Network Working Group Nat Brown INTERNET-DRAFT Charlie Kindel Microsoft Corporation Expires in six months May 2, 1996 Distributed Component Object Model Protocol -- DCOM/1.0 Status of this Memo This document is an Internet-Draft. 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." To learn the current status of any Internet-Draft, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe), munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast). Distribution of this document is unlimited. Please send comments to the authors at . General discussions about DCOM and its applications should take place on the mailing list. To subscribe, send a piece of mail to , and in the body of the message, first type "subscribe" for non-digest or "digest" for digest version, followed by "dcom", and finally by your email address. (example: "subscribe dcom johndoe@gizmo.com") Abstract The Distributed Component Object Model protocol is an application- level protocol for object-oriented remote procedure calls useful for distributed, component-based systems of all types. It is a generic protocol layered on the distributed computing environment (DCE) RPC specification and facilitates the construction of task-specific communication protocols through features such as: a platform neutral argument/parameter marshaling format (NDR), the ability for objects to support multiple interfaces with a safe, interface-level versioning scheme suited to independent evolution by multiple parties, the ability to make authenticated connections and to choose levels of channel security, and a transport-neutral data representation for references (including by-value) to objects. Table of Contents 1. Introduction 1.1 Purpose 2. Overall Operation 2.1 Object Calls 2.2 OXIDs and Object Exporters 2.3 Marshaled Interface References 2.4 Reference Counting 2.5 Pinging 2.5.1 Delta Pinging 2.6 QueryInterface 2.7 Causality ID 3. Data Types and Structures 3.1 DCE Packet Headers 3.2 ORPC Base Definitions 3.3 OBJREF Structure 3.3.1 OBJREF_STANDARD form 3.3.2 OBJREF_HANDLER form 3.3.3 OBJREF_CUSTOM form 3.4 STDOBJREF Structure 3.5 SORFLAGS Constants 3.6 ORPCINFOFLAGS Constants 3.7 ORPCTHIS Structure 3.8 ORPCTHAT Structure 3.9 HRESULTs 3.10 Body Extensions 3.10.1 Debugging Extension 3.10.2 Extended Error Extension 4. IRemUnknown Interface 4.1 IRemUnknown::RemQueryInterface 4.1.1 REMQIRESULT Structure 4.2 IRemUnknown::RemAddRef 4.3 IRemUnknown::RemRelease 5. The Object Exporter 5.1 Object Exporter Reserved Ports 5.2 The IObjectExporter Interface 5.2.1 IObjectExporter::ResolveOxid 5.2.1.1 Lookup Between Friends 5.2.1.2 My First Lookup 5.2.1.3 Middleman Lookup 5.2.1.4 Lazy Use Protseq 5.2.2 IObjectExporter::SimplePing 5.2.3 IObjectExporter::ComplexPing 6. Security Considerations 7. Acknowledgements 8. References 9. Author's Address 1. Introduction The Distributed Component Object Model protocol (DCOM) is an application-level protocol for object-oriented remote procedure calls and is thus also called "Object RPC" or ORPC. The protocol consists of a set of extensions, layered on the distributed computing environment (DCE) RPC specification [CAE RPC], with which familiarity is assumed. Familiarity is also assumed with the COM (Component Object Model) specification [COM]. Object RPC specifies: * How calls are made on an object * How object references are represented, communicated, and maintained 1.1 Purpose There is a natural tendency in a networked environment to create entirely new application-level protocols as each new or seemingly unique combination of client, user agent, and server requirement arises. While in many situations the definition of a new protocol is useful and justifiable, there are numerous features which have eventually been added to or required from each new protocol (or which become layered above them) as they evolve and become used in broader contexts. A design goal of the DCOM protocol is the inherent support of standard features required by any distributed application communication protocol. In other words, to act as a framework to facilitate the construction of task-specific communication paths between distributed applications. Data Marshaling A common occurrence among user agents using the HTTP protocol today is the use of complex, task-specific Query URL syntax and HTTP POSTs. Also increasingly common is the POSTing and response with custom MIME types to and from resources which interpret the format and reply in same. While workable, there are drawbacks to this approach including increased complexity and work to produce and consume each new (and unique) format in the client and server, lessened ability to build task-specific firewalls for administration and security purposes, and in many cases definition of platform-centric formats. DCOM utilizes the Network Data Representation for arbitrary data types supported by DCE RPC. Security DCOM leverages the authentication, authorization, and message integrity capabilities of DCE RPC. An implementation may support any level of DCE RPC security. Any connection or call can be made as secure or as insecure as negotiated by the client and the server. Safe Non-Coordinated Versioning of Interfaces In DCOM versioning of interfaces is done through identifiers which are universally unique (UUID's). To version a published interface, a new interface is defined with a different UUID to the updated specification. Multiple parties can simultaneously introduce "revisions" to interfaces by defining related but distinct interfaces without fear of colliding with each other's version numbers and without fear of breaking each other's down-level or up-level clients. To date, the bulk of task-specific protocols (such as custom POSTs or MIME types using HTTP) have little or no concept of versioning at all, and simply "narrow" the incompatibility window by updating clients (typically pages which are being downloaded anyway) and servers (CGI scripts or other HTTP server infrastructure) simultaneously. 2. Overall Operation The Object RPC protocol highly leverages the OSF DCE RPC network protocol (see the reference [CAE RPC]). This leverage occurs at both the specification level and the implementation level: the bulk of the implementation effort involved in implementing the DCOM network protocol is in fact that of implementing the DCE RPC network protocol on which it is built. 2.1 Object Calls An actual COM network remote procedure call (hereinafter referred to as "an ORPC") is in fact a true DCE remote procedure call (herein termed "a DCE RPC"), a "Request PDU" conforming to the specification for such calls per [CAE RPC]. In an ORPC, the object ID field of the invocation header as specified in [CAE RPC] contains an "IPID". An IPID is a 128-bit identifier known as an interface pointer identifier which represents a particular interface on a particular object in a particular server. As it is passed in the object ID fields of a DCE RPC, the static type of an IPID is in fact a UUID. However, IPIDs are scoped not globally but rather only relative to the machine on which the server is located (and which thus originally allocated them); IPIDs do not necessarily use the standard UUID allocation algorithm, but rather may use a machine-specific algorithm which can assist with dispatching. In an ORPC, the interface ID field of the RPC header specifies the IID, and arguments are found in the body, as usual. However, when viewed from the DCE RPC perspective an additional first argument is always present that is absent in the corresponding interface specification. This argument is of type ORPCTHIS, which is described in Section 3.7. It is placed first in the body of the Request PDU, before the actual arguments of the ORPC. It is specifically legal for an ORPC to attempt a call a method number on a given interface which is beyond the number of methods believed by the server to be in that interface. Such calls should cause a fault. Similarly, in a reply to an ORPC (a DCE RPC "Response PDU"), when viewed from the DCE RPC perspective, an additional first return value is always present that is absent in the corresponding COM interface specification. This argument is of type ORPCTHAT, which is described in Section 3.8. It is placed first in the body of the Response PDU, before the actual return values of the ORPC. An ORPCTHAT may also be present in a "Fault PDU." In the Connectionless (CL) Fault PDU, it is placed four bytes after the 32- bit fault code which normally comprises the entire body of the PDU, thus achieving eight byte alignment for the ORPCTHAT; the intervening padding bytes are presently reserved and must be zero. The PDU body length is of course set to encompass the entire body of the Fault PDU, including the ORPCTHAT. In the Connection- Oriented (CO) Fault PDU, the ORPCTHAT is placed in the standard location allocated for the "stub data." In a Fault PDU of either form that results from an ORPC, if an ORPCTHAT is not present then no other data may be substituted in its here-specified location in the PDU. 2.2 OXIDs and Object Exporters Although an IPID from a logical perspective semantically determines the server, object and interface to which a particular call should be directed, it does not by itself indicate the binding information necessary to actually carry out an invocation. The protocol represents this "how-to" communication information in a UUID called an object exporter identifier, otherwise known as an OXID. Conceptually, an OXID can be thought of as an implementation scope for an object, which may be a whole machine, a given process, a thread within that process, or other more esoteric implementation scope, but the exact definition of such scopes has no bearing on the protocol itself. A given machine at any moment may support several OXIDs; however there is always a unique Object Exporter service per machine which coordinates the management of all the OXIDs on the machine. Data structures in each Object Exporter keep track of the IPIDs exported and imported by that Object Exporter. The Object Exporter resides at well-known ports (or endpoints, depending on your terminology -- one per protocol, of course) on the machine. It supports a DCE RPC interface known as IObjectExporter, described in Section 5.2. An OXID is used to determine the RPC string bindings that allow calls to reach their target IPID. Before making a call, the calling process must translate an OXID into a set of bindings that the underlying RPC implementation understands. It accomplishes by maintaining a cache of these mappings. When the destination application receives an object reference, it checks to see if it recognizes the OXID. If it does not, then it asks the source of the object reference (the server machine from which the object reference was acquired, which is not necessarily the home machine for the interface pointer) for the translation, and saves the resulting set of string bindings in a local table that maps OXIDs to string bindings. Associated with each OXID (not each Object Exporter) is COM object termed an "OXID object." OXID objects implement (at least) the IRemUnknown interface, a COM (not DCE-RPC) interface through which remote management of reference counts and requests for interfaces are returned. 2.3 Marshaled Interface References The DCOM protocol extends the Network Data Representation standard specified in [CAE RPC] by defining what can be thought of as a new primitive data type that can be marshaled: that of an interface reference to an object. This is the only extension to NDR made by the DCOM protocol. A marshaled interface reference is described by a type known as an OBJREF, which is described in detail in Section 3.3. An OBJREF in actuality has several variations: NULL This is a reference to no object. STANDARD A standard remote reference. Known as a STDOBJREF. A STDOBJREF contains: * An IPID, which uniquely specifies the interface and object. * An object ID (OID), which uniquely specifies the identity of the object on which the IPID is found (scoped to the object exporter with which the object is associated). * An OXID, which identifies the scope where the implementation of the object is active, and can be used to reach the interface pointer. * A reference count, indicating the number of references to this IPID that are conveyed by this marshaling. This count, though typically a value of one, may in fact be zero, one, or more (see the next section). * Some flags, explained later. CUSTOM Contains a class ID (CLSID) and class-specific information. The Custom format gives an object control over the representation of references to itself. For example, an immutable object might be passed by value, in which case the class-specific information would contain the object's immutable data. HANDLER A sub-case of the custom reference in which the class- specific information is standardized. For example, an object wishes to be represented in client address spaces by a proxy object that caches state. In this case, the class-specific information is just a standard reference to an interface pointer that the handler (proxy object) will use to communicate with the original object. Interface references are always marshaled in little-endian byte order, irrespective of the byte order prevailing in the remainder of the data being marshaled. 2.4 Reference Counting In the DCOM protocol, remote reference counting is conducted per interface (per IPID). The actual increment and decrement calls are carried out using (respectively) the RemAddRef and RemRelease methods in a COM interface known as IRemUnknown found on an object associated with each OXID, the IPID of which is returned from the function IObjectExporter::ResolveOxid (section 5.2.1). In contrast to their analogues in IUnknown, RemAddRef and RemRelease can in one call increment or decrement the reference count of many different IPIDs by an arbitrary amount; this allows for greater network efficiency. In the interests of performance, client COM implementations typically do not immediately translate each local AddRef and Release into a remote RemAddRef and RemRelease. Rather, the actual remote release of all interfaces on an object is typically deferred until all local references to all interfaces on that object have been released. Further, one actual remote reference count may be used to service many local reference counts; that is, the client infrastructure may multiplex zero or more local references to an interface into zero or one remote references on the actual IPID. 2.5 Pinging The above reference counting scheme would be entirely adequate on its own if clients never terminated abnormally, but in fact they do, and the system needs to be robust in the face of clients terminating abnormally when they hold remote references. In a DCE RPC, one typically addresses this issue through the use of context handles. Context handles are not used, however, by the DCOM protocol, for reasons of expense. The basic underlying technology used in virtually all protocols for detecting remote abnormal termination is that of periodic pings. Naive use of RPC context handles would result in per object per client process pings being sent to the server. The DCOM protocol includes a pinging infrastructure to reduce network traffic by relying on the client Object Exporter implementation to do local management of client liveness detection, and having the actual pings be sent only on a machine to machine basis. Pinging is carried out on a per-object (per OID), not a per- interface (per-IPID) basis. Architecturally, at its server machine, each exported object (each exported OID) has associated with it a pingPeriod time value and a numPingsToTimeOut count which together (through their product) determine the overall amount of time known as the "ping period" that must elapse without receiving a ping on that OID before all the remote references to IPIDs associated with that OID can be considered to have expired. Once expiration has occurred, the interfaces behind the IPIDs can as would be expected be reclaimed solely on the basis local knowledge, though the timeliness with which this is carried out, if at all, is an implementation specific. If the server COM infrastructure defers such garbage collection in this situation (perhaps because it has local references keeping the interface pointer alive) and it later hears a ping or receives a remote call, then it knows a network partition healed. It can consider the extant remote references to be reactivated and can continue remote operations. When interface pointers are conveyed from one client to another, such as being passed as either [in] or [out] parameters to a call, the interface pointer is marshaled in one client and unmarshaled in the other. In order to successfully unmarshal the interface, the destination client must obtain at least one reference count on the interface. This is usually accomplished by passing in the marshaled interface STDOBJREF a cRefs of (at least) one; the destination client then takes ownership of that many (more) reference counts to the indicated IPID, and the source client then owns that many fewer reference counts on the IPID. It is legal, however, for zero reference counts to be passed in the STDOBJREF; here, the destination client must (if it does not already have access to that IPID and thus have a non-zero reference count for it) before it successfully unmarshals the interface reference (concretely, e.g., before CoUnmarshalInterface returns) call to the object's exporter using IRemUnknown::RemAddRef to obtain a reference count for it. If the destination client is in fact the object's server, then special processing is required by the destination client. The remote reference counts being passed to it should, in effect, be "taken out of circulation," as what where heretofore remote references are being converted into local references. Thus, the reference counts present in the STDOBJREF are in fact decremented from the remote reference count for the IPID in question. Some objects have a usage model such that they do not need to be pinged at all; such objects are indicated by the presence of a flag in a STDOBJREF to an interface on the object. Objects which are not pinged in fact need not be reference counted either, though it is legal (but pointless) for a client to reference count the IPIDs of such objects. For all other objects, assuming a non-zero ping period, it is the responsibility of the holder of an interface reference on some object to ensure that pings reach the server frequently enough to prevent expiration of the object. The frequency used by a client depends on the ping period, the reliability of the channel between the client and the server, and the probability of failure (no pings getting through and possible premature garbage-collection) that the client is willing to tolerate. The ping packet and / or its reply may both request changes to the ping period. Through this mechanism, network traffic may be reduced in the face of slow links to busy servers. 