1 //===- FileAnalysis.h -------------------------------------------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 10 #ifndef LLVM_CFI_VERIFY_FILE_ANALYSIS_H 11 #define LLVM_CFI_VERIFY_FILE_ANALYSIS_H 12 13 #include "llvm/ADT/DenseMap.h" 14 #include "llvm/BinaryFormat/ELF.h" 15 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 16 #include "llvm/MC/MCAsmInfo.h" 17 #include "llvm/MC/MCContext.h" 18 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 19 #include "llvm/MC/MCInst.h" 20 #include "llvm/MC/MCInstPrinter.h" 21 #include "llvm/MC/MCInstrAnalysis.h" 22 #include "llvm/MC/MCInstrDesc.h" 23 #include "llvm/MC/MCInstrInfo.h" 24 #include "llvm/MC/MCObjectFileInfo.h" 25 #include "llvm/MC/MCRegisterInfo.h" 26 #include "llvm/MC/MCSubtargetInfo.h" 27 #include "llvm/Object/Binary.h" 28 #include "llvm/Object/COFF.h" 29 #include "llvm/Object/ELFObjectFile.h" 30 #include "llvm/Object/ObjectFile.h" 31 #include "llvm/Support/Casting.h" 32 #include "llvm/Support/CommandLine.h" 33 #include "llvm/Support/Error.h" 34 #include "llvm/Support/MemoryBuffer.h" 35 #include "llvm/Support/TargetRegistry.h" 36 #include "llvm/Support/TargetSelect.h" 37 #include "llvm/Support/raw_ostream.h" 38 39 #include <functional> 40 #include <set> 41 #include <string> 42 #include <unordered_map> 43 44 namespace llvm { 45 namespace cfi_verify { 46 47 struct GraphResult; 48 49 extern bool IgnoreDWARFFlag; 50 51 enum class CFIProtectionStatus { 52 // This instruction is protected by CFI. 53 PROTECTED, 54 // The instruction is not an indirect control flow instruction, and thus 55 // shouldn't be protected. 56 FAIL_NOT_INDIRECT_CF, 57 // There is a path to the instruction that was unexpected. 58 FAIL_ORPHANS, 59 // There is a path to the instruction from a conditional branch that does not 60 // properly check the destination for this vcall/icall. 61 FAIL_BAD_CONDITIONAL_BRANCH, 62 // One of the operands of the indirect CF instruction is modified between the 63 // CFI-check and execution. 64 FAIL_REGISTER_CLOBBERED, 65 // The instruction referenced does not exist. This normally indicates an 66 // error in the program, where you try and validate a graph that was created 67 // in a different FileAnalysis object. 68 FAIL_INVALID_INSTRUCTION, 69 }; 70 71 StringRef stringCFIProtectionStatus(CFIProtectionStatus Status); 72 73 // Disassembler and analysis tool for machine code files. Keeps track of non- 74 // sequential control flows, including indirect control flow instructions. 75 class FileAnalysis { 76 public: 77 // A metadata struct for an instruction. 78 struct Instr { 79 uint64_t VMAddress; // Virtual memory address of this instruction. 80 MCInst Instruction; // Instruction. 81 uint64_t InstructionSize; // Size of this instruction. 82 bool Valid; // Is this a valid instruction? If false, Instr::Instruction is 83 // undefined. 84 }; 85 86 // Construct a FileAnalysis from a file path. 87 static Expected<FileAnalysis> Create(StringRef Filename); 88 89 // Construct and take ownership of the supplied object. Do not use this 90 // constructor, prefer to use FileAnalysis::Create instead. 91 FileAnalysis(object::OwningBinary<object::Binary> Binary); 92 FileAnalysis() = delete; 93 FileAnalysis(const FileAnalysis &) = delete; 94 FileAnalysis(FileAnalysis &&Other) = default; 95 96 // Returns the instruction at the provided address. Returns nullptr if there 97 // is no instruction at the provided address. 98 const Instr *getInstruction(uint64_t Address) const; 99 100 // Returns the instruction at the provided adress, dying if the instruction is 101 // not found. 102 const Instr &getInstructionOrDie(uint64_t Address) const; 103 104 // Returns a pointer to the previous/next instruction in sequence, 105 // respectively. Returns nullptr if the next/prev instruction doesn't exist, 106 // or if the provided instruction doesn't exist. 107 const Instr *getPrevInstructionSequential(const Instr &InstrMeta) const; 108 const Instr *getNextInstructionSequential(const Instr &InstrMeta) const; 109 110 // Returns whether this instruction is used by CFI to trap the program. 111 bool isCFITrap(const Instr &InstrMeta) const; 112 113 // Returns whether this function can fall through to the next instruction. 114 // Undefined (and bad) instructions cannot fall through, and instruction that 115 // modify the control flow can only fall through if they are conditional 116 // branches or calls. 117 bool canFallThrough(const Instr &InstrMeta) const; 118 119 // Returns the definitive next instruction. This is different from the next 120 // instruction sequentially as it will follow unconditional branches (assuming 121 // they can be resolved at compile time, i.e. not indirect). This method 122 // returns nullptr if the provided instruction does not transfer control flow 123 // to exactly one instruction that is known deterministically at compile time. 124 // Also returns nullptr if the deterministic target does not exist in this 125 // file. 126 const Instr *getDefiniteNextInstruction(const Instr &InstrMeta) const; 127 128 // Get a list of deterministic control flows that lead to the provided 129 // instruction. This list includes all static control flow cross-references as 130 // well as the previous instruction if it can fall through. 