Internet Draft James Pinkerton Document: draft-ietf-rddp-security-00.txt Microsoft Corporation Expires: April, 2004 Ellen Deleganes Intel Corporation Allyn Romanow Cisco Systems Bernard Aboba Microsoft Corporation October 2003 DDP/RDMAP Security 1 Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. 2 Abstract This document analyzes security issues around implementation and use of the Direct Data Placement Protocol(DDP) and Remote Direct Memory Access Protocol (RDMAP). It first defines an architectural model for an RDMA Network Interface Card (RNIC), which can implement DDP or RDMAP and DDP. The model includes a definition of resources that can be attacked. This document then introduces various Trust Models between a local peer and a remote peer and the tools that can be used to create countermeasures against attacks. Finally, the document reviews various attacks and the countermeasures to be used against them, grouping the attacks into spoofing, tampering, information disclosure, denial of service, and elevation of privilege. J. Pinkerton, et al. Expires April 2004 1 Internet-Draft RDDP/RDMAP Security October 2003 Table of Contents 1 Status of this Memo.........................................1 2 Abstract....................................................1 2.1 Issues......................................................3 3 Introduction................................................5 4 Architectural Model.........................................7 4.1 Components..................................................8 4.2 Resources..................................................10 4.2.1 Connection Context Memory.................................10 4.2.2 Data Buffers..............................................10 4.2.3 Page Translation Tables...................................11 4.2.4 STag Namespace............................................11 4.2.5 Completion Queues.........................................11 4.2.6 RDMA Read Request Queue...................................11 4.2.7 RDMA Asynchronous Event Queue.............................11 4.2.8 RNIC Control Interactions.................................12 4.2.9 Initialization of RNIC Data Structures for Data Transfer..12 4.2.10 RNIC Data Transfer Interactions.........................13 4.3 System Properties..........................................14 5 Trust Models...............................................15 6 Attacker Capabilities......................................17 7 Attacks and Countermeasures................................18 7.1 Tools for Countermeasures..................................18 7.1.1 Protection Domain (PD)....................................18 7.1.2 Limiting STag Scope.......................................19 7.1.3 Access Rights.............................................19 7.1.4 Limiting the Scope of the Completion Queue................20 7.1.5 Limiting the Scope of an Error............................20 7.2 Spoofing...................................................20 7.2.1 Connection Hijacking......................................20 7.2.2 Using an STag on a different connection...................21 7.3 Tampering..................................................22 7.3.1 RDMA Write or Read Response to Memory Outside the Buffer..22 7.3.2 Modifying a Buffer After Indicating Contents Are Ready....23 7.3.3 Using Multiple Stags to access the same buffer............23 7.4 Information Disclosure.....................................23 7.4.1 Probing memory outside of the buffer bounds...............24 7.4.2 Using RDMA Read to Access Stale Data......................24 7.4.3 Accessing a buffer after the transfer is over.............24 7.4.4 Accessing data within a valid STag that was unintended....24 7.4.5 Using RDMA Read on a buffer meant only for RDMA Write.....25 7.4.6 Using Multiple Stags to access the same buffer............25 7.4.7 Remote node loading firmware onto the RNIC................26 7.4.8 Controlling Access to Page Translation Table and STag Mapping 26 7.5 Denial of Service (DOS)....................................26 7.5.1 RNIC Resource Consumption.................................27 7.5.2 Resource Consumption By Active Applications...............27 7.5.2.1 Receive Data Buffers..................................28 J. Pinkerton, et al. Expires - April 2004 [Page 2] Internet-Draft RDDP/RDMAP Security October 2003 7.5.2.2 Completion Queue (CQ) Resource Consumption............29 7.5.2.3 RDMA Read Request Queue...............................31 7.5.3 Resource Consumption by Idle Applications.................32 7.5.4 Exercise of non-optimal code paths........................32 7.5.5 Remote Invalidation of an STag Shared Across Multiple Connections......................................................32 7.5.6 Remote Peer Consumes too many Untagged Receive Buffers....33 7.6 Elevation of Privilege.....................................33 7.6.1 Loading Firmware into the RNIC............................34 8 Security Services for RDDP.................................35 8.1 Introduction to Security Options...........................35 8.1.1 Introduction to IPsec.....................................35 8.1.2 Introduction to SSL Limitations on RDMAP..................36 8.1.3 Applications Which Provide Security.......................36 8.1.4 Authentication Only.......................................36 8.1.5 Privacy...................................................36 8.2 Recommendations for IPsec Encapsulation of RDDP............36 9 Summary Table of Attacks and Trust Models..................38 10 References.................................................40 10.1 Normative References......................................40 10.2 Informative References....................................40 11 AuthorÆs Addresses.........................................41 12 Acknowledgments............................................42 13 Full Copyright Statement...................................43 Table of Figures Figure 1 - RDMA Security Model....................................8 Figure 2 - Summary Attacks and Trust Model Table.................39 2.1 Issues This section is temporary and will go away when all issues have been resolved. Note: this is far from a complete list of issues; as more are raised, they will be added to this list until some sort of consensus is reached. They are in the order found in the specification. Issue: Discuss issues in allowing the non-privileged consumer to control mapping the Page Translation Table and Stag..............26 Issue: Need to analyze the case of sharing a queue of Untagged receive buffers across multiple connections, and that the Remote Peer can mount a denial of service attack........................33 Issue: Security Services section is a placeholder for now........35 Issue: Guidance for application protocols like NFS which implement security.........................................................36 J. Pinkerton, et al. Expires - April 2004 [Page 3] Internet-Draft RDDP/RDMAP Security October 2003 Issue: IPsec recommendations for RDMAP/DDP.......................36 Issue: Finish Summary table of Attacks/Trust Models..............38 J. Pinkerton, et al. Expires - April 2004 [Page 4] Internet-Draft RDDP/RDMAP Security October 2003 3 Introduction RDMA enables new levels of flexibility when communicating between two parties compared to current conventional networking practice (e.g. a stream-based model or datagram model). This flexibility brings new security issues that must be carefully understood when designing application protocols utilizing RDMA and when implementing RDMA-aware NICs (RNICs). Note that for the purposes of this security analysis, an RNIC may implement RDMAP and DDP, or just DDP. The specification first develops an architectural model that is relevant for the security analysis - it details components, resources, and system properties that may be attacked. The specification then defines Partial Trust Models. Partial Trust is defined as: Partial Trust - When one party is willing to assume that another party will refrain from a specific attack or set of attacks, the parties are said to be in a state of Partial Trust. Note that the partially trusted peer may attempt a different set of attacks. This may be appropriate for many applications where any adverse effects of the betrayal is easily confined and does not place other clients or applications at risk. An Untrusted Peer relationship is appropriate when an application wishes to ensure that it will be robust and uncompromised even in the face of a deliberate attack by its peer. For example, a single application that concurrently supports multiple unrelated sessions would presumably treat each of its peers as an Untrusted Peer. For the Untrusted peer, a brief list of capabilities is enumerated. The rest of the specification is focused on analyzing attacks. First, the tools for mitigating attacks are listed, and then a series of attacks on components, resources, or system properties is enumerated. For each attack, possible countermeasures are reviewed. Applications within a host are divided into two categories - Privileged and Non-Privileged. Both application types can send and receive data and request resources. The key differences between the two are: The Privileged Application is Partially Trusted. It is assumed that the Privileged Application will not intentionally attack the system (e.g., it is a kernel application), although it may be greedy for resources. A Non-Privileged ApplicationÆs capabilities are a logical sub- set of the Privileged ApplicationÆs. It is assumed by the local host infrastructure that a Non-Privileged Application is J. Pinkerton, et al. Expires - April 2004 [Page 5] Internet-Draft RDDP/RDMAP Security October 2003 Untrusted. All Non-Privileged Application interactions with the RNIC Engine that could affect other applications need to be done through a Trusted intermediary that can verify the Non- Privileged Application