1//===-- sanitizer_symbolizer.h ----------------------------------*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// Symbolizer is used by sanitizers to map instruction address to a location in
11// source code at run-time. Symbolizer either uses __sanitizer_symbolize_*
12// defined in the program, or (if they are missing) tries to find and
13// launch "llvm-symbolizer" commandline tool in a separate process and
14// communicate with it.
15//
16// Generally we should try to avoid calling system library functions during
17// symbolization (and use their replacements from sanitizer_libc.h instead).
18//===----------------------------------------------------------------------===//
19#ifndef SANITIZER_SYMBOLIZER_H
20#define SANITIZER_SYMBOLIZER_H
21
22#include "sanitizer_common.h"
23#include "sanitizer_mutex.h"
24
25namespace __sanitizer {
26
27struct AddressInfo {
28  // Owns all the string members. Storage for them is
29  // (de)allocated using sanitizer internal allocator.
30  uptr address;
31
32  char *module;
33  uptr module_offset;
34  ModuleArch module_arch;
35
36  static const uptr kUnknown = ~(uptr)0;
37  char *function;
38  uptr function_offset;
39
40  char *file;
41  int line;
42  int column;
43
44  AddressInfo();
45  // Deletes all strings and resets all fields.
46  void Clear();
47  void FillModuleInfo(const char *mod_name, uptr mod_offset, ModuleArch arch);
48};
49
50// Linked list of symbolized frames (each frame is described by AddressInfo).
51struct SymbolizedStack {
52  SymbolizedStack *next;
53  AddressInfo info;
54  static SymbolizedStack *New(uptr addr);
55  // Deletes current, and all subsequent frames in the linked list.
56  // The object cannot be accessed after the call to this function.
57  void ClearAll();
58
59 private:
60  SymbolizedStack();
61};
62
63// For now, DataInfo is used to describe global variable.
64struct DataInfo {
65  // Owns all the string members. Storage for them is
66  // (de)allocated using sanitizer internal allocator.
67  char *module;
68  uptr module_offset;
69  ModuleArch module_arch;
70
71  char *file;
72  uptr line;
73  char *name;
74  uptr start;
75  uptr size;
76
77  DataInfo();
78  void Clear();
79};
80
81class SymbolizerTool;
82
83class Symbolizer final {
84 public:
85  /// Initialize and return platform-specific implementation of symbolizer
86  /// (if it wasn't already initialized).
87  static Symbolizer *GetOrInit();
88  static void LateInitialize();
89  // Returns a list of symbolized frames for a given address (containing
90  // all inlined functions, if necessary).
91  SymbolizedStack *SymbolizePC(uptr address);
92  bool SymbolizeData(uptr address, DataInfo *info);
93
94  // The module names Symbolizer returns are stable and unique for every given
95  // module.  It is safe to store and compare them as pointers.
96  bool GetModuleNameAndOffsetForPC(uptr pc, const char **module_name,
97                                   uptr *module_address);
98  const char *GetModuleNameForPc(uptr pc) {
99    const char *module_name = nullptr;
100    uptr unused;
101    if (GetModuleNameAndOffsetForPC(pc, &module_name, &unused))
102      return module_name;
103    return nullptr;
104  }
105
106  // Release internal caches (if any).
107  void Flush();
108  // Attempts to demangle the provided C++ mangled name.
109  const char *Demangle(const char *name);
110
111  // Allow user to install hooks that would be called before/after Symbolizer
112  // does the actual file/line info fetching. Specific sanitizers may need this
113  // to distinguish system library calls made in user code from calls made
114  // during in-process symbolization.
115  typedef void (*StartSymbolizationHook)();
116  typedef void (*EndSymbolizationHook)();
117  // May be called at most once.
118  void AddHooks(StartSymbolizationHook start_hook,
119                EndSymbolizationHook end_hook);
120
121  void RefreshModules();
122  const LoadedModule *FindModuleForAddress(uptr address);
123
124  void InvalidateModuleList();
125
126 private:
127  // GetModuleNameAndOffsetForPC has to return a string to the caller.
128  // Since the corresponding module might get unloaded later, we should create
129  // our owned copies of the strings that we can safely return.
130  // ModuleNameOwner does not provide any synchronization, thus calls to
131  // its method should be protected by |mu_|.
132  class ModuleNameOwner {
133   public:
134    explicit ModuleNameOwner(BlockingMutex *synchronized_by)
135        : last_match_(nullptr), mu_(synchronized_by) {
136      storage_.reserve(kInitialCapacity);
137    }
138    const char *GetOwnedCopy(const char *str);
139
140   private:
141    static const uptr kInitialCapacity = 1000;
142    InternalMmapVector<const char*> storage_;
143    const char *last_match_;
144
145    BlockingMutex *mu_;
146  } module_names_;
147
148  /// Platform-specific function for creating a Symbolizer object.
149  static Symbolizer *PlatformInit();
150
151  bool FindModuleNameAndOffsetForAddress(uptr address, const char **module_name,
152                                         uptr *module_offset,
153                                         ModuleArch *module_arch);
154  ListOfModules modules_;
155  ListOfModules fallback_modules_;
156  // If stale, need to reload the modules before looking up addresses.
157  bool modules_fresh_;
158
159  // Platform-specific default demangler, must not return nullptr.
160  const char *PlatformDemangle(const char *name);
161
162  static Symbolizer *symbolizer_;
163  static StaticSpinMutex init_mu_;
164
165  // Mutex locked from public methods of |Symbolizer|, so that the internals
166  // (including individual symbolizer tools and platform-specific methods) are
167  // always synchronized.
168  BlockingMutex mu_;
169
170  IntrusiveList<SymbolizerTool> tools_;
171
172  explicit Symbolizer(IntrusiveList<SymbolizerTool> tools);
173
174  static LowLevelAllocator symbolizer_allocator_;
175
176  StartSymbolizationHook start_hook_;
177  EndSymbolizationHook end_hook_;
178  class SymbolizerScope {
179   public:
180    explicit SymbolizerScope(const Symbolizer *sym);
181    ~SymbolizerScope();
182   private:
183    const Symbolizer *sym_;
184  };
185};
186
187#ifdef SANITIZER_WINDOWS
188void InitializeDbgHelpIfNeeded();
189#endif
190
191}  // namespace __sanitizer
192
193#endif  // SANITIZER_SYMBOLIZER_H
194