131 std::set<const Instr *> 132 getDirectControlFlowXRefs(const Instr &InstrMeta) const; 133 134 // Returns whether this instruction uses a register operand. 135 bool usesRegisterOperand(const Instr &InstrMeta) const; 136 137 // Returns the list of indirect instructions. 138 const std::set<uint64_t> &getIndirectInstructions() const; 139 140 const MCRegisterInfo *getRegisterInfo() const; 141 const MCInstrInfo *getMCInstrInfo() const; 142 const MCInstrAnalysis *getMCInstrAnalysis() const; 143 144 // Returns the inlining information for the provided address. 145 Expected<DIInliningInfo> symbolizeInlinedCode(uint64_t Address); 146 147 // Returns whether the provided Graph represents a protected indirect control 148 // flow instruction in this file. 149 CFIProtectionStatus validateCFIProtection(const GraphResult &Graph) const; 150 151 // Returns the first place the operand register is clobbered between the CFI- 152 // check and the indirect CF instruction execution. We do this by walking 153 // backwards from the indirect CF and ensuring there is at most one load 154 // involving the operand register (which is the indirect CF itself on x86). 155 // If the register is not modified, returns the address of the indirect CF 156 // instruction. The result is undefined if the provided graph does not fall 157 // under either the FAIL_REGISTER_CLOBBERED or PROTECTED status (see 158 // CFIProtectionStatus). 159 uint64_t indirectCFOperandClobber(const GraphResult& Graph) const; 160 161 // Prints an instruction to the provided stream using this object's pretty- 162 // printers. 163 void printInstruction(const Instr &InstrMeta, raw_ostream &OS) const; 164 165 protected: 166 // Construct a blank object with the provided triple and features. Used in 167 // testing, where a sub class will dependency inject protected methods to 168 // allow analysis of raw binary, without requiring a fully valid ELF file. 169 FileAnalysis(const Triple &ObjectTriple, const SubtargetFeatures &Features); 170 171 // Add an instruction to this object. 172 void addInstruction(const Instr &Instruction); 173 174 // Disassemble and parse the provided bytes into this object. Instruction 175 // address calculation is done relative to the provided SectionAddress. 176 void parseSectionContents(ArrayRef<uint8_t> SectionBytes, 177 uint64_t SectionAddress); 178 179 // Constructs and initialises members required for disassembly. 180 Error initialiseDisassemblyMembers(); 181 182 // Parses code sections from the internal object file. Saves them into the 183 // internal members. Should only be called once by Create(). 184 Error parseCodeSections(); 185 186 private: 187 // Members that describe the input file. 188 object::OwningBinary<object::Binary> Binary; 189 const object::ObjectFile *Object = nullptr; 190 Triple ObjectTriple; 191 std::string ArchName; 192 std::string MCPU; 193 const Target *ObjectTarget = nullptr; 194 SubtargetFeatures Features; 195 196 // Members required for disassembly. 197 std::unique_ptr<const MCRegisterInfo> RegisterInfo; 198 std::unique_ptr<const MCAsmInfo> AsmInfo; 199 std::unique_ptr<MCSubtargetInfo> SubtargetInfo; 200 std::unique_ptr<const MCInstrInfo> MII; 201 MCObjectFileInfo MOFI; 202 std::unique_ptr<MCContext> Context; 203 std::unique_ptr<const MCDisassembler> Disassembler; 204 std::unique_ptr<const MCInstrAnalysis> MIA; 205 std::unique_ptr<MCInstPrinter> Printer; 206 207 // Symbolizer used for debug information parsing. 208 std::unique_ptr<symbolize::LLVMSymbolizer> Symbolizer; 209 210 // A mapping between the virtual memory address to the instruction metadata 211 // struct. TODO(hctim): Reimplement this as a sorted vector to avoid per- 212 // insertion allocation. 213 std::map<uint64_t, Instr> Instructions; 214 215 // Contains a mapping between a specific address, and a list of instructions 216 // that use this address as a branch target (including call instructions). 217 DenseMap<uint64_t, std::vector<uint64_t>> StaticBranchTargetings; 218 219 // A list of addresses of indirect control flow instructions. 220 std::set<uint64_t> IndirectInstructions; 221 }; 222 223 class UnsupportedDisassembly : public ErrorInfo<UnsupportedDisassembly> { 224 public: 225 static char ID; 226 std::string Text; 227 228 UnsupportedDisassembly(StringRef Text); 229 230 void log(raw_ostream &OS) const override; 231 std::error_code convertToErrorCode() const override; 232 }; 233 234 } // namespace cfi_verify 235 } // namespace llvm 236 237 #endif // LLVM_CFI_VERIFY_FILE_ANALYSIS_H 238