Method and apparatus to allow remotely located computer programs and/or data to be accessed on a local computer in a secure, time-limited manner, with persistent caching

[0002] This application claims priority from provisional application No. 60/049,759, filed Jun. 16, 1997.

[0003] The present invention relates to network file systems, and in particular, methods of allowing remotely located programs and/or data to be accessed on a local computer without exposing the remote program and/or data (henceforth collectively referred to as the target program) to indiscriminant copying. In addition, methods of limiting the length of time that the target program can be executed or accessed on the local computer are applicable. A persistent caching scheme is also described which allows subsequent accesses to the program or data to proceed with reduced download requirements even across different connection sessions.

[0004] There are many instances in which it would be desirable to allow a remotely located target program to be executed on a local computer without exposing the target program in question to the threat of indiscriminate copying and propagation. Additionally, time-limiting the execution of the remotely located target program would add additional benefits and control. Some of the areas that could benefit from such a capability are the rental of computer programs over the Internet, try-before-you-buy evaluation of computer programs, and pay-per-view of internet-based information sources, among others.

[0005] Traditional network file system drivers allow local computers to access remotely-located programs and/or data as if they were local, but also allowed those programs and data to be copied to the local computer, and then be propagated indiscriminately.

[0006] FIGS. 1A to 1C—Flowchart of Decimation Procedure, showing how the target program on the install system is transformed into an index file and a set of chunk files on the runtime server.

[0007] FIGS. 2A to 2D—Flowchart of Client-Agent Procedure, showing how the client-agent is initialized, and the steps taken to initiate access to the target program.

[0008] FIG. 3—Flowchart of Client-Agent Registration Procedure, showing how the client-agent determines whether it is the primary client-agent or a secondary client-agent.

[0009] FIG. 4—Flowchart of Index File Insertion Procedure, showing how an index file is inserted into the FSD, thereby creating a virtual directory structure.

[0010] FIG. 5—Flowchart of Add Process Procedure, showing how a process is added to the process access list, thereby granting access to the process to use the virtual directory structure.

[0011] FIG. 6—Flowchart of Index File to Cache File Binding Procedure, showing how a downloaded index file is synchronized with the cache file.

[0012] FIG. 7—Flowchart of Set Cache Info during File Open Procedure, showing the steps necessary to link a given file to a cache handle, allowing the file's contents to be cached.

[0013] FIGS. 8A to 8C—Flowchart of Retrieve Cache Block Procedure, showing how remote data blocks are migrated from the remote system, to the on-disk cache, to the in-memory cache, and finally used by the system.

[0014] FIG. 9—Flowchart of General Access Control Procedure, showing how access to the virtual directory contents is affected by request type, process ID, and process access lists.

[0015] FIG. 10—Flowchart of Primary Client-Agent Deregistration Procedure, showing how the client-agent is shutdown, and all target programs terminated.

[0016] This invention extends a traditional network file system driver with fine-grain, process-specific access control, time-limited access, and a persistent caching scheme. Only those portions of the program or data actually used are downloaded, and the parts that are downloaded are compressed for improved network performance and encrypted for security. A persistent caching scheme that allows the working-set of the target program to reside on the client machine (limited, by the user configurable size of the local cache) in a secure manner allows subsequent executions of the target program to load and execute with reduced requirements for downloading from the remote computer.

[0017] The present invention provides a method and apparatus for allowing a remotely-located computer program or data to be accessed on a local computer in such a way as to severely limit the possibility of said program or data being indiscriminately copied and propagated, plus providing time limits on the access to said program or data. A persistent caching scheme improves performance of subsequent accesses.

[0018] A core component of the present invention is a network file system driver that simulates a local drive, but only allows access to that drive to designated computer processes. Process ID's are used to limit access to the program and/or data. A client program referred to as the client-agent accomplishes control of the network file system driver.

[0019] The persistent cache retains sufficient information about the blocks contained within to allow subsequent accesses, even when separated by disjoint sessions, to continue to be referenceable. All data in the cache is maintained in an encrypted state; at no time does the decrypted version of the program and/or data reside on the local disk.

[0020] The method whereby a remotely located program and/or data can be accessed securely in a time-constrained manner includes several interrelated pieces and/or phases. Some off these are as follows:

[0021] 1. Installation of the target program on a clean install system.

[0022] 2. Preprocessing of the target program after it has been installed.

[0023] 3. Installation of the software components that implement this invention onto the client machine.

[0024] 4. The selection and activation of a particular target program.

[0025] 5. Launch, initialization, and execution of the client-agent software component.

