1 //===-- TSanRuntime.cpp -----------------------------------------*- 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 #include "TSanRuntime.h"
10 
11 #include "Plugins/Process/Utility/HistoryThread.h"
12 #include "lldb/Breakpoint/StoppointCallbackContext.h"
13 #include "lldb/Core/Debugger.h"
14 #include "lldb/Core/Module.h"
15 #include "lldb/Core/PluginInterface.h"
16 #include "lldb/Core/PluginManager.h"
17 #include "lldb/Core/StreamFile.h"
18 #include "lldb/Core/ValueObject.h"
19 #include "lldb/Expression/UserExpression.h"
20 #include "lldb/Interpreter/CommandReturnObject.h"
21 #include "lldb/Symbol/Symbol.h"
22 #include "lldb/Symbol/SymbolContext.h"
23 #include "lldb/Symbol/Variable.h"
24 #include "lldb/Symbol/VariableList.h"
25 #include "lldb/Target/InstrumentationRuntimeStopInfo.h"
26 #include "lldb/Target/SectionLoadList.h"
27 #include "lldb/Target/StopInfo.h"
28 #include "lldb/Target/Target.h"
29 #include "lldb/Target/Thread.h"
30 #include "lldb/Utility/RegularExpression.h"
31 #include "lldb/Utility/Stream.h"
32 
33 #include <memory>
34 
35 using namespace lldb;
36 using namespace lldb_private;
37 
38 lldb::InstrumentationRuntimeSP
CreateInstance(const lldb::ProcessSP & process_sp)39 ThreadSanitizerRuntime::CreateInstance(const lldb::ProcessSP &process_sp) {
40   return InstrumentationRuntimeSP(new ThreadSanitizerRuntime(process_sp));
41 }
42 
Initialize()43 void ThreadSanitizerRuntime::Initialize() {
44   PluginManager::RegisterPlugin(
45       GetPluginNameStatic(), "ThreadSanitizer instrumentation runtime plugin.",
46       CreateInstance, GetTypeStatic);
47 }
48 
Terminate()49 void ThreadSanitizerRuntime::Terminate() {
50   PluginManager::UnregisterPlugin(CreateInstance);
51 }
52 
GetPluginNameStatic()53 lldb_private::ConstString ThreadSanitizerRuntime::GetPluginNameStatic() {
54   return ConstString("ThreadSanitizer");
55 }
56 
GetTypeStatic()57 lldb::InstrumentationRuntimeType ThreadSanitizerRuntime::GetTypeStatic() {
58   return eInstrumentationRuntimeTypeThreadSanitizer;
59 }
60 
~ThreadSanitizerRuntime()61 ThreadSanitizerRuntime::~ThreadSanitizerRuntime() { Deactivate(); }
62 
63 const char *thread_sanitizer_retrieve_report_data_prefix = R"(
64 extern "C"
65 {
66     void *__tsan_get_current_report();
67     int __tsan_get_report_data(void *report, const char **description, int *count,
68                                int *stack_count, int *mop_count, int *loc_count,
69                                int *mutex_count, int *thread_count,
70                                int *unique_tid_count, void **sleep_trace,
71                                unsigned long trace_size);
72     int __tsan_get_report_stack(void *report, unsigned long idx, void **trace,
73                                 unsigned long trace_size);
74     int __tsan_get_report_mop(void *report, unsigned long idx, int *tid, void **addr,
75                               int *size, int *write, int *atomic, void **trace,
76                               unsigned long trace_size);
77     int __tsan_get_report_loc(void *report, unsigned long idx, const char **type,
78                               void **addr, unsigned long *start, unsigned long *size, int *tid,
79                               int *fd, int *suppressable, void **trace,
80                               unsigned long trace_size);
81     int __tsan_get_report_mutex(void *report, unsigned long idx, unsigned long *mutex_id, void **addr,
82                                 int *destroyed, void **trace, unsigned long trace_size);
83     int __tsan_get_report_thread(void *report, unsigned long idx, int *tid, unsigned long *os_id,
84                                  int *running, const char **name, int *parent_tid,
85                                  void **trace, unsigned long trace_size);
86     int __tsan_get_report_unique_tid(void *report, unsigned long idx, int *tid);
87 
88     // TODO: dlsym won't work on Windows.
