1 //===-- sanitizer_symbolizer_posix_libcdep.cc -----------------------------===//
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
7 //===----------------------------------------------------------------------===//
9 // This file is shared between AddressSanitizer and ThreadSanitizer
10 // run-time libraries.
11 // POSIX-specific implementation of symbolizer parts.
12 //===----------------------------------------------------------------------===//
14 #include "sanitizer_platform.h"
16 #include "sanitizer_allocator_internal.h"
17 #include "sanitizer_common.h"
18 #include "sanitizer_file.h"
19 #include "sanitizer_flags.h"
20 #include "sanitizer_internal_defs.h"
21 #include "sanitizer_linux.h"
22 #include "sanitizer_placement_new.h"
23 #include "sanitizer_posix.h"
24 #include "sanitizer_procmaps.h"
25 #include "sanitizer_symbolizer_internal.h"
26 #include "sanitizer_symbolizer_libbacktrace.h"
27 #include "sanitizer_symbolizer_mac.h"
29 #include <dlfcn.h> // for dlsym()
37 #include <util.h> // for forkpty()
38 #endif // SANITIZER_MAC
40 // C++ demangling function, as required by Itanium C++ ABI. This is weak,
41 // because we do not require a C++ ABI library to be linked to a program
42 // using sanitizers; if it's not present, we'll just use the mangled name.
43 namespace __cxxabiv1 {
44 extern "C" SANITIZER_WEAK_ATTRIBUTE
45 char *__cxa_demangle(const char *mangled, char *buffer,
46 size_t *length, int *status);
49 namespace __sanitizer {
51 // Attempts to demangle the name via __cxa_demangle from __cxxabiv1.
52 const char *DemangleCXXABI(const char *name) {
53 // FIXME: __cxa_demangle aggressively insists on allocating memory.
54 // There's not much we can do about that, short of providing our
55 // own demangler (libc++abi's implementation could be adapted so that
56 // it does not allocate). For now, we just call it anyway, and we leak
57 // the returned value.
58 if (&__cxxabiv1::__cxa_demangle)
59 if (const char *demangled_name =
60 __cxxabiv1::__cxa_demangle(name, 0, 0, 0))
61 return demangled_name;
66 // As of now, there are no headers for the Swift runtime. Once they are
67 // present, we will weakly link since we do not require Swift runtime to be
69 typedef char *(*swift_demangle_ft)(const char *mangledName,
70 size_t mangledNameLength, char *outputBuffer,
71 size_t *outputBufferSize, uint32_t flags);
72 static swift_demangle_ft swift_demangle_f;
74 // This must not happen lazily at symbolication time, because dlsym uses
75 // malloc and thread-local storage, which is not a good thing to do during
77 static void InitializeSwiftDemangler() {
78 swift_demangle_f = (swift_demangle_ft)dlsym(RTLD_DEFAULT, "swift_demangle");
79 (void)dlerror(); // Cleanup error message in case of failure
82 // Attempts to demangle a Swift name. The demangler will return nullptr if a
83 // non-Swift name is passed in.
84 const char *DemangleSwift(const char *name) {
85 if (!name) return nullptr;
87 // Check if we are dealing with a Swift mangled name first.
88 if (name[0] != '_' || name[1] != 'T') {
93 return swift_demangle_f(name, internal_strlen(name), 0, 0, 0);
98 const char *DemangleSwiftAndCXX(const char *name) {
99 if (!name) return nullptr;
100 if (const char *swift_demangled_name = DemangleSwift(name))
101 return swift_demangled_name;
102 return DemangleCXXABI(name);
105 static bool CreateTwoHighNumberedPipes(int *infd_, int *outfd_) {
108 // The client program may close its stdin and/or stdout and/or stderr
109 // thus allowing socketpair to reuse file descriptors 0, 1 or 2.
110 // In this case the communication between the forked processes may be
111 // broken if either the parent or the child tries to close or duplicate
112 // these descriptors. The loop below produces two pairs of file
113 // descriptors, each greater than 2 (stderr).