[0026] 6. Execution of the network file system driver.

[0027] Each of these steps or phases is now described in more detail, below.

[0028] The program to be remotely executed (hereafter referred to as the target program) is first installed on a clean system on which the contents of the registry is known, and all files prior to the installation of the target program are known. (The description is in the context of a Windows 95—type system.) By knowing the state of the machine before the installation of the target program, changes made to the install machine caused by the installation of the target program can be determined. It is sometimes required to run the target program on the install system so that any final installation steps needed by the target program can be determined. This need is determined manually on a case-by-case basis.

[0029] The purpose of the preprocessing phase is twofold: To create an index containing information about all the registry changes, files, and directories created by the installation of the target program, and to break-up the program and all its data into “chunks” which are then encrypted, compressed, and placed on the runtime server. The index is used by the client-side file system driver (hereafter the FSD) to present a virtual directory structure to the client machine, including all file properties (file size, last modification time, etc.) and the overall directory structure. The chunks are downloaded by the FSD and cached, decompressed, and decrypted as needed during the execution of the target program on the client machine. The files are preferably broken into compressed, encrypted chunks in a preprocessing phase, rather than during runtime, to improve runtime performance. The index also contains the information used by the FSD to compute which chunk to download at runtime as it is needed. As illustrated in figure !, one possible procedure that could be used to create the index and produce the chunks is as follows:

[0030] Decimation Procedure—Creates the index and produces the downloadable chunks.

[0031] Phase 1—Initialization

[0032] 1. Initialize the Index to be empty (step 102)

[0033] Phase 2—Record Registry Changes (steps 104, 106, 108 and 110)

[0034] 1. Assume that a database 101 has been previously created containing all the key/value pairs in the registry before the target program was installed. Note that performing steps 2 and 3 (suitably modified) with an initially empty database can create this database.

[0035] 2. (step 106) Enumerate all the key/value pairs in the registry. For each key/value pair that is not in the database of known key/value pairs, perform step 3. If there are no more key/value pairs, go to the next phase, Phase 3.

[0036] 3. (step 108) Add the key and value to the index. Replace any references to a drive letter with a special code that will replaced with the virtual drive letter created by the FSD at runtime.

[0037] Phase 3—Record File/Directory Changes

[0038] 1. Assume that a database 111 has been previously created containing all the files and directories, including their properties, that were on the install system before the target program was installed. Note that performing steps 2 and 3 (suitably modified) with an initially empty database can create this database.

[0039] 2. Traverse the directory structure of the drive on the install system (step 112). For each directory/file that is not in the database of known directories/files (step 114), perform step 3. If there are no more directories/files, go to the next phase, Phase 4.

[0040] 3. (step 116) Add the directory/file to the index. Include all the properties of the file/directory, including but not limited to: file creation time, last modification time, last access time, file size, short form of the name, and long form of the name. The location of the file/directory in the overall directory hierarchy is also recorded, using an appropriate tree data structure. Some additional fields which are later filled-in by the runtime system (described in that section) are zeroed out during preprocessing. An unique ID is also assigned to each file, and stored in the index, which will later be used by the FSD to calculate which chunk to download when the contents of that region of the file are needed by the runtime. If the file is not a directory, also perform step 4 below.

[0041] 4. For each program and data file that is not a directory (steps 118, 120, 124 and 126), break the file up into chunks and write the chunks to individual files on the runtime server. The initial size of each chunk (hereafter chunk-size) is a parameter to the Decimation Procedure, and can be tuned to increase performance; a typical value might be 32768 (i.e. 2**15). The actual decimation into chunks is accomplished (step 122) by opening the file, reading a block of chunk-size bytes, encrypting the block, compressing the block, calculating a unique file name for the block from the unique ID of the file and the block number, writing the just processed block to a new file in the specified directory of the runtime server (the destination directory on the runtime server is another parameter to the Decimation Procedure) using the file name previously generated, and then repeating for the next block in the input file until the end of file is reached.

[0042] Phase 4—Record Additional Index Information (step 132)

[0043] 1. In addition to the registry and file/directory changes recorded in the index by the previous phases, the index also contains the client version number, the target program version number, an expiration date for the index file, a program ID identifying the target program that is used when contacting the authentication server, the directory on the runtime server containing the chunks, and the virtual command line containing the program component that contains the image of the program that is to be used to create the runnable process on the client machine as well as any command line arguments to that program. This data is obtained as parameters to the Decimation Procedure, and are provided manually by the operator running the Decimation Procedure.