89     void *dlsym(void* handle, const char* symbol);
90     int (*ptr__tsan_get_report_loc_object_type)(void *report, unsigned long idx, const char **object_type);
91 }
92 
93 const int REPORT_TRACE_SIZE = 128;
94 const int REPORT_ARRAY_SIZE = 4;
95 
96 struct data {
97     void *report;
98     const char *description;
99     int report_count;
100 
101     void *sleep_trace[REPORT_TRACE_SIZE];
102 
103     int stack_count;
104     struct {
105         int idx;
106         void *trace[REPORT_TRACE_SIZE];
107     } stacks[REPORT_ARRAY_SIZE];
108 
109     int mop_count;
110     struct {
111         int idx;
112         int tid;
113         int size;
114         int write;
115         int atomic;
116         void *addr;
117         void *trace[REPORT_TRACE_SIZE];
118     } mops[REPORT_ARRAY_SIZE];
119 
120     int loc_count;
121     struct {
122         int idx;
123         const char *type;
124         void *addr;
125         unsigned long start;
126         unsigned long size;
127         int tid;
128         int fd;
129         int suppressable;
130         void *trace[REPORT_TRACE_SIZE];
131         const char *object_type;
132     } locs[REPORT_ARRAY_SIZE];
133 
134     int mutex_count;
135     struct {
136         int idx;
137         unsigned long mutex_id;
138         void *addr;
139         int destroyed;
140         void *trace[REPORT_TRACE_SIZE];
141     } mutexes[REPORT_ARRAY_SIZE];
142 
143     int thread_count;
144     struct {
145         int idx;
146         int tid;
147         unsigned long os_id;
148         int running;
149         const char *name;
150         int parent_tid;
151         void *trace[REPORT_TRACE_SIZE];
152     } threads[REPORT_ARRAY_SIZE];
153 
154     int unique_tid_count;
155     struct {
156         int idx;
157         int tid;
158     } unique_tids[REPORT_ARRAY_SIZE];
159 };
160 )";
161 
162 const char *thread_sanitizer_retrieve_report_data_command = R"(
163 data t = {0};
164 
165 ptr__tsan_get_report_loc_object_type = (typeof(ptr__tsan_get_report_loc_object_type))(void *)dlsym((void*)-2 /*RTLD_DEFAULT*/, "__tsan_get_report_loc_object_type");
166 
167 t.report = __tsan_get_current_report();
168 __tsan_get_report_data(t.report, &t.description, &t.report_count, &t.stack_count, &t.mop_count, &t.loc_count, &t.mutex_count, &t.thread_count, &t.unique_tid_count, t.sleep_trace, REPORT_TRACE_SIZE);
169 
170 if (t.stack_count > REPORT_ARRAY_SIZE) t.stack_count = REPORT_ARRAY_SIZE;
171 for (int i = 0; i < t.stack_count; i++) {
172     t.stacks[i].idx = i;
173     __tsan_get_report_stack(t.report, i, t.stacks[i].trace, REPORT_TRACE_SIZE);
174 }
175 
176 if (t.mop_count > REPORT_ARRAY_SIZE) t.mop_count = REPORT_ARRAY_SIZE;
177 for (int i = 0; i < t.mop_count; i++) {
178     t.mops[i].idx = i;
179     __tsan_get_report_mop(t.report, i, &t.mops[i].tid, &t.mops[i].addr, &t.mops[i].size, &t.mops[i].write, &t.mops[i].atomic, t.mops[i].trace, REPORT_TRACE_SIZE);
180 }
181 
182 if (t.loc_count > REPORT_ARRAY_SIZE) t.loc_count = REPORT_ARRAY_SIZE;
183 for (int i = 0; i < t.loc_count; i++) {
184     t.locs[i].idx = i;
185     __tsan_get_report_loc(t.report, i, &t.locs[i].type, &t.locs[i].addr, &t.locs[i].start, &t.locs[i].size, &t.locs[i].tid, &t.locs[i].fd, &t.locs[i].suppressable, t.locs[i].trace, REPORT_TRACE_SIZE);
186     if (ptr__tsan_get_report_loc_object_type)
187         ptr__tsan_get_report_loc_object_type(t.report, i, &t.locs[i].object_type);
188 }
189 
190 if (t.mutex_count > REPORT_ARRAY_SIZE) t.mutex_count = REPORT_ARRAY_SIZE;
191 for (int i = 0; i < t.mutex_count; i++) {
192     t.mutexes[i].idx = i;
193     __tsan_get_report_mutex(t.report, i, &t.mutexes[i].mutex_id, &t.mutexes[i].addr, &t.mutexes[i].destroyed, t.mutexes[i].trace, REPORT_TRACE_SIZE);
194 }
195 
196 if (t.thread_count > REPORT_ARRAY_SIZE) t.thread_count = REPORT_ARRAY_SIZE;
197 for (int i = 0; i < t.thread_count; i++) {
198     t.threads[i].idx = i;
199     __tsan_get_report_thread(t.report, i, &t.threads[i].tid, &t.threads[i].os_id, &t.threads[i].running, &t.threads[i].name, &t.threads[i].parent_tid, t.threads[i].trace, REPORT_TRACE_SIZE);
200 }
201 
202 if (t.unique_tid_count > REPORT_ARRAY_SIZE) t.unique_tid_count = REPORT_ARRAY_SIZE;
203 for (int i = 0; i < t.unique_tid_count; i++) {
204     t.unique_tids[i].idx = i;
205     __tsan_get_report_unique_tid(t.report, i, &t.unique_tids[i].tid);
206 }
207 
208 t;
209 )";
210 
211 static StructuredData::Array *
CreateStackTrace(ValueObjectSP o,const std::string & trace_item_name=".trace")212 CreateStackTrace(ValueObjectSP o,
213                  const std::string &trace_item_name = ".trace") {
214   StructuredData::Array *trace = new StructuredData::Array();
215   ValueObjectSP trace_value_object =
216       o->GetValueForExpressionPath(trace_item_name.c_str());
217   size_t count = trace_value_object->GetNumChildren();
218   for (size_t j = 0; j < count; j++) {
219     addr_t trace_addr =
220         trace_value_object->GetChildAtIndex(j, true)->GetValueAsUnsigned(0);
221     if (trace_addr == 0)
222       break;
223     trace->AddItem(
224         StructuredData::ObjectSP(new StructuredData::Integer(trace_addr)));
225   }
226   return trace;
227 }
228 
ConvertToStructuredArray(ValueObjectSP return_value_sp,const std::string & items_name,const std::string & count_name,std::function<void (ValueObjectSP o,StructuredData::Dictionary * dict)> const & callback)229 static StructuredData::Array *ConvertToStructuredArray(
230     ValueObjectSP return_value_sp, const std::string &items_name,
231     const std::string &count_name,
232     std::function<void(ValueObjectSP o, StructuredData::Dictionary *dict)> const
233         &callback) {
234   StructuredData::Array *array = new StructuredData::Array();
235   unsigned int count =
236       return_value_sp->GetValueForExpressionPath(count_name.c_str())
237           ->GetValueAsUnsigned(0);
238   ValueObjectSP objects =
239       return_value_sp->GetValueForExpressionPath(items_name.c_str());
