1 //==- CoreEngine.h - Path-Sensitive Dataflow Engine ----------------*- 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 file defines a generic engine for intraprocedural, path-sensitive, 11 // dataflow analysis via graph reachability. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #ifndef LLVM_CLANG_GR_COREENGINE 16 #define LLVM_CLANG_GR_COREENGINE 17 18 #include "clang/AST/Expr.h" 19 #include "clang/Analysis/AnalysisContext.h" 20 #include "clang/StaticAnalyzer/Core/PathSensitive/BlockCounter.h" 21 #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" 22 #include "clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h" 23 #include "clang/StaticAnalyzer/Core/PathSensitive/WorkList.h" 24 #include "llvm/ADT/OwningPtr.h" 25 26 namespace clang { 27 28 class ProgramPointTag; 29 30 namespace ento { 31 32 class NodeBuilder; 33 34 //===----------------------------------------------------------------------===// 35 /// CoreEngine - Implements the core logic of the graph-reachability 36 /// analysis. It traverses the CFG and generates the ExplodedGraph. 37 /// Program "states" are treated as opaque void pointers. 38 /// The template class CoreEngine (which subclasses CoreEngine) 39 /// provides the matching component to the engine that knows the actual types 40 /// for states. Note that this engine only dispatches to transfer functions 41 /// at the statement and block-level. The analyses themselves must implement 42 /// any transfer function logic and the sub-expression level (if any). 43 class CoreEngine { 44 friend struct NodeBuilderContext; 45 friend class NodeBuilder; 46 friend class ExprEngine; 47 friend class CommonNodeBuilder; 48 friend class IndirectGotoNodeBuilder; 49 friend class SwitchNodeBuilder; 50 friend class EndOfFunctionNodeBuilder; 51 public: 52 typedef std::vector<std::pair<BlockEdge, const ExplodedNode*> > 53 BlocksExhausted; 54 55 typedef std::vector<std::pair<const CFGBlock*, const ExplodedNode*> > 56 BlocksAborted; 57 58 private: 59 60 SubEngine& SubEng; 61 62 /// G - The simulation graph. Each node is a (location,state) pair. 63 OwningPtr<ExplodedGraph> G; 64 65 /// WList - A set of queued nodes that need to be processed by the 66 /// worklist algorithm. It is up to the implementation of WList to decide 67 /// the order that nodes are processed. 68 OwningPtr<WorkList> WList; 69 70 /// BCounterFactory - A factory object for created BlockCounter objects. 71 /// These are used to record for key nodes in the ExplodedGraph the 72 /// number of times different CFGBlocks have been visited along a path. 73 BlockCounter::Factory BCounterFactory; 74 75 /// The locations where we stopped doing work because we visited a location 76 /// too many times. 77 BlocksExhausted blocksExhausted; 78 79 /// The locations where we stopped because the engine aborted analysis, 80 /// usually because it could not reason about something. 81 BlocksAborted blocksAborted; 82 83 /// The information about functions shared by the whole translation unit. 84 /// (This data is owned by AnalysisConsumer.) 85 FunctionSummariesTy *FunctionSummaries; 86 87 void generateNode(const ProgramPoint &Loc, 88 ProgramStateRef State, 89 ExplodedNode *Pred); 90 91 void HandleBlockEdge(const BlockEdge &E, ExplodedNode *Pred); 92 void HandleBlockEntrance(const BlockEntrance &E, ExplodedNode *Pred); 93 void HandleBlockExit(const CFGBlock *B, ExplodedNode *Pred); 94 void HandlePostStmt(const CFGBlock *B, unsigned StmtIdx, ExplodedNode *Pred); 95 96 void HandleBranch(const Stmt *Cond, const Stmt *Term, const CFGBlock *B, 97 ExplodedNode *Pred); 98 99 /// Handle conditional logic for running static initializers. 