This invention relates to computer security, and more particularly to secure and/or protected computer execution environments. Still more specifically, the present invention relates to computer security techniques based at least in part on cryptography, that protect a computer processing environment against potentially harmful computer executables, programs and/or data; and to techniques for certifying load modules such as executable computer programs or fragments thereof as being authorized for use by a protected or secure processing environment.
Computers have become increasingly central to business, finance and other important aspects of our lives. It is now more important than ever to protect computers from “bad” or harmful computer programs. Unfortunately, since many of our most critical business, financial and governmental tasks now rely heavily on computers, dishonest people have a great incentive to use increasingly sophisticated and ingenious computer attacks.
Imagine, for example, if a dishonest customer of a major bank could reprogram the bank's computer so it adds to instead of subtracts from the customer's account—or diverts a penny to the customer's account from anyone else's bank deposit in excess of $10,000. If successful, such attacks would not only allow dishonest people to steal, but could also undermine society's confidence in the integrity and reliability of the banking system.
Terrorists can also try to attack us through our computers. We cannot afford to have harmful computer programs destroy the computers driving the greater San Francisco metropolitan air traffic controller network, the New York Stock Exchange, the life support systems of a major hospital, or the Northern Virginia metropolitan area fire and paramedic emergency dispatch service.
There are many different kinds of “bad” computer programs, which in general are termed “Trojan horses”—programs that cause a computer to act in a manner not intended by its operator, named after the famous wooden horse of Troy that delivered an attacking army disguised as an attractive gift. One of the most notorious kinds is so-called “computer viruses”—“diseases” that a computer can “catch” from another computer. A computer virus is a computer program that instructs the computer to do harmful or spurious things instead of useful things—and can also replicate itself to spread from one computer to another. Since the computer does whatever its instructions tell it to do, it will carry out the bad intent of a malicious human programmer who wrote the computer virus program—unless the computer is protected from the computer virus program. Special “anti-virus” protection software exists, but it unfortunately is only partially effective—for example, because new viruses can escape detection until they become widely known and recognized, and because sophisticated viruses can escape detection by masquerading as tasks the computer is supposed to be performing.
Computer security risks of all sorts—including the risks from computer viruses—have increased dramatically as computers have become increasingly connected to one another over the Internet and by other means. Increased computer connectivity provides increased capabilities, but also creates a host of computer security problems that haven't been fully solved. For example, electronic networks are an obvious path for spreading computer viruses. In October 1988, a university student used the Internet (a network of computer networks connected to millions of computers worldwide) to infect thousands of university and business computers with a self-replicating “worm” virus that took over the infected computers and caused them to execute the computer virus instead of performing the tasks they were supposed to perform. This computer virus outbreak (which resulted in a criminal prosecution) caused widespread panic throughout the electronic community.
Computer viruses are by no means the only computer security risk made even more significant by increased computer connectivity. For example, a significant percentage of the online electronic community has recently become committed to a new “portable” computer language called Java™ developed by Sun Microsystems of Mountain View, Calif. Java was designed to allow computers to interactively and dynamically download computer program code fragments (called “applets”) over an electronic network such as the internet, and execute the downloaded code fragments locally. Java's “download and execute” capability is valuable because it allows certain tasks to be performed locally on local equipment using local resources. For example, a user's computer could run a particularly computationally or data-intensive routine—relieving the provider's computer from having to run the task and/or eliminating the need to transmit large amounts of data over the communications path.
While Java's “download and execute” capability has great potential, it raises significant computer security concerns. For example, Java applets could be written to damage hardware, software or information on the recipient computer, make the computer unstable by depleting its resources, and/or access confidential information on the computer and send it to someone else without first getting the computer owner's permission. People have expended lots of time and effort trying to solve Java's security problems. To alleviate some of these concerns, Sun Microsystems has developed a Java interpreter providing certain built-in security features such as:
Numerous security flaws have been found despite these techniques. Moreover, a philosophy underlying this overall security design is that a user will have no incentive to compromise the security of her own locally installed Java interpreter—and that any such compromise is inconsequential from a system security standpoint because only the user's own computer (and its contents) are at risk. This philosophy—which is typical of many security system designs—is seriously flawed in many useful electronic commerce contexts for reasons described below in connection with the above-referenced Ginter et al. patent specification.
