xref: /freebsd/contrib/llvm-project/compiler-rt/lib/interception/interception.h (revision a8089ea5aee578e08acab2438e82fc9a9ae50ed8)
1 //===-- interception.h ------------------------------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file is a part of AddressSanitizer, an address sanity checker.
10 //
11 // Machinery for providing replacements/wrappers for system functions.
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef INTERCEPTION_H
15 #define INTERCEPTION_H
16 
17 #include "sanitizer_common/sanitizer_asm.h"
18 #include "sanitizer_common/sanitizer_internal_defs.h"
19 
20 #if !SANITIZER_LINUX && !SANITIZER_FREEBSD && !SANITIZER_APPLE &&    \
21     !SANITIZER_NETBSD && !SANITIZER_WINDOWS && !SANITIZER_FUCHSIA && \
22     !SANITIZER_SOLARIS
23 #  error "Interception doesn't work on this operating system."
24 #endif
25 
26 // These typedefs should be used only in the interceptor definitions to replace
27 // the standard system types (e.g. SSIZE_T instead of ssize_t)
28 typedef __sanitizer::uptr    SIZE_T;
29 typedef __sanitizer::sptr    SSIZE_T;
30 typedef __sanitizer::sptr    PTRDIFF_T;
31 typedef __sanitizer::s64     INTMAX_T;
32 typedef __sanitizer::u64     UINTMAX_T;
33 typedef __sanitizer::OFF_T   OFF_T;
34 typedef __sanitizer::OFF64_T OFF64_T;
35 
36 // How to add an interceptor:
37 // Suppose you need to wrap/replace system function (generally, from libc):
38 //      int foo(const char *bar, double baz);
39 // You'll need to:
40 //      1) define INTERCEPTOR(int, foo, const char *bar, double baz) { ... } in
41 //         your source file. See the notes below for cases when
42 //         INTERCEPTOR_WITH_SUFFIX(...) should be used instead.
43 //      2) Call "INTERCEPT_FUNCTION(foo)" prior to the first call of "foo".
44 //         INTERCEPT_FUNCTION(foo) evaluates to "true" iff the function was
45 //         intercepted successfully.
46 // You can access original function by calling REAL(foo)(bar, baz).
47 // By default, REAL(foo) will be visible only inside your interceptor, and if
48 // you want to use it in other parts of RTL, you'll need to:
49 //      3a) add DECLARE_REAL(int, foo, const char*, double) to a
50 //          header file.
51 // However, if the call "INTERCEPT_FUNCTION(foo)" and definition for
52 // INTERCEPTOR(..., foo, ...) are in different files, you'll instead need to:
53 //      3b) add DECLARE_REAL_AND_INTERCEPTOR(int, foo, const char*, double)
54 //          to a header file.
55 
56 // Notes: 1. Things may not work properly if macro INTERCEPTOR(...) {...} or
57 //           DECLARE_REAL(...) are located inside namespaces.
58 //        2. On Mac you can also use: "OVERRIDE_FUNCTION(foo, zoo)" to
59 //           effectively redirect calls from "foo" to "zoo". In this case
60 //           you aren't required to implement
61 //           INTERCEPTOR(int, foo, const char *bar, double baz) {...}
62 //           but instead you'll have to add
63 //           DECLARE_REAL(int, foo, const char *bar, double baz) in your
64 //           source file (to define a pointer to overriden function).
65 //        3. Some Mac functions have symbol variants discriminated by
66 //           additional suffixes, e.g. _$UNIX2003 (see
67 //           https://developer.apple.com/library/mac/#releasenotes/Darwin/SymbolVariantsRelNotes/index.html
68 //           for more details). To intercept such functions you need to use the
69 //           INTERCEPTOR_WITH_SUFFIX(...) macro.
70 
71 // How it works on Linux
72 // ---------------------
73 //
74 // To replace system functions on Linux we just need to declare functions with
75 // the same names in our library and then obtain the real function pointers
76 // using dlsym().
77 //
78 // There is one complication: a user may also intercept some of the functions we
79 // intercept. To allow for up to 3 interceptors (including ours) of a given
80 // function "func", the interceptor implementation is in ___interceptor_func,
81 // which is aliased by a weak function __interceptor_func, which in turn is
82 // aliased (via a trampoline) by weak wrapper function "func".
