| description |
|---|
Building a secure Hackintosh |
Let's be honest: most x86 systems are garbage when platform security is concerned. Intel does not have the best track record with building secure chips and AMD doesn't do much better. That's why Apple has recently been improving their boot and platform security with the T2 chip. The common belief is that Hackintoshes cannot be as secure as a Mac due to various patches and that it is miles behind a Mac with T2. But how true is that belief?
In this article, we will explore the boot and platform security of Intel x86, the Hades Canyon NUC, and Macs. We will compare the security offerings on these similar platforms and present an argument that we can make the Hades Canyon Hackintosh as secure as most Macs.
Before starting, it is important to clarify what part of the security architecture is being discussed here. Specifically, we mainly focus on boot-time security from remote attackers. Specifically, we wish to protect against an attack where a compromised OS (OSX/Windows/Linux) installs a "bootkit" that persists across a reboot and across a fresh re-install (with a wiped disk). We do not consider the security of OSX vs Windows. Finally, we do not consider the security of a local attacker with physical access to the device. Note that the T2 does attempt to protect against local attackers, but this is a much more niche issue.
The first link in the secure boot chain is owned solely by Intel on every motherboard with an Intel chipset. When your computer first powers up, a small 32-bit x86 CPU that is in the Platform Controller Hub (PCH or chipset) called the Intel Management Engine (ME or CSME) starts running from its unpatchable on-die boot ROM. First it reads the Serial Flash Discoverable Parameters (SFDP) from the SPI flash chip connected to the PCH to determine how to read from the flash, and it then reads the Intel Flash Descriptor (IFD) from offset 0 to locate the ME region of the flash.
The ME boot ROM code then loads the ME firmware partion table (FPT) from the region offset in the SPI flash chip into the ME's on-die SRAM. The FPT is not signed, which allows the various partitons to be relocated around the flash. The ME finds the Factory Partition (FTPR) in the table and copies the signed header from this partion from the flash into SRAM and validates the signature on the partition against a Intel's public key, whose hash is fixed in the on-die ROM. If these do not match, then the ME shuts down and the computer won't boot.
If the FTPR signature matches, then the ME finds the Bring Up (BUP) module in the FTPR and copies it into SRAM. It compares the hash of the module with the hash stored in the signed FTPR, and if these match then it starts executing the code in BUP, which transitions away from the on-die code. However, the on-die code remains resident, similar to a shared library of functions for the modules stored in the flash. Also note that the other modules in the FTPR are not validated at this time, allowing limited editing of the contents of the ME's flash partitions to remove functionality.
The ME's BUP module is what begins booting the real CPU in the system by releasing it from reset. At reset time, the last few megabytes of the BIOS region of the SPI flash are mapped into the top 4GB of memory, so reads to the addresses from 0xFF000000 to 0xFFFFFFFF are routed to the PCH and turned into reads from the SPI flash. The first thing the CPU does is locate the Firmware Interface Table (FIT) pointer, stored at 0xFFFFFFC0. The FIT, which is not signed directly, contains pointers to the microcode updates for various CPUs, as well as pointers to the reset vector and BIOS entry points.
The CPU's internal boot ROM walks the FIT to find the microcode updates and reads their headers to determine if they apply to this CPU, and if so the microcode update is copied linearly into cache and Intel's signatures are validated by the boot ROM. If this passes, then they are decrypted with an Intel key that is also in the boot ROM and the update is applied to the boot CPU.
The boot ROM then goes back to the FIT to find the Startup Authenticated Code Module (ACM) in the SPI flash chip. This is copied into cache by the boot ROM and Intel's signature on the ACM is validated by the boot ROM before executing it in 32-bit non-evict (aka Cache-as-RAM) based on the GDT stored in the ACM. The ACM is signed with an Intel owned key and the hash of the public key to verify the signature is inside the boot ROM, although the signature checks are only enforced if the OEM has set the Bootguard configuration fuses in the PCH to do so. It is especially dangerous if the configuration is left unfused in the PCH since malware could then fuse its own keys and the system would not boot without the malware in the SPI flash.
The Startup ACM contacts the ME to receive the Bootguard OEM hash and fuse configuration. The ACM returns to the FIT to locate the Bootguard Key Manifest (KEYM), which contains an OEM public key that matches the hash provided by the ME as well as other public keys the OEM uses to sign the firmware. The OEM hash is used to by the ACM to validate that the KEYM is provided by the OEM, and the ACM then loads the Bootguard Boot Policy (ACBP) from the FIT, which is signed by the Key Manifest public key in the KEYM.
