X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=tools%2Fllvm-objdump%2FMachODump.cpp;h=50afab7b54c10b7696d82268de963206cd90da01;hb=a25032214d92adb613a5db84adeec88e71d59b30;hp=fe9513a4c77553937c73a9c4aa24750af091913e;hpb=9bf0327ea77669212b10325f934c6efe1d011f3e;p=oota-llvm.git diff --git a/tools/llvm-objdump/MachODump.cpp b/tools/llvm-objdump/MachODump.cpp index fe9513a4c77..50afab7b54c 100644 --- a/tools/llvm-objdump/MachODump.cpp +++ b/tools/llvm-objdump/MachODump.cpp @@ -12,9 +12,11 @@ //===----------------------------------------------------------------------===// #include "llvm-objdump.h" +#include "llvm-c/Disassembler.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/StringExtras.h" #include "llvm/ADT/Triple.h" +#include "llvm/Config/config.h" #include "llvm/DebugInfo/DIContext.h" #include "llvm/MC/MCAsmInfo.h" #include "llvm/MC/MCContext.h" @@ -35,20 +37,34 @@ #include "llvm/Support/GraphWriter.h" #include "llvm/Support/MachO.h" #include "llvm/Support/MemoryBuffer.h" +#include "llvm/Support/FormattedStream.h" #include "llvm/Support/TargetRegistry.h" #include "llvm/Support/TargetSelect.h" #include "llvm/Support/raw_ostream.h" #include #include #include + +#if HAVE_CXXABI_H +#include +#endif + using namespace llvm; using namespace object; static cl::opt - UseDbg("g", cl::desc("Print line information from debug info if available")); + UseDbg("g", + cl::desc("Print line information from debug info if available")); -static cl::opt - DSYMFile("dsym", cl::desc("Use .dSYM file for debug info")); +static cl::opt DSYMFile("dsym", + cl::desc("Use .dSYM file for debug info")); + +static cl::opt FullLeadingAddr("full-leading-addr", + cl::desc("Print full leading address")); + +static cl::opt + PrintImmHex("print-imm-hex", + cl::desc("Use hex format for immediate values")); static std::string ThumbTripleName; @@ -108,9 +124,8 @@ typedef std::pair DiceTableEntry; typedef std::vector DiceTable; typedef DiceTable::iterator dice_table_iterator; -static bool -compareDiceTableEntries(const DiceTableEntry i, - const DiceTableEntry j) { +static bool compareDiceTableEntries(const DiceTableEntry i, + const DiceTableEntry j) { return i.first == j.first; } @@ -122,15 +137,11 @@ static void DumpDataInCode(const char *bytes, uint64_t Size, case MachO::DICE_KIND_DATA: switch (Size) { case 4: - Value = bytes[3] << 24 | - bytes[2] << 16 | - bytes[1] << 8 | - bytes[0]; + Value = bytes[3] << 24 | bytes[2] << 16 | bytes[1] << 8 | bytes[0]; outs() << "\t.long " << Value; break; case 2: - Value = bytes[1] << 8 | - bytes[0]; + Value = bytes[1] << 8 | bytes[0]; outs() << "\t.short " << Value; break; case 1: @@ -145,15 +156,11 @@ static void DumpDataInCode(const char *bytes, uint64_t Size, outs() << "\t.byte " << Value << "\t@ KIND_JUMP_TABLE8"; break; case MachO::DICE_KIND_JUMP_TABLE16: - Value = bytes[1] << 8 | - bytes[0]; + Value = bytes[1] << 8 | bytes[0]; outs() << "\t.short " << Value << "\t@ KIND_JUMP_TABLE16"; break; case MachO::DICE_KIND_JUMP_TABLE32: - Value = bytes[3] << 24 | - bytes[2] << 16 | - bytes[1] << 8 | - bytes[0]; + Value = bytes[3] << 24 | bytes[2] << 16 | bytes[1] << 8 | bytes[0]; outs() << "\t.long " << Value << "\t@ KIND_JUMP_TABLE32"; break; default: @@ -180,20 +187,18 @@ static void getSectionsAndSymbols(const MachO::mach_header Header, MachOObjectFile::LoadCommandInfo Command = MachOObj->getFirstLoadCommandInfo(); bool BaseSegmentAddressSet = false; - for (unsigned i = 0; ; ++i) { + for (unsigned i = 0;; ++i) { if (Command.C.cmd == MachO::LC_FUNCTION_STARTS) { // We found a function starts segment, parse the addresses for later // consumption. MachO::linkedit_data_command LLC = - MachOObj->getLinkeditDataLoadCommand(Command); + MachOObj->getLinkeditDataLoadCommand(Command); MachOObj->ReadULEB128s(LLC.dataoff, FoundFns); - } - else if (Command.C.cmd == MachO::LC_SEGMENT) { - MachO::segment_command SLC = - MachOObj->getSegmentLoadCommand(Command); + } else if (Command.C.cmd == MachO::LC_SEGMENT) { + MachO::segment_command SLC = MachOObj->getSegmentLoadCommand(Command); StringRef SegName = SLC.segname; - if(!BaseSegmentAddressSet && SegName != "__PAGEZERO") { + if (!BaseSegmentAddressSet && SegName != "__PAGEZERO") { BaseSegmentAddressSet = true; BaseSegmentAddress = SLC.vmaddr; } @@ -224,6 +229,1047 @@ void llvm::DisassembleInputMachO(StringRef Filename) { DisassembleInputMachO2(Filename, MachOOF.get()); } +typedef DenseMap SymbolAddressMap; +typedef std::pair BindInfoEntry; +typedef std::vector BindTable; +typedef BindTable::iterator bind_table_iterator; + +// The block of info used by the Symbolizer call backs. +struct DisassembleInfo { + bool verbose; + MachOObjectFile *O; + SectionRef S; + SymbolAddressMap *AddrMap; + std::vector *Sections; + const char *class_name; + const char *selector_name; + char *method; + char *demangled_name; + BindTable *bindtable; +}; + +// GuessSymbolName is passed the address of what might be a symbol and a +// pointer to the DisassembleInfo struct. It returns the name of a symbol +// with that address or nullptr if no symbol is found with that address. +static const char *GuessSymbolName(uint64_t value, + struct DisassembleInfo *info) { + const char *SymbolName = nullptr; + // A DenseMap can't lookup up some values. + if (value != 0xffffffffffffffffULL && value != 0xfffffffffffffffeULL) { + StringRef name = info->AddrMap->lookup(value); + if (!name.empty()) + SymbolName = name.data(); + } + return SymbolName; +} + +// SymbolizerGetOpInfo() is the operand information call back function. +// This is called to get the symbolic information for operand(s) of an +// instruction when it is being done. This routine does this from +// the relocation information, symbol table, etc. That block of information +// is a pointer to the struct DisassembleInfo that was passed when the +// disassembler context was created and passed to back to here when +// called back by the disassembler for instruction operands that could have +// relocation information. The address of the instruction containing operand is +// at the Pc parameter. The immediate value the operand has is passed in +// op_info->Value and is at Offset past the start of the instruction and has a +// byte Size of 1, 2 or 4. The symbolc information is returned in TagBuf is the +// LLVMOpInfo1 struct defined in the header "llvm-c/Disassembler.h" as symbol +// names and addends of the symbolic expression to add for the operand. The +// value of TagType is currently 1 (for the LLVMOpInfo1 struct). If symbolic +// information is returned then this function returns 1 else it returns 0. +int SymbolizerGetOpInfo(void *DisInfo, uint64_t Pc, uint64_t Offset, + uint64_t Size, int TagType, void *TagBuf) { + struct DisassembleInfo *info = (struct DisassembleInfo *)DisInfo; + struct LLVMOpInfo1 *op_info = (struct LLVMOpInfo1 *)TagBuf; + unsigned int value = op_info->Value; + + // Make sure all fields returned are zero if we don't set them. + memset((void *)op_info, '\0', sizeof(struct LLVMOpInfo1)); + op_info->Value = value; + + // If the TagType is not the value 1 which it code knows about or if no + // verbose symbolic information is wanted then just return 0, indicating no + // information is being returned. + if (TagType != 1 || info->verbose == false) + return 0; + + unsigned int Arch = info->O->getArch(); + if (Arch == Triple::x86) { + if (Size != 1 && Size != 2 && Size != 4 && Size != 0) + return 0; + // First search the section's relocation entries (if any) for an entry + // for this section offset. + uint32_t sect_addr = info->S.getAddress(); + uint32_t sect_offset = (Pc + Offset) - sect_addr; + bool reloc_found = false; + DataRefImpl Rel; + MachO::any_relocation_info RE; + bool isExtern = false; + SymbolRef Symbol; + bool r_scattered = false; + uint32_t r_value, pair_r_value, r_type; + for (const RelocationRef &Reloc : info->S.relocations()) { + uint64_t RelocOffset; + Reloc.getOffset(RelocOffset); + if (RelocOffset == sect_offset) { + Rel = Reloc.getRawDataRefImpl(); + RE = info->O->getRelocation(Rel); + r_scattered = info->O->isRelocationScattered(RE); + if (r_scattered) { + r_value = info->O->getScatteredRelocationValue(RE); + r_type = info->O->getScatteredRelocationType(RE); + if (r_type == MachO::GENERIC_RELOC_SECTDIFF || + r_type == MachO::GENERIC_RELOC_LOCAL_SECTDIFF) { + DataRefImpl RelNext = Rel; + info->O->moveRelocationNext(RelNext); + MachO::any_relocation_info RENext; + RENext = info->O->getRelocation(RelNext); + if (info->O->isRelocationScattered(RENext)) + pair_r_value = info->O->getPlainRelocationSymbolNum(RENext); + else + return 0; + } + } else { + isExtern = info->O->getPlainRelocationExternal(RE); + if (isExtern) { + symbol_iterator RelocSym = Reloc.getSymbol(); + Symbol = *RelocSym; + } + } + reloc_found = true; + break; + } + } + if (reloc_found && isExtern) { + StringRef SymName; + Symbol.getName(SymName); + const char *name = SymName.data(); + op_info->AddSymbol.Present = 1; + op_info->AddSymbol.Name = name; + // For i386 extern relocation entries the value in the instruction is + // the offset from the symbol, and value is already set in op_info->Value. + return 1; + } + if (reloc_found && (r_type == MachO::GENERIC_RELOC_SECTDIFF || + r_type == MachO::GENERIC_RELOC_LOCAL_SECTDIFF)) { + const char *add = GuessSymbolName(r_value, info); + const char *sub = GuessSymbolName(pair_r_value, info); + uint32_t offset = value - (r_value - pair_r_value); + op_info->AddSymbol.Present = 1; + if (add != nullptr) + op_info->AddSymbol.Name = add; + else + op_info->AddSymbol.Value = r_value; + op_info->SubtractSymbol.Present = 1; + if (sub != nullptr) + op_info->SubtractSymbol.Name = sub; + else + op_info->SubtractSymbol.Value = pair_r_value; + op_info->Value = offset; + return 1; + } + // TODO: + // Second search the external relocation entries of a fully linked image + // (if any) for an entry that matches this segment offset. + // uint32_t seg_offset = (Pc + Offset); + return 0; + } else if (Arch == Triple::x86_64) { + if (Size != 1 && Size != 2 && Size != 4 && Size != 0) + return 0; + // First search the section's relocation entries (if any) for an entry + // for this section offset. + uint64_t sect_addr = info->S.getAddress(); + uint64_t sect_offset = (Pc + Offset) - sect_addr; + bool reloc_found = false; + DataRefImpl Rel; + MachO::any_relocation_info RE; + bool isExtern = false; + SymbolRef Symbol; + for (const RelocationRef &Reloc : info->S.relocations()) { + uint64_t RelocOffset; + Reloc.getOffset(RelocOffset); + if (RelocOffset == sect_offset) { + Rel = Reloc.getRawDataRefImpl(); + RE = info->O->getRelocation(Rel); + // NOTE: Scattered relocations don't exist on x86_64. + isExtern = info->O->getPlainRelocationExternal(RE); + if (isExtern) { + symbol_iterator RelocSym = Reloc.getSymbol(); + Symbol = *RelocSym; + } + reloc_found = true; + break; + } + } + if (reloc_found && isExtern) { + // The Value passed in will be adjusted by the Pc if the instruction + // adds the Pc. But for x86_64 external relocation entries the Value + // is the offset from the external symbol. + if (info->O->getAnyRelocationPCRel(RE)) + op_info->Value -= Pc + Offset + Size; + StringRef SymName; + Symbol.getName(SymName); + const char *name = SymName.data(); + unsigned Type = info->O->getAnyRelocationType(RE); + if (Type == MachO::X86_64_RELOC_SUBTRACTOR) { + DataRefImpl RelNext = Rel; + info->O->moveRelocationNext(RelNext); + MachO::any_relocation_info RENext = info->O->getRelocation(RelNext); + unsigned TypeNext = info->O->getAnyRelocationType(RENext); + bool isExternNext = info->O->getPlainRelocationExternal(RENext); + unsigned SymbolNum = info->O->getPlainRelocationSymbolNum(RENext); + if (TypeNext == MachO::X86_64_RELOC_UNSIGNED && isExternNext) { + op_info->SubtractSymbol.Present = 1; + op_info->SubtractSymbol.Name = name; + symbol_iterator RelocSymNext = info->O->getSymbolByIndex(SymbolNum); + Symbol = *RelocSymNext; + StringRef SymNameNext; + Symbol.getName(SymNameNext); + name = SymNameNext.data(); + } + } + // TODO: add the VariantKinds to op_info->VariantKind for relocation types + // like: X86_64_RELOC_TLV, X86_64_RELOC_GOT_LOAD and X86_64_RELOC_GOT. + op_info->AddSymbol.Present = 1; + op_info->AddSymbol.Name = name; + return 1; + } + // TODO: + // Second search the external relocation entries of a fully linked image + // (if any) for an entry that matches this segment offset. + // uint64_t seg_offset = (Pc + Offset); + return 0; + } else if (Arch == Triple::arm) { + return 0; + } else if (Arch == Triple::aarch64) { + return 0; + } else { + return 0; + } +} + +// GuessCstringPointer is passed the address of what might be a pointer to a +// literal string in a cstring section. If that address is in a cstring section +// it returns a pointer to that string. Else it returns nullptr. +const char *GuessCstringPointer(uint64_t ReferenceValue, + struct DisassembleInfo *info) { + uint32_t LoadCommandCount = info->O->getHeader().ncmds; + MachOObjectFile::LoadCommandInfo Load = info->O->getFirstLoadCommandInfo(); + for (unsigned I = 0;; ++I) { + if (Load.C.cmd == MachO::LC_SEGMENT_64) { + MachO::segment_command_64 Seg = info->O->getSegment64LoadCommand(Load); + for (unsigned J = 0; J < Seg.nsects; ++J) { + MachO::section_64 Sec = info->O->getSection64(Load, J); + uint32_t section_type = Sec.flags & MachO::SECTION_TYPE; + if (section_type == MachO::S_CSTRING_LITERALS && + ReferenceValue >= Sec.addr && + ReferenceValue < Sec.addr + Sec.size) { + uint64_t sect_offset = ReferenceValue - Sec.addr; + uint64_t object_offset = Sec.offset + sect_offset; + StringRef MachOContents = info->O->getData(); + uint64_t object_size = MachOContents.size(); + const char *object_addr = (const char *)MachOContents.data(); + if (object_offset < object_size) { + const char *name = object_addr + object_offset; + return name; + } else { + return nullptr; + } + } + } + } else if (Load.C.cmd == MachO::LC_SEGMENT) { + MachO::segment_command Seg = info->O->getSegmentLoadCommand(Load); + for (unsigned J = 0; J < Seg.nsects; ++J) { + MachO::section Sec = info->O->getSection(Load, J); + uint32_t section_type = Sec.flags & MachO::SECTION_TYPE; + if (section_type == MachO::S_CSTRING_LITERALS && + ReferenceValue >= Sec.addr && + ReferenceValue < Sec.addr + Sec.size) { + uint64_t sect_offset = ReferenceValue - Sec.addr; + uint64_t object_offset = Sec.offset + sect_offset; + StringRef MachOContents = info->O->getData(); + uint64_t object_size = MachOContents.size(); + const char *object_addr = (const char *)MachOContents.data(); + if (object_offset < object_size) { + const char *name = object_addr + object_offset; + return name; + } else { + return nullptr; + } + } + } + } + if (I == LoadCommandCount - 1) + break; + else + Load = info->O->getNextLoadCommandInfo(Load); + } + return nullptr; +} + +// GuessIndirectSymbol returns the name of the indirect symbol for the +// ReferenceValue passed in or nullptr. This is used when ReferenceValue maybe +// an address of a symbol stub or a lazy or non-lazy pointer to associate the +// symbol name being referenced by the stub or pointer. +static const char *GuessIndirectSymbol(uint64_t ReferenceValue, + struct DisassembleInfo *info) { + uint32_t LoadCommandCount = info->O->getHeader().ncmds; + MachOObjectFile::LoadCommandInfo Load = info->O->getFirstLoadCommandInfo(); + MachO::dysymtab_command Dysymtab = info->O->getDysymtabLoadCommand(); + MachO::symtab_command Symtab = info->O->getSymtabLoadCommand(); + for (unsigned I = 0;; ++I) { + if (Load.C.cmd == MachO::LC_SEGMENT_64) { + MachO::segment_command_64 Seg = info->O->getSegment64LoadCommand(Load); + for (unsigned J = 0; J < Seg.nsects; ++J) { + MachO::section_64 Sec = info->O->getSection64(Load, J); + uint32_t section_type = Sec.flags & MachO::SECTION_TYPE; + if ((section_type == MachO::S_NON_LAZY_SYMBOL_POINTERS || + section_type == MachO::S_LAZY_SYMBOL_POINTERS || + section_type == MachO::S_LAZY_DYLIB_SYMBOL_POINTERS || + section_type == MachO::S_THREAD_LOCAL_VARIABLE_POINTERS || + section_type == MachO::S_SYMBOL_STUBS) && + ReferenceValue >= Sec.addr && + ReferenceValue < Sec.addr + Sec.size) { + uint32_t stride; + if (section_type == MachO::S_SYMBOL_STUBS) + stride = Sec.reserved2; + else + stride = 8; + if (stride == 0) + return nullptr; + uint32_t index = Sec.reserved1 + (ReferenceValue - Sec.addr) / stride; + if (index < Dysymtab.nindirectsyms) { + uint32_t indirect_symbol = + info->O->getIndirectSymbolTableEntry(Dysymtab, index); + if (indirect_symbol < Symtab.nsyms) { + symbol_iterator Sym = info->O->getSymbolByIndex(indirect_symbol); + SymbolRef Symbol = *Sym; + StringRef SymName; + Symbol.getName(SymName); + const char *name = SymName.data(); + return name; + } + } + } + } + } else if (Load.C.cmd == MachO::LC_SEGMENT) { + MachO::segment_command Seg = info->O->getSegmentLoadCommand(Load); + for (unsigned J = 0; J < Seg.nsects; ++J) { + MachO::section Sec = info->O->getSection(Load, J); + uint32_t section_type = Sec.flags & MachO::SECTION_TYPE; + if ((section_type == MachO::S_NON_LAZY_SYMBOL_POINTERS || + section_type == MachO::S_LAZY_SYMBOL_POINTERS || + section_type == MachO::S_LAZY_DYLIB_SYMBOL_POINTERS || + section_type == MachO::S_THREAD_LOCAL_VARIABLE_POINTERS || + section_type == MachO::S_SYMBOL_STUBS) && + ReferenceValue >= Sec.addr && + ReferenceValue < Sec.addr + Sec.size) { + uint32_t stride; + if (section_type == MachO::S_SYMBOL_STUBS) + stride = Sec.reserved2; + else + stride = 4; + if (stride == 0) + return nullptr; + uint32_t index = Sec.reserved1 + (ReferenceValue - Sec.addr) / stride; + if (index < Dysymtab.nindirectsyms) { + uint32_t indirect_symbol = + info->O->getIndirectSymbolTableEntry(Dysymtab, index); + if (indirect_symbol < Symtab.nsyms) { + symbol_iterator Sym = info->O->getSymbolByIndex(indirect_symbol); + SymbolRef Symbol = *Sym; + StringRef SymName; + Symbol.getName(SymName); + const char *name = SymName.data(); + return name; + } + } + } + } + } + if (I == LoadCommandCount - 1) + break; + else + Load = info->O->getNextLoadCommandInfo(Load); + } + return nullptr; +} + +// method_reference() is called passing it the ReferenceName that might be +// a reference it to an Objective-C method call. If so then it allocates and +// assembles a method call string with the values last seen and saved in +// the DisassembleInfo's class_name and selector_name fields. This is saved +// into the method field of the info and any previous string is free'ed. +// Then the class_name field in the info is set to nullptr. The method call +// string is set into ReferenceName and ReferenceType is set to +// LLVMDisassembler_ReferenceType_Out_Objc_Message. If this not a method call +// then both ReferenceType and ReferenceName are left unchanged. +static void method_reference(struct DisassembleInfo *info, + uint64_t *ReferenceType, + const char **ReferenceName) { + if (*ReferenceName != nullptr) { + if (strcmp(*ReferenceName, "_objc_msgSend") == 0) { + if (info->selector_name != NULL) { + if (info->method != nullptr) + free(info->method); + if (info->class_name != nullptr) { + info->method = (char *)malloc(5 + strlen(info->class_name) + + strlen(info->selector_name)); + if (info->method != nullptr) { + strcpy(info->method, "+["); + strcat(info->method, info->class_name); + strcat(info->method, " "); + strcat(info->method, info->selector_name); + strcat(info->method, "]"); + *ReferenceName = info->method; + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Message; + } + } else { + info->method = (char *)malloc(9 + strlen(info->selector_name)); + if (info->method != nullptr) { + strcpy(info->method, "-[%rdi "); + strcat(info->method, info->selector_name); + strcat(info->method, "]"); + *ReferenceName = info->method; + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Message; + } + } + info->class_name = nullptr; + } + } else if (strcmp(*ReferenceName, "_objc_msgSendSuper2") == 0) { + if (info->selector_name != NULL) { + if (info->method != nullptr) + free(info->method); + info->method = (char *)malloc(17 + strlen(info->selector_name)); + if (info->method != nullptr) { + strcpy(info->method, "-[[%rdi super] "); + strcat(info->method, info->selector_name); + strcat(info->method, "]"); + *ReferenceName = info->method; + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Message; + } + info->class_name = nullptr; + } + } + } +} + +// GuessPointerPointer() is passed the address of what might be a pointer to +// a reference to an Objective-C class, selector, message ref or cfstring. +// If so the value of the pointer is returned and one of the booleans are set +// to true. If not zero is returned and all the booleans are set to false. +static uint64_t GuessPointerPointer(uint64_t ReferenceValue, + struct DisassembleInfo *info, + bool &classref, bool &selref, bool &msgref, + bool &cfstring) { + classref = false; + selref = false; + msgref = false; + cfstring = false; + uint32_t LoadCommandCount = info->O->getHeader().ncmds; + MachOObjectFile::LoadCommandInfo Load = info->O->getFirstLoadCommandInfo(); + for (unsigned I = 0;; ++I) { + if (Load.C.cmd == MachO::LC_SEGMENT_64) { + MachO::segment_command_64 Seg = info->O->getSegment64LoadCommand(Load); + for (unsigned J = 0; J < Seg.nsects; ++J) { + MachO::section_64 Sec = info->O->getSection64(Load, J); + if ((strncmp(Sec.sectname, "__objc_selrefs", 16) == 0 || + strncmp(Sec.sectname, "__objc_classrefs", 16) == 0 || + strncmp(Sec.sectname, "__objc_superrefs", 16) == 0 || + strncmp(Sec.sectname, "__objc_msgrefs", 16) == 0 || + strncmp(Sec.sectname, "__cfstring", 16) == 0) && + ReferenceValue >= Sec.addr && + ReferenceValue < Sec.addr + Sec.size) { + uint64_t sect_offset = ReferenceValue - Sec.addr; + uint64_t object_offset = Sec.offset + sect_offset; + StringRef MachOContents = info->O->getData(); + uint64_t object_size = MachOContents.size(); + const char *object_addr = (const char *)MachOContents.data(); + if (object_offset < object_size) { + uint64_t pointer_value; + memcpy(&pointer_value, object_addr + object_offset, + sizeof(uint64_t)); + if (info->O->isLittleEndian() != sys::IsLittleEndianHost) + sys::swapByteOrder(pointer_value); + if (strncmp(Sec.sectname, "__objc_selrefs", 16) == 0) + selref = true; + else if (strncmp(Sec.sectname, "__objc_classrefs", 16) == 0 || + strncmp(Sec.sectname, "__objc_superrefs", 16) == 0) + classref = true; + else if (strncmp(Sec.sectname, "__objc_msgrefs", 16) == 0 && + ReferenceValue + 8 < Sec.addr + Sec.size) { + msgref = true; + memcpy(&pointer_value, object_addr + object_offset + 8, + sizeof(uint64_t)); + if (info->O->isLittleEndian() != sys::IsLittleEndianHost) + sys::swapByteOrder(pointer_value); + } else if (strncmp(Sec.sectname, "__cfstring", 16) == 0) + cfstring = true; + return pointer_value; + } else { + return 0; + } + } + } + } + // TODO: Look for LC_SEGMENT for 32-bit Mach-O files. + if (I == LoadCommandCount - 1) + break; + else + Load = info->O->getNextLoadCommandInfo(Load); + } + return 0; +} + +// get_pointer_64 returns a pointer to the bytes in the object file at the +// Address from a section in the Mach-O file. And indirectly returns the +// offset into the section, number of bytes left in the section past the offset +// and which section is was being referenced. If the Address is not in a +// section nullptr is returned. +const char *get_pointer_64(uint64_t Address, uint32_t &offset, uint32_t &left, + SectionRef &S, DisassembleInfo *info) { + offset = 0; + left = 0; + S = SectionRef(); + for (unsigned SectIdx = 0; SectIdx != info->Sections->size(); SectIdx++) { + uint64_t SectAddress = ((*(info->Sections))[SectIdx]).getAddress(); + uint64_t SectSize = ((*(info->Sections))[SectIdx]).getSize(); + if (Address >= SectAddress && Address < SectAddress + SectSize) { + S = (*(info->Sections))[SectIdx]; + offset = Address - SectAddress; + left = SectSize - offset; + StringRef SectContents; + ((*(info->Sections))[SectIdx]).getContents(SectContents); + return SectContents.data() + offset; + } + } + return nullptr; +} + +// get_symbol_64() returns the name of a symbol (or nullptr) and the address of +// the symbol indirectly through n_value. Based on the relocation information +// for the specified section offset in the specified section reference. +const char *get_symbol_64(uint32_t sect_offset, SectionRef S, + DisassembleInfo *info, uint64_t &n_value) { + n_value = 0; + if (info->verbose == false) + return nullptr; + + // See if there is an external relocation entry at the sect_offset. + bool reloc_found = false; + DataRefImpl Rel; + MachO::any_relocation_info RE; + bool isExtern = false; + SymbolRef Symbol; + for (const RelocationRef &Reloc : S.relocations()) { + uint64_t RelocOffset; + Reloc.getOffset(RelocOffset); + if (RelocOffset == sect_offset) { + Rel = Reloc.getRawDataRefImpl(); + RE = info->O->getRelocation(Rel); + if (info->O->isRelocationScattered(RE)) + continue; + isExtern = info->O->getPlainRelocationExternal(RE); + if (isExtern) { + symbol_iterator RelocSym = Reloc.getSymbol(); + Symbol = *RelocSym; + } + reloc_found = true; + break; + } + } + // If there is an external relocation entry for a symbol in this section + // at this section_offset then use that symbol's value for the n_value + // and return its name. + const char *SymbolName = nullptr; + if (reloc_found && isExtern) { + Symbol.getAddress(n_value); + StringRef name; + Symbol.getName(name); + if (!name.empty()) { + SymbolName = name.data(); + return SymbolName; + } + } + + // TODO: For fully linked images, look through the external relocation + // entries off the dynamic symtab command. For these the r_offset is from the + // start of the first writeable segment in the Mach-O file. So the offset + // to this section from that segment is passed to this routine by the caller, + // as the database_offset. Which is the difference of the section's starting + // address and the first writable segment. + // + // NOTE: need add passing the database_offset to this routine. + + // TODO: We did not find an external relocation entry so look up the + // ReferenceValue as an address of a symbol and if found return that symbol's + // name. + // + // NOTE: need add passing the ReferenceValue to this routine. Then that code + // would simply be this: + // SymbolName = GuessSymbolName(ReferenceValue, info); + + return SymbolName; +} + +// These are structs in the Objective-C meta data and read to produce the +// comments for disassembly. While these are part of the ABI they are no +// public defintions. So the are here not in include/llvm/Support/MachO.h . + +// The cfstring object in a 64-bit Mach-O file. +struct cfstring64_t { + uint64_t isa; // class64_t * (64-bit pointer) + uint64_t flags; // flag bits + uint64_t characters; // char * (64-bit pointer) + uint64_t length; // number of non-NULL characters in above +}; + +// The class object in a 64-bit Mach-O file. +struct class64_t { + uint64_t isa; // class64_t * (64-bit pointer) + uint64_t superclass; // class64_t * (64-bit pointer) + uint64_t cache; // Cache (64-bit pointer) + uint64_t vtable; // IMP * (64-bit pointer) + uint64_t data; // class_ro64_t * (64-bit pointer) +}; + +struct class_ro64_t { + uint32_t flags; + uint32_t instanceStart; + uint32_t instanceSize; + uint32_t reserved; + uint64_t ivarLayout; // const uint8_t * (64-bit pointer) + uint64_t name; // const char * (64-bit pointer) + uint64_t baseMethods; // const method_list_t * (64-bit pointer) + uint64_t baseProtocols; // const protocol_list_t * (64-bit pointer) + uint64_t ivars; // const ivar_list_t * (64-bit pointer) + uint64_t weakIvarLayout; // const uint8_t * (64-bit pointer) + uint64_t baseProperties; // const struct objc_property_list (64-bit pointer) +}; + +inline void swapStruct(struct cfstring64_t &cfs) { + sys::swapByteOrder(cfs.isa); + sys::swapByteOrder(cfs.flags); + sys::swapByteOrder(cfs.characters); + sys::swapByteOrder(cfs.length); +} + +inline void swapStruct(struct class64_t &c) { + sys::swapByteOrder(c.isa); + sys::swapByteOrder(c.superclass); + sys::swapByteOrder(c.cache); + sys::swapByteOrder(c.vtable); + sys::swapByteOrder(c.data); +} + +inline void swapStruct(struct class_ro64_t &cro) { + sys::swapByteOrder(cro.flags); + sys::swapByteOrder(cro.instanceStart); + sys::swapByteOrder(cro.instanceSize); + sys::swapByteOrder(cro.reserved); + sys::swapByteOrder(cro.ivarLayout); + sys::swapByteOrder(cro.name); + sys::swapByteOrder(cro.baseMethods); + sys::swapByteOrder(cro.baseProtocols); + sys::swapByteOrder(cro.ivars); + sys::swapByteOrder(cro.weakIvarLayout); + sys::swapByteOrder(cro.baseProperties); +} + +static const char *get_dyld_bind_info_symbolname(uint64_t ReferenceValue, + struct DisassembleInfo *info); + +// get_objc2_64bit_class_name() is used for disassembly and is passed a pointer +// to an Objective-C class and returns the class name. It is also passed the +// address of the pointer, so when the pointer is zero as it can be in an .o +// file, that is used to look for an external relocation entry with a symbol +// name. +const char *get_objc2_64bit_class_name(uint64_t pointer_value, + uint64_t ReferenceValue, + struct DisassembleInfo *info) { + const char *r; + uint32_t offset, left; + SectionRef S; + + // The pointer_value can be 0 in an object file and have a relocation + // entry for the class symbol at the ReferenceValue (the address of the + // pointer). + if (pointer_value == 0) { + r = get_pointer_64(ReferenceValue, offset, left, S, info); + if (r == nullptr || left < sizeof(uint64_t)) + return nullptr; + uint64_t n_value; + const char *symbol_name = get_symbol_64(offset, S, info, n_value); + if (symbol_name == nullptr) + return nullptr; + const char *class_name = strrchr(symbol_name, '$'); + if (class_name != nullptr && class_name[1] == '_' && class_name[2] != '\0') + return class_name + 2; + else + return nullptr; + } + + // The case were the pointer_value is non-zero and points to a class defined + // in this Mach-O file. + r = get_pointer_64(pointer_value, offset, left, S, info); + if (r == nullptr || left < sizeof(struct class64_t)) + return nullptr; + struct class64_t c; + memcpy(&c, r, sizeof(struct class64_t)); + if (info->O->isLittleEndian() != sys::IsLittleEndianHost) + swapStruct(c); + if (c.data == 0) + return nullptr; + r = get_pointer_64(c.data, offset, left, S, info); + if (r == nullptr || left < sizeof(struct class_ro64_t)) + return nullptr; + struct class_ro64_t cro; + memcpy(&cro, r, sizeof(struct class_ro64_t)); + if (info->O->isLittleEndian() != sys::IsLittleEndianHost) + swapStruct(cro); + if (cro.name == 0) + return nullptr; + const char *name = get_pointer_64(cro.name, offset, left, S, info); + return name; +} + +// get_objc2_64bit_cfstring_name is used for disassembly and is passed a +// pointer to a cfstring and returns its name or nullptr. +const char *get_objc2_64bit_cfstring_name(uint64_t ReferenceValue, + struct DisassembleInfo *info) { + const char *r, *name; + uint32_t offset, left; + SectionRef S; + struct cfstring64_t cfs; + uint64_t cfs_characters; + + r = get_pointer_64(ReferenceValue, offset, left, S, info); + if (r == nullptr || left < sizeof(struct cfstring64_t)) + return