1 //== MIGChecker.cpp - MIG calling convention checker ------------*- C++ -*--==// 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 defines MIGChecker, a Mach Interface Generator calling convention 10 // checker. Namely, in MIG callback implementation the following rules apply: 11 // - When a server routine returns an error code that represents success, it 12 // must take ownership of resources passed to it (and eventually release 13 // them). 14 // - Additionally, when returning success, all out-parameters must be 15 // initialized. 16 // - When it returns any other error code, it must not take ownership, 17 // because the message and its out-of-line parameters will be destroyed 18 // by the client that called the function. 19 // For now we only check the last rule, as its violations lead to dangerous 20 // use-after-free exploits. 21 // 22 //===----------------------------------------------------------------------===// 23 24 #include "clang/Analysis/AnyCall.h" 25 #include "clang/StaticAnalyzer/Checkers/BuiltinCheckerRegistration.h" 26 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 27 #include "clang/StaticAnalyzer/Core/Checker.h" 28 #include "clang/StaticAnalyzer/Core/CheckerManager.h" 29 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h" 30 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" 31 32 using namespace clang; 33 using namespace ento; 34 35 namespace { 36 class MIGChecker : public Checker<check::PostCall, check::PreStmt<ReturnStmt>, 37 check::EndFunction> { 38 BugType BT{this, "Use-after-free (MIG calling convention violation)", 39 categories::MemoryError}; 40 41 // The checker knows that an out-of-line object is deallocated if it is 42 // passed as an argument to one of these functions. If this object is 43 // additionally an argument of a MIG routine, the checker keeps track of that 44 // information and issues a warning when an error is returned from the 45 // respective routine. 46 std::vector<std::pair<CallDescription, unsigned>> Deallocators = { 47 #define CALL(required_args, deallocated_arg, ...) \ 48 {{{__VA_ARGS__}, required_args}, deallocated_arg} 49 // E.g., if the checker sees a C function 'vm_deallocate' that is 50 // defined on class 'IOUserClient' that has exactly 3 parameters, it knows 51 // that argument #1 (starting from 0, i.e. the second argument) is going 52 // to be consumed in the sense of the MIG consume-on-success convention. 53 CALL(3, 1, "vm_deallocate"), 54 CALL(3, 1, "mach_vm_deallocate"), 55 CALL(2, 0, "mig_deallocate"), 56 CALL(2, 1, "mach_port_deallocate"), 57 CALL(1, 0, "device_deallocate"), 58 CALL(1, 0, "iokit_remove_connect_reference"), 59 CALL(1, 0, "iokit_remove_reference"), 60 CALL(1, 0, "iokit_release_port"), 61 CALL(1, 0, "ipc_port_release"), 62 CALL(1, 0, "ipc_port_release_sonce"), 63 CALL(1, 0, "ipc_voucher_attr_control_release"), 64 CALL(1, 0, "ipc_voucher_release"), 65 CALL(1, 0, "lock_set_dereference"), 66 CALL(1, 0, "memory_object_control_deallocate"), 67 CALL(1, 0, "pset_deallocate"), 68 CALL(1, 0, "semaphore_dereference"), 69 CALL(1, 0, "space_deallocate"), 70 CALL(1, 0, "space_inspect_deallocate"), 71 CALL(1, 0, "task_deallocate"), 72 CALL(1, 0, "task_inspect_deallocate"), 73 CALL(1, 0, "task_name_deallocate"), 74 CALL(1, 0, "thread_deallocate"), 75 CALL(1, 0, "thread_inspect_deallocate"), 76 CALL(1, 0, "upl_deallocate"), 77 CALL(1, 0, "vm_map_deallocate"), 78 // E.g., if the checker sees a method 'releaseAsyncReference64()' that is 79 // defined on class 'IOUserClient' that takes exactly 1 argument, it knows 80 // that the argument is going to be consumed in the sense of the MIG 81 // consume-on-success convention. 82 CALL(1, 0, "IOUserClient", "releaseAsyncReference64"), 83 CALL(1, 0, "IOUserClient", "releaseNotificationPort"), 84 #undef CALL 85 }; 86 87 void checkReturnAux(const ReturnStmt *RS, CheckerContext &C) const; 88 89 public: 90 void checkPostCall(const CallEvent &Call, CheckerContext &C) const; 91 92 // HACK: We're making two attempts to find the bug: checkEndFunction 93 // should normally be enough but it fails when the return value is a literal 94 // that never gets put into the Environment and ends of function with multiple 95 // returns get agglutinated across returns, preventing us from obtaining 96 // the return value. The problem is similar to https://reviews.llvm.org/D25326 97 // but now we step into it in the top-level function. 