240   for (unsigned int i = 0; i < count; i++) {
241     ValueObjectSP o = objects->GetChildAtIndex(i, true);
242     StructuredData::Dictionary *dict = new StructuredData::Dictionary();
243 
244     callback(o, dict);
245 
246     array->AddItem(StructuredData::ObjectSP(dict));
247   }
248   return array;
249 }
250 
RetrieveString(ValueObjectSP return_value_sp,ProcessSP process_sp,const std::string & expression_path)251 static std::string RetrieveString(ValueObjectSP return_value_sp,
252                                   ProcessSP process_sp,
253                                   const std::string &expression_path) {
254   addr_t ptr =
255       return_value_sp->GetValueForExpressionPath(expression_path.c_str())
256           ->GetValueAsUnsigned(0);
257   std::string str;
258   Status error;
259   process_sp->ReadCStringFromMemory(ptr, str, error);
260   return str;
261 }
262 
263 static void
GetRenumberedThreadIds(ProcessSP process_sp,ValueObjectSP data,std::map<uint64_t,user_id_t> & thread_id_map)264 GetRenumberedThreadIds(ProcessSP process_sp, ValueObjectSP data,
265                        std::map<uint64_t, user_id_t> &thread_id_map) {
266   ConvertToStructuredArray(
267       data, ".threads", ".thread_count",
268       [process_sp, &thread_id_map](ValueObjectSP o,
269                                    StructuredData::Dictionary *dict) {
270         uint64_t thread_id =
271             o->GetValueForExpressionPath(".tid")->GetValueAsUnsigned(0);
272         uint64_t thread_os_id =
273             o->GetValueForExpressionPath(".os_id")->GetValueAsUnsigned(0);
274         user_id_t lldb_user_id = 0;
275 
276         bool can_update = true;
277         ThreadSP lldb_thread = process_sp->GetThreadList().FindThreadByID(
278             thread_os_id, can_update);
279         if (lldb_thread) {
280           lldb_user_id = lldb_thread->GetIndexID();
281         } else {
282           // This isn't a live thread anymore.  Ask process to assign a new
283           // Index ID (or return an old one if we've already seen this
284           // thread_os_id). It will also make sure that no new threads are
285           // assigned this Index ID.
286           lldb_user_id = process_sp->AssignIndexIDToThread(thread_os_id);
287         }
288 
289         thread_id_map[thread_id] = lldb_user_id;
290       });
291 }
292 
Renumber(uint64_t id,std::map<uint64_t,user_id_t> & thread_id_map)293 static user_id_t Renumber(uint64_t id,
294                           std::map<uint64_t, user_id_t> &thread_id_map) {
295   auto IT = thread_id_map.find(id);
296   if (IT == thread_id_map.end())
297     return 0;
298 
299   return IT->second;
300 }
301 
302 StructuredData::ObjectSP
RetrieveReportData(ExecutionContextRef exe_ctx_ref)303 ThreadSanitizerRuntime::RetrieveReportData(ExecutionContextRef exe_ctx_ref) {
304   ProcessSP process_sp = GetProcessSP();
305   if (!process_sp)
306     return StructuredData::ObjectSP();
307 
308   ThreadSP thread_sp = exe_ctx_ref.GetThreadSP();
309   StackFrameSP frame_sp = thread_sp->GetSelectedFrame();
310 
311   if (!frame_sp)
312     return StructuredData::ObjectSP();
313 
314   EvaluateExpressionOptions options;
315   options.SetUnwindOnError(true);
316   options.SetTryAllThreads(true);
317   options.SetStopOthers(true);
318   options.SetIgnoreBreakpoints(true);
319   options.SetTimeout(process_sp->GetUtilityExpressionTimeout());
320   options.SetPrefix(thread_sanitizer_retrieve_report_data_prefix);
321   options.SetAutoApplyFixIts(false);
322   options.SetLanguage(eLanguageTypeObjC_plus_plus);
323 
324   ValueObjectSP main_value;
325   ExecutionContext exe_ctx;
326   Status eval_error;
327   frame_sp->CalculateExecutionContext(exe_ctx);
328   ExpressionResults result = UserExpression::Evaluate(
329       exe_ctx, options, thread_sanitizer_retrieve_report_data_command, "",
330       main_value, eval_error);
331   if (result != eExpressionCompleted) {
332     process_sp->GetTarget().GetDebugger().GetAsyncOutputStream()->Printf(
333         "Warning: Cannot evaluate ThreadSanitizer expression:\n%s\n",
334         eval_error.AsCString());
335     return StructuredData::ObjectSP();
336   }
337 
338   std::map<uint64_t, user_id_t> thread_id_map;
339   GetRenumberedThreadIds(process_sp, main_value, thread_id_map);
340 
341   StructuredData::Dictionary *dict = new StructuredData::Dictionary();
342   dict->AddStringItem("instrumentation_class", "ThreadSanitizer");
343   dict->AddStringItem("issue_type",
344                       RetrieveString(main_value, process_sp, ".description"));
345   dict->AddIntegerItem("report_count",
346                        main_value->GetValueForExpressionPath(".report_count")
347                            ->GetValueAsUnsigned(0));
348   dict->AddItem("sleep_trace", StructuredData::ObjectSP(CreateStackTrace(
349                                    main_value, ".sleep_trace")));
350 
351   StructuredData::Array *stacks = ConvertToStructuredArray(
352       main_value, ".stacks", ".stack_count",
353       [thread_sp](ValueObjectSP o, StructuredData::Dictionary *dict) {
354         dict->AddIntegerItem(
355             "index",
356             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
357         dict->AddItem("trace", StructuredData::ObjectSP(CreateStackTrace(o)));
358         // "stacks" happen on the current thread
359         dict->AddIntegerItem("thread_id", thread_sp->GetIndexID());
360       });
361   dict->AddItem("stacks", StructuredData::ObjectSP(stacks));
362 
363   StructuredData::Array *mops = ConvertToStructuredArray(
364       main_value, ".mops", ".mop_count",
365       [&thread_id_map](ValueObjectSP o, StructuredData::Dictionary *dict) {
366         dict->AddIntegerItem(
367             "index",
368             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
369         dict->AddIntegerItem(
370             "thread_id",
371             Renumber(