nullptr; + memcpy(&cfs, r, sizeof(struct cfstring64_t)); + if (info->O->isLittleEndian() != sys::IsLittleEndianHost) + swapStruct(cfs); + if (cfs.characters == 0) { + uint64_t n_value; + const char *symbol_name = get_symbol_64( + offset + offsetof(struct cfstring64_t, characters), S, info, n_value); + if (symbol_name == nullptr) + return nullptr; + cfs_characters = n_value; + } else + cfs_characters = cfs.characters; + name = get_pointer_64(cfs_characters, offset, left, S, info); + + return name; +} + +// get_objc2_64bit_selref() is used for disassembly and is passed a the address +// of a pointer to an Objective-C selector reference when the pointer value is +// zero as in a .o file and is likely to have a external relocation entry with +// who's symbol's n_value is the real pointer to the selector name. If that is +// the case the real pointer to the selector name is returned else 0 is +// returned +uint64_t get_objc2_64bit_selref(uint64_t ReferenceValue, + struct DisassembleInfo *info) { + uint32_t offset, left; + SectionRef S; + + const char *r = get_pointer_64(ReferenceValue, offset, left, S, info); + if (r == nullptr || left < sizeof(uint64_t)) + return 0; + uint64_t n_value; + const char *symbol_name = get_symbol_64(offset, S, info, n_value); + if (symbol_name == nullptr) + return 0; + return n_value; +} + +// GuessLiteralPointer returns a string which for the item in the Mach-O file +// for the address passed in as ReferenceValue for printing as a comment with +// the instruction and also returns the corresponding type of that item +// indirectly through ReferenceType. +// +// If ReferenceValue is an address of literal cstring then a pointer to the +// cstring is returned and ReferenceType is set to +// LLVMDisassembler_ReferenceType_Out_LitPool_CstrAddr . +// +// If ReferenceValue is an address of an Objective-C CFString, Selector ref or +// Class ref that name is returned and the ReferenceType is set accordingly. +// +// Lastly, literals which are Symbol address in a literal pool are looked for +// and if found the symbol name is returned and ReferenceType is set to +// LLVMDisassembler_ReferenceType_Out_LitPool_SymAddr . +// +// If there is no item in the Mach-O file for the address passed in as +// ReferenceValue nullptr is returned and ReferenceType is unchanged. +const char *GuessLiteralPointer(uint64_t ReferenceValue, uint64_t ReferencePC, + uint64_t *ReferenceType, + struct DisassembleInfo *info) { + // TODO: This rouine's code and the routines it calls are only work with + // x86_64 Mach-O files for now. + unsigned int Arch = info->O->getArch(); + if (Arch != Triple::x86_64) + return nullptr; + + // First see if there is an external relocation entry at the ReferencePC. + uint64_t sect_addr = info->S.getAddress(); + uint64_t sect_offset = ReferencePC - sect_addr; + bool reloc_found = false; + DataRefImpl Rel; + MachO::any_relocation_info RE; + bool isExtern = false; + SymbolRef Symbol; + for (const RelocationRef &Reloc : info->S.relocations()) { + uint64_t RelocOffset; + Reloc.getOffset(RelocOffset); + if (RelocOffset == sect_offset) { + Rel = Reloc.getRawDataRefImpl(); + RE = info->O->getRelocation(Rel); + if (info->O->isRelocationScattered(RE)) + continue; + isExtern = info->O->getPlainRelocationExternal(RE); + if (isExtern) { + symbol_iterator RelocSym = Reloc.getSymbol(); + Symbol = *RelocSym; + } + reloc_found = true; + break; + } + } + // If there is an external relocation entry for a symbol in a section + // then used that symbol's value for the value of the reference. + if (reloc_found && isExtern) { + if (info->O->getAnyRelocationPCRel(RE)) { + unsigned Type = info->O->getAnyRelocationType(RE); + if (Type == MachO::X86_64_RELOC_SIGNED) { + Symbol.getAddress(ReferenceValue); + } + } + } + + // Look for literals such as Objective-C CFStrings refs, Selector refs, + // Message refs and Class refs. + bool classref, selref, msgref, cfstring; + uint64_t pointer_value = GuessPointerPointer(ReferenceValue, info, classref, + selref, msgref, cfstring); + if (classref == true && pointer_value == 0) { + // Note the ReferenceValue is a pointer into the __objc_classrefs section. + // And the pointer_value in that section is typically zero as it will be + // set by dyld as part of the "bind information". + const char *name = get_dyld_bind_info_symbolname(ReferenceValue, info); + if (name != nullptr) { + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Class_Ref; + const char *class_name = strrchr(name, '$'); + if (class_name != nullptr && class_name[1] == '_' && + class_name[2] != '\0') { + info->class_name = class_name + 2; + return name; + } + } + } + + if (classref == true) { + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Class_Ref; + const char *name = + get_objc2_64bit_class_name(pointer_value, ReferenceValue, info); + if (name != nullptr) + info->class_name = name; + else + name = "bad class ref"; + return name; + } + + if (cfstring == true) { + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_CFString_Ref; + const char *name = get_objc2_64bit_cfstring_name(ReferenceValue, info); + return name; + } + + if (selref == true && pointer_value == 0) + pointer_value = get_objc2_64bit_selref(ReferenceValue, info); + + if (pointer_value != 0) + ReferenceValue = pointer_value; + + const char *name = GuessCstringPointer(ReferenceValue, info); + if (name) { + if (pointer_value != 0 && selref == true) { + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Selector_Ref; + info->selector_name = name; + } else if (pointer_value != 0 && msgref == true) { + info->class_name = nullptr; + *ReferenceType = LLVMDisassembler_ReferenceType_Out_Objc_Message_Ref; + info->selector_name = name; + } else + *ReferenceType = LLVMDisassembler_ReferenceType_Out_LitPool_CstrAddr; + return name; + } + + // Lastly look for an indirect symbol with this ReferenceValue which is in + // a literal pool. If found return that symbol name. + name = GuessIndirectSymbol(ReferenceValue, info); + if (name) { + *ReferenceType = LLVMDisassembler_ReferenceType_Out_LitPool_SymAddr; + return name; + } + + return nullptr; +} + +// SymbolizerSymbolLookUp is the symbol lookup function passed when creating +// the Symbolizer. It looks up the ReferenceValue using the info passed via the +// pointer to the struct DisassembleInfo that was passed when MCSymbolizer +// is created and returns the symbol name that matches the ReferenceValue or +// nullptr if none. The ReferenceType is passed in for the IN type of +// reference the instruction is making from the values in defined in the header +// "llvm-c/Disassembler.h". On return the ReferenceType can set to a specific +// Out type and the ReferenceName will also be set which is added as a comment +// to the disassembled instruction. +// +#if HAVE_CXXABI_H +// If the symbol name is a C++ mangled name then the demangled name is +// returned through ReferenceName and ReferenceType is set to +// LLVMDisassembler_ReferenceType_DeMangled_Name . +#endif +// +// When this is called to get a symbol name for a branch target then the +// ReferenceType will be LLVMDisassembler_ReferenceType_In_Branch and then +// SymbolValue will be looked for in the indirect symbol table to determine if +// it is an address for a symbol stub. If so then the symbol name for that +// stub is returned indirectly through ReferenceName and then ReferenceType is +// set to LLVMDisassembler_ReferenceType_Out_SymbolStub. +// +// When this is called with an value loaded via a PC relative load then +// ReferenceType will be LLVMDisassembler_ReferenceType_In_PCrel_Load then the +// SymbolValue is checked to be an address of literal pointer, symbol pointer, +// or an Objective-C meta data reference. If so the output ReferenceType is +// set to correspond to that as well as setting the ReferenceName. +const char *SymbolizerSymbolLookUp(void *DisInfo, uint64_t ReferenceValue, + uint64_t *ReferenceType, + uint64_t ReferencePC, + const char **ReferenceName) { + struct DisassembleInfo *info = (struct DisassembleInfo *)DisInfo; + // If no verbose symbolic information is wanted then just return nullptr. + if (info->verbose == false) { + *ReferenceName = nullptr; + *ReferenceType = LLVMDisassembler_ReferenceType_InOut_None; + return nullptr; + } + + const char *SymbolName = GuessSymbolName(ReferenceValue, info); + + if (*ReferenceType == LLVMDisassembler_ReferenceType_In_Branch) { + *ReferenceName = GuessIndirectSymbol(ReferenceValue, info); + if (*ReferenceName != nullptr) { + method_reference(info, ReferenceType, ReferenceName); + if (*ReferenceType != LLVMDisassembler_ReferenceType_Out_Objc_Message) + *ReferenceType = LLVMDisassembler_ReferenceType_Out_SymbolStub; + } else +#if HAVE_CXXABI_H + if (SymbolName != nullptr && strncmp(SymbolName, "__Z", 3) == 0) { + if (info->demangled_name != nullptr) + free(info->demangled_name); + int status; + info->demangled_name = + abi::__cxa_demangle(SymbolName + 1, nullptr, nullptr, &status); + if (info->demangled_name != nullptr) { + *ReferenceName = info->demangled_name; + *ReferenceType = LLVMDisassembler_ReferenceType_DeMangled_Name; + } else + *ReferenceType = LLVMDisassembler_ReferenceType_InOut_None; + } else +#endif + *ReferenceType = LLVMDisassembler_ReferenceType_InOut_None; + } else if (*ReferenceType == LLVMDisassembler_ReferenceType_In_PCrel_Load) { + *ReferenceName = + GuessLiteralPointer(ReferenceValue, ReferencePC, ReferenceType, info); + if (*ReferenceName) + method_reference(info, ReferenceType, ReferenceName); + else + *ReferenceType = LLVMDisassembler_ReferenceType_InOut_None; + } +#if HAVE_CXXABI_H + else if (SymbolName != nullptr && strncmp(SymbolName, "__Z", 3) == 0) { + if (info->demangled_name != nullptr) + free(info->demangled_name); + int status; + info->demangled_name = + abi::__cxa_demangle(SymbolName + 1, nullptr, nullptr, &status); + if (info->demangled_name != nullptr) { + *ReferenceName = info->demangled_name; + *ReferenceType = LLVMDisassembler_ReferenceType_DeMangled_Name; + } + } +#endif + else { + *ReferenceName = nullptr; + *ReferenceType = LLVMDisassembler_ReferenceType_InOut_None; + } + + return SymbolName; +} + +// +// This is the memory object used by DisAsm->getInstruction() which has its +// BasePC. This then allows the 'address' parameter to getInstruction() to +// be the actual PC of the instruction. Then when a branch dispacement is +// added to the PC