11 ShadowCallStack is an instrumentation pass, currently only implemented for
12 aarch64 and RISC-V, that protects programs against return address overwrites
13 (e.g. stack buffer overflows.) It works by saving a function's return address
14 to a separately allocated 'shadow call stack' in the function prolog in
15 non-leaf functions and loading the return address from the shadow call stack
16 in the function epilog. The return address is also stored on the regular stack
17 for compatibility with unwinders, but is otherwise unused.
19 The aarch64 implementation is considered production ready, and
20 an `implementation of the runtime`_ has been added to Android's libc
21 (bionic). An x86_64 implementation was evaluated using Chromium and was found
22 to have critical performance and security deficiencies--it was removed in
23 LLVM 9.0. Details on the x86_64 implementation can be found in the
24 `Clang 7.0.1 documentation`_.
26 .. _`implementation of the runtime`: https://android.googlesource.com/platform/bionic/+/808d176e7e0dd727c7f929622ec017f6e065c582/libc/bionic/pthread_create.cpp#128
27 .. _`Clang 7.0.1 documentation`: https://releases.llvm.org/7.0.1/tools/clang/docs/ShadowCallStack.html
32 To optimize for memory consumption and cache locality, the shadow call
33 stack stores only an array of return addresses. This is in contrast to other
34 schemes, like :doc:`SafeStack`, that mirror the entire stack and trade-off
35 consuming more memory for shorter function prologs and epilogs with fewer
38 `Return Flow Guard`_ is a pure software implementation of shadow call stacks
39 on x86_64. Like the previous implementation of ShadowCallStack on x86_64, it is
40 inherently racy due to the architecture's use of the stack for calls and
43 Intel `Control-flow Enforcement Technology`_ (CET) is a proposed hardware
44 extension that would add native support to use a shadow stack to store/check
45 return addresses at call/return time. Being a hardware implementation, it
46 would not suffer from race conditions and would not incur the overhead of
47 function instrumentation, but it does require operating system support.
49 .. _`Return Flow Guard`: https://xlab.tencent.com/en/2016/11/02/return-flow-guard/
50 .. _`Control-flow Enforcement Technology`: https://software.intel.com/sites/default/files/managed/4d/2a/control-flow-enforcement-technology-preview.pdf
55 A runtime is not provided in compiler-rt so one must be provided by the
56 compiled application or the operating system. Integrating the runtime into
57 the operating system should be preferred since otherwise all thread creation
58 and destruction would need to be intercepted by the application.
60 The instrumentation makes use of the platform register ``x18`` on AArch64,
61 ``x3`` (``gp``) on RISC-V with software shadow stack and ``ssp`` on RISC-V with
62 hardware shadow stack, which needs `Zicfiss`_ and ``-fcf-protection=return``.
63 Users can choose between the software and hardware based shadow stack
64 implementation on RISC-V backend by passing ``-fsanitize=shadowcallstack``
65 or ``Zicfiss`` with ``-fcf-protection=return``.
66 For simplicity we will refer to this as the ``SCSReg``. On some platforms,
67 ``SCSReg`` is reserved, and on others, it is designated as a scratch register.
68 This generally means that any code that may run on the same thread as code
69 compiled with ShadowCallStack must either target one of the platforms whose ABI
70 reserves ``SCSReg`` (currently Android, Darwin, Fuchsia and Windows) or be
71 compiled with a flag to reserve that register (e.g., ``-ffixed-x18``). If
72 absolutely necessary, code compiled without reserving the register may be run on
73 the same thread as code that uses ShadowCallStack by saving the register value
74 temporarily on the stack (`example in Android`_) but this should be done with
75 care since it risks leaking the shadow call stack address.
77 .. _`Zicfiss`: https://github.com/riscv/riscv-cfi/blob/main/cfi_backward.adoc
78 .. _`example in Android`: https://android-review.googlesource.com/c/platform/frameworks/base/+/803717
80 Because it requires a dedicated register, the ShadowCallStack feature is
81 incompatible with any other feature that may use ``SCSReg``. However, there is
82 no inherent reason why ShadowCallStack needs to use a specific register; in
83 principle, a platform could choose to reserve and use another register for
84 ShadowCallStack, but this would be incompatible with the ABI standards
85 published in AAPCS64 and the RISC-V psABI.
87 Special unwind information is required on functions that are compiled
88 with ShadowCallStack and that may be unwound, i.e. functions compiled with
89 ``-fexceptions`` (which is the default in C++). Some unwinders (such as the
90 libgcc 4.9 unwinder) do not understand this unwind info and will segfault
91 when encountering it. LLVM libunwind processes this unwind info correctly,
92 however. This means that if exceptions are used together with ShadowCallStack,
93 the program must use a compatible unwinder.
98 ShadowCallStack is intended to be a stronger alternative to
99 ``-fstack-protector``. It protects from non-linear overflows and arbitrary
100 memory writes to the return address slot.