100 void HandleStaticInit(const DeclStmt *DS, const CFGBlock *B, 101 ExplodedNode *Pred); 102 103 private: 104 CoreEngine(const CoreEngine &) LLVM_DELETED_FUNCTION; 105 void operator=(const CoreEngine &) LLVM_DELETED_FUNCTION; 106 107 ExplodedNode *generateCallExitBeginNode(ExplodedNode *N); 108 109 public: 110 /// Construct a CoreEngine object to analyze the provided CFG. CoreEngine(SubEngine & subengine,FunctionSummariesTy * FS)111 CoreEngine(SubEngine& subengine, 112 FunctionSummariesTy *FS) 113 : SubEng(subengine), G(new ExplodedGraph()), 114 WList(WorkList::makeDFS()), 115 BCounterFactory(G->getAllocator()), 116 FunctionSummaries(FS){} 117 118 /// getGraph - Returns the exploded graph. getGraph()119 ExplodedGraph& getGraph() { return *G.get(); } 120 121 /// takeGraph - Returns the exploded graph. Ownership of the graph is 122 /// transferred to the caller. takeGraph()123 ExplodedGraph* takeGraph() { return G.take(); } 124 125 /// ExecuteWorkList - Run the worklist algorithm for a maximum number of 126 /// steps. Returns true if there is still simulation state on the worklist. 127 bool ExecuteWorkList(const LocationContext *L, unsigned Steps, 128 ProgramStateRef InitState); 129 /// Returns true if there is still simulation state on the worklist. 130 bool ExecuteWorkListWithInitialState(const LocationContext *L, 131 unsigned Steps, 132 ProgramStateRef InitState, 133 ExplodedNodeSet &Dst); 134 135 /// Dispatch the work list item based on the given location information. 136 /// Use Pred parameter as the predecessor state. 137 void dispatchWorkItem(ExplodedNode* Pred, ProgramPoint Loc, 138 const WorkListUnit& WU); 139 140 // Functions for external checking of whether we have unfinished work wasBlockAborted()141 bool wasBlockAborted() const { return !blocksAborted.empty(); } wasBlocksExhausted()142 bool wasBlocksExhausted() const { return !blocksExhausted.empty(); } hasWorkRemaining()143 bool hasWorkRemaining() const { return wasBlocksExhausted() || 144 WList->hasWork() || 145 wasBlockAborted(); } 146 147 /// Inform the CoreEngine that a basic block was aborted because 148 /// it could not be completely analyzed. addAbortedBlock(const ExplodedNode * node,const CFGBlock * block)149 void addAbortedBlock(const ExplodedNode *node, const CFGBlock *block) { 150 blocksAborted.push_back(std::make_pair(block, node)); 151 } 152 getWorkList()153 WorkList *getWorkList() const { return WList.get(); } 154 blocks_exhausted_begin()155 BlocksExhausted::const_iterator blocks_exhausted_begin() const { 156 return blocksExhausted.begin(); 157 } blocks_exhausted_end()158 BlocksExhausted::const_iterator blocks_exhausted_end() const { 159 return blocksExhausted.end(); 160 } blocks_aborted_begin()161 BlocksAborted::const_iterator blocks_aborted_begin() const { 162 return blocksAborted.begin(); 163 } blocks_aborted_end()164 BlocksAborted::const_iterator blocks_aborted_end() const { 165 return blocksAborted.end(); 166 } 167 168 /// \brief Enqueue the given set of nodes onto the work list. 169 void enqueue(ExplodedNodeSet &Set); 170 171 /// \brief Enqueue nodes that were created as a result of processing 172 /// a statement onto the work list. 173 void enqueue(ExplodedNodeSet &Set, const CFGBlock *Block, unsigned Idx); 174 175 /// \brief enqueue the nodes corresponding to the end of function onto the 176 /// end of path / work list. 177 void enqueueEndOfFunction(ExplodedNodeSet &Set); 178 179 /// \brief Enqueue a single node created as a result of statement processing. 