of an instruction, the 'ReferenceValue' passed to the +// SymbolizerSymbolLookUp() routine is the correct target addresses. As in +// the case of a fully linked Mach-O file where a section being disassembled +// generally not linked at address zero. +// +class DisasmMemoryObject : public MemoryObject { + const uint8_t *Bytes; + uint64_t Size; + uint64_t BasePC; + +public: + DisasmMemoryObject(const uint8_t *bytes, uint64_t size, uint64_t basePC) + : Bytes(bytes), Size(size), BasePC(basePC) {} + + uint64_t getBase() const override { return BasePC; } + uint64_t getExtent() const override { return Size; } + + int readByte(uint64_t Addr, uint8_t *Byte) const override { + if (Addr - BasePC >= Size) + return -1; + *Byte = Bytes[Addr - BasePC]; + return 0; + } +}; + +/// \brief Emits the comments that are stored in the CommentStream. +/// Each comment in the CommentStream must end with a newline. +static void emitComments(raw_svector_ostream &CommentStream, + SmallString<128> &CommentsToEmit, + formatted_raw_ostream &FormattedOS, + const MCAsmInfo &MAI) { + // Flush the stream before taking its content. + CommentStream.flush(); + StringRef Comments = CommentsToEmit.str(); + // Get the default information for printing a comment. + const char *CommentBegin = MAI.getCommentString(); + unsigned CommentColumn = MAI.getCommentColumn(); + bool IsFirst = true; + while (!Comments.empty()) { + if (!IsFirst) + FormattedOS << '\n'; + // Emit a line of comments. + FormattedOS.PadToColumn(CommentColumn); + size_t Position = Comments.find('\n'); + FormattedOS << CommentBegin << ' ' << Comments.substr(0, Position); + // Move after the newline character. + Comments = Comments.substr(Position + 1); + IsFirst = false; + } + FormattedOS.flush(); + + // Tell the comment stream that the vector changed underneath it. + CommentsToEmit.clear(); + CommentStream.resync(); +} + static void DisassembleInputMachO2(StringRef Filename, MachOObjectFile *MachOOF) { const char *McpuDefault = nullptr; @@ -264,11 +1310,27 @@ static void DisassembleInputMachO2(StringRef Filename, std::unique_ptr STI( TheTarget->createMCSubtargetInfo(TripleName, MCPU, FeaturesStr)); MCContext Ctx(AsmInfo.get(), MRI.get(), nullptr); - std::unique_ptr DisAsm( + std::unique_ptr DisAsm( TheTarget->createMCDisassembler(*STI, Ctx)); + std::unique_ptr Symbolizer; + struct DisassembleInfo SymbolizerInfo; + std::unique_ptr RelInfo( + TheTarget->createMCRelocationInfo(TripleName, Ctx)); + if (RelInfo) { + Symbolizer.reset(TheTarget->createMCSymbolizer( + TripleName, SymbolizerGetOpInfo, SymbolizerSymbolLookUp, + &SymbolizerInfo, &Ctx, RelInfo.release())); + DisAsm->setSymbolizer(std::move(Symbolizer)); + } int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); std::unique_ptr IP(TheTarget->createMCInstPrinter( AsmPrinterVariant, *AsmInfo, *InstrInfo, *MRI, *STI)); + // Set the display preference for hex vs. decimal immediates. + IP->setPrintImmHex(PrintImmHex); + // Comment stream and backing vector. + SmallString<128> CommentsToEmit; + raw_svector_ostream CommentStream(CommentsToEmit); + IP->setCommentStream(CommentStream); if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) { errs() << "error: couldn't initialize disassembler for target " @@ -291,10 +1353,13 @@ static void DisassembleInputMachO2(StringRef Filename, ThumbTarget->createMCSubtargetInfo(ThumbTripleName, MCPU, FeaturesStr)); ThumbCtx.reset(new MCContext(ThumbAsmInfo.get(), ThumbMRI.get(), nullptr)); ThumbDisAsm.reset(ThumbTarget->createMCDisassembler(*ThumbSTI, *ThumbCtx)); + // TODO: add MCSymbolizer here for the ThumbTarget like above for TheTarget. int ThumbAsmPrinterVariant = ThumbAsmInfo->getAssemblerDialect(); ThumbIP.reset(ThumbTarget->createMCInstPrinter( ThumbAsmPrinterVariant, *ThumbAsmInfo, *ThumbInstrInfo, *ThumbMRI, *ThumbSTI)); + // Set the display preference for hex vs. decimal immediates. + ThumbIP->setPrintImmHex(PrintImmHex); } if (ThumbTarget && (!ThumbInstrAnalysis || !ThumbAsmInfo || !ThumbSTI || @@ -308,8 +1373,6 @@ static void DisassembleInputMachO2(StringRef Filename, MachO::mach_header Header = MachOOF->getHeader(); - // FIXME: FoundFns isn't used anymore. Using symbols/LC_FUNCTION_STARTS to - // determine function locations will eventually go in MCObjectDisassembler. // FIXME: Using the -cfg command line option, this code used to be able to // annotate relocations with the referenced symbol's name, and if this was // inside a __[cf]string section, the data it points to. This is now replaced @@ -328,7 +1391,7 @@ static void DisassembleInputMachO2(StringRef Filename, // Build a data in code table that is sorted on by the address of each entry. uint64_t BaseAddress = 0; if (Header.filetype == MachO::MH_OBJECT) - Sections[0].getAddress(BaseAddress); + BaseAddress = Sections[0].getAddress(); else BaseAddress = BaseSegmentAddress; DiceTable Dices; @@ -371,14 +1434,12 @@ static void DisassembleInputMachO2(StringRef Filename, for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) { - bool SectIsText = false; - Sections[SectIdx].isText(SectIsText); + bool SectIsText = Sections[SectIdx].isText(); if (SectIsText == false) continue; StringRef SectName; - if (Sections[SectIdx].getName(SectName) || - SectName != "__text") + if (Sections[SectIdx].getName(SectName) || SectName != "__text") continue; // Skip non-text sections DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl(); @@ -389,15 +1450,17 @@ static void DisassembleInputMachO2(StringRef Filename, StringRef Bytes; Sections[SectIdx].getContents(Bytes); - StringRefMemoryObject memoryObject(Bytes); + uint64_t SectAddress = Sections[SectIdx].getAddress(); + DisasmMemoryObject MemoryObject((const uint8_t *)Bytes.data(), Bytes.size(), + SectAddress); bool symbolTableWorked = false; // Parse relocations. std::vector> Relocs; for (const RelocationRef &Reloc : Sections[SectIdx].relocations()) { - uint64_t RelocOffset, SectionAddress; + uint64_t RelocOffset; Reloc.getOffset(RelocOffset); - Sections[SectIdx].getAddress(SectionAddress); + uint64_t SectionAddress = Sections[SectIdx].getAddress(); RelocOffset -= SectionAddress; symbol_iterator RelocSym = Reloc.getSymbol(); @@ -406,6 +1469,33 @@ static void DisassembleInputMachO2(StringRef Filename, } array_pod_sort(Relocs.begin(), Relocs.end()); + // Create a map of symbol addresses to symbol names for use by + // the SymbolizerSymbolLookUp() routine. + SymbolAddressMap AddrMap; + for (const SymbolRef &Symbol : MachOOF->symbols()) { + SymbolRef::Type ST; + Symbol.getType(ST); + if (ST == SymbolRef::ST_Function || ST == SymbolRef::ST_Data || + ST == SymbolRef::ST_Other) { + uint64_t Address; + Symbol.getAddress(Address); + StringRef SymName; + Symbol.getName(SymName); + AddrMap[Address] = SymName; + } + } + // Set up the block of info used by the Symbolizer call backs. + SymbolizerInfo.verbose = true; + SymbolizerInfo.O = MachOOF; + SymbolizerInfo.S = Sections[SectIdx]; + SymbolizerInfo.AddrMap = &AddrMap; + SymbolizerInfo.Sections = &Sections; + SymbolizerInfo.class_name = nullptr; + SymbolizerInfo.selector_name = nullptr; + SymbolizerInfo.method = nullptr; + SymbolizerInfo.demangled_name = nullptr; + SymbolizerInfo.bindtable = nullptr; + // Disassemble symbol by symbol. for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) { StringRef SymName; @@ -417,15 +1507,13 @@ static void DisassembleInputMachO2(StringRef Filename, continue; // Make sure the symbol is defined in this section. - bool containsSym = false; - Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym); + bool containsSym = Sections[SectIdx].containsSymbol(Symbols[SymIdx]); if (!containsSym) continue; // Start at the address of the symbol relative to the section's address. - uint64_t SectionAddress = 0; uint64_t Start = 0; - Sections[SectIdx].getAddress(SectionAddress); + uint64_t SectionAddress = Sections[SectIdx].getAddress(); Symbols[SymIdx].getAddress(Start); Start -= SectionAddress; @@ -433,13 +1521,13 @@ static void DisassembleInputMachO2(StringRef Filename, // the end of the section. bool containsNextSym = false; uint64_t NextSym = 0; - uint64_t NextSymIdx = SymIdx+1; + uint64_t NextSymIdx = SymIdx + 1; while (Symbols.size() > NextSymIdx) { SymbolRef::Type NextSymType; Symbols[NextSymIdx].getType(NextSymType); if (NextSymType == SymbolRef::ST_Function) { - Sections[SectIdx].containsSymbol(Symbols[NextSymIdx], - containsNextSym); + containsNextSym = + Sections[SectIdx].containsSymbol(Symbols[NextSymIdx]); Symbols[NextSymIdx].getAddress(NextSym); NextSym -= SectionAddress; break; @@ -447,12 +1535,14 @@ static void DisassembleInputMachO2(StringRef Filename, ++NextSymIdx; } - uint64_t SectSize; - Sections[SectIdx].getSize(SectSize); - uint64_t End = containsNextSym ? NextSym : SectSize; + uint64_t SectSize = Sections[SectIdx].getSize(); + uint64_t End = containsNextSym ? NextSym : SectSize; uint64_t Size; symbolTableWorked = true; + DisasmMemoryObject SectionMemoryObject((const uint8_t *)Bytes.data() + + Start, + End - Start, SectAddress + Start); DataRefImpl Symb = Symbols[SymIdx].getRawDataRefImpl(); bool isThumb = @@ -463,18 +1553,26 @@ static void DisassembleInputMachO2(StringRef Filename, for (uint64_t Index = Start; Index < End; Index += Size) { MCInst Inst; - uint64_t SectAddress = 0; - Sections[SectIdx].getAddress(SectAddress); - outs() << format("%8" PRIx64 ":\t", SectAddress + Index); + uint64_t PC = SectAddress + Index; + if (FullLeadingAddr) { + if (MachOOF->is64Bit()) + outs() << format("%016" PRIx64, PC); + else + outs() << format("%08" PRIx64, PC); + } else { + outs() << format("%8" PRIx64 ":", PC); + } + if (!NoShowRawInsn) + outs() << "\t"; // Check the data in code table here to see if this is data not an // instruction to be disassembled. DiceTable Dice; - Dice.push_back(std::make_pair(SectAddress + Index, DiceRef())); - dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(), - Dice.begin(), Dice.end(), - compareDiceTableEntries); - if (DTI != Dices.end()){ + Dice.push_back(std::make_pair(PC, DiceRef())); + dice_table_iterator DTI = + std::search(Dices.begin(), Dices.end(), Dice.begin(), Dice.end(), + compareDiceTableEntries); + if (DTI != Dices.end()) { uint16_t Length; DTI->second.getLength(Length); DumpBytes(StringRef(Bytes.data() + Index, Length)); @@ -484,24 +1582,32 @@ static void DisassembleInputMachO2(StringRef Filename, continue; } + SmallVector AnnotationsBytes; + raw_svector_ostream