The ACBP contains the list of the Initial Boot Block (IBB) segments that are to be copied into cache as well as the overall hash of the regions and the real entry point for the legacy BIOS (typically 0xFFFFFFF0). Depending on the Bootguard fuse configuration, a hash comparison failure might lead to an immediate CPU halt, a PCR extension with the failure code, or just a global variable set in memory. If the result is not a halt, the ACM loads the legacy BIOS entry point from the ACBP and executes a GETSEC instruction to shift back into real mode and exit the ACM, jumping into the traditional x86 CPU reset vector.
Note that any vulnerability in the ME boot ROM, the x86 boot ROM, or the ACM can be used to exploit every Intel system out there including all Macs (including T2 Macs). There have been bugs found in the ME's boot ROM, TOCTOU attacks against the ACM have been successful, and many vendors have made mistakes with Bootguard configurations or executed code from unsigned regions of the flash.
The ME is the most important part of x86 security because not only does it have full DRAM access, it also feeds the OEM public key and security policies to the ACM on the CPU. The ACM will load the next stage called the Initial Boot Block from the SPI flash. The ACM code then will check the signature of the IBB against the OEM public key from the ME only if enabled. Note that many/most OEMs do not enable this and therefore ANY IBB can run if flashed into the SPI flash. On the Hades Canyon, only an Intel signed IBB can be executed (and this cannot be disabled). On T2 Macs, Apple does additional verification of the IBB and won't even attempt to feed it to the Intel CPU if it fails. Although there is a dearth of details in Apple's whitepaper, presumably this allows for previous IBB to be revoked while on Intel-only systems like the NUC any vulnerable IBB can be downgraded to by an attacker with physical SPI access.
This whole process is referred to as Intel Boot Guard by Intel and is only found on relatively recent (Haswell+) systems. There is also an adjacent technology called Intel BIOS Guard which can prevent unauthorized BIOS updates from being flashed if Boot Guard is disabled. In systems with Boot Guard is enabled, BIOS Guard can also be used to prevent downgrades to an vulnerable IBB. However, this is strictly worse than T2 because BIOS Guard requires the integrity of the boot chain up until the OS is booted while the T2 can assume that everything down to the IBB is compromised. The Hades Canyon has both Boot Guard and BIOS Guard enabled, which, from a non-scientific survey of Google search results seems to be quite rare.
This is a very high-level overview of the first part of the boot chain. Primary references used for this section are as followed:
- Bypassing Intel Boot Guard
- Safeguarding rootkits: Intel BIOS Guard part 2
- Betraying the BIOS
- Intel x86 Root of Trust: loss of trust
- ME Cleaner: How does it work
- Intel ME: Flash file system explained
- Intel Management Engine deep dive
- TOCTOU attacks on Bootguard
The IBB is also the start of the UEFI firmware on UEFI systems (such as the Hades Canyon). Specifically, it is the Security (SEC) Phase and the Pre EFI Initalization (PEI) Phase. The IBB is (if enabled by the OEM) verified by the ACM with the OEM key passed from the ME. If that mess of acronym soup is nonsense, the section above attempts to explain it in more detail.
The SEC phase sets up some security policies and the PEI phase does the initial setup to bring the Intel CPU to a usable state. The Driver Execution Environment (DXE) phase is loaded from the SPI flash which sets up many more devices and brings the computer to a usable state. At this point, the display turns on and you see the OEM logo. More details can be found in the PI Boot Flow wiki page.
The PEI modules checks the signatures of the DXE modules (if such security is even enabled). Most "BIOS" vulnerability we know of are in this phase. The latest NUC Hades Canyon BIOS is free of known vulnerabilities and configuration issues. You can check your system with the chipsec framework.
The DXE modules sets up the boot device (NVMe for example) and then loads the boot manager from the selected boot device. This is usually /EFI/BOOT/BOOTx64.efi. When Secure Boot is enabled from the BIOS, the .efi must either contain a signature verified from a public key in the platform DB or the hash of the .efi itself is in the platform DB.
The DB is a database of allowed public keys and hashes stored in the SPI flash. Along with the DBX (database of revoked keys/hashes) and the DBT (database of timestamps when the keys are valid), the databases themselves are signed with a key inside the KEK (Key Exchange Key) database (also in the SPI flash). Finally, the KEK is signed with the Platform Key. All this is explained in more detail in Diving into Secure Boot.