372                 o->GetValueForExpressionPath(".tid")->GetValueAsUnsigned(0),
373                 thread_id_map));
374         dict->AddIntegerItem(
375             "size",
376             o->GetValueForExpressionPath(".size")->GetValueAsUnsigned(0));
377         dict->AddBooleanItem(
378             "is_write",
379             o->GetValueForExpressionPath(".write")->GetValueAsUnsigned(0));
380         dict->AddBooleanItem(
381             "is_atomic",
382             o->GetValueForExpressionPath(".atomic")->GetValueAsUnsigned(0));
383         dict->AddIntegerItem(
384             "address",
385             o->GetValueForExpressionPath(".addr")->GetValueAsUnsigned(0));
386         dict->AddItem("trace", StructuredData::ObjectSP(CreateStackTrace(o)));
387       });
388   dict->AddItem("mops", StructuredData::ObjectSP(mops));
389 
390   StructuredData::Array *locs = ConvertToStructuredArray(
391       main_value, ".locs", ".loc_count",
392       [process_sp, &thread_id_map](ValueObjectSP o,
393                                    StructuredData::Dictionary *dict) {
394         dict->AddIntegerItem(
395             "index",
396             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
397         dict->AddStringItem("type", RetrieveString(o, process_sp, ".type"));
398         dict->AddIntegerItem(
399             "address",
400             o->GetValueForExpressionPath(".addr")->GetValueAsUnsigned(0));
401         dict->AddIntegerItem(
402             "start",
403             o->GetValueForExpressionPath(".start")->GetValueAsUnsigned(0));
404         dict->AddIntegerItem(
405             "size",
406             o->GetValueForExpressionPath(".size")->GetValueAsUnsigned(0));
407         dict->AddIntegerItem(
408             "thread_id",
409             Renumber(
410                 o->GetValueForExpressionPath(".tid")->GetValueAsUnsigned(0),
411                 thread_id_map));
412         dict->AddIntegerItem(
413             "file_descriptor",
414             o->GetValueForExpressionPath(".fd")->GetValueAsUnsigned(0));
415         dict->AddIntegerItem("suppressable",
416                              o->GetValueForExpressionPath(".suppressable")
417                                  ->GetValueAsUnsigned(0));
418         dict->AddItem("trace", StructuredData::ObjectSP(CreateStackTrace(o)));
419         dict->AddStringItem("object_type",
420                             RetrieveString(o, process_sp, ".object_type"));
421       });
422   dict->AddItem("locs", StructuredData::ObjectSP(locs));
423 
424   StructuredData::Array *mutexes = ConvertToStructuredArray(
425       main_value, ".mutexes", ".mutex_count",
426       [](ValueObjectSP o, StructuredData::Dictionary *dict) {
427         dict->AddIntegerItem(
428             "index",
429             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
430         dict->AddIntegerItem(
431             "mutex_id",
432             o->GetValueForExpressionPath(".mutex_id")->GetValueAsUnsigned(0));
433         dict->AddIntegerItem(
434             "address",
435             o->GetValueForExpressionPath(".addr")->GetValueAsUnsigned(0));
436         dict->AddIntegerItem(
437             "destroyed",
438             o->GetValueForExpressionPath(".destroyed")->GetValueAsUnsigned(0));
439         dict->AddItem("trace", StructuredData::ObjectSP(CreateStackTrace(o)));
440       });
441   dict->AddItem("mutexes", StructuredData::ObjectSP(mutexes));
442 
443   StructuredData::Array *threads = ConvertToStructuredArray(
444       main_value, ".threads", ".thread_count",
445       [process_sp, &thread_id_map](ValueObjectSP o,
446                                    StructuredData::Dictionary *dict) {
447         dict->AddIntegerItem(
448             "index",
449             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
450         dict->AddIntegerItem(
451             "thread_id",
452             Renumber(
453                 o->GetValueForExpressionPath(".tid")->GetValueAsUnsigned(0),
454                 thread_id_map));
455         dict->AddIntegerItem(
456             "thread_os_id",
457             o->GetValueForExpressionPath(".os_id")->GetValueAsUnsigned(0));
458         dict->AddIntegerItem(
459             "running",
460             o->GetValueForExpressionPath(".running")->GetValueAsUnsigned(0));
461         dict->AddStringItem("name", RetrieveString(o, process_sp, ".name"));
462         dict->AddIntegerItem(
463             "parent_thread_id",
464             Renumber(o->GetValueForExpressionPath(".parent_tid")
465                          ->GetValueAsUnsigned(0),
466                      thread_id_map));
467         dict->AddItem("trace", StructuredData::ObjectSP(CreateStackTrace(o)));
468       });
469   dict->AddItem("threads", StructuredData::ObjectSP(threads));
470 
471   StructuredData::Array *unique_tids = ConvertToStructuredArray(
472       main_value, ".unique_tids", ".unique_tid_count",
473       [&thread_id_map](ValueObjectSP o, StructuredData::Dictionary *dict) {
474         dict->AddIntegerItem(
475             "index",
476             o->GetValueForExpressionPath(".idx")->GetValueAsUnsigned(0));
477         dict->AddIntegerItem(
478             "tid",
479             Renumber(
480                 o->GetValueForExpressionPath(".tid")->GetValueAsUnsigned(0),
481                 thread_id_map));
482       });
483   dict->AddItem("unique_tids", StructuredData::ObjectSP(unique_tids));
484 
485   return StructuredData::ObjectSP(dict);
486 }
487 
488 std::string
FormatDescription(StructuredData::ObjectSP report)489 ThreadSanitizerRuntime::FormatDescription(StructuredData::ObjectSP report) {
490   std::string description = report->GetAsDictionary()
491                                 ->GetValueForKey("issue_type")
492                                 ->GetAsString()
493                                 ->GetValue();
494 
495   if (description == "data-race") {
496     return "Data race";
497   } else if (description == "data-race-vptr") {