The Ginter et al. specification describes a “virtual distribution environment” comprehensively providing overall systems and wide arrays of methods, techniques, structures and arrangements that enable secure, efficient electronic commerce and rights management, including on the Internet or other “Information Super Highway.”
The Ginter et al. patent disclosure describes, among other things, techniques for providing a secure, tamper resistant execution spaces within a “protected processing environment” for computer programs and data. The protected processing environment described in Ginter et al. may be hardware-based, software-based, or a hybrid. It can execute computer code the Ginter et al. disclosure refers to as “load modules.” See, for example, Ginter et al. FIG. 23 and corresponding text. These load modules—which can be transmitted from remote locations within secure cryptographic wrappers or “containers”—are used to perform the basic operations of the “virtual distribution environment.” Load modules may contain algorithms, data, cryptographic keys, shared secrets, and/or other information that permits a load module to interact with other system components (e.g., other load modules and/or computer programs operating in the same or different protected processing environment). For a load module to operate and interact as intended, it must execute without unauthorized modification and its contents may need to be protected from disclosure.
Unlike many other computer security scenarios, there may be a significant incentive for an owner of a Ginter et al. type protected processing environment to attack his or her own protected processing environment. For example:
Security experts can often be heard to say that to competently do their job, they must “think like an attacker.” For example, a successful home security system installer must try to put herself in the place of a burglar trying to break in. Only by anticipating how a burglar might try to break into a house can the installer successfully defend the house against burglary. Similarly, computer security experts must try to anticipate the sorts of attacks that might be brought against a presumably secure computer system.
From this “think like an attacker” viewpoint, introducing a bogus load module is one of the strongest possible forms of attack (by a protected processing environment user or anyone else) on the virtual distribution environment disclosed in the Ginter et al. patent specification. Because load modules have access to internal protected data structures within protected processing environments and also (at least to an extent) control the results brought about by those protected processing environments, bogus load modules can (putting aside for the moment additional possible local protections such as addressing and/or ring protection and also putting aside system level fraud and other security related checks) perform almost any action possible in the virtual distribution environment without being subject to intended electronic controls. Especially likely attacks may range from straightforward changes to protected data (for example, adding budget, billing for nothing instead of the desired amount, etc.) to wholesale compromise (for example, using a load module to expose a protected processing environment's cryptographic keys). For at least these reasons, the methods for validating the origin and soundness of a load module are critically important.
The Ginter et al. patent specification discloses important techniques for securing protected processing environments against inauthentic load modules introduced by the computer owner, user, or any other party, including for example:
Although the Ginter et al. patent specification comprehensively solves a host of load module (and other) security related problems, any computer system—no matter how secure—can be “cracked” if enough time, money and effort is devoted to the project. Therefore, even a very secure system such as that disclosed in Ginter et al. can be improved to provide even greater security and protection against attack.
The present invention provides improved techniques for protecting secure computation and/or execution spaces (as one important but non-limiting example, the protected processing environments as disclosed in Ginter et al) from unauthorized (and potentially harmful) load modules or other “executables” or associated data. In one particular preferred embodiment, these techniques build upon, enhance and/or extend in certain respects, the load module security techniques, arrangements and systems provided in the Ginter et al. specification.
In accordance with one aspect provided by the present invention, one or more trusted verifying authorities validate load modules or other executables by analyzing and/or testing them. A verifying authority digitally “signs” and “certifies” those load modules or other executables it has verified (using a public key based digital signature and/or certificate based thereon, for example).
Protected execution spaces such as protected processing environments can be programmed or otherwise conditioned to accept only those load modules or other executables bearing a digital signature/certificate of an accredited (or particular) verifying authority. Tamper resistant barriers may be used to protect this programming or other conditioning. The assurance levels described below are a measure or assessment of the effectiveness with which this programming or other conditioning is protected.
A web of trust may stand behind a verifying authority. For example, a verifying authority may be an independent organization that can be trusted by all electronic value chain participants not to collaborate with any particular participant to the disadvantage of other participants. A given load module or other executable may be independently certified by any number of authorized verifying authority participants. If a load module or other executable is signed, for example, by five different verifying authority participants, a user will have (potentially) a higher likelihood of finding one that they trust. General commercial users may insist on several different certifiers, and government users, large corporations, and international trading partners may each have their own unique “web of trust” requirements. This “web of trust” prevents value chain participants from conspiring to defraud other value chain participants.
