The present invention relates to a method, system and computer-readable medium for providing a secure time source for trusted execution environments.
Trusted execution environments (TEEs)—such as INTEL software guard extensions (SGX) or ARM TrustZone—provide hardware supported shielded environments for applications executing inside them, which ensures isolation from the rest of the software stack and even prevents privileged software from directly manipulating the trusted application runtime state. TEEs also benefit from some services usually supported by an untrusted OS—such as sealing (encrypted and authenticated storage) or access to a coarse-grained clock service.
When using TEEs, only certain privileged and trusted code has access to hardware resources. Therefore, in order to prevent code running inside TEEs from hijacking hardware resources, the code runs in user privilege mode and does not have direct access to such resources. This approach is also helpful to maximize compatibility with existing operating systems (OSs). As a result, TEEs need to either request temporary and exclusive access to a peripheral device or invoke an untrusted party via an outside call (ocall—i.e., an outside call to a predefined function).
In the first scenario—common in ARM or RISC-V based architectures, TEEs request access to the secure-monitor. In the latter case—e.g. the SGX scenario—the untrusted party take cares of performing the task. In both cases, the OS usually acts as an intermediary because the OS is typically the party in charge of managing the hardware resources or handling inter-process communication. Because the OS is not part of group of trusted systems that can provide a secure environment for the execution of operations, e.g., is not part of the trusted computing base (TCB)—and therefore is not trusted by a TEE—a compromised OS can intentionally mount delay attacks by generating some interruptions that alter the regular execution flow of the code inside a TEE, or by not delivering the messages exchanged between the trusted party and the TEEs in a timely and reliable way. The messages exchanged among processes can be sent encrypted to ensure confidentiality and integrity of the exchanged data. However, encryption does not protect against such delaying attacks.
Many information technology (IT) applications—e.g., in areas like internet-of-things (IoT), banking, or connected vehicles—rely on the existence of accurate timing information and would suffer from such attacks. Furthermore, the inventors have recognized that accurate and reliable measurements of elapsed time are essential to ensure safe operation of the hardware or the network system (e.g., when measuring round trip times of packages), to evaluate the performance of critical operations, and even to detect malware or microarchitecture attacks as they cause timing anomalies on the execution of some functions.
While TEEs provide confidentiality and integrity guarantees for applications, the inventors have identified that current TEEs do not offer access to a trusted source of time, which is particularly important for TEE applications to be able to self-monitor their performance and security.
Because TEEs do not have access to a trusted (and fine-grained) source of time, currently the TEEs need to rely on an external party to get such information. Such an external party could be the untrusted OS (as aforementioned) or a remote party, either of which making the timer vulnerable to attacks and manipulations or increasing latency in an unpredictable way. Alternatively, the TEE enclave might try to build its own fine-grained timer by using a counting thread. The inventors have determined that this approach is not reliable for accurate timing measurements as an untrusted OS might interrupt or de-schedule such thread. Further, solutions based on timing threads are not scalable as they consume one virtual core per application implementing such timer.
An aspect of the present disclosure provides a method enabling trusted timing services to an enclave of a computer having memory and a trusted hardware timer. The computer executes a privileged management program and an untrusted operating system. The privileged management program has access to the memory and the trusted hardware timer, has higher privileges than the untrusted operating system, and exposes a system call to the enclave for requesting the trusted timing services. The method includes: receiving, by the privileged management program, a request for timing services from the enclave, via the system call; reserving, by the privileged management program, a memory region of the memory for tracking time; and writing, by the privileged management program, at least one value of the trusted hardware timer into the memory region.
Subject matter of the present disclosure will be described in even greater detail below based on the exemplary figures. All features described and/or illustrated herein can be used alone or combined in different combinations. The features and advantages of various embodiments will become apparent by reading the following detailed description with reference to the attached drawings, which illustrate the following:
One or more aspects of the present disclosure provide a trusted and secure source of time measurement for TEE applications (even in the presence interruptions), which may avoid potential disruptions to—and increases the security of—the functioning of the TEE (e.g., protecting against an OS that delays the passage of messages between the TEE and a trusted party, or against an OS that provides interruptions to the execution of functions in the TEE).