[0044] 2. Write the index data to an index file on the runtime server. (step 134)

[0045] After the completion of the Decimation Procedure, the runtime server contains an index file, and a set of preprocessed chunks.

[0046] Before a program can be executed on a client machine using the present invention, a set of software components must be installed on the client machine. The three major components are a file system driver (FSD), a network provider system library, and a client-side agent program (hereafter client-agent). These components are usually downloaded from a web site. Depending on the intended use of this invention, billing information may also be gathered from the user prior to the download of the client-side components. This billing information would be used if the present invention was to be used for, say, software rental.

[0047] The components are usually supplied with an installation program. The installation program copies the FSD and network provider library to the usual places for network file system drivers, and places the client-agent in a user-specified directory. The registry is updated to reflect the addition of a new network file system driver, providing the customary information usually given for network file system drivers.

[0048] An additional part of the installation involves registering a new file type with the Windows Explorer, also referred to as the Shell. The index file (described above) is given a unique file extension used exclusively by the client-agent. The Windows Explorer registry settings are updated to associate this unique file extension with the client-agent. Thus, whenever an index file with the given extension is opened, the client-agent is started and the index file supplied as a command line argument to the client-agent.

[0049] A final installation step is to associate a new MIME type, application/x-gamewave, with the index file and update the settings for the web browser so that when a web page contains a link to an index file, and that link is clicked by the user, the web browser downloads the index file to the client machine, starts the client-agent, and passes the index file location to the client-agent. The details of the changes required are different for different web browsers, but can be obtained from the respective browser publishers.

[0050] The client machine is typically rebooted after the installation of the client-side components.

[0051] After the preceding steps, the user may then use their web browser to navigate to a web site, and click on a link indicating a target program listed on a web page. The link points to the index file for that target program. The web browser initiates retrieval of the index file, and based on the MIME type for the index file, which was setup during installation, knows that the index file should be downloaded to the client machine and the client-agent started with the location of the index file given as an argument to the client-agent. The web browser downloads the index file, the web browser starts the client-agent, and the location of the index file that was downloaded is given to the client-agent.

[0052] Optionally, once an index file has been downloaded to the client machine, the user can bypass navigating to the web site and selecting the link for the target program, and can instead just directly open the index file already present on the client machine. Since the index file's file extension was registered with the Windows Explorer, opening the index file will cause the client-agent to be started with the location of the index file given as an argument to the client-agent.

[0053] When the client-agent is launched, as described in the previous section, it performs steps similar to the ones described here as illustrated in FIGS. 2A, 2B, 2C and 2D.

[0054] Client-Agent Procedure—Describes actions taken by client-agent.

[0055] Phase 1—FSD loading and primary client-agent checks

[0056] 1. Attempt to load the FSD (step 202). If it isn't already loaded and can't be loaded, signal an error condition (step 204 and 206) and stop further execution of the client-agent.

[0057] 2. (step 208) Contact the FSD, via a DeviceloControl 0 call, and determine the primary client-agent, as described below in the Client-Agent Registration Procedure. The primary client-agent is the first client-agent that begins execution and contacts the FSD. All subsequent client-agents are referred to as secondary client-agents. Because of how the client-agent is activated, it is possible for more than one client-agent to start execution. The primary client-agent is responsible for performing all client-agent functions, while the secondary client-agents will merely pass on requests to the primary client-agent and eventually exit.

[0058] 3. (step 210) Regardless of whether this is a primary or secondary client-agent, send a message to the primary client-agent giving the location of the index file. One possible way to do this would be to have the FSD return the window handle for the primary client-agent, and then use SendMessage ( ) to send an appropriate WM_USER-derived message with a parameter containing a string describing the index file location.

[0059] 4. If this is a secondary client-agent (step 212), exit at this point. Otherwise, continue to the next step, step 5.

[0060] 5. (step 214) Create a Win32 event that will later be used by the user-interface to signal that the client-agent Is to begin shutting down. The event is a manual reset event, and is initially in the non-signaled state. This event variable will hereafter be referred as the e_ShutdownClientAgent event.

[0061] 6. (step 216) Display a dialog, and retrieve the usemame and password from the user. The authentication server is contacted, and an initial verification is made that the usemame and password are valid.

[0062] 7. If a billing server is part of the intended use of the present invention, a notification is sent to the billing server at this point indicating that the specified user has started a session. (step 218)

[0063] Phase 2—Processing of incoming index file location notifications

[0064] 1. A new thread is created (step 220) and begins execution with the steps outlined in the next phase, Phase 3. The main thread continues to execute the following steps, remaining in Phase 2.

