@samitouri / QOSamiQemu
Code Commits Issues Pulls Sessions CI/CD
master
master
rst 139 lines 5.63 KB
Raw
1 .. _cfi:
2
3 ============================
4 Control-Flow Integrity (CFI)
5 ============================
6
7 This document describes the current control-flow integrity (CFI) mechanism in
8 QEMU. How it can be enabled, its benefits and deficiencies, and how it affects
9 new and existing code in QEMU
10
11 Basics
12 ------
13
14 CFI is a hardening technique that focusing on guaranteeing that indirect
15 function calls have not been altered by an attacker.
16 The type used in QEMU is a forward-edge control-flow integrity that ensures
17 function calls performed through function pointers, always call a "compatible"
18 function. A compatible function is a function with the same signature of the
19 function pointer declared in the source code.
20
21 This type of CFI is entirely compiler-based and relies on the compiler knowing
22 the signature of every function and every function pointer used in the code.
23 As of now, the only compiler that provides support for CFI is Clang.
24
25 CFI is best used on production binaries, to protect against unknown attack
26 vectors.
27
28 In case of a CFI violation (i.e. call to a non-compatible function) QEMU will
29 terminate abruptly, to stop the possible attack.
30
31 Building with CFI
32 -----------------
33
34 NOTE: CFI requires the use of link-time optimization. Therefore, when CFI is
35 selected, LTO will be automatically enabled.
36
37 To build with CFI, the minimum requirement is Clang 6+. If you
38 are planning to also enable fuzzing, then Clang 11+ is needed (more on this
39 later).
40
41 Given the use of LTO, a version of AR that supports LLVM IR is required.
42 The easies way of doing this is by selecting the AR provided by LLVM::
43
44 AR=llvm-ar-9 CC=clang-9 CXX=clang++-9 /path/to/configure --enable-cfi
45
46 CFI is enabled on every binary produced.
47
48 If desired, an additional flag to increase the verbosity of the output in case
49 of a CFI violation is offered (``--enable-debug-cfi``).
50
51 Using QEMU built with CFI
52 -------------------------
53
54 A binary with CFI will work exactly like a standard binary. In case of a CFI
55 violation, the binary will terminate with an illegal instruction signal.
56
57 Incompatible code with CFI
58 --------------------------
59
60 As mentioned above, CFI is entirely compiler-based and therefore relies on
61 compile-time knowledge of the code. This means that, while generally supported
62 for most code, some specific use pattern can break CFI compatibility, and
63 create false-positives. The two main patterns that can cause issues are:
64
65 * Just-in-time compiled code: since such code is created at runtime, the jump
66 to the buffer containing JIT code will fail.
67
68 * Libraries loaded dynamically, e.g. with dlopen/dlsym, since the library was
69 not known at compile time.
70
71 Current areas of QEMU that are not entirely compatible with CFI are:
72
73 1. TCG, since the idea of TCG is to pre-compile groups of instructions at
74 runtime to speed-up interpretation, quite similarly to a JIT compiler
75
76 2. TCI, where the interpreter has to interpret the generic *call* operation
77
78 3. Plugins, since a plugin is implemented as an external library
79
80 4. Modules, since they are implemented as an external library
81
82 5. Directly calling signal handlers from the QEMU source code, since the
83 signal handler may have been provided by an external library or even plugged
84 at runtime.
85
86 Disabling CFI for a specific function
87 -------------------------------------
88
89 If you are working on function that is performing a call using an
90 incompatible way, as described before, you can selectively disable CFI checks
91 for such function by using the decorator ``QEMU_DISABLE_CFI`` at function
92 definition, and add an explanation on why the function is not compatible
93 with CFI. An example of the use of ``QEMU_DISABLE_CFI`` is provided here::
94
95 /*
96 * Disable CFI checks.
97 * TCG creates binary blobs at runtime, with the transformed code.
98 * A TB is a blob of binary code, created at runtime and called with an
99 * indirect function call. Since such function did not exist at compile time,
100 * the CFI runtime has no way to verify its signature and would fail.
101 * TCG is not considered a security-sensitive part of QEMU so this does not
102 * affect the impact of CFI in environment with high security requirements
103 */
104 QEMU_DISABLE_CFI
105 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
106
107 NOTE: CFI needs to be disabled at the **caller** function, (i.e. a compatible
108 cfi function that calls a non-compatible one), since the check is performed
109 when the function call is performed.
110
111 CFI and fuzzing
112 ---------------
113
114 There is generally no advantage of using CFI and fuzzing together, because
115 they target different environments (production for CFI, debug for fuzzing).
116
117 CFI could be used in conjunction with fuzzing to identify a broader set of
118 bugs that may not end immediately in a segmentation fault or triggering
119 an assertion. However, other sanitizers such as address and ub sanitizers
120 can identify such bugs in a more precise way than CFI.
121
122 There is, however, an interesting use case in using CFI in conjunction with
123 fuzzing, that is to make sure that CFI is not triggering any false positive
124 in remote-but-possible parts of the code.
125
126 CFI can be enabled with fuzzing, but with some caveats:
127 1. Fuzzing relies on the linker performing function wrapping at link-time.
128 The standard BFD linker does not support function wrapping when LTO is
129 also enabled. The workaround is to use LLVM's lld linker.
130 2. Fuzzing also relies on a custom linker script, which is only supported by
131 lld with version 11+.
132
133 In other words, to compile with fuzzing and CFI, clang 11+ is required, and
134 lld needs to be used as a linker::
135
136 AR=llvm-ar-11 CC=clang-11 CXX=clang++-11 /path/to/configure --enable-cfi \
137 -enable-fuzzing --extra-ldflags="-fuse-ld=lld"
138
139 and then, compile the fuzzers as usual.