498     return "Data race on C++ virtual pointer";
499   } else if (description == "heap-use-after-free") {
500     return "Use of deallocated memory";
501   } else if (description == "heap-use-after-free-vptr") {
502     return "Use of deallocated C++ virtual pointer";
503   } else if (description == "thread-leak") {
504     return "Thread leak";
505   } else if (description == "locked-mutex-destroy") {
506     return "Destruction of a locked mutex";
507   } else if (description == "mutex-double-lock") {
508     return "Double lock of a mutex";
509   } else if (description == "mutex-invalid-access") {
510     return "Use of an uninitialized or destroyed mutex";
511   } else if (description == "mutex-bad-unlock") {
512     return "Unlock of an unlocked mutex (or by a wrong thread)";
513   } else if (description == "mutex-bad-read-lock") {
514     return "Read lock of a write locked mutex";
515   } else if (description == "mutex-bad-read-unlock") {
516     return "Read unlock of a write locked mutex";
517   } else if (description == "signal-unsafe-call") {
518     return "Signal-unsafe call inside a signal handler";
519   } else if (description == "errno-in-signal-handler") {
520     return "Overwrite of errno in a signal handler";
521   } else if (description == "lock-order-inversion") {
522     return "Lock order inversion (potential deadlock)";
523   } else if (description == "external-race") {
524     return "Race on a library object";
525   } else if (description == "swift-access-race") {
526     return "Swift access race";
527   }
528 
529   // for unknown report codes just show the code
530   return description;
531 }
532 
Sprintf(const char * format,...)533 static std::string Sprintf(const char *format, ...) {
534   StreamString s;
535   va_list args;
536   va_start(args, format);
537   s.PrintfVarArg(format, args);
538   va_end(args);
539   return s.GetString();
540 }
541 
GetSymbolNameFromAddress(ProcessSP process_sp,addr_t addr)542 static std::string GetSymbolNameFromAddress(ProcessSP process_sp, addr_t addr) {
543   lldb_private::Address so_addr;
544   if (!process_sp->GetTarget().GetSectionLoadList().ResolveLoadAddress(addr,
545                                                                        so_addr))
546     return "";
547 
548   lldb_private::Symbol *symbol = so_addr.CalculateSymbolContextSymbol();
549   if (!symbol)
550     return "";
551 
552   std::string sym_name = symbol->GetName().GetCString();
553   return sym_name;
554 }
555 
GetSymbolDeclarationFromAddress(ProcessSP process_sp,addr_t addr,Declaration & decl)556 static void GetSymbolDeclarationFromAddress(ProcessSP process_sp, addr_t addr,
557                                             Declaration &decl) {
558   lldb_private::Address so_addr;
559   if (!process_sp->GetTarget().GetSectionLoadList().ResolveLoadAddress(addr,
560                                                                        so_addr))
561     return;
562 
563   lldb_private::Symbol *symbol = so_addr.CalculateSymbolContextSymbol();
564   if (!symbol)
565     return;
566 
567   ConstString sym_name = symbol->GetMangled().GetName(
568       lldb::eLanguageTypeUnknown, Mangled::ePreferMangled);
569 
570   ModuleSP module = symbol->CalculateSymbolContextModule();
571   if (!module)
572     return;
573 
574   VariableList var_list;
575   module->FindGlobalVariables(sym_name, nullptr, 1U, var_list);
576   if (var_list.GetSize() < 1)
577     return;
578 
579   VariableSP var = var_list.GetVariableAtIndex(0);
580   decl = var->GetDeclaration();
581 }
582 
GetFirstNonInternalFramePc(StructuredData::ObjectSP trace,bool skip_one_frame)583 addr_t ThreadSanitizerRuntime::GetFirstNonInternalFramePc(
584     StructuredData::ObjectSP trace, bool skip_one_frame) {
585   ProcessSP process_sp = GetProcessSP();
586   ModuleSP runtime_module_sp = GetRuntimeModuleSP();
587 
588   StructuredData::Array *trace_array = trace->GetAsArray();
589   for (size_t i = 0; i < trace_array->GetSize(); i++) {
590     if (skip_one_frame && i == 0)
591       continue;
592 
593     addr_t addr;
594     if (!trace_array->GetItemAtIndexAsInteger(i, addr))
595       continue;
596 
597     lldb_private::Address so_addr;
598     if (!process_sp->GetTarget().GetSectionLoadList().ResolveLoadAddress(
599             addr, so_addr))
600       continue;
601 
602     if (so_addr.GetModule() == runtime_module_sp)
603       continue;
604 
605     return addr;
606   }
607 
608   return 0;
609 }
610 
611 std::string
GenerateSummary(StructuredData::ObjectSP report)612 ThreadSanitizerRuntime::GenerateSummary(StructuredData::ObjectSP report) {
613   ProcessSP process_sp = GetProcessSP();
614 
615   std::string summary = report->GetAsDictionary()
616                             ->GetValueForKey("description")
617                             ->GetAsString()
618                             ->GetValue();
619   bool skip_one_frame =
620       report->GetObjectForDotSeparatedPath("issue_type")->GetStringValue() ==
621       "external-race";
622 
623   addr_t pc = 0;
624   if (report->GetAsDictionary()
625           ->GetValueForKey("mops")
626           ->GetAsArray()
627           ->GetSize() > 0)
628     pc = GetFirstNonInternalFramePc(report->GetAsDictionary()
629                                         ->GetValueForKey("mops")
630                                         ->GetAsArray()
631                                         ->GetItemAtIndex(0)
632                                         ->GetAsDictionary()
633                                         ->GetValueForKey("trace"),
634                                     skip_one_frame);
635 
636   if (report->GetAsDictionary()
637           ->GetValueForKey("stacks")
638           ->GetAsArray()
639           ->GetSize() > 0)
640     pc = GetFirstNonInternalFramePc(report->GetAsDictionary()
641                                         ->GetValueForKey("stacks")
642                                         ->GetAsArray()