[0065] 2. (step 222) Wait for notification of an incoming index file location notification. When a new notification arrives, proceed with step 3, below.

[0066] 3. (step 224) Retrieve the index file location from the notification. Open the index file and perform a quick verification that the index file is indeed an index file. If the index file is invalid (step 226), display a message (step 228) and go back to step 2, above.

[0067] 4. Contact the authentication server, (step 230) giving the program ID specified in the index file and the usemame and password entered in step 6 of Phase 1. If access is denied (step 232), display a message (step 228) and go back to step 2, above.

[0068] 5. If access is granted (step 232), the authentication server will respond with a session key for this execution of the target program, and the remaining time left in which to allow execution of the target program. The reply may also optionally contain an Internet address of a machine to use as the runtime server and the directory path to use on that runtime server. (step 240)

[0069] 6. The index file location and the information returned by the authentication server are then given to the FSD using a DeviceloControl ( ) call. The FSD adds the index file information and the information returned by the authentication server to it's internal tables (which is described in greater detail below in the Index File Insertion Procedure). The FSD will return a string containing a command line to be executed by a call to CreateProcess 0 that will actually start the execution of the target program. The drive letter to use to reference the virtual drive, as well as a pointer to the program descriptor block (also described in the Index File Insertion Procedure) is also returned. (step 242)

[0070] 7. Merge any registry changes specified in the index file into the registry, replacing any references to drive letters with the drive letter returned In step 6, above. (step 244)

[0071] 8. Using the command line returned in step 6, create a new process using the CreateProcess ( ) system call. In the call to CreateProcess ( ), indicate that the process is to initially be created In a suspended state. Retrieve the process ID for the newly created process. (step 246)

[0072] 9. Call the FSD via DeviceloControl ( ) again, this time indicating that the newly created process is to be given access to the target program specified by the index file used in step 6 above. The parameters to the FSD are the process ID of the new process, and the pointer to the program descriptor block returned in step 6 above. This is covered in more detail in the Add Process Access Procedure below. The FSD will add the process to its list of processes that can access the files referenced by the index file inserted in step 6. (step 248)

[0073] 10. Create a deathwatch thread that will notify the client-agent when the newly created process exits. (step 250) The newly created thread starts executing the instructions described in phase 4 below. (step 252) The current thread continues execution with the following step, step 10.

[0074] 11. Allow the newly created process to resume execution using the ResumeThread ( ) system call. (step 254) At this point the operating system begins to execute the newly created process. All file operations are handled by the FSD, which downloads, caches, decompresses, and decrypts the pieces of the program as needed. (This is described in more detail in the FSD section). The FSD will only allow the newly created process to “see”the files referenced by the index file; all other processes will see either nothing or will receive access violation errors, depending on the type of access attempted.

[0075] 12. If a billing server is part of the intended use of the present invention, a notification message is sent at this point to the billing server indicating that the user has started execution of the designated target program, as specified in the index file. (step 254)

[0076] 13. Go back to step 2 and wait for another index file location notification.

[0077] Phase 3—Handle display updates for client-agent.

[0078] 1. While the main thread waits for index file location notification messages (as described in the previous phase), the new thread created to handle phase 3 processing is responsible for performing display updates and user-interface functionality of the client-agent. (step 233)

[0079] 2. Display the dialog used as the user-interface for the client-agent. Handle user-interface events by updating the user-interface. (step 234)

[0080] 3. If the user indicates that the client-agent is to be terminated, set the e_ShutdownClientAgent event, and wait for all deathwatch threads (described above and in phase 4 below) to terminate. (step 236)

[0081] 4. Since other user-interface events of the client-agent are not relevant to the description of this invention, no further explanation of them will be provided here. (step 238)

[0082] Phase 4—Process deathwatch threads

[0083] 1. A deathwatch thread is started for each new process created. The function of the deathwatch thread is to detect when the newly created process exits, so that the primary client-agent can perform any dean up associated with the process exiting. (step 254)

[0084] 2. The death-watch thread waits for one of three things: the demise of the watched process, a timeout when the time allowed for the process to access the program expires (this time value was retrieved in phase 2, step 5, above, from the authentication server), or the e_ShutdownClientAgent event becomes signaled. Based on the reason why the death-watch thread was awakened, perform the appropriate step below: (step 256)

[0085] 3. If the thread was awakened by the demise of the watched process, notify the FSD that the process Is terminating and should be removed from the program descriptor blocks process access list (described in the FSD section below). Undo any registry changes specified in the index file. If a billing server is part of the intended use of the present invention, send a notification message to the billing server indicating that the user has ended use of the designated target program, as specified in the index file. Perform any additional clean up and notifications required by the user-interface, then let the deathwatch thread terminate. (step 262)

[0086] 4. If the thread was awakened by the timeout of the time allowed for the process to access the target program, terminate the process (step 260) and then perform the processing described in step 3, above. (step 262) As an option, the client-agent could prompt the user if additional time is desired, and then takes steps to reset the timer. If the authentication server granted additional time, go back to step 2.