requests. J. Pinkerton, et al. Expires - April 2004 [Page 6] Internet-Draft RDDP/RDMAP Security October 2003 4 Architectural Model This section describes an RDMA architectural reference model that is used as security issues are examined. It introduces the components of the model, the resources that can be attacked, the types of interactions possible between components and resources, and the system properties, which should be preserved when under attack. Figure 1 shows the components comprising the architecture and the interfaces where potential security attacks could be launched. External attacks can be injected into the system from an application that sits above the RI or from the Internet. The intent here is to describe high level components and capabilities which affect threat analysis, and not focus on specific implementation options. Also note that the architectural model is an abstraction, and an actual implementation may choose to subdivide its components along different boundary lines than defined here. For example, the Privileged Resource Manager may be partially or completely encapsulated in the Privileged Application. Regardless, it is expected that the security analysis of the potential threats and countermeasures still apply. J. Pinkerton, et al. Expires - April 2004 [Page 7] Internet-Draft RDDP/RDMAP Security October 2003 +-------------+ Request Proxy Interface | Privileged |<--------------------------------+ | Resource | | Admin<-->| Manager | App Control Interface | | |<------+-------------------+ | +-------------+ | | | ^ v v v | +-------------+ +-----------------+ | | Privileged | | Non-Privileged | | | Application | | Application | | +-------------+ +-----------------+ | ^ ^ |Privileged |Privileged |Non-Privileged |Control |Data |Data |Interface |Interface |Interface RNIC | | | Interface(RI) v v v ================================================================= +-----------------------------------------+ | | | RNIC Engine | <------ Firmware | | +-----------------------------------------+ ^ | v Internet Figure 1 - RDMA Security Model 4.1 Components The components shown in Figure 1 - RDMA Security Model are: * RNIC Engine - the component that implements the RDMA protocol and/or DDP protocol. * Privileged Resource Manager - the component responsible for managing and allocating resources associated with the RNIC Engine. The Resource Manager does not send or receive data. Note that whether the Resource Manager is an independent component, part of the RNIC, or part of the application is implementation dependent. If a specific implementation does not wish to address security issues resolved by the Resource Manager, there may in fact be no resource manager at all. * Privileged Application - See Section 3 Introduction for a definition of Privileged Application. The local host J. Pinkerton, et al. Expires - April 2004 [Page 8] Internet-Draft RDDP/RDMAP Security October 2003 infrastructure can enable the Privileged Application to map a data buffer directly from the RNIC Engine to the host through the RNIC Interface, but it does not allow the Privileged Application to directly consume RNIC Engine resources. * Non-Privileged Application - See Section 3 Introduction for a definition of Non-Privileged Application. All Non- Privileged Application interactions with the RNIC Engine that could affect other applications MUST be done using the Privileged Resource Manager as a proxy. A design goal of the DDP and RDMAP protocols is to allow, under constrained conditions, Non-Privileged applications to send and receive data directly to/from the RDMA Engine without Privileged Resource Manager intervention - while ensuring that the host remains secure. Thus, one of the primary goals of this paper is to analyze this usage model for the enforcement that is required in the RNIC Engine to ensure the system remains secure. The host interfaces that could be exercised include: * Control Interface - A Privileged Resource Manager uses the RI to allocate and manage RNIC Engine resources, control the state within the RNIC Engine, and monitor various events from the RNIC Engine. It also uses this interface to act as a proxy for some operations that a Non-Privileged Application may require (after performing appropriate countermeasures). * Non-Privileged Data Transfer Interface - A Non-Privileged Application uses this interface to initiate and to check the status of data transfer operations. * Privileged Data Transfer Interface - A superset of the functionality provided by the Non-Privileged Data Transfer Interface. The application is allowed to directly manipulate RNIC Engine mapping resources to map an STag to an application data buffer. * Request Proxy Interface - a Non-Privileged Application uses this interface to control RNIC Engine resources that could affect other applications - such as manipulating the RNIC Engine's mapping of an STag to an application data buffer. The Privileged Resource Manager implements countermeasures to ensure that if the Non-Privileged Application launches an attack it can prevent the attack from affecting other applications. J. Pinkerton, et al. Expires - April 2004 [Page 9] Internet-Draft RDDP/RDMAP Security October 2003 * Figure 1 also shows the ability to load new firmware in the RNIC Engine. Not all RNICs will support this, but it is shown for completeness and is also reviewed under potential attacks. If Internet control messages, such as ICMP, ARP, RIPv4, etc. are processed by the RNIC Engine, the threat analyses for those protocols is also applicable, but outside the scope of this paper. 4.2 Resources This section describes the primary resources in the RNIC Engine that could be affected if under attack. For RDMAP, all of the defined resources apply. For DDP, all of the resources except the RDMA Read Queue apply. 4.2.1 Connection Context Memory The state information for each connection is maintained in memory, which could be located in a number of places - on the NIC, inside RAM attached to the NIC, in host memory, or in any combination of the three, depending on the implementation. Connection Context Memory includes state associated with Data Buffers. For Tagged Buffers, this includes how STag names, Data Buffers, and Page Translation Tables inter-relate. It also includes the FIFO list of Untagged Data Buffers posted for reception of Untagged Messages (referred to in some contexts as the Receive Queue), and a list of operations to perform to send data (referred to in some contexts as the Send Queue). 4.2.2 Data Buffers There are two different ways to expose a data buffer; a buffer can be exposed for receiving RDMAP Send Type Messages (a.k.a. DDP Untagged Messages) on DDP Queue zero or the buffer can be exposed for remote access through STags (a.k.a. DDP Tagged Messages). This distinction is important because the attacks and the countermeasures used to protect against the attack are different depending on the method for exposing the buffer to the Internet. For the purposes of the security discussion, a single logical Data Buffer is exposed with a single STag. Actual implementations may support scatter/gather capabilities to enable multiple physical data buffers to be accessed with a single STag, but from a threat analysis perspective it is assumed that a single STag enables access to a single logical Data Buffer. J. Pinkerton, et al. Expires - April 2004 [Page 10] Internet-Draft RDDP/RDMAP Security October 2003 In any event, it is the responsibility of the RI to ensure that no STag can be created that exposes memory that the consumer had no authority to expose. 4.2.3 Page Translation Tables Page Translation Tables are the structures used by the RNIC to be able to access application memory for data transfer operations. Even though these structures are called "Page" Translation Tables, they may not reference a page at all - conceptually they are used to map an application address space representation of a buffer to the physical addresses that are used by the RNIC Engine to move data. If on a specific system, a mapping is not used, then a subset of the attacks examined may be appropriate. 4.2.4 STag Namespace The DDP specification defines a 32-bit namespace for the STag. Implementations may vary in terms of the actual number of STags that are supported. In any case, this is a bounded resource that can come under attack. Depending upon Stag namespace allocation algorithms, the actual name space to attack may be significantly less than 2^32. 4.2.5 Completion Queues Completion Queues are used in this specification to conceptually represent how the RNIC Engine notifies the Application of the completion of the transmission of data, or the completion of the reception of data through the Data Transfer Interface. Because there could be many transmissions or receptions in flight at any one time, completions are modeled as a queue rather than a single event. An implementation may also use the Completion Queue to notify the application of other activities, for example, the completion of a mapping of an STag to a specific application buffer. 4.2.6 RDMA Read Request Queue The RDMA Read Request Queue is the memory holding state information for one or more RDMA Read Request Messages that have arrived, but for which the RDMA Read Response Messages have not yet been completely sent. 4.2.7 RDMA Asynchronous Event Queue The Asynchronous Event Queue is a queue from the RNIC to the Privileged Resource Manager of bounded size. It is used by the RNIC to notify the host of various events which might require management action, including protocol violations, connection state changes, local operation errors, low water marks on receive queues, and possibly other events. J. Pinkerton, et al. Expires - April 2004 [Page 11] Internet-Draft RDDP/RDMAP Security October 2003 The RDMA Event Queue is a resource that can be attacked because Remote or Local Peers can cause events to occur which have the potential of overflowing the queue. Note that an implementation is at liberty to implement the functions of the RDMA Asynchronous Event Queue in a variety of ways, including multiple queues or even simple callbacks. All vulnerabilities identified for a single Asynchronous Event Queue apply to specific- purpose subsets. A callback function is simply a very short queue. 