102 The instrumentation makes use of the ``SCSReg`` register to reference the shadow
103 call stack, meaning that references to the shadow call stack do not have
104 to be stored in memory. This makes it possible to implement a runtime that
105 avoids exposing the address of the shadow call stack to attackers that can
106 read arbitrary memory. However, attackers could still try to exploit side
107 channels exposed by the operating system `[1]`_ `[2]`_ or processor `[3]`_
108 to discover the address of the shadow call stack.
110 .. _`[1]`: https://eyalitkin.wordpress.com/2017/09/01/cartography-lighting-up-the-shadows/
111 .. _`[2]`: https://www.blackhat.com/docs/eu-16/materials/eu-16-Goktas-Bypassing-Clangs-SafeStack.pdf
112 .. _`[3]`: https://www.vusec.net/projects/anc/
114 Unless care is taken when allocating the shadow call stack, it may be
115 possible for an attacker to guess its address using the addresses of
116 other allocations. Therefore, the address should be chosen to make this
117 difficult. One way to do this is to allocate a large guard region without
118 read/write permissions, randomly select a small region within it to be
119 used as the address of the shadow call stack and mark only that region as
120 read/write. This also mitigates somewhat against processor side channels.
121 The intent is that the Android runtime `will do this`_, but the platform will
122 first need to be `changed`_ to avoid using ``setrlimit(RLIMIT_AS)`` to limit
123 memory allocations in certain processes, as this also limits the number of
124 guard regions that can be allocated.
126 .. _`will do this`: https://android-review.googlesource.com/c/platform/bionic/+/891622
127 .. _`changed`: https://android-review.googlesource.com/c/platform/frameworks/av/+/837745
129 The runtime will need the address of the shadow call stack in order to
130 deallocate it when destroying the thread. If the entire program is compiled
131 with ``SCSReg`` reserved, this is trivial: the address can be derived from the
132 value stored in ``SCSReg`` (e.g. by masking out the lower bits). If a guard
133 region is used, the address of the start of the guard region could then be
134 stored at the start of the shadow call stack itself. But if it is possible
135 for code compiled without reserving ``SCSReg`` to run on a thread managed by the
136 runtime, which is the case on Android for example, the address must be stored
137 somewhere else instead. On Android we store the address of the start of the
138 guard region in TLS and deallocate the entire guard region including the
139 shadow call stack at thread exit. This is considered acceptable given that
140 the address of the start of the guard region is already somewhat guessable.
142 One way in which the address of the shadow call stack could leak is in the
143 ``jmp_buf`` data structure used by ``setjmp`` and ``longjmp``. The Android
144 runtime `avoids this`_ by only storing the low bits of ``SCSReg`` in the
145 ``jmp_buf``, which requires the address of the shadow call stack to be
148 .. _`avoids this`: https://android.googlesource.com/platform/bionic/+/808d176e7e0dd727c7f929622ec017f6e065c582/libc/arch-arm64/bionic/setjmp.S#49
150 The architecture's call and return instructions (``bl`` and ``ret``) operate on
151 a register rather than the stack, which means that leaf functions are generally
152 protected from return address overwrites even without ShadowCallStack.
157 To enable ShadowCallStack, just pass the ``-fsanitize=shadow-call-stack`` flag
158 to both compile and link command lines. On aarch64, you also need to pass
159 ``-ffixed-x18`` unless your target already reserves ``x18``. No additional flags
160 need to be passed on RISC-V because the software based shadow stack uses
161 ``x3`` (``gp``), which is always reserved, and the hardware based shadow call
162 stack uses a dedicated register, ``ssp``.
163 However, it is important to disable GP relaxation in the linker when using the
164 software based shadow call stack on RISC-V. This can be done with the
165 ``--no-relax-gp`` flag in GNU ld, and is off by default in LLD.
170 ``__has_feature(shadow_call_stack)``
171 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
173 In some cases one may need to execute different code depending on whether
174 ShadowCallStack is enabled. The macro ``__has_feature(shadow_call_stack)`` can
175 be used for this purpose.
179 #if defined(__has_feature)
180 # if __has_feature(shadow_call_stack)
181 // code that builds only under ShadowCallStack
185 ``__attribute__((no_sanitize("shadow-call-stack")))``
186 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
188 Use ``__attribute__((no_sanitize("shadow-call-stack")))`` on a function
189 declaration to specify that the shadow call stack instrumentation should not be
190 applied to that function, even if enabled globally.
195 The following example code:
203 Generates the following aarch64 assembly when compiled with ``-O2``:
207 stp x29, x30, [sp, #-16]!
211 ldp x29, x30, [sp], #16
214 Adding ``-fsanitize=shadow-call-stack`` would output the following assembly:
219 stp x29, x30, [sp, #-16]!
223 ldp x29, x30, [sp], #16