1 //===-- DynamicLoaderDarwin.cpp -----------------------------*- 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 #include "DynamicLoaderDarwin.h"
11 
12 #include "lldb/Breakpoint/StoppointCallbackContext.h"
13 #include "lldb/Core/Debugger.h"
14 #include "lldb/Core/Module.h"
15 #include "lldb/Core/ModuleSpec.h"
16 #include "lldb/Core/PluginManager.h"
17 #include "lldb/Core/Section.h"
18 #include "lldb/Expression/DiagnosticManager.h"
19 #include "lldb/Host/FileSystem.h"
20 #include "lldb/Symbol/ClangASTContext.h"
21 #include "lldb/Symbol/Function.h"
22 #include "lldb/Symbol/ObjectFile.h"
23 #include "lldb/Target/ABI.h"
24 #include "lldb/Target/ObjCLanguageRuntime.h"
25 #include "lldb/Target/RegisterContext.h"
26 #include "lldb/Target/StackFrame.h"
27 #include "lldb/Target/Target.h"
28 #include "lldb/Target/Thread.h"
29 #include "lldb/Target/ThreadPlanCallFunction.h"
30 #include "lldb/Target/ThreadPlanRunToAddress.h"
31 #include "lldb/Utility/DataBuffer.h"
32 #include "lldb/Utility/DataBufferHeap.h"
33 #include "lldb/Utility/Log.h"
34 #include "lldb/Utility/State.h"
35 
36 //#define ENABLE_DEBUG_PRINTF // COMMENT THIS LINE OUT PRIOR TO CHECKIN
37 #ifdef ENABLE_DEBUG_PRINTF
38 #include <stdio.h>
39 #define DEBUG_PRINTF(fmt, ...) printf(fmt, ##__VA_ARGS__)
40 #else
41 #define DEBUG_PRINTF(fmt, ...)
42 #endif
43 
44 #ifndef __APPLE__
45 #include "Utility/UuidCompatibility.h"
46 #else
47 #include <uuid/uuid.h>
48 #endif
49 
50 using namespace lldb;
51 using namespace lldb_private;
52 
53 //----------------------------------------------------------------------
54 // Constructor
55 //----------------------------------------------------------------------
56 DynamicLoaderDarwin::DynamicLoaderDarwin(Process *process)
57     : DynamicLoader(process), m_dyld_module_wp(), m_libpthread_module_wp(),
58       m_pthread_getspecific_addr(), m_tid_to_tls_map(), m_dyld_image_infos(),
59       m_dyld_image_infos_stop_id(UINT32_MAX), m_dyld(), m_mutex() {}
60 
61 //----------------------------------------------------------------------
62 // Destructor
63 //----------------------------------------------------------------------
64 DynamicLoaderDarwin::~DynamicLoaderDarwin() {}
65 
66 //------------------------------------------------------------------
67 /// Called after attaching a process.
68 ///
69 /// Allow DynamicLoader plug-ins to execute some code after
70 /// attaching to a process.
71 //------------------------------------------------------------------
72 void DynamicLoaderDarwin::DidAttach() {
73   PrivateInitialize(m_process);
74   DoInitialImageFetch();
75   SetNotificationBreakpoint();
76 }
77 
78 //------------------------------------------------------------------
79 /// Called after attaching a process.
80 ///
81 /// Allow DynamicLoader plug-ins to execute some code after
82 /// attaching to a process.
83 //------------------------------------------------------------------
84 void DynamicLoaderDarwin::DidLaunch() {
85   PrivateInitialize(m_process);
86   DoInitialImageFetch();
87   SetNotificationBreakpoint();
88 }
89 
90 //----------------------------------------------------------------------
91 // Clear out the state of this class.
92 //----------------------------------------------------------------------
93 void DynamicLoaderDarwin::Clear(bool clear_process) {
94   std::lock_guard<std::recursive_mutex> guard(m_mutex);
95   if (clear_process)
96     m_process = NULL;
97   m_dyld_image_infos.clear();
98   m_dyld_image_infos_stop_id = UINT32_MAX;
99   m_dyld.Clear(false);
100 }
101 
102 ModuleSP DynamicLoaderDarwin::FindTargetModuleForImageInfo(
103     ImageInfo &image_info, bool can_create, bool *did_create_ptr) {
104   if (did_create_ptr)
105     *did_create_ptr = false;
106 
107   Target &target = m_process->GetTarget();
108   const ModuleList &target_images = target.GetImages();
109   ModuleSpec module_spec(image_info.file_spec);
110   module_spec.GetUUID() = image_info.uuid;
111   ModuleSP module_sp(target_images.FindFirstModule(module_spec));
112 
113   if (module_sp && !module_spec.GetUUID().IsValid() &&
114       !module_sp->GetUUID().IsValid()) {
115     // No UUID, we must rely upon the cached module modification time and the
116     // modification time of the file on disk
117     if (module_sp->GetModificationTime() !=
118         FileSystem::GetModificationTime(module_sp->GetFileSpec()))
119       module_sp.reset();
120   }
121 
122   if (!module_sp) {
123     if (can_create) {
124       module_sp = target.GetSharedModule(module_spec);
125       if (!module_sp || module_sp->GetObjectFile() == NULL)
126         module_sp = m_process->ReadModuleFromMemory(image_info.file_spec,
127                                                     image_info.address);
128 
129       if (did_create_ptr)
130         *did_create_ptr = (bool)module_sp;
131     }
132   }
133   return module_sp;
134 }
135 
136 void DynamicLoaderDarwin::UnloadImages(
137     const std::vector<lldb::addr_t> &solib_addresses) {
138   std::lock_guard<std::recursive_mutex> guard(m_mutex);
139   if (m_process->GetStopID() == m_dyld_image_infos_stop_id)
140     return;
141 
142   Log *log(lldb_private::GetLogIfAnyCategoriesSet(LIBLLDB_LOG_DYNAMIC_LOADER));
143   Target &target = m_process->GetTarget();
144   if (log)
145     log->Printf("Removing %" PRId64 " modules.",
146                 (uint64_t)solib_addresses.size());
147 
148   ModuleList unloaded_module_list;
149 
150   for (addr_t solib_addr : solib_addresses) {
151     Address header;
152     if (header.SetLoadAddress(solib_addr, &target)) {
153       if (header.GetOffset() == 0) {
154         ModuleSP module_to_remove(header.GetModule());
155         if (module_to_remove.get()) {
156           if (log)
157             log->Printf("Removing module at address 0x%" PRIx64, solib_addr);
158           // remove the sections from the Target
159           UnloadSections(module_to_remove);
160           // add this to the list of modules to remove