115 for (int i = 0; i < 5; i++) {
116 if (pipe(sock_pair[i]) == -1) {
117 for (int j = 0; j < i; j++) {
118 internal_close(sock_pair[j][0]);
119 internal_close(sock_pair[j][1]);
122 } else if (sock_pair[i][0] > 2 && sock_pair[i][1] > 2) {
126 outfd = sock_pair[i];
127 for (int j = 0; j < i; j++) {
128 if (sock_pair[j] == infd) continue;
129 internal_close(sock_pair[j][0]);
130 internal_close(sock_pair[j][1]);
140 outfd_[0] = outfd[0];
141 outfd_[1] = outfd[1];
145 bool SymbolizerProcess::StartSymbolizerSubprocess() {
146 if (!FileExists(path_)) {
147 if (!reported_invalid_path_) {
148 Report("WARNING: invalid path to external symbolizer!\n");
149 reported_invalid_path_ = true;
157 internal_memset(&infd, 0, sizeof(infd));
159 internal_memset(&outfd, 0, sizeof(outfd));
160 if (!CreateTwoHighNumberedPipes(infd, outfd)) {
161 Report("WARNING: Can't create a socket pair to start "
162 "external symbolizer (errno: %d)\n", errno);
168 fd_t fd = kInvalidFd;
170 // forkpty redirects stdout and stderr into a single stream, so we would
171 // receive error messages as standard replies. To avoid that, let's dup
172 // stderr and restore it in the child.
173 int saved_stderr = dup(STDERR_FILENO);
174 CHECK_GE(saved_stderr, 0);
176 // We only need one pipe, for stdin of the child.
180 // Use forkpty to disable buffering in the new terminal.
181 pid = internal_forkpty(&fd);
184 Report("WARNING: failed to fork external symbolizer (errno: %d)\n",
187 } else if (pid == 0) {
190 // infd[0] is the child's reading end.
193 // Set up stdin to read from the pipe.
194 CHECK_GE(dup2(infd[0], STDIN_FILENO), 0);
198 CHECK_GE(dup2(saved_stderr, STDERR_FILENO), 0);
201 const char *argv[kArgVMax];
202 GetArgV(path_, argv);
203 execv(path_, const_cast<char **>(&argv[0]));
207 // Input for the child, infd[1] is our writing end.
208 output_fd_ = infd[1];
211 // Continue execution in parent process.
216 // Disable echo in the new terminal, disable CR.
217 struct termios termflags;
218 tcgetattr(fd, &termflags);
219 termflags.c_oflag &= ~ONLCR;
220 termflags.c_lflag &= ~ECHO;
221 tcsetattr(fd, TCSANOW, &termflags);
222 #else // SANITIZER_MAC
224 #endif // SANITIZER_MAC
226 const char *argv[kArgVMax];
227 GetArgV(path_, argv);
228 pid = StartSubprocess(path_, argv, /* stdin */ outfd[0],
229 /* stdout */ infd[1]);
231 internal_close(infd[0]);
232 internal_close(outfd[1]);
237 output_fd_ = outfd[1];
242 // Check that symbolizer subprocess started successfully.
243 SleepForMillis(kSymbolizerStartupTimeMillis);
244 if (!IsProcessRunning(pid)) {
245 // Either waitpid failed, or child has already exited.
246 Report("WARNING: external symbolizer didn't start up correctly!\n");
253 class Addr2LineProcess : public SymbolizerProcess {
255 Addr2LineProcess(const char *path, const char *module_name)
256 : SymbolizerProcess(path), module_name_(internal_strdup(module_name)) {}
258 const char *module_name() const { return module_name_; }
261 void GetArgV(const char *path_to_binary,
262 const char *(&argv)[kArgVMax]) const override {
264 argv[i++] = path_to_binary;
266 argv[i++] = module_name_;
270 bool ReachedEndOfOutput(const char *buffer, uptr length) const override;
272 bool ReadFromSymbolizer(char *buffer, uptr max_length) override {
273 if (!SymbolizerProcess::ReadFromSymbolizer(buffer, max_length))
275 // The returned buffer is empty when output is valid, but exceeds
279 // We should cut out output_terminator_ at the end of given buffer,
280 // appended by addr2line to mark the end of its meaningful output.
281 // We cannot scan buffer from it's beginning, because it is legal for it
282 // to start with output_terminator_ in case given offset is invalid. So,
283 // scanning from second character.
284 char *garbage = internal_strstr(buffer + 1, output_terminator_);
285 // This should never be NULL since buffer must end up with
286 // output_terminator_.
293 const char *module_name_; // Owned, leaked.
