Home | History | Annotate | Download | only in PathSensitive
      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/ExplodedGraph.h"
     21 #include "clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h"
     22 #include "clang/StaticAnalyzer/Core/PathSensitive/WorkList.h"
     23 #include "clang/StaticAnalyzer/Core/PathSensitive/BlockCounter.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 private:
    100   CoreEngine(const CoreEngine&); // Do not implement.
    101   CoreEngine& operator=(const CoreEngine&);
    102 
    103   ExplodedNode *generateCallExitBeginNode(ExplodedNode *N);
    104 
    105 public:
    106   /// Construct a CoreEngine object to analyze the provided CFG.
    107   CoreEngine(SubEngine& subengine,
    108              FunctionSummariesTy *FS)
    109     : SubEng(subengine), G(new ExplodedGraph()),
    110       WList(WorkList::makeDFS()),
    111       BCounterFactory(G->getAllocator()),
    112       FunctionSummaries(FS){}
    113 
    114   /// getGraph - Returns the exploded graph.
    115   ExplodedGraph& getGraph() { return *G.get(); }
    116 
    117   /// takeGraph - Returns the exploded graph.  Ownership of the graph is
    118   ///  transferred to the caller.
    119   ExplodedGraph* takeGraph() { return G.take(); }
    120 
    121   /// ExecuteWorkList - Run the worklist algorithm for a maximum number of
    122   ///  steps.  Returns true if there is still simulation state on the worklist.
    123   bool ExecuteWorkList(const LocationContext *L, unsigned Steps,
    124                        ProgramStateRef InitState);
    125   /// Returns true if there is still simulation state on the worklist.
    126   bool ExecuteWorkListWithInitialState(const LocationContext *L,
    127                                        unsigned Steps,
    128                                        ProgramStateRef InitState,
    129                                        ExplodedNodeSet &Dst);
    130 
    131   /// Dispatch the work list item based on the given location information.
    132   /// Use Pred parameter as the predecessor state.
    133   void dispatchWorkItem(ExplodedNode* Pred, ProgramPoint Loc,
    134                         const WorkListUnit& WU);
    135 
    136   // Functions for external checking of whether we have unfinished work
    137   bool wasBlockAborted() const { return !blocksAborted.empty(); }
    138   bool wasBlocksExhausted() const { return !blocksExhausted.empty(); }
    139   bool hasWorkRemaining() const { return wasBlocksExhausted() ||
    140                                          WList->hasWork() ||
    141                                          wasBlockAborted(); }
    142 
    143   /// Inform the CoreEngine that a basic block was aborted because
    144   /// it could not be completely analyzed.
    145   void addAbortedBlock(const ExplodedNode *node, const CFGBlock *block) {
    146     blocksAborted.push_back(std::make_pair(block, node));
    147   }
    148 
    149   WorkList *getWorkList() const { return WList.get(); }
    150 
    151   BlocksExhausted::const_iterator blocks_exhausted_begin() const {
    152     return blocksExhausted.begin();
    153   }
    154   BlocksExhausted::const_iterator blocks_exhausted_end() const {
    155     return blocksExhausted.end();
    156   }
    157   BlocksAborted::const_iterator blocks_aborted_begin() const {
    158     return blocksAborted.begin();
    159   }
    160   BlocksAborted::const_iterator blocks_aborted_end() const {
    161     return blocksAborted.end();
    162   }
    163 
    164   /// \brief Enqueue the given set of nodes onto the work list.
    165   void enqueue(ExplodedNodeSet &Set);
    166 
    167   /// \brief Enqueue nodes that were created as a result of processing
    168   /// a statement onto the work list.
    169   void enqueue(ExplodedNodeSet &Set, const CFGBlock *Block, unsigned Idx);
    170 
    171   /// \brief enqueue the nodes corresponding to the end of function onto the
    172   /// end of path / work list.
    173   void enqueueEndOfFunction(ExplodedNodeSet &Set);
    174 
    175   /// \brief Enqueue a single node created as a result of statement processing.
    176   void enqueueStmtNode(ExplodedNode *N, const CFGBlock *Block, unsigned Idx);
    177 };
    178 
    179 // TODO: Turn into a calss.
    180 struct NodeBuilderContext {
    181   const CoreEngine &Eng;
    182   const CFGBlock *Block;
    183   ExplodedNode *Pred;
    184   NodeBuilderContext(const CoreEngine &E, const CFGBlock *B, ExplodedNode *N)
    185     : Eng(E), Block(B), Pred(N) { assert(B); assert(!N->isSink()); }
    186 
    187   ExplodedNode *getPred() const { return Pred; }
    188 
    189   /// \brief Return the CFGBlock associated with this builder.
    190   const CFGBlock *getBlock() const { return Block; }
    191 
    192   /// \brief Returns the number of times the current basic block has been
    193   /// visited on the exploded graph path.