In accordance with another aspect provided by this invention, each load module or other executable has specifications associated with it describing the executable, its operations, content, and functions. Such specifications could be represented by any combination of specifications, formal mathematical descriptions that can be verified in an automated or other well-defined manner, or any other forms of description that can be processed, verified, and/or tested in an automated or other well-defined manner. The load module or other executable is preferably constructed using a programming language (e.g., languages such as Java and Python) and/or design/implementation methodology (e.g., Gypsy, FDM) that can facilitate automated analysis, validation, verification, inspection, and/or testing.
A verifying authority analyzes, validates, verifies, inspects, and/or tests the load module or other executable, and compares its results with the specifications associated with the load module or other executable. A verifying authority may digitally sign or certify only those load modules or other executables having proper specifications—and may include the specifications as part of the material being signed or certified.
A verifying authority may instead, or in addition, selectively be given the responsibility for analyzing the load module and generating a specification for it. Such a specification could be reviewed by the load module's originator and/or any potential users of the load module.
A verifying authority may selectively be given the authority to generate an additional specification for the load module, for example by translating a formal mathematical specification to other kinds of specifications. This authority could be granted, for example, by a load module originator wishing to have a more accessible, but verified (certified), description of the load module for purposes of informing other potential users of the load module.
Additionally, a verifying authority may selectively be empowered to modify the specifications to make it accurate—but may refuse to sign or certify load modules or other executables that are harmful or dangerous irrespective of the accuracy of their associated specifications. The specifications may in some instances be viewable by ultimate users or other value chain participants providing a high degree of assurance that load modules or other executables are not subverting the system and/or the legitimate interest of any participant in an electronic value chain the system supports.
In accordance with another aspect provided by the present invention, an execution environment protects itself by deciding—based on digital signatures, for example—which load modules or other executables it is willing to execute. A digital signature allows the execution environment to test both the authenticity and the integrity of the load module or other executables, as well permitting a user of such executables to determine their correctness with respect to their associated specifications or other description of their behavior, if such descriptions are included in the verification process.
A hierarchy of assurance levels may be provided for different protected processing environment security levels. Load modules or other executables can be provided with digital signatures associated with particular assurance levels. Appliances assigned to particular assurance levels can protect themselves from executing load modules or other executables associated with different assurance levels. Different digital signatures and/or certificates may be used to distinguish between load modules or other executables intended for different assurance levels. This strict assurance level hierarchy provides a framework to help ensure that a more trusted environment can protect itself from load modules or other executables exposed to environments with different work factors (e.g., less trusted or tamper resistant environments). This can be used to provide a high degree of security compartmentalization that helps protect the remainder of the system should parts of the system become compromised.
For example, protected processing environments or other secure execution spaces that are more impervious to tampering (such as those providing a higher degree of physical security) may use an assurance level that isolates it from protected processing environments or other secure execution spaces that are relatively more susceptible to tampering (such as those constructed solely by software executing on a general purpose digital computer in a non-secure location).
A verifying authority may digitally sign load modules or other executables with a digital signature that indicates or implies assurance level. A verifying authority can use digital signature techniques to distinguish between assurance levels. As one example, each different digital signature may be encrypted using a different verification key and/or fundamentally different encryption, one-way hash and/or other techniques. A protected processing environment or other secure execution space protects itself by executing only those load modules or other executables that have been digitally signed for its corresponding assurance level.
The present invention may use a verifying authority and the digital signatures it provides to compartmentalize the different electronic appliances depending on their level of security (e.g., work factor or relative tamper resistance). In particular, a verifying authority and the digital signatures it provides isolate appliances with significantly different work factors—preventing the security of high work factor appliances from collapsing into the security of low work factor appliances due to free exchange of load modules or other executables.
Encryption can be used in combination with the assurance level scheme discussed above to ensure that load modules or other executables can be executed only in specific environments or types of environments. The secure way to ensure that a load module or other executable can't execute in a particular environment is to ensure that the environment doesn't have the key(s) necessary to decrypt it. Encryption can rely on multiple public keys and/or algorithms to transport basic key(s). Such encryption protects the load module or other executable from disclosure to environments (or assurance levels of environments) other than the one it is intended to execute in.