161           unloaded_module_list.AppendIfNeeded(module_to_remove);
162           // remove the entry from the m_dyld_image_infos
163           ImageInfo::collection::iterator pos, end = m_dyld_image_infos.end();
164           for (pos = m_dyld_image_infos.begin(); pos != end; pos++) {
165             if (solib_addr == (*pos).address) {
166               m_dyld_image_infos.erase(pos);
167               break;
168             }
169           }
170         }
171       }
172     }
173   }
174 
175   if (unloaded_module_list.GetSize() > 0) {
176     if (log) {
177       log->PutCString("Unloaded:");
178       unloaded_module_list.LogUUIDAndPaths(
179           log, "DynamicLoaderDarwin::UnloadModules");
180     }
181     m_process->GetTarget().GetImages().Remove(unloaded_module_list);
182     m_dyld_image_infos_stop_id = m_process->GetStopID();
183   }
184 }
185 
186 void DynamicLoaderDarwin::UnloadAllImages() {
187   Log *log(lldb_private::GetLogIfAnyCategoriesSet(LIBLLDB_LOG_DYNAMIC_LOADER));
188   ModuleList unloaded_modules_list;
189 
190   Target &target = m_process->GetTarget();
191   const ModuleList &target_modules = target.GetImages();
192   std::lock_guard<std::recursive_mutex> guard(target_modules.GetMutex());
193 
194   size_t num_modules = target_modules.GetSize();
195   ModuleSP dyld_sp(GetDYLDModule());
196 
197   for (size_t i = 0; i < num_modules; i++) {
198     ModuleSP module_sp = target_modules.GetModuleAtIndexUnlocked(i);
199 
200     // Don't remove dyld - else we'll lose our breakpoint notifying us about
201     // libraries being re-loaded...
202     if (module_sp.get() != nullptr && module_sp.get() != dyld_sp.get()) {
203       UnloadSections(module_sp);
204       unloaded_modules_list.Append(module_sp);
205     }
206   }
207 
208   if (unloaded_modules_list.GetSize() != 0) {
209     if (log) {
210       log->PutCString("Unloaded:");
211       unloaded_modules_list.LogUUIDAndPaths(
212           log, "DynamicLoaderDarwin::UnloadAllImages");
213     }
214     target.GetImages().Remove(unloaded_modules_list);
215     m_dyld_image_infos.clear();
216     m_dyld_image_infos_stop_id = m_process->GetStopID();
217   }
218 }
219 
220 //----------------------------------------------------------------------
221 // Update the load addresses for all segments in MODULE using the updated INFO
222 // that is passed in.
223 //----------------------------------------------------------------------
224 bool DynamicLoaderDarwin::UpdateImageLoadAddress(Module *module,
225                                                  ImageInfo &info) {
226   bool changed = false;
227   if (module) {
228     ObjectFile *image_object_file = module->GetObjectFile();
229     if (image_object_file) {
230       SectionList *section_list = image_object_file->GetSectionList();
231       if (section_list) {
232         std::vector<uint32_t> inaccessible_segment_indexes;
233         // We now know the slide amount, so go through all sections and update
234         // the load addresses with the correct values.
235         const size_t num_segments = info.segments.size();
236         for (size_t i = 0; i < num_segments; ++i) {
237           // Only load a segment if it has protections. Things like __PAGEZERO
238           // don't have any protections, and they shouldn't be slid
239           SectionSP section_sp(
240               section_list->FindSectionByName(info.segments[i].name));
241 
242           if (info.segments[i].maxprot == 0) {
243             inaccessible_segment_indexes.push_back(i);
244           } else {
245             const addr_t new_section_load_addr =
246                 info.segments[i].vmaddr + info.slide;
247             static ConstString g_section_name_LINKEDIT("__LINKEDIT");
248 
249             if (section_sp) {
250               // __LINKEDIT sections from files in the shared cache can overlap
251               // so check to see what the segment name is and pass "false" so
252               // we don't warn of overlapping "Section" objects, and "true" for
253               // all other sections.
254               const bool warn_multiple =
255                   section_sp->GetName() != g_section_name_LINKEDIT;
256 
257               changed = m_process->GetTarget().SetSectionLoadAddress(
258                   section_sp, new_section_load_addr, warn_multiple);
259             } else {
260               Host::SystemLog(
261                   Host::eSystemLogWarning,
262                   "warning: unable to find and load segment named '%s' at "
263                   "0x%" PRIx64 " in '%s' in macosx dynamic loader plug-in.\n",
264                   info.segments[i].name.AsCString("<invalid>"),
265                   (uint64_t)new_section_load_addr,
266                   image_object_file->GetFileSpec().GetPath().c_str());
267             }
268           }
269         }
270 
271         // If the loaded the file (it changed) and we have segments that are
272         // not readable or writeable, add them to the invalid memory region
273         // cache for the process. This will typically only be the __PAGEZERO
274         // segment in the main executable. We might be able to apply this more
275         // generally to more sections that have no protections in the future,
276         // but for now we are going to just do __PAGEZERO.
277         if (changed && !inaccessible_segment_indexes.empty()) {
278           for (uint32_t i = 0; i < inaccessible_segment_indexes.size(); ++i) {
279             const uint32_t seg_idx = inaccessible_segment_indexes[i];
280             SectionSP section_sp(
281                 section_list->FindSectionByName(info.segments[seg_idx].name));
282 
283             if (section_sp) {
284               static ConstString g_pagezero_section_name("__PAGEZERO");
285               if (g_pagezero_section_name == section_sp->GetName()) {
286                 // __PAGEZERO never slides...