294 static const char output_terminator_[];
297 const char Addr2LineProcess::output_terminator_[] = "??\n??:0\n";
299 bool Addr2LineProcess::ReachedEndOfOutput(const char *buffer,
301 const size_t kTerminatorLen = sizeof(output_terminator_) - 1;
302 // Skip, if we read just kTerminatorLen bytes, because Addr2Line output
303 // should consist at least of two pairs of lines:
304 // 1. First one, corresponding to given offset to be symbolized
305 // (may be equal to output_terminator_, if offset is not valid).
306 // 2. Second one for output_terminator_, itself to mark the end of output.
307 if (length <= kTerminatorLen) return false;
308 // Addr2Line output should end up with output_terminator_.
309 return !internal_memcmp(buffer + length - kTerminatorLen,
310 output_terminator_, kTerminatorLen);
313 class Addr2LinePool : public SymbolizerTool {
315 explicit Addr2LinePool(const char *addr2line_path,
316 LowLevelAllocator *allocator)
317 : addr2line_path_(addr2line_path), allocator_(allocator) {
318 addr2line_pool_.reserve(16);
321 bool SymbolizePC(uptr addr, SymbolizedStack *stack) override {
322 if (const char *buf =
323 SendCommand(stack->info.module, stack->info.module_offset)) {
324 ParseSymbolizePCOutput(buf, stack);
330 bool SymbolizeData(uptr addr, DataInfo *info) override {
335 const char *SendCommand(const char *module_name, uptr module_offset) {
336 Addr2LineProcess *addr2line = 0;
337 for (uptr i = 0; i < addr2line_pool_.size(); ++i) {
339 internal_strcmp(module_name, addr2line_pool_[i]->module_name())) {
340 addr2line = addr2line_pool_[i];
346 new(*allocator_) Addr2LineProcess(addr2line_path_, module_name);
347 addr2line_pool_.push_back(addr2line);
349 CHECK_EQ(0, internal_strcmp(module_name, addr2line->module_name()));
350 char buffer[kBufferSize];
351 internal_snprintf(buffer, kBufferSize, "0x%zx\n0x%zx\n",
352 module_offset, dummy_address_);
353 return addr2line->SendCommand(buffer);
356 static const uptr kBufferSize = 64;
357 const char *addr2line_path_;
358 LowLevelAllocator *allocator_;
359 InternalMmapVector<Addr2LineProcess*> addr2line_pool_;
360 static const uptr dummy_address_ =
361 FIRST_32_SECOND_64(UINT32_MAX, UINT64_MAX);
364 #if SANITIZER_SUPPORTS_WEAK_HOOKS
366 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
367 bool __sanitizer_symbolize_code(const char *ModuleName, u64 ModuleOffset,
368 char *Buffer, int MaxLength);
369 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
370 bool __sanitizer_symbolize_data(const char *ModuleName, u64 ModuleOffset,
371 char *Buffer, int MaxLength);
372 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
373 void __sanitizer_symbolize_flush();
374 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
375 int __sanitizer_symbolize_demangle(const char *Name, char *Buffer,
379 class InternalSymbolizer : public SymbolizerTool {
381 static InternalSymbolizer *get(LowLevelAllocator *alloc) {
382 if (__sanitizer_symbolize_code != 0 &&
383 __sanitizer_symbolize_data != 0) {
384 return new(*alloc) InternalSymbolizer();
389 bool SymbolizePC(uptr addr, SymbolizedStack *stack) override {
390 bool result = __sanitizer_symbolize_code(
391 stack->info.module, stack->info.module_offset, buffer_, kBufferSize);
392 if (result) ParseSymbolizePCOutput(buffer_, stack);
396 bool SymbolizeData(uptr addr, DataInfo *info) override {
397 bool result = __sanitizer_symbolize_data(info->module, info->module_offset,
398 buffer_, kBufferSize);
400 ParseSymbolizeDataOutput(buffer_, info);
401 info->start += (addr - info->module_offset); // Add the base address.