4.2.8 RNIC Control Interactions With RNIC resources and interfaces defined, it is now possible to examine the interactions supported by the generic RNIC functional interfaces through each of the 3 interfaces - Privileged Control Interface, Privileged Data Interface, and Non-Privileged Data Interface. Generically, the Privileged Control Interface controls the RNICÆs allocation, deallocation, and initialization of RNIC global resources. This includes allocation and deallocation of Connection Context Memory, Page Translation Tables, STag names, Completion Queues, and RDMA Read Request Queues. Event Queue Initialization and removal of Page Translation Table resources. 4.2.9 Initialization of RNIC Data Structures for Data Transfer Initialization of the mapping between an STag and a Data Buffer can be viewed in the abstract as two separate opertions: a. Initialization of the allocated Page Translation Table entries with the location of the Data Buffer, and b. Initialization of a mapping from an allocated STag name to a set of Page Translation Table entry(s) or partial-entries. Note that an implementation may not have a Page Translation Table (i.e. it may support a direct mapping between an STag and a Data Buffer). In this case threats and mitigations associated with the Page Translation Table are not relevent. Initialization of the contents of the Page Translation Table can be done by either the Privileged Application or by the Privileged Resource Manager as a proxy for the Non-Privileged Application. By definition the Non-Privileged Application is not trusted to directly J. Pinkerton, et al. Expires - April 2004 [Page 12] Internet-Draft RDDP/RDMAP Security October 2003 manipulate the Page Translation Table. In general the concern is that the Non-Privileged application may try to maliciously initialize the Page Translation Table to access a buffer for which it does not have permission. The exact resource allocation algorithm for the Page Translation Table is outside the scope of this specification. It may be allocated for a specific Data Buffer, or be allocated as a pooled resource to be consumed by potentially multiple Data Buffers, or be managed in some other way. This paper attempts to abstract implementation dependent issues, and focus on higher level security issues such as resource starvation and sharing of resources between connections. The next issue is how an STag name is associated with a Data Buffer. For the case of an Untagged Data Buffer, there is no wire visible mapping between an STag name and a Data Buffer. Note that there may, in fact, be a mapping that is not visible from the wire, but this is a local host specific issue which should be analyzed in the context of local host implementation specific security analysis, and thus is outside the scope of this paper. For a Tagged Data Buffer, either the Privileged Application, the Non-Privileged Application, or the Privileged Resource Manager acting on behalf of the Non-Privileged Resource Manager may initialize a mapping from an STag to a Page Translation Table, or may have the ability to simply enable/disable an existing STag to Page Translation Table mapping. There may also be multiple STag names which map to a specific group of Page Translation Table entries (or sub-entries). Specific security issues with this level of flexibility are examined later. There are a variety of implementation options for initialization of Page Translation Table entries and mapping an STag to a group of Page Translation Table entries which have security repercussions. This includes support for separation of Mapping an STag verses mapping a set of Page Translation Table entries, and support for Applications directly manipulating STag to Page Translation Table entry mappings (verses requiring access through the Privileged Resource Manager). 4.2.10 RNIC Data Transfer Interactions RNIC Data Transfer operations can be subdivided into send operations and receive operations. For send operations, there is typically a queue that enables the Application to post multiple operations. Depending upon the implementation, Data Buffers used in the operations may or may not have Page Translation Table entries associated with them, and may or J. Pinkerton, et al. Expires - April 2004 [Page 13] Internet-Draft RDDP/RDMAP Security October 2003 may not have STags associated with them. Because this is a local host specific implementation issue rather than a protocol issue, the security analysis of threats and mitigations is left to the host implementation. Receive operations are different for Tagged Data Buffers verses Untagged Data Buffers. If more than one Untagged Data Buffer can be posted by the Application, the DDP specification requires that they be consumed in FIFO order. Thus the most general implementation is that there is a FIFO queue of receive Untagged Data Buffers. Some implementations may also support sharing of the FIFO queue between multiple connections. In this case defining ææFIFOÆÆ becomes non- trivial - in general the buffers for a single stream are consumed from the queue in the order that they were placed on the queue, but there is no order guarantee between streams. For receive Tagged Data Buffers, at some time prior to data transfer, the mapping of the STag to specific Page Translation Table entries (if present) and the mapping from the Page Translation Table entries to the Data Buffer must have been initialized (see the prior section for interaction details). 4.3 System Properties System properties that can be attacked included system integrity, system stability (liveness, large fluctuations in performance), and confidentiality. J. Pinkerton, et al. Expires - April 2004 [Page 14] Internet-Draft RDDP/RDMAP Security October 2003 5 Trust Models The Trust Models described in this section have three primary distinguishing characteristics. * Local Resource Sharing (yes/no) - When local resources are shared, they are shared across a grouping of RDMAP/DDP Streams. If local resources are not shared, the resources are dedicated on a per Stream basis. Resources are defined in Section 4.2 - Resources on page 10. The advantage of not sharing resources between Streams is that it reduces the types of attacks that are possible. The disadvantage is that applications might run out of resources. * Local Partial Trust (yes/no) - Local Partial Trust is determined based on whether the local grouping of RDMAP/DDP Streams (which typically equates to one application or group of applications) mutually trust each other to not perform a specific set of attacks. * Remote Partial Trust (yes/no) - The Remote Partial Trust level is determined based on whether the Local Peer of a specific RDMAP/DDP Stream partially trusts the Remote Peer of the Stream (see the definition of Partial Trust in Section 3 Introduction). It is assumed in this paper that the component that implements the mechanism to control sharing of RNIC Engine resources is the Privileged Resource Manager. The RNIC Engine exposes its resources through the RI to the Privileged Resource Manager. All Privileged and Non-Privileged applications request resources from the Resource Manager. The Resource Manager implements resource management policies to ensure fair access to resources. The Resource Manager should be designed to take into account security attacks detailed in this specification. The sharing of resources across connections should be under the control of the application, both in terms of the Trust Model the application wishes to operate under, as well as the level of resource sharing the application wishes to give Local Peer processes. Not all of the combinations of the trust characteristics are expected to be used by applications. This paper specifically analyzes five application Trust Models that are expected to be in common use. The Trust Models are as follows: 1. NS-NT - Non-Shared Local Resources, no Local Trust, no Remote Trust - typically a server application that wants to run in a mode that has the least number of potential attacks. J. Pinkerton, et al. Expires - April 2004 [Page 15] Internet-Draft RDDP/RDMAP Security October 2003 2. NS-RT - Non-Shared Local Resources, no Local Trust, Remote Partial Trust - typically a peer-to-peer application, which has, by some method outside of the scope of this specification, authenticated the Remote Peer. Note that unless some form of key based authentication is used on a per packet basis, it may not be possible to bind the authentication result to the RDMA packet. 3. S-NT - Shared Local Resources, no Local Trust, no Remote Trust - typically a server application that runs in an untrusted environment where the amount of resources required is either too large or too dynamic to dedicate for each RDMAP/DDP Stream. 