ObjCSelfInitChecker.cpp revision 2f3017f9cbd3774f690c979410bfec38423d03af
1//== ObjCSelfInitChecker.cpp - Checker for 'self' initialization -*- 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// This defines ObjCSelfInitChecker, a builtin check that checks for uses of 11// 'self' before proper initialization. 12// 13//===----------------------------------------------------------------------===// 14 15// This checks initialization methods to verify that they assign 'self' to the 16// result of an initialization call (e.g. [super init], or [self initWith..]) 17// before using 'self' or any instance variable. 18// 19// To perform the required checking, values are tagged with flags that indicate 20// 1) if the object is the one pointed to by 'self', and 2) if the object 21// is the result of an initializer (e.g. [super init]). 22// 23// Uses of an object that is true for 1) but not 2) trigger a diagnostic. 24// The uses that are currently checked are: 25// - Using instance variables. 26// - Returning the object. 27// 28// Note that we don't check for an invalid 'self' that is the receiver of an 29// obj-c message expression to cut down false positives where logging functions 30// get information from self (like its class) or doing "invalidation" on self 31// when the initialization fails. 32// 33// Because the object that 'self' points to gets invalidated when a call 34// receives a reference to 'self', the checker keeps track and passes the flags 35// for 1) and 2) to the new object that 'self' points to after the call. 36// 37//===----------------------------------------------------------------------===// 38 39#include "ClangSACheckers.h" 40#include "clang/StaticAnalyzer/Core/Checker.h" 41#include "clang/StaticAnalyzer/Core/CheckerManager.h" 42#include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h" 43#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" 44#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" 45#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 46#include "clang/AST/ParentMap.h" 47 48using namespace clang; 49using namespace ento; 50 51static bool shouldRunOnFunctionOrMethod(const NamedDecl *ND); 52static bool isInitializationMethod(const ObjCMethodDecl *MD); 53static bool isInitMessage(const ObjCMethodCall &Msg); 54static bool isSelfVar(SVal location, CheckerContext &C); 55 56namespace { 57class ObjCSelfInitChecker : public Checker< check::PostObjCMessage, 58 check::PostStmt<ObjCIvarRefExpr>, 59 check::PreStmt<ReturnStmt>, 60 check::PreCall, 61 check::PostCall, 62 check::Location, 63 check::Bind > { 64public: 65 void checkPostObjCMessage(const ObjCMethodCall &Msg, CheckerContext &C) const; 66 void checkPostStmt(const ObjCIvarRefExpr *E, CheckerContext &C) const; 67 void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; 68 void checkLocation(SVal location, bool isLoad, const Stmt *S, 69 CheckerContext &C) const; 70 void checkBind(SVal loc, SVal val, const Stmt *S, CheckerContext &C) const; 71 72 void checkPreCall(const CallEvent &CE, CheckerContext &C) const; 73 void checkPostCall(const CallEvent &CE, CheckerContext &C) const; 74 75 void printState(raw_ostream &Out, ProgramStateRef State, 76 const char *NL, const char *Sep) const; 77}; 78} // end anonymous namespace 79 80namespace { 81 82class InitSelfBug : public BugType { 83 const std::string desc; 84public: 85 InitSelfBug() : BugType("Missing \"self = [(super or self) init...]\"", 86 categories::CoreFoundationObjectiveC) {} 87}; 88 89} // end anonymous namespace 90 91namespace { 92enum SelfFlagEnum { 93 /// \brief No flag set. 94 SelfFlag_None = 0x0, 95 /// \brief Value came from 'self'. 96 SelfFlag_Self = 0x1, 97 /// \brief Value came from the result of an initializer (e.g. [super init]). 