Annotations(AnnotationsBytes); + bool gotInst; if (isThumb) - gotInst = ThumbDisAsm->getInstruction(Inst, Size, memoryObject, Index, - DebugOut, nulls()); + gotInst = ThumbDisAsm->getInstruction(Inst, Size, SectionMemoryObject, + PC, DebugOut, Annotations); else - gotInst = DisAsm->getInstruction(Inst, Size, memoryObject, Index, - DebugOut, nulls()); + gotInst = DisAsm->getInstruction(Inst, Size, SectionMemoryObject, PC, + DebugOut, Annotations); if (gotInst) { - DumpBytes(StringRef(Bytes.data() + Index, Size)); + if (!NoShowRawInsn) { + DumpBytes(StringRef(Bytes.data() + Index, Size)); + } + formatted_raw_ostream FormattedOS(outs()); + Annotations.flush(); + StringRef AnnotationsStr = Annotations.str(); if (isThumb) - ThumbIP->printInst(&Inst, outs(), ""); + ThumbIP->printInst(&Inst, FormattedOS, AnnotationsStr); else - IP->printInst(&Inst, outs(), ""); + IP->printInst(&Inst, FormattedOS, AnnotationsStr); + emitComments(CommentStream, CommentsToEmit, FormattedOS, *AsmInfo); // Print debug info. if (diContext) { - DILineInfo dli = - diContext->getLineInfoForAddress(SectAddress + Index); + DILineInfo dli = diContext->getLineInfoForAddress(PC); // Print valid line info if it changed. if (dli != lastLine && dli.Line != 0) outs() << "\t## " << dli.FileName << ':' << dli.Line << ':' @@ -510,39 +1616,67 @@ static void DisassembleInputMachO2(StringRef Filename, } outs() << "\n"; } else { - errs() << "llvm-objdump: warning: invalid instruction encoding\n"; - if (Size == 0) - Size = 1; // skip illegible bytes + unsigned int Arch = MachOOF->getArch(); + if (Arch == Triple::x86_64 || Arch == Triple::x86) { + outs() << format("\t.byte 0x%02x #bad opcode\n", + *(Bytes.data() + Index) & 0xff); + Size = 1; // skip exactly one illegible byte and move on. + } else { + errs() << "llvm-objdump: warning: invalid instruction encoding\n"; + if (Size == 0) + Size = 1; // skip illegible bytes + } } } } if (!symbolTableWorked) { - // Reading the symbol table didn't work, disassemble the whole section. - uint64_t SectAddress; - Sections[SectIdx].getAddress(SectAddress); - uint64_t SectSize; - Sections[SectIdx].getSize(SectSize); + // Reading the symbol table didn't work, disassemble the whole section. + uint64_t SectAddress = Sections[SectIdx].getAddress(); + uint64_t SectSize = Sections[SectIdx].getSize(); uint64_t InstSize; for (uint64_t Index = 0; Index < SectSize; Index += InstSize) { MCInst Inst; - if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index, - DebugOut, nulls())) { - outs() << format("%8" PRIx64 ":\t", SectAddress + Index); - DumpBytes(StringRef(Bytes.data() + Index, InstSize)); + uint64_t PC = SectAddress + Index; + if (DisAsm->getInstruction(Inst, InstSize, MemoryObject, PC, DebugOut, + nulls())) { + if (FullLeadingAddr) { + if (MachOOF->is64Bit()) + outs() << format("%016" PRIx64, PC); + else + outs() << format("%08" PRIx64, PC); + } else { + outs() << format("%8" PRIx64 ":", PC); + } + if (!NoShowRawInsn) { + outs() << "\t"; + DumpBytes(StringRef(Bytes.data() + Index, InstSize)); + } IP->printInst(&Inst, outs(), ""); outs() << "\n"; } else { - errs() << "llvm-objdump: warning: invalid instruction encoding\n"; - if (InstSize == 0) - InstSize = 1; // skip illegible bytes + unsigned int Arch = MachOOF->getArch(); + if (Arch == Triple::x86_64 || Arch == Triple::x86) { + outs() << format("\t.byte 0x%02x #bad opcode\n", + *(Bytes.data() + Index) & 0xff); + InstSize = 1; // skip exactly one illegible byte and move on. + } else { + errs() << "llvm-objdump: warning: invalid instruction encoding\n"; + if (InstSize == 0) + InstSize = 1; // skip illegible bytes + } } } } + if (SymbolizerInfo.method != nullptr) + free(SymbolizerInfo.method); + if (SymbolizerInfo.demangled_name != nullptr) + free(SymbolizerInfo.demangled_name); + if (SymbolizerInfo.bindtable != nullptr) + delete SymbolizerInfo.bindtable; } } - //===----------------------------------------------------------------------===// // __compact_unwind section dumping //===----------------------------------------------------------------------===// @@ -550,13 +1684,13 @@ static void DisassembleInputMachO2(StringRef Filename, namespace { template static uint64_t readNext(const char *&Buf) { - using llvm::support::little; - using llvm::support::unaligned; + using llvm::support::little; + using llvm::support::unaligned; - uint64_t Val = support::endian::read(Buf); - Buf += sizeof(T); - return Val; - } + uint64_t Val = support::endian::read(Buf); + Buf += sizeof(T); + return Val; +} struct CompactUnwindEntry { uint32_t OffsetInSection; @@ -572,7 +1706,7 @@ struct CompactUnwindEntry { RelocationRef LSDAReloc; CompactUnwindEntry(StringRef Contents, unsigned Offset, bool Is64) - : OffsetInSection(Offset) { + : OffsetInSection(Offset) { if (Is64) read(Contents.data() + Offset); else @@ -580,8 +1714,7 @@ struct CompactUnwindEntry { } private: - template - void read(const char *Buf) { + template void read(const char *Buf) { FunctionAddr = readNext(Buf); Length = readNext(Buf); CompactEncoding = readNext(Buf); @@ -603,8 +1736,7 @@ private: /// referenced section. static void findUnwindRelocNameAddend(const MachOObjectFile *Obj, std::map &Symbols, - const RelocationRef &Reloc, - uint64_t Addr, + const RelocationRef &Reloc, uint64_t Addr, StringRef &Name, uint64_t &Addend) { if (Reloc.getSymbol() != Obj->symbol_end()) { Reloc.getSymbol()->getName(Name); @@ -615,8 +1747,7 @@ static void findUnwindRelocNameAddend(const MachOObjectFile *Obj, auto RE = Obj->getRelocation(Reloc.getRawDataRefImpl()); SectionRef RelocSection = Obj->getRelocationSection(RE); - uint64_t SectionAddr; - RelocSection.getAddress(SectionAddr); + uint64_t SectionAddr = RelocSection.getAddress(); auto Sym = Symbols.upper_bound(Addr); if (Sym == Symbols.begin()) { @@ -647,11 +1778,13 @@ static void findUnwindRelocNameAddend(const MachOObjectFile *Obj, static void printUnwindRelocDest(const MachOObjectFile *Obj, std::map &Symbols, - const RelocationRef &Reloc, - uint64_t Addr) { + const RelocationRef &Reloc, uint64_t Addr) { StringRef Name; uint64_t Addend; + if (!Reloc.getObjectFile()) + return; + findUnwindRelocNameAddend(Obj, Symbols, Reloc, Addr, Name, Addend); outs() << Name; @@ -709,15 +1842,14 @@ printMachOCompactUnwindSection(const MachOObjectFile *Obj, << format("0x%" PRIx32, Entry.OffsetInSection) << ":\n"; // 1. Start of the region this entry applies to. - outs() << " start: " - << format("0x%" PRIx64, Entry.FunctionAddr) << ' '; - printUnwindRelocDest(Obj, Symbols, Entry.FunctionReloc, - Entry.FunctionAddr); + outs() << " start: " << format("0x%" PRIx64, + Entry.FunctionAddr) << ' '; + printUnwindRelocDest(Obj, Symbols, Entry.FunctionReloc, Entry.FunctionAddr); outs() << '\n'; // 2. Length of the region this entry applies to. - outs() << " length: " - << format("0x%" PRIx32, Entry.Length) << '\n'; + outs() << " length: " << format("0x%" PRIx32, Entry.Length) + << '\n'; // 3. The 32-bit compact encoding. outs() << " compact encoding: " << format("0x%08" PRIx32, Entry.CompactEncoding) << '\n'; @@ -733,8 +1865,8 @@ printMachOCompactUnwindSection(const MachOObjectFile *Obj, // 5. This entry's language-specific data area. if (Entry.LSDAReloc.getObjectFile()) { - outs() << " LSDA: " - << format("0x%" PRIx64, Entry.LSDAAddr) << ' '; + outs() << " LSDA: " << format("0x%" PRIx64, + Entry.LSDAAddr) << ' '; printUnwindRelocDest(Obj, Symbols, Entry.LSDAReloc, Entry.LSDAAddr); outs() << '\n'; } @@ -760,11 +1892,9 @@ static void printRegularSecondLevelUnwindPage(const char *PageStart) { uint32_t Encoding = readNext(Pos); outs() << " [" << i << "]: " - << "function offset=" - << format("0x%08" PRIx32, FunctionOffset) << ", " - << "encoding=" - << format("0x%08" PRIx32, Encoding) - << '\n'; + << "function offset=" << format("0x%08" PRIx32, FunctionOffset) + << ", " + << "encoding=" << format("0x%08" PRIx32, Encoding) << '\n'; } } @@ -797,18 +1927,16 @@ static void printCompressedSecondLevelUnwindPage( Encoding = PageEncodings[EncodingIdx - CommonEncodings.size()]; outs() << " [" << i << "]: " - << "function offset=" - << format("0x%08" PRIx32, FunctionOffset) << ", " - << "encoding[" << EncodingIdx << "]=" - << format("0x%08" PRIx32, Encoding) - << '\n'; + << "function offset=" << format("0x%08" PRIx32, FunctionOffset) + << ", " + << "encoding[" << EncodingIdx + << "]=" << format("0x%08" PRIx32, Encoding) << '\n'; } } -static void -printMachOUnwindInfoSection(const MachOObjectFile *Obj, - std::map &Symbols, - const SectionRef &UnwindInfo) { +static void printMachOUnwindInfoSection(const MachOObjectFile *Obj, + std::map &Symbols, + const SectionRef &UnwindInfo) { assert(Obj->isLittleEndian() && "There should not be a big-endian .o with __unwind_info"); @@ -869,7 +1997,6 @@ printMachOUnwindInfoSection(const MachOObjectFile *Obj, << '\n'; } - //===---------------------------------- // Personality functions used in this executable //===---------------------------------- @@ -911,15 +2038,13 @@ printMachOUnwindInfoSection(const MachOObjectFile *Obj, IndexEntries.push_back(Entry); outs() << " [" << i << "]: " - << "function offset=" - << format("0x%08" PRIx32, Entry.FunctionOffset) << ", " + << "function offset=" << format("0x%08" PRIx32, Entry.FunctionOffset) + << ", " << "2nd level page offset=" << format("0x%08" PRIx32, Entry.SecondLevelPageStart) << ", " - << "LSDA offset=" - << format("0x%08" PRIx32, Entry.LSDAStart) << '\n'; + << "LSDA offset=" << format("0x%08" PRIx32, Entry.LSDAStart) << '\n'; } - //===---------------------------------- // Next come the LSDA tables //===---------------------------------- @@ -935,10 +2060,9 @@ printMachOUnwindInfoSection(const MachOObjectFile *Obj, uint32_t FunctionOffset = readNext(Pos); uint32_t LSDAOffset = readNext(Pos); outs() << " [" << i << "]: " - << "function offset=" - << format("0x%08" PRIx32, FunctionOffset) << ", " - << "LSDA offset=" - << format("0x%08" PRIx32, LSDAOffset) << '\n'; + << "function offset=" << format("0x%08" PRIx32, FunctionOffset) + << ", " + << "LSDA offset=" << format("0x%08" PRIx32, LSDAOffset) << '\n'; } //===---------------------------------- @@ -971,7 +2095,6 @@ printMachOUnwindInfoSection(const MachOObjectFile *Obj, CommonEncodings); else llvm_unreachable("Do not know how to print this kind of 2nd level page"); - } } @@ -999,7 +2122,6 @@ void llvm::printMachOUnwindInfo(const MachOObjectFile *Obj) { printMachOUnwindInfoSection(Obj, Symbols, Section); else