4. S-LT - Shared Local Resources, Local Partial Trust, no Remote Trust - typically an application, which provides a session layer and uses multiple Streams, to provide additional throughput or fail-over capabilities. All of the Streams within the local application partially trust each other, but do not trust the remote peer. 5. S-T - Shared Local Resources, Local Partial Trust, Remote Partial Trust - typically a distributed application, such as a distributed database application or a High Performance Computer (HPC) application, which is intended to run on a cluster. Due to extreme resource and performance requirements, the application typically authenticates with all of its peers and then runs in a highly trusted environment. The application peers are all in a single application fault domain and depend on one another to be well-behaved when accessing data structures. If a trusted Remote Peer has an implementation defect that results in poor behavior, the entire application could be corrupted. Models NS-NT and S-NT above are typical for Internet networking - neither Local Peers nor the Remote Peer is trusted. Sometimes optimizations can be done that enable sharing of Page Translation Tables across multiple Local Peers, thus Model S-LT can be advantageous. Model S-T is typically used when resource scaling across a large parallel application makes it infeasible to use any other model. Resource scaling issues can either be due to performance around scaling or because there simply are not enough resources. Model NS-RT is probably the least likely model to be used, but is presented for completeness. J. Pinkerton, et al. Expires - April 2004 [Page 16] Internet-Draft RDDP/RDMAP Security October 2003 6 Attacker Capabilities An attackerÆs capabilities delimit the types of attacks that attacker is able to launch. RDMAP and DDP require that the initial LLP Stream (and connection) be set up prior to transferring RDMAP/DDP Messages. For the attacker to actively generate an RDMAP/DDP protocol attack, it must have the capability to both send and receive messages. Attackers with send only capabilities should be addressed by the LLP, not by RDMAP/DDP. J. Pinkerton, et al. Expires - April 2004 [Page 17] Internet-Draft RDDP/RDMAP Security October 2003 7 Attacks and Countermeasures This section describes the attacks that are possible against the RDMA system defined in Figure 1 - RDMA Security Model and the RNIC Engine resources defined in Section 4.2. The analysis includes a detailed description of each attack, the Trust Models the attack applies to (see Section 5 for a description of the Trust Models), and a description of the countermeasures appropriate to the Trust Model(s) that can be taken to thwart the attack. Note that, connection setup and teardown is presumed to be done in stream mode (i.e. no RDMA encapsulation of the payload), so there are no new attacks related to connection setup/teardown beyond what is already present in the LLP (e.g. TCP or SCTP). Consequently, any existing analysis of Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, or Elevation of Privilege continues to apply. Thus, the analysis in this section focuses on attacks that are present regardless of the LLP Stream type. 7.1 Tools for Countermeasures The tools described in this section are the primary mechanisms that can be used to provide countermeasures to potential attacks. 7.1.1 Protection Domain (PD) Protection Domains are associated with two of the resources of concern, connection context memory and STags associated with Page Translation Table entries and data buffers. Protection Domains are used mainly to ensure that an STag can only be used to access the associated data buffer through connections in the same Protection Domain as that STag. For the Trust Models that are defined to have non-shared resources (Trust Models NS-NT and NS-RT), it is recommended that each Stream be associated with its own, unique Protection Domain. For those Trust Models where resources are shared (Trust Models S-NT, S-LT and S-T), it is recommended that Protection Domain be limited to the number of Streams that share the same Trust Model. Note that an application (either Privileged or Non-Privileged) can potentially have multiple Protection Domains. This could be used, for example, to ensure that multiple clients of a server do not have the ability to corrupt each other. This can apply to any combination of the Trust Models, because Partial Trust does not imply complete trust. J. Pinkerton, et al. Expires - April 2004 [Page 18] Internet-Draft RDDP/RDMAP Security October 2003 7.1.2 Limiting STag Scope The key to protecting a local data buffer is to limit the scope of its STag to the level appropriate for the Trust Model. The scope of the STag can be measured in multiple ways. * Number of Connections and/or Streams on which the STag is valid. One way to limit the scope of the STag is to limit the connections and/or Streams that are allowed to use the STag. As noted in the previous section, use of Protection Domains appropriately can limit the scope of the STag. It is also possible to create an STag that is valid only on a single connection, even in the case where several connections are associated with the Protection Domain of the STag. * Limit the time an STag is valid. By Invalidating an Advertised STag (e.g., revoking remote access to the buffers described by an STag when done with the transfer), an entire class of attacks can be eliminated. * Limit the buffer the STag can reference. Limiting the scope of an STag access to *just* the intended buffers to be exposed is critical to prevent certain forms of attacks. * Allocating Stag numbers in an unpredictable way. If STags are allocated using an algorithm which makes it hard for the Remote Peer to guess which STag(s) are currently in use, it makes it more difficult for an attacker to guess the correct value. 7.1.3 Access Rights Access Rights associated with a specific Advertised STag or RDMAP/DDP Stream provide another mechanism for applications to limit the attack capabilities of the Remote Peer. The Local Peer can control whether a data buffer is exposed for local only, or local and remote access, and assign specific access privileges (read, write, read and write). For DDP, when an STag is advertised, the Remote Peer is presumably given write access rights to the data (otherwise there was not much point to the advertisement). For RDMAP, when an application advertises an STag, it can enable write-only, read-only, or both write and read access rights. Similarly, some applications may wish to provide a single buffer with different access rights on a per-connection or per-Stream basis. For example, some connections may have read-only access, some J. Pinkerton, et al. Expires - April 2004 [Page 19] Internet-Draft RDDP/RDMAP Security October 2003 may have remote read and write access, while on other connections only the Local Peer is allowed access. 7.1.4 Limiting the Scope of the Completion Queue Completions associated with sending and receiving data, or setting up buffers for sending and receiving data, could be accumulated in a shared Completion Queue for a group of RDMAP/DDP Streams, or a specific RDMAP/DDP Stream could have a dedicated Completion Queue. Limiting Completion Queue association to one, or a small number of RDMAP/DDP Streams can prevent several forms of Denial of Service attacks. 7.1.5 Limiting the Scope of an Error To prevent a variety of attacks, it is important that an RDMAP/DDP implementation be robust in the face of errors. If an error on a specific Stream can cause other unrelated Streams to fail, then a broad class of attacks are enabled against the implementation. 7.2 Spoofing Because the RDMAP Stream is only offloaded if it is in the ESTABLISHED state, certain types of traditional forms of wire attacks do not apply -- an end-to-end handshake must have occurred to establish the RDMAP Stream. So, the only form of spoofing that applies is one when a remote node can both send and receive packets. 7.2.1 Connection Hijacking If a man-in-the-middle attacker has the ability to inject packets which will be accepted by the LLP (e.g., TCP sequence number is correct) then the connection can essentially be hijacked. One style of attack is for the man-in-the-middle to send Tagged Messages (either RDMAP or DDP). If it can discover a buffer that has been exposed for STag enabled access, then the man-in-the-middle can use an RDMA Read operation to read the contents of the associated data buffer, perform an RDMA Write Operation to modify the contents of the associated data buffer, or invalidate the STag to disable further access to the buffer. The only countermeasure for this form of attack is to either secure the RDMAP/DDP Stream or attempt to provide physical security to prevent man-in-the-middle type access. The best protection against this form of attack is end-to-end integrity protection and authentication, such as IPsec (see Section 8 Security Services for RDDP on page 35), to prevent spoofing or tampering. If authentication is not used, then a man-in-the-middle attack can occur, enabling spoofing, tampering, and repudiation. J. Pinkerton, et al. Expires - April 2004 [Page 20] Internet-Draft RDDP/RDMAP Security October 2003 Because the connection itself is established by the LLP, some LLPs are more difficult to hijack than others. Please see the relevant LLP documentation on security issues around connection hijacking . Another approach is to restrict access to only the local subnet/link, and provide some mechanism to limit access, such as physical security or 802.1.x. This model is an extremely limited deployment scenario, and will not be further examined here. 7.2.2 Using an STag on a different connection One style of attack from the Remote Peer is for it to attempt to use STag values that it is not authorized to use. Note that, if the Remote Peer sends an invalid STag to the Local Peer, per the DDP and RDMAP specifications, the connection must be torn down. Thus, the threat exists if a STag has been enabled for Remote Access on one connection and a Remote Peer is able to use it on an unrelated connection. If the attack is successful, the attacker could potentially be able to perform either RDMA Read Operations to read the contents of the associated data buffer, perform RDMA Write Operations to modify the contents of the associated data buffer, or to Invalidate the STag to disable further access to the buffer. An attempt by a Remote Peer to access a buffer with an STag on a different connection in the same Protection Domain may or may not be an attack depending on the Trust Model employed by the application. For Trust Model S-T, where resources are shared between connections, and both Local and Remote Peers are Trusted, using an STag on