98 SelfFlag_InitRes = 0x2 99}; 100} 101 102REGISTER_MAP_WITH_PROGRAMSTATE(SelfFlag, SymbolRef, unsigned) 103REGISTER_TRAIT_WITH_PROGRAMSTATE(CalledInit, bool) 104 105/// \brief A call receiving a reference to 'self' invalidates the object that 106/// 'self' contains. This keeps the "self flags" assigned to the 'self' 107/// object before the call so we can assign them to the new object that 'self' 108/// points to after the call. 109REGISTER_TRAIT_WITH_PROGRAMSTATE(PreCallSelfFlags, unsigned) 110 111static SelfFlagEnum getSelfFlags(SVal val, ProgramStateRef state) { 112 if (SymbolRef sym = val.getAsSymbol()) 113 if (const unsigned *attachedFlags = state->get<SelfFlag>(sym)) 114 return (SelfFlagEnum)*attachedFlags; 115 return SelfFlag_None; 116} 117 118static SelfFlagEnum getSelfFlags(SVal val, CheckerContext &C) { 119 return getSelfFlags(val, C.getState()); 120} 121 122static void addSelfFlag(ProgramStateRef state, SVal val, 123 SelfFlagEnum flag, CheckerContext &C) { 124 // We tag the symbol that the SVal wraps. 125 if (SymbolRef sym = val.getAsSymbol()) 126 state = state->set<SelfFlag>(sym, getSelfFlags(val, state) | flag); 127 C.addTransition(state); 128} 129 130static bool hasSelfFlag(SVal val, SelfFlagEnum flag, CheckerContext &C) { 131 return getSelfFlags(val, C) & flag; 132} 133 134/// \brief Returns true of the value of the expression is the object that 'self' 135/// points to and is an object that did not come from the result of calling 136/// an initializer. 137static bool isInvalidSelf(const Expr *E, CheckerContext &C) { 138 SVal exprVal = C.getState()->getSVal(E, C.getLocationContext()); 139 if (!hasSelfFlag(exprVal, SelfFlag_Self, C)) 140 return false; // value did not come from 'self'. 141 if (hasSelfFlag(exprVal, SelfFlag_InitRes, C)) 142 return false; // 'self' is properly initialized. 143 144 return true; 145} 146 147static void checkForInvalidSelf(const Expr *E, CheckerContext &C, 148 const char *errorStr) { 149 if (!E) 150 return; 151 152 if (!C.getState()->get<CalledInit>()) 153 return; 154 155 if (!isInvalidSelf(E, C)) 156 return; 157 158 // Generate an error node. 159 ExplodedNode *N = C.generateSink(); 160 if (!N) 161 return; 162 163 BugReport *report = 164 new BugReport(*new InitSelfBug(), errorStr, N); 165 C.emitReport(report); 166} 167 168void ObjCSelfInitChecker::checkPostObjCMessage(const ObjCMethodCall &Msg, 169 CheckerContext &C) const { 170 // When encountering a message that does initialization (init rule), 171 // tag the return value so that we know later on that if self has this value 172 // then it is properly initialized. 173 174 // FIXME: A callback should disable checkers at the start of functions. 175 if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( 176 C.getCurrentAnalysisDeclContext()->getDecl()))) 177 return; 178 179 if (isInitMessage(Msg)) { 180 // Tag the return value as the result of an initializer. 181 ProgramStateRef state = C.getState(); 182 183 // FIXME this really should be context sensitive, where we record 184 // the current stack frame (for IPA). Also, we need to clean this 185 // value out when we return from this method. 186 state = state->set<CalledInit>(true); 187 188 SVal V = state->getSVal(Msg.getOriginExpr(), C.getLocationContext()); 189 addSelfFlag(state, V, SelfFlag_InitRes, C); 190 return; 191 } 192 193 // We don't check for an invalid 'self' in an obj-c message expression to cut 194 // down false positives where logging functions get information from self 195 // (like its class) or doing "invalidation" on self when the initialization 196 // fails. 197} 198 199void ObjCSelfInitChecker::checkPostStmt(const ObjCIvarRefExpr *E, 200 CheckerContext &C) const { 201 // FIXME: A callback should disable checkers at the start of functions. 202 if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( 203 C.getCurrentAnalysisDeclContext()->getDecl()))) 204 return; 205 206 checkForInvalidSelf(E->getBase(), C, 207 "Instance variable used while 'self' is not set to the result of " 208 "'[(super or self) init...]'"); 209} 210 211void ObjCSelfInitChecker::checkPreStmt(const ReturnStmt *S, 212 CheckerContext &C) const { 213 // FIXME: A callback should disable checkers at the start of functions. 