get Return outerDoorOpenValue End Get Set(ByVal value As Boolean) If (value & innerDoorOpenValue) Then Throw New Exception("Cannot open both doors!") Else outerDoorOpenValue = value End If End Set End Property End Class
An attacker may send a malicious message like this, getting around the constraints and getting the object into an invalid state, which may have unintended and unpredictable consequences.
<SpaceStationAirlock> <innerDoorOpen>true</innerDoorOpen> <outerDoorOpen>true</outerDoorOpen> </SpaceStationAirlock>
This situation can be avoided by being aware of the following points:
When the DataContractSerializer deserializes most classes, constructors do not run. Therefore, do not rely on any state management done in the constructor.
Use callbacks to ensure that the object is in a valid state. The callback marked with the OnDeserializedAttribute attribute is especially useful because it runs after deserialization is complete and has a chance to examine and correct the overall state. For more information, see Version-Tolerant Serialization Callbacks.
Do not design data contract types to rely on any particular order in which property setters must be called.
Take care using legacy types marked with the SerializableAttribute attribute. Many of them were designed to work with .NET Framework remoting for use with trusted data only. Existing types marked with this attribute may not have been designed with state safety in mind.
Do not rely on the IsRequired property of the DataMemberAttribute attribute to guarantee presence of data as far as state safety is concerned. Data could always be null, zero, or invalid.
Never trust an object graph deserialized from an untrusted data source without validating it first. Each individual object may be in a consistent state, but the object graph as a whole may not be. Furthermore, even if the object graph preservation mode is disabled, the deserialized graph may have multiple references to the same object or have circular references. For more information, see Serialization and Deserialization.
Using the NetDataContractSerializer Securely
The NetDataContractSerializer is a serialization engine that uses tight coupling to types. This is similar to the BinaryFormatter and the SoapFormatter. That is, it determines which type to instantiate by reading the .NET Framework assembly and type name from the incoming data. Although it is a part of WCF, there is no supplied way of plugging in this serialization engine; custom code must be written. The NetDataContractSerializer is provided primarily to ease migration from .NET Framework remoting to WCF. For more information, see the relevant section in Serialization and Deserialization.
Because the message itself may indicate any type can be loaded, the NetDataContractSerializer mechanism is inherently insecure and should be used only with trusted data. It is possible to make it secure by writing a secure, type-limiting type binder that allows only safe types to load (using the Binder property).
Even when used with trusted data, the incoming data may insufficiently specify the type to load, especially if the AssemblyFormat property is set to Simple. Anyone with access to the application’s directory or to the global assembly cache can substitute a malicious type in place of the one that is supposed to load. Always ensure the security of your application’s directory and of the global assembly cache by correctly setting permissions.
In general, if you allow partially trusted code access to your NetDataContractSerializer instance or otherwise control the surrogate selector (ISurrogateSelector) or the serialization binder (SerializationBinder), the code may exercise a great deal of control over the serialization/deserialization process. For example, it may inject arbitrary types, lead to information disclosure, tamper with the resulting object graph or serialized data, or overflow the resultant serialized stream.
Another security concern with the NetDataContractSerializer is a denial of service, not a malicious code execution threat. When using the NetDataContractSerializer, always set the MaxItemsInObjectGraph quota to a safe value. It is easy to construct a small malicious message that allocates an array of objects whose size is limited only by this quota.
The XmlSerializer security model is similar to that of the DataContractSerializer. A few threats, however, are unique to the XmlSerializer.
The XmlSerializer generates serialization assemblies at runtime that contain code that actually serializes and deserializes; these assemblies are created in a temporary files directory. If some other process or user has access rights to that directory, they may overwrite the serialization/deserialization code with arbitrary code. The XmlSerializer then runs this code using its security context, instead of the serialization/deserialization code. Make sure the permissions are set correctly on the temporary files directory to prevent this from happening.
The XmlSerializer also has a mode in which it uses pre-generated serialization assemblies instead of generating them at runtime. This mode is triggered whenever the XmlSerializer can find a suitable serialization assembly. The XmlSerializer checks whether or not the serialization assembly was signed by the same key that was used to sign the assembly that contains the types being serialized. This offers protection from malicious assemblies being disguised as serialization assemblies. However, if the assembly that contains your serializable types is not signed, the XmlSerializer cannot perform this check and uses any assembly with the correct name. This makes running malicious code possible. Always sign the assemblies that contain your serializable types, or tightly control access to your application’s directory and the global assembly cache to prevent the introduction of malicious assemblies.
The XmlSerializer can be subject to a denial of service attack. The XmlSerializer does not have a MaxItemsInObjectGraph quota (as is available on the DataContractSerializer). Thus, it deserializes an arbitrary amount of objects, limited only by the message size.
Partial Trust Threats
Note the following concerns regarding threats related to code running with partial trust. These threats include malicious partially-trusted code as well as malicious partially-trusted code in combination with other attack scenarios (for example, partially-trusted code that constructs a specific string and then deserializing it).