Compared to previous approaches in the art, embodiments of the present disclosure preserve computing resources, are scalable, and do not require the consumption of a virtual core to utilize a trusted time source. Embodiments according the present disclosure advantageously also do not require any modifications to the OS and can be integrated with existing TEE instantiations. Embodiments of the present disclosure provide technological improvements to the functioning of a computer. For example, embodiments can improve the security of TEEs (including the securing of applications running at least partially in the TEEs) and improve processing time, as compared to prior approaches in the art with TEEs that do not have the secured time measurement mechanism offered by the present disclosure. Embodiments of the present disclosure also improve the function of IoT systems by ensuring access to a trusted and reliable timing resource, which if fine grained, allows for distribution of the timing across multiple components, reducing latency and increasing security.
An aspect of the present disclosure provides a mechanism to reliably devise a secure source of time for TEE applications. A mechanism provided according to an aspect of the present disclosure securely measures elapsed time between two events executed within the TEE (also known as enclave in some contexts), both measuring the time when the enclave was executing and the actual real elapsed time. For example, whenever the enclave is involved in a context switch (e.g., because of the OS scheduler, an interrupt, or a syscall), the mechanism is triggered updates the timing information for the enclave. The timing information may include the elapsed time and/or interrupt counter.
Aspects of the present disclosure enable enclaves to measure elapsed time reliably by directly exposing, to the enclaves, certain interfaces to a privileged and trusted system management program (referred to herein as a privileged management program for short and a primary example of which is system firmware) in the form of system calls (syscalls), where the direct exposure means that the shielded environment allows for communication with a program that would not otherwise interact with the enclave. The enclaves can thereby directly communicate with the system's trusted program, avoiding the interference of the OS.
According to an aspect of the present disclosure, the privileged and trusted system program (e.g., system firmware) introduces a protected memory region for each of the enclaves on the machine, where timing information and interruptions that alter the regular execution flow of the enclave are stored. The privileged and trusted system program (e.g., system firmware) has direct and privileged (e.g., sole write access) to protected memory, which is used for the protected memory region.
According to a preferred implementation of the present disclosure: (1) trusted firmware and trusted hardware are provided, (2) the firmware has direct access to a hardware clock/timer and protected memory, (3) the firmware runs in a privileged mode (e.g., most privileged mode) of the machine, and (4) the firmware exposes one or more direct system calls to the enclaves for requesting timing data.
According to an aspect of the present disclosure, the firmware reserves different private memory regions—one for each enclave—where it stores the information referring to timing measurements. These regions lay on the firmware private memory and are not accessible by any other software. Whenever the enclave requests to start a measurement, the firmware writes the current value of the hardware clock (in cycles, e.g., the number of cycles from an oscillator of the hardware clock determined from a signal thereof) into the dedicated memory region and the hardware clock keeps counting cycles as usual. If there is an interruption, or the execution of the enclave is stopped, the firmware updates the corresponding elapsed time value in the private memory region accordingly. When the execution is resumed, the firmware writes down the resume time and so on till the enclave decides to stop the timer. At that time, there are two available values: one refers to the actual elapsed clock cycles and the other one refers to the time that the enclave has been running. If there are no interruptions, the two values are equal.
A first aspect of the present disclosure provides a method enabling trusted timing services to an enclave of a computer having memory and a trusted hardware timer. The computer executes a privileged management program and an untrusted operating system. The privileged management program has access to the memory and the trusted hardware timer, has higher privileges than the untrusted operating system, and exposes a system call to the enclave for requesting the trusted timing services. The method includes: receiving, by the privileged management program, a request for timing services from the enclave, via the system call; reserving, by the privileged management program, a memory region of the memory for tracking time; and writing, by the privileged management program, at least one value of the trusted hardware timer into the memory region
According to a second aspect of the present disclosure, the method according to the first aspect may further include: sending, by the privileged management program in response to receiving a request from the enclave, a time measurement based on the at least one value written in the memory region to the enclave.
According to a third aspect of the present disclosure, in a method according to at least one of the preceding aspects, the memory includes protected memory having the memory region, and the privileged management program has exclusive access to the protected memory.
