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