180 void enqueueStmtNode(ExplodedNode *N, const CFGBlock *Block, unsigned Idx); 181 }; 182 183 // TODO: Turn into a calss. 184 struct NodeBuilderContext { 185 const CoreEngine &Eng; 186 const CFGBlock *Block; 187 const LocationContext *LC; NodeBuilderContextNodeBuilderContext188 NodeBuilderContext(const CoreEngine &E, const CFGBlock *B, ExplodedNode *N) 189 : Eng(E), Block(B), LC(N->getLocationContext()) { assert(B); } 190 191 /// \brief Return the CFGBlock associated with this builder. getBlockNodeBuilderContext192 const CFGBlock *getBlock() const { return Block; } 193 194 /// \brief Returns the number of times the current basic block has been 195 /// visited on the exploded graph path. blockCountNodeBuilderContext196 unsigned blockCount() const { 197 return Eng.WList->getBlockCounter().getNumVisited( 198 LC->getCurrentStackFrame(), 199 Block->getBlockID()); 200 } 201 }; 202 203 /// \class NodeBuilder 204 /// \brief This is the simplest builder which generates nodes in the 205 /// ExplodedGraph. 206 /// 207 /// The main benefit of the builder is that it automatically tracks the 208 /// frontier nodes (or destination set). This is the set of nodes which should 209 /// be propagated to the next step / builder. They are the nodes which have been 210 /// added to the builder (either as the input node set or as the newly 211 /// constructed nodes) but did not have any outgoing transitions added. 212 class NodeBuilder { 213 virtual void anchor(); 214 protected: 215 const NodeBuilderContext &C; 216 217 /// Specifies if the builder results have been finalized. For example, if it 218 /// is set to false, autotransitions are yet to be generated. 219 bool Finalized; 220 bool HasGeneratedNodes; 221 /// \brief The frontier set - a set of nodes which need to be propagated after 222 /// the builder dies. 223 ExplodedNodeSet &Frontier; 224 225 /// Checkes if the results are ready. checkResults()226 virtual bool checkResults() { 227 if (!Finalized) 228 return false; 229 return true; 230 } 231 hasNoSinksInFrontier()232 bool hasNoSinksInFrontier() { 233 for (iterator I = Frontier.begin(), E = Frontier.end(); I != E; ++I) { 234 if ((*I)->isSink()) 235 return false; 236 } 237 return true; 238 } 239 240 /// Allow subclasses to finalize results before result_begin() is executed. finalizeResults()241 virtual void finalizeResults() {} 242 243 ExplodedNode *generateNodeImpl(const ProgramPoint &PP, 244 ProgramStateRef State, 245 ExplodedNode *Pred, 246 bool MarkAsSink = false); 247 248 public: 249 NodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet, 250 const NodeBuilderContext &Ctx, bool F = true) C(Ctx)251 : C(Ctx), Finalized(F), HasGeneratedNodes(false), Frontier(DstSet) { 252 Frontier.Add(SrcNode); 253 } 254 255 NodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet, 256 const NodeBuilderContext &Ctx, bool F = true) C(Ctx)257 : C(Ctx), Finalized(F), HasGeneratedNodes(false), Frontier(DstSet) { 258 Frontier.insert(SrcSet); 259 assert(hasNoSinksInFrontier()); 260 } 261 ~NodeBuilder()262 virtual ~NodeBuilder() {} 263 264 /// \brief Generates a node in the ExplodedGraph. generateNode(const ProgramPoint & PP,ProgramStateRef State,ExplodedNode * Pred)265 ExplodedNode *generateNode(const ProgramPoint &PP, 266 ProgramStateRef State, 267 ExplodedNode *Pred) { 268 return generateNodeImpl(PP, State, Pred, false); 269 } 270 271 /// \brief Generates a sink in the ExplodedGraph. 