83 //
84 // Most user interceptors should define a foreign interceptor as follows:
85 //
86 //  - provide a non-weak function "func" that performs interception;
87 //  - if __interceptor_func exists, call it to perform the real functionality;
88 //  - if it does not exist, figure out the real function and call it instead.
89 //
90 // In rare cases, a foreign interceptor (of another dynamic analysis runtime)
91 // may be defined as follows (on supported architectures):
92 //
93 //  - provide a non-weak function __interceptor_func that performs interception;
94 //  - if ___interceptor_func exists, call it to perform the real functionality;
95 //  - if it does not exist, figure out the real function and call it instead;
96 //  - provide a weak function "func" that is an alias to __interceptor_func.
97 //
98 // With this protocol, sanitizer interceptors, foreign user interceptors, and
99 // foreign interceptors of other dynamic analysis runtimes, or any combination
100 // thereof, may co-exist simultaneously.
101 //
102 // How it works on Mac OS
103 // ----------------------
104 //
105 // This is not so on Mac OS, where the two-level namespace makes our replacement
106 // functions invisible to other libraries. This may be overcomed using the
107 // DYLD_FORCE_FLAT_NAMESPACE, but some errors loading the shared libraries in
108 // Chromium were noticed when doing so.
109 //
110 // Instead we create a dylib containing a __DATA,__interpose section that
111 // associates library functions with their wrappers. When this dylib is
112 // preloaded before an executable using DYLD_INSERT_LIBRARIES, it routes all the
113 // calls to interposed functions done through stubs to the wrapper functions.
114 //
115 // As it's decided at compile time which functions are to be intercepted on Mac,
116 // INTERCEPT_FUNCTION() is effectively a no-op on this system.
117 
118 #if SANITIZER_APPLE
119 #include <sys/cdefs.h>  // For __DARWIN_ALIAS_C().
120 
121 // Just a pair of pointers.
122 struct interpose_substitution {
123   const __sanitizer::uptr replacement;
124   const __sanitizer::uptr original;
125 };
126 
127 // For a function foo() create a global pair of pointers { wrap_foo, foo } in
128 // the __DATA,__interpose section.
129 // As a result all the calls to foo() will be routed to wrap_foo() at runtime.
130 #define INTERPOSER(func_name) __attribute__((used))     \
131 const interpose_substitution substitution_##func_name[] \
132     __attribute__((section("__DATA, __interpose"))) = { \
133     { reinterpret_cast<const uptr>(WRAP(func_name)),    \
134       reinterpret_cast<const uptr>(func_name) }         \
135 }
136 
137 // For a function foo() and a wrapper function bar() create a global pair
138 // of pointers { bar, foo } in the __DATA,__interpose section.
139 // As a result all the calls to foo() will be routed to bar() at runtime.
140 #define INTERPOSER_2(func_name, wrapper_name) __attribute__((used)) \
141 const interpose_substitution substitution_##func_name[]             \
142     __attribute__((section("__DATA, __interpose"))) = {             \
143     { reinterpret_cast<const uptr>(wrapper_name),                   \
144       reinterpret_cast<const uptr>(func_name) }                     \
145 }
146 
147 # define WRAP(x) wrap_##x
148 # define TRAMPOLINE(x) WRAP(x)
149 # define INTERCEPTOR_ATTRIBUTE
150 # define DECLARE_WRAPPER(ret_type, func, ...)
151 
152 #elif SANITIZER_WINDOWS
153 # define WRAP(x) __asan_wrap_##x
154 # define TRAMPOLINE(x) WRAP(x)
155 # define INTERCEPTOR_ATTRIBUTE __declspec(dllexport)
156 # define DECLARE_WRAPPER(ret_type, func, ...)         \
157     extern "C" ret_type func(__VA_ARGS__);
158 # define DECLARE_WRAPPER_WINAPI(ret_type, func, ...)  \
159     extern "C" __declspec(dllimport) ret_type __stdcall func(__VA_ARGS__);
160 #elif !SANITIZER_FUCHSIA  // LINUX, FREEBSD, NETBSD, SOLARIS
161 # define INTERCEPTOR_ATTRIBUTE __attribute__((visibility("default")))
162 # if ASM_INTERCEPTOR_TRAMPOLINE_SUPPORT
163 // Weak aliases of weak aliases do not work, therefore we need to set up a
164 // trampoline function. The function "func" is a weak alias to the trampoline
165 // (so that we may check if "func" was overridden), which calls the weak
166 // function __interceptor_func, which in turn aliases the actual interceptor
167 // implementation ___interceptor_func:
168 //
169 //    [wrapper "func": weak] --(alias)--> [TRAMPOLINE(func)]
170 //                                                |
171 //                     +--------(tail call)-------+
172 //                     |
173 //                     v
174 //      [__interceptor_func: weak] --(alias)--> [WRAP(func)]
175 //
176 // We use inline assembly to define most of this, because not all compilers
177 // support functions with the "naked" attribute with every architecture.