287                 const lldb::addr_t vmaddr = info.segments[seg_idx].vmaddr;
288                 const lldb::addr_t vmsize = info.segments[seg_idx].vmsize;
289                 Process::LoadRange pagezero_range(vmaddr, vmsize);
290                 m_process->AddInvalidMemoryRegion(pagezero_range);
291               }
292             }
293           }
294         }
295       }
296     }
297   }
298   // We might have an in memory image that was loaded as soon as it was created
299   if (info.load_stop_id == m_process->GetStopID())
300     changed = true;
301   else if (changed) {
302     // Update the stop ID when this library was updated
303     info.load_stop_id = m_process->GetStopID();
304   }
305   return changed;
306 }
307 
308 //----------------------------------------------------------------------
309 // Unload the segments in MODULE using the INFO that is passed in.
310 //----------------------------------------------------------------------
311 bool DynamicLoaderDarwin::UnloadModuleSections(Module *module,
312                                                ImageInfo &info) {
313   bool changed = false;
314   if (module) {
315     ObjectFile *image_object_file = module->GetObjectFile();
316     if (image_object_file) {
317       SectionList *section_list = image_object_file->GetSectionList();
318       if (section_list) {
319         const size_t num_segments = info.segments.size();
320         for (size_t i = 0; i < num_segments; ++i) {
321           SectionSP section_sp(
322               section_list->FindSectionByName(info.segments[i].name));
323           if (section_sp) {
324             const addr_t old_section_load_addr =
325                 info.segments[i].vmaddr + info.slide;
326             if (m_process->GetTarget().SetSectionUnloaded(
327                     section_sp, old_section_load_addr))
328               changed = true;
329           } else {
330             Host::SystemLog(Host::eSystemLogWarning,
331                             "warning: unable to find and unload segment named "
332                             "'%s' in '%s' in macosx dynamic loader plug-in.\n",
333                             info.segments[i].name.AsCString("<invalid>"),
334                             image_object_file->GetFileSpec().GetPath().c_str());
335           }
336         }
337       }
338     }
339   }
340   return changed;
341 }
342 
343 // Given a JSON dictionary (from debugserver, most likely) of binary images
344 // loaded in the inferior process, add the images to the ImageInfo collection.
345 
346 bool DynamicLoaderDarwin::JSONImageInformationIntoImageInfo(
347     StructuredData::ObjectSP image_details,
348     ImageInfo::collection &image_infos) {
349   StructuredData::ObjectSP images_sp =
350       image_details->GetAsDictionary()->GetValueForKey("images");
351   if (images_sp.get() == nullptr)
352     return false;
353 
354   image_infos.resize(images_sp->GetAsArray()->GetSize());
355 
356   for (size_t i = 0; i < image_infos.size(); i++) {
357     StructuredData::ObjectSP image_sp =
358         images_sp->GetAsArray()->GetItemAtIndex(i);
359     if (image_sp.get() == nullptr || image_sp->GetAsDictionary() == nullptr)
360       return false;
361     StructuredData::Dictionary *image = image_sp->GetAsDictionary();
362     if (image->HasKey("load_address") == false ||
363         image->HasKey("pathname") == false ||
364         image->HasKey("mod_date") == false ||
365         image->HasKey("mach_header") == false ||
366         image->GetValueForKey("mach_header")->GetAsDictionary() == nullptr ||
367         image->HasKey("segments") == false ||
368         image->GetValueForKey("segments")->GetAsArray() == nullptr ||
369         image->HasKey("uuid") == false) {
370       return false;
371     }
372     image_infos[i].address =
373         image->GetValueForKey("load_address")->GetAsInteger()->GetValue();
374     image_infos[i].mod_date =
375         image->GetValueForKey("mod_date")->GetAsInteger()->GetValue();
376     image_infos[i].file_spec.SetFile(
377         image->GetValueForKey("pathname")->GetAsString()->GetValue(), false,
378         FileSpec::Style::native);
379 
380     StructuredData::Dictionary *mh =
381         image->GetValueForKey("mach_header")->GetAsDictionary();
382     image_infos[i].header.magic =
383         mh->GetValueForKey("magic")->GetAsInteger()->GetValue();
384     image_infos[i].header.cputype =
385         mh->GetValueForKey("cputype")->GetAsInteger()->GetValue();
386     image_infos[i].header.cpusubtype =
387         mh->GetValueForKey("cpusubtype")->GetAsInteger()->GetValue();
388     image_infos[i].header.filetype =
389         mh->GetValueForKey("filetype")->GetAsInteger()->GetValue();
390 
391     if (image->HasKey("min_version_os_name")) {
392       std::string os_name = image->GetValueForKey("min_version_os_name")
393                                 ->GetAsString()
394                                 ->GetValue();
395       if (os_name == "macosx")
396         image_infos[i].os_type = llvm::Triple::MacOSX;
397       else if (os_name == "ios" || os_name == "iphoneos")
398         image_infos[i].os_type = llvm::Triple::IOS;
399       else if (os_name == "tvos")
400         image_infos[i].os_type = llvm::Triple::TvOS;
401       else if (os_name == "watchos")
402         image_infos[i].os_type = llvm::Triple::WatchOS;
403     }
404     if (image->HasKey("min_version_os_sdk")) {
405       image_infos[i].min_version_os_sdk =
406           image->GetValueForKey("min_version_os_sdk")
407               ->GetAsString()
408               ->GetValue();
409     }
410 
411     // Fields that aren't used by DynamicLoaderDarwin so debugserver doesn't
412     // currently send them in the reply.