In accordance with another aspect provided by this invention, a verifying authority can digitally sign a load module or other executable with several different digital signatures and/or signature schemes. A protected processing environment or other secure execution space may require a load module or other executable to present multiple digital signatures before accepting it. An attacker would have to “break” each (all) of the several digital signatures and/or signature schemes to create an unauthorized load module or other executable that would be accepted by the protected processing environment or other secure execution space. Different protected processing environments (secure execution spaces) might examine different subsets of the multiple digital signatures—so that compromising one protected processing environment (secure execution space) will not compromise all of them. As an optimization, a protected processing environment or other secure execution space might verify only one of the several digital signatures (for example, chosen at random each time an executable is used)—thereby speeding up the digital signature verification while still maintaining a high degree of security.
These and other features and advantages provided in accordance with this invention may be better and more completely understood by referring to the following detailed description of example preferred embodiments in conjunction with the drawings, of which:
Protected processing environments 108 can use this digital “seal of approval” 106 (which may comprise one or more “digital signatures”) to distinguish between authorized and unauthorized load modules 54.
Verifying authority 100 uses an analyzing tool(s) 112 to analyze and test load module 54 and determine whether it performs as specified by its associated specifications 110—that is, whether the specifications are both accurate and complete.
Verifying authority 100 is preferably a trusted, independent third party such as an impartial, well respected independent testing laboratory. Therefore, all participants in an electronic transaction involving load module 54 can trust a verifying authority 100 as performing its testing and analysis functions competently and completely objectively and impartially. As described above, there may be several different verifying authorities 100 that together provide a “web of trust”. Several different verifying authorities may each verify and digitally sign the same load module—increasing the likelihood that a particular value chain participant will trust one of them and decreasing the likelihood of collusion or fraud Electronic value chain participants may rely upon different verifying authorities 100 to certify different types of load modules. For example, one verifying authority 100 trusted by and known to financial participants might verify load modules relating to financial aspects of a transaction (e.g., billing), whereas another verifying authority 100′ trusted by and known to participants involved in using the “information exhaust” provided by an electronic transaction might be used to verify load modules relating to usage metering aspects of the same transaction.
Once verifying authority 100 is satisfied with load module 54, it affixes its digital “seal of approval” 106 to the load module.
In the
Message digest 116 may then be encrypted using asymmetric key cryptography.
In this case the first key is owned by verifying authority 100 and is kept highly secure (for example, using standard physical and procedural measures typically employed to keep an important private key secret while preventing it from being lost). Once message digest 116 is locked into strong box 118 using the first key 122 the strong box can be opened only by using the corresponding second key 124. Note that other items (e.g., further identification information, a time/date stamp, etc.) can also be placed within strong box 106.
Maintaining “public” verification key 124 as a secret within tamper resistant protected processing environment 108 greatly complicates the job of generating bogus digital signatures 106. If the attacker does not possess second key 124, the difficulty of an algorithmic attack or cryptanalytic attack on the verification digital signature algorithm is significantly increased, and the attacker might be reduced to exhaustive search (brute force) type attacks which would be even less practical because the search trials would require attempting to present a bogus load module 54 to protected processing environment 108—which, after a few such attempts is likely to refuse all further attempts. Keeping second key 124 secret also requires a multi-disciplinary attack: an attacker must both (A) extract the secret from protected processing environment 108, and (B) attack the algorithm. It may be substantially less likely that a single attacker may have expertise in each of these two specialized disciplines.
In addition, maintaining the “public” key within a tamper-resistant environment forecloses the significant threat that the owner of protected processing environment 108 may himself attack the environment. For example, if the owner could replace the appropriate “public” key 124 with his own substitute public key, the owner could force the protected processing environment 108 to execute load modules 54 of his own design—thereby compromising the interests of others in enforcing their own controls within the owner's protected processing environment. For example, the owner could turn off the control that required him to pay for watching or prohibited him from copying content. Since protected processing environment 108 can support a “virtual business presence” by parties other than the owner, it is important for the protected processing environment to be protected against attacks from the owner.