98 void checkPreStmt(const ReturnStmt *RS, CheckerContext &C) const { 99 checkReturnAux(RS, C); 100 } 101 void checkEndFunction(const ReturnStmt *RS, CheckerContext &C) const { 102 checkReturnAux(RS, C); 103 } 104 105 }; 106 } // end anonymous namespace 107 108 REGISTER_TRAIT_WITH_PROGRAMSTATE(ReleasedParameter, bool) 109 110 static const ParmVarDecl *getOriginParam(SVal V, CheckerContext &C) { 111 SymbolRef Sym = V.getAsSymbol(); 112 if (!Sym) 113 return nullptr; 114 115 // If we optimistically assume that the MIG routine never re-uses the storage 116 // that was passed to it as arguments when it invalidates it (but at most when 117 // it assigns to parameter variables directly), this procedure correctly 118 // determines if the value was loaded from the transitive closure of MIG 119 // routine arguments in the heap. 120 while (const MemRegion *MR = Sym->getOriginRegion()) { 121 const auto *VR = dyn_cast<VarRegion>(MR); 122 if (VR && VR->hasStackParametersStorage() && 123 VR->getStackFrame()->inTopFrame()) 124 return cast<ParmVarDecl>(VR->getDecl()); 125 126 const SymbolicRegion *SR = MR->getSymbolicBase(); 127 if (!SR) 128 return nullptr; 129 130 Sym = SR->getSymbol(); 131 } 132 133 return nullptr; 134 } 135 136 static bool isInMIGCall(CheckerContext &C) { 137 const LocationContext *LC = C.getLocationContext(); 138 const StackFrameContext *SFC; 139 // Find the top frame. 140 while (LC) { 141 SFC = LC->getStackFrame(); 142 LC = SFC->getParent(); 143 } 144 145 const Decl *D = SFC->getDecl(); 146 147 if (Optional<AnyCall> AC = AnyCall::forDecl(D)) { 148 // Even though there's a Sema warning when the return type of an annotated 149 // function is not a kern_return_t, this warning isn't an error, so we need 150 // an extra sanity check here. 151 // FIXME: AnyCall doesn't support blocks yet, so they remain unchecked 152 // for now. 153 if (!AC->getReturnType(C.getASTContext()) 154 .getCanonicalType()->isSignedIntegerType()) 155 return false; 156 } 157 158 if (D->hasAttr<MIGServerRoutineAttr>()) 159 return true; 160 161 // See if there's an annotated method in the superclass. 162 if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) 163 for (const auto *OMD: MD->overridden_methods()) 164 if (OMD->hasAttr<MIGServerRoutineAttr>()) 165 return true; 166 167 return false; 168 } 169 170 void MIGChecker::checkPostCall(const CallEvent &Call, CheckerContext &C) const { 171 if (!isInMIGCall(C)) 172 return; 173 174 auto I = llvm::find_if(Deallocators, 175 [&](const std::pair<CallDescription, unsigned> &Item) { 176 return Call.isCalled(Item.first); 177 }); 178 if (I == Deallocators.end()) 179 return; 180 181 unsigned ArgIdx = I->second; 182 SVal Arg = Call.getArgSVal(ArgIdx); 183 const ParmVarDecl *PVD = getOriginParam(Arg, C); 184 if (!PVD) 185 return; 186 187 const NoteTag *T = C.getNoteTag([this, PVD](BugReport &BR) -> std::string { 188 if (&BR.getBugType() != &BT) 189 return ""; 190 SmallString<64> Str; 191 llvm::raw_svector_ostream OS(Str); 192 OS << "Value passed through parameter '" << PVD->getName() 193 << "\' is deallocated"; 194 return OS.str(); 195 }); 196 C.addTransition(C.getState()->set<ReleasedParameter>(true), T); 197 } 198 199 // Returns true if V can potentially represent a "successful" kern_return_t. 200 static bool mayBeSuccess(SVal V, CheckerContext &C) { 201 ProgramStateRef State = C.getState(); 202 203 // Can V represent KERN_SUCCESS? 204 if (!State->isNull(V).isConstrainedFalse()) 205 return true; 206 207 SValBuilder &SVB = C.getSValBuilder(); 208 ASTContext &ACtx = C.getASTContext(); 209 210 // Can V represent MIG_NO_REPLY? 211 static const int MigNoReply = -305; 212 V = SVB.evalEQ(C.getState(), V, SVB.makeIntVal(MigNoReply, ACtx.IntTy)); 213 if (!State->isNull(V).isConstrainedTrue()) 214 return true; 215 216 // If none of the above, it's definitely an error. 217 return false; 218 } 219 220 void MIGChecker::checkReturnAux(const ReturnStmt *RS, CheckerContext &C) const { 221 // It is very unlikely that a MIG callback will be called from anywhere 222 // within the project under analysis and the caller isn't itself a routine 223 // that follows the MIG calling convention. Therefore we're safe to believe 224 // that it's always the top frame that is of interest. There's a slight chance 225 // that the user would want to enforce the MIG calling convention upon 226 // a random routine in the middle of nowhere, but given that the convention is 227 // fairly weird and hard to follow in the first place, there's relatively 228 // little motivation to spread it this way. 229 if (!C.inTopFrame()) 230 return; 231 232 if (!isInMIGCall(C)) 233 return; 234 235 // We know that the function is non-void, but what if the return statement 236 // is not there in the code? It's not a compile error, we should not crash. 237 if (!RS) 238 return; 239 240 ProgramStateRef State = C.getState(); 241 if (!State->get<ReleasedParameter>()) 242 return; 243 244 SVal V = C.getSVal(RS); 245 if (mayBeSuccess(V, C)) 246 return; 247 248 ExplodedNode *N = C.generateErrorNode(); 249 if (!N) 250 return; 251 252 auto R = llvm::make_unique<BugReport>( 253 BT, 254 "MIG callback fails with error after deallocating argument value. " 255 "This is a use-after-free vulnerability because the caller will try to " 256 "deallocate it again", 257 N); 258 259 R->addRange(RS->getSourceRange()); 260 bugreporter::trackExpressionValue(N, RS->getRetValue(), *R, false); 261 C.emitReport(std::move(R)); 262 } 263 264 void ento::registerMIGChecker(CheckerManager &Mgr) { 265 Mgr.registerChecker<MIGChecker>(); 266 } 267 268 bool ento::shouldRegisterMIGChecker(const LangOptions &LO) { 269 return true; 270 } 271