if (SectName == "__eh_frame") outs() << "llvm-objdump: warning: unhandled __eh_frame section\n"; - } } @@ -1114,7 +2236,7 @@ static void PrintMachHeader(uint32_t magic, uint32_t cputype, break; } if ((cpusubtype & MachO::CPU_SUBTYPE_MASK) == MachO::CPU_SUBTYPE_LIB64) { - outs() << " LIB64 "; + outs() << " LIB64"; } else { outs() << format(" 0x%02" PRIx32, (cpusubtype & MachO::CPU_SUBTYPE_MASK) >> 24); @@ -1698,6 +2820,275 @@ static void PrintDysymtabLoadCommand(MachO::dysymtab_command dyst, outs() << "\n"; } +static void PrintDyldInfoLoadCommand(MachO::dyld_info_command dc, + uint32_t object_size) { + if (dc.cmd == MachO::LC_DYLD_INFO) + outs() << " cmd LC_DYLD_INFO\n"; + else + outs() << " cmd LC_DYLD_INFO_ONLY\n"; + outs() << " cmdsize " << dc.cmdsize; + if (dc.cmdsize != sizeof(struct MachO::dyld_info_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + outs() << " rebase_off " << dc.rebase_off; + if (dc.rebase_off > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " rebase_size " << dc.rebase_size; + uint64_t big_size; + big_size = dc.rebase_off; + big_size += dc.rebase_size; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " bind_off " << dc.bind_off; + if (dc.bind_off > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " bind_size " << dc.bind_size; + big_size = dc.bind_off; + big_size += dc.bind_size; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " weak_bind_off " << dc.weak_bind_off; + if (dc.weak_bind_off > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " weak_bind_size " << dc.weak_bind_size; + big_size = dc.weak_bind_off; + big_size += dc.weak_bind_size; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " lazy_bind_off " << dc.lazy_bind_off; + if (dc.lazy_bind_off > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " lazy_bind_size " << dc.lazy_bind_size; + big_size = dc.lazy_bind_off; + big_size += dc.lazy_bind_size; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " export_off " << dc.export_off; + if (dc.export_off > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " export_size " << dc.export_size; + big_size = dc.export_off; + big_size += dc.export_size; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; +} + +static void PrintDyldLoadCommand(MachO::dylinker_command dyld, + const char *Ptr) { + if (dyld.cmd == MachO::LC_ID_DYLINKER) + outs() << " cmd LC_ID_DYLINKER\n"; + else if (dyld.cmd == MachO::LC_LOAD_DYLINKER) + outs() << " cmd LC_LOAD_DYLINKER\n"; + else if (dyld.cmd == MachO::LC_DYLD_ENVIRONMENT) + outs() << " cmd LC_DYLD_ENVIRONMENT\n"; + else + outs() << " cmd ?(" << dyld.cmd << ")\n"; + outs() << " cmdsize " << dyld.cmdsize; + if (dyld.cmdsize < sizeof(struct MachO::dylinker_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + if (dyld.name >= dyld.cmdsize) + outs() << " name ?(bad offset " << dyld.name << ")\n"; + else { + const char *P = (const char *)(Ptr) + dyld.name; + outs() << " name " << P << " (offset " << dyld.name << ")\n"; + } +} + +static void PrintUuidLoadCommand(MachO::uuid_command uuid) { + outs() << " cmd LC_UUID\n"; + outs() << " cmdsize " << uuid.cmdsize; + if (uuid.cmdsize != sizeof(struct MachO::uuid_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + outs() << " uuid "; + outs() << format("%02" PRIX32, uuid.uuid[0]); + outs() << format("%02" PRIX32, uuid.uuid[1]); + outs() << format("%02" PRIX32, uuid.uuid[2]); + outs() << format("%02" PRIX32, uuid.uuid[3]); + outs() << "-"; + outs() << format("%02" PRIX32, uuid.uuid[4]); + outs() << format("%02" PRIX32, uuid.uuid[5]); + outs() << "-"; + outs() << format("%02" PRIX32, uuid.uuid[6]); + outs() << format("%02" PRIX32, uuid.uuid[7]); + outs() << "-"; + outs() << format("%02" PRIX32, uuid.uuid[8]); + outs() << format("%02" PRIX32, uuid.uuid[9]); + outs() << "-"; + outs() << format("%02" PRIX32, uuid.uuid[10]); + outs() << format("%02" PRIX32, uuid.uuid[11]); + outs() << format("%02" PRIX32, uuid.uuid[12]); + outs() << format("%02" PRIX32, uuid.uuid[13]); + outs() << format("%02" PRIX32, uuid.uuid[14]); + outs() << format("%02" PRIX32, uuid.uuid[15]); + outs() << "\n"; +} + +static void PrintVersionMinLoadCommand(MachO::version_min_command vd) { + if (vd.cmd == MachO::LC_VERSION_MIN_MACOSX) + outs() << " cmd LC_VERSION_MIN_MACOSX\n"; + else if (vd.cmd == MachO::LC_VERSION_MIN_IPHONEOS) + outs() << " cmd LC_VERSION_MIN_IPHONEOS\n"; + else + outs() << " cmd " << vd.cmd << " (?)\n"; + outs() << " cmdsize " << vd.cmdsize; + if (vd.cmdsize != sizeof(struct MachO::version_min_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + outs() << " version " << ((vd.version >> 16) & 0xffff) << "." + << ((vd.version >> 8) & 0xff); + if ((vd.version & 0xff) != 0) + outs() << "." << (vd.version & 0xff); + outs() << "\n"; + if (vd.sdk == 0) + outs() << " sdk n/a\n"; + else { + outs() << " sdk " << ((vd.sdk >> 16) & 0xffff) << "." + << ((vd.sdk >> 8) & 0xff); + } + if ((vd.sdk & 0xff) != 0) + outs() << "." << (vd.sdk & 0xff); + outs() << "\n"; +} + +static void PrintSourceVersionCommand(MachO::source_version_command sd) { + outs() << " cmd LC_SOURCE_VERSION\n"; + outs() << " cmdsize " << sd.cmdsize; + if (sd.cmdsize != sizeof(struct MachO::source_version_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + uint64_t a = (sd.version >> 40) & 0xffffff; + uint64_t b = (sd.version >> 30) & 0x3ff; + uint64_t c = (sd.version >> 20) & 0x3ff; + uint64_t d = (sd.version >> 10) & 0x3ff; + uint64_t e = sd.version & 0x3ff; + outs() << " version " << a << "." << b; + if (e != 0) + outs() << "." << c << "." << d << "." << e; + else if (d != 0) + outs() << "." << c << "." << d; + else if (c != 0) + outs() << "." << c; + outs() << "\n"; +} + +static void PrintEntryPointCommand(MachO::entry_point_command ep) { + outs() << " cmd LC_MAIN\n"; + outs() << " cmdsize " << ep.cmdsize; + if (ep.cmdsize != sizeof(struct MachO::entry_point_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + outs() << " entryoff " << ep.entryoff << "\n"; + outs() << " stacksize " << ep.stacksize << "\n"; +} + +static void PrintDylibCommand(MachO::dylib_command dl, const char *Ptr) { + if (dl.cmd == MachO::LC_ID_DYLIB) + outs() << " cmd LC_ID_DYLIB\n"; + else if (dl.cmd == MachO::LC_LOAD_DYLIB) + outs() << " cmd LC_LOAD_DYLIB\n"; + else if (dl.cmd == MachO::LC_LOAD_WEAK_DYLIB) + outs() << " cmd LC_LOAD_WEAK_DYLIB\n"; + else if (dl.cmd == MachO::LC_REEXPORT_DYLIB) + outs() << " cmd LC_REEXPORT_DYLIB\n"; + else if (dl.cmd == MachO::LC_LAZY_LOAD_DYLIB) + outs() << " cmd LC_LAZY_LOAD_DYLIB\n"; + else if (dl.cmd == MachO::LC_LOAD_UPWARD_DYLIB) + outs() << " cmd LC_LOAD_UPWARD_DYLIB\n"; + else + outs() << " cmd " << dl.cmd << " (unknown)\n"; + outs() << " cmdsize " << dl.cmdsize; + if (dl.cmdsize < sizeof(struct MachO::dylib_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + if (dl.dylib.name < dl.cmdsize) { + const char *P = (const char *)(Ptr) + dl.dylib.name; + outs() << " name " << P << " (offset " << dl.dylib.name << ")\n"; + } else { + outs() << " name ?(bad offset " << dl.dylib.name << ")\n"; + } + outs() << " time stamp " << dl.dylib.timestamp << " "; + time_t t = dl.dylib.timestamp; + outs() << ctime(&t); + outs() << " current version "; + if (dl.dylib.current_version == 0xffffffff) + outs() << "n/a\n"; + else + outs() << ((dl.dylib.current_version >> 16) & 0xffff) << "." + << ((dl.dylib.current_version >> 8) & 0xff) << "." + << (dl.dylib.current_version & 0xff) << "\n"; + outs() << "compatibility version "; + if (dl.dylib.compatibility_version == 0xffffffff) + outs() << "n/a\n"; + else + outs() << ((dl.dylib.compatibility_version >> 16) & 0xffff) << "." + << ((dl.dylib.compatibility_version >> 8) & 0xff) << "." + << (dl.dylib.compatibility_version & 0xff) << "\n"; +} + +static void PrintLinkEditDataCommand(MachO::linkedit_data_command ld, + uint32_t object_size) { + if (ld.cmd == MachO::LC_CODE_SIGNATURE) + outs() << " cmd LC_FUNCTION_STARTS\n"; + else if (ld.cmd == MachO::LC_SEGMENT_SPLIT_INFO) + outs() << " cmd LC_SEGMENT_SPLIT_INFO\n"; + else if (ld.cmd == MachO::LC_FUNCTION_STARTS) + outs() << " cmd LC_FUNCTION_STARTS\n"; + else if (ld.cmd == MachO::LC_DATA_IN_CODE) + outs() << " cmd LC_DATA_IN_CODE\n"; + else if (ld.cmd == MachO::LC_DYLIB_CODE_SIGN_DRS) + outs() << " cmd LC_DYLIB_CODE_SIGN_DRS\n"; + else if (ld.cmd == MachO::LC_LINKER_OPTIMIZATION_HINT) + outs() << " cmd LC_LINKER_OPTIMIZATION_HINT\n"; + else + outs() << " cmd " << ld.cmd << " (?)\n"; + outs() << " cmdsize " << ld.cmdsize; + if (ld.cmdsize != sizeof(struct MachO::linkedit_data_command)) + outs() << " Incorrect size\n"; + else + outs() << "\n"; + outs() << " dataoff " << ld.dataoff; + if (ld.dataoff > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; + outs() << " datasize " << ld.datasize; + uint64_t big_size = ld.dataoff; + big_size += ld.datasize; + if (big_size > object_size) + outs() << " (past end of file)\n"; + else + outs() << "\n"; +} + static void PrintLoadCommands(const MachOObjectFile *Obj, uint32_t ncmds, uint32_t filetype, uint32_t cputype, bool verbose) { @@ -1739,6 +3130,44 @@ static void PrintLoadCommands(const MachOObjectFile *Obj, uint32_t ncmds, MachO::dysymtab_command Dysymtab = Obj->getDysymtabLoadCommand(); MachO::symtab_command Symtab = Obj->getSymtabLoadCommand(); PrintDysymtabLoadCommand(Dysymtab, Symtab.nsyms, Buf.size(), cputype); + } else if (Command.C.cmd == MachO::LC_DYLD_INFO || + Command.C.cmd == MachO::LC_DYLD_INFO_ONLY) { + MachO::dyld_info_command DyldInfo = Obj->getDyldInfoLoadCommand(Command); + PrintDyldInfoLoadCommand(DyldInfo, Buf.size()); + } else if (Command.C.cmd == MachO::LC_LOAD_DYLINKER || + Command.C.cmd == MachO::LC_ID_DYLINKER || + Command.C.cmd == MachO::LC_DYLD_ENVIRONMENT) { + MachO::dylinker_command Dyld = Obj->getDylinkerCommand(Command); + PrintDyldLoadCommand(Dyld, Command.Ptr); + } else if (Command.C.cmd == MachO::LC_UUID) { + MachO::uuid_command Uuid = Obj->getUuidCommand(Command); + PrintUuidLoadCommand(Uuid); + } else if (Command.C.cmd == MachO::LC_VERSION_MIN_MACOSX) { + MachO::version_min_command Vd = Obj->getVersionMinLoadCommand(Command); + PrintVersionMinLoadCommand(Vd); + } else if (Command.C.cmd == MachO::LC_SOURCE_VERSION) { + MachO::source_version_command Sd = Obj->getSourceVersionCommand(Command); + PrintSourceVersionCommand(Sd); + } else if (Command.C.cmd == MachO::LC_MAIN) { + MachO::entry_point_command Ep = Obj->getEntryPointCommand(Command); + PrintEntryPointCommand(Ep); + } else if (Command.C.cmd == MachO::LC_LOAD_DYLIB || + Command.C.cmd == MachO::LC_ID_DYLIB || + Command.C.cmd == MachO::LC_LOAD_WEAK_DYLIB || + Command.C.cmd == MachO::LC_REEXPORT_DYLIB || + Command.C.cmd == MachO::LC_LAZY_LOAD_DYLIB || + Command.C.cmd == MachO::LC_LOAD_UPWARD_DYLIB) { + MachO::dylib_command Dl = Obj->getDylibIDLoadCommand(Command); + PrintDylibCommand(Dl, Command.Ptr); + } else if (Command.C.cmd == MachO::LC_CODE_SIGNATURE || + Command.C.cmd == MachO::LC_SEGMENT_SPLIT_INFO || + Command.C.cmd == MachO::LC_FUNCTION_STARTS || + Command.C.cmd == MachO::LC_DATA_IN_CODE || + Command.C.cmd == MachO::LC_DYLIB_CODE_SIGN_DRS || + Command.C.cmd == MachO::LC_LINKER_OPTIMIZATION_HINT) { + MachO::linkedit_data_command Ld = + Obj->getLinkeditDataLoadCommand(Command); + PrintLinkEditDataCommand(Ld, Buf.size()); } else { outs() << " cmd ?