2.5.1 Delta Pinging Without any further refinements, ping messages could be quite hefty. If machine A held 1024 remote object references (OIDs) on machine B, then it would send 16K byte ping messages. This would be annoying if the set of remote objects was relatively stable and the ping messages were the same from ping to ping. The delta mechanism reduces the size of ping messages. It uses a ping-set interface that allows the pinging of a single set to replace the pinging of multiple OIDs. Instead of pinging each OID, the client defines a set. Each ping contains only the set id and the list of additions and subtractions to the set. Objects that come and go within one ping period are removed from the set without ever having been added. The pinging protocol is carried out using two methods in the (DCE) interface IObjectExporter on the Object Exporter: ComplexPing, and SimplePing. ComplexPing is used by clients to group the set of OIDs that they must ping into UUID-tagged sets known to the server. These entire sets of OIDs can then be subsequently pinged with a single, short, call to SimplePing. 2.6 QueryInterface The IRemUnknown interface on the object-exporter specified object, in addition to servicing reference counting as described above also services QueryInterface calls for remote clients for IPIDs managed by that exporter. IRemUnknown::RemQueryInterface differs from IUnknown::QueryInterface in much the same way as RemAddRef and RemRelease differ from AddRef and Release, in that it is optimized for network access by being able to retrieve many interfaces at once. 2.7 Causality ID Each ORPC carries with it a UUID known as the causality id that connects together the chain of ORPC calls that are causally related. If an outgoing ORPC is made while servicing an incoming ORPC, the outgoing call is to have the same causality id as the incoming call. If an outgoing ORPC is made while not servicing an incoming ORPC, then a new causality id is allocated for it. Causality ids may in theory be reused as soon as it is certain that no transitively outstanding call is still in progress which uses that call. In practice, however, in the face of transitive calls and the possibility of network failures in the middle of such call chains, it is difficult to know for certain when this occurs. Thus, pragmatically, causality ids are not reusable. The causality id can be used by servers to understand when blocking or deferring an incoming call (supported in some COM server programming models) is very highly probable to cause a deadlock, and thus should be avoided. The causality id for maybe, idempotent, and broadcast calls must be set to null. If a server makes a ORPC call while processing such a call, a new causality id must be generated as if it were a top level call. 3. Data Types and Structures This following several sections present the technical details of the DCOM protocol. 3.1 DCE Packet Headers Object RPC sits entirely on top of DCE RPC. The following list describes the elements of ORPC that are specified above and beyond DCE RPC. * The object id field of the header must contain the IPID. * The interface id of the RPC header must contain the IID, even though it is not needed given the IPID. This allows ORPC to sit on top of DCE RPC. An unmodified DCE RPC implementation will correctly dispatch based on IID and IPID. An optimized RPC need only dispatch based on IPID. * An IPID uniquely identifies a particular interface on a particular object on a machine. The converse is not true; a particular interface on a particular object may be represented by multiple IPIDs. IPIDs are unique on their OXID. IPIDs may be reused, however reuse of IPIDs should be avoided. * Datagram, maybe, and idempotent calls are all allowed in ORPC. * Interface pointers may not be passed on maybe or idempotent calls. * Datagram broadcasts are not allowed in ORPC. * Faults are returned in the stub fault field of the DCE RPC fault packet. Any 32 bit value may be returned. Only the following value is pre-specified: RPC_E_VERSION_MISMATCH * DCE cancel is supported. * All interface version numbers must be 0.0. 3.2 ORPC Base Definitions There are several fundamental data types and structures on which the COM network protocol is built. These types are shown here in standard C header format. //////////////////////////////////////////////////////////// // Basic Definitions //////////////////////////////////////////////////////////// typedef unsiged long HRESULT; // 32-bit integer: success/failure typedef t_uuid UUID; // rename DCE-RPC type typedef UUID GUID; // Globally Unique IDentifier typedef unsigned hyper ID; // 64-bit integer typedef ID OXID; // Object Exporter Identifier typedef ID OID; // Object Identifer typedef ID SETID; // Ping Set Identifier typedef GUID IPID; // Interface Pointer Identifier typedef GUID* REFIPID; typedef GUID CID; // Causality Identifier #define CID_NULL uuid_null; // All zeros ////////////////////////////////////////////////////////////////// // ORPC Call Packet Format ////////////////////////////////////////////////////////////////// const unsigned short COM_MAJOR_VERSION = 4; const unsigned short COM_MINOR_VERSION = 1; typedef struct tagCOMVERSION { unsigned short MajorVersion; // Major version number unsigned short MinorVersion; // Minor version number } COMVERSION; const unsigned long ORPCF_NULL = 0; // no additional info // in packet const unsigned long ORPCF_LOCAL = 1; // call is local to this // machine const unsigned long ORPCF_RESERVED1 = 2; // reserved for local use const unsigned long ORPCF_RESERVED2 = 4; // reserved for local use const unsigned long ORPCF_RESERVED3 = 8; // reserved for local use const unsigned long ORPCF_RESERVED4 = 16; // reserved for local use // Extension to implicit parameters. typedef struct tagORPC_EXTENT { GUID id; // Extension identifier unsigned long size; // Extension size byte data[]; // [size_is((size+7)&~7)] } ORPC_EXTENT; // Array of extensions. typedef struct tagORPC_EXTENT_ARRAY { unsigned long size; // Num extents. unsigned long reserved; // Must be zero. ORPC_EXTENT** extent; // [size_is((size+1)&~1), unique] } ORPC_EXTENT_ARRAY; // implicit 'this' pointer which is the first [in] parameter on // every ORPC call. typedef struct tagORPCTHIS { COMVERSION version; // COM version number unsigned long flags; // ORPCF flags for presence of // other data unsigned long reserved1; // set to zero CID cid; // causality id of caller ORPC_EXTENT_ARRAY* extensions; // [unique] extensions } ORPCTHIS; // implicit 'that' pointer which is the first [out] parameter on // every ORPC call. typedef struct tagORPCTHAT { unsigned long flags; // ORPCF flags for presence // of other data ORPC_EXTENT_ARRAY *extensions; // [unique] extensions } ORPCTHAT; ////////////////////////////////////////////////////////////////// // Marshaled COM Interface Wire Format ////////////////////////////////////////////////////////////////// typedef enum tagMSHLFLAGS { MSHLFLAGS_NORMAL = 0, MSHLFLAGS_TABLESTRONG = 1, MSHLFLAGS_TABLEWEAK = 2, } MSHLFLAGS; // Tower IDs for common protocols const unsigned short NCADG_IP_UDP = 0x08; const unsigned short NCACN_IP_TCP = 0x07; const unsigned short NCADG_IPX = 0x0E; const unsigned short NCACN_SPX = 0x0C; const unsigned short NCACN_NB_NB = 0x12; const unsigned short NCACN_NB_IPX = 0x0D; const unsigned short NCACN_DNET_NSP = 0x04; const unsigned short NCALRPC = 0x10; // This is the return type for arrays of string bindings or protseqs // used by many ORPC interfaces. typedef struct tagSTRINGBINDING { unsigned short wTowerId; // Cannot be zero. unsigned short aNetworkAddr; // Zero terminated. } STRINGBINDING; // this value (invalid in DCE RPC) indicates to use default authz const unsigned short COM_C_AUTHZ_NONE = 0xffff; typedef struct tagSECURITYBINDING { unsigned short wAuthnSvc; // Must not be zero unsigned short wAuthzSvc; // Must not be zero unsigned short aPrincName; // NULL terminated } SECURITYBINDING; // DUALSTRINGARRAYS are the return type for arrays of network // addresses, arrays of endpoints and arrays of both used in // many ORPC interfaces typedef struct tagDUALSTRINGARRAY { unsigned short wNumEntries; // # of entries in array unsigned short wSecurityOffset; // Offset of security info // The array contains two parts, a set of STRINGBINDINGs // and a set of SECURITYBINDINGs. Each set is terminated by an // extra zero. The shortest array contains four zeros. unsigned short aStringArray[]; // [size_is(wNumEntries)] } DUALSTRINGARRAY; // arbitrary value to help ensure validity const unsigned long OBJREF_SIGNATURE = 0x574f454d; const unsigned long OBJREF_STANDARD = 0x1; const unsigned long OBJREF_HANDLER = 0x2; const unsigned long OBJREF_CUSTOM = 0x4; // Flag values for a STDOBJREF (standard part of an OBJREF). // SORF_OXRES1 - SORF_OXRES8 are reserved for the object exporters // use only, object importers must ignore them and must not enforce // MBZ. const unsigned long SORF_NULL = 0x0000; // convenient for init const