[0087] 5. If the signaling of the e_ShutdownClientAgent event awakened the thread, terminate the process (step 260) and then perform the additional actions specified in step 3, above. (step 262)

[0088] Since the file system driver (FSD) described by this invention is in essence an enhanced network file system driver, it will be assumed that the reader is familiar with the implementation of network file system drivers, and only the differences and enhancements to a baseline network file system driver will be described. For more information on implementing network file system drivers, see the Microsoft DDK documentation, Walter Oney's book, “Systems Programming for Windows 95”, Microsoft Press, 1996, or Stan Mitchell's book, “Inside the Windows 95 File System”, O'Reilly & Associates, Inc., 1997.

[0089] The primary difference between a baseline network file system driver and the present invention is in controlling access to the remote components, and in maintaining a persistent but secure cache. The FSD creates a virtual drive, complete with drive letter, and places each target program that is to be accessed under a top-level directory on that virtual drive.

[0090] Descriptions of each of these major FSD functionality groups follow.

[0091] As was described in a previous section on the launch, initialization, and execution of the client-agent, the client-agent attempts to register with the FSD and to determine if it is the primary client-agent One possible procedure on how this could be done is given here and illustrated in FIG. 3.

[0092] Client-Agent Registration with FSD Procedure—Registers Client-Agent with FSD.

[0093] 1. Assume that the FSD maintains a global variable, hereafter referred to as g_PrimaryClientID, that initially is zero. (step 302) After the primary client-agent has been successfully registered, this variable will hold the process ID of the primary clientagent. agent Also assume that another global variable, g_PrimaryClientHwnd, is also initially zero. After the primary client-agent has registered, this variable will hold the HWND of the primary client-agent main window. The use of the HWND is described in the procedure for launching, initializing, and executing the client-agent.

[0094] 2. (step 304) After the client-agent has started execution and opened the FSD, it uses DeviceloControl ( ) to call the FSD, passing in a code indicating this is a request to register the client-agent, the client-agent's process ID, and the HWND of the client-agent's main window.

[0095] 3. If g_PrimaryClientID is zero, (step 306) set it to be the value of the process ID passed in by the calling client-agent. (step 308) If g_PrimaryClientHwnd is zero (step 310), set it to be the value of the HWND passed in by the calling client-agent. (step 312)

[0096] 4. If the value of g_PrimaryClientID is the same as the value of the process ID passed in by the calling client-agent (step 314), return a result code indicating that the calling client-agent is the primary client-agent. (step 316) Otherwise, return a result code indicating that the calling client-agent is a secondary client-agent, and also return the value of g_PrimaryClientHwnd as the HWND of the primary client-agent. (step 318)

[0097] As was described in the section on the launch, initialization, and execution of the client-agent, whenever a new index file location notification is received by the client-agent, it attempts to insert the specified index file into the FSD so that the target program specified by the index file may be accessed. The FSD maintains a list of all the index files thus inserted. The elements of this list are called program descriptor blocks, and contain the following fields: forward and backward links to the other program descriptor blocks in the list, a pointer to the file and directory information tree containing all the data read from the index file, the top level directory name this program descriptor block will correspond to on the virtual drive created by the FSD, and a map of valid process ID's that have access to files in this program descriptor block.

[0098] A procedure for inserting the index file information into the FSD could be described as follows and as illustrated in FIG. 4:

[0099] Index File Insertion Procedure—Adds the data contained In an index file to the FSD.

[0100] 1. Read and verify the data in the index file to be inserted. (step 402)

[0101] 2. If a program descriptor block already exists for the given index file (step 404), return the command line (as specified in the index file), the drive letter of the virtual drive, and a pointer to the already existing program descriptor block (step 406).

