Overview of Spectre and Meltdown Vulnerabilities

Cyber security researcher have highlighted two processor level flaws named as Spectre and Meltdown. Almost all operating systems such as Windows, Linux, Android, MacOS and iOS are affected by these vulnerabilities. The researchers have proposed that these vulnerabilities are occurring due to a feature named as Speculative execution which is present in most of the processors used nowadays.

According to Apple, the vulnerabilities cannot exploit data if a malicious software is not present in the affected system. According to Project Zero run by Google, the attackers needs to have physical access to the device before running the vulnerabilities. Till now, most of the companies have denied any allegation that these vulnerabilities have been used to extrapolate sensitive information from consumer devices but Project Zero had already shown a working example of the attack which was used to cripple an entire server network. Apple has accepted that out of the two attacks meltdown has more potential to cause damage. As many devices do not support updates any more, this puts a lot of people at risk immediately. According to a security blog by Google, the devices with latest security updates are safe from this vulnerability (Gruss et al. 2016). Both of the vulnerabilities use speculative execution but the main difference between the two is that where Meltdown exploits Intel privilege escalation, Spectre exploits two processes in combination namely Branch Prediction and Speculative Execution.

In the following report, the details about Spectre and Meltdown has been discussed in details and possible mitigation techniques have been proposed.

The aim of the report is written as follows:-

• Analyse the threats from Spectre and Meltdown
• Evaluate the security techniques and policies that are implemented for combating the vulnerabilities
• Predicting the future impact of the vulnerabilities
• Recommend counter measures to Meltdown and Spectre

The scope of the report is to provide a detailed analysis to researchers for understanding the threats and create proper preventive measures to combat these issues. 

About Spectre

The vulnerability named Spectre utilizes random locations in the memory space of the program to trick other programs.  The contents of the accessed memory space can be looked up by the potential attacker to gather sensitive data. The spectre is a list of vulnerabilities instead of a single vulnerability and are related to the speculative execution exploit. To be precise, Spectre runs on a special case of speculative execution known as branch prediction (Hruska 2018). It does not rely on the memory management of a single processor which makes it different form the other vulnerability called Meltdown. The initiation point of the attack is triggered by a side channel timing attack in the modern microprocessors and utilizes its branch prediction machinery. Even after the vulnerability is mitigated, the side effects of the speculative execution can leave side effects such as loaded cache lines. The non-functional elements of the computing environment can be affected due to this.

Spectre Vulnerability

To understand about spectre, the working of a microprocessor needs to be understood. Suppose a simple program is present which adds two variables x and y. In order to execute the processor has to perform four basic things.  First it needs to load the value of x from the main memory into the processor (here into a register called R1). Same thing happens to variable y which goes into register R2. The processor then adds the two values together and stores the result in R1. Then the computer needs to simply store R1 back to the main memory to complete the program (Lipp et al. 2016). As the processor is really fast and the main memory is slow, processor designers add a number of features to hide this discrepancies.

One feature is known as cache and the other feature is known as speculation and Spectre uses this two features in combination to leak secret data to people who are not supposed to have it. Caches are small memories that are much faster than main memory which can be used to store frequently used values. The values of X and Y can be updated in the cache so the need to write back the values in the slow main memory can be avoided which results in faster execution of the program. This difference in access speeds is exploited by Spectre vulnerability to track data (Trippel, Lustig and Martonosi  2018).

The Spectre vulnerability is based on two major flaws. First, the logic of branch prediction can be trained to hit a program’s internal workings. Second, it shows that a non-functional difference can be moved to a covert channel which can collect personal information from the inner workings of a processor.

The spectre vulnerability is remotely accessible. All the user has to do is to visit the website containing the spectre code and the leaking of sensitive information starts instantly from the browser such as passwords and cookies. To make matters worse, it is not a software vulnerability but a hardware bug and no software workarounds can be dispatched to erase it. Intel has released a micro code to reduce exposure to the bug but it does not eliminate it entirely. The bug relies on caching and speculation as the primary means to extract data and the problem is that this feature has been present in most processors that are used in performance sensitive systems (laptops, smartphones and servers in clouds). This makes all the devices a potential candidate for the vulnerability (Theregister.co.uk 2018). Without entirely removing speculation or caching (which can cause systems to slow down), it is still unclear if the vulnerability can be removed at all.