unsigned long SORF_OXRES1 = 0x0001; // reserved by exporter const unsigned long SORF_OXRES2 = 0x0020; // reserved by exporter const unsigned long SORF_OXRES3 = 0x0040; // reserved by exporter const unsigned long SORF_OXRES4 = 0x0080; // reserved by exporter const unsigned long SORF_OXRES5 = 0x0100; // reserved by exporter const unsigned long SORF_OXRES6 = 0x0200; // reserved by exporter const unsigned long SORF_OXRES7 = 0x0400; // reserved by exporter const unsigned long SORF_OXRES8 = 0x0800; // reserved by exporter const unsigned long SORF_NOPING = 0x1000; // Pinging not required typedef struct tagSTDOBJREF { unsigned long flags; // SORF_ flags (see above) unsigned long cPublicRefs; // count of references passed OXID oxid; // oxid of server with this oid OID oid; // oid of object with this ipid IPID ipid; // ipid of Interface } STDOBJREF; // although this structure is conformant, it is always marshaled // in little-endian byte-order. typedef struct tagOBJREF { unsigned long signature; // must be OBJREF_SIGNATURE unsigned long flags; // OBJREF flags (see above) GUID iid; // interface identifier union { // [switch_is(flags), switch_type(unsigned long)] struct { // [case(OBJREF_STANDARD)] STDOBJREF std; // standard objref DUALSTRINGARRAY saResAddr; // resolver address } u_standard; struct { // [case(OBJREF_HANDLER)] STDOBJREF std; // standard objref CLSID clsid; // Clsid of handler code DUALSTRINGARRAY saResAddr; // resolver address } u_handler; struct { // [case(OBJREF_CUSTOM)] CLSID clsid; // Clsid of unmarshaling code unsigned long size; // size of data that follows byte* pData; // class specific data, // [size_is(size), ref] } u_custom; } u_objref; } OBJREF; // wire representation of a marshalled interface pointer, // always the little-endian form of an OBJREF typedef struct tagMInterfacePointer { ULONG ulCntData; // size of data byte abData[]; // [size_is(ulCntData)] data } MInterfacePointer, *PMInterfacePointer; // OXID Resolver information associated with each OXID. typedef struct tagOXID_INFO { IPID ipidRemUnknown; // IRemUnknown IPID for // object exporter unsigned long dwAuthnHint; // 32-bit RPC_C_AUTHN value DUALSTRINGARRAY* psa; // [unique] protocol and // security info } OXID_INFO; ////////////////////////////////////////////////////////////////// 3.3 OBJREF An OBJREF is the data type used to represent an actual marshaled object reference. An OBJREF can either be empty or assume one of three variations, depending on the degree to which the object being marshaled uses the hook architecture (IMarshal, etc.) in the marshaling infrastructure. The OBJREF structure is a union consisting of a switch flag followed by the appropriate data. 3.3.1 OBJREF_STANDARD Contains one interface of an object marshaled in standard form. Contains a standard reference, along with a set of protocol sequences and network addresses that can be used to bind to an OXID resolver that is able to resolve the OXID in the STDOBJREF. This is useful when marshaling a proxy to give to another machine (a.k.a. the "middleman" case). The marshaling machine can specify the saResAddr for the resolver on the server machine so that the unmarshaler does not need to call the marshaler (middleman) back to get this information. Further, the marshaler does not need to keep the OXID in its cache beyond the lifetime of its own references in order to satisfy requests from parties that it just gave the OBJREF to. Member Type Semantic ------------------------------------------------------------------ std STDOBJREF A standard object reference used to connect to the source object. saResAddr STRINGARRAY The resolver address. 3.3.2 OBJREF_HANDLER A marshaling of an object that wishes to use handler marshaling. For example, an object wishes to be represented in client address spaces by a proxy object that caches state. In this case, the class- specific information is just a standard reference to an interface pointer that the handler (proxy object) will use to communicate with the original object. See the IStdMarshalInfo interface. Member Type Semantic ------------------------------------------------------------------ std STDOBJREF A standard object reference used to connect to the source object (Section 3.4). clsid CLSID The CLSID of handler to create in the destination client. 3.3.3 OBJREF_CUSTOM A marshaling of an object which supports custom marshaling. The Custom format gives an object control over the representation of references to itself. For example, an immutable object might be passed by value, in which case the class-specific information would contain the object's immutable data. See the IMarshal interface. Member Type Semantic ------------------------------------------------------------------ clsid CLSID The CLSID of the object to create in the destination client. size unsigned long The size of the marshaled data provided by the source object and passed here in pData. pData byte* The data bytes that should be passed to IMarshal::UnmarshalInterface on a new instance of class clsid in order to initialize it and complete the unmarshal process. 3.4 STDOBJREF An instance of a STDOBJREF represents a COM interface pointer that has been marshaled using the standard COM network protocol. The members and semantics of the STDOBJREF structure are as follows: Member Semantic ------------------------------------------------------------------ flags Flag values taken from the enumeration SORFFLAGS. These are described in Section 3.5. crefs The number of reference counts on ipid that being transferred in this marshaling. Ipid The IPID of the interface being marshaled. oid The OID of the object to which ipid corresponds. oxid The OXID of the server that owns this OID. 3.5 SORFLAGS The various SORFLAGS values have the following meanings. The SORF_OXRESxxx bit flags are reserved for the object exporter's use only, and must be ignored by object importers. They need not be passed through when marshaling an interface proxy. Flag Value Meaning ------------------------------------------------------------------ SORF_NULL 0 Convenient for initialization. SORF_OXRES1 1 Reserved for exporter. SORF_OXRES2 32 Reserved for exporter. SORF_OXRES3 64 Reserved for exporter. SORF_OXRES4 128 Reserved for exporter. SORF_OXRES5 256 Reserved for exporter. SORF_OXRES6 512 Reserved for exporter. SORF_OXRES7 1024 Reserved for exporter. SORF_OXRES8 2048 Reserved for exporter. SORF_NOPING 4096 This OID does not require pinging. Further, all interfaces on this OID, including this IPID, need not be reference counted. Pinging and reference counting on this object and its interfaces are still permitted, however, though such action is pointless. 3.6 ORPCINFOFLAGS The various ORPCINFOFLAGS have the following meanings. Flag Meaning ------------------------------------------------------------------ INFO_NULL (Not a real flag. Merely a defined constant indicating the absence of any flag values.) INFO_LOCAL The destination of this call is on the same machine on which it originates. This value is never to be specified in calls which are not in fact local. INFO_RESERVED1 If INFO_LOCAL is set, then reserved for local use; otherwise, reserved for future use. INFO_RESERVED2 If INFO_LOCAL is set, then reserved for local use; otherwise, reserved for future use. INFO_RESERVED3 If INFO_LOCAL is set, then reserved for local use; otherwise, reserved for future use. INFO_RESERVED4 If INFO_LOCAL is set, then reserved for local use; otherwise, reserved for future use. Implementations may use the local and reserved flags to indicate any extra information needed for local calls. Note that if the INFO_LOCAL bit is not set and any of the other bits are set then the receiver should return a fault. 3.7 ORPCTHIS In every Request PDU that is an ORPC, the body (CL case) or the stub data (CO case) which normally contains the marshaled arguments in fact begins with an instance of the ORPCTHIS structure. The marshaled arguments of the COM interface invocation follow the ORPCTHIS; thus, viewed at the DCE RPC perspective, the call has an additional first argument. The ORPCTHIS is padded with zero-bytes if necessary to achieve an overall size that is a multiple of eight bytes; thus, the remaining arguments are as a whole eight byte aligned. As in regular calls, the causality id must be propagated. If A calls ComputePi on B, B calls Release on C (which gets converted to RemRelease), and C calls Add on A, A will see the same causality id that it called B with. Member Type Semantic ------------------------------------------------------------------ version COMVERSION The version number of the COM protocol used to make this particular ORPC. The initial value will be 4.1. Each packet contains the sender's major and minor ORPC version numbers. The client's and server's major versions must be equal. Backward compatible changes in the protocol are indicated by higher minor version numbers. Therefore, a server's minor version must be greater than or equal to the client's. However, if the server's minor version exceeds the client's minor version, it must return the client's minor version and restrict its use of the protocol to the minor version specified by the client. A protocol version mismatch causes the RPC_E_VERSION_MISMATCH ORPC fault to be returned. Flags unsigned long Flag values taken from the enumeration ORPCINFOFLAGS (section 3.6). Reserved unsigned long Must be set to zero. Cid CID The causality id of this ORPC. Extensions ORPC_EXTENT_ARRAY* The body extensions, if any, passed with this call. Body extensions are GUID-tagged blobs of data which are marshaled as an array of bytes. Extensions are always marshaled with initial eight byte alignment. Body extensions which are presently defined are described in Section 3.10. The cid field contains the causality id. Each time a client makes a unique call, a new causality id is generated. If a server makes a call while processing a request from a client, the new call must have the same causality id. This allows simple servers to avoid working on more then one thing at a time (for example A calls B calls A again, meanwhile C tries to call A with a new causality id). It tells the server that he is being called because he asked someone to do something for him. There are several interesting exceptions. The causality id for maybe and idempotent calls must be set to CID_NULL. If a server makes a ORPC call while processing such a call, a new causality id must be generated. In the face of network failures, the same causality id may end up in use by two independent processes at the same time. If A calls B calls C calls D and C fails, both B and D can independently, simultaneously make calls to E with the same causality id. The extensions field contains extensions to the channel header, described in Section 3.10. Note that in order to force the ORPCTHIS header to be 8 byte aligned an even number of extensions must be present and the size of the extension data must be a multiple of 8. 3.8 ORPCTHAT In every Response PDU that is an ORPC, the body (CL case) or the stub data (CO case) which normally contains the marshaled output parameters in fact begins with an instance of the ORPCTHAT structure. The marshaled output parameters of the COM interface invocation follow the ORPCTHAT; thus, viewed at the DCE RPC perspective, the call has an additional output parameters. The ORPCTHAT is padded with zero-bytes if necessary to achieve an overall size that is a multiple of eight bytes; thus, the remaining output parameters as a whole eight byte aligned. Member Type Semantic ------------------------------------------------------------------ flags unsigned long Flag values taken from the enumeration ORPCINFOFLAGS (section 3.6). Extensions ORPC_EXTENT_ARRAY* The body extensions, if any, returned by this call. See Section 3.10 for a general description of body extensions as well as a description of existing well-known extensions. 3.9 HRESULTs HRESULTs are the 32-bit return value from all ORPC methods. The following is a partial list of already defined HRESULTs. Further details TBS. S_OK Success. (0x00000000) E_OUTOFMEMORY Insufficient memory to complete the call. (0x80000002) E_INVALIDARG One or more arguments are invalid. (0x80000003) E_NOINTERFACE No such interface supported (0x80000004) E_ACCESSDENIED A secured operation has failed due to (0x80070005) inadequate security privileges. E_UNEXPECTED Unknown, but relatively catastrophic (0x8000FFFF) error. RPC_E_INVALID_OXID The object exporter was not found. (0x80070776) RPC_E_INVALID_OID The object specified was not found or (0x80070777) recognized. RPC_E_INVALID_SET The object exporter set specified was (0x80070778) not found. 3.10 Body Extensions Body Extensions are UUID-tagged blobs of data which are useful for conveying additional, typically out-of-band, information on incoming invocations (within ORPCTHIS, Section 3.7) and in replies (within ORPCTHAT, Section 3.8). Any implementations of the DCOM protocol may define its own extensions with their own UUIDs. Implementations should skip over extensions which they do not recognize or do not wish to support. Body Extensions are marshaled as an array of bytes with initial eight byte alignment. The following sections descrobe several existing body extensions. 3.10.1 Debugging Extension: {f1f19680-4d2a-11ce-a66a-0020af6e72f4} This extension aids in debugging ORPC. In particular it is designed to allow single stepping over an ORPC call into the server and out of the server into the client. Further details TBS. 3.10.2 Extended Error Extension: {f1f19681-4d2a-11ce-a66a-0020af6e72f4} The extended error information body extension conveys extended error information concerning the original root cause of a error back to a caller can deal with it. This extension is only semantically useful in Response and Fault PDUs. It is intended that this error information is suitable for displaying information to a human being who is the user; this information is not intended to be the basis for logic decisions in a piece of client code, for doing so couples the client code to the implementation of the server. Rather, client code should act semantically only on the information returned through the interface that it invokes. Further details TBS. 4. IRemUnknown interface The IRemUnknown interface is used by remote clients for manipulating reference counts on the IPIDs that they hold and for obtaining additional interfaces on the objects on which those IPIDs are found. References are kept per interface rather then per object. This interface is implemented by the COM "OXID object" associated with each OXID (nb. not each Object Exporter). The IPID for the IRemUnknown interface on this object is returned from IObjectExporter::ResolveOxid; see Section 5.2.1. An OXID object need never be pinged; its interfaces (this IPID included) need never be reference counted. IRemUnknown is specified as follows: //--------------------------------------------------------------- // The remote version of IUnknown. Once instance of this // interface exists per OXID (whether an OXID represents either // a thread or a process is implementation specific). This // interface is passed along during OXID resolution. It is used // by clients to query for new interfaces, get additional // references (for marshalling), and release outstanding // references. //--------------------------------------------------------------- // return structure from a QI call typedef struct tagREMQIRESULT { HRESULT hResult; // result of call STDOBJREF std; // data for returned interface } REMQIRESULT; // structure passed to AddRef/Release to specify interface and // count of references to Add/Release. typedef struct tagREMINTERFACEREF { IPID ipid; // ipid to AddRef/Release unsigned long cPublicRefs; unsigned long cPrivateRefs; } REMINTERFACEREF; [object, uuid(99fcff28-5260-101b-bbcb-00aa0021347a)] interface IRemUnknown : IUnknown { HRESULT RemQueryInterface( [in] IPID ipid, // interface to QI on [in] unsigned long cRefs, // count of AddRefs requested // for each interface [in] unsigned short cIids, // count of IIDs that follow [in, size_is(cIids)] IID* iids, // IIDs to query for [out, size_is(cIids)] REMQIRESULT** ppQIResults // results returned ); HRESULT RemAddRef( [in] unsigned short cInterfaceRefs, [in, size_is(cInterfaceRefs)] REMINTERFACEREF InterfaceRefs[], [out, size_is(cInterfaceRefs)] HRESULT* pResults ); HRESULT RemRelease( [in] unsigned short cInterfaceRefs, [in, size_is(cInterfaceRefs)] REMINTERFACEREF InterfaceRefs[], ); } 4.1 IRemUnknown::RemQueryInterface QueryInterface for and return the result thereof for zero or more interfaces from the interface behind the IPID ipid. ipid must designate an interface derived from IUnknown (recall that all remoted interfaces must derive from IUnknown). The QueryInterface calls on the object that are used to service this request are conducted on the IUnknown interface of the object. Argument Type Semantic ------------------------------------------------------------------ ipid IPID The interface on an object from whom more interfaces are desired. cRefs REFCNT The number of references sought on each of the returned IIDs. cIids USHORT The number of interfaces being requested. iids IID* The list of IIDs that name the interfaces sought on this object. ppQIResults REMQIRESULT** The place at which the array of the results of the various QueryInterface calls are returned. Return Value Meaning --------------------------------------------------------------- S_OK Success. An attempt was made to retrieve each of the requested interfaces from the indicated object; that is, QueryInterface was actually invoked for each IID. E_INVALIDARG One or more arguments (likely ipid) were invalid. No result values are returned. E_UNEXPECTED An unspecified error occurred. No result values are returned. 4.1.1 REMQIRESULT The REMQIRESULT structure contains the following members: Member Type Semantic ------------------------------------------------------------------ hResult HRESULT The result code from the QueryInterface call made for the requested IID. std STDOBJREF The data for the returned interface. Note that if hResult indicates failure then the contents of STDOBJREF are undefined. 