1 //===- GraphBuilder.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_GRAPH_BUILDER_H 11 #define LLVM_CFI_VERIFY_GRAPH_BUILDER_H 12 13 #include "FileAnalysis.h" 14 15 #include "llvm/ADT/DenseMap.h" 16 #include "llvm/BinaryFormat/ELF.h" 17 #include "llvm/MC/MCAsmInfo.h" 18 #include "llvm/MC/MCContext.h" 19 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 20 #include "llvm/MC/MCInst.h" 21 #include "llvm/MC/MCInstPrinter.h" 22 #include "llvm/MC/MCInstrAnalysis.h" 23 #include "llvm/MC/MCInstrDesc.h" 24 #include "llvm/MC/MCInstrInfo.h" 25 #include "llvm/MC/MCObjectFileInfo.h" 26 #include "llvm/MC/MCRegisterInfo.h" 27 #include "llvm/MC/MCSubtargetInfo.h" 28 #include "llvm/Object/Binary.h" 29 #include "llvm/Object/COFF.h" 30 #include "llvm/Object/ELFObjectFile.h" 31 #include "llvm/Object/ObjectFile.h" 32 #include "llvm/Support/Casting.h" 33 #include "llvm/Support/CommandLine.h" 34 #include "llvm/Support/Error.h" 35 #include "llvm/Support/MemoryBuffer.h" 36 #include "llvm/Support/TargetRegistry.h" 37 #include "llvm/Support/TargetSelect.h" 38 #include "llvm/Support/raw_ostream.h" 39 40 #include <functional> 41 #include <set> 42 #include <string> 43 #include <unordered_map> 44 45 using Instr = llvm::cfi_verify::FileAnalysis::Instr; 46 47 namespace llvm { 48 namespace cfi_verify { 49 50 extern unsigned long long SearchLengthForUndef; 51 extern unsigned long long SearchLengthForConditionalBranch; 52 53 struct ConditionalBranchNode { 54 uint64_t Address; 55 uint64_t Target; 56 uint64_t Fallthrough; 57 // Does this conditional branch look like it's used for CFI protection? i.e. 58 // - The exit point of a basic block whos entry point is {target|fallthrough} 59 // is a CFI trap, and... 60 // - The exit point of the other basic block is an undirect CF instruction. 61 bool CFIProtection; 62 bool IndirectCFIsOnTargetPath; 63 }; 64 65 // The canonical graph result structure returned by GraphBuilder. The members 66 // in this structure encapsulate all possible code paths to the instruction 67 // located at `BaseAddress`. 68 struct GraphResult { 69 uint64_t BaseAddress; 70 71 // Map between an instruction address, and the address of the next instruction 72 // that will be executed. This map will contain all keys in the range: 73 // - [orphaned node, base address) 74 // - [conditional branch node {target|fallthrough}, base address) 75 DenseMap<uint64_t, uint64_t> IntermediateNodes; 76 77 // A list of orphaned nodes. A node is an 'orphan' if it meets any of the 78 // following criteria: 79 // - The length of the path from the base to this node has exceeded 80 // `SearchLengthForConditionalBranch`. 81 // - The node has no cross references to it. 82 // - The path from the base to this node is cyclic. 83 std::vector<uint64_t> OrphanedNodes; 84 85 // A list of top-level conditional branches that exist at the top of any 86 // non-orphan paths from the base. 87 std::vector<ConditionalBranchNode> ConditionalBranchNodes; 88 89 // Returns an in-order list of the path between the address provided and the 90 // base. The provided address must be part of this graph, and must not be a 91 // conditional branch. 92 std::vector<uint64_t> flattenAddress(uint64_t Address) const; 93 94 // Print the DOT representation of this result. 95 void printToDOT(const FileAnalysis &Analysis, raw_ostream &OS) const; 96 }; 97 98 class GraphBuilder { 99 public: 100 // Build the control flow graph for a provided control flow node. This method 101 // will enumerate all branch nodes that can lead to this node, and place them 102 // into GraphResult::ConditionalBranchNodes. It will also provide any orphaned 103 // (i.e. the upwards traversal did not make it to a branch node) flows to the 104 // provided node in GraphResult::OrphanedNodes. 105 static GraphResult buildFlowGraph(const FileAnalysis &Analysis, 106 uint64_t Address); 107 108 private: 109 // Implementation function that actually builds the flow graph. Retrieves a 110 // list of cross references to instruction referenced in `Address`. If any of 111 // these XRefs are conditional branches, it will build the other potential 112 // path (fallthrough or target) using `buildFlowsToUndefined`. Otherwise, this 113 // function will recursively call itself where `Address` in the recursive call 114 // is now the XRef. If any XRef is an orphan, it is added to 115 // `Result.OrphanedNodes`. `OpenedNodes` keeps track of the list of nodes 116 // in the current path and is used for cycle-checking. If the path is found 117 // to be cyclic, it will be added to `Result.OrphanedNodes`. 118 static void buildFlowGraphImpl(const FileAnalysis &Analysis, 119 DenseSet<uint64_t> &OpenedNodes, 120 GraphResult &Result, uint64_t Address, 121 uint64_t Depth); 122 123 // Utilised by buildFlowGraphImpl to build the tree out from the provided 124 // conditional branch node to an undefined instruction. The provided 125 // conditional branch node must have exactly one of its subtrees set, and will 126 // update the node's CFIProtection field if a deterministic flow can be found 127 // to an undefined instruction. 128 static void buildFlowsToUndefined(const FileAnalysis &Analysis, 129 GraphResult &Result, 130 ConditionalBranchNode &BranchNode, 131 const Instr &BranchInstrMeta); 132 }; 133 134 } // end namespace cfi_verify 135 } // end namespace llvm 136 137 #endif // LLVM_CFI_VERIFY_GRAPH_BUILDER_H 138