643                                         ->GetItemAtIndex(0)
644                                         ->GetAsDictionary()
645                                         ->GetValueForKey("trace"),
646                                     skip_one_frame);
647 
648   if (pc != 0) {
649     summary = summary + " in " + GetSymbolNameFromAddress(process_sp, pc);
650   }
651 
652   if (report->GetAsDictionary()
653           ->GetValueForKey("locs")
654           ->GetAsArray()
655           ->GetSize() > 0) {
656     StructuredData::ObjectSP loc = report->GetAsDictionary()
657                                        ->GetValueForKey("locs")
658                                        ->GetAsArray()
659                                        ->GetItemAtIndex(0);
660     std::string object_type = loc->GetAsDictionary()
661                                   ->GetValueForKey("object_type")
662                                   ->GetAsString()
663                                   ->GetValue();
664     if (!object_type.empty()) {
665       summary = "Race on " + object_type + " object";
666     }
667     addr_t addr = loc->GetAsDictionary()
668                       ->GetValueForKey("address")
669                       ->GetAsInteger()
670                       ->GetValue();
671     if (addr == 0)
672       addr = loc->GetAsDictionary()
673                  ->GetValueForKey("start")
674                  ->GetAsInteger()
675                  ->GetValue();
676 
677     if (addr != 0) {
678       std::string global_name = GetSymbolNameFromAddress(process_sp, addr);
679       if (!global_name.empty()) {
680         summary = summary + " at " + global_name;
681       } else {
682         summary = summary + " at " + Sprintf("0x%llx", addr);
683       }
684     } else {
685       int fd = loc->GetAsDictionary()
686                    ->GetValueForKey("file_descriptor")
687                    ->GetAsInteger()
688                    ->GetValue();
689       if (fd != 0) {
690         summary = summary + " on file descriptor " + Sprintf("%d", fd);
691       }
692     }
693   }
694 
695   return summary;
696 }
697 
698 addr_t
GetMainRacyAddress(StructuredData::ObjectSP report)699 ThreadSanitizerRuntime::GetMainRacyAddress(StructuredData::ObjectSP report) {
700   addr_t result = (addr_t)-1;
701 
702   report->GetObjectForDotSeparatedPath("mops")->GetAsArray()->ForEach(
703       [&result](StructuredData::Object *o) -> bool {
704         addr_t addr =
705             o->GetObjectForDotSeparatedPath("address")->GetIntegerValue();
706         if (addr < result)
707           result = addr;
708         return true;
709       });
710 
711   return (result == (addr_t)-1) ? 0 : result;
712 }
713 
GetLocationDescription(StructuredData::ObjectSP report,addr_t & global_addr,std::string & global_name,std::string & filename,uint32_t & line)714 std::string ThreadSanitizerRuntime::GetLocationDescription(
715     StructuredData::ObjectSP report, addr_t &global_addr,
716     std::string &global_name, std::string &filename, uint32_t &line) {
717   std::string result = "";
718 
719   ProcessSP process_sp = GetProcessSP();
720 
721   if (report->GetAsDictionary()
722           ->GetValueForKey("locs")
723           ->GetAsArray()
724           ->GetSize() > 0) {
725     StructuredData::ObjectSP loc = report->GetAsDictionary()
726                                        ->GetValueForKey("locs")
727                                        ->GetAsArray()
728                                        ->GetItemAtIndex(0);
729     std::string type =
730         loc->GetAsDictionary()->GetValueForKey("type")->GetStringValue();
731     if (type == "global") {
732       global_addr = loc->GetAsDictionary()
733                         ->GetValueForKey("address")
734                         ->GetAsInteger()
735                         ->GetValue();
736       global_name = GetSymbolNameFromAddress(process_sp, global_addr);
737       if (!global_name.empty()) {
738         result = Sprintf("'%s' is a global variable (0x%llx)",
739                          global_name.c_str(), global_addr);
740       } else {
741         result = Sprintf("0x%llx is a global variable", global_addr);
742       }
743 
744       Declaration decl;
745       GetSymbolDeclarationFromAddress(process_sp, global_addr, decl);
746       if (decl.GetFile()) {
747         filename = decl.GetFile().GetPath();
748         line = decl.GetLine();
749       }
750     } else if (type == "heap") {
751       addr_t addr = loc->GetAsDictionary()
752                         ->GetValueForKey("start")
753                         ->GetAsInteger()
754                         ->GetValue();
755       long size = loc->GetAsDictionary()
756                       ->GetValueForKey("size")
757                       ->GetAsInteger()
758                       ->GetValue();
759       std::string object_type = loc->GetAsDictionary()
760                                     ->GetValueForKey("object_type")
761                                     ->GetAsString()
762                                     ->GetValue();
763       if (!object_type.empty()) {
764         result = Sprintf("Location is a %ld-byte %s object at 0x%llx", size,
765                          object_type.c_str(), addr);
766       } else {
767         result =
768             Sprintf("Location is a %ld-byte heap object at 0x%llx", size, addr);
769       }
770     } else if (type == "stack") {
771       int tid = loc->GetAsDictionary()
772                     ->GetValueForKey("thread_id")
773                     ->GetAsInteger()
774                     ->GetValue();
775       result = Sprintf("Location is stack of thread %d", tid);
776     } else if (type == "tls") {
777       int tid = loc->GetAsDictionary()
778                     ->GetValueForKey("thread_id")
779                     ->GetAsInteger()
780                     ->GetValue();
781       result = Sprintf("Location is TLS of thread %d", tid);