413 
414     if (mh->HasKey("flags"))
415       image_infos[i].header.flags =
416           mh->GetValueForKey("flags")->GetAsInteger()->GetValue();
417     else
418       image_infos[i].header.flags = 0;
419 
420     if (mh->HasKey("ncmds"))
421       image_infos[i].header.ncmds =
422           mh->GetValueForKey("ncmds")->GetAsInteger()->GetValue();
423     else
424       image_infos[i].header.ncmds = 0;
425 
426     if (mh->HasKey("sizeofcmds"))
427       image_infos[i].header.sizeofcmds =
428           mh->GetValueForKey("sizeofcmds")->GetAsInteger()->GetValue();
429     else
430       image_infos[i].header.sizeofcmds = 0;
431 
432     StructuredData::Array *segments =
433         image->GetValueForKey("segments")->GetAsArray();
434     uint32_t segcount = segments->GetSize();
435     for (size_t j = 0; j < segcount; j++) {
436       Segment segment;
437       StructuredData::Dictionary *seg =
438           segments->GetItemAtIndex(j)->GetAsDictionary();
439       segment.name =
440           ConstString(seg->GetValueForKey("name")->GetAsString()->GetValue());
441       segment.vmaddr =
442           seg->GetValueForKey("vmaddr")->GetAsInteger()->GetValue();
443       segment.vmsize =
444           seg->GetValueForKey("vmsize")->GetAsInteger()->GetValue();
445       segment.fileoff =
446           seg->GetValueForKey("fileoff")->GetAsInteger()->GetValue();
447       segment.filesize =
448           seg->GetValueForKey("filesize")->GetAsInteger()->GetValue();
449       segment.maxprot =
450           seg->GetValueForKey("maxprot")->GetAsInteger()->GetValue();
451 
452       // Fields that aren't used by DynamicLoaderDarwin so debugserver doesn't
453       // currently send them in the reply.
454 
455       if (seg->HasKey("initprot"))
456         segment.initprot =
457             seg->GetValueForKey("initprot")->GetAsInteger()->GetValue();
458       else
459         segment.initprot = 0;
460 
461       if (seg->HasKey("flags"))
462         segment.flags =
463             seg->GetValueForKey("flags")->GetAsInteger()->GetValue();
464       else
465         segment.flags = 0;
466 
467       if (seg->HasKey("nsects"))
468         segment.nsects =
469             seg->GetValueForKey("nsects")->GetAsInteger()->GetValue();
470       else
471         segment.nsects = 0;
472 
473       image_infos[i].segments.push_back(segment);
474     }
475 
476     image_infos[i].uuid.SetFromStringRef(
477         image->GetValueForKey("uuid")->GetAsString()->GetValue());
478 
479     // All sections listed in the dyld image info structure will all either be
480     // fixed up already, or they will all be off by a single slide amount that
481     // is determined by finding the first segment that is at file offset zero
482     // which also has bytes (a file size that is greater than zero) in the
483     // object file.
484 
485     // Determine the slide amount (if any)
486     const size_t num_sections = image_infos[i].segments.size();
487     for (size_t k = 0; k < num_sections; ++k) {
488       // Iterate through the object file sections to find the first section
489       // that starts of file offset zero and that has bytes in the file...
490       if ((image_infos[i].segments[k].fileoff == 0 &&
491            image_infos[i].segments[k].filesize > 0) ||
492           (image_infos[i].segments[k].name == ConstString("__TEXT"))) {
493         image_infos[i].slide =
494             image_infos[i].address - image_infos[i].segments[k].vmaddr;
495         // We have found the slide amount, so we can exit this for loop.
496         break;
497       }
498     }
499   }
500 
501   return true;
502 }
503 
504 void DynamicLoaderDarwin::UpdateSpecialBinariesFromNewImageInfos(
505     ImageInfo::collection &image_infos) {
506   uint32_t exe_idx = UINT32_MAX;
507   uint32_t dyld_idx = UINT32_MAX;
508   Target &target = m_process->GetTarget();
509   Log *log(lldb_private::GetLogIfAnyCategoriesSet(LIBLLDB_LOG_DYNAMIC_LOADER));
510   ConstString g_dyld_sim_filename("dyld_sim");
511 
512   ArchSpec target_arch = target.GetArchitecture();
513   const size_t image_infos_size = image_infos.size();
514   for (size_t i = 0; i < image_infos_size; i++) {
515     if (image_infos[i].header.filetype == llvm::MachO::MH_DYLINKER) {
516       // In a "simulator" process (an x86 process that is ios/tvos/watchos) we
517       // will have two dyld modules -- a "dyld" that we want to keep track of,
518       // and a "dyld_sim" which we don't need to keep track of here. If the
519       // target is an x86 system and the OS of the dyld binary is
520       // ios/tvos/watchos, then we are looking at dyld_sym.
521 
522       // debugserver has only recently (late 2016) started sending up the os
523       // type for each binary it sees -- so if we don't have an os type, use a
524       // filename check as our next best guess.
525       if (image_infos[i].os_type == llvm::Triple::OSType::UnknownOS) {
526         if (image_infos[i].file_spec.GetFilename() != g_dyld_sim_filename) {
527           dyld_idx = i;
528         }
529       } else if (target_arch.GetTriple().getArch() == llvm::Triple::x86 ||
530                  target_arch.GetTriple().getArch() == llvm::Triple::x86_64) {
531         if (image_infos[i].os_type != llvm::Triple::OSType::IOS &&
532             image_infos[i].os_type != llvm::Triple::TvOS &&
533             image_infos[i].os_type != llvm::Triple::WatchOS) {
534           dyld_idx = i;
535         }
536       }
537       else {
538         // catch-all for any other environment -- trust that dyld is actually
539         // dyld
540         dyld_idx = i;
541       }
542     } else if (image_infos[i].header.filetype == llvm::MachO::MH_EXECUTE) {
543       exe_idx = i;
544     }
545   }
546 
547   if (exe_idx != UINT32_MAX) {
548     const bool can_create = true;
549     ModuleSP exe_module_sp(
550         FindTargetModuleForImageInfo(image_infos[exe_idx], can_create, NULL));
551     if (exe_module_sp) {
552       if (log)
553         log->Printf("Found executable module: %s",
554                     exe_module_sp->GetFileSpec().GetPath().c_str());
555       target.GetImages().AppendIfNeeded(exe_module_sp);
556       UpdateImageLoadAddress(exe_module_sp.get(), image_infos[exe_idx]);
557       if (exe_module_sp.get() != target.GetExecutableModulePointer()) {
558         const bool get_dependent_images = false;
559         target.SetExecutableModule(exe_module_sp, get_dependent_images);
560       }
561     }
562   }
563 
564   if (dyld_idx != UINT32_MAX) {
565     const bool can_create = true;
566     ModuleSP dyld_sp =
567         FindTargetModuleForImageInfo(image_infos[dyld_idx], can_create, NULL);
568     if (dyld_sp.get()) {
569       if (log)
570         log->Printf("Found dyld module: %s",
571                     dyld_sp->GetFileSpec().GetPath().c_str());
572       target.GetImages().AppendIfNeeded(dyld_sp);
573       UpdateImageLoadAddress(dyld_sp.get(), image_infos[dyld_idx]);
574       SetDYLDModule(dyld_sp);
575     }
576   }
577 }
578 
579 void DynamicLoaderDarwin::UpdateDYLDImageInfoFromNewImageInfo(
580     ImageInfo &image_info) {
581   if (image_info.header.filetype == llvm::MachO::MH_DYLINKER) {
582     const bool can_create = true;
583     ModuleSP dyld_sp =
584         FindTargetModuleForImageInfo(image_info, can_create, NULL);
585     if (dyld_sp.get()) {
586       Target &target = m_process->GetTarget();
587       target.GetImages().AppendIfNeeded(dyld_sp);
588       UpdateImageLoadAddress(dyld_sp.get(), image_info);
589       SetDYLDModule(dyld_sp);
590     }
591   }
592 }
593 
594 void DynamicLoaderDarwin::SetDYLDModule(lldb::ModuleSP &dyld_module_sp) {
595   m_dyld_module_wp = dyld_module_sp;
596 }
597 
598 ModuleSP DynamicLoaderDarwin::GetDYLDModule() {
599   ModuleSP dyld_sp(m_dyld_module_wp.lock());
600   return dyld_sp;
601 }
602 
603 bool DynamicLoaderDarwin::AddModulesUsingImageInfos(
604     ImageInfo::collection &image_infos) {
605   std::lock_guard<std::recursive_mutex> guard(m_mutex);
606   // Now add these images to the main list.