According to a fourth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the enclave is one of a plurality of enclaves of the computer, and the method further includes: receiving, by the privileged management program, a request for timing services from a second enclave of the enclaves via the system call; reserving, by the privileged management program, a second memory region of the protected memory for tracking time for the second enclave; and writing, by the privileged management program, at least one value of the trusted hardware timer into the second memory region.
According to a fifth aspect of the present disclosure, in a method according to the at least one of preceding aspects, the privileged management program is firmware of the computer.
According to a sixth aspect of the present disclosure, a method according to at least one of the preceding aspects further includes, based on detecting an interruption, an exception, or a context switch involving the enclave, writing, by the privileged management program, at least one further value of the trusted hardware timer and information on the interruption or context switch into the memory region.
According to a seventh aspect of the present disclosure, in a method according to at least one of the preceding aspects, reserving the memory region occurs prior to receiving the system call and includes: reserving a page of protected memory, the page having an enclave identification field, a running field, a CPU start time field, and an enclave start time field; and associating the page to the enclave by writing to the enclave identification field. The privileged management program exposes a plurality of system calls to the enclave for requesting the trusted timing services. The system calls include the system call, and the system call is a timer start system call. The writing, by the privileged management program, the at least one value of the trusted hardware timer into the memory region is in response to receiving the timer start system call and comprises writing in the CPU start time field and the enclave start time field based on a current value of the trusted hardware timer, and the method further includes updating the running field to true.
According to an eighth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the page of protected memory further includes an enclave elapsed time field and at least one timer interruption cause counter, and the method further includes: based an interruption, an exception, or a context switch involving the enclave occurring; updating, by the privileged management program, the enclave elapsed time field based on the enclave start time field, a current value of the trusted hardware timer, and a current value of the enclave elapsed time field; and incrementing, by the privileged management program, the at least one timer interruption cause counter, and based on execution of the enclave resuming after the interruption, the exception, or the context switch: overwriting, by the privileged management program, the enclave start time field based on a current value from the trusted hardware timer.
According to a ninth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the page of protected memory further includes an enclave elapsed time field, and a total elapsed time field. The method further includes: receiving, by the privileged management program from the enclave, a timer stop system call, the system calls include the timer stop system call; updating, by the privileged management program, the running field to false; updating, by the privileged management program, the total elapsed time field based on a current value of the CPU start time field and a current value of the trusted hardware timer; updating, by the privileged management program, the enclave elapsed time field based on the enclave start time field, the current value of the trusted hardware timer, and a current value of the enclave elapsed time field; and sending, by the privileged management program, a report to the enclave, the report having an indication on whether or not the enclave was interrupted and one or more of current values of the enclave elapsed time field, the total elapsed time field; the CPU start time field, or the enclave start time field.
According to a tenth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the page of protected memory further has an enclave elapsed time field, and a total elapsed time field, and the method further includes: receiving, by the privileged management program from the enclave, a timer report system call, the system calls comprising the timer report system call; calculating, by the privileged management program, a total elapsed time based on a current value of the CPU start time field and a current value of the trusted hardware timer; calculating, by the privileged management program, an enclave elapsed time based on the enclave start time field, the current value of the trusted hardware timer, and a current value of the enclave elapsed time field; generating, by the privileged management program, a report to the enclave, the report comprising one or more of the enclave elapsed time, the total elapsed time; an indication on whether or not the enclave was interrupted, or an interruption count; signing, by the privileged management program, the report with a cryptographic key; and sending, by the privileged management program, the signed report to the enclave.
According to an eleventh aspect of the present disclosure, in a method according to at least one of the preceding aspects, during the reserving, by the privileged management program, the memory region is executed based on the enclave being instantiated; and the method further includes: based on determining that the exclave has requested termination, releasing the reserved memory region.
According to a twelfth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the method further includes: determining, by the privileged management program, a time measurement based on the at least one value of the trusted hardware timer written in the memory region; encrypting, by the privileged management program, the time measurement; and sending, by the privileged management program, the encrypted time measurement to the untrusted operating system for storage.
According to a thirteenth aspect of the present disclosure, in a method according to at least one of the preceding aspects, the time measurements are encrypted using a current key, the current key being derived based on a current state of the enclave and a base key or a previous key.
