DE102016207855A1 - Method and device for averting computer time attacks on a secret calculation - Google Patents

Abstract

Method (40) for averting computer time attacks on a secret calculation, characterized by the following features: - before the calculation, a certain waiting period (accidental) is triggered, - the calculation is made while the waiting period (accidentally) elapses, and - as soon as the calculation and the waiting period (random) has expired, a result of the calculation for shipment is provided (48).

Description

The present invention relates to a method for averting computational time attacks on a secret computation. The present invention also relates to a corresponding controller for a bus, a corresponding computer program and a corresponding storage medium.

State of the art

A cryptoanalytic method which exploits the physical implementation of a cryptosystem in a device is referred to in cryptology as a side channel attack. A widespread variant of the side channel attack is known to the person skilled in the art as a timing attack.

To perform a computational time attack, the attacker measures the computational time of the implemented cryptographic method for various inputs, usually selected by the attacker himself. In this case, the attacker takes advantage of the fact that certain cryptosystems require slightly different computing times from case to case, which may depend on the one hand on the key chosen, on the other hand on the input data, ie the plain text or cipher. Through a repeated runtime analysis, the key can be reconstructed step by step in this way. Known computing time attacks are directed against a wide variety of hardware and software.

DE 10 2013 205 542 A1 relates to a device for processing data, the device having an input interface for receiving input data, a processing unit for processing data and a coding unit which is designed to encode data words received at the input interface as input data in order to obtain coded data words, wherein for coded data words the coded data words and / or their processing by the device characterizing measured values as a function of at least one physical size of the device can be determined, wherein the coding unit is adapted to encode the data words so that a predetermined portion of all measured values, preferably at least approximately 50% of all measured values, a difference from a default value which is less than or equal to a predefinable threshold, and the processing unit is adapted to process the coded data words.

Disclosure of the invention

The invention provides a method for repelling computing time attacks on a secret calculation, a corresponding device, a corresponding computer program and a corresponding storage medium according to the independent claims.

The approach according to the invention is based on the knowledge that the in 1 Attack scenario shown is conceivable. An attacker (node A) can therefore use a bus ( 10 ) Messages with a request ( 11 ) for decrypting, encrypting or signing a message, checking a personal identification number (PIN) or similar to the security software ( 12 ) of a legitimate control device (node B), which by means of a communication controller ( 14 ) and transceiver (transceiver, 15 ) to the bus ( 10 ), thus triggering the corresponding calculation on the attacked controller (Node B) and measuring the response time ( 16 ). Any functions can be controlled via various physical control devices ( 17 ), which can then be called up according to a client-server model.

2 shows the total runtime measured by the attacker (node A) with special consideration of the internal runtimes within the attack target (node B). In this case, "t _propAB = const" designates the runtime from node A to B, "t _queue = 0..X" the time until the message from the security software ( 12 ), "t _{secure_algo} = V..W" the computation time to be protected, "t _{sw_hprio} = 0..Z" any preemptions or interrupts, "t _tx = const" the time required to reply messages in the communication controller ( 14 ) and trigger their dispatch, "t _{arb_int} = 0..M" the time required for the internal arbitration of the reply message against other messages to be sent, "t _{arb_bus} = 0..L", which is to arbitrate the response message against other messages to be sent by other nodes take time and finally "t _propBA = const" the runtime from node B to A. Many parts of the total run time are thus variable. Therefore, the attacker (node A) usually determines the minimum of the measured response times for a computing time attack for each hypothesis. With a sufficient number of requests, the attacker (node A) can thus determine the minimum total run time for the response time (node A). 16 ). The thus determined minimum response time ( 16 ) allows the attacker (node A) to draw conclusions about the internal calculation time.

To obscure the internal calculation time, it would be conceivable to insert a random waiting time (t _random = P..Q) after the time period t _{secure_algo} ( 2 ). Since such an additional random delay also has a minimum, an attacker could still be able to determine the minimum total run time with a sufficiently large sample. Thus, one would However, such an approach may make it difficult to identify the minimum maturity, but it may not prevent it altogether.

An advantage of the proposed solution, in comparison to such an implementation - and certainly a completely unprotected one - is that computing time attacks are completely prevented by an embodiment of the invention. Consequently, the danger emanating from a computing time attack need not be excluded by other software measures. An embodiment of the invention is - comparatively easy to implement - from the point of view of the required hardware resources. For this purpose, the state of the art is expanded insofar as the random waiting time before or with the calculation time to be protected (t _{secure_algo} = V..W) is started ( 3 ). Thus, on the one hand, the calculation time to be protected (t _{secure_algo} = V..W) and, on the other hand, the "inserted" time (t _random = P..Q) run in parallel. In this way, the calculation time to be protected (t _{secure_algo} = V..W) is completely obfuscated. In other words, the response time of the attack target (node B) is no longer related to the calculation time to be protected (t _{secure_algo} = V..W).