    194   unsigned blockCount() const {
    195     return Eng.WList->getBlockCounter().getNumVisited(
    196                     Pred->getLocationContext()->getCurrentStackFrame(),
    197                     Block->getBlockID());
    198   }
    199 };
    200 
    201 /// \class NodeBuilder
    202 /// \brief This is the simplest builder which generates nodes in the
    203 /// ExplodedGraph.
    204 ///
    205 /// The main benefit of the builder is that it automatically tracks the
    206 /// frontier nodes (or destination set). This is the set of nodes which should
    207 /// be propagated to the next step / builder. They are the nodes which have been
    208 /// added to the builder (either as the input node set or as the newly
    209 /// constructed nodes) but did not have any outgoing transitions added.
    210 class NodeBuilder {
    211   virtual void anchor();
    212 protected:
    213   const NodeBuilderContext &C;
    214 
    215   /// Specifies if the builder results have been finalized. For example, if it
    216   /// is set to false, autotransitions are yet to be generated.
    217   bool Finalized;
    218   bool HasGeneratedNodes;
    219   /// \brief The frontier set - a set of nodes which need to be propagated after
    220   /// the builder dies.
    221   ExplodedNodeSet &Frontier;
    222 
    223   /// Checkes if the results are ready.
    224   virtual bool checkResults() {
    225     if (!Finalized)
    226       return false;
    227     return true;
    228   }
    229 
    230   bool hasNoSinksInFrontier() {
    231     for (iterator I = Frontier.begin(), E = Frontier.end(); I != E; ++I) {
    232       if ((*I)->isSink())
    233         return false;
    234     }
    235     return true;
    236   }
    237 
    238   /// Allow subclasses to finalize results before result_begin() is executed.
    239   virtual void finalizeResults() {}
    240 
    241   ExplodedNode *generateNodeImpl(const ProgramPoint &PP,
    242                                  ProgramStateRef State,
    243                                  ExplodedNode *Pred,
    244                                  bool MarkAsSink = false);
    245 
    246 public:
    247   NodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
    248               const NodeBuilderContext &Ctx, bool F = true)
    249     : C(Ctx), Finalized(F), HasGeneratedNodes(false), Frontier(DstSet) {
    250     Frontier.Add(SrcNode);
    251   }
    252 
    253   NodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
    254               const NodeBuilderContext &Ctx, bool F = true)
    255     : C(Ctx), Finalized(F), HasGeneratedNodes(false), Frontier(DstSet) {
    256     Frontier.insert(SrcSet);
    257     assert(hasNoSinksInFrontier());
    258   }
    259 
    260   virtual ~NodeBuilder() {}
    261 
    262   /// \brief Generates a node in the ExplodedGraph.
    263   ExplodedNode *generateNode(const ProgramPoint &PP,
    264                              ProgramStateRef State,
    265                              ExplodedNode *Pred) {
    266     return generateNodeImpl(PP, State, Pred, false);
    267   }
    268 
    269   /// \brief Generates a sink in the ExplodedGraph.
    270   ///
    271   /// When a node is marked as sink, the exploration from the node is stopped -
    272   /// the node becomes the last node on the path and certain kinds of bugs are
    273   /// suppressed.
    274   ExplodedNode *generateSink(const ProgramPoint &PP,
    275                              ProgramStateRef State,
    276                              ExplodedNode *Pred) {
    277     return generateNodeImpl(PP, State, Pred, true);
    278   }
    279 
    280   const ExplodedNodeSet &getResults() {
    281     finalizeResults();
    282     assert(checkResults());
    283     return Frontier;
    284   }
    285 
    286   typedef ExplodedNodeSet::iterator iterator;
    287   /// \brief Iterators through the results frontier.
    288   inline iterator begin() {
    289     finalizeResults();
    290     assert(checkResults());
    291     return Frontier.begin();
    292   }
    293   inline iterator end() {
    294     finalizeResults();
    295     return Frontier.end();
    296   }
    297 
    298   const NodeBuilderContext &getContext() { return C; }
    299   bool hasGeneratedNodes() { return HasGeneratedNodes; }
    300 
    301   void takeNodes(const ExplodedNodeSet &S) {
    302     for (ExplodedNodeSet::iterator I = S.begin(), E = S.end(); I != E; ++I )
    303       Frontier.erase(*I);
    304   }
    305   void takeNodes(ExplodedNode *N) { Frontier.erase(N); }
    306   void addNodes(const ExplodedNodeSet &S) { Frontier.insert(S); }
    307   void addNodes(ExplodedNode *N) { Frontier.Add(N); }
    308 };
    309 
    310 /// \class NodeBuilderWithSinks
    311 /// \brief This node builder keeps track of the generated sink nodes.