Why do we need this chain of certificates? The reason is ownership of the private keys in each step is different. The PK private key is owned by the owner of the system. This is either you, the end user, or the administrator if the computer belongs to an organization. The PK public key is used to verify the KEK. The KEK private keys are owned by the OS provider. For example, if the administrator wishes to permit Windows 10 to be booted on the computer, she shall put Microsoft's certificate into the KEK. Microsoft will then use their private key to sign all the different valid versions of the Windows Boot Manager, which are then placed into the DB.
Let's talk about the PK, the root key for UEFI Secure Boot. Because it is owned by the system administrator/end user, it is important that an attacker does not overwrite the system's PK or they can bypass the Secure Boot policy. The PK is stored in the NVRAM (which is usually in the same SPI chip as the BIOS) as a special variable. Write access to this UEFI variable is permitted only in "user present" context. In Hades Canyon NUC, there are two ways to change the PK: you can reset the PK with the BIOS security jumper inside the case and you can set the PK inside the Visual BIOS program by pressing F2 during boot. Access restriction to the PK variable (and the Secure Boot enable BIOS option) is done by the UEFI firmware meaning that (barring a BIOS vulnerability), you cannot change the PK from Windows/Linux/OSX.
Note that on Macs (including T2 Macs), the usage of PK/KEK/etc are disabled. There is an option to enable "secure boot" on T2 Macs which only checks Microsoft's certificates, but there is no way to add any other certificate or change the "PK" (which does not exist as Microsoft's certificate is essentially the root key). There is no way to have a "personalized" secure boot. You either trust only Apple & Microsoft or you trust everyone.
All of this so far applies to every x86 platform. Now, we will discuss security specific to Hackintoshes. OpenCore is a new bootloader designed to run OSX on non-Apple hardware. In addition to being better designed and better coded than Clover, the legacy UEFI bootloader for Hackintoshes, OpenCore introduces security options to the boot process.
In order to boot OSX on a variety of x86 hardware, OpenCore provides an interface to patch ACPI tables, kernel memory, inject additional OSX drivers/kexts, load additional UEFI DXE modules, and more. Now, OSX with System Integrity Protection (SIP) enabled protects the system from unauthorized modifications, but OpenCore is outside of SIP's control (and needs to bypass SIP to make modifications). An attacker who takes over OSX will usually have a hard time gaining persistence due to SIP, but if they discover the computer is a Hackintosh, they can just use OpenCore as a conduit for boot-persistance.
To counteract this, OpenCore provides an optional feature called the "vault" which allows you to sign a hash of every file OpenCore uses (all the patches, configuration, injected drivers, etc) and then embed the public key into the OpenCore EFI binary itself. OpenCore will then check the integrity of all the files on boot. Currently, OpenCore does not check the integrity of Apple's boot.efi for loading XNU, but I believe that is a future addition.
The administrator can then use their custom PK to sign a KEK to sign OpenCore and place the OpenCore certificate inside the DB. With Secure Boot enabled, we can then complete the chain of trust by loading a verified OpenCore (whose binary contains the vault public key) which verifies all the items in the vault before booting into a SIP enabled OSX.
This is how the final chain of trust looks for the Hades Canyon running OpenCore with UEFI Secure Boot enabled:
- Intel ME boot ROM is implicitly trusted as the root.
- ME ROM with proper fuses blown verifies ME firmware with Intel's key
- ME firmware passes security policy and OEM key to Intel CPU
- CPU boot ROM is implicitly trusted and it verifies the ACM with Intel's key
- ACM verifies IBB with OEM key from ME firmware
- IBB verifies UEFI DXE with OEM key
- The KEK is verified with the PK (owned by the administrator)
- A certificate in KEK verifies the DB
- A certificate in the DB verifies the OpenCore binary
- OpenCore verifies the vault files with its embedded public key
- OSX boots with SIP enabled and all binaries fall under Apple's signing policies
How does this compare to a T2 Mac? The main difference is that with the lack of the (locally owned) PK, there is no attack vector for a local attacker to compromise chain at that point. The second difference is that the T2's policy checks could allow for downgrade protection against a SPI re-flash attack (to an older vulnerable BIOS). The third difference is that the chain of trust is shorter for Apple because they trust less 3rd parties (OEM + system administrator), and therefore there is less room for mistakes. It is also important to point out that OpenCore + third party plugins and extensions are more likely to have security bugs as there has been less security research into those components compared to parts shared by Macs and other x86 platforms.
How does the Hades Canyon compare to other systems? It is still rare for Intel BIOS Guard and Intel Boot Guard to be enabled, and without those two the chain breaks very early on and the rest is not worth considering. You can check the status of your platform security using chipsec to see if the security features are enabled and configured properly.