1//===- CoreEngine.h - Path-Sensitive Dataflow Engine ------------*- 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 a generic engine for intraprocedural, path-sensitive,
10// dataflow analysis via graph reachability.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_CLANG_STATICANALYZER_CORE_PATHSENSITIVE_COREENGINE_H
15#define LLVM_CLANG_STATICANALYZER_CORE_PATHSENSITIVE_COREENGINE_H
16
17#include "clang/AST/Stmt.h"
18#include "clang/Analysis/AnalysisDeclContext.h"
19#include "clang/Analysis/CFG.h"
20#include "clang/Analysis/ProgramPoint.h"
21#include "clang/Basic/LLVM.h"
22#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h"
23#include "clang/StaticAnalyzer/Core/PathSensitive/BlockCounter.h"
24#include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h"
25#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState_Fwd.h"
26#include "clang/StaticAnalyzer/Core/PathSensitive/WorkList.h"
27#include "llvm/ADT/SmallVector.h"
28#include "llvm/ADT/iterator_range.h"
29#include "llvm/Support/Casting.h"
30#include <cassert>
31#include <memory>
32#include <utility>
33#include <vector>
34
35namespace clang {
36
37class AnalyzerOptions;
38class CXXBindTemporaryExpr;
39class Expr;
40class LabelDecl;
41
42namespace ento {
43
44class FunctionSummariesTy;
45class ExprEngine;
46
47//===----------------------------------------------------------------------===//
48/// CoreEngine - Implements the core logic of the graph-reachability
49/// analysis. It traverses the CFG and generates the ExplodedGraph.
50/// Program "states" are treated as opaque void pointers.
51/// The template class CoreEngine (which subclasses CoreEngine)
52/// provides the matching component to the engine that knows the actual types
53/// for states. Note that this engine only dispatches to transfer functions
54/// at the statement and block-level. The analyses themselves must implement
55/// any transfer function logic and the sub-expression level (if any).
56class CoreEngine {
57 friend class CommonNodeBuilder;
58 friend class EndOfFunctionNodeBuilder;
59 friend class ExprEngine;
60 friend class IndirectGotoNodeBuilder;
61 friend class NodeBuilder;
62 friend class NodeBuilderContext;
63 friend class SwitchNodeBuilder;
64
65public:
66 using BlocksExhausted =
67 std::vector<std::pair<BlockEdge, const ExplodedNode *>>;
68
69 using BlocksAborted =
70 std::vector<std::pair<const CFGBlock *, const ExplodedNode *>>;
71
72private:
73 ExprEngine &ExprEng;
74
75 /// G - The simulation graph. Each node is a (location,state) pair.
76 mutable ExplodedGraph G;
77
78 /// WList - A set of queued nodes that need to be processed by the
79 /// worklist algorithm. It is up to the implementation of WList to decide
80 /// the order that nodes are processed.
81 std::unique_ptr<WorkList> WList;
82 std::unique_ptr<WorkList> CTUWList;
83
84 /// BCounterFactory - A factory object for created BlockCounter objects.
85 /// These are used to record for key nodes in the ExplodedGraph the
86 /// number of times different CFGBlocks have been visited along a path.
87 BlockCounter::Factory BCounterFactory;
88
89 /// The locations where we stopped doing work because we visited a location
90 /// too many times.
91 BlocksExhausted blocksExhausted;
92
93 /// The locations where we stopped because the engine aborted analysis,
94 /// usually because it could not reason about something.
95 BlocksAborted blocksAborted;
96
97 /// The information about functions shared by the whole translation unit.
98 /// (This data is owned by AnalysisConsumer.)
99 FunctionSummariesTy *FunctionSummaries;
100
101 /// Add path tags with some useful data along the path when we see that
102 /// something interesting is happening. This field is the allocator for such
103 /// tags.
104 DataTag::Factory DataTags;
105
106 void setBlockCounter(BlockCounter C);
107
108 void generateNode(const ProgramPoint &Loc,
109 ProgramStateRef State,
110 ExplodedNode *Pred);
111
112 void HandleBlockEdge(const BlockEdge &E, ExplodedNode *Pred);
113 void HandleBlockEntrance(const BlockEntrance &E, ExplodedNode *Pred);
114 void HandleBlockExit(const CFGBlock *B, ExplodedNode *Pred);
115
116 void HandleCallEnter(const CallEnter &CE, ExplodedNode *Pred);
117
118 void HandlePostStmt(const CFGBlock *B, unsigned StmtIdx, ExplodedNode *Pred);
119
120 void HandleBranch(const Stmt *Cond, const Stmt *Term, const CFGBlock *B,
121 ExplodedNode *Pred);
122 void HandleCleanupTemporaryBranch(const CXXBindTemporaryExpr *BTE,
123 const CFGBlock *B, ExplodedNode *Pred);
124
125 /// Handle conditional logic for running static initializers.