multiple connections within the same Protection Domain is allowed, and could be desired behavior. For the other four Trust Models where the Remote Peer is not Trusted, and/or resources are not intended to be shared, attempting to use an STag on a different connection could be considered to be an attack. In the case where the Trust Model is defined with no shared resources between connections (Trust Models NS-NT and NS-RT), this attack can be defeated by assigning each connection to a different Protection Domain. Before allowing remote access to the buffer, the Protection Domain of the connection where the access attempt was made is matched against the Protection Domain of the STag. If the Protection Domains do not match, access to the buffer is denied, an error is generated, and the RDMAP Stream associated with the attacking connection should be terminated. Thus, for Trust Models NS-NT and NS-RT, it is RECOMMENDED that each connection be in a separate Protection Domain. For Trust Models S-NT and S-LT, where resources are shared, but the Remote Peers are Untrusted, it may not be practical to separate each connection into its own Protection Domain. In this case, the J. Pinkerton, et al. Expires - April 2004 [Page 21] Internet-Draft RDDP/RDMAP Security October 2003 application can still limit the scope of any of the STags it is enabling for remote access to a single connection. If the STag scope has been limited to a single connection, any attempt to use that STag on a different connection will result in an error, and the RDMA Stream associated with that connection should be terminated. Thus, for Trust Models S-NT and S-LT, it is RECOMMENDED that the scope of an STag be limited to a single connection. For Trust Models S-NT and S-LT (Untrusted Remote Peers), if it is not possible to use Protection Domains or to limit the scope of an STag to a single connection, it is RECOMMENDED that STag allocators select an STag using an algorithm which makes it difficult to guess the next allocated STag number. This approach is good practice in general. Allocation methods which always start with the same number (e.g. zero) after Stream initialization or simply allocate the next STag in a monotonically increasing namespace should be avoided. 7.3 Tampering A Remote Peer can attempt to tamper with the contents of data buffers on a Local Peer that have been enabled for remote write access. The types of tampering attacks that are possible are outlined in the sections that follow. 7.3.1 RDMA Write or Read Response to Memory Outside the Buffer This attack is an attempt by the Remote Peer to perform an RDMA Write or RDMA Read Response to memory outside of the valid length range of the data buffer enabled for remote write access. This attack applies primarily to Trust Models with Untrusted Remote Peers (NS-NT, S-NT and S-LT), and can occur even when no resources are shared across connections. This issue can also arise for Trust Models NS-RT and S-T, which assume remote Partial Trust, if the application has a bug. Thus it is RECOMMENDED that all Trust Models ensure this countermeasures are in place against this form of attack. The countermeasure for this type of attack must be in the RNIC implementation, using the STag. When the Local Peer specifies to the RI, the base and the number of bytes in the buffer that it wishes to make accessible, the RI must ensure that the base and bounds check are applied to any access to the buffer referenced by the STag before the STag is enabled for access. When an RDMA data transfer operation (which includes an STag) arrives on a connection, a base and bounds byte granularity access check must be performed to ensure the operation accesses only memory locations within the buffer described by that STag. Thus, it is RECOMMENDED that an RI implementation ensure that a Remote Peer, regardless of Trust Model, will not be able to access J. Pinkerton, et al. Expires - April 2004 [Page 22] Internet-Draft RDDP/RDMAP Security October 2003 memory outside of the buffer specified when the STag was enabled for remote access. 7.3.2 Modifying a Buffer After Indicating Contents Are Ready This attack occurs if a Remote Peer attempts to modify the contents by performing an RDMA Write or an RDMA Read Response after it had indicated to the Local Peer that the data buffer contents were ready for use. This attack applies primarily to the Trust Models where the Remote Peers are not Trusted (Trust Models NS-NT, S-NT and S-LT), and can occur even when no resources are shared across connections. Note that, an error on the part of a Trusted Remote Peer could also result in this problem. The Local Peer can protect itself from this type of attack by revoking remote access when the original data transfer has completed and before it validates the contents of the buffer. The Local Peer can either do this by explicitly revoking remote access rights for the STag when the Remote Peer indicates the operation has completed, or by checking to make sure the Remote Peer Invalidated the STag through the RDMAP Invalidate capability, and if it did not, the Local Peer then explicitly revokes the STag remote access rights. It is RECOMMENDED that the Local Peer follow the above procedure for Trust Models NS-NT, S-NT, and S-LT to protect the buffer. The Local Peer MAY also wish to use this procedure for Trust Models NS-RT and S-T to protect itself from unintended tampering due to an error in the Remote Peer. 7.3.3 Using Multiple Stags to access the same buffer See section 7.4.6 on page 25 for this analysis. 7.4 Information Disclosure The main potential source for information disclosure is through a local buffer that has been enabled for remote access. If the buffer can be probed by a Remote Peer on another connection, then there is potential for information disclosure. Information disclosure attacks mainly apply to the Trust Models that include Untrusted Remote Peers (Trust Models NS-NT, S-NT, and S-LT as defined in Section 5). Trusted Remote Peers are assumed not to purposely attempt such attacks - any attempt is assumed to be due to an error or other unexpected failure in the Remote Peer. The potential attacks that could result in unintended information disclosure and countermeasures are as follows: J. Pinkerton, et al. Expires - April 2004 [Page 23] Internet-Draft RDDP/RDMAP Security October 2003 7.4.1 Probing memory outside of the buffer bounds This is essentially the same attack as described in Section 7.3.1, except an RDMA Read Request is used to mount the attack. The same countermeasure applies. 7.4.2 Using RDMA Read to Access Stale Data If a buffer is being used for a combination of reads and writes (either remote or local), and is exposed to the Remote Peer with at least remote read access rights, the Remote Peer may be able to examine the contents of the buffer before they are initialized with the correct data. In this situation, whatever contents were present in the buffer before the buffer is initialized can be viewed by the Remote Peer, if the Remote Peer performs an RDMA Read. This threat applies to Trust Models NS-NT, S-NT, and S-LT. Because of this, it is RECOMMENDED that the Local Peer ensure that no stale data is contained in the buffer when remote read access rights are initially granted (this can be done by zeroing the contents of the memory, for example). 7.4.3 Accessing a buffer after the transfer is over If the Remote Peer has remote read access to a buffer, and by some mechanism tells the Local Peer that the transfer has been completed, but the Local Peer does not disable remote access to the buffer before modifying the data, it is possible for the Remote Peer to retrieve the new data. This is similar to the attack defined in Section 7.3.2 Modifying a Buffer After Indicating Contents Are Ready on page 23. The same countermeasures apply. In addition, it is RECOMMENDED that the Local Peer should grant remote read access rights only for the amount of time needed to retrieve the data. 7.4.4 Accessing data within a valid STag that was unintended If the Local Peer enables remote access to a buffer using an STag that references the entire buffer, but intends only a portion of the buffer to be accessed, it is possible for the Remote Peer to access the other parts of the buffer anyway. This threat applies to Trust Models NS-NT, S-NT, and S-LT. To prevent this attack, it is RECOMMENDED that the Local Peer set the base and bounds of the buffer when the STag is initialized to expose only the data to be retrieved. J. Pinkerton, et al. Expires - April 2004 [Page 24] Internet-Draft RDDP/RDMAP Security October 2003 7.4.5 Using RDMA Read on a buffer meant only for RDMA Write One form of disclosure can occur if the access rights on the buffer were set for remote read, when only remote write access was intended. This attack applies primarily to Trust Models with Untrusted Remote Peers (NS-NT, S-NT and S-LT). If the buffer contained application data, or data from a transfer on an unrelated connection, the Remote Peer could retrieve the data through an RDMA Read operation. The most obvious countermeasure for this attack is to not grant remote read access if the buffer is intended to be write-only. The Remote Peer would not be able to retrieve data associated with the buffer. An attempt to do so would result in an error and the RDMAP Stream associated with the connection would be terminated. Thus, it is RECOMMENDED that if an application only intends a buffer to be exposed for remote write access, it set the access rights to the buffer to only enable remote write access. 