178 #  define WRAP(x) ___interceptor_ ## x
179 #  define TRAMPOLINE(x) __interceptor_trampoline_ ## x
180 #  if SANITIZER_FREEBSD || SANITIZER_NETBSD
181 // FreeBSD's dynamic linker (incompliantly) gives non-weak symbols higher
182 // priority than weak ones so weak aliases won't work for indirect calls
183 // in position-independent (-fPIC / -fPIE) mode.
184 #   define __ASM_WEAK_WRAPPER(func) ".globl " #func "\n"
185 #  else
186 #   define __ASM_WEAK_WRAPPER(func) ".weak " #func "\n"
187 #  endif  // SANITIZER_FREEBSD || SANITIZER_NETBSD
188 #  if defined(__arm__) || defined(__aarch64__)
189 #   define ASM_TYPE_FUNCTION_STR "%function"
190 #  else
191 #   define ASM_TYPE_FUNCTION_STR "@function"
192 #  endif
193 // Keep trampoline implementation in sync with sanitizer_common/sanitizer_asm.h
194 #  define DECLARE_WRAPPER(ret_type, func, ...)                                 \
195      extern "C" ret_type func(__VA_ARGS__);                                    \
196      extern "C" ret_type TRAMPOLINE(func)(__VA_ARGS__);                        \
197      extern "C" ret_type __interceptor_##func(__VA_ARGS__)                     \
198        INTERCEPTOR_ATTRIBUTE __attribute__((weak)) ALIAS(WRAP(func));          \
199      asm(                                                                      \
200        ".text\n"                                                               \
201        __ASM_WEAK_WRAPPER(func)                                                \
202        ".set " #func ", " SANITIZER_STRINGIFY(TRAMPOLINE(func)) "\n"           \
203        ".globl " SANITIZER_STRINGIFY(TRAMPOLINE(func)) "\n"                    \
204        ".type  " SANITIZER_STRINGIFY(TRAMPOLINE(func)) ", "                    \
205          ASM_TYPE_FUNCTION_STR "\n"                                            \
206        SANITIZER_STRINGIFY(TRAMPOLINE(func)) ":\n"                             \
207        SANITIZER_STRINGIFY(CFI_STARTPROC) "\n"                                 \
208        C_ASM_TAIL_CALL(SANITIZER_STRINGIFY(TRAMPOLINE(func)),                  \
209                        "__interceptor_"                                        \
210                          SANITIZER_STRINGIFY(ASM_PREEMPTIBLE_SYM(func))) "\n"  \
211        SANITIZER_STRINGIFY(CFI_ENDPROC) "\n"                                   \
212        ".size  " SANITIZER_STRINGIFY(TRAMPOLINE(func)) ", "                    \
213             ".-" SANITIZER_STRINGIFY(TRAMPOLINE(func)) "\n"                    \
214      );
215 # else  // ASM_INTERCEPTOR_TRAMPOLINE_SUPPORT
216 // Some architectures cannot implement efficient interceptor trampolines with
217 // just a plain jump due to complexities of resolving a preemptible symbol. In
218 // those cases, revert to just this scheme:
219 //
220 //    [wrapper "func": weak] --(alias)--> [WRAP(func)]
221 //
222 #  define WRAP(x) __interceptor_ ## x
223 #  define TRAMPOLINE(x) WRAP(x)
224 #  if SANITIZER_FREEBSD || SANITIZER_NETBSD
225 #   define __ATTRIBUTE_WEAK_WRAPPER
226 #  else
227 #   define __ATTRIBUTE_WEAK_WRAPPER __attribute__((weak))
228 #  endif  // SANITIZER_FREEBSD || SANITIZER_NETBSD
229 #  define DECLARE_WRAPPER(ret_type, func, ...)                                 \
230      extern "C" ret_type func(__VA_ARGS__)                                     \
231        INTERCEPTOR_ATTRIBUTE __ATTRIBUTE_WEAK_WRAPPER ALIAS(WRAP(func));
232 # endif  // ASM_INTERCEPTOR_TRAMPOLINE_SUPPORT
233 #endif
234 
235 #if SANITIZER_FUCHSIA
236 // There is no general interception at all on Fuchsia.