According to a fourteenth aspect of the present disclosure, a system is provided that is configured to enable trusted timing services to an enclave, the system having: a trusted hardware timer; memory, the memory storing a privileged management program and an untrusted operating system, the privileged management program being configured to: have access the memory and the trusted hardware, to expose a system call to the enclave for requesting the trusted timing services, and have higher system privileges than the untrusted operating system; and one or more hardware processors in communication with the memory and the trusted hardware timer, the hardware processors, alone or in combination, are configured to provide the enclave, and are configured to execute the privileged management program such that the method according to at least one of the first through thirteen aspects are executed.
According to a fifteenth aspect of the present disclosure, a tangible, non-transitory computer-readable medium is provided, which has instructions thereon which, upon being executed by one or more hardware processors, alone or in combination, are configured to execute a method according to at least one of the first through thirteenth aspects. The computer readable medium may be included in the system according to the fourteenth aspect.
According to another aspect of the present disclosure, a method is provided that provides a trusted mechanism to enclaves to measure elapsed time via system calls to a piece of trusted firmware running on the machine. The method includes at least one of the following operations:
An “enclave” according to the present disclosure is a hardware-supported shielded environment for executing applications, which ensures isolation from the rest of the software stack; a trusted execution environment (TEE) is an enclave. “Firmware” according to the present disclosure is (generally) the highest privileged software running on the device, which provides low-level control of the hardware and higher-level hardware abstractions services to other components. As used in the present disclosure, a “component” refers to any portion of a system, e.g., hardware, software, firmware, an OS, an application, a timer, etc. “Privileged” according to the present disclosure means that the relevant component (e.g., application) has direct access (e.g., read and/or write) to hardware or software resources. A “higher privileged” element has greater access than a “lower privileged” element. A “most privileged” element is the element having the greatest access in the system as compared to other elements, particularly the highest access to trusted hardware resources (e.g., firmware is generally the “most privileged” application, with direct access to trusted hardware resources, including protected memory) and whose execution cannot be interrupted by other elements running at lower privilege levels. “Trusted” according to the present disclosure means that the element (firmware, software, hardware, application, program, etc.) can be relied upon to a pre-specified extent in relation to other components to perform tasks without secretly executing harmful, malicious, or unauthorized programs. “Reliable” according to the present disclosure means that the reliable component can perform its function at least under specified conditions for a specified time, without interruption.
A “coarse grain” clock service, provides time resolution in orders of magnitude coarser than the time it takes to execute the piece of code being measured, e.g., milliseconds, rendering such clock service less preferred for accurately measuring the execution time of the aforementioned piece of code whereas a “fine-grained” clock service provides resolution in a similar or the same order of magnitude as hardware clock cycles which enables the accurate measurement of the execution time of even a single instruction. A “protected memory region,” “private memory region,” or “exclusive memory region” is a region of hardware memory accessible by a specific component (e.g., the firmware), but is inaccessible other components (e.g., the OS). Being able to “access” a specific resource or component, or having “access to” a specific resource or component according to the present disclosure means that the component that has such access to the specific resource can directly utilize the resource itself or utilize information in the resource, without having to make a request to any intermediate component having higher privileges to the specific resource or component.
Accordingly, aspects of the present disclosure provide technological improvements in the field of computer processing units (ComPUs), and more particularly to ComPUs having trusted execution environments (TEEs). For example, aspects of the present disclosure improve the security of TEEs (including the securing of applications running at least partially in the TEEs) and improve processing time, as compared to ComPUs with TEEs that do not have the secured time measurement mechanism offered by the present disclosure.
