Apple-platform mitigations

Apple-platform mitigations

TL;DR: Apple Silicon stacks hardware mitigations — Pointer Authentication (PAC), Branch Target Identification (BTI), the page-protection layer (PPL/SPTM), and shared-cache/dyld closures — that together kill the easy ROP/JOP/code-injection patterns of the x86 era.

What it is

A bundle of hardware- and OS-level controls Apple ships on arm64e and modern macOS/iOS:

  • PAC signs pointers (return addresses, function pointers, C++ vtables) with a per-process key embedded in unused virtual-address bits; the CPU strips and authenticates on use.
  • BTI requires indirect-branch targets to start with a special instruction, blocking arbitrary jumps into the middle of gadgets.
  • PPL (Page Protection Layer) and on newer SoCs SPTM (Secure Page Table Monitor) wall off page-table updates and code-signing enforcement from compromised EL1 kernel code.
  • dyld closures / shared cache pre-compute bind/rebase for system libraries, removing many writable function-pointer tables from process memory.

Preconditions / where it applies

  • Apple Silicon Macs (M1/M2/M3…), iOS, iPadOS — Intel Macs do not get PAC/BTI/PPL but do get the dyld shared-cache benefits.
  • Relevant when developing kernel exploits (iokit-attack-surface, macos-kernel-debugging), sandbox escapes (macos-sandbox-escape), or jailbreaks.
  • Userland mitigations (PAC for return/call pointers) apply only to arm64e binaries; arm64 third-party apps still get coarse mitigations but not PAC on return addresses.

Technique

What each mitigation forces an attacker to do:

  • PAC — straight pop {pc} / ret gadgets fail because the popped return address has no valid signature. Workarounds: leak a signed pointer with a memory disclosure, forge with a PAC-signing oracle in the target, or pivot to a JOP-style chain that reuses existing signed pointers. PACMAN-class attacks brute-force PAC bits via speculative execution but require specific gadgets and timing.
  • BTI — indirect jumps must land on BTI c/BTI j instructions. Cuts the gadget catalogue dramatically; combined with PAC, classic ROP is largely dead.
  • PPL / SPTM — even with arbitrary kernel R/W, an attacker cannot map writable+executable pages, modify code signatures, or patch the kernel text. To run unsigned code post-exploit you must defeat PPL itself (historically via flaws in the PPL trampolines) or stay in data-only exploitation.
  • dyld shared cache — __DATA slots for system libs are computed at build time; tampering shows up as cache-mismatch panics. Userland injection has moved toward DYLD_INSERT_LIBRARIES (blocked for hardened/platform binaries) and entitlement-gated attach.

Quick check on a binary:

1
2

otool -hv /usr/bin/some_tool | grep -E "PAC|cputype"
codesign -dv --entitlements - /usr/bin/some_tool 2>&1 | grep -i hardened

Detection and defence

  • Defenders rely on Apple’s vuln-fix cadence — these mitigations move primitives, they do not eliminate bugs. Track Apple security releases and CVE notes for kernel and dyld.
  • For enterprise: enforce Hardened Runtime, Library Validation, and notarisation on internal apps; do not ship binaries with com.apple.security.cs.disable-library-validation unless absolutely required.
  • On the offensive side, exploit research now leans data-only and entitlement-abuse rather than classic code-injection. See sip-bypasses and tcc-bypasses.

References