272 /// 273 /// When a node is marked as sink, the exploration from the node is stopped - 274 /// the node becomes the last node on the path and certain kinds of bugs are 275 /// suppressed. generateSink(const ProgramPoint & PP,ProgramStateRef State,ExplodedNode * Pred)276 ExplodedNode *generateSink(const ProgramPoint &PP, 277 ProgramStateRef State, 278 ExplodedNode *Pred) { 279 return generateNodeImpl(PP, State, Pred, true); 280 } 281 getResults()282 const ExplodedNodeSet &getResults() { 283 finalizeResults(); 284 assert(checkResults()); 285 return Frontier; 286 } 287 288 typedef ExplodedNodeSet::iterator iterator; 289 /// \brief Iterators through the results frontier. begin()290 inline iterator begin() { 291 finalizeResults(); 292 assert(checkResults()); 293 return Frontier.begin(); 294 } end()295 inline iterator end() { 296 finalizeResults(); 297 return Frontier.end(); 298 } 299 getContext()300 const NodeBuilderContext &getContext() { return C; } hasGeneratedNodes()301 bool hasGeneratedNodes() { return HasGeneratedNodes; } 302 takeNodes(const ExplodedNodeSet & S)303 void takeNodes(const ExplodedNodeSet &S) { 304 for (ExplodedNodeSet::iterator I = S.begin(), E = S.end(); I != E; ++I ) 305 Frontier.erase(*I); 306 } takeNodes(ExplodedNode * N)307 void takeNodes(ExplodedNode *N) { Frontier.erase(N); } addNodes(const ExplodedNodeSet & S)308 void addNodes(const ExplodedNodeSet &S) { Frontier.insert(S); } addNodes(ExplodedNode * N)309 void addNodes(ExplodedNode *N) { Frontier.Add(N); } 310 }; 311 312 /// \class NodeBuilderWithSinks 313 /// \brief This node builder keeps track of the generated sink nodes. 314 class NodeBuilderWithSinks: public NodeBuilder { 315 virtual void anchor(); 316 protected: 317 SmallVector<ExplodedNode*, 2> sinksGenerated; 318 ProgramPoint &Location; 319 320 public: NodeBuilderWithSinks(ExplodedNode * Pred,ExplodedNodeSet & DstSet,const NodeBuilderContext & Ctx,ProgramPoint & L)321 NodeBuilderWithSinks(ExplodedNode *Pred, ExplodedNodeSet &DstSet, 322 const NodeBuilderContext &Ctx, ProgramPoint &L) 323 : NodeBuilder(Pred, DstSet, Ctx), Location(L) {} 324 325 ExplodedNode *generateNode(ProgramStateRef State, 326 ExplodedNode *Pred, 327 const ProgramPointTag *Tag = 0) { 328 const ProgramPoint &LocalLoc = (Tag ? Location.withTag(Tag) : Location); 329 return NodeBuilder::generateNode(LocalLoc, State, Pred); 330 } 331 332 ExplodedNode *generateSink(ProgramStateRef State, ExplodedNode *Pred, 333 const ProgramPointTag *Tag = 0) { 334 const ProgramPoint &LocalLoc = (Tag ? Location.withTag(Tag) : Location); 335 ExplodedNode *N = NodeBuilder::generateSink(LocalLoc, State, Pred); 336 if (N && N->isSink()) 337 sinksGenerated.push_back(N); 338 return N; 339 } 340 getSinks()341 const SmallVectorImpl<ExplodedNode*> &getSinks() const { 342 return sinksGenerated; 343 } 344 }; 345 346 /// \class StmtNodeBuilder 347 /// \brief This builder class is useful for generating nodes that resulted from 348 /// visiting a statement. The main difference from its parent NodeBuilder is 349 /// that it creates a statement specific ProgramPoint. 350 class StmtNodeBuilder: public NodeBuilder { 351 NodeBuilder *EnclosingBldr; 352 public: 353 354 /// \brief Constructs a StmtNodeBuilder. If the builder is going to process 355 /// nodes currently owned by another builder(with larger scope), use 356 /// Enclosing builder to transfer ownership. 