Meltdown Vulnerability

About Meltdown

Meltdown specifically attacks the Intel x86 Microprocessors, ARM Microprocessors and IBM processors. It breaks the fundamental wall between the operating system and user applications. It allows a program to access memory which is extremely dangerous. The vulnerability is hardware based and gives permission to a rogue process to read memory without authentication. AMD processors are not affected by this vulnerability. Any Intel processors that were made in the last 10 years are potentially vulnerable to the attack. The vulnerability exploits a race condition of the CPU that happens between privilege checking and instruction execution (Griffin 2018). Before the privilege check can occur, it reads the unauthorised mapped data in a normal way.

Figure 1: How Meltdown works

(Source:  Created by the author)

Meltdown convinces the processor to load some secret data. The processor is eventually going to block this access and not allow the attacking process to see the results explicitly in registers or in memory that’s under the attacker’s control. However in step 2 of the attack the attacker convinces the processor to index the array that is under the attacker control using the data that was loaded in step 1 and then the attacker observes the timing differences between the accesses to this array in order to extract the secret data even though the processor didn’t explicitly load it. Step 2 and 3 are known as cache side channel attack.  Meltdown specifically uses flush reload side channel attack (Kocher et al. 2018). The way this work is that first the vulnerability clears the processor cache through flush operation. The next step is to trick the processor into loading a particular element of an array under the attackers control using secret data. This access causes the processor to load into the cache the element of the rogue array (O’Donnell et al. 2018). The attacker then observes the array and observes the amount of time to make each access very reliably. Step 1 is very important to Meltdown as it gives the vulnerability access to the kernel.

The kernel is the heart of the operating system and controls the hardware by authenticating who gets access to which memory slot.  Meltdown computes the kernel address and asks the processor to load it through speculative process in a register (Fruhlinger 2018). Before loading the kernel data in the register from the main memory, the data is saved temporarily in the cache memory present in the processor. This data is then exploited by the attacker to get the access time of kernel data without any privileges. The kernel data leakage contains sensitive information such as contents of files that were recently read. The vulnerability creates a virtual environment inside the physical memory and the attacker gets to move exploit data whenever he wants (Simakov et al. 2018). Therefore, any program which is running in the user context can force the CPU to collect data in the caches through speculative execution and use it as an Oracle database to trickle Kernel data by bypassing the operating system’s standard controls.

Preventive Measures and Mitigation Techniques

Counter measures to Spectre and Meltdown 

As the vulnerability is fundamentally present at the hardware level, a complete patch is not possible. But security patches that work around the vulnerability are released by several vendors like Apple, Microsoft and Google. For Linux operating system, the KAISER patch that was incidentally released in 2017 prevents the Meltdown vulnerability but provides little security for Spectre. Cloud servers are being patched to protect them from the vulnerabilities. Rendition Infosec has provided a strategy for organizations about how they can protect their systems from potential damage from the vulnerabilities (Gras et al. 2014). Keeping the browsers updated is essential as Spectre normally uses JavaScript to conduct its malicious activities. Protection of older operating systems such as Windows XP are not possible.  Similarly older android operating systems or budget phones are exempted from getting any software updates for the issue (Support.microsoft.com 2018).

Microsoft has released several system patches for its operating systems like Windows 7 and up. It also patches the edge and explorer browsers. Firmware updates have been provided as well for all the major processor manufacturers including AMD.  Some patches were not working with AMD so they have been temporarily removed. The patches were released from 11^th January (Benger et al. 2014). 

On January 3^rd, Apple released similar patches for its operating systems namely MacOS, TvOS and iOS as well as its Safari browsers (Meltdownattack.com 2018).

Although ChromeOS is not affected as much as other operating systems, Google has released patches for some of its chromebooks too.

Firefox also has a patch ready for its browser on January 23^rd whose beta version can be downloaded now.