    312 class NodeBuilderWithSinks: public NodeBuilder {
    313   virtual void anchor();
    314 protected:
    315   SmallVector<ExplodedNode*, 2> sinksGenerated;
    316   ProgramPoint &Location;
    317 
    318 public:
    319   NodeBuilderWithSinks(ExplodedNode *Pred, ExplodedNodeSet &DstSet,
    320                        const NodeBuilderContext &Ctx, ProgramPoint &L)
    321     : NodeBuilder(Pred, DstSet, Ctx), Location(L) {}
    322 
    323   ExplodedNode *generateNode(ProgramStateRef State,
    324                              ExplodedNode *Pred,
    325                              const ProgramPointTag *Tag = 0) {
    326     const ProgramPoint &LocalLoc = (Tag ? Location.withTag(Tag) : Location);
    327     return NodeBuilder::generateNode(LocalLoc, State, Pred);
    328   }
    329 
    330   ExplodedNode *generateSink(ProgramStateRef State, ExplodedNode *Pred,
    331                              const ProgramPointTag *Tag = 0) {
    332     const ProgramPoint &LocalLoc = (Tag ? Location.withTag(Tag) : Location);
    333     ExplodedNode *N = NodeBuilder::generateSink(LocalLoc, State, Pred);
    334     if (N && N->isSink())
    335       sinksGenerated.push_back(N);
    336     return N;
    337   }
    338 
    339   const SmallVectorImpl<ExplodedNode*> &getSinks() const {
    340     return sinksGenerated;
    341   }
    342 };
    343 
    344 /// \class StmtNodeBuilder
    345 /// \brief This builder class is useful for generating nodes that resulted from
    346 /// visiting a statement. The main difference from its parent NodeBuilder is
    347 /// that it creates a statement specific ProgramPoint.
    348 class StmtNodeBuilder: public NodeBuilder {
    349   NodeBuilder *EnclosingBldr;
    350 public:
    351 
    352   /// \brief Constructs a StmtNodeBuilder. If the builder is going to process
    353   /// nodes currently owned by another builder(with larger scope), use
    354   /// Enclosing builder to transfer ownership.
    355   StmtNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
    356                       const NodeBuilderContext &Ctx, NodeBuilder *Enclosing = 0)
    357     : NodeBuilder(SrcNode, DstSet, Ctx), EnclosingBldr(Enclosing) {
    358     if (EnclosingBldr)
    359       EnclosingBldr->takeNodes(SrcNode);
    360   }
    361 
    362   StmtNodeBuilder(ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
    363                       const NodeBuilderContext &Ctx, NodeBuilder *Enclosing = 0)
    364     : NodeBuilder(SrcSet, DstSet, Ctx), EnclosingBldr(Enclosing) {
    365     if (EnclosingBldr)
    366       for (ExplodedNodeSet::iterator I = SrcSet.begin(),
    367                                      E = SrcSet.end(); I != E; ++I )
    368         EnclosingBldr->takeNodes(*I);
    369   }
    370 
    371   virtual ~StmtNodeBuilder();
    372 
    373   using NodeBuilder::generateNode;
    374   using NodeBuilder::generateSink;
    375 
    376   ExplodedNode *generateNode(const Stmt *S,
    377                              ExplodedNode *Pred,
    378                              ProgramStateRef St,
    379                              const ProgramPointTag *tag = 0,
    380                              ProgramPoint::Kind K = ProgramPoint::PostStmtKind){
    381     const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K,
    382                                   Pred->getLocationContext(), tag);
    383     return NodeBuilder::generateNode(L, St, Pred);
    384   }
    385 
    386   ExplodedNode *generateSink(const Stmt *S,
    387                              ExplodedNode *Pred,
    388                              ProgramStateRef St,
    389                              const ProgramPointTag *tag = 0,
    390                              ProgramPoint::Kind K = ProgramPoint::PostStmtKind){
    391     const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K,
    392                                   Pred->getLocationContext(), tag);
    393     return NodeBuilder::generateSink(L, St, Pred);
    394   }
    395 };
    396 
    397 /// \brief BranchNodeBuilder is responsible for constructing the nodes
    398 /// corresponding to the two branches of the if statement - true and false.
    399 class BranchNodeBuilder: public NodeBuilder {
    400   virtual void anchor();
    401   const CFGBlock *DstT;
    402   const CFGBlock *DstF;
    403 
    404   bool InFeasibleTrue;
    405   bool InFeasibleFalse;
    406 
    407 public:
    408   BranchNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
    409                     const NodeBuilderContext &C,
    410                     const CFGBlock *dstT, const CFGBlock *dstF)
    411   : NodeBuilder(SrcNode, DstSet, C), DstT(dstT), DstF(dstF),
    412     InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) {
    413     // The branch node builder does not generate autotransitions.
    414     // If there are no successors it means that both branches are infeasible.