[0102] 3. If a drive letter for the virtual drive has not already been created (step 409), create one now, and map in the new drive letter. (step 410)

[0103] 4. (step 412) Create a new program descriptor block using the data read from the index file, and insert it Into the global list of program descriptor blocks. The map of valid process ID's for this program descriptor block contains only the process ID of the primary client-agent at this point No other process can see the files or directories hiding beneath this program descriptor block. Adding a new program descriptor block causes a new top-level directory to appear on the virtual drive. Underneath this top-level directory is a virtual directory structure containing all the files and directories that were decimated in the pre-processing phase. The data in the index file is used to produce the virtual directory structure.

[0104] 5. Return the command line, virtual drive letter, and pointer to the program descriptor block. (step 414)

[0105] Access to the virtual directory structure underneath a top-level directory on the virtual drive is granted based on process ID's. In order for a process to be created that can access the files in the virtual directory, the process ID for that process must be added to the process access list of the program descriptor block controlling access to the virtual directory structure. Adding a process to a program descriptor block's process access list could be performed with a procedure similar to this and as illustrated in FIG. 5:

[0106] Add Process Access Procedure—Grants access to the given process to the target program.

[0107] 1. As was described in the section above on launching, initializing, and executing the client-agent, this procedure is called with a process ID of a suspended process, and a pointer to a program descriptor block returned from the Index File Insertion Procedure, above. (step 502)

[0108] 2. Add the process ID to the given program descriptor block's process access list. The specified process can now see the virtual directory structure beneath the top-level directory corresponding to the given program descriptor block on the virtual drive. (step 504)

[0109] The FSD maintains a two level cache. The first level is an in-memory cache consisting of blocks each containing 4K of data, and is implemented using the VCACHE block cache module provided by the operating system. The second level is an on-disk cache implemented as described here.

[0110] All the data for the on-disk cache is maintained in a single file, the maximum size of which can be adjusted by the user using the user-interface of the client-agent. The on-disk cache file contains the following items: a cache file version number, a unique cache file signature (derived from the creation time of the cache file), a counter specifying the next cache block handle (described later) to allocate, the current number of on-disk cache blocks in use, the maximum number of on-disk cache blocks to allow, an array (with enough entries to hold the maximum number of allowable on-disk cache blocks) of block descriptors (described later) in LRU order, followed by the actual cache blocks themselves. The size of each cache block is fixed, and is larger enough to contain an encrypted version of a file chunk as produced by the Decimation Procedure, described earlier. A block descriptor consists of a cache block handle, a block number, and the actual size of the data within the indicated block. The first two items in the block descriptor are sufficient to uniquely address any block in any file referenced by the index file, as will be explained subsequently.

[0111] When the Decimation Procedure creates an index file (as described earlier), one of the fields in the index file is reserved for a cache file signature. The value of the cache file signature as initially produced by the Decimation Procedure is zero. Additionally, each file entry contained within the index file has a field reserved for a cache handle, which is also initialized to zero. When the index file is downloaded from the server to the client machine by the web browser, all the mentioned values are still zero. When an index file is inserted into the FSD (as described below), the index file is bound to the cache file by a procedure similar to the following:

[0112] Index File to Cache File Binding Procedure—Connects Index file to the cache file.

[0113] 1. If the cache file doesn't exist (step 602), create one (step 604). The number of blocks currently in use will be zero, and the cache file signature will be set based on the creation date of the cache file.

[0114] 2. Read the index file (step 606) and look at its recorded cache file signature (step 608). If it doesn't match the signature of the actual cache file, reset the index file's cache file signature to zero (step 610), and set the cache handle of every file entry contained within the Index file to zero. What this situation indicates is that an existing index file is referencing an older cache file, and so the cache info contained within the Index file is no longer valid, and must be reset.

[0115] 3. (step 612) If the cache file signature in the index file is zero (either because it was zero initially or because it was reset in step 2 above), set it to the signature of the actual cache file (step 614).

[0116] Binding the index file to the cache file serves to keep the index file and cache file in agreement with respect to the identity of the cache handles. Note that the user can choose to delete the cache file if so desired; any existing index files will then re-bind to a new cache file the next time the index file is used to start the target program.

[0117] Once a binding has been established between an index file and the cache file, files specified by the index file can be opened (assuming access has been granted by the FSD). The cache processing required when a file is opened can be described by a procedure similar to the following and as illustrated in FIG. 7:

[0118] Set Cache Info during File Open Procedure—Handles cache when a virtual file Is opened.