237 // Sanitizer runtimes just define functions directly to preempt them,
238 // and have bespoke ways to access the underlying libc functions.
239 # include <zircon/sanitizer.h>
240 # define INTERCEPTOR_ATTRIBUTE __attribute__((visibility("default")))
241 # define REAL(x) __unsanitized_##x
242 # define DECLARE_REAL(ret_type, func, ...)
243 #elif !SANITIZER_APPLE
244 # define PTR_TO_REAL(x) real_##x
245 # define REAL(x) __interception::PTR_TO_REAL(x)
246 # define FUNC_TYPE(x) x##_type
247 
248 # define DECLARE_REAL(ret_type, func, ...)            \
249     typedef ret_type (*FUNC_TYPE(func))(__VA_ARGS__); \
250     namespace __interception {                        \
251     extern FUNC_TYPE(func) PTR_TO_REAL(func);         \
252     }
253 # define ASSIGN_REAL(dst, src) REAL(dst) = REAL(src)
254 #else  // SANITIZER_APPLE
255 # define REAL(x) x
256 # define DECLARE_REAL(ret_type, func, ...) \
257     extern "C" ret_type func(__VA_ARGS__);
258 # define ASSIGN_REAL(x, y)
259 #endif  // SANITIZER_APPLE
260 
261 #if !SANITIZER_FUCHSIA
262 # define DECLARE_REAL_AND_INTERCEPTOR(ret_type, func, ...)  \
263     DECLARE_REAL(ret_type, func, __VA_ARGS__)               \
264     extern "C" ret_type TRAMPOLINE(func)(__VA_ARGS__);      \
265     extern "C" ret_type WRAP(func)(__VA_ARGS__);
266 // Declare an interceptor and its wrapper defined in a different translation
267 // unit (ex. asm).
268 # define DECLARE_EXTERN_INTERCEPTOR_AND_WRAPPER(ret_type, func, ...)  \
269     extern "C" ret_type TRAMPOLINE(func)(__VA_ARGS__);                \
270     extern "C" ret_type WRAP(func)(__VA_ARGS__);                      \
271     extern "C" ret_type func(__VA_ARGS__);
272 #else
273 # define DECLARE_REAL_AND_INTERCEPTOR(ret_type, func, ...)
274 # define DECLARE_EXTERN_INTERCEPTOR_AND_WRAPPER(ret_type, func, ...)
275 #endif
276 
277 // Generally, you don't need to use DEFINE_REAL by itself, as INTERCEPTOR
278 // macros does its job. In exceptional cases you may need to call REAL(foo)
279 // without defining INTERCEPTOR(..., foo, ...). For example, if you override
280 // foo with an interceptor for other function.
281 #if !SANITIZER_APPLE && !SANITIZER_FUCHSIA
282 #  define DEFINE_REAL(ret_type, func, ...)            \
283     typedef ret_type (*FUNC_TYPE(func))(__VA_ARGS__); \
284     namespace __interception {                        \
285     FUNC_TYPE(func) PTR_TO_REAL(func);                \
286     }
287 #else
288 # define DEFINE_REAL(ret_type, func, ...)
289 #endif
290 
291 #if SANITIZER_FUCHSIA
292 
293 // We need to define the __interceptor_func name just to get
294 // sanitizer_common/scripts/gen_dynamic_list.py to export func.
295 // But we don't need to export __interceptor_func to get that.