4.2 IRemUnknown::RemAddRef Obtain and grant ownership to the caller of one or more reference counts on one or more IPIDs managed by the corresponding OXID. Argument Type Semantic ------------------------------------------------------------------ cRefs unsigned short The size of the rgRefs array. rgRefs REMINTERFACEREF An array of IPID, cRefs pairs, cRefs large. Each IPID indicates an interface managed by this OXID on whom more reference counts are sought. The corresponding reference count (cRefs), which may not be zero (and thus is one or more), indicates the number of reference counts sought on that IPID. Return Value Meaning --------------------------------------------------------------- S_OK Success. An attempt was made to retrieve each of the requested interface references. E_INVALIDARG One or more of the IPIDs indicated were not in fact managed by this OXID, or one or more of the requested reference counts was zero. None of the requested reference counts have been granted to the caller; the call is a no-op. E_UNEXPECTED An unspecified error occurred. It is unknown whether any or all of the requested reference counts have been granted. A useful optimization is for a caller to RemAddRef more than needed. When a process receives an out marshaled interface, it receives one reference count. If the process wishes to pass that interface as an out parameter, it must get another reference to pass along. Instead, the process (or middleman) should get a large number of references. Then if the interface is passed out multiple times, no new remote calls are needed to gain additional references. A marshaler may optionally specify more than one reference in the STDOBJREF when marshaling an interface. This allows the middle man case to pre-fill its cache of references without making an extra RemAddRef call. The number of references passed is always specified in the STDOBJREF field. 4.3 IRemUnknown::RemRelease Release ownership of one or more reference counts on one or more IPIDs managed by the corresponding OXID. Argument Type Semantic ------------------------------------------------------------------ cRefs unsigned short The size of the rgRefs array. rgRefs REMINTERFACEREF An array of IPID, cRefs pairs, cRefs large. Each IPID indicates an interface managed by this OXID on whom more reference counts are being returned. The corresponding reference count, which may not be zero (and thus is one or more), indicates the number of reference counts returned on that IPID. Return Value Meaning --------------------------------------------------------------- S_OK Success. An attempt was made to release each of the requested interface references. E_INVALIDARG One or more of the IPIDs indicated were not in fact managed by this OXID, or one or more of the requested reference counts was zero. None of the offered reference counts have been accepted by the server; the call is a no-op. E_UNEXPECTED An unspecified error occurred. It is unknown whether any or all of the offered reference counts have been accepted. 5. The Object Exporter Each machine that supports the COM network protocol supports a one- per-machine service known as the machine's `Object Exporter.' Communication with an Object Exporter is via a DCE RPC, not an ORPC. The Object Exporter performs several services: It caches and returns to clients when asked the string bindings necessary to connect to OXIDs of exported objects for which this machine is it either itself a client or is the server. It receives pings from remote client machines to keep its own objects alive. These services are carried out through an RPC interface (not a COM interface) known as IObjectExporter. An Object Exporter may be asked for the information required to connect to one of two different kinds of OXIDs, either the OXIDs associated with its own objects, or the OXIDs associated with objects for which it itself is a client, and which it has passed on to a second client machine. This second case, where one marshals an object from one client machine to a second, is colloquially referred to the "middleman" case. In the middleman case, the exporter is required to retain the connection information associated with the OXIDs that it passes on until it is certain that that the second client machine no longer needs them. 5.1 Object Exporter Ports/Endpoints The Object Exporter resides at different endpoints (ports) depending on the transport being used. The Object Exporter optimally resides at the same endpoints as the DCE RPC Endpoint Mapper (EPM). To accommodate systems where DCOM will coexist with existing DCE RPC installations (i.e., where an EPM and presumably a complete DCE RPC runtime already exists), the DCOM implementation on that system will register its interfaces with the DCE EPM and all DCOM implementations must be able to fall back if they make DCOM-specific calls on the DCE EPM endpoint which fail. Protocol String Name(s) Description Endpoint ------------------------------------------------------------------ ncadg_ip_udp, ip CL over UDP/IP 135 ncacn_ip_tcp CO over TCP/IP 135 ncadg_ipx CL over IPX TBD ncacn_spx CO over SPX TBD ncacn_nb_nb CO over NetBIOS over NetBEUI TBD ncacn_nb_tcp CO over NetBIOS over TCP/IP 135 ncacn_np CO over Named Pipes TBD ncacn_dnet_nsp CO over DECNet Network Services 96 Protocol (DECNet Phase IV) ncacn_osi_dna CO over Open Systems 69 Interconnection (DECNet Phase V) ncadg_dds, dds CL over Domain Datagram Service 12 ncahttp Hybrid over HTTP (TBS) 80 5.2 The IObjectExporter Interface IObjectExporter is defined as follows: [uuid(99fcfec4-5260-101b-bbcb-00aa0021347a)] interface IObjectExporter { [idempotent] error_status_t ResolveOxid( [in] handle_t hRpc, [in] OXID *pOxid, [in] unsigned short cRequestedProtseqs, [in, ref, size_is(cRequestedProtseqs)] unsigned short arRequestedProtseqs[], [out, ref] DUALSTRINGARRAY **psaOxidBindings, [out, ref] IPID *pipidRemUnknown, [out, ref] unsigned long *pdwAuthnHint ); [idempotent] error_status_t SimplePing( [in] handle_t hRpc, [in] SETID *pSetId ); [idempotent] error_status_t ComplexPing( [in] handle_t hRpc, [in] SETID *pSetId, [in] unsigned short SequenceNum, [in] unsigned short cAddToSet, [in] unsigned short cDelFromSet, [in, unique, size_is(cAddToSet)] OID AddToSet[], [in, unique, size_is(cDelFromSet)] OID DelFromSet[] [out] unsigned short *pPingBackoffFactor ); } 5.2.1 IObjectExporter::ResolveOxid Return the string bindings necessary to connect to a given OXID object. On entry, arRequestedProtseqs contains the protocol sequences the client is willing to use to reach the server. These should be decreasing order of protocol preference, with no duplicates permitted. Local protocols (such as "ncalrpc") are not permitted. On exit, psaOxidBindings contains the string bindings that may be used to connect to the indicated OXID; if no such protocol bindings exist which match the requested protocol sequences, NULL may be returned. The returned string bindings are in decreasing order of preference of the server, with duplicate string bindings permitted (and not necessarily of the same preferential priority), though of course duplicates are of no utility. Local protocol sequences may not be present; however, protocol sequences that were not in the set of protocol sequences requested by the client may be. The string bindings returned need not contain endpoints; the endpoint mapper will be used as usual to obtain these dynamically. If a ResolveOxid call is received for which the recipient Object Exporter is a middleman, the action required of the middleman depends on how the ordered list of requested protocol sequences (arRequested-Protseqs) relate to lists of protocol sequences previously known by the middleman to have been previously requested of the server. If the list of requested protocol sequences is a (perhaps non-proper) subset in order of a protocol sequence list previously requested of the server, then the corresponding cached string bindings may be returned immediately to the caller without actually communicating with the server. Otherwise, the actual psaRequestedProtseqs must be forwarded to the server, and the returned string bindings propagated back to the client. In such cases, it behooves the middleman to cache the returned string bindings for use in later calls. Argument Type Description ------------------------------------------------------------------ hRpc handle_t An RPC binding handle used to make the request. pOxid OXID* The OXID for whom string bindings are requested. The OXID may or may not represent a process on the machine that receives the ResolveOxid call. cRequestedProtseqs unsigned short The number of protocol sequences requested. arRequestedProtseqs unsigned short[] arRequestedProtseqs must be initialized with all the protocol id's the client is willing to use to reach the server. It cannot contain local protocol sequences. The object exporter must take care of local lookups privately. The protocol sequences are in order of preference or random order. No duplicates are allowed. See the Lazy Use Protseq section for more details. psaOxidBindings STRINGARRAY** The string bindings supported by this OXID, in preferential order. Note that these are Unicode strings. pipidRemUnknown IPID* The IPID to the IRemUnknown interface the OXID object for this OXID. pdwAuthnHint unsigned long* A value taken from the RPC_C_AUTHN constants. A hint to the caller as to the minimum authentication level which the server will accept. Return Value Meaning --------------------------------------------------------------- S_OK Success. The requested information was returned. RPC_E_INVALID_OXID This OXID is unknown to this Object Exporter, and thus no information was returned. E_UNEXPECTED An unspecified error occurred. Some of the requested information may not be returned. Since the object exporter ages string bindings and discards them, object references are transient things. They are not meant to be stored in files or otherwise kept persistently. In any case, well known object references can be constructed from well known string bindings, IPIDs and OIDs. Conversely, since object references are aged, it is the responsibility of each client to unmarshal them and begin pinging them in a timely fashion. The basic use of the ResolveOxid method is to translate an OXID to string bindings. Put another way, this method translates an opaque process and machine identifier to the information needed to reach that machine and process. There are four interesting cases: 1. Looking up an OXID the first time an interface is unmarshaled on a machine, 2. Looking up an OXID between a pair of machines that already have connections, and 3. Looking up an OXID from a middleman, and 4. Looking up string bindings with unresolved endpoints (lazy use protseq). Another interesting topic is garbage collection of stored string binding vectors. 