214 if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( 215 C.getCurrentAnalysisDeclContext()->getDecl()))) 216 return; 217 218 checkForInvalidSelf(S->getRetValue(), C, 219 "Returning 'self' while it is not set to the result of " 220 "'[(super or self) init...]'"); 221} 222 223// When a call receives a reference to 'self', [Pre/Post]Call pass 224// the SelfFlags from the object 'self' points to before the call to the new 225// object after the call. This is to avoid invalidation of 'self' by logging 226// functions. 227// Another common pattern in classes with multiple initializers is to put the 228// subclass's common initialization bits into a static function that receives 229// the value of 'self', e.g: 230// @code 231// if (!(self = [super init])) 232// return nil; 233// if (!(self = _commonInit(self))) 234// return nil; 235// @endcode 236// Until we can use inter-procedural analysis, in such a call, transfer the 237// SelfFlags to the result of the call. 238 239void ObjCSelfInitChecker::checkPreCall(const CallEvent &CE, 240 CheckerContext &C) const { 241 // FIXME: A callback should disable checkers at the start of functions. 242 if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( 243 C.getCurrentAnalysisDeclContext()->getDecl()))) 244 return; 245 246 ProgramStateRef state = C.getState(); 247 unsigned NumArgs = CE.getNumArgs(); 248 // If we passed 'self' as and argument to the call, record it in the state 249 // to be propagated after the call. 250 // Note, we could have just given up, but try to be more optimistic here and 251 // assume that the functions are going to continue initialization or will not 252 // modify self. 253 for (unsigned i = 0; i < NumArgs; ++i) { 254 SVal argV = CE.getArgSVal(i); 255 if (isSelfVar(argV, C)) { 256 unsigned selfFlags = getSelfFlags(state->getSVal(cast<Loc>(argV)), C); 257 C.addTransition(state->set<PreCallSelfFlags>(selfFlags)); 258 return; 259 } else if (hasSelfFlag(argV, SelfFlag_Self, C)) { 260 unsigned selfFlags = getSelfFlags(argV, C); 261 C.addTransition(state->set<PreCallSelfFlags>(selfFlags)); 262 return; 263 } 264 } 265} 266 267void ObjCSelfInitChecker::checkPostCall(const CallEvent &CE, 268 CheckerContext &C) const { 269 // FIXME: A callback should disable checkers at the start of functions. 270 if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( 271 C.getCurrentAnalysisDeclContext()->getDecl()))) 272 return; 273 274 ProgramStateRef state = C.getState(); 275 SelfFlagEnum prevFlags = (SelfFlagEnum)state->get<PreCallSelfFlags>(); 276 if (!prevFlags) 277 return; 278 state = state->remove<PreCallSelfFlags>(); 279 280 unsigned NumArgs = CE.getNumArgs(); 281 for (unsigned i = 0; i < NumArgs; ++i) { 282 SVal argV = CE.getArgSVal(i); 283 if (isSelfVar(argV, C)) { 284 // If the address of 'self' is being passed to the call, assume that the 285 // 'self' after the call will have the same flags. 286 // EX: log(&self) 287 addSelfFlag(state, state->getSVal(cast<Loc>(argV)), prevFlags, C); 288 return; 289 } else if (hasSelfFlag(argV, SelfFlag_Self, C)) { 290 // If 'self' is passed to the call by value, assume that the function 291 // returns 'self'. So assign the flags, which were set on 'self' to the 292 // return value. 293 // EX: self = performMoreInitialization(self) 294 addSelfFlag(state, CE.getReturnValue(), prevFlags, C); 295 return; 296 } 297 } 298 299 C.addTransition(state); 300} 301 302void ObjCSelfInitChecker::checkLocation(SVal location, bool isLoad, 303 const Stmt *S, 304 CheckerContext &C) const { 305 // Tag the result of a load from 'self' so that we can easily know that the 306 // value is the object that 'self' points to. 307 ProgramStateRef state = C.getState(); 308 if (isSelfVar(location, C)) 309 addSelfFlag(state, state->getSVal(cast<Loc>(location)), SelfFlag_Self, C); 310} 311 312 313void ObjCSelfInitChecker::checkBind(SVal loc, SVal val, const Stmt *S, 314 CheckerContext &C) const { 315 // Allow assignment of anything to self. Self is a local variable in the 316 // initializer, so it is legal to assign anything to it, like results of 317 // static functions/method calls. After self is assigned something we cannot 318 // reason about, stop enforcing the rules. 