To mitigate Meltdown, the memory management between the operating system and the application software needs to be fundamentally changed.  Kernel page table isolation or KPTI is a technology which makes sure that secure data cannot be loaded in the internal caches of a microchip when the user code is running.  KPTI takes some extra steps when an application software asks the operating system to do something for the software. Apple has also released mitigation updates in iOS 13.2, 11.2 and 10. Microsoft has released software patches that are incompatible with antivirus programmes that use kernels calls which are unsupported. As Meltdown involves no software vulnerability, mist of the software patches will leave a little amount of memory exposed. It can be prevented by serializing the register fetch and permission check (Lipp et al. 2018). This process however imposes a lot of overhead to memory addresses stalling the fetching process sometimes.

Operating System Specific Mitigations

A better mitigation process would be to provide a hard split between the kernel space and the user space. Introducing a new kernel bit in the control register of the CPU can be enabled by modern kernels. The kernel has to stay in the upper part of the address space when the hard split bit is set. This will enable the system to identify an unauthorized memory fetch which is violating the security boundary of the privilege level (Azab et al. 2014). The performance impact is expected to be minimal (around 10 percent). Another countermeasure that can be imposed for Meltdown is introducing KAISER (a modified kernel where the kernel is situated outside the user space). It prevent Meltdown by not providing any valid mapping space to physical memory kernel space (Leonhard 2018). The modification will be available to all Linux based system under the alias of page table isolation. Similar patches have been discharged for other operating systems. It has some limitations but at least it can prevent attacks by not allowing the attacker to have any memory locations or kernel pointers in the user space that can leak sensitive information.

Spectre has two types: Variant 1 and Variant 2.  To mitigate Variant 1, load fences are imposed around the kernel. This prevents the speculation programme from loading a second load when a first load is already performed (HPE 2018). The mitigation technique requires putting small and minimum performance impacting changes in the source of the kernel. To mitigate Variant 2 of Spectre, the hardware related to branch predicting needs to be trained to understand which code should be favoured over what for execution. The operating system can even deactivate the branch predicting hardware when the operating system is asked by a program to attempt malicious activities. This approach is quite reliable but the performance of the system suffers. The patches that have been implemented by various vendors needs to be downloaded in the required systems even at the cost of performance issues and for organizations, a system administrator needs to be present who can toggle the patches on and off (Watson et al. 2018). Intel announced in January, 2018 that they will ship new processors which are not vulnerable to Spectre as well as Meltdown. The variant 1 of spectre will be mitigated with software changes while the variant 2 will be mitigated with hardware changes. Intel announced that they have redesigned certain part of the processors and introduced new security levels to prevent both variants of the Spectre (Gibbs 2018). Micro codes which are software based has been launched for all Intel products that has been launched in the last five years. 

Conclusion

Permanent counter measure for Spectre is still not available yet.

Future Impacts of Spectre and Meltdown

Cloud providers will be severely impacted due to Spectre rather than Meltdown Although Meltdown uses unauthorised programs to obtain personal data from physical memory processes on the cloud, Spectre utilizes a programs to send data to a guest system by inducing a hypervisor.

In the future, the impacts of Spectre and Meltdown will result in more hardware based attacks rather than software based attacks. Last year, in Intel’s remote administrative feature named Management Engine a new vulnerability was discovered (Gens et al. 2017). They affected the chips just like Spectre and Meltdown. As these vulnerabilities are comparatively new, criminals, intelligence agencies and security researchers will be on the hunt to research this new area.

Secondly, releasing patches for microprocessors require a coordinated effort of all the major companies working in this area. Although patches have been released by AMD and Intel, it will be a while before application vendors and computer manufacturers customize the patch to make it functional for the user (Design News 2018). This makes it very difficult for the vulnerabilities to remain a secret before appropriate patches are pushed out. As Spectre and Meltdown were announced early, it gives hackers ample time to attack the system before they are properly secured.

Thirdly, the patches for Spectre and Meltdown will affect the functionality of computers. It was previously estimated that the patches will reduce the system performance by as much as 30% but that was later proved to be true for only cloud based systems (Payer 2016). Still, as more vulnerabilities are discovered, further hardware patches will definitely affect the core performance of the systems.

Moreover, these vulnerabilities will not only infect the computers but also the global infrastructure, medical devices, appliances, cars and smartphones. Information from the infrastructure of cloud providers can be compromised to steal customer data. Personal pictures, emails, critical documents and passwords can be stolen by exploiting these vulnerabilities. These vulnerabilities will also impact future designs of microprocessors. Blockchain technologies will be used widely in the future. Businesses will suffer as most of them which handle sensitive information have to upgrade their entire system increasing the total cost of the implementation (Engadget 2018). The appreciation and understanding for the requirement of a secure system will change as researchers are pretty much sure that speculative execution will be exploited further in the near future. More side channels attacks can be expected as well. 