5.2.1.2 Lookup Between Friends The case of a lookup between two machines that have already established communication is the easiest. In this scenario there are two machines, A and B. Process D already has an interface pointer to process F. Object exporter C already knows the string bindings for object exporter E and process F, but not process G. +=========+===========+==+=========+===========+===========+ | OX C | Process D | | OX E | Process F | Process G | +=========+===========+==+=========+===========+===========+ | | call F | | | | | +---------+-----------+--+---------+-----------+-----------+ | | | | | pass out | | | | | | | ref to G | | +---------+-----------+--+---------+-----------+-----------+ | | receive | | | | | | | out ref | | | | | | | to G | | | | | +---------+-----------+--+---------+-----------+-----------+ | | ask local | | | | | | | OX to | | | | | | | resolve G | | | | | +---------+-----------+--+---------+-----------+-----------+ | ask OX | | | | | | | E to | | | | | | | resolve | | | | | | | OXID G | | | | | | +---------+-----------+--+---------+-----------+-----------+ | | | | lookup | | | | | | | G and | | | | | | | return | | | | | | | endpts | | | +---------+-----------+--+---------+-----------+-----------+ | cache | | | | | | | binding | | | | | | | vector | | | | | | | for G, | | | | | | | return | | | | | | | to D | | | | | | +---------+-----------+--+---------+-----------+-----------+ | | ready to | | | | | | | call G | | | | | +---------+-----------+--+---------+-----------+-----------+ Object exporter E knows the string bindings for all the servers on its machine, i.e. processes F and G. Process D calls process F and gets a reference to process G. Since process D has never seen the OXID for G before, it asks its local object exporter to resolve G. Process D also has to tell object exporter C where it got the reference from, in this case, process F. Object exporter C does not recognize the OXID G. However it does recognize the OXID F and knows the object exporter E is on the same machine as process F. So OX C calls ResolveOxid on OX E. OX E recognizes G and passes the string bindings back to OX C with the machine id B. OX C caches this information so that if D ever gets a reference from G, it knows who to ask to resolve that reference. 5.2.1.2 My First Lookup The previous example assumes that OX C already knows about OX E and process D is already talking to process F. Setting up the first connection between D and F (as well as C and E) is a tricky business known as activation. ORPC as described in this specification does not include activation models. Thus different vendors may have different activation models. However there is one basic form of activation shared by all ORPC. If two processes can communicate via DCE RPC, they can pass long standard object references. While this is not expected to be a common form of activation, it is a simple one that should certainly work across all ORPC implementations. Thus if D and E have established DCE RPC (or raw RPC) communication, they can bootstrap ORPC communication as follows. +==============+==============+==+==============+==============+ | OX C | Process D | | OX E | Process F | +==============+==============+==+==============+==============+ | | | | register | | | | | | endpointts & | | | | | | OXID for F | | |--------------+--------------+--+--------------+--------------+ | | call F with | | | | | | raw RPC | | | | |--------------+--------------+--+--------------+--------------+ | | | | | pass an out | | | | | | ref to F, | | | | | | pass IID as | | | | | | additional | | | | | | parameter | |--------------+--------------+--+--------------+--------------+ | | tell C the | | | | | | OXID_INFO | | | | | | for F. | | | | | | includes net | | | | | | address(es) | | | | |--------------+--------------+--+--------------+--------------+ | compute the | | | | | | string bind- | | | | | | ings for OX | | | | | | E from F. | | | | | | ask E to | | | | | | resolve F | | | | | | to get end- | | | | | | points for F | | | | | |--------------+--------------+--+--------------+--------------+ | | | | return end- | | | | | | points | | |--------------+--------------+--+--------------+--------------+ | | ready to | | | | | | call F | | | | |--------------+--------------+--+--------------+--------------+ This example points out that there has to be a local interface between processes and the local object exporter. 5.2.1.3 Middleman Lookup The next case shows how lookup works between multiple machines. Suppose that E has a reference to G and G has a reference to I. Similarly, D knows about F and G and F knows about H and I. What happens if G passes a reference to I over to E? +=============+===========+======+============+========+===========+ | OX D | Process E | OX F | Process G | OX H | Process I | +=============+===========+======+============+========+===========+ | | call G | | | | | +-------------+-----------+------+------------+--------+-----------+ | | | | return a | | | | | | | long ref | | | | | | | to I | | | +-------------+-----------+------+------------+--------+-----------+ | | ask D to | | | | | | | lookup I | | | | | +-------------+-----------+------+------------+--------+-----------+ | since it's | | | | | | | long, call | | | | | | | ResolveOxid | | | | | | | on H | | | | | | +-------------+-----------+------+------------+--------+-----------+ | | | | | return | | | | | | | end- | | | | | | | points | | | | | | | to I | | +-------------+-----------+------+------------+--------+-----------+ | compute | | | | | | | bindings to | | | | | | | I from end- | | | | | | | points & | | | | | | | network | | | | | | | addresses | | | | | | | return | | | | | | | string | | | | | | | bindings to | | | | | | | E | | | | | | +-------------+-----------+------+------------+--------+-----------+ Note that when process G returned a reference to I, it used he long form of the OBJREF which includes the protocol id's and network addresses of the OXID resolver for process I (in this example, the addresses for OX H). This would results in OX D calling OX H directly, rather than needing to call OX F. The advantage of this is that if no references to process I needed by OX F, it could remove it from its OXID cache at any time, rather than keeping it around at least until OX D has had a chance to call it back to resolve OXID I. 5.2.1.4 Lazy Use Protseq In a homogeneous network, all machines communication via the same protocol sequence. In a heterogeneous network, machines may support multiple protocol sequences. Since it is often expensive in resources to allocate endpoints (RpcServerUseProtseq) for all available protocol sequences, ORPC provides a mechanism where they may be allocated on demand. To implement this extension fully, there are some changes in the server. However, changes are optional. If not implemented, ORPC will still work correctly if less optimally in heterogeneous networks. There are two cases: the server implements the lazy use protocol or it does not. If the server is using the lazy use protseq protocol, the use of ResolveOxid is modified slightly. When the client OX calls the server OX, it passes the requested protseq vector. If none of the requested protseqs have endpoints allocated in the server, the server OX performs some local magic to get one allocated. If the server does not implement the lazy use protseq protocol, then all protseqs are registered by the server and contain complete endpoints. However, if they are not, the endpoint mapper can be used to forward calls to the server. This requires that all server IIDs are registered in the endpoint mapper. It also allows a different lazy use protseq mechanism. The endpoint mapper can perform some local magic to force the server to allocate an endpoint. This is less efficient since no OXs ever learn the new endpoints. The client will always pass in a vector of requested protseqs which the server can ignore if it does not implement the lazy use protseq protocol. 