357 StmtNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet, 358 const NodeBuilderContext &Ctx, NodeBuilder *Enclosing = 0) NodeBuilder(SrcNode,DstSet,Ctx)359 : NodeBuilder(SrcNode, DstSet, Ctx), EnclosingBldr(Enclosing) { 360 if (EnclosingBldr) 361 EnclosingBldr->takeNodes(SrcNode); 362 } 363 364 StmtNodeBuilder(ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet, 365 const NodeBuilderContext &Ctx, NodeBuilder *Enclosing = 0) NodeBuilder(SrcSet,DstSet,Ctx)366 : NodeBuilder(SrcSet, DstSet, Ctx), EnclosingBldr(Enclosing) { 367 if (EnclosingBldr) 368 for (ExplodedNodeSet::iterator I = SrcSet.begin(), 369 E = SrcSet.end(); I != E; ++I ) 370 EnclosingBldr->takeNodes(*I); 371 } 372 373 virtual ~StmtNodeBuilder(); 374 375 using NodeBuilder::generateNode; 376 using NodeBuilder::generateSink; 377 378 ExplodedNode *generateNode(const Stmt *S, 379 ExplodedNode *Pred, 380 ProgramStateRef St, 381 const ProgramPointTag *tag = 0, 382 ProgramPoint::Kind K = ProgramPoint::PostStmtKind){ 383 const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K, 384 Pred->getLocationContext(), tag); 385 return NodeBuilder::generateNode(L, St, Pred); 386 } 387 388 ExplodedNode *generateSink(const Stmt *S, 389 ExplodedNode *Pred, 390 ProgramStateRef St, 391 const ProgramPointTag *tag = 0, 392 ProgramPoint::Kind K = ProgramPoint::PostStmtKind){ 393 const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K, 394 Pred->getLocationContext(), tag); 395 return NodeBuilder::generateSink(L, St, Pred); 396 } 397 }; 398 399 /// \brief BranchNodeBuilder is responsible for constructing the nodes 400 /// corresponding to the two branches of the if statement - true and false. 401 class BranchNodeBuilder: public NodeBuilder { 402 virtual void anchor(); 403 const CFGBlock *DstT; 404 const CFGBlock *DstF; 405 406 bool InFeasibleTrue; 407 bool InFeasibleFalse; 408 409 public: BranchNodeBuilder(ExplodedNode * SrcNode,ExplodedNodeSet & DstSet,const NodeBuilderContext & C,const CFGBlock * dstT,const CFGBlock * dstF)410 BranchNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet, 411 const NodeBuilderContext &C, 412 const CFGBlock *dstT, const CFGBlock *dstF) 413 : NodeBuilder(SrcNode, DstSet, C), DstT(dstT), DstF(dstF), 414 InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) { 415 // The branch node builder does not generate autotransitions. 416 // If there are no successors it means that both branches are infeasible. 417 takeNodes(SrcNode); 418 } 419 BranchNodeBuilder(const ExplodedNodeSet & SrcSet,ExplodedNodeSet & DstSet,const NodeBuilderContext & C,const CFGBlock * dstT,const CFGBlock * dstF)420 BranchNodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet, 421 const NodeBuilderContext &C, 422 const CFGBlock *dstT, const CFGBlock *dstF) 423 : NodeBuilder(SrcSet, DstSet, C), DstT(dstT), DstF(dstF), 424 InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) { 425 takeNodes(SrcSet); 426 } 427 428 ExplodedNode *generateNode(ProgramStateRef State, bool branch, 429 ExplodedNode *Pred); 430 getTargetBlock(bool branch)431 const CFGBlock *getTargetBlock(bool branch) const { 432 return branch ? DstT : DstF; 433 } 434 markInfeasible(bool branch)435 void markInfeasible(bool branch) { 436 if (branch) 437 InFeasibleTrue = true; 438 else 439 InFeasibleFalse = true; 440 } 441 isFeasible(bool branch)442 bool isFeasible(bool branch) { 