406 void Flush() override {
407 if (__sanitizer_symbolize_flush)
408 __sanitizer_symbolize_flush();
411 const char *Demangle(const char *name) override {
412 if (__sanitizer_symbolize_demangle) {
413 for (uptr res_length = 1024;
414 res_length <= InternalSizeClassMap::kMaxSize;) {
415 char *res_buff = static_cast<char*>(InternalAlloc(res_length));
417 __sanitizer_symbolize_demangle(name, res_buff, res_length);
418 if (req_length > res_length) {
419 res_length = req_length + 1;
420 InternalFree(res_buff);
430 InternalSymbolizer() { }
432 static const int kBufferSize = 16 * 1024;
433 char buffer_[kBufferSize];
435 #else // SANITIZER_SUPPORTS_WEAK_HOOKS
437 class InternalSymbolizer : public SymbolizerTool {
439 static InternalSymbolizer *get(LowLevelAllocator *alloc) { return 0; }
442 #endif // SANITIZER_SUPPORTS_WEAK_HOOKS
444 const char *Symbolizer::PlatformDemangle(const char *name) {
445 return DemangleSwiftAndCXX(name);
448 static SymbolizerTool *ChooseExternalSymbolizer(LowLevelAllocator *allocator) {
449 const char *path = common_flags()->external_symbolizer_path;
450 const char *binary_name = path ? StripModuleName(path) : "";
451 if (path && path[0] == '\0') {
452 VReport(2, "External symbolizer is explicitly disabled.\n");
454 } else if (!internal_strcmp(binary_name, "llvm-symbolizer")) {
455 VReport(2, "Using llvm-symbolizer at user-specified path: %s\n", path);
456 return new(*allocator) LLVMSymbolizer(path, allocator);
457 } else if (!internal_strcmp(binary_name, "atos")) {
459 VReport(2, "Using atos at user-specified path: %s\n", path);
460 return new(*allocator) AtosSymbolizer(path, allocator);
461 #else // SANITIZER_MAC
462 Report("ERROR: Using `atos` is only supported on Darwin.\n");
464 #endif // SANITIZER_MAC
465 } else if (!internal_strcmp(binary_name, "addr2line")) {
466 VReport(2, "Using addr2line at user-specified path: %s\n", path);
467 return new(*allocator) Addr2LinePool(path, allocator);
469 Report("ERROR: External symbolizer path is set to '%s' which isn't "
470 "a known symbolizer. Please set the path to the llvm-symbolizer "
471 "binary or other known tool.\n", path);
475 // Otherwise symbolizer program is unknown, let's search $PATH
476 CHECK(path == nullptr);
478 if (const char *found_path = FindPathToBinary("atos")) {
479 VReport(2, "Using atos found at: %s\n", found_path);
480 return new(*allocator) AtosSymbolizer(found_path, allocator);
482 #endif // SANITIZER_MAC
483 if (const char *found_path = FindPathToBinary("llvm-symbolizer")) {
484 VReport(2, "Using llvm-symbolizer found at: %s\n", found_path);
485 return new(*allocator) LLVMSymbolizer(found_path, allocator);
487 if (common_flags()->allow_addr2line) {
488 if (const char *found_path = FindPathToBinary("addr2line")) {
489 VReport(2, "Using addr2line found at: %s\n", found_path);
490 return new(*allocator) Addr2LinePool(found_path, allocator);
496 static void ChooseSymbolizerTools(IntrusiveList<SymbolizerTool> *list,
497 LowLevelAllocator *allocator) {
498 if (!common_flags()->symbolize) {
499 VReport(2, "Symbolizer is disabled.\n");
502 if (IsAllocatorOutOfMemory()) {
503 VReport(2, "Cannot use internal symbolizer: out of memory\n");
504 } else if (SymbolizerTool *tool = InternalSymbolizer::get(allocator)) {
505 VReport(2, "Using internal symbolizer.\n");
506 list->push_back(tool);
509 if (SymbolizerTool *tool = LibbacktraceSymbolizer::get(allocator)) {
510 VReport(2, "Using libbacktrace symbolizer.\n");
511 list->push_back(tool);
515 if (SymbolizerTool *tool = ChooseExternalSymbolizer(allocator)) {
516 list->push_back(tool);
520 VReport(2, "Using dladdr symbolizer.\n");
521 list->push_back(new(*allocator) DlAddrSymbolizer());
522 #endif // SANITIZER_MAC
525 Symbolizer *Symbolizer::PlatformInit() {
526 IntrusiveList<SymbolizerTool> list;
528 ChooseSymbolizerTools(&list, &symbolizer_allocator_);
529 return new(symbolizer_allocator_) Symbolizer(list);
532 void Symbolizer::LateInitialize() {
533 Symbolizer::GetOrInit();
534 InitializeSwiftDemangler();
537 } // namespace __sanitizer
539 #endif // SANITIZER_POSIX