The measures listed in the dependent claims advantageous refinements and improvements of the independent claim basic idea are possible. Thus, the inventive prevention of a computing time attack in the case of the communication controller ( 14 ) by a relatively simple extension according to 3 be achieved. A basic idea of this refinement is to determine the transmission time for selected messages in the communication controller ( 14 ) so that the attacker (node A) can not conclude on the calculation time to be protected (t _{secure_algo} = V..W). This aspect also allows the secure concealment of the calculation time to be protected (t _{secure_algo} = V..W).

According to a further aspect, it may be provided that the configuration of the communication controller ( 14 ) prescribes an essentially random _{selection of} the waiting period (t _random ), which is fixed in advance, but at fixed term. This is a simple realization of the basic idea according to the invention and also provides a deterministic behavior, but nevertheless differs substantially from known approaches.

According to another aspect, it may be provided that the configuration ( 42 ) _specifies a minimum length P and a maximum length Q of the waiting period (t _random = P..Q), the latter being determined at random so as to comply with the minimum length and the maximum length. If the waiting period (t _random = P..Q) expires before the calculation is completed, the minimum length and / or the maximum length are changed. This variant detects if, after a software change, the originally set limits P and Q are no longer correct under all conditions. In the described embodiment, the implementation in such an application could essentially self-configure or at least adapt.

This object is achieved according to the invention very easily by means of adapted hardware or software. The advantage of a possible hardware solution is that such a solution requires only minor changes in the software. Typically, it is sufficient to read and write new registers for configuration and operation.

Brief description of the drawings

Embodiments of the invention are illustrated in the drawings and explained in more detail in the following description. It shows:

1 a computing time attack in a vehicle via its CAN bus.

2 schematically the durations occurring in the case of an alternative solution starting with the dispatch of the request of the attacker to the receipt of the answer.

3 the 2 corresponding maturities in the case of an embodiment of the invention.

4 a representation of the implementation of this embodiment on the part of the attack target. This unit is the communication controller ( 14 ) or part of it. This unit is responsible for the delay at _random . The function of this unit can also be realized completely in software.

Embodiments of the invention

3 shows the total runtime that the attacker (node A) can measure in the case of a computational time attack. In order to disguise the calculation time to be protected (t _{secure_algo} = V..W), a random or pseudo-random waiting period (t _random = P..Q) is now determined before the start of the calculation. This waiting period (t _coincidence = P..Q) runs as in 3 shown parallel to the calculation time (t _{secure_algo} = V..W) of the security software ( 12 ). The time "Start" marks the beginning of the calculation time (t _{secure_algo} = V..W). "End" marks the time when the "Upstream Unit" ( 4 ) causes the result of the calculation to be protected to be sent to the attacker (node A). The difference between times "end" and "start" can also be greater than by t _coincidence maximum reach - this happens when z. For example, t _{sw_hprio becomes} unexpectedly large. Starting with the time "end" is the communication controller ( 14 ) responsible for the shipment.

For the procedure to work, the following inequality must be observed: t _random > t _{secure_algo} - t _tk + δ _t

In the above formula δ _t denotes all other constant time shares between the start (start) of the calculation time (t _{secure_algo} = V..W) and the time (end) of the result. By "constant time share" is meant the minimum time for an action; According to this sense of the word, in the exemplary case of a stochastic time variable t ∈ [a, b], the lower limit a would be regarded as a constant component, while the interval length b - a would be considered as the variable component of the random variable t. "Purely variable" times and the variable parts of the times - constant parts have been considered in the inequality - between start and end are to be regarded as a kind of natural boundary condition, comparable to a known from the network technology variance of the duration of data packets (jitter).

