1 ======================================= 2 Pointer authentication in AArch64 Linux 3 ======================================= 4 5 Author: Mark Rutland <mark.rutland@arm.com> 6 7 Date: 2017-07-19 8 9 This document briefly describes the provision of pointer authentication 10 functionality in AArch64 Linux. 11 12 13 Architecture overview 14 --------------------- 15 16 The ARMv8.3 Pointer Authentication extension adds primitives that can be 17 used to mitigate certain classes of attack where an attacker can corrupt 18 the contents of some memory (e.g. the stack). 19 20 The extension uses a Pointer Authentication Code (PAC) to determine 21 whether pointers have been modified unexpectedly. A PAC is derived from 22 a pointer, another value (such as the stack pointer), and a secret key 23 held in system registers. 24 25 The extension adds instructions to insert a valid PAC into a pointer, 26 and to verify/remove the PAC from a pointer. The PAC occupies a number 27 of high-order bits of the pointer, which varies dependent on the 28 configured virtual address size and whether pointer tagging is in use. 29 30 A subset of these instructions have been allocated from the HINT 31 encoding space. In the absence of the extension (or when disabled), 32 these instructions behave as NOPs. Applications and libraries using 33 these instructions operate correctly regardless of the presence of the 34 extension. 35 36 The extension provides five separate keys to generate PACs - two for 37 instruction addresses (APIAKey, APIBKey), two for data addresses 38 (APDAKey, APDBKey), and one for generic authentication (APGAKey). 39 40 41 Basic support 42 ------------- 43 44 When CONFIG_ARM64_PTR_AUTH is selected, and relevant HW support is 45 present, the kernel will assign random key values to each process at 46 exec*() time. The keys are shared by all threads within the process, and 47 are preserved across fork(). 48 49 Presence of address authentication functionality is advertised via 50 HWCAP_PACA, and generic authentication functionality via HWCAP_PACG. 51 52 The number of bits that the PAC occupies in a pointer is 55 minus the 53 virtual address size configured by the kernel. For example, with a 54 virtual address size of 48, the PAC is 7 bits wide. 55 56 When ARM64_PTR_AUTH_KERNEL is selected, the kernel will be compiled 57 with HINT space pointer authentication instructions protecting 58 function returns. Kernels built with this option will work on hardware 59 with or without pointer authentication support. 60 61 In addition to exec(), keys can also be reinitialized to random values 62 using the PR_PAC_RESET_KEYS prctl. A bitmask of PR_PAC_APIAKEY, 63 PR_PAC_APIBKEY, PR_PAC_APDAKEY, PR_PAC_APDBKEY and PR_PAC_APGAKEY 64 specifies which keys are to be reinitialized; specifying 0 means "all 65 keys". 66 67 68 Debugging 69 --------- 70 71 When CONFIG_ARM64_PTR_AUTH is selected, and HW support for address 72 authentication is present, the kernel will expose the position of TTBR0 73 PAC bits in the NT_ARM_PAC_MASK regset (struct user_pac_mask), which 74 userspace can acquire via PTRACE_GETREGSET. 75 76 The regset is exposed only when HWCAP_PACA is set. Separate masks are 77 exposed for data pointers and instruction pointers, as the set of PAC 78 bits can vary between the two. Note that the masks apply to TTBR0 79 addresses, and are not valid to apply to TTBR1 addresses (e.g. kernel 80 pointers). 81 82 Additionally, when CONFIG_CHECKPOINT_RESTORE is also set, the kernel 83 will expose the NT_ARM_PACA_KEYS and NT_ARM_PACG_KEYS regsets (struct 84 user_pac_address_keys and struct user_pac_generic_keys). These can be 85 used to get and set the keys for a thread. 86 87 88 Virtualization 89 -------------- 90 91 Pointer authentication is enabled in KVM guest when each virtual cpu is 92 initialised by passing flags KVM_ARM_VCPU_PTRAUTH_[ADDRESS/GENERIC] and 93 requesting these two separate cpu features to be enabled. The current KVM 94 guest implementation works by enabling both features together, so both 95 these userspace flags are checked before enabling pointer authentication. 96 The separate userspace flag will allow to have no userspace ABI changes 97 if support is added in the future to allow these two features to be 98 enabled independently of one another. 99 100 As Arm Architecture specifies that Pointer Authentication feature is 101 implemented along with the VHE feature so KVM arm64 ptrauth code relies 102 on VHE mode to be present. 103 104 Additionally, when these vcpu feature flags are not set then KVM will 105 filter out the Pointer Authentication system key registers from 106 KVM_GET/SET_REG_* ioctls and mask those features from cpufeature ID 107 register. Any attempt to use the Pointer Authentication instructions will 108 result in an UNDEFINED exception being injected into the guest. 109 110 111 Enabling and disabling keys 112 --------------------------- 113 114 The prctl PR_PAC_SET_ENABLED_KEYS allows the user program to control which 115 PAC keys are enabled in a particular task. It takes two arguments, the 116 first being a bitmask of PR_PAC_APIAKEY, PR_PAC_APIBKEY, PR_PAC_APDAKEY 117 and PR_PAC_APDBKEY specifying which keys shall be affected by this prctl, 118 and the second being a bitmask of the same bits specifying whether the key 119 should be enabled or disabled. For example:: 120 121 prctl(PR_PAC_SET_ENABLED_KEYS, 122 PR_PAC_APIAKEY | PR_PAC_APIBKEY | PR_PAC_APDAKEY | PR_PAC_APDBKEY, 123 PR_PAC_APIBKEY, 0, 0); 124 125 disables all keys except the IB key. 126 127 The main reason why this is useful is to enable a userspace ABI that uses PAC 128 instructions to sign and authenticate function pointers and other pointers 129 exposed outside of the function, while still allowing binaries conforming to 130 the ABI to interoperate with legacy binaries that do not sign or authenticate 131 pointers. 132 133 The idea is that a dynamic loader or early startup code would issue this 134 prctl very early after establishing that a process may load legacy binaries, 135 but before executing any PAC instructions. 136 137 For compatibility with previous kernel versions, processes start up with IA, 138 IB, DA and DB enabled, and are reset to this state on exec(). Processes created 139 via fork() and clone() inherit the key enabled state from the calling process. 140 141 It is recommended to avoid disabling the IA key, as this has higher performance 142 overhead than disabling any of the other keys.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.