[0119] 1. Assume that the file entry within the index file for the file that is to be opened has already been located by the FSD's Open File Procedure. (step 702)

[0120] 2. If the cache handle in the located file entry is zero (step 704), allocate a new cache handle from the cache file (step 706). This is done by incrementing the cache file's next available cache handle counter, and using the new value obtained. The cache handle is also written back to the index file on disk, so that it can be used again the next time that index file is used, and the incremented next available cache handle counter is written back out to the cache file. The cache handle uniquely identifies one file. The correspondence between the cache handle and the file within the index file is maintained by the index file, not by the cache file. The cache file just records the next available cache handle to use.

[0121] After a cache handle has been associated with a particular file in the index file, the FSD can use the cache handle to retrieve blocks from the cache. The FSD specifies which block to obtain by giving the cache handle of the file, along with the relative block number of which block to retrieve, to a procedure similar to the following as illustrated in FIGS. 8A, 8B and 8C.

[0122] Retrieve Cache Block Procedure—Given a cache handle and a block number, get the block.

[0123] 1. Using the given cache handle and block number, create a key by multiplying the cache handle by 32768, and then adding in the block number. (step 802)

[0124] 2. Using the key generated above, use the VCACHE block cache to determine if a block with the given key exists in the in-memory block cache. (step 804)

[0125] 3. If a block with the given key exists (step 806), retrieve the block from VCACHE and compare the actual cache handle and block number associated with the retrieved block to the cache handle and block number passed in to this procedure (step 808). If the cache handle and block number match the requested block (step 810), then the block has been found in the in-memory cache. In this case, update the VCACHE MRU info for this block (step 812), and then return the found block. If a block with the given key doesn't exist, or if the cache handle and block number do not match, continue with step 4.

[0126] 4. (step 814) The requested block Is not in the In-memory cache. Use VCACHE to locate the block in the in-memory cache that has been used least recently, and retrieve this block. Remove the block from the MRU list of VCACHE. Set the cache handle and block number of the block to the requested cache handle and block number. An attempt will now be made to retrieve the data for the block from the on-disk cache.

[0127] 5. The on-disk cache maintains a map from key values to cache block numbers. A cache block number is just an index into the cache blocks stored in the cache file. Using this map, see if the key calculated in step 1 is already associated with a cache block number (step 818). If it is, compare the cache handle and block number associated with the found cache block number to the requested cache handle and block number (step 820)0. If they match (step 822), we have found the requested block in the on-disk cache. In this case, read the cache block given by the found cache block number into a temporary buffer, and then decrypt the temporary buffer into the buffer allocated in step 4 (step 824). Blocks are always stored in the cache file in encrypted form. Update the on-disk cache's MRU info for the found cache block number. Add the newly acquired block to the MRU of VCACHE, and then return the found block with its data (step 826). If a cache block number was not in the map for this key, or if the cache handle and block number of the found cache block number did not match the requested cache handle and block number, proceed to step 6.

[0128] 6. The requested block was not in the on-disk cache as well (step 822). Get the cache block numbers of the 8 least recently used on-disk cache blocks, and remove them from the on-disk MRU list and from the key map. The reason that 8 on-disk cache blocks are allocated is because the block size used by VCACHE and the on-disk cache is 4K, while the chunk files downloaded from the runtime server are 32K in length. Thus, each chunk file downloaded from the runtime server is used to fill 8 on-disk cache buffers. The on-disk cache block numbers just allocated will be used to contain the data downloaded from the runtime server. (step 828)

[0129] 7. Using the key calculated in step 1, along with the session keys retrieved from the authentication server when the client-agent started processing the index file being used to access this target program, and the runtime server source directory also stored with the index file, contact the runtime server and request the file containing the chunk with the data for the sought-after block. The actual communication protocol used to retrieve the block can be FTP, HTTP, or some other suitable protocol. If the chunk cannot be found, or some other communication error occurs, return a failure code and exit this procedure. (steps 832 and 834) Otherwise, after the data from the chunk file has been downloaded into a temporary buffer, decompress the temporary buffer into the 8 buffers allocated from the on-disk cache in step 6. Write the buffers back out to the cache file, and update the MRU and key map of the on-disk cache. Decrypt the data from the one on-disk cache buffer corresponding to the requested block and place it in the in-memory buffer allocated in step 4. Update the VCACHE MRU and map for the in-memory buffer, and then return the found block. (step 826)

[0130] Thus, in general, the search for a request block begins with the in-memory cache. If it isn't in the in-memory cache, the on-disk cache is searched. If it isn't in the on-disk cache, the data is downloaded from the runtime server.