Along with the TEEs 8, 10, the hardware 4 and firmware 22 (e.g., as stored trusted program), below the OS 20 and the trust line 6 of
According to an embodiment, the application operating in the most privileged operation mode of the system model 1 is the firmware 22 running on the machine, which can get different names depending on the architecture, and has complete access to all components of hardware 4. Such firmware 22 is included in the trusted computing base (TCB). It guarantees that the hardware 4 providing security guarantees for the TEE, e.g., TEE 8, is properly configured and used. For instance, firmware 22 will be the one either configuring and launching the TEE 8, 10 or verifying that the OS 20 has done so in an appropriate way. The firmware 22 is notified whenever there is a hardware 4 interruption or an exception and firmware 22 either executes some software routine as a response and/or delegates software routine to another entity, e.g., the OS 20. Generally, if one component is more privileged than another component (e.g., as the firmware is more privileged than the OS in the present embodiment), this means that the more privileged operation or component has access to or can change the less privileged operation or component, but the less privileged component is restricted from accessing or changing the more privileged component. In the system model 1, as the most privileged software running on the machine, firmware 22 cannot be interrupted while executing unless it allows such interruptions. The rest of the components can communicate with the firmware via a clearly defined interface, which embodiments of the present disclosure enhance to include functions to carry out timing measurements for enclaves, and to get a signed report verifiable by a third party.
The OS 20 runs on top of the firmware 22. It runs at higher privilege level than regular applications, e.g., normal app 16 and normal app 18, but at lower privilege level than the firmware 22. The OS may be a general purpose OS 20, that is, embodiments of the present disclosure do not require modification of the OS 20 to be implemented. The OS 20 commonly handles inter-process communication, task scheduling, most of the interruptions and exceptions, and holds the majority of the drivers for handling peripherals. As a result, it can schedule and de-schedule an enclave process, e.g., TEE 8 process, at will, interrupt it, or except it from execution. In all of these cases, the privileged firmware 22 can get notified of the changes, and firmware 22 will not prevent them. In any case, the OS 20 is not part of the TCB. Hence, the OS 20 cannot act as timekeeper because it may not accurately report timing information. For example, a compromised OS 20 may try to delay or modify timing measurements requested by an enclave, e.g., TEE 8.
Finally, the enclaves, e.g., TEE 8, 10, run in isolated containers with their code protected from the rest of the untrusted software of the system, including the OS 20. They can only communicate with the outside world using specific entry and exit points, e.g., the system calls 34 exposed by embodiments of the present disclosure. As shown in
Reliable Time Measurements
Timers can be used with each memory region, among other things, to either (1) generate reports (e.g., attestation or proof of elapsed time) for benchmarking, to prevent, e.g., time of check, time of use exploits, e.g., a file-based race condition that occurs when a resource is checked for a particular value, such as whether a file exists or not, and that value then changes before the resource is used, invalidating the results of the check; or (2) to provide temporal forensic capabilities to enclave developers, to detect attacks on code integrity and even for time based attestation protocols. The timers can utilize HW timer 24, and can operate in conjunction with a memory region, e.g., memory region 36, 38.
An embodiment of the present disclosure reserves a memory region 36, e.g., a page, which is then associated to an enclave, e.g., TEE 8, and can do so for each of the enclaves, e.g., for TEE 8, 10, memory regions 36, 38 are reserved. Such memory region 36 can contain private information referring to TEE 8, for example, the memory region 36 that is accessible from TEE 8 and the input/output interfaces of TEE 8, and memory region 36 is formed from a portion of memory only accessible for the firmware 22 of the machine. If the entity or firmware 22 handling TEE 8 already holds some structure to keep track of some other information referring to TEE 8, that structure can be enhanced so that the structure can hold the necessary information to compute the timing measurements. A malicious or compromised OS might try to remove from firmware 22 pages, e.g., memory regions 36, from the dynamic random access memory (DRAM) in order to cause delays on timing measurements. To avoid this situation, the firmware 22 will ensure that these memory regions 36 are not removed from the DRAM by any other entity. Additionally, the firmware 22 can monitor the notifications received regarding the OS 20 and the TEE 8 and issue commands in response to the notifications. For example, the firmware 22 can monitor the notifications of actions taken by the OS 20 and issue calls, e.g., timer stop, timer start (below), on its own, instead of having the TEE 8 issue those calls. In some of these embodiments where firmware 22 issues calls on behalf or in addition to TEE 8, the entity tracking the time could be the shared memory, using an additional and independent portion of the memory, for example for tracking or monitoring the performance of the system. These embodiments could utilize memory, e.g., shared memory, independent of the timing information provided to the TEE 8 as per its request.