When using any serialization components, never assert any permissions before such usage, even if the entire serialization scenario is within the scope of your assert, and you are not dealing with any untrusted data or objects. Such usage may lead to security vulnerabilities.
In cases where partially-trusted code has control over the serialization process, either through extensibility points (surrogates), types being serialized, or through other means, the partially-trusted code may cause the serializer to output a large amount of data into the serialized stream, which may cause Denial of Service (DoS) to the receiver of this stream. If you are serializing data intended for a target that is sensitive to DoS threats, do not serialize partially-trusted types or otherwise let partially-trusted code control serialization.
If you allow partially-trusted code access to your DataContractSerializer instance or otherwise control the Data Contract Surrogates, it may exercise a great deal of control over the serialization/deserialization process. For example, it may inject arbitrary types, lead to information disclosure, tamper with the resulting object graph or serialized data, or overflow the resultant serialized stream. An equivalent NetDataContractSerializer threat is described in the "Using the NetDataContractSerializer Securely" section.
If the DataContractAttribute attribute is applied to a type (or the type marked as
[Serializable]but is not
ISerializable), the deserializer can create an instance of such a type even if all constructors are non-public or protected by demands.
Never trust the result of deserialization unless the data to be deserialized is trusted and you are certain that all known types are types that you trust. Note that known types are not loaded from the application configuration file, (but are loaded from the machine configuration file) when running in partial trust.
If you pass a DataContractSerializer instance with a surrogate added to partially-trusted code, the code can change any modifiable settings on that surrogate.
For a deserialized object, if the XML reader (or the data therein) comes from partially-trusted code, treat the resulting deserialized object as untrusted data.
The fact that the ExtensionDataObject type has no public members does not mean that data within it is secure. For example, if you deserialize from a privileged data source into an object in which some data resides, then hand that object to partially-trusted code, the partially-trusted code can read the data in the ExtensionDataObject by serializing the object. Consider setting IgnoreExtensionDataObject to true when deserializing from a privileged data source into an object that is later passed to partially-trusted code.
Other State Management Concerns
A few other concerns regarding object state management are worth mentioning:
When using the stream-based programming model with a streaming transport, processing of the message occurs as the message arrives. The sender of the message may abort the send operation in the middle of the stream, leaving your code in an unpredictable state if more content was expected. In general, do not rely on the stream being complete, and do not perform any work in a stream-based operation that cannot be rolled back in case the stream is aborted. This also applies to the situation where a message may be malformed after the streaming body (for example, it may be missing an end tag for the SOAP envelope or may have a second message body).
Using the IExtensibleDataObject feature may cause sensitive data to be emitted. If you are accepting data from an untrusted source into data contracts with IExtensibleObjectData and later re-emitting it on a secure channel where messages are signed, you are potentially vouching for data you know nothing about. Moreover, the overall state you are sending may be invalid if you take both the known and unknown pieces of data into account. Avoid this situation by either selectively setting the extension data property to null or by selectively disabling the IExtensibleObjectData feature.
Normally, the process of importing schema to generate types happens only at design time, for example, when using the ServiceModel Metadata Utility Tool (Svcutil.exe) on a Web service to generate a client class. However, in more advanced scenarios, you may process schema at runtime. Be aware that doing so can expose you to denial-of-service risks. Some schema may take a long time to be imported. Never use the XmlSerializer schema import component in such scenarios if schemas are possibly coming from an untrusted source.
Threats Specific to ASP.NET AJAX Integration
When the user implements WebScriptEnablingBehavior or WebHttpBehavior, WCF exposes an endpoint that can accept both XML and JSON messages. However, there is only one set of reader quotas, used both by the XML reader and the JSON reader. Some quota settings may be appropriate for one reader but too large for the other.
A Note on Components
WCF is a flexible and customizable system. Most of the contents of this topic focus on the most common WCF usage scenarios. However, it is possible to compose components WCF provides in many different ways. It is important to understand the security implications of using each component. In particular:
When you must use XML readers, use the readers the XmlDictionaryReader class provides as opposed to any other readers. Safe readers are created using CreateTextReader, CreateBinaryReader, or CreateMtomReader methods. Do not use the Create method. Always configure the readers with safe quotas. The serialization engines in WCF are secure only when used with secure XML readers from WCF.
When using the DataContractSerializer to deserialize potentially untrusted data, always set the MaxItemsInObjectGraph property.
When creating a message, set the maxSizeOfHeaders parameter if MaxReceivedMessageSize does not offer enough protection.
When creating an encoder, always configure the relevant quotas, such as MaxSessionSize and MaxBufferSize.
When using an XPath message filter, set the NodeQuota to limit the amount of XML nodes the filter visits. Do not use XPath expressions that could take a long time to compute without visiting many nodes.
In general, when using any component that accepts a quota, understand its security implications and set it to a safe value.