[0131] After the target program begins execution, it can begin to access other files on the virtual drive using the volume and file functions provided by the FSD. For the most part, these volume and file functions are similar to those found in any basic network file system. One difference is in how the data for the files is retrieved, which has been explained briefly in the section describing the cache architecture. The other difference is in what any given process can see when it looks at the virtual drive. The general procedure for whether access is granted or not depends on the type of access being requested (i.e. which volume or file function is being executed), the process ID of the process attempting access, and optionally the path specified in the request for those requests containing paths. The general procedure for determining access can be described as in the following procedure (and is illustrated in FIG. 9). Some exceptions to this general procedure will be noted in later sections.

[0132] General Access Control Procedure—Performs general process access checks.

[0133] 1. Since the virtual drive is implemented as a read-only device (an extension to this embodiment includes a write capability), any type of request that attempts to modify the content of the virtual drive is denied access. Examples of these types of requests would be: deleting a file on the virtual drive, renaming a file, opening a file as write-only, attempting to set file attributes, etc. (step 902)

[0134] 2. For requests that contain a path, and that path has one or more directory components, the first directory component in the path is used to locate the program descriptor block for that first level directory. (steps 904, 906 and 908) If the current process ID associated with the request is not in the process access list in the specified program descriptor block (step 910), access is denied (step 512). If no program descriptor block matches the first directory component (step 914), access is denied (step 912). Otherwise, in general, access is allowed (step 916).

[0135] 3. For requests that contain a path, but the path is referring to the root of the virtual drive, access will be granted to each top level directory in which the corresponding program descriptor block contains the current process ID in it's process access list. An example of this case would be a FS_FindFile request specifying the root of the virtual drive and using a wildcard search. In this case, all top-level directories that allow access to the current process would become visible to the search. (steps 918, 920, 922, 924 and 926)

[0136] 4. Requests that do not contain a path are handled on a case-by-case manner. (step 928)

[0137] Some exceptions to this general procedure will now be described on a case-by-case manner, based on the request type:

[0138] FS_Dir—Depending on the sub-function requested, the request will be denied if it involves modifying the contents of the virtual drive, and will follow the general procedure above if the request only requires read-only access.

[0139] FS_FlushVolume—Allowed for all processes, but ignored.

[0140] FS_GetDisklnfo—Allowed for all processes.

[0141] FS_QueryResourcelnfo—Allowed for all processes.

[0142] FS_loctl16Drive—Allowed for all processes but only for subfunction 0: Get Device Data

[0143] When each client-agent begins execution, it opens a handle to the FSD. The FSD is notified whenever a new process opens a handle to it. Likewise, whenever any process with an open handle to the FSD exits, the handle is closed and the FSD is notified of the closure. The FSD uses these notification events to track when the primary client-agent exits. The processing done at the time the primary client-agent exits is similar to the following as illustrated in FIGS. 10A and 10B:

[0144] Primary Client-Agent Deregistration Procedure—Deregisters Primary Client-Agent with FSD.

[0145] 1. Whenever an open handle to the FSD is dosed, the FSD is notified of the closure via it's OnW32Deviceiocontrol ( ) entry point (step 1002). When such an event is received, the process ID of the process dosing its handle to the FSD is compared to g_PrimaryClientID (step 1004).

[0146] 2. If the two process ID's match, the global list of program descriptor blocks is traversed, and all the process ID's in the process access list (except for the process ID of the primary client-agent) for each program descriptor block are used to terminate all processes thus referenced. The rationale for terminating all processes currently using the FSD is to disallow target programs from running if the primary client-agent is no longer running. If a billing server is being used as part of the intended use of this invention, a notification message is sent to the billing server for each terminated process indicating that the user has ended use of the specified target program. (steps 1006, 1008, 1010, 1012, 1014 and 1016)

[0147] 3. When each process is terminated, it's process ID is removed from the program descriptor block's process access list, and when the list only contains the primary client-agent process ID, the entire program descriptor block is removed and deallocated. (step 1018)

[0148] 4. The g_PrimaryClientID and g_PrimaryClientHwnd variables are then reset to zero. The drive letter for the virtual drive is unmapped (step 1020). At this point, all processes formerly referencing target programs have all been terminated, and all program descriptor blocks have been removed, thus bringing the FSD back to it's initial empty state.

[0149] 5. If a billing server is being used as part of the intended use of this invention, a notification message is sent to the billing server indicating that the user has completed the session. (step 1022)

[0150] A complete copy of provisional application No. 601049,759, filed Jun. 16, 1997, is attached herein as Appendix A and is to be considered an integral part of this specification.

[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]



[Inline image omitted. See PDF.]