An example layout 40 with different fields which can be used to track time is depicted in
In order to simplify the management of the timers for all the possible enclaves, e.g., TEE 8, 10, that might be running on the system model 1, the firmware 22 could have some variables indicating the active timers. Thus, a variable can be consulted whenever there is an interruption or a context switch from TEE 8 to another process, and then only the memory regions 38 of the active timers for the active TEE 10 will be updated. An easy way to implement this is to assign to each timer an id equal to the position of the bit that will be set to 1 in the variables. For example, one variable whose binary representation is equal to 0xb000010010 indicates that the timers 1 and 4 are running. Alternatively, the firmware 22 could check the running field of each of the enclaves, e.g., TEE 8, 10. Pseudocode is provided below as an example of how a variable can be consulted whenever there is an interruption or a context switch from TEE 8 to another process.
Assuming the system has just booted, and TEE IDs, e.g., IDs for TEE 8, 10, are assigned sequentially by the firmware 22, then we have:
The pseudocode for updating a memory region, e.g., memory region 36, 38, if there is an interruption or exception could be executed by the firmware 22 and may be directly invoked by the TEEs, e.g., TEE 8, 10 (although an interruption or exception could be indirectly invoked if the TEEs cause an interruption, or if they make an external call to another function). The following pseudocode assumes the TEE is running as the function that is calling it has already checked that condition. So the update function can be as follows:
According to an embodiment, the firmware 22 is modified to expose four different calls to the software 2 running in TEE 8, 10, namely: timer_start, timer_stop, time_report and timer_reset. In the embodiment of
An embodiment of a procedure 100 to measure time is described by the diagrams in
In the embodiment of procedure 100, at step 101 an enclave process can start, e.g., code is executed in or by the TEE, such as TEE 8 of
In the embodiment of procedure 100, if a timer_stop request is received by the firmware at step 110, a check can be performed to verify if the timer is running at step 112. If the timer is not running, the enclave continues to execute as normal, whether normal execution is active execution of code or not. If the timer is running at step 112, then in step 114 the total elapsed time of the enclave or timer and the total elapsed time of the hardware clock, e.g., HW Timer 24 of
If a timer_start request is received at step 116, a check can be performed at step 118 to verify if the timer is running. If the timer is running, a notification can be sent that the timer is running. If the timer is not running at step 118, the CPU start time and enclave or timer start time fields can be updated, and the running value in the table of
If a timer_report request is received at step 122, the total elapsed time of the enclave or timer and the total elapsed time of the hardware clock can be computed in step 124, and the running value of
If a timer_reset request is received at step 126, the total elapsed time of the enclave or timer and the total elapsed time of the hardware clock can be set to 0 in step 128, and the counters for interruption causes similarly reset. The enclave then proceeds to execute as normal.
If an interruption is detected at any point, as in step 130, a check to verify if the timer is running is performed in step 132. If the timer is not running, the enclave can proceed to execute as normal. If the timer is running in step 132, the total elapsed time of the enclave or timer can be computed in step 134, and the counter for interruption causes can be updated. Once execution of the enclave resumes, the start time of the enclave or timer can be updated in step 136.
An enclave, e.g., TEE 8 is not limited to use one single timer. TEE 8 it could use as many as the firmware 22 allows, as it might prefer to use the available memory for other purposes. For example, in the embodiment of
Note that some registers that hold the time of the system can be accessed speculatively or out of order. Embodiments of the present disclosure can avoid this scenario by placing a fence instruction, or a constraint on the order of memory operations issued before or after the instruction, before actually storing the timer value in the memory region, e.g., memory region 36, associated to the enclave, e.g., TEE 8.
Since counters, e.g., hardware counters, usually expose an interface for obtaining or modifying the actual count (e.g., if the system needs to add timestamps to the files and manages to get a reference from a trusted party) and that service could be actually used, embodiments of the present disclosure do not limit this ability. As described before, embodiments can do so by just updating the values on the structure in a procedure similar to interruption or exception handling. For example, this can be similar to internal processes performed by the firmware 22 each time there is an interruption or exception, wherein the TEE does not need to call the update, but firmware 22 will register the interruption or exception to obtain accurate timing measurements.