607   ModuleList loaded_module_list;
608   Log *log(lldb_private::GetLogIfAnyCategoriesSet(LIBLLDB_LOG_DYNAMIC_LOADER));
609   Target &target = m_process->GetTarget();
610   ModuleList &target_images = target.GetImages();
611 
612   for (uint32_t idx = 0; idx < image_infos.size(); ++idx) {
613     if (log) {
614       log->Printf("Adding new image at address=0x%16.16" PRIx64 ".",
615                   image_infos[idx].address);
616       image_infos[idx].PutToLog(log);
617     }
618 
619     m_dyld_image_infos.push_back(image_infos[idx]);
620 
621     ModuleSP image_module_sp(
622         FindTargetModuleForImageInfo(image_infos[idx], true, NULL));
623 
624     if (image_module_sp) {
625       ObjectFile *objfile = image_module_sp->GetObjectFile();
626       if (objfile) {
627         SectionList *sections = objfile->GetSectionList();
628         if (sections) {
629           ConstString commpage_dbstr("__commpage");
630           Section *commpage_section =
631               sections->FindSectionByName(commpage_dbstr).get();
632           if (commpage_section) {
633             ModuleSpec module_spec(objfile->GetFileSpec(),
634                                    image_infos[idx].GetArchitecture());
635             module_spec.GetObjectName() = commpage_dbstr;
636             ModuleSP commpage_image_module_sp(
637                 target_images.FindFirstModule(module_spec));
638             if (!commpage_image_module_sp) {
639               module_spec.SetObjectOffset(objfile->GetFileOffset() +
640                                           commpage_section->GetFileOffset());
641               module_spec.SetObjectSize(objfile->GetByteSize());
642               commpage_image_module_sp = target.GetSharedModule(module_spec);
643               if (!commpage_image_module_sp ||
644                   commpage_image_module_sp->GetObjectFile() == NULL) {
645                 commpage_image_module_sp = m_process->ReadModuleFromMemory(
646                     image_infos[idx].file_spec, image_infos[idx].address);
647                 // Always load a memory image right away in the target in case
648                 // we end up trying to read the symbol table from memory... The
649                 // __LINKEDIT will need to be mapped so we can figure out where
650                 // the symbol table bits are...
651                 bool changed = false;
652                 UpdateImageLoadAddress(commpage_image_module_sp.get(),
653                                        image_infos[idx]);
654                 target.GetImages().Append(commpage_image_module_sp);
655                 if (changed) {
656                   image_infos[idx].load_stop_id = m_process->GetStopID();
657                   loaded_module_list.AppendIfNeeded(commpage_image_module_sp);
658                 }
659               }
660             }
661           }
662         }
663       }
664 
665       // UpdateImageLoadAddress will return true if any segments change load
666       // address. We need to check this so we don't mention that all loaded
667       // shared libraries are newly loaded each time we hit out dyld breakpoint
668       // since dyld will list all shared libraries each time.
669       if (UpdateImageLoadAddress(image_module_sp.get(), image_infos[idx])) {
670         target_images.AppendIfNeeded(image_module_sp);
671         loaded_module_list.AppendIfNeeded(image_module_sp);
672       }
673     }
674   }
675 
676   if (loaded_module_list.GetSize() > 0) {
677     if (log)
678       loaded_module_list.LogUUIDAndPaths(log,
679                                          "DynamicLoaderDarwin::ModulesDidLoad");
680     m_process->GetTarget().ModulesDidLoad(loaded_module_list);
681   }
682   return true;
683 }
684 
685 //----------------------------------------------------------------------
686 // On Mac OS X libobjc (the Objective-C runtime) has several critical dispatch
687 // functions written in hand-written assembly, and also have hand-written
688 // unwind information in the eh_frame section.  Normally we prefer analyzing
689 // the assembly instructions of a currently executing frame to unwind from that
690 // frame -- but on hand-written functions this profiling can fail.  We should
691 // use the eh_frame instructions for these functions all the time.
692 //
693 // As an aside, it would be better if the eh_frame entries had a flag (or were
694 // extensible so they could have an Apple-specific flag) which indicates that
695 // the instructions are asynchronous -- accurate at every instruction, instead
696 // of our normal default assumption that they are not.
697 //----------------------------------------------------------------------
698 
699 bool DynamicLoaderDarwin::AlwaysRelyOnEHUnwindInfo(SymbolContext &sym_ctx) {
700   ModuleSP module_sp;
701   if (sym_ctx.symbol) {
702     module_sp = sym_ctx.symbol->GetAddressRef().GetModule();
703   }
704   if (module_sp.get() == NULL && sym_ctx.function) {
705     module_sp =
706         sym_ctx.function->GetAddressRange().GetBaseAddress().GetModule();
707   }
708   if (module_sp.get() == NULL)
709     return false;
710 
711   ObjCLanguageRuntime *objc_runtime = m_process->GetObjCLanguageRuntime();
712   if (objc_runtime != NULL && objc_runtime->IsModuleObjCLibrary(module_sp)) {
713     return true;
714   }
715 
716   return false;
717 }
718 
719 //----------------------------------------------------------------------
720 // Dump a Segment to the file handle provided.