7.4.6 Using Multiple Stags to access the same buffer Multiple STags accessing the same buffer at the same time can result in unintentional information disclosure if the STags are used by different Remote Peers. Because an RDMA implementation could allow an STag to have read, write, or read and write access associated with an Stag, it is possible to have unintended information disclosure if the Remote Peers do not share the same Trust Model. If only read access is enabled, then the Local Peer has complete control over the information disclosure and multiple Stags to the same buffer creates no new security issues. When one STag has remote read access enabled and a different STag has remote write access enabled to the same buffer, it is possible for one connection to view the contents that have been written by another Remote Peer. If both Stags have remote write access enabled and the two Remote Peers do not mutually trust each other, it is possible for one Remote Peer to overwrite the contents that have been written by the other Remote Peer. For Trust Models NS-NT, S-NT, S-LT it is RECOMMENDED that multiple Remote Peers not be granted access to the same buffer through different STags at the same time. A buffer should be exposed to only one Untrusted Remote Peer at a time to ensure that no information disclosure or information tampering occurs between peers. J. Pinkerton, et al. Expires - April 2004 [Page 25] Internet-Draft RDDP/RDMAP Security October 2003 7.4.7 Remote node loading firmware onto the RNIC If the Remote Peer can cause firmware to be loaded onto the RNIC, there is an opportunity for information disclosure. See Elevation of Privilege in Section 7.6 for this analysis. 7.4.8 Controlling Access to Page Translation Table and STag Mapping Issue: Discuss issues in allowing the non-privileged consumer to control mapping the Page Translation Table and Stag. It is RECOMMENDED that the Privileged Resource Manager verify that the Non-Privileged application has the right to access a specific Data Buffer before allowing an STag for which the application has access rights to be associated with a specific Data Buffer. This can be done when the Page Translation Table is initialized to access the Data Buffer or when the STag is initialized to point to a group of Page Translation Table entries, or both. 7.5 Denial of Service (DOS) A DOS attack is one of the primary security risks of RDMAP. This is because RNIC resources are valuable and scarce, and many application environments require communication with Untrusted Remote Peers. If the remote application can be authenticated or encrypted, clearly, the DOS profile can be reduced. For the purposes of this analysis, it is assumed that the RNIC must be able to operate in Untrusted environments, which are open to DOS style attacks. Denial of service attacks against RI resources are not the typical unknown party spraying packets at a random host (such as a TCP SYN attack). Because the connection must be fully established, the attacker must be able to both send and receive messages over that connection, or be able to guess a valid packet on an existing RDMAP Stream. This section outlines the potential attacks and the countermeasures available for dealing with each attack. For each attack, the Trust Model that it applies to is described. J. Pinkerton, et al. Expires - April 2004 [Page 26] Internet-Draft RDDP/RDMAP Security October 2003 7.5.1 RNIC Resource Consumption This section covers attacks that fall into the general category of a Local Peer attempting to unfairly allocate scarce RNIC resources. The Local Peer may be attempting to allocate resources on its own behalf, or on behalf of a Remote Peer. Resources that fall into this category include: Protection Domains, Connection Context Memory, Translation and Protection Tables, and STag namespace. These can be attacks by currently active Local Peers or ones that allocated resources earlier, but are now idle. These attacks generally apply to any Trust Model that includes Untrusted Local Peers (Trust Models NS-NT, NS-RT and S-NT). This type of attack can occur even when resources are not shared across connections. It is RECOMMENDED that the allocation of all scarce resources be placed under the control of a Resource Manager. This allows the Resource Manager to: * prevent a Local Peer from allocating more than its fair share of resources, and * detect if a Remote Peer is attempting to launch a DOS attack by attempting to create an excessive number of connections and take corrective action (such as refusing the request or applying network layer filters against the Remote Peer). Note that, placing scarce resource management under the control of a Resource Manager also prevents other Trusted Local Peers from attempting to allocate more than their fair share of resources. This analysis assumes that the Resource Manager is responsible for handing out Protection Domains, and RNIC implementations will provide enough Protection Domains to allow the Resource Manager to be able to assign a unique Protection Domain for each unrelated, Untrusted Local Peer (for a bounded, reasonable number of Local Peers). This analysis further assumes that the Resource Manager implements policies to ensure that Untrusted Local Peers are not able to consume all of the Protection Domains through a DOS attack. Note that Protection Domain consumption cannot result from a DOS attack launched by a Remote Peer, unless a Local Peer is acting on the Remote PeerÆs behalf. 7.5.2 Resource Consumption By Active Applications This section describes DOS attacks from Local and Remote Peers that are actively exchanging messages. Attacks on each RDMA NIC resource are examined, including the Trust Models that apply, and the specific countermeasures. Note that, attacks on Connection Context J. Pinkerton, et al. Expires - April 2004 [Page 27] Internet-Draft RDDP/RDMAP Security October 2003 Memory, Page Translation Tables, and STag namespace are covered in Section 7.5.1 RNIC Resource Consumption, so are not included here. 7.5.2.1 Receive Data Buffers The Remote Peer can attempt to consume more than its fair share of receive data buffers (Untagged DDP buffers or for RDMAP buffers consumed with Send Type Messages). If resources are not shared across multiple connections (Trust Models NS-NT, NS-RT), then this attack is not possible because the Remote Peer will not be able to consume more buffers than were allocated to the Stream. The worst case scenario is that the Remote Peer can consume more receive buffers than the Local Peer allowed, resulting in no buffers to be available, which would cause the Remote PeerÆs connection to the Local Peer to be torn down. If local receive data buffers are shared among multiple Streams and the Remote Peer is not Trusted (Trust Models S-NT, S-LT), then the Remote Peer can attempt to consume more than its fair share of the receive buffers, causing a different Stream to be short of receive buffers, thus possibly causing the other Stream to be torn down. One method the Local Peer could use is to recognize that a Remote Peer is attempting to use more than its fair share of resources and terminate the Stream. However, if the Local Peer is sufficiently slow, it may be possible for the Remote Peer to still mount a denial of service attack. An RNIC Engine implementation that enables a more robust countermeasure is one that provides high and low-water notifications to enable the Local Peer to detect and prevent DOS attacks against shared data buffers. If a low-water notification is enabled, and the Local Peer is able to bound the amount of time that it takes to replenish receive buffers, and the Local Peer maintains statistics to determine which Remote Peer is consuming buffers, then the Local Peer can size the amount of local receive buffers posted on the receive queue such that the low-water notification can arrive before resources are depleted and corrective action can be taken (e.g., terminate the Stream of the attacking Remote Peer). Enabling the high-water notification can help the Local Peer detect a Remote Peer that is launching an attack by sending a large number of out- of-order packets. The notification is generated when more than the specified number of buffers are in process simultaneously on a Stream (i.e., packets have started to arrive for the buffer, but have not yet been delivered to the ULP). A different countermeasure is for the RNIC Engine to provide the capability to limit the Remote PeerÆs ability to consume receive buffers on a per Stream basis. Unfortunately this requires a large amount of state to be tracked on a per RNIC basis. J. Pinkerton, et al. Expires - April 2004 [Page 28] Internet-Draft RDDP/RDMAP Security October 2003 Thus, if an RNIC Engine provides the ability to share receive buffers across multiple Streams, it is RECOMMENDED that it enable the Local Peer to detect if the Remote Peer is attempting to consume more than its fair share of resources so that the application can apply countermeasures to detect and prevent the attack. 7.5.2.2 Completion Queue (CQ) Resource Consumption DOS attacks against the Completion Queue can be caused by either the Local Peer or the Remote Peer if either attempts to cause more completions than its fair share, thus potentially starving another unrelated Stream such that no Completion Queue entries are available. A Completion Queue entry can potentially be consumed by a completion from the send queue or a receive completion. In the former, the attacker is the Local Peer (Trust Models S-NT). In the later, the attacker is the Remote Peer (S-NT, S-LT). The potential attacks and the countermeasures for each are described in the subsections that follow. 7.5.2.2.1 Local Peer Attacking a Shared CQ A form of attack can occur for Trust Models NS-NT, NS-RT, and S-NT, where the Local Peers are Untrusted, and Local Peers can consume resources on the CQ. Sharing a CQ across connections that belong to different Protection Domains is NOT RECOMMENDED in cases where any of the Local Peers are Untrusted. A Local Peer that is slow to free resources on the CQ by not reaping the completion status quick enough could stall all other Local Peers attempting to use that CQ. One of two countermeasures can be used to avoid this kind of attack. The first is to use a different CQ per Untrusted Local Peer. The other is to use a Trusted Local Peer to act as a third party to free resources on the CQ and place the status in intermediate storage until the Untrusted Local Peer reaps the status information. 