In order to prevent enclaves, e.g., TEE 8, 10, from abusing the timing measurements to carry out microarchitectural attacks, embodiments of the present disclosure can let the firmware 22 decide when it deems that an enclave is behaving maliciously, such as by requesting a timing measurement each few cycles. In embodiments where the enclave is deemed to behave maliciously, and under the assumption that the enclave is not running a critical piece of code like the break control system on an autonomous car, it could deny the access to the enclave. In an embodiment, the firmware 22 can set a threshold amount of requests per a certain amount of cycles, e.g., by setting the threshold amount depending on a specific type of behavior to be identified, and identify malicious behavior by comparing the amount of requests received to the threshold number of requests. For example, if the amount of requests received exceeds the threshold amount, the firmware 22 can determine that the enclave is acting maliciously. In an embodiment, the firmware 22 can acquire information from another source, e.g., hardware performance counters, and correlate that information with the requests or amounts thereof. For example, in a cache attack, if multiple requests are occurring simultaneously with cache misses, the firmware 22 can determine that the enclave is acting maliciously as it is likely that the enclave is measuring and generating cache misses as it happens.
Implementation for Different Architectures
Different architectures have different ways to expose time measurements and also have different requirements for the implementation of the trusted functions. Thus, the present disclosure provides three embodiments, among the various embodiments, to cover three well-known architectures, namely x86, ARM and Risc-V. For example ARM exposes the EL1 physical timer through the CNTPCT_EL0 register. Intel (x86) includes a high resolution timestamp counter accessible through the RDTSC instruction and Risc-V platforms have to provide a real-time counter exposed as a memory register called MTIME to machine mode software. As a person of ordinary skill in the art would understand, the embodiments may be combined and/or adapted to other architectures without departing from the spirit or scope of the present disclosure.
X86
As of now, Intel SGX is the most used platform that supports enclaves. SGX used to offer access to a low resolution timer to the enclaves, although such feature is now deprecated due to some implementation issues. Further, the procedure SGX was using was shown to be vulnerable to delay attacks, since the service was provided by an architectural enclave, which is a permanent enclave of the platform, and required inter-enclave communication which was only possible through the OS. An alternative for enclaves that wished to use timers was to implement a counting thread, but since it can be interrupted by the OS, it provides no guarantee at all.
SGX already stores some information for each of the enclaves on its private memory, like the assigned pages to each enclave and where are these pages placed on memory. According to the present disclosure, fields are added to that information to keep track of the time. Differently to their design, embodiments of the present disclosure may not use an architectural enclave to offer the timing information to the rest of the enclaves; instead, the most privileged firmware is modified so it can be directly called by the enclaves, effectively avoiding IPC and delays introduced by the OS. Further, in this case, besides tracking interruptions and exception this invention also tracks time for the ocalls executed by the enclave and the report is signed with the SGX key. The firmware runs either at system management mode (SMM) or as management engine (ME) mode, the most privileged mode. Both modes are more privileged than the OS or hypervisor and therefore the code running on such modes is trusted.
RISC-V
Risc-V based TEEs usually provide memory isolation to the enclaves by using some physical memory protection (PMP) registers that hold information referring to the memory regions that are accessible to that particular enclave and their access privileges (read, write, execute). In this case there are also different privilege modes for the code, the most privileged one is known as machine mode. The security monitor is the entity in charge of configuring the PMP registers and providing all the guarantees to the enclaves and it is therefore trusted. As such it is the only code with access to the mtime register, which in turn holds the timing information.
Every Risc-v based machine implements some sort of security monitor because it has complete access to the hardware and the only mode capable of configuring it. According to an aspect of the present disclosure, the code to support the aforementioned system calls is added to the security monitor, and the security monitor in turn uses the information from the mtime register for the measurements.
ARM
ARM Trustzone distinguishes two different worlds, the secure world where the code runs in trusted execution environments (TEEs) and the normal world or rich execution environment. A monitor running in so-called monitor mode takes care of switching between the two modes. A problem with Trustzone is that it does not have a dedicated timer for the secure world, so the OS running in the TEE could modify the timer even for legitimate purpose. Therefore, according to an aspect of the present disclosure, each modification is tracked so it is possible to undo them to measure the actual elapsed time.
Implementations for Trustzone are similar to implementations for Risc-V. The monitor can be modified to support the exposed system calls and can register any interruption or alteration of the execution of the enclave on the area reserved to that end.