721 //----------------------------------------------------------------------
722 void DynamicLoaderDarwin::Segment::PutToLog(Log *log,
723                                             lldb::addr_t slide) const {
724   if (log) {
725     if (slide == 0)
726       log->Printf("\t\t%16s [0x%16.16" PRIx64 " - 0x%16.16" PRIx64 ")",
727                   name.AsCString(""), vmaddr + slide, vmaddr + slide + vmsize);
728     else
729       log->Printf("\t\t%16s [0x%16.16" PRIx64 " - 0x%16.16" PRIx64
730                   ") slide = 0x%" PRIx64,
731                   name.AsCString(""), vmaddr + slide, vmaddr + slide + vmsize,
732                   slide);
733   }
734 }
735 
736 const DynamicLoaderDarwin::Segment *
737 DynamicLoaderDarwin::ImageInfo::FindSegment(const ConstString &name) const {
738   const size_t num_segments = segments.size();
739   for (size_t i = 0; i < num_segments; ++i) {
740     if (segments[i].name == name)
741       return &segments[i];
742   }
743   return NULL;
744 }
745 
746 //----------------------------------------------------------------------
747 // Dump an image info structure to the file handle provided.
748 //----------------------------------------------------------------------
749 void DynamicLoaderDarwin::ImageInfo::PutToLog(Log *log) const {
750   if (!log)
751     return;
752   if (address == LLDB_INVALID_ADDRESS) {
753     LLDB_LOG(log, "modtime={0:x+8} uuid={1} path='{2}' (UNLOADED)", mod_date,
754              uuid.GetAsString(), file_spec.GetPath());
755   } else {
756     LLDB_LOG(log, "address={0:x+16} modtime={1:x+8} uuid={2} path='{3}'",
757              address, mod_date, uuid.GetAsString(), file_spec.GetPath());
758     for (uint32_t i = 0; i < segments.size(); ++i)
759       segments[i].PutToLog(log, slide);
760   }
761 }
762 
763 void DynamicLoaderDarwin::PrivateInitialize(Process *process) {
764   DEBUG_PRINTF("DynamicLoaderDarwin::%s() process state = %s\n", __FUNCTION__,
765                StateAsCString(m_process->GetState()));
766   Clear(true);
767   m_process = process;
768   m_process->GetTarget().ClearAllLoadedSections();
769 }
770 
771 //----------------------------------------------------------------------
772 // Member function that gets called when the process state changes.
773 //----------------------------------------------------------------------
774 void DynamicLoaderDarwin::PrivateProcessStateChanged(Process *process,
775                                                      StateType state) {
776   DEBUG_PRINTF("DynamicLoaderDarwin::%s(%s)\n", __FUNCTION__,
777                StateAsCString(state));
778   switch (state) {
779   case eStateConnected:
780   case eStateAttaching:
781   case eStateLaunching:
782   case eStateInvalid:
783   case eStateUnloaded:
784   case eStateExited:
785   case eStateDetached:
786     Clear(false);
787     break;
788 
789   case eStateStopped:
790     // Keep trying find dyld and set our notification breakpoint each time we
791     // stop until we succeed
792     if (!DidSetNotificationBreakpoint() && m_process->IsAlive()) {
793       if (NeedToDoInitialImageFetch())
794         DoInitialImageFetch();
795 
796       SetNotificationBreakpoint();
797     }
798     break;
799 
800   case eStateRunning:
801   case eStateStepping:
802   case eStateCrashed:
803   case eStateSuspended:
804     break;
805   }
806 }
807 
808 ThreadPlanSP
809 DynamicLoaderDarwin::GetStepThroughTrampolinePlan(Thread &thread,
810                                                   bool stop_others) {
811   ThreadPlanSP thread_plan_sp;
812   StackFrame *current_frame = thread.GetStackFrameAtIndex(0).get();
813   const SymbolContext &current_context =
814       current_frame->GetSymbolContext(eSymbolContextSymbol);
815   Symbol *current_symbol = current_context.symbol;
816   Log *log(lldb_private::GetLogIfAllCategoriesSet(LIBLLDB_LOG_STEP));
817   TargetSP target_sp(thread.CalculateTarget());
818 
819   if (current_symbol != NULL) {
820     std::vector<Address> addresses;
821 
822     if (current_symbol->IsTrampoline()) {
823       const ConstString &trampoline_name = current_symbol->GetMangled().GetName(
824           current_symbol->GetLanguage(), Mangled::ePreferMangled);
825 
826       if (trampoline_name) {
827         const ModuleList &images = target_sp->GetImages();
828 
829         SymbolContextList code_symbols;
830         images.FindSymbolsWithNameAndType(trampoline_name, eSymbolTypeCode,
831                                           code_symbols);
832         size_t num_code_symbols = code_symbols.GetSize();
833 
834         if (num_code_symbols > 0) {
835           for (uint32_t i = 0; i < num_code_symbols; i++) {
836             SymbolContext context;
837             AddressRange addr_range;
838             if (code_symbols.GetContextAtIndex(i, context)) {
839               context.GetAddressRange(eSymbolContextEverything, 0, false,
840                                       addr_range);
841               addresses.push_back(addr_range.GetBaseAddress());
842               if (log) {
843                 addr_t load_addr =
844                     addr_range.GetBaseAddress().GetLoadAddress(target_sp.get());
845 
846                 log->Printf("Found a trampoline target symbol at 0x%" PRIx64
847                             ".",
848                             load_addr);
849               }
850             }
851           }
852         }
853 
854         SymbolContextList reexported_symbols;
855         images.FindSymbolsWithNameAndType(
856             trampoline_name, eSymbolTypeReExported, reexported_symbols);
857         size_t num_reexported_symbols = reexported_symbols.GetSize();
858         if (num_reexported_symbols > 0) {
859           for (uint32_t i = 0; i < num_reexported_symbols; i++) {
860             SymbolContext context;
861             if (reexported_symbols.GetContextAtIndex(i, context)) {
862               if (context.symbol) {
863                 Symbol *actual_symbol =