5.4 IObjectExporter::SimplePing Pings provide a mechanism to garbage collect interfaces. If an interface has references but is not being pinged, it may be released. Conversely, if an interface has no references, it may be released even though it has recently been pinged. SimplePing just pings the contents of a set. The set must be created with ComplexPing (section 5.2.3). Ping a set, previously created with IObjectExporter::ComplexPing, of OIDs owned by this Object Exporter. Note that neither IPIDs nor OIDs may be pinged, only explicitly created SETIDs. Argument Type Description ------------------------------------------------------------------ hRpc handle_t An RPC binding handle used to make the request. pSetId SETID* A SETID previously created with IObjectExporter::ComplexPing on this same Object Exporter. Return Value Meaning --------------------------------------------------------------- S_OK Success. The set was pinged. RPC_E_INVALID_SET This SETID is unknown to this Object Exporter, and thus the ping did not occur. E_UNEXPECTED An unspecified error occurred. It is not known whether the ping was done or not. 5.5 IObjectExporter::ComplexPing Ping a ping set. Optionally, add and/or remove some OIDs from the set. Optionally, adjust some ping timing parameters associated with the set. After a set is defined, a SimplePing will mark the entire contents of the set as active. After a set is defined, SimplePing should be used to ping the set. ComplexPing need only be used to adjust the contents of the set (or to adjust the time-out). Ping set ids (SETIDs) are allocated unilaterally by a client Object Exporter. The client Object Exporter then communicates with the server Object Exporter to add (and later remove) OIDs from the ping set. Clients must ensure the SETIDs pinged at a given server are unique over all of that server's clients. Thus, the client must only use SETIDs that it knows not to be in use as SETIDs by other clients on that server. (In practice, clients allocate SETIDs as globally unique). A client may use as many sets as it likes, though using fewer sets is more efficient. Each OID owned by a server Object Exporter may be placed in zero or more ping sets by the various clients of the OID. The client owner of each such set will set a ping period and a ping time-out count for the set, thus determining an overall time-out period for the set as the product of these two values. The time-out period is implicitly applied to each OID contained in the set and to future OIDs that might add be added to it. The server Object Exporter is responsible for ensuring that an OID that it owns does not expire until at least a period of time t has elapsed without that OID being pinged, where t is the maximum time-out period over all the sets which presently contain the given OID, or, if OID is not presently in any such sets but was previously, t is the time-out period for the last set from which OID was removed at the instant that that removal was done; otherwise, OID has never been in a set, and t is a default value (ping period equals 120 seconds, ping time-out count equals three (3), t equals 360 seconds, or six (6) minutes). Clients are responsible for pinging servers often enough to ensure that they do not expire given the possibility of network delays, lost packets, and so on. If a client only requires access to a given object for what it would consider less than a time-out period for the object (that is, it receives and release the object in that period of time), then unless it is certain it has not itself passed the object to another client it must be sure to nevertheless ping the object (a ComplexPing that both adds and removes the OID will suffice). This ensures that an object will not expire as it is passed through a chain of calls from one client to another. An OID is said to be pinged when a set into which it was previously added and presently still resides is pinged with either a SimplePing or a ComplexPing, or when it is newly added to a set with ComplexPing. Note that these rules imply that a ComplexPing that removes an OID from a set still counts as a ping on that OID. In addition to pinging the set SETID, this call sets the time-out period of the set as the product of a newly-specified ping period and a newly-specified "ping count to expiration;" these values take effect immediately. Ping periods are specified in tenths of a second, yielding a maximum allowable ping period of about 1 hr 50 min. Adjustment of the time-out period of the set is considered to happen before the addition of any new OIDs to the set, which is in turn considered to happen before the removal of any OIDs from the set. Thus, an OID that is added and removed in a single call no longer resides in the set, but is considered to have been pinged, and will have as its time-out at least the time-out period specified in that ComplexPing call. On exit, the server may request that the client adjust the time-out period; that is, ask it to specify a different time-out period in subsequent calls to ComplexPing. This capability can be used to reduce traffic in busy servers or over slow links. The server indicates its desire through the values it returns through the variables pReqSetPingPeriod and pReqSetNumPingsToTimeOut. If the server seeks no change, it simply returns the corresponding values passed by the client; if it wishes a longer time-out period, it indicates larger values for one or both of these variables; if it wishes a smaller period, it indicates smaller values. When indicating a larger value, the server must start immediately acting on that larger value by adjusting the time-out period of the set. However, when indicating a smaller value, it must consider its request as purely advice to the client, and not take any action: if the client wishes to oblige, it will do so in a subsequent call to ComplexPing by specifying an appropriate time-out period. Argument Type Description ------------------------------------------------------------------ hRpc handle_t An RPC binding handle used to make the request. pSetId SETID The SETID being manipulated. SequenceNum unsigned short The sequence number allows the object exporter to detect duplicate packets. Since the call is idempotent, it is possible for duplicates to get executed and for calls to arrive out of order when one ping is delayed. cAddToSet unsigned short The size of the array AddToSet. cDelFromSet unsigned short The size of the array DelFromSet. AddToSet OID[] The list of OIDs which are to be added to this set. Adding an OID to a set in which it already exists is permitted; such an action, as would be expected, is considered to ping the OID. DelFromSet OID[] The list of OIDs which are to be removed from this set. Removal counts as a ping. An IPID removed from a set will expire after the number of ping periods has expired without any pings (not the number of ping periods - 1). If an id is added and removed from a set in the same ComplexPing, the id is considered to have been deleted. pPingBackoffFactor unsigned short* Acts as a hint (only) from the server to the client in order to reduce ping traffic. Clients are requested to not ping more often than (1<<*pPingBackoffFactor)* (BasePingInterval=120) seconds, and the number of pings until timeout remains unchanged at the default of 3. Clients may choose to assume that this parameter is always zero. Return Value Meaning --------------------------------------------------------------- S_OK Success. The set was pinged, etc. RPC_E_INVALID_OID Indicates that some OID was not recognized. There is no recovery action for this error, it is informational only. E_ACCESSDENIED Access is denied. E_OUTOFMEMORY There was not enough memory to service the call. The caller may retry adding OIDs to the set on the next ping. E_UNEXPECTED An unspecified error occurred. It is not known whether the ping or any of the other actions were done or not. 6. Security Considerations In general, like any generic data transfer protocol, DCOM cannot regulate the content of the data that is transferred, nor is there any a priori method of determining the sensitivity of any particular piece of information within the context of any given ORPC. Specifically, however, DCOM entirely leverages the security infrastructure defined by DCE RPC, which allows for various forms of authentication, authorization, and message integrity. Further details TBS. 7. Acknowledgements As previously noted, the DCOM protocol highly leverages the DCE RPC Specification [CAE RPC], and we again acknowledge its usefulness to this specification. The DCOM protocol itself is the combined effort of a large number of people. The following individuals in particular were critical to the definitions which appear in this specification. Bob Atkinson Deborah Black Vibhas Chandorkar Richard Draves Mario Goertzel Rick Hill Gregory Jensenworth David Kays Paul Leach Alex Mitchell Kevin Ross Mark Ryland Bharat Shah Tony Williams 8. References [CAE RPC] CAE Specification, X/Open DCE: Remote Procedure Call, X/Open Company Limited, Reading, Berkshire, UK (xopen.co.uk), 1994. X/Open Document Number C309. ISBN 1-85912-041-5. (also available online through from the OSF at after registration) [COM] The Component Object Model Specification, Version 0.9, October 24, 1995, Microsoft Corporation. (also available online from Microsoft at 9. Author's Addresses Nat Brown One Microsoft Way Redmond, WA 98052-6399, U.S.A. Fax: +1 (206) 936 7329 Email: natbro@microsoft.com Charlie Kindel One Microsoft Way Redmond, WA 98052-6399, U.S.A. Fax: +1 (206) 936 7329 Email: ckindel@microsoft.com