126 void HandleStaticInit(const DeclStmt *DS, const CFGBlock *B,
127 ExplodedNode *Pred);
128
129 void HandleVirtualBaseBranch(const CFGBlock *B, ExplodedNode *Pred);
130
131private:
132 ExplodedNode *generateCallExitBeginNode(ExplodedNode *N,
133 const ReturnStmt *RS);
134
135public:
136 /// Construct a CoreEngine object to analyze the provided CFG.
137 CoreEngine(ExprEngine &exprengine,
138 FunctionSummariesTy *FS,
139 AnalyzerOptions &Opts);
140
141 CoreEngine(const CoreEngine &) = delete;
142 CoreEngine &operator=(const CoreEngine &) = delete;
143
144 /// getGraph - Returns the exploded graph.
145 ExplodedGraph &getGraph() { return G; }
146
147 /// ExecuteWorkList - Run the worklist algorithm for a maximum number of
148 /// steps. Returns true if there is still simulation state on the worklist.
149 bool ExecuteWorkList(const LocationContext *L, unsigned Steps,
150 ProgramStateRef InitState);
151
152 /// Dispatch the work list item based on the given location information.
153 /// Use Pred parameter as the predecessor state.
154 void dispatchWorkItem(ExplodedNode* Pred, ProgramPoint Loc,
155 const WorkListUnit& WU);
156
157 // Functions for external checking of whether we have unfinished work
158 bool wasBlockAborted() const { return !blocksAborted.empty(); }
159 bool wasBlocksExhausted() const { return !blocksExhausted.empty(); }
160 bool hasWorkRemaining() const { return wasBlocksExhausted() ||
161 WList->hasWork() ||
162 wasBlockAborted(); }
163
164 /// Inform the CoreEngine that a basic block was aborted because
165 /// it could not be completely analyzed.
166 void addAbortedBlock(const ExplodedNode *node, const CFGBlock *block) {
167 blocksAborted.push_back(x: std::make_pair(x&: block, y&: node));
168 }
169
170 WorkList *getWorkList() const { return WList.get(); }
171 WorkList *getCTUWorkList() const { return CTUWList.get(); }
172
173 auto exhausted_blocks() const {
174 return llvm::iterator_range(blocksExhausted);
175 }
176
177 auto aborted_blocks() const { return llvm::iterator_range(blocksAborted); }
178
179 /// Enqueue the given set of nodes onto the work list.
180 void enqueue(ExplodedNodeSet &Set);
181
182 /// Enqueue nodes that were created as a result of processing
183 /// a statement onto the work list.
184 void enqueue(ExplodedNodeSet &Set, const CFGBlock *Block, unsigned Idx);
185
186 /// enqueue the nodes corresponding to the end of function onto the
187 /// end of path / work list.
188 void enqueueEndOfFunction(ExplodedNodeSet &Set, const ReturnStmt *RS);
189
190 /// Enqueue a single node created as a result of statement processing.
191 void enqueueStmtNode(ExplodedNode *N, const CFGBlock *Block, unsigned Idx);
192
193 DataTag::Factory &getDataTags() { return DataTags; }
194};
195
196class NodeBuilderContext {
197 const CoreEngine &Eng;
198 const CFGBlock *Block;
199 const LocationContext *LC;
200
201public:
202 NodeBuilderContext(const CoreEngine &E, const CFGBlock *B,
203 const LocationContext *L)
204 : Eng(E), Block(B), LC(L) {
205 assert(B);
206 }
207
208 NodeBuilderContext(const CoreEngine &E, const CFGBlock *B, ExplodedNode *N)
209 : NodeBuilderContext(E, B, N->getLocationContext()) {}
210
211 /// Return the CoreEngine associated with this builder.
212 const CoreEngine &getEngine() const { return Eng; }
213
214 /// Return the CFGBlock associated with this builder.
215 const CFGBlock *getBlock() const { return Block; }
216
217 /// Return the location context associated with this builder.
218 const LocationContext *getLocationContext() const { return LC; }
219
220 /// Returns the number of times the current basic block has been
221 /// visited on the exploded graph path.
222 unsigned blockCount() const {
223 return Eng.WList->getBlockCounter().getNumVisited(
224 CallSite: LC->getStackFrame(),
225 BlockID: Block->getBlockID());
226 }
227};
228
229/// \class NodeBuilder
230/// This is the simplest builder which generates nodes in the
231/// ExplodedGraph.
232///
233/// The main benefit of the builder is that it automatically tracks the
234/// frontier nodes (or destination set). This is the set of nodes which should
235/// be propagated to the next step / builder. They are the nodes which have been
236/// added to the builder (either as the input node set or as the newly
237/// constructed nodes) but did not have any outgoing transitions added.
238class NodeBuilder {
239 virtual void anchor();
240
241protected:
242 const NodeBuilderContext &C;
243
244 /// Specifies if the builder results have been finalized. For example, if it
245 /// is set to false, autotransitions are yet to be generated.
246 bool Finalized;
247
248 bool HasGeneratedNodes = false;
249
250 /// The frontier set - a set of nodes which need to be propagated after
251 /// the builder dies.
252 ExplodedNodeSet &Frontier;
253
254 /// Checks if the results are ready.