782     } else if (type == "fd") {
783       int fd = loc->GetAsDictionary()
784                    ->GetValueForKey("file_descriptor")
785                    ->GetAsInteger()
786                    ->GetValue();
787       result = Sprintf("Location is file descriptor %d", fd);
788     }
789   }
790 
791   return result;
792 }
793 
NotifyBreakpointHit(void * baton,StoppointCallbackContext * context,user_id_t break_id,user_id_t break_loc_id)794 bool ThreadSanitizerRuntime::NotifyBreakpointHit(
795     void *baton, StoppointCallbackContext *context, user_id_t break_id,
796     user_id_t break_loc_id) {
797   assert(baton && "null baton");
798   if (!baton)
799     return false;
800 
801   ThreadSanitizerRuntime *const instance =
802       static_cast<ThreadSanitizerRuntime *>(baton);
803 
804   ProcessSP process_sp = instance->GetProcessSP();
805 
806   if (process_sp->GetModIDRef().IsLastResumeForUserExpression())
807     return false;
808 
809   StructuredData::ObjectSP report =
810       instance->RetrieveReportData(context->exe_ctx_ref);
811   std::string stop_reason_description;
812   if (report) {
813     std::string issue_description = instance->FormatDescription(report);
814     report->GetAsDictionary()->AddStringItem("description", issue_description);
815     stop_reason_description = issue_description + " detected";
816     report->GetAsDictionary()->AddStringItem("stop_description",
817                                              stop_reason_description);
818     std::string summary = instance->GenerateSummary(report);
819     report->GetAsDictionary()->AddStringItem("summary", summary);
820     addr_t main_address = instance->GetMainRacyAddress(report);
821     report->GetAsDictionary()->AddIntegerItem("memory_address", main_address);
822 
823     addr_t global_addr = 0;
824     std::string global_name = "";
825     std::string location_filename = "";
826     uint32_t location_line = 0;
827     std::string location_description = instance->GetLocationDescription(
828         report, global_addr, global_name, location_filename, location_line);
829     report->GetAsDictionary()->AddStringItem("location_description",
830                                              location_description);
831     if (global_addr != 0) {
832       report->GetAsDictionary()->AddIntegerItem("global_address", global_addr);
833     }
834     if (!global_name.empty()) {
835       report->GetAsDictionary()->AddStringItem("global_name", global_name);
836     }
837     if (location_filename != "") {
838       report->GetAsDictionary()->AddStringItem("location_filename",
839                                                location_filename);
840       report->GetAsDictionary()->AddIntegerItem("location_line", location_line);
841     }
842 
843     bool all_addresses_are_same = true;
844     report->GetObjectForDotSeparatedPath("mops")->GetAsArray()->ForEach(
845         [&all_addresses_are_same,
846          main_address](StructuredData::Object *o) -> bool {
847           addr_t addr =
848               o->GetObjectForDotSeparatedPath("address")->GetIntegerValue();
849           if (main_address != addr)
850             all_addresses_are_same = false;
851           return true;
852         });
853     report->GetAsDictionary()->AddBooleanItem("all_addresses_are_same",
854                                               all_addresses_are_same);
855   }
856 
857   // Make sure this is the right process
858   if (process_sp && process_sp == context->exe_ctx_ref.GetProcessSP()) {
859     ThreadSP thread_sp = context->exe_ctx_ref.GetThreadSP();
860     if (thread_sp)
861       thread_sp->SetStopInfo(
862           InstrumentationRuntimeStopInfo::
863               CreateStopReasonWithInstrumentationData(
864                   *thread_sp, stop_reason_description, report));
865 
866     StreamFile &s = process_sp->GetTarget().GetDebugger().GetOutputStream();
867     s.Printf("ThreadSanitizer report breakpoint hit. Use 'thread "
868              "info -s' to get extended information about the "
869              "report.\n");
870 
871     return true; // Return true to stop the target
872   } else
873     return false; // Let target run
874 }
875 
GetPatternForRuntimeLibrary()876 const RegularExpression &ThreadSanitizerRuntime::GetPatternForRuntimeLibrary() {
877   static RegularExpression regex(llvm::StringRef("libclang_rt.tsan_"));
878   return regex;
879 }
880 
CheckIfRuntimeIsValid(const lldb::ModuleSP module_sp)881 bool ThreadSanitizerRuntime::CheckIfRuntimeIsValid(
882     const lldb::ModuleSP module_sp) {
883   static ConstString g_tsan_get_current_report("__tsan_get_current_report");
884   const Symbol *symbol = module_sp->FindFirstSymbolWithNameAndType(
885       g_tsan_get_current_report, lldb::eSymbolTypeAny);
886   return symbol != nullptr;
887 }
888 
Activate()889 void ThreadSanitizerRuntime::Activate() {
890   if (IsActive())
891     return;
892 
893   ProcessSP process_sp = GetProcessSP();
894   if (!process_sp)
895     return;
896 
897   ConstString symbol_name("__tsan_on_report");
898   const Symbol *symbol = GetRuntimeModuleSP()->FindFirstSymbolWithNameAndType(
899       symbol_name, eSymbolTypeCode);
900 
901   if (symbol == nullptr)
902     return;
903 
904   if (!symbol->ValueIsAddress() || !symbol->GetAddressRef().IsValid())
905     return;
906 
907   Target &target = process_sp->GetTarget();
908   addr_t symbol_address = symbol->GetAddressRef().GetOpcodeLoadAddress(&target);
909 
910   if (symbol_address == LLDB_INVALID_ADDRESS)
911     return;
912 
913   bool internal = true;
914   bool hardware = false;
915   Breakpoint *breakpoint =
916       process_sp->GetTarget()
917           .CreateBreakpoint(symbol_address, internal, hardware)
918           .get();
919   breakpoint->SetCallback(ThreadSanitizerRuntime::NotifyBreakpointHit, this,
920                           true);
921   breakpoint->SetBreakpointKind("thread-sanitizer-report");
922   SetBreakpointID(breakpoint->GetID());
923 