Conclusions 

To conclude the report, it can be stated that the research results have fundamentally changed how software and hardware will be designed 10 to 20 years from now as the new vulnerabilities of the leakage of data through the cache side channels have been discovered. The implications of Spectre and Meltdown will have serious repercussions in the coming years for all PC users. The mitigation strategies will even impact the performance of the machines depending on the device’s hardware and workload specifications. Desktop users are even more vulnerable to these attacks as the vulnerabilities use browser plugins to get access to the user’s personal information with the help of JavaScript. Moreover, as the vulnerabilities are very complex in nature, researchers may be able to find new exploits in the future that are not covered in the current mitigation strategies. Researchers to have to be quick to find the exploits properly and assign the same project to different teams so that the vulnerabilities are confirmed independently. As speculative execution which is hardware specific cannot be deactivated, the vulnerabilities may never get patched totally. But still researchers are working hard to change the entire operating systems so the kernel memory is rendered invisible from user contexts. This will result in the kernel working harder for the system which will definitely impact the performance. Some vendors are even considering implementing BIOS modifications to decrease the risk of exposure. As per the mitigation strategies proposed in the report, the short term strategies are to patch the operating systems s ell as the firmware of all devices which are vulnerable to these types of attacks. These vulnerabilities cannot be misused through a network. To steal kernel data, the attacker needs to run the code in his own machine.  Cloud providers such as Microsoft Azure and Amazon AWS are even more vulnerable to these attacks as the vulnerabilities can be executed remotely from one user to another. More research have to be considered in the future to understand the extent and vectors of these vulnerabilities. 

References 

Azab, A.M., Ning, P., Shah, J., Chen, Q., Bhutkar, R., Ganesh, G., Ma, J. and Shen, W., 2014, November. Hypervision across worlds: Real-time kernel protection from the arm trustzone secure world. In Proceedings of the 2014 ACM SIGSAC Conference on Computer and Communications Security (pp. 90-102). ACM.

Benger, N., Van de Pol, J., Smart, N.P. and Yarom, Y., 2014, September. “Ooh Aah… Just a Little Bit”: A small amount of side channel can go a long way. In International Workshop on Cryptographic Hardware and Embedded Systems (pp. 75-92). Springer, Berlin, Heidelberg.

Design News. (2018). Chipmakers Discuss a Future After Meltdown and Spectre. [online] Available at: https://www.designnews.com/content/chipmakers-discuss-future-after-meltdown-and-spectre/42684598058203 [Accessed 17 Mar. 2018].

Engadget. (2018). Intel currently facing 32 class-action lawsuits for Spectre and Meltdown. [online] Available at: https://www.engadget.com/2018/02/16/intel-face-32-lawsuits-spectre-meltdown/ [Accessed 17 Mar. 2018].

Fruhlinger, J. (2018). Spectre and Meltdown explained: What they are, how they work, what’s at risk. [online] CSO Online. Available at: https://www.csoonline.com/article/3247868/vulnerabilities/spectre-and-meltdown-explained-what-they-are-how-they-work-whats-at-risk.html [Accessed 17 Mar. 2018].

Gens, D., Arias, O., Sullivan, D., Liebchen, C., Jin, Y. and Sadeghi, A.R., 2017, September. LAZARUS: Practical Side-Channel Resilient Kernel-Space Randomization. In International Symposium on Research in Attacks, Intrusions, and Defenses (pp. 238-258). Springer, Cham.

Gras, B., Razavi, K., Bosman, E., Bos, H. and Giuffrida, C., 2017. ASLR on the Line: Practical Cache Attacks on the MMU. NDSS (Feb. 2017).

Griffin, C. (2018). The Latest: Spectre And Meltdown | CRN. [online] CRN. Available at: https://www.crn.com/spectre-meltdown [Accessed 18 Mar. 2018].

Gruss, D., Maurice, C., Fogh, A., Lipp, M. and Mangard, S., 2016, October. Prefetch side-channel attacks: Bypassing SMAP and kernel ASLR. In Proceedings of the 2016 ACM SIGSAC conference on computer and communications security (pp. 368-379). ACM.