The load module 54 and its associated digital signature 106 is then delivered to the protected processing environment 108. (These items can be provided together at the same time, independently, or at different times.) Protected processing environment 115 applies the same one way hash transformation on load module 54 that a verifying authority 100 applied. Since protected processing environment 108 starts with the same load module 54 and uses the same one-way hash function 115, it should generate the same message digest 116′.
Protected processing environment 108 then decrypts digital signature 106 using the second key 124—i.e., it opens strongbox 118 to retrieve the message digest 116 a verifying authority 100 placed in there. Protected processing environment 108 compares the version of message digest 116 it obtains from the digital signature 106 with the version of message digest 116′ it calculates itself from load module 54 using the one way hash transformation 115. The message digests 116, 116′ should be identical. If they do not match, digital signature 106 is not authentic or load module 54 has been changed—and protected processing environment 108 rejects load module 54.
The public key 124(1) corresponding to private key 122(1) acts only to decrypt (authenticate) digital signature 106(1). Similarly, digital signature 106′ can only be decrypted (authenticated) using public key 124(2) corresponding to the private 122(2). Public key 124(1) will not “unlock” digital signature 106(2) and public key 124(2) will not “unlock” digital signature 106(1).
Different digital signatures 106(1), 106(N) can also be made by using different one way hash functions 115 and/or different encryption algorithms. As shown in
For example, as shown in
Multiple signatures as shown in
These three signatures 55(1), 55(2), 55(3) could all be affixed by the same verifying authority 100, or they could be affixed by three different verifying authorities (providing a “web of trust”). (In another model, a load module is verified in its entirety by multiple parties—if a user trusts any of them, she can trust the load module.) A protected processing environment 108 would need to have all three corresponding “public” keys 124(1), 124(2), 124(3) to authenticate the entire load module 54—or the different load module segments could be used by different protected processing environments possessing the corresponding different keys 124(1), 124(2), 124(3). Different signatures 55(1), 55(2), 55(3) could be calculated using different signature and/or one-way hash algorithms to increase the difficulty of defeating them by cryptanalytic attack.
Verifying authority 100 can use different digital signing techniques to provide different “assurance levels” for different kinds of electronic appliances 61 having different “work factors” or levels of tamper resistance.
An assurance level II electronic appliance 61B may provide a protected processing environment 108 based on a hybrid of software security techniques and hardware-based security techniques. An example of an assurance level II electronic appliance 61B might be a general purpose personal computer equipped with a hardware integrated circuit secure processing unit (“SPU”) that performs some secure processing outside of the SPU (see Ginter et al. patent disclosure
In this example, verifying authority 100 digitally signs load modules 54 using different digital signature techniques (for example, different “private” keys 122) based on assurance level. The digital signatures 106 applied by verifying authority 100 thus securely encode the same (or different) load module 54 for use by appropriate corresponding assurance level electronic appliances 61
Assurance level in this example may be assigned to a particular protected processing environment 108 at initialization (e.g., at the factory in the case of hardware-based secure processing units). Assigning assurance level at initialization time facilitates the use of key management (e.g., secure key exchange protocols) to enforce isolation based on assurance level. For example, since establishment of assurance level is done at initialization time, rather than in the field in this example, the key exchange mechanism can be used to provide new keys (assuming an assurance level has been established correctly).
Within a protected processing environment 108, as shown in
As shown in
In this example, electronic appliances 61 of different assurance levels can communicate with one another and pass load modules 54 between one another—an important feature providing a scaleable virtual distribution environment involving all sorts of different appliances (e.g., personal computers, laptop computers, handheld computers, television sets, media players, set top boxes, internet browser appliances, smart cards, mainframe computers, etc.) The present invention uses verifying authority 100 and the digital signatures it provides to compartmentalize the different electronic appliances depending on their level of security (e.g., work factor or relative tamper resistance). In particular, verifying authority 100 and the digital signatures it provides isolate appliances with significantly different work factors—preventing the security of high work factor appliances from collapsing into the security of low work factor appliances due to free exchange of load modules 54.