296 #define INTERCEPTOR(ret_type, func, ...)                                \
297   extern "C"[[ gnu::alias(#func), gnu::visibility("hidden") ]] ret_type \
298       __interceptor_##func(__VA_ARGS__);                                \
299   extern "C" INTERCEPTOR_ATTRIBUTE ret_type func(__VA_ARGS__)
300 
301 #elif !SANITIZER_APPLE
302 
303 #define INTERCEPTOR(ret_type, func, ...)        \
304   DEFINE_REAL(ret_type, func, __VA_ARGS__)      \
305   DECLARE_WRAPPER(ret_type, func, __VA_ARGS__)  \
306   extern "C" INTERCEPTOR_ATTRIBUTE ret_type WRAP(func)(__VA_ARGS__)
307 
308 // We don't need INTERCEPTOR_WITH_SUFFIX on non-Darwin for now.
309 #define INTERCEPTOR_WITH_SUFFIX(ret_type, func, ...) \
310   INTERCEPTOR(ret_type, func, __VA_ARGS__)
311 
312 #else  // SANITIZER_APPLE
313 
314 #define INTERCEPTOR_ZZZ(suffix, ret_type, func, ...)  \
315   extern "C" ret_type func(__VA_ARGS__) suffix;       \
316   extern "C" ret_type WRAP(func)(__VA_ARGS__);        \
317   INTERPOSER(func);                                   \
318   extern "C" INTERCEPTOR_ATTRIBUTE ret_type WRAP(func)(__VA_ARGS__)
319 
320 #define INTERCEPTOR(ret_type, func, ...) \
321   INTERCEPTOR_ZZZ(/*no symbol variants*/, ret_type, func, __VA_ARGS__)
322 
323 #define INTERCEPTOR_WITH_SUFFIX(ret_type, func, ...) \
324   INTERCEPTOR_ZZZ(__DARWIN_ALIAS_C(func), ret_type, func, __VA_ARGS__)
325 
326 // Override |overridee| with |overrider|.
327 #define OVERRIDE_FUNCTION(overridee, overrider) \
328   INTERPOSER_2(overridee, WRAP(overrider))
329 #endif
330 
331 #if SANITIZER_WINDOWS
332 # define INTERCEPTOR_WINAPI(ret_type, func, ...)                \
333     typedef ret_type (__stdcall *FUNC_TYPE(func))(__VA_ARGS__); \
334     namespace __interception {                                  \
335       FUNC_TYPE(func) PTR_TO_REAL(func);                        \
336     }                                                           \
337     extern "C" INTERCEPTOR_ATTRIBUTE ret_type __stdcall WRAP(func)(__VA_ARGS__)
338 #endif
339 
340 // ISO C++ forbids casting between pointer-to-function and pointer-to-object,
341 // so we use casting via an integral type __interception::uptr,
342 // assuming that system is POSIX-compliant. Using other hacks seem
343 // challenging, as we don't even pass function type to
344 // INTERCEPT_FUNCTION macro, only its name.
345 namespace __interception {
346 #if defined(_WIN64)
347 typedef unsigned long long uptr;
348 #else
349 typedef unsigned long uptr;
350 #endif  // _WIN64
351 }  // namespace __interception
352 
353 #define INCLUDED_FROM_INTERCEPTION_LIB
354 
355 #if SANITIZER_LINUX || SANITIZER_FREEBSD || SANITIZER_NETBSD || \
356     SANITIZER_SOLARIS
357 
358 # include "interception_linux.h"
359 # define INTERCEPT_FUNCTION(func) INTERCEPT_FUNCTION_LINUX_OR_FREEBSD(func)
360 # define INTERCEPT_FUNCTION_VER(func, symver) \
361     INTERCEPT_FUNCTION_VER_LINUX_OR_FREEBSD(func, symver)
362 #elif SANITIZER_APPLE
363 # include "interception_mac.h"
364 # define INTERCEPT_FUNCTION(func) INTERCEPT_FUNCTION_MAC(func)
365 # define INTERCEPT_FUNCTION_VER(func, symver) \
366     INTERCEPT_FUNCTION_VER_MAC(func, symver)
367 #elif SANITIZER_WINDOWS
368 # include "interception_win.h"
369 # define INTERCEPT_FUNCTION(func) INTERCEPT_FUNCTION_WIN(func)
370 # define INTERCEPT_FUNCTION_VER(func, symver) \
371     INTERCEPT_FUNCTION_VER_WIN(func, symver)
372 #endif
373 
374 #undef INCLUDED_FROM_INTERCEPTION_LIB
375 
376 #endif  // INTERCEPTION_H
377