255 virtual bool checkResults() {
256 return Finalized;
257 }
258
259 bool hasNoSinksInFrontier() {
260 for (const auto I : Frontier)
261 if (I->isSink())
262 return false;
263 return true;
264 }
265
266 /// Allow subclasses to finalize results before result_begin() is executed.
267 virtual void finalizeResults() {}
268
269 ExplodedNode *generateNodeImpl(const ProgramPoint &PP,
270 ProgramStateRef State,
271 ExplodedNode *Pred,
272 bool MarkAsSink = false);
273
274public:
275 NodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
276 const NodeBuilderContext &Ctx, bool F = true)
277 : C(Ctx), Finalized(F), Frontier(DstSet) {
278 Frontier.Add(N: SrcNode);
279 }
280
281 NodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
282 const NodeBuilderContext &Ctx, bool F = true)
283 : C(Ctx), Finalized(F), Frontier(DstSet) {
284 Frontier.insert(S: SrcSet);
285 assert(hasNoSinksInFrontier());
286 }
287
288 virtual ~NodeBuilder() = default;
289
290 /// Generates a node in the ExplodedGraph.
291 ExplodedNode *generateNode(const ProgramPoint &PP,
292 ProgramStateRef State,
293 ExplodedNode *Pred) {
294 return generateNodeImpl(
295 PP, State, Pred,
296 /*MarkAsSink=*/MarkAsSink: State->isPosteriorlyOverconstrained());
297 }
298
299 /// Generates a sink in the ExplodedGraph.
300 ///
301 /// When a node is marked as sink, the exploration from the node is stopped -
302 /// the node becomes the last node on the path and certain kinds of bugs are
303 /// suppressed.
304 ExplodedNode *generateSink(const ProgramPoint &PP,
305 ProgramStateRef State,
306 ExplodedNode *Pred) {
307 return generateNodeImpl(PP, State, Pred, MarkAsSink: true);
308 }
309
310 const ExplodedNodeSet &getResults() {
311 finalizeResults();
312 assert(checkResults());
313 return Frontier;
314 }
315
316 using iterator = ExplodedNodeSet::iterator;
317
318 /// Iterators through the results frontier.
319 iterator begin() {
320 finalizeResults();
321 assert(checkResults());
322 return Frontier.begin();
323 }
324
325 iterator end() {
326 finalizeResults();
327 return Frontier.end();
328 }
329
330 const NodeBuilderContext &getContext() { return C; }
331 bool hasGeneratedNodes() { return HasGeneratedNodes; }
332
333 void takeNodes(const ExplodedNodeSet &S) {
334 for (const auto I : S)
335 Frontier.erase(N: I);
336 }
337
338 void takeNodes(ExplodedNode *N) { Frontier.erase(N); }
339 void addNodes(const ExplodedNodeSet &S) { Frontier.insert(S); }
340 void addNodes(ExplodedNode *N) { Frontier.Add(N); }
341};
342
343/// \class NodeBuilderWithSinks
344/// This node builder keeps track of the generated sink nodes.
345class NodeBuilderWithSinks: public NodeBuilder {
346 void anchor() override;
347
348protected:
349 SmallVector<ExplodedNode*, 2> sinksGenerated;
350 ProgramPoint &Location;
351
352public:
353 NodeBuilderWithSinks(ExplodedNode *Pred, ExplodedNodeSet &DstSet,
354 const NodeBuilderContext &Ctx, ProgramPoint &L)
355 : NodeBuilder(Pred, DstSet, Ctx), Location(L) {}
356
357 ExplodedNode *generateNode(ProgramStateRef State,
358 ExplodedNode *Pred,
359 const ProgramPointTag *Tag = nullptr) {
360 const ProgramPoint &LocalLoc = (Tag ? Location.withTag(tag: Tag) : Location);
361 return NodeBuilder::generateNode(PP: LocalLoc, State, Pred);
362 }
363
364 ExplodedNode *generateSink(ProgramStateRef State, ExplodedNode *Pred,
365 const ProgramPointTag *Tag = nullptr) {
366 const ProgramPoint &LocalLoc = (Tag ? Location.withTag(tag: Tag) : Location);
367 ExplodedNode *N = NodeBuilder::generateSink(PP: LocalLoc, State, Pred);
368 if (N && N->isSink())
369 sinksGenerated.push_back(Elt: N);
370 return N;
371 }
372
373 const SmallVectorImpl<ExplodedNode*> &getSinks() const {
374 return sinksGenerated;
375 }
376};
377
378/// \class StmtNodeBuilder
379/// This builder class is useful for generating nodes that resulted from
380/// visiting a statement. The main difference from its parent NodeBuilder is
381/// that it creates a statement specific ProgramPoint.
382class StmtNodeBuilder: public NodeBuilder {
383 NodeBuilder *EnclosingBldr;
384
385public:
386 /// Constructs a StmtNodeBuilder. If the builder is going to process
387 /// nodes currently owned by another builder(with larger scope), use
388 /// Enclosing builder to transfer ownership.
389 StmtNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
390 const NodeBuilderContext &Ctx,
391 NodeBuilder *