443 return branch ? !InFeasibleTrue : !InFeasibleFalse; 444 } 445 }; 446 447 class IndirectGotoNodeBuilder { 448 CoreEngine& Eng; 449 const CFGBlock *Src; 450 const CFGBlock &DispatchBlock; 451 const Expr *E; 452 ExplodedNode *Pred; 453 454 public: IndirectGotoNodeBuilder(ExplodedNode * pred,const CFGBlock * src,const Expr * e,const CFGBlock * dispatch,CoreEngine * eng)455 IndirectGotoNodeBuilder(ExplodedNode *pred, const CFGBlock *src, 456 const Expr *e, const CFGBlock *dispatch, CoreEngine* eng) 457 : Eng(*eng), Src(src), DispatchBlock(*dispatch), E(e), Pred(pred) {} 458 459 class iterator { 460 CFGBlock::const_succ_iterator I; 461 462 friend class IndirectGotoNodeBuilder; iterator(CFGBlock::const_succ_iterator i)463 iterator(CFGBlock::const_succ_iterator i) : I(i) {} 464 public: 465 466 iterator &operator++() { ++I; return *this; } 467 bool operator!=(const iterator &X) const { return I != X.I; } 468 getLabel()469 const LabelDecl *getLabel() const { 470 return cast<LabelStmt>((*I)->getLabel())->getDecl(); 471 } 472 getBlock()473 const CFGBlock *getBlock() const { 474 return *I; 475 } 476 }; 477 begin()478 iterator begin() { return iterator(DispatchBlock.succ_begin()); } end()479 iterator end() { return iterator(DispatchBlock.succ_end()); } 480 481 ExplodedNode *generateNode(const iterator &I, 482 ProgramStateRef State, 483 bool isSink = false); 484 getTarget()485 const Expr *getTarget() const { return E; } 486 getState()487 ProgramStateRef getState() const { return Pred->State; } 488 getLocationContext()489 const LocationContext *getLocationContext() const { 490 return Pred->getLocationContext(); 491 } 492 }; 493 494 class SwitchNodeBuilder { 495 CoreEngine& Eng; 496 const CFGBlock *Src; 497 const Expr *Condition; 498 ExplodedNode *Pred; 499 500 public: SwitchNodeBuilder(ExplodedNode * pred,const CFGBlock * src,const Expr * condition,CoreEngine * eng)501 SwitchNodeBuilder(ExplodedNode *pred, const CFGBlock *src, 502 const Expr *condition, CoreEngine* eng) 503 : Eng(*eng), Src(src), Condition(condition), Pred(pred) {} 504 505 class iterator { 506 CFGBlock::const_succ_reverse_iterator I; 507 508 friend class SwitchNodeBuilder; iterator(CFGBlock::const_succ_reverse_iterator i)509 iterator(CFGBlock::const_succ_reverse_iterator i) : I(i) {} 510 511 public: 512 iterator &operator++() { ++I; return *this; } 513 bool operator!=(const iterator &X) const { return I != X.I; } 514 bool operator==(const iterator &X) const { return I == X.I; } 515 getCase()516 const CaseStmt *getCase() const { 517 return cast<CaseStmt>((*I)->getLabel()); 518 } 519 getBlock()520 const CFGBlock *getBlock() const { 521 return *I; 522 } 523 }; 524 begin()525 iterator begin() { return iterator(Src->succ_rbegin()+1); } end()526 iterator end() { return iterator(Src->succ_rend()); } 527 getSwitch()528 const SwitchStmt *getSwitch() const { 529 return cast<SwitchStmt>(Src->getTerminator()); 530 } 531 532 ExplodedNode *generateCaseStmtNode(const iterator &I, 533 ProgramStateRef State); 534 535 ExplodedNode *generateDefaultCaseNode(ProgramStateRef State, 536 bool isSink = false); 537 getCondition()538 const Expr *getCondition() const { return Condition; } 539 getState()540 ProgramStateRef getState() const { return Pred->State; } 541 getLocationContext()542 const LocationContext *getLocationContext() const { 543 return Pred->getLocationContext(); 544 } 545 }; 546 547 } // end ento namespace 548 } // end clang namespace 549 550 #endif 551