HPE. (2018). Coping with Spectre and Meltdown: What sysadmins are doing. [online] Available at: https://www.hpe.com/us/en/insights/articles/coping-with-spectre-and-meltdown-what-sysadmins-are-doing-1802.html [Accessed 17 Mar. 2018].

Hruska, J. (2018). Intel Didn’t Disclose Spectre, Meltdown to US Government Until News Went Public – ExtremeTech. [online] ExtremeTech. Available at: https://www.extremetech.com/computing/264490-intel-didnt-disclose-spectre-meltdown-us-government-news-went-public [Accessed 17 Mar. 2018].

Information Management. (2018). 6 steps firms can take to mitigate Spectre and Meltdown risks. [online] Available at: https://www.information-management.com/slideshow/6-steps-firms-can-take-to-defend-against-spectre-and-meltdown?slide=3 [Accessed 16 Mar. 2018].

Kocher, P., Genkin, D., Gruss, D., Haas, W., Hamburg, M., Lipp, M., Mangard, S., Prescher, T., Schwarz, M. and Yarom, Y., 2018. Spectre Attacks: Exploiting Speculative Execution. arXiv preprint arXiv:1801.01203.

Leonhard, W. (2018). Intel releases more Meltdown/Spectre fixes, Microsoft feints SP3 patch. [online] Computerworld. Available at: https://www.computerworld.com/article/3257225/microsoft-windows/intel-releases-more-meltdownspectre-firmware-fixes-microsoft-feints-an-sp3-patch.html [Accessed 16 Mar. 2018].

Lipp, M., Gruss, D., Spreitzer, R., Maurice, C. and Mangard, S., 2016, August. ARMageddon: Cache Attacks on Mobile Devices. In USENIX Security Symposium (pp. 549-564).

Lipp, M., Schwarz, M., Gruss, D., Prescher, T., Haas, W., Mangard, S., Kocher, P., Genkin, D., Yarom, Y. and Hamburg, M., 2018. Meltdown. arXiv preprint arXiv:1801.01207.

Meltdownattack.com. (2018). Meltdown and Spectre. [online] Available at: https://meltdownattack.com/ [Accessed 18 Mar. 2018].

O’Donnell, L., O’Donnell, L., Spring, T. and O’Donnell, L. (2018). Intel Details CPU ‘Virtual Fences’ Fix As Safeguard Against Spectre, Meltdown Flaws. [online] Threatpost | The first stop for security news. Available at: https://threatpost.com/intel-details-cpu-virtual-fences-fix-as-safeguard-against-spectre-meltdown-flaws/130501/ [Accessed 18 Mar. 2018].

Payer, M., 2016, April. HexPADS: a platform to detect “stealth” attacks. In International Symposium on Engineering Secure Software and Systems (pp. 138-154). Springer, Cham.

Simakov, N.A., Innus, M.D., Jones, M.D., White, J.P., Gallo, S.M., DeLeon, R.L. and Furlani, T.R., 2018. Effect of Meltdown and Spectre Patches on the Performance of HPC Applications. arXiv preprint arXiv:1801.04329.

Support.microsoft.com. (2018). [online] Available at: https://support.microsoft.com/en-in/help/4073757/protect-your-windows-devices-against-spectre-meltdown [Accessed 16 Mar. 2018].

Theregister.co.uk. (2018). Intel didn’t tell CERTS, govs, about Meltdown and Spectre because they couldn’t help fix it. [online] Available at: https://www.theregister.co.uk/2018/02/23/meltdown_spectre_letters_to_congress/ [Accessed 16 Mar. 2018].

Trippel, C., Lustig, D. and Martonosi, M., 2018. MeltdownPrime and SpectrePrime: Automatically-Synthesized Attacks Exploiting Invalidation-Based Coherence Protocols. arXiv preprint arXiv:1802.03802.

Watson, R.N., Woodruff, J., Roe, M., Moore, S.W. and Neumann, P.G., 2018. Capability Hardware Enhanced RISC Instructions (CHERI): Notes on the Meltdown and Spectre Attacks (No. UCAM-CL-TR-916). University of Cambridge, Computer Laboratory