In one example, verifying authority 100 may digitally sign identical copies of load module 54 for use by different classes or “assurance levels” of electronic appliances 61. If the sharing of a load module 54 between different electronic appliances is regarded as an open communications channel between the protected processing environments 108 of the two appliances, it becomes apparent that there is a high degree of risk in permitting such sharing to occur. In particular, the extra security assurances and precautions of the more trusted environment are collapsed into the those of the less trusted environment because an attacker who compromises a load module within a less trusted environment is then be able to launch the same load module to attack the more trusted environment. Hence, although compartmentalization based on encryption and key management can be used to restrict certain kinds of load modules 54 to execute only on certain types of electronic appliances 61, a significant application in this context is to compartmentalize the different types of electronic appliances and thereby allow an electronic appliance to protect itself against load modules 54 of different assurance levels.
In accordance with this feature of the invention, verifying authority 100 supports all of these various categories of digital signatures, and system 50 uses key management to distribute the appropriate verification keys to different assurance level devices. For example, verifying authority 100 may digitally sign a particular load module 54 such that only hardware-only based server(s) 402(3) at assurance level XI may authenticate it. This compartmentalization prevents any load module executable on hardware-only servers 402(3) from executing on any other assurance level appliance (for example, software-only protected processing environment based support service 404(1)).
To simplify key management and distribution, execution environments having significantly similar work factors can be classified in the same assurance level.
To show this type of isolation,
The “desert islands” are created by the use of different digital signatures on each of load modules 54a, 54b, 54c. In this example, all of the appliances 61 may freely communicate with one another (as indicated by the barges—which represent electronic or other communications between the various devices. However, because particular assurance level load modules 54 will be authenticated only by appliances 60 having corresponding assurance levels, the load modules cannot leave their associated “desert island”—providing isolation between the different assurance level execution environments. More specifically, a particular assurance level appliance 61 thus protects itself from using a load module 54 of a different assurance level. Digital signatures (and/or signature algorithms) 106 in this sense create the isolated “desert islands” shown—since they allow execution environments to protect themselves from “off island” load modules 54 of different assurance levels.
A load module or other executable may be certified for multiple assurance levels. Different digital signatures may be used to certify the same load module or other executable for different respective assurance levels. The load module or other executable could also be encrypted differently (e.g. using different keys to encrypt the load module) based on assurance level. If a load module is encrypted differently for different assurance levels, and the keys and/or algorithms that are used to decrypt such load modules are only distributed to environments of the same assurance level, an additional measure of security is provided. The risk associated with disclosing the load module or other executable contents (e.g., by decrypting encrypted code before execution) in a lower assurance environment does not compromise the security of higher assurance level systems directly, but it may help the attacker learn how the load module or other executable works and how to encrypt them—which can be important in making bogus load modules or other executables (although not in certifying them—since certification requires keys that would only become available to an attacker who has compromised the keys of a corresponding appropriate assurance level environment). Commercially, it may be important for administrative ease and consistency to take this risk. In other cases, it will not be (e.g. provider sensitivities, government uses, custom functions, etc.)
If the load module is found to satisfy its specifications, a verifying authority 100 determines whether it is authorized to generate one or more new specifications for the load module (
If the load module fails the test (“N” exit to decision block 508), verifying authority 100 determines whether it is authorized and able to create new specifications corresponding to the actual load module performance, and whether it is desirable to create the conforming specifications (
A verifying authority 100 may then digitally sign the load module 54 to indicate approval (
This is a continuation of commonly assigned co-pending application Ser. No. 11/454,072, which is a continuation of application Ser. No. 09/925,072, filed on Aug. 6, 2001, Publication No. US 2002-0023214 A1, which is a continuation of application Ser. No. 09/678,830, filed on Oct. 4, 2000, now U.S. Pat. No. 6,292,569, which is a continuation of application Ser. No. 08/689,754, filed on Aug. 12, 1996, now U.S. Pat. No. 6,157,721, all of which are incorporated herein by reference. This application is related to application Ser. No. 08/388,107, filed 13 Feb. 1995, abandoned via file wrapper continuation, now U.S. Pat. No. 5,982,891 incorporated herein by reference.
Number | Date | Country | |
---|---|---|---|
Parent | 11454072 | Jun 2006 | US |
Child | 11781062 | US | |
Parent | 09925072 | Aug 2001 | US |
Child | 11454072 | US | |
Parent | 09678830 | Oct 2000 | US |
Child | 09925072 | US | |
Parent | 08689754 | Aug 1996 | US |
Child | 09678830 | US |