Preferably, the time t _is random, but to choose between P and Q. The upper limit Q is arbitrary, as long as resulting response times are tolerable; the lower limit, ie the minimum random waiting time P, is determined as a function of the estimated computing time (t _{secure_algo} = V..W) of the algorithm to be protected. This determination can be nonlinear, arbitrarily simple or complicated. For example, if the calculation time (t _{secure_algo} = V..W) of the algorithm is estimated to be very short, a very high value may be chosen for Q; if the calculation time (t _{secure_algo} = V..W) of the algorithm is considered short, a mean value for Q may be chosen; and if the calculation time (t _{secure_algo} = V..W) of the algorithm is estimated to be long, a large value for Q may be chosen. In any case, the following inequality must be fulfilled: t _random > t _{secure_algo} - t _tx + δ _t

A dynamic adaptation of the limits P and Q is considered. Once the inequality is violated, the implementation adjusts the limits. If, for example, the value t _random -P-1 is selected and the implementation - be it hardware or software - determines that a transmission request signaling status indicator (flag) for the response message only after this time from the software set, the implementation makes the following assignments: P _new : - P _alt + x and Q _new : - P _new + y

An implementation of the inventive concept is possible in software and hardware. In this case, a software implementation requires a change in the security software ( 12 ), although the algorithm to be protected itself does not have to be adapted.

In an exemplary hardware implementation, however, additional configuration registers in the "upstream unit" (FIG. 4 ) store the identifiers of the messages for which the send time is to be randomized. In addition, the boundary conditions for the randomization can be stored in further configuration registers. The boundary conditions include, for example, the minimum (P) and maximum (Q) durations of the randomly selected waiting period (t _random = P..Q). These boundary conditions could with respect to a given time unit -. B. μs -, as a multiple of a clock period of the microcontroller ( 13 ) or a bit interval of the bus ( 10 ).

4 FIG. 12 schematically illustrates the exemplary implementation of this invention as a digital hardware extension in a communications controller (FIG. 14 ). The procedure ( 40 ) could z. For example, the potential attack target (Node B) receives a message asking it to calculate something according to the algorithm to be protected. Immediately before the start (start) of the calculation time (t _{secure_algo} = V..W), the security software ( 12 ) of the potential attack target (Node B) by write access ( 46 ) on the time unit ( 41 ) start the calculation with. The time unit ( 41 ) determined by its configuration ( 42 ) a waiting period (t _zufall = P..Q), the monitoring by means of a suitable timer (timer) takes place. The waiting period can be constant, purely random or dependent on an estimated calculation time (t _{secure_algo} ) as described at the beginning. The security software ( 12 ) copies the reply message to the communication controller ( 14 ) and sets ( 47 ) then the said status indicator for the send request ( 43 ) in the delay unit according to the invention ( 4 ). The test unit ( 45 ) checks whether the waiting period (t _coincidence = P..Q) has expired and also the send request ( 43 ) is present. If both are given, then it sets ( 45 ) finally the delayed send request ( 44 ) for the reply message in the communication controller ( 14 ).

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102013205542 A1 [0004]

Claims (10)

Procedure ( 40 ) to _{avert computing time attacks} on a secret calculation, characterized by the following features: - before the calculation a certain waiting period (t _random ) is triggered, - the calculation is made while the waiting period (t _random ) elapses, and - as soon as the calculation is completed and the waiting period (t _random ) has expired, a result of the calculation for shipment is provided ( 48 ). Procedure ( 40 ) according to claim 1, characterized by the following features: - the waiting period (t _random ) is determined by a communication controller ( 14 ) for a bus ( 10 ), in particular a CAN, determines and monitors and - delivering the result ( 48 ) is done by a delayed send request ( 44 ) in the communication controller ( 14 ). Procedure ( 40 ) according to claim 2, characterized by the following feature: - the waiting period (t _random ) is determined on the basis of a configuration ( 42 ). Procedure ( 40 ) according to claim 3, characterized by the following feature: - the configuration ( 42 ) specifies a fixed length of the waiting period (t _random ). Procedure ( 40 ) according to claim 3, characterized by the following features: - the configuration ( 42 ) specifies a minimum length and a maximum length of the waiting period (t _random ) and - the waiting period (t _random ) is determined at random or depending on an estimated calculation time such that it complies with the minimum length and the maximum length. Procedure ( 40 ) according to claim 5, characterized by the following features: - _If the waiting period (t _random ) _expires before the calculation is completed, then the minimum length and / or the maximum length are changed. Procedure ( 40 ) according to claim 5 or 6, characterized by the following features: - the minimum length and the maximum length are in the configuration ( 42 ) with respect to a given time unit, a clock period or a bit interval of the bus ( 10 ). Computer program ( 12 ), which is set up, the method ( 40 ) according to one of claims 1 to 7. Machine-readable storage medium on which the computer program ( 12 ) is stored according to claim 8. Communication Controller ( 14 ) for a bus ( 10 ), which is set up, the procedure ( 40 ) according to one of claims 1 to 7.

Images