924   SetActive(true);
925 }
926 
Deactivate()927 void ThreadSanitizerRuntime::Deactivate() {
928   if (GetBreakpointID() != LLDB_INVALID_BREAK_ID) {
929     ProcessSP process_sp = GetProcessSP();
930     if (process_sp) {
931       process_sp->GetTarget().RemoveBreakpointByID(GetBreakpointID());
932       SetBreakpointID(LLDB_INVALID_BREAK_ID);
933     }
934   }
935   SetActive(false);
936 }
GenerateThreadName(const std::string & path,StructuredData::Object * o,StructuredData::ObjectSP main_info)937 static std::string GenerateThreadName(const std::string &path,
938                                       StructuredData::Object *o,
939                                       StructuredData::ObjectSP main_info) {
940   std::string result = "additional information";
941 
942   if (path == "mops") {
943     int size = o->GetObjectForDotSeparatedPath("size")->GetIntegerValue();
944     int thread_id =
945         o->GetObjectForDotSeparatedPath("thread_id")->GetIntegerValue();
946     bool is_write =
947         o->GetObjectForDotSeparatedPath("is_write")->GetBooleanValue();
948     bool is_atomic =
949         o->GetObjectForDotSeparatedPath("is_atomic")->GetBooleanValue();
950     addr_t addr = o->GetObjectForDotSeparatedPath("address")->GetIntegerValue();
951 
952     std::string addr_string = Sprintf(" at 0x%llx", addr);
953 
954     if (main_info->GetObjectForDotSeparatedPath("all_addresses_are_same")
955             ->GetBooleanValue()) {
956       addr_string = "";
957     }
958 
959     if (main_info->GetObjectForDotSeparatedPath("issue_type")
960             ->GetStringValue() == "external-race") {
961       result = Sprintf("%s access by thread %d",
962                        is_write ? "mutating" : "read-only", thread_id);
963     } else if (main_info->GetObjectForDotSeparatedPath("issue_type")
964                    ->GetStringValue() == "swift-access-race") {
965       result = Sprintf("modifying access by thread %d", thread_id);
966     } else {
967       result = Sprintf("%s%s of size %d%s by thread %d",
968                        is_atomic ? "atomic " : "", is_write ? "write" : "read",
969                        size, addr_string.c_str(), thread_id);
970     }
971   }
972 
973   if (path == "threads") {
974     int thread_id =
975         o->GetObjectForDotSeparatedPath("thread_id")->GetIntegerValue();
976     result = Sprintf("Thread %d created", thread_id);
977   }
978 
979   if (path == "locs") {
980     std::string type =
981         o->GetAsDictionary()->GetValueForKey("type")->GetStringValue();
982     int thread_id =
983         o->GetObjectForDotSeparatedPath("thread_id")->GetIntegerValue();
984     int fd =
985         o->GetObjectForDotSeparatedPath("file_descriptor")->GetIntegerValue();
986     if (type == "heap") {
987       result = Sprintf("Heap block allocated by thread %d", thread_id);
988     } else if (type == "fd") {
989       result =
990           Sprintf("File descriptor %d created by thread %t", fd, thread_id);
991     }
992   }
993 
994   if (path == "mutexes") {
995     int mutex_id =
996         o->GetObjectForDotSeparatedPath("mutex_id")->GetIntegerValue();
997 
998     result = Sprintf("Mutex M%d created", mutex_id);
999   }
1000 
1001   if (path == "stacks") {
1002     int thread_id =
1003         o->GetObjectForDotSeparatedPath("thread_id")->GetIntegerValue();
1004     result = Sprintf("Thread %d", thread_id);
1005   }
1006 
1007   result[0] = toupper(result[0]);
1008 
1009   return result;
1010 }
1011 
AddThreadsForPath(const std::string & path,ThreadCollectionSP threads,ProcessSP process_sp,StructuredData::ObjectSP info)1012 static void AddThreadsForPath(const std::string &path,
1013                               ThreadCollectionSP threads, ProcessSP process_sp,
1014                               StructuredData::ObjectSP info) {
1015   info->GetObjectForDotSeparatedPath(path)->GetAsArray()->ForEach(
1016       [process_sp, threads, path, info](StructuredData::Object *o) -> bool {
1017         std::vector<lldb::addr_t> pcs;
1018         o->GetObjectForDotSeparatedPath("trace")->GetAsArray()->ForEach(
1019             [&pcs](StructuredData::Object *pc) -> bool {
1020               pcs.push_back(pc->GetAsInteger()->GetValue());
1021               return true;
1022             });
1023 
1024         if (pcs.size() == 0)
1025           return true;
1026 
1027         StructuredData::ObjectSP thread_id_obj =
1028             o->GetObjectForDotSeparatedPath("thread_os_id");
1029         tid_t tid = thread_id_obj ? thread_id_obj->GetIntegerValue() : 0;
1030 
1031         HistoryThread *history_thread =
1032             new HistoryThread(*process_sp, tid, pcs);
1033         ThreadSP new_thread_sp(history_thread);
1034         new_thread_sp->SetName(GenerateThreadName(path, o, info).c_str());
1035 
1036         // Save this in the Process' ExtendedThreadList so a strong pointer
1037         // retains the object
1038         process_sp->GetExtendedThreadList().AddThread(new_thread_sp);
1039         threads->AddThread(new_thread_sp);
1040 
1041         return true;
1042       });
1043 }
1044 
1045 lldb::ThreadCollectionSP
GetBacktracesFromExtendedStopInfo(StructuredData::ObjectSP info)1046 ThreadSanitizerRuntime::GetBacktracesFromExtendedStopInfo(
1047     StructuredData::ObjectSP info) {
1048   ThreadCollectionSP threads;
1049   threads = std::make_shared<ThreadCollection>();
1050 
1051   if (info->GetObjectForDotSeparatedPath("instrumentation_class")
1052           ->GetStringValue() != "ThreadSanitizer")
1053     return threads;
1054 
1055   ProcessSP process_sp = GetProcessSP();
1056 
1057   AddThreadsForPath("stacks", threads, process_sp, info);
1058   AddThreadsForPath("mops", threads, process_sp, info);
1059   AddThreadsForPath("locs", threads, process_sp, info);
1060   AddThreadsForPath("mutexes", threads, process_sp, info);
1061   AddThreadsForPath("threads", threads, process_sp, info);
1062 
1063   return threads;
1064 }
1065