1 //===-- CodeGen/AsmPrinter/WinCFGuard.cpp - Control Flow Guard Impl ------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file contains support for writing the metadata for Windows Control Flow
10 // Guard, including address-taken functions and valid longjmp targets.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "WinCFGuard.h"
15 #include "llvm/CodeGen/AsmPrinter.h"
16 #include "llvm/CodeGen/MachineFunction.h"
17 #include "llvm/CodeGen/MachineModuleInfo.h"
18 #include "llvm/IR/Constants.h"
19 #include "llvm/IR/InstrTypes.h"
20 #include "llvm/MC/MCObjectFileInfo.h"
21 #include "llvm/MC/MCStreamer.h"
22 
23 #include <vector>
24 
25 using namespace llvm;
26 
WinCFGuard(AsmPrinter * A)27 WinCFGuard::WinCFGuard(AsmPrinter *A) : Asm(A) {}
28 
29 WinCFGuard::~WinCFGuard() = default;
30 
endFunction(const MachineFunction * MF)31 void WinCFGuard::endFunction(const MachineFunction *MF) {
32 
33   // Skip functions without any longjmp targets.
34   if (MF->getLongjmpTargets().empty())
35     return;
36 
37   // Copy the function's longjmp targets to a module-level list.
38   llvm::append_range(LongjmpTargets, MF->getLongjmpTargets());
39 }
40 
41 /// Returns true if this function's address is escaped in a way that might make
42 /// it an indirect call target. Function::hasAddressTaken gives different
43 /// results when a function is called directly with a function prototype
44 /// mismatch, which requires a cast.
isPossibleIndirectCallTarget(const Function * F)45 static bool isPossibleIndirectCallTarget(const Function *F) {
46   SmallVector<const Value *, 4> Users{F};
47   while (!Users.empty()) {
48     const Value *FnOrCast = Users.pop_back_val();
49     for (const Use &U : FnOrCast->uses()) {
50       const User *FnUser = U.getUser();
51       if (isa<BlockAddress>(FnUser))
52         continue;
53       if (const auto *Call = dyn_cast<CallBase>(FnUser)) {
54         if (!Call->isCallee(&U))
55           return true;
56       } else if (isa<Instruction>(FnUser)) {
57         // Consider any other instruction to be an escape. This has some weird
58         // consequences like no-op intrinsics being an escape or a store *to* a
59         // function address being an escape.
60         return true;
61       } else if (const auto *C = dyn_cast<Constant>(FnUser)) {
62         // If this is a constant pointer cast of the function, don't consider
63         // this escape. Analyze the uses of the cast as well. This ensures that
64         // direct calls with mismatched prototypes don't end up in the CFG
65         // table. Consider other constants, such as vtable initializers, to
66         // escape the function.
67         if (C->stripPointerCasts() == F)
68           Users.push_back(FnUser);
69         else
70           return true;
71       }
72     }
73   }
74   return false;
75 }
76 
lookupImpSymbol(const MCSymbol * Sym)77 MCSymbol *WinCFGuard::lookupImpSymbol(const MCSymbol *Sym) {
78   if (Sym->getName().starts_with("__imp_"))
79     return nullptr;
80   return Asm->OutContext.lookupSymbol(Twine("__imp_") + Sym->getName());
81 }
82 
endModule()83 void WinCFGuard::endModule() {
84   const Module *M = Asm->MMI->getModule();
85   std::vector<const MCSymbol *> GFIDsEntries;
86   std::vector<const MCSymbol *> GIATsEntries;
87   for (const Function &F : *M) {
88     if (isPossibleIndirectCallTarget(&F)) {
89       // If F is a dllimport and has an "__imp_" symbol already defined, add the
90       // "__imp_" symbol to the .giats section.
91       if (F.hasDLLImportStorageClass()) {
92         if (MCSymbol *impSym = lookupImpSymbol(Asm->getSymbol(&F))) {
93           GIATsEntries.push_back(impSym);
94         }
95       }
96       // Add the function's symbol to the .gfids section.
97       // Note: For dllimport functions, MSVC sometimes does not add this symbol
98       // to the .gfids section, but only adds the corresponding "__imp_" symbol
99       // to the .giats section. Here we always add the symbol to the .gfids
100       // section, since this does not introduce security risks.
101       GFIDsEntries.push_back(Asm->getSymbol(&F));
102     }
103   }
104 
105   if (GFIDsEntries.empty() && GIATsEntries.empty() && LongjmpTargets.empty())
106     return;
107 
108   // Emit the symbol index of each GFIDs entry to form the .gfids section.
109   auto &OS = *Asm->OutStreamer;
110   OS.switchSection(Asm->OutContext.getObjectFileInfo()->getGFIDsSection());
111   for (const MCSymbol *S : GFIDsEntries)
112     OS.emitCOFFSymbolIndex(S);
113 
114   // Emit the symbol index of each GIATs entry to form the .giats section.
115   OS.switchSection(Asm->OutContext.getObjectFileInfo()->getGIATsSection());
116   for (const MCSymbol *S : GIATsEntries) {
117     OS.emitCOFFSymbolIndex(S);
118   }
119 
120   // Emit the symbol index of each longjmp target to form the .gljmp section.
121   OS.switchSection(Asm->OutContext.getObjectFileInfo()->getGLJMPSection());
122   for (const MCSymbol *S : LongjmpTargets) {
123     OS.emitCOFFSymbolIndex(S);
124   }
125 }
126