7.5.2.2.2 Remote Peer Attacking a Shared CQ The Remote Peer can attack a CQ by consuming more than its fair share of CQ entries by using one of two methods. The first method can only be used if the ULP protocol allows the Remote Peer to reserve a specified number of CQ entries, possibly leaving insufficient entries for other connections that are sharing the CQ. The other method is if the Remote Peer can attack the CQ by overwhelming the CQ with completions, which can affect completion processing on other Streams sharing that connection. J. Pinkerton, et al. Expires - April 2004 [Page 29] Internet-Draft RDDP/RDMAP Security October 2003 The first method of attack can be avoided if the ULP does not allow a Remote Peer to reserve CQ entries. This is RECOMMENDED particularly for Trust Models S-NT and S-LT, with shared resources and Untrusted Remote Peers. If a Local Peer allows this type of resource allocation, and it has any Untrusted Remote Peers, then the Local Peer it is RECOMMENDED that the CQ be isolated to connections within a single Protection Domain. One way that a Remote Peer can attempt to overwhelm its CQ with completions is by sending minimum length RDMAP/DDP Messages to cause as many completions per second as possible. Assuming that the Local Peer does not run out of receive buffers (if they do, then this is a different attack, documented in Section 7.5.2.1 Receive Data Buffers on page 28), then it might be possible for the Remote Peer to consume more than its fair share of Completion Queue entries. Depending upon the CQ implementation, this could either cause the CQ to overflow (if it is not large enough to handle all of the completions generated) or for another Stream to not be able to generate CQ entries (if the RNIC had flow control on generation of CQ entries into the CQ). In either case, the CQ will stop functioning correctly and any connections expecting completions on the CQ will stop functioning. This attack can occur regardless of whether all of the connections associated with the CQ are in the same Protection Domain or are in different Protection Domains. Because this attack assumes a shared local resource and an Untrusted Remote Peer, Trust Models S-NT, S-LT apply. The Local Peer can protect itself from this type of attack using either of the following methods: * Resize the CQ to the appropriate level(note that resizing the CQ can fail, so the CQ resize should be done before sizing the buffers on the connection), OR * Grant fewer resources than the Remote Peer requested (not supplying the number of Receive Data Buffers requested). The proper sizing of the CQ is dependent on the Trust Model. For the Trust Model described in Section 5, with Trusted Local Peers and Untrusted Remote Peers (Trust Model S-LT), a correctly sized CQ means that the CQ is large enough to hold completion status for all of the outstanding Receive Data Buffers, or: J. Pinkerton, et al. Expires - April 2004 [Page 30] Internet-Draft RDDP/RDMAP Security October 2003 CQ_MIN_SIZE = SUM(MaxPostedOnEachRQ) + SUM(MaxPostedOnEachS-RQ) + SUM(MaxPostedOnEachSQ) If the Trust Model assumes neither the Local Peer nor the Remote Peer is trusted (Trust Model S-NT or S-LT), then the CQ must be sized to accommodate the maximum number of operations or Receive Data Buffers that it is possible to post at any one time, thus the equation becomes: CQ_MIN_SIZE = SUM(SizeOfEachRQ) + SUM(SizeOfEachS-RQ) + SUM(SizeOfEachSQ) Where MaxPosted*OnEach*Q and SizeOfEach*Q varies on a per connection or per Shared Receive Queue basis. 7.5.2.3 RDMA Read Request Queue Two types of attacks are possible against resources associated with RDMA Read Request Queues. One style of attack can only occur when the RDMA Read Request Queue resources are pooled across multiple connections. This attack occurs when an Untrusted Local Peer attempts to unfairly allocate RDMA Read Request Queue resources for its connections. It is RECOMMENDED that access to interfaces that allocate RDMA Read Request Queue entries be restricted to a Trusted Local Peer, such as a Resource Manager. The Resource Manager should prevent a Local Peer from allocating more than its fair share of resources. Another form of attack is the Remote Peer sending more RDMA Read Requests than the depth of the RDMA Read Request Queue at the Local Peer. This attack can be prevented by properly configuring the connection when the connection is established. The Remote PeerÆs end of the connection should be configured by a trusted agent such that the RNIC will not transmit RDMA Read Requests that exceed the depth of the RDMA Read Request Queue at the Local Peer. If the connection is correctly configured, and if the Remote Peer submits more requests than the Local PeerÆs RDMA Read Request Queue can handle, the requests will be queued at the Remote PeerÆs connection until previous requests complete. If the Remote PeerÆs connection is not configured correctly, the RDMAP Stream for that connection is terminated when more RDMA Read Requests arrive at the Local Peer than the Local Peer can handle. Thus, the Remote Peer is able to only affect its own connection. J. Pinkerton, et al. Expires - April 2004 [Page 31] Internet-Draft RDDP/RDMAP Security October 2003 7.5.3 Resource Consumption by Idle Applications The simplest form of a DOS attack given a fixed amount of resources is for the Remote Peer to create a RDMAP Stream to a Local Peer, request dedicated resources then do no actual work. This allows the Remote Peer to be very light weight (i.e. only negotiate resources, but do no data transfer) and consumes a disproportionate amount of resources in the server. A general countermeasure for this style of attack is to monitor active RDMAP Streams and if resources are getting low, reap the resources from RDMAP Streams that are not transferring data and possibly terminate the connection. This needs to be under administrative control, and demonstrates the need for a MIB for RDMAP so this condition can be detected and acted upon. Refer to Section 7.5.1 for the analysis and countermeasures for this style of attack on the following RNIC resources: Connection Context Memory, Page Translation Tables and STag namespace. Note that some RNIC resources are not at risk of this type of attack from a Remote Peer because an attack requires the Remote Peer to send messages in order to consume the resource. Receive Data Buffers, Completion Queue, and RDMA Read Request Queue resources are examples. These resources are, however, at risk form a Local Peer that attempts to allocate resources, then goes idle. The general countermeasure described in this section can be used to free resources allocated by an idle Local Peer. 7.5.4 Exercise of non-optimal code paths Another form of DOS attack is to attempt to exercise data paths that can consume a disproportionate amount of resources. An example might be if error cases are handled on a ææslow pathÆÆ (consuming either host or RNIC computational resources), and an attacker generates excessive numbers of errors in an attempt to consume these resources. It is RECOMMENDED that an implementation provide the ability to detect the above condition and allow an administrator to act, including potentially administratively tearing down the RDMAP Stream associated with the connection exercising data paths consuming a disproportionate amount of resources. 7.5.5 Remote Invalidation of an STag Shared Across Multiple Connections If a Local Peer has enabled an STag for remote access, the Remote Peer could attempt to invalidate the STag by using the RDMAP Send with Invalidate or Send with SE and Invalidate Message. If the STag J. Pinkerton, et al. Expires - April 2004 [Page 32] Internet-Draft RDDP/RDMAP Security October 2003 is only valid on the current connection (NS-NT or NS-RT, S-NT), then the only side effect is that the Remote Peer can no longer use the STag, thus there are no security issues. If the STag is valid across multiple connections, then the Remote Peer can prevent other connections from using that STag by using the remote invalidate functionality. Thus for Trust Models where the Remote Peer may attempt to invalidate the STag prematurely, the application SHOULD NOT allow an STag to be valid across multiple connections. 7.5.6 Remote Peer Consumes too many Untagged Receive Buffers Issue: Need to analyze the case of sharing a queue of Untagged receive buffers across multiple connections, and that the Remote Peer can mount a denial of service attack. Below are some notes. Many ways to attack here. If receive queue is not shared, itÆs a simple queue overflow attack on a dedicated resource. Make sure when the queue is empty and a DDP segment arrives nothing bad happens. For a shared receive queue, one node attacks with single byte Untagged Messages to consume large Untagged Buffers (this maximizes packet arrival rate). One node provides DDP segments out of order to consume out-of-order resources (this is only possible if out-of- order placement is supported within a merged TCP/SCTP and DDP implementation). 