864                     context.symbol->ResolveReExportedSymbol(*target_sp.get());
865                 if (actual_symbol) {
866                   const Address actual_symbol_addr =
867                       actual_symbol->GetAddress();
868                   if (actual_symbol_addr.IsValid()) {
869                     addresses.push_back(actual_symbol_addr);
870                     if (log) {
871                       lldb::addr_t load_addr =
872                           actual_symbol_addr.GetLoadAddress(target_sp.get());
873                       log->Printf(
874                           "Found a re-exported symbol: %s at 0x%" PRIx64 ".",
875                           actual_symbol->GetName().GetCString(), load_addr);
876                     }
877                   }
878                 }
879               }
880             }
881           }
882         }
883 
884         SymbolContextList indirect_symbols;
885         images.FindSymbolsWithNameAndType(trampoline_name, eSymbolTypeResolver,
886                                           indirect_symbols);
887         size_t num_indirect_symbols = indirect_symbols.GetSize();
888         if (num_indirect_symbols > 0) {
889           for (uint32_t i = 0; i < num_indirect_symbols; i++) {
890             SymbolContext context;
891             AddressRange addr_range;
892             if (indirect_symbols.GetContextAtIndex(i, context)) {
893               context.GetAddressRange(eSymbolContextEverything, 0, false,
894                                       addr_range);
895               addresses.push_back(addr_range.GetBaseAddress());
896               if (log) {
897                 addr_t load_addr =
898                     addr_range.GetBaseAddress().GetLoadAddress(target_sp.get());
899 
900                 log->Printf("Found an indirect target symbol at 0x%" PRIx64 ".",
901                             load_addr);
902               }
903             }
904           }
905         }
906       }
907     } else if (current_symbol->GetType() == eSymbolTypeReExported) {
908       // I am not sure we could ever end up stopped AT a re-exported symbol.
909       // But just in case:
910 
911       const Symbol *actual_symbol =
912           current_symbol->ResolveReExportedSymbol(*(target_sp.get()));
913       if (actual_symbol) {
914         Address target_addr(actual_symbol->GetAddress());
915         if (target_addr.IsValid()) {
916           if (log)
917             log->Printf(
918                 "Found a re-exported symbol: %s pointing to: %s at 0x%" PRIx64
919                 ".",
920                 current_symbol->GetName().GetCString(),
921                 actual_symbol->GetName().GetCString(),
922                 target_addr.GetLoadAddress(target_sp.get()));
923           addresses.push_back(target_addr.GetLoadAddress(target_sp.get()));
924         }
925       }
926     }
927 
928     if (addresses.size() > 0) {
929       // First check whether any of the addresses point to Indirect symbols,
930       // and if they do, resolve them:
931       std::vector<lldb::addr_t> load_addrs;
932       for (Address address : addresses) {
933         Symbol *symbol = address.CalculateSymbolContextSymbol();
934         if (symbol && symbol->IsIndirect()) {
935           Status error;
936           Address symbol_address = symbol->GetAddress();
937           addr_t resolved_addr = thread.GetProcess()->ResolveIndirectFunction(
938               &symbol_address, error);
939           if (error.Success()) {
940             load_addrs.push_back(resolved_addr);
941             if (log)
942               log->Printf("ResolveIndirectFunction found resolved target for "
943                           "%s at 0x%" PRIx64 ".",
944                           symbol->GetName().GetCString(), resolved_addr);
945           }
946         } else {
947           load_addrs.push_back(address.GetLoadAddress(target_sp.get()));
948         }
949       }
950       thread_plan_sp.reset(
951           new ThreadPlanRunToAddress(thread, load_addrs, stop_others));
952     }
953   } else {
954     if (log)
955       log->Printf("Could not find symbol for step through.");
956   }
957 
958   return thread_plan_sp;
959 }
960 
961 size_t DynamicLoaderDarwin::FindEquivalentSymbols(
962     lldb_private::Symbol *original_symbol, lldb_private::ModuleList &images,
963     lldb_private::SymbolContextList &equivalent_symbols) {
964   const ConstString &trampoline_name = original_symbol->GetMangled().GetName(
965       original_symbol->GetLanguage(), Mangled::ePreferMangled);
966   if (!trampoline_name)
967     return 0;
968 
969   size_t initial_size = equivalent_symbols.GetSize();
970 
971   static const char *resolver_name_regex = "(_gc|_non_gc|\\$[A-Za-z0-9\\$]+)$";
972   std::string equivalent_regex_buf("^");
973   equivalent_regex_buf.append(trampoline_name.GetCString());
974   equivalent_regex_buf.append(resolver_name_regex);
975 
976   RegularExpression equivalent_name_regex(equivalent_regex_buf);
977   const bool append = true;
978   images.FindSymbolsMatchingRegExAndType(equivalent_name_regex, eSymbolTypeCode,
979                                          equivalent_symbols, append);
980 
981   return equivalent_symbols.GetSize() - initial_size;
982 }
983 
984 lldb::ModuleSP DynamicLoaderDarwin::GetPThreadLibraryModule() {
985   ModuleSP module_sp = m_libpthread_module_wp.lock();
986   if (!module_sp) {
987     SymbolContextList sc_list;
988     ModuleSpec module_spec;
989     module_spec.GetFileSpec().GetFilename().SetCString(
990         "libsystem_pthread.dylib");
991     ModuleList module_list;
992     if (m_process->GetTarget().GetImages().FindModules(module_spec,
993                                                        module_list)) {
994       if (module_list.GetSize() == 1) {
995         module_sp = module_list.GetModuleAtIndex(0);
996         if (module_sp)
997           m_libpthread_module_wp = module_sp;