(" << format("0x%08" PRIx32, Command.C.cmd) << ")\n"; @@ -1801,12 +3230,12 @@ void llvm::printMachOExportsTrie(const object::MachOObjectFile *Obj) { if (ReExport) outs() << "[re-export] "; else - outs() - << format("0x%08llX ", Entry.address()); // FIXME:add in base address + outs() << format("0x%08llX ", + Entry.address()); // FIXME:add in base address outs() << Entry.name(); if (WeakDef || ThreadLocal || Resolver || Abs) { bool NeedsComma = false; - printf(" ["); + outs() << " ["; if (WeakDef) { outs() << "weak_def"; NeedsComma = true; @@ -1843,3 +3272,252 @@ void llvm::printMachOExportsTrie(const object::MachOObjectFile *Obj) { outs() << "\n"; } } + +//===----------------------------------------------------------------------===// +// rebase table dumping +//===----------------------------------------------------------------------===// + +namespace { +class SegInfo { +public: + SegInfo(const object::MachOObjectFile *Obj); + + StringRef segmentName(uint32_t SegIndex); + StringRef sectionName(uint32_t SegIndex, uint64_t SegOffset); + uint64_t address(uint32_t SegIndex, uint64_t SegOffset); + +private: + struct SectionInfo { + uint64_t Address; + uint64_t Size; + StringRef SectionName; + StringRef SegmentName; + uint64_t OffsetInSegment; + uint64_t SegmentStartAddress; + uint32_t SegmentIndex; + }; + const SectionInfo &findSection(uint32_t SegIndex, uint64_t SegOffset); + SmallVector Sections; +}; +} + +SegInfo::SegInfo(const object::MachOObjectFile *Obj) { + // Build table of sections so segIndex/offset pairs can be translated. + uint32_t CurSegIndex = Obj->hasPageZeroSegment() ? 1 : 0; + StringRef CurSegName; + uint64_t CurSegAddress; + for (const SectionRef &Section : Obj->sections()) { + SectionInfo Info; + if (error(Section.getName(Info.SectionName))) + return; + Info.Address = Section.getAddress(); + Info.Size = Section.getSize(); + Info.SegmentName = + Obj->getSectionFinalSegmentName(Section.getRawDataRefImpl()); + if (!Info.SegmentName.equals(CurSegName)) { + ++CurSegIndex; + CurSegName = Info.SegmentName; + CurSegAddress = Info.Address; + } + Info.SegmentIndex = CurSegIndex - 1; + Info.OffsetInSegment = Info.Address - CurSegAddress; + Info.SegmentStartAddress = CurSegAddress; + Sections.push_back(Info); + } +} + +StringRef SegInfo::segmentName(uint32_t SegIndex) { + for (const SectionInfo &SI : Sections) { + if (SI.SegmentIndex == SegIndex) + return SI.SegmentName; + } + llvm_unreachable("invalid segIndex"); +} + +const SegInfo::SectionInfo &SegInfo::findSection(uint32_t SegIndex, + uint64_t OffsetInSeg) { + for (const SectionInfo &SI : Sections) { + if (SI.SegmentIndex != SegIndex) + continue; + if (SI.OffsetInSegment > OffsetInSeg) + continue; + if (OffsetInSeg >= (SI.OffsetInSegment + SI.Size)) + continue; + return SI; + } + llvm_unreachable("segIndex and offset not in any section"); +} + +StringRef SegInfo::sectionName(uint32_t SegIndex, uint64_t OffsetInSeg) { + return findSection(SegIndex, OffsetInSeg).SectionName; +} + +uint64_t SegInfo::address(uint32_t SegIndex, uint64_t OffsetInSeg) { + const SectionInfo &SI = findSection(SegIndex, OffsetInSeg); + return SI.SegmentStartAddress + OffsetInSeg; +} + +void llvm::printMachORebaseTable(const object::MachOObjectFile *Obj) { + // Build table of sections so names can used in final output. + SegInfo sectionTable(Obj); + + outs() << "segment section address type\n"; + for (const llvm::object::MachORebaseEntry &Entry : Obj->rebaseTable()) { + uint32_t SegIndex = Entry.segmentIndex(); + uint64_t OffsetInSeg = Entry.segmentOffset(); + StringRef SegmentName = sectionTable.segmentName(SegIndex); + StringRef SectionName = sectionTable.sectionName(SegIndex, OffsetInSeg); + uint64_t Address = sectionTable.address(SegIndex, OffsetInSeg); + + // Table lines look like: __DATA __nl_symbol_ptr 0x0000F00C pointer + outs() << format("%-8s %-18s 0x%08" PRIX64 " %s\n", + SegmentName.str().c_str(), SectionName.str().c_str(), + Address, Entry.typeName().str().c_str()); + } +} + +static StringRef ordinalName(const object::MachOObjectFile *Obj, int Ordinal) { + StringRef DylibName; + switch (Ordinal) { + case MachO::BIND_SPECIAL_DYLIB_SELF: + return "this-image"; + case MachO::BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE: + return "main-executable"; + case MachO::BIND_SPECIAL_DYLIB_FLAT_LOOKUP: + return "flat-namespace"; + default: + if (Ordinal > 0) { + std::error_code EC = + Obj->getLibraryShortNameByIndex(Ordinal - 1, DylibName); + if (EC) + return "<>"; + return DylibName; + } + } + return "<>"; +} + +//===----------------------------------------------------------------------===// +// bind table dumping +//===----------------------------------------------------------------------===// + +void llvm::printMachOBindTable(const object::MachOObjectFile *Obj) { + // Build table of sections so names can used in final output. + SegInfo sectionTable(Obj); + + outs() << "segment section address type " + "addend dylib symbol\n"; + for (const llvm::object::MachOBindEntry &Entry : Obj->bindTable()) { + uint32_t SegIndex = Entry.segmentIndex(); + uint64_t OffsetInSeg = Entry.segmentOffset(); + StringRef SegmentName = sectionTable.segmentName(SegIndex); + StringRef SectionName = sectionTable.sectionName(SegIndex, OffsetInSeg); + uint64_t Address = sectionTable.address(SegIndex, OffsetInSeg); + + // Table lines look like: + // __DATA __got 0x00012010 pointer 0 libSystem ___stack_chk_guard + StringRef Attr; + if (Entry.flags() & MachO::BIND_SYMBOL_FLAGS_WEAK_IMPORT) + Attr = " (weak_import)"; + outs() << left_justify(SegmentName, 8) << " " + << left_justify(SectionName, 18) << " " + << format_hex(Address, 10, true) << " " + << left_justify(Entry.typeName(), 8) << " " + << format_decimal(Entry.addend(), 8) << " " + << left_justify(ordinalName(Obj, Entry.ordinal()), 16) << " " + << Entry.symbolName() << Attr << "\n"; + } +} + +//===----------------------------------------------------------------------===// +// lazy bind table dumping +//===----------------------------------------------------------------------===// + +void llvm::printMachOLazyBindTable(const object::MachOObjectFile *Obj) { + // Build table of sections so names can used in final output. + SegInfo sectionTable(Obj); + + outs() << "segment section address " + "dylib symbol\n"; + for (const llvm::object::MachOBindEntry &Entry : Obj->lazyBindTable()) { + uint32_t SegIndex = Entry.segmentIndex(); + uint64_t OffsetInSeg = Entry.segmentOffset(); + StringRef SegmentName = sectionTable.segmentName(SegIndex); + StringRef SectionName = sectionTable.sectionName(SegIndex, OffsetInSeg); + uint64_t Address = sectionTable.address(SegIndex, OffsetInSeg); + + // Table lines look like: + // __DATA __got 0x00012010 libSystem ___stack_chk_guard + outs() << left_justify(SegmentName, 8) << " " + << left_justify(SectionName, 18) << " " + << format_hex(Address, 10, true) << " " + << left_justify(ordinalName(Obj, Entry.ordinal()), 16) << " " + << Entry.symbolName() << "\n"; + } +} + +//===----------------------------------------------------------------------===// +// weak bind table dumping +//===----------------------------------------------------------------------===// + +void llvm::printMachOWeakBindTable(const object::MachOObjectFile *Obj) { + // Build table of sections so names can used in final output. + SegInfo sectionTable(Obj); + + outs() << "segment section address " + "type addend symbol\n"; + for (const llvm::object::MachOBindEntry &Entry : Obj->weakBindTable()) { + // Strong symbols don't have a location to update. + if (Entry.flags() & MachO::BIND_SYMBOL_FLAGS_NON_WEAK_DEFINITION) { + outs() << " strong " + << Entry.symbolName() << "\n"; + continue; + } + uint32_t SegIndex = Entry.segmentIndex(); + uint64_t OffsetInSeg = Entry.segmentOffset(); + StringRef SegmentName = sectionTable.segmentName(SegIndex); + StringRef SectionName = sectionTable.sectionName(SegIndex, OffsetInSeg); + uint64_t Address = sectionTable.address(SegIndex, OffsetInSeg); + + // Table lines look like: + // __DATA __data 0x00001000 pointer 0 _foo + outs() << left_justify(SegmentName, 8) << " " + << left_justify(SectionName, 18) << " " + << format_hex(Address, 10, true) << " " + << left_justify(Entry.typeName(), 8) << " " + << format_decimal(Entry.addend(), 8) << " " << Entry.symbolName() + << "\n"; + } +} + +// get_dyld_bind_info_symbolname() is used for disassembly and passed an +// address, ReferenceValue, in the Mach-O file and looks in the dyld bind +// information for that address. If the address is found its binding symbol +// name is returned. If not nullptr is returned. +static const char *get_dyld_bind_info_symbolname(uint64_t ReferenceValue, + struct DisassembleInfo *info) { + if (info->bindtable == nullptr) { + info->bindtable = new (BindTable); + SegInfo sectionTable(info->O); + for (const llvm::object::MachOBindEntry &Entry : info->O->bindTable()) { + uint32_t SegIndex = Entry.segmentIndex(); + uint64_t OffsetInSeg = Entry.segmentOffset(); + uint64_t Address = sectionTable.address(SegIndex, OffsetInSeg); + const char *SymbolName = nullptr; + StringRef name = Entry.symbolName(); + if (!name.empty()) + SymbolName = name.data(); + info->bindtable->push_back(std::make_pair(Address, SymbolName)); + } + } + for (bind_table_iterator BI = info->bindtable->begin(), + BE = info->bindtable->end(); + BI != BE; ++BI) { + uint64_t Address = BI->first; + if (ReferenceValue == Address) { + const char *SymbolName = BI->second; + return SymbolName; + } + } + return nullptr; +}