7.6 Elevation of Privilege The RDMAP/DDP Security Architecture explicitly differentiates between three levels of privilege - Non-Privileged, Privileged, and the Privileged Resource Manager. If a Non-Privileged Application is able to elevate its privilege level to a Privileged Application, then mapping a physical address list to an STag can provide local and remote access to any physical address location on the node. If a Privileged Mode Application is able to promote itself to be a Resource Manager, then it is possible for it to perform denial of service type attacks where substantial amounts of local resources could be consumed. There is only one mechanism discovered to date, other than implementation defects, which would potentially allow an elevation of privilege. J. Pinkerton, et al. Expires - April 2004 [Page 33] Internet-Draft RDDP/RDMAP Security October 2003 7.6.1 Loading Firmware into the RNIC If the RI implementation, by some insecure mechanism (or implementation defect), can enable a Remote Peer or un-trusted Local Peer to load firmware into the RNIC Engine, it is possible to use the RNIC to attack the host. Thus, it is RECOMMENDED that an implementation not enable firmware to be loaded on the RNIC Engine directly from a Remote Peer, unless the update is done via a secure protocol, such as IPsec (See Section 8 Security Services for RDDP on page 35). It is RECOMMENDED that an implementation not allow a Non- Privileged Local Peer to update firmware in the RNIC Engine. J. Pinkerton, et al. Expires - April 2004 [Page 34] Internet-Draft RDDP/RDMAP Security October 2003 8 Security Services for RDDP Issue: Security Services section is a placeholder for now. 8.1 Introduction to Security Options 8.1.1 Introduction to IPsec IPsec is a protocol suite which is used to secure communication at the network layer between two peers. The IPsec protocol suite is specified within the IP Security Architecture [RFC2401], IKE [RFC2409], IPsec Authentication Header (AH) [RFC2402] and IPsec Encapsulating Security Payload (ESP) [RFC2406] documents. IKE is the key management protocol while AH and ESP are used to protect IP traffic. An IPsec SA is a one-way security association, uniquely identified by the 3-tuple: Security Parameter Index (SPI), protocol (ESP) and destination IP. The parameters for an IPsec security association are typically established by a key management protocol. These include the encapsulation mode, encapsulation type, session keys and SPI values. IKE is a two phase negotiation protocol based on the modular exchange of messages defined by ISAKMP [RFC2408],and the IP Security Domain of Interpretation (DOI) [RFC2407]. IKE has two phases, and accomplishes the following functions: 1. Protected cipher suite and options negotiation - using keyed MACs and encryption and anti-replay mechanisms 2. Master key generation - such as via MODP Diffie-Hellman calculations 3. Authentication of end-points 4. IPsec SA management (selector negotiation, options negotiation, create, delete, and rekeying) Items 1 through 3 are accomplished in IKE Phase 1, while item 4 is handled in IKE Phase 2. An IKE Phase 2 negotiation is performed to establish both an inbound and an outbound IPsec SA. The traffic to be protected by an IPsec SA is determined by a selector which has been proposed by the IKE J. Pinkerton, et al. Expires - April 2004 [Page 35] Internet-Draft RDDP/RDMAP Security October 2003 initiator and accepted by the IKE Responder. In IPsec transport mode, the IPsec SA selector can be a "filter" or traffic classifier, defined as the 5-tuple: . The successful establishment of a IKE Phase-2 SA results in the creation of two uni-directional IPsec SAs fully qualified by the tuple . The session keys for each IPsec SA are derived from a master key, typically via a MODP Diffie-Hellman computation. Rekeying of an existing IPsec SA pair is accomplished by creating two new IPsec SAs, making them active, and then optionally deleting the older IPsec SA pair. Typically the new outbound SA is used immediately, and the old inbound SA is left active to receive packets for some locally defined time, perhaps 30 seconds or 1 minute. 8.1.2 Introduction to SSL Limitations on RDMAP 8.1.3 Applications Which Provide Security Issue: Guidance for application protocols like NFS which implement security. 8.1.4 Authentication Only 8.1.5 Privacy 8.2 Recommendations for IPsec Encapsulation of RDDP Issue: IPsec recommendations for RDMAP/DDP This is work that is still to be done. Hopefully this wonÆt be terribly complex. One possible thought on the approach: a. Use IPsec ESP with authentication to provide authentication, integrity and replay protection. b. Use IKE for key management. J. Pinkerton, et al. Expires - April 2004 [Page 36] Internet-Draft RDDP/RDMAP Security October 2003 c. (optionally) use a non-null transform for encryption. This should be something other than DES. J. Pinkerton, et al. Expires - April 2004 [Page 37] Internet-Draft RDDP/RDMAP Security October 2003 9 Summary Table of Attacks and Trust Models Issue: Finish Summary table of Attacks/Trust Models Rows are the attack (grouped into categories) Columns are the: * Sec - Section the attack is discussed * Attack Name - short name for the attack * Threat - threat type (DOS, etc) * Columns labeled 1-5 are the Trust Model number (see section 5 Trust Models on page 15). Each entry has a value of +, -, and R (research). J. Pinkerton, et al. Expires - April 2004 [Page 38] Internet-Draft RDDP/RDMAP Security October 2003 +-------+--------------------------+-------+---+---+---+---+---+ | Sec | Attack Name |Threat | 1 | 2 | 3 | 4 | 5 | +-------+--------------------------+-------+---+---+---+---+---+ | 7.2.1 | STag use on different | Spoof | | | | | | | | connection in same PD | | | | | | | +-------+--------------------------+-------+---+---+---+---+---+ | 7.3.1 | Memory write outside of | Tamper| | | | | | | | buffer range | | | | | | | | 7.3.2 | Modify Buffer after | Tamper| | | | | | | | contents ready | | | | | | | +-------+--------------------------+-------+---+---+---+---+---+ | 7.4.1 | Probe memory outside of | ID | | | | | | | | buffer bounds | | | | | | | | 7.4.2 | Access stale data | ID | | | | | | | 7.4.3 | Access buffer after | ID | | | | | | | | transfer over | | | | | | | | 7.4.4 | Unintended data access | ID | | | | | | | | using valid STag | | | | | | | | 7.4.5 | Using RDMA Read on a | ID | | | | | | | | buffer meant only for | | | | | | | | | RDMA Write | | | | | | | | 7.4.6 | Using multiple STags to | ID | | | | | | | | access the same buffer | | | | | | | | 7.4.7 | Remote node loading | ID | | | | | | | | firmware onto RNIC | | | | | | | +-------+--------------------------+-------+---+---+---+---+---+ | 7.5.1 | RNIC resource consumption| DOS | | | | | | | 7.5.2 | Resource consumption by | DOS | | | | | | | | active processes | | | | | | | | 7.5.3 | Resource consumption by | DOS | | | | | | | | idle processes | | | | | | | | 7.5.4 | Non-optimal code paths | DOS | | | | | | +-------+--------------------------+-------+---+---+---+---+---+ | 7.6.1 | Loading firmware onto | Elev | | | | | | | | RNIC | | | | | | | +-------+--------------------------+-------+---+---+---+---+---+ Figure 2 - Summary Attacks and Trust Model Table J. Pinkerton, et al. Expires - April 2004 [Page 39] Internet-Draft RDDP/RDMAP Security October 2003 10 References 10.1 Normative References [RFC2828] Shirley, R., "Internet Security Glossary", FYI 36, RFC 2828, May 2000. [DDP] Shah, H., J. Pinkerton, R.Recio, and P. Culley, "Direct Data Placement over Reliable Transports", Internet-Draft draft-ietf- rddp-ddp-01.txt, February 2003. [RDMAP] Recio, R., P. Culley, D. Garcia, J. Hilland, "An RDMA Protocol Specification", Internet-Draft draft-ietf-rddp-rdmap- 01.txt, February 2003. [SEC-CONS] Rescorla, E., B. Korver, IAB, "Guidelines for Writing RFC Text on Security Considerations", Internet-Draft draft-ab- sec-cons-03.txt, January 2003. [RFC2401] Atkinson, R. and Kent, S., "Security Architecture for the Internet Protocol", RFC 2401, November 1998 [RFC2402] Kent, S., Atkinson, R., "IP Authentication Header", RFC 2402, November 1998 [RFC2404] Madson, C., Glenn, R., "The Use of HMAC-SHA-1-96 within ESP and AH", RFC 2404, November 1998 [RFC2406] Kent, S., Atkinson, R., "IP Encapsulating Security Payload (ESP)", RFC 2406, November 1998 [RFC2407] Piper, D., "The Internet IP Security Domain of Interpretation of ISAKMP", RFC 2407, November 1998 [RFC2408] Maughan, D., Schertler, M., Schneider, M., Turner, J., "Internet Security Association and Key Management Protocol (ISAKMP), RFC 2408, November 1998 [RFC2409] Harkins, D., Carrel, D., "The Internet Key Exchange (IKE)", RFC 2409, November 1998 10.2 Informative References [IPv6-Trust] Nikander, P., J.Kempf, E. Nordmark, "IPv6 Neighbor Discovery trust modelsTrust Models and threats", Internet-Draft draft-ietf-send-psreq-01.txt, January 2003. J. Pinkerton, et al. Expires - April 2004 [Page 40] Internet-Draft RDDP/RDMAP Security October 2003 11 AuthorÆs Addresses James Pinkerton Microsoft Corporation One Microsoft Way Redmond, WA. 98052 USA Phone: +1 (425) 705-5442 Email: jpink@windows.microsoft.com Ellen Deleganes Intel Corporation MS JF5-355 2111 NE 25th Ave. Hillsboro, OR 97124 USA Phone: +1 (503) 712-4173 Email: ellen.m.deleganes@intel.com Allyn Romanow Cisco Systems 170 W Tasman Drive San Jose, CA 95134 USA Phone: +1 408 525 8836 Email: allyn@cisco.com Bernard Aboba Microsoft Corporation One Microsoft Way Redmond, WA. 98052 USA Phone: +1 (425) 706-6606 Email: bernarda@windows.microsoft.com J. Pinkerton, et al. Expires - April 2004 [Page 41] Internet-Draft RDDP/RDMAP Security October 2003 12 Acknowledgments Catherine Meadows Naval Research Laboratory Code 5543 Washington, DC 20375 Email: meadows@itd.nrl.navy.mil Patricia Thaler Agilent Technologies, Inc. 1101 Creekside Ridge Drive, #100 M/S-RG10 Roseville, CA 95678 Phone: +1-916-788-5662 email: pat_thaler@agilent.com James Livingston NEC Solutions (America), Inc. 7525 166th Ave. N.E., Suite D210 Redmond, WA 98052-7811 Phone: +1 (425) 897-2033 Email: james.livingston@necsam.com John Carrier Adaptec, Inc. 691 S. Milpitas Blvd. Milpitas, CA 95035 USA Phone: +1 (360) 378-8526 Email: john_carrier@adaptec.com Caitlin Bestler Email: cait@asomi.com J. Pinkerton, et al. Expires - April 2004 [Page 42] Internet-Draft RDDP/RDMAP Security October 2003 13 Full Copyright Statement Copyright (C) The Internet Society (2001). All Rights Reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Funding for the RFC Editor function is currently provided by the Internet Society. J. Pinkerton, et al. Expires - April 2004 [Page 43]