998       }
999     }
1000   }
1001   return module_sp;
1002 }
1003 
1004 Address DynamicLoaderDarwin::GetPthreadSetSpecificAddress() {
1005   if (!m_pthread_getspecific_addr.IsValid()) {
1006     ModuleSP module_sp = GetPThreadLibraryModule();
1007     if (module_sp) {
1008       lldb_private::SymbolContextList sc_list;
1009       module_sp->FindSymbolsWithNameAndType(ConstString("pthread_getspecific"),
1010                                             eSymbolTypeCode, sc_list);
1011       SymbolContext sc;
1012       if (sc_list.GetContextAtIndex(0, sc)) {
1013         if (sc.symbol)
1014           m_pthread_getspecific_addr = sc.symbol->GetAddress();
1015       }
1016     }
1017   }
1018   return m_pthread_getspecific_addr;
1019 }
1020 
1021 lldb::addr_t
1022 DynamicLoaderDarwin::GetThreadLocalData(const lldb::ModuleSP module_sp,
1023                                         const lldb::ThreadSP thread_sp,
1024                                         lldb::addr_t tls_file_addr) {
1025   if (!thread_sp || !module_sp)
1026     return LLDB_INVALID_ADDRESS;
1027 
1028   std::lock_guard<std::recursive_mutex> guard(m_mutex);
1029 
1030   const uint32_t addr_size = m_process->GetAddressByteSize();
1031   uint8_t buf[sizeof(lldb::addr_t) * 3];
1032 
1033   lldb_private::Address tls_addr;
1034   if (module_sp->ResolveFileAddress(tls_file_addr, tls_addr)) {
1035     Status error;
1036     const size_t tsl_data_size = addr_size * 3;
1037     Target &target = m_process->GetTarget();
1038     if (target.ReadMemory(tls_addr, false, buf, tsl_data_size, error) ==
1039         tsl_data_size) {
1040       const ByteOrder byte_order = m_process->GetByteOrder();
1041       DataExtractor data(buf, sizeof(buf), byte_order, addr_size);
1042       lldb::offset_t offset = addr_size; // Skip the first pointer
1043       const lldb::addr_t pthread_key = data.GetAddress(&offset);
1044       const lldb::addr_t tls_offset = data.GetAddress(&offset);
1045       if (pthread_key != 0) {
1046         // First check to see if we have already figured out the location of
1047         // TLS data for the pthread_key on a specific thread yet. If we have we
1048         // can re-use it since its location will not change unless the process
1049         // execs.
1050         const tid_t tid = thread_sp->GetID();
1051         auto tid_pos = m_tid_to_tls_map.find(tid);
1052         if (tid_pos != m_tid_to_tls_map.end()) {
1053           auto tls_pos = tid_pos->second.find(pthread_key);
1054           if (tls_pos != tid_pos->second.end()) {
1055             return tls_pos->second + tls_offset;
1056           }
1057         }
1058         StackFrameSP frame_sp = thread_sp->GetStackFrameAtIndex(0);
1059         if (frame_sp) {
1060           ClangASTContext *clang_ast_context =
1061               target.GetScratchClangASTContext();
1062 
1063           if (!clang_ast_context)
1064             return LLDB_INVALID_ADDRESS;
1065 
1066           CompilerType clang_void_ptr_type =
1067               clang_ast_context->GetBasicType(eBasicTypeVoid).GetPointerType();
1068           Address pthread_getspecific_addr = GetPthreadSetSpecificAddress();
1069           if (pthread_getspecific_addr.IsValid()) {
1070             EvaluateExpressionOptions options;
1071 
1072             lldb::ThreadPlanSP thread_plan_sp(new ThreadPlanCallFunction(
1073                 *thread_sp, pthread_getspecific_addr, clang_void_ptr_type,
1074                 llvm::ArrayRef<lldb::addr_t>(pthread_key), options));
1075 
1076             DiagnosticManager execution_errors;
1077             ExecutionContext exe_ctx(thread_sp);
1078             lldb::ExpressionResults results = m_process->RunThreadPlan(
1079                 exe_ctx, thread_plan_sp, options, execution_errors);
1080 
1081             if (results == lldb::eExpressionCompleted) {
1082               lldb::ValueObjectSP result_valobj_sp =
1083                   thread_plan_sp->GetReturnValueObject();
1084               if (result_valobj_sp) {
1085                 const lldb::addr_t pthread_key_data =
1086                     result_valobj_sp->GetValueAsUnsigned(0);
1087                 if (pthread_key_data) {
1088                   m_tid_to_tls_map[tid].insert(
1089                       std::make_pair(pthread_key, pthread_key_data));
1090                   return pthread_key_data + tls_offset;
1091                 }
1092               }
1093             }
1094           }
1095         }
1096       }
1097     }
1098   }
1099   return LLDB_INVALID_ADDRESS;
1100 }
1101 
1102 bool DynamicLoaderDarwin::UseDYLDSPI(Process *process) {
1103   Log *log(lldb_private::GetLogIfAnyCategoriesSet(LIBLLDB_LOG_DYNAMIC_LOADER));
1104   bool use_new_spi_interface = false;
1105 
1106   llvm::VersionTuple version = process->GetHostOSVersion();
1107   if (!version.empty()) {
1108     const llvm::Triple::OSType os_type =
1109         process->GetTarget().GetArchitecture().GetTriple().getOS();
1110 
1111     // macOS 10.12 and newer
1112     if (os_type == llvm::Triple::MacOSX &&
1113         version >= llvm::VersionTuple(10, 12))
1114       use_new_spi_interface = true;
1115 
1116     // iOS 10 and newer
1117     if (os_type == llvm::Triple::IOS && version >= llvm::VersionTuple(10))
1118       use_new_spi_interface = true;
1119 
1120     // tvOS 10 and newer
1121     if (os_type == llvm::Triple::TvOS && version >= llvm::VersionTuple(10))
1122       use_new_spi_interface = true;
1123 
1124     // watchOS 3 and newer
1125     if (os_type == llvm::Triple::WatchOS && version >= llvm::VersionTuple(3))
1126       use_new_spi_interface = true;
1127   }
1128 
1129   if (log) {
1130     if (use_new_spi_interface)
1131       log->Printf(
1132           "DynamicLoaderDarwin::UseDYLDSPI: Use new DynamicLoader plugin");
1133     else
1134       log->Printf(
1135           "DynamicLoaderDarwin::UseDYLDSPI: Use old DynamicLoader plugin");
1136   }
1137   return use_new_spi_interface;
1138 }
1139