319 // (Only continue checking if the assigned value should be treated as self.) 320 if ((isSelfVar(loc, C)) && 321 !hasSelfFlag(val, SelfFlag_InitRes, C) && 322 !hasSelfFlag(val, SelfFlag_Self, C) && 323 !isSelfVar(val, C)) { 324 325 // Stop tracking the checker-specific state in the state. 326 ProgramStateRef State = C.getState(); 327 State = State->remove<CalledInit>(); 328 if (SymbolRef sym = loc.getAsSymbol()) 329 State = State->remove<SelfFlag>(sym); 330 C.addTransition(State); 331 } 332} 333 334void ObjCSelfInitChecker::printState(raw_ostream &Out, ProgramStateRef State, 335 const char *NL, const char *Sep) const { 336 SelfFlagTy FlagMap = State->get<SelfFlag>(); 337 bool DidCallInit = State->get<CalledInit>(); 338 SelfFlagEnum PreCallFlags = (SelfFlagEnum)State->get<PreCallSelfFlags>(); 339 340 if (FlagMap.isEmpty() && !DidCallInit && !PreCallFlags) 341 return; 342 343 Out << Sep << NL << "ObjCSelfInitChecker:" << NL; 344 345 if (DidCallInit) 346 Out << " An init method has been called." << NL; 347 348 if (PreCallFlags != SelfFlag_None) { 349 if (PreCallFlags & SelfFlag_Self) { 350 Out << " An argument of the current call came from the 'self' variable." 351 << NL; 352 } 353 if (PreCallFlags & SelfFlag_InitRes) { 354 Out << " An argument of the current call came from an init method." 355 << NL; 356 } 357 } 358 359 Out << NL; 360 for (SelfFlagTy::iterator I = FlagMap.begin(), E = FlagMap.end(); 361 I != E; ++I) { 362 Out << I->first << " : "; 363 364 if (I->second == SelfFlag_None) 365 Out << "none"; 366 367 if (I->second & SelfFlag_Self) 368 Out << "self variable"; 369 370 if (I->second & SelfFlag_InitRes) { 371 if (I->second != SelfFlag_InitRes) 372 Out << " | "; 373 Out << "result of init method"; 374 } 375 376 Out << NL; 377 } 378} 379 380 381// FIXME: A callback should disable checkers at the start of functions. 382static bool shouldRunOnFunctionOrMethod(const NamedDecl *ND) { 383 if (!ND) 384 return false; 385 386 const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(ND); 387 if (!MD) 388 return false; 389 if (!isInitializationMethod(MD)) 390 return false; 391 392 // self = [super init] applies only to NSObject subclasses. 393 // For instance, NSProxy doesn't implement -init. 394 ASTContext &Ctx = MD->getASTContext(); 395 IdentifierInfo* NSObjectII = &Ctx.Idents.get("NSObject"); 396 ObjCInterfaceDecl *ID = MD->getClassInterface()->getSuperClass(); 397 for ( ; ID ; ID = ID->getSuperClass()) { 398 IdentifierInfo *II = ID->getIdentifier(); 399 400 if (II == NSObjectII) 401 break; 402 } 403 if (!ID) 404 return false; 405 406 return true; 407} 408 409/// \brief Returns true if the location is 'self'. 410static bool isSelfVar(SVal location, CheckerContext &C) { 411 AnalysisDeclContext *analCtx = C.getCurrentAnalysisDeclContext(); 412 if (!analCtx->getSelfDecl()) 413 return false; 414 if (!isa<loc::MemRegionVal>(location)) 415 return false; 416 417 loc::MemRegionVal MRV = cast<loc::MemRegionVal>(location); 418 if (const DeclRegion *DR = dyn_cast<DeclRegion>(MRV.stripCasts())) 419 return (DR->getDecl() == analCtx->getSelfDecl()); 420 421 return false; 422} 423 424static bool isInitializationMethod(const ObjCMethodDecl *MD) { 425 return MD->getMethodFamily() == OMF_init; 426} 427 428static bool isInitMessage(const ObjCMethodCall &Call) { 429 return Call.getMethodFamily() == OMF_init; 430} 431 432//===----------------------------------------------------------------------===// 433// Registration. 434//===----------------------------------------------------------------------===// 435 436void ento::registerObjCSelfInitChecker(CheckerManager &mgr) { 437 mgr.registerChecker<ObjCSelfInitChecker>(); 438} 439