Persistent Storage
For storing timing measurements persistently, an enclave partially relies on the OS, because it is in charge of handling the devices to which the data is written to, e.g., a hard drive. The time measurements that are stored persistently might comprise multiple measurements of the enclave over a period of time. The confidentially and also the integrity of the time measurements can be protected by encrypting them before handing them over to the OS for writing them to disk. However, although the firmware and also the enclave both have a sense of time, they cannot determine the order in which time measurements were written to disk. In particular, the enclave cannot infer whether an encrypted time measurement provided by the OS to the enclave is the latest one that the enclave previously requested to be written to disk. This opens the possibility for a malicious OS to mount so-called rollback attacks by proving stale state information to enclaves.
To prevent the OS for providing stale time measurements, the firmware may have access to some private and persistent storage. The storage size may be limited, however. Note that a Trusted Platform Module (TPM) already provides such limited storage capabilities. However, the access to TPM storage can be slow and the storage can wear out quickly. Furthermore, there can be a significant overhead when communicating with a TPM. For performance reasons, the firmware has its own persistent storage, which it controls and which it can access directly.
Instead of using the same key for encrypting timing measurements, the key is altered to infer in which order timing measurements were stored. A base key, which is a combination of the machine key, is provided by the underlying hardware, and the identity of the enclave is known. The firmware or the underlying hardware derives an encryption key from the enclave state (e.g. a hash of the data to be stored), and the previous key (if there is no previous key, its value is set to the base key). This newly derived key is used by the firmware to encrypt the current time measurements. The firmware stores the current key in its persistent storage once the OS has written the encrypted time measurements to disk. This allows the firmware later to decrypt the latest encrypted time measurements. It also prevents the OS from providing stale data. If it tries to do so, the data will not be decrypted by the firmware as the current key is not the key that was used to encrypt this data. A different approach would be to always use the same key and store a freshness tag that summarizes the execution history of the enclave as they do in Parno et al., “Memoir: Practical State Continuity for Protected Modules,” IEEE Symposium on Security and Privacy (2011) (the entire contents of which is hereby incorporated by reference herein). In such approach, data is decrypted and the tag is compared to deem the data as valid. An example embodiment of this approach is also provided as pseudocode:
Embodiments of the present disclosure are not limited to the storage of time measurements persistently on any piece of hardware controlled by an untrusted OS. Indeed, embodiments can be used to store any data the enclave wants to store persistently, i.e., it can be used to seal enclave data in general to any kind of persistent storage, e.g., flash memory, hard disk drive, solid state drive. Embodiments can also be used to track or assist with any regular process running at user mode, e.g., any process that could be completely under the control of the OS. In that case, the protection could be against other unprivileged processes, even if vulnerable to malicious attacks from the same or higher privilege as the OS, including the OS itself.
While subject matter of the present disclosure has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. Any statement made herein characterizing the invention is also to be considered illustrative or exemplary and not restrictive as the invention is defined by the claims. It will be understood that changes and modifications may be made, by those of ordinary skill in the art, within the scope of the following claims, which may include any combination of features from different embodiments described above.
The terms used in the claims should be construed to have the broadest reasonable interpretation consistent with the foregoing description. For example, the use of the article “a” or “the” in introducing an element should not be interpreted as being exclusive of a plurality of elements. Likewise, the recitation of “or” should be interpreted as being inclusive, such that the recitation of “A or B” is not exclusive of “A and B,” unless it is clear from the context or the foregoing description that only one of A and B is intended. Further, the recitation of “at least one of A, B and C” should be interpreted as one or more of a group of elements consisting of A, B and C, and should not be interpreted as requiring at least one of each of the listed elements A, B and C, regardless of whether A, B and C are related as categories or otherwise. Moreover, the recitation of “A, B and/or C” or “at least one of A, B or C” should be interpreted as including any singular entity from the listed elements, e.g., A, any subset from the listed elements, e.g., A and B, or the entire list of elements A, B and C.
Priority is claimed to U.S. Provisional Patent Application No. 63/347,033, filed on May 31, 2022, the entire disclosure of which is hereby incorporated by reference herein.
Number | Date | Country | |
---|---|---|---|
63347033 | May 2022 | US |