    415     takeNodes(SrcNode);
    416   }
    417 
    418   BranchNodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
    419                     const NodeBuilderContext &C,
    420                     const CFGBlock *dstT, const CFGBlock *dstF)
    421   : NodeBuilder(SrcSet, DstSet, C), DstT(dstT), DstF(dstF),
    422     InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) {
    423     takeNodes(SrcSet);
    424   }
    425 
    426   ExplodedNode *generateNode(ProgramStateRef State, bool branch,
    427                              ExplodedNode *Pred);
    428 
    429   const CFGBlock *getTargetBlock(bool branch) const {
    430     return branch ? DstT : DstF;
    431   }
    432 
    433   void markInfeasible(bool branch) {
    434     if (branch)
    435       InFeasibleTrue = true;
    436     else
    437       InFeasibleFalse = true;
    438   }
    439 
    440   bool isFeasible(bool branch) {
    441     return branch ? !InFeasibleTrue : !InFeasibleFalse;
    442   }
    443 };
    444 
    445 class IndirectGotoNodeBuilder {
    446   CoreEngine& Eng;
    447   const CFGBlock *Src;
    448   const CFGBlock &DispatchBlock;
    449   const Expr *E;
    450   ExplodedNode *Pred;
    451 
    452 public:
    453   IndirectGotoNodeBuilder(ExplodedNode *pred, const CFGBlock *src,
    454                     const Expr *e, const CFGBlock *dispatch, CoreEngine* eng)
    455     : Eng(*eng), Src(src), DispatchBlock(*dispatch), E(e), Pred(pred) {}
    456 
    457   class iterator {
    458     CFGBlock::const_succ_iterator I;
    459 
    460     friend class IndirectGotoNodeBuilder;
    461     iterator(CFGBlock::const_succ_iterator i) : I(i) {}
    462   public:
    463 
    464     iterator &operator++() { ++I; return *this; }
    465     bool operator!=(const iterator &X) const { return I != X.I; }
    466 
    467     const LabelDecl *getLabel() const {
    468       return llvm::cast<LabelStmt>((*I)->getLabel())->getDecl();
    469     }
    470 
    471     const CFGBlock *getBlock() const {
    472       return *I;
    473     }
    474   };
    475 
    476   iterator begin() { return iterator(DispatchBlock.succ_begin()); }
    477   iterator end() { return iterator(DispatchBlock.succ_end()); }
    478 
    479   ExplodedNode *generateNode(const iterator &I,
    480                              ProgramStateRef State,
    481                              bool isSink = false);
    482 
    483   const Expr *getTarget() const { return E; }
    484 
    485   ProgramStateRef getState() const { return Pred->State; }
    486 
    487   const LocationContext *getLocationContext() const {
    488     return Pred->getLocationContext();
    489   }
    490 };
    491 
    492 class SwitchNodeBuilder {
    493   CoreEngine& Eng;
    494   const CFGBlock *Src;
    495   const Expr *Condition;
    496   ExplodedNode *Pred;
    497 
    498 public:
    499   SwitchNodeBuilder(ExplodedNode *pred, const CFGBlock *src,
    500                     const Expr *condition, CoreEngine* eng)
    501   : Eng(*eng), Src(src), Condition(condition), Pred(pred) {}
    502 
    503   class iterator {
    504     CFGBlock::const_succ_reverse_iterator I;
    505 
    506     friend class SwitchNodeBuilder;
    507     iterator(CFGBlock::const_succ_reverse_iterator i) : I(i) {}
    508 
    509   public:
    510     iterator &operator++() { ++I; return *this; }
    511     bool operator!=(const iterator &X) const { return I != X.I; }
    512     bool operator==(const iterator &X) const { return I == X.I; }
    513 
    514     const CaseStmt *getCase() const {
    515       return llvm::cast<CaseStmt>((*I)->getLabel());
    516     }
    517 
    518     const CFGBlock *getBlock() const {
    519       return *I;
    520     }
    521   };
    522 
    523   iterator begin() { return iterator(Src->succ_rbegin()+1); }
    524   iterator end() { return iterator(Src->succ_rend()); }
    525 
    526   const SwitchStmt *getSwitch() const {
    527     return llvm::cast<SwitchStmt>(Src->getTerminator());
    528   }
    529 
    530   ExplodedNode *generateCaseStmtNode(const iterator &I,
    531                                      ProgramStateRef State);
    532 
    533   ExplodedNode *generateDefaultCaseNode(ProgramStateRef State,
    534                                         bool isSink = false);
    535 
    536   const Expr *getCondition() const { return Condition; }
    537 
    538   ProgramStateRef getState() const { return Pred->State; }
    539 
    540   const LocationContext *getLocationContext() const {
    541     return Pred->getLocationContext();
    542   }
    543 };
    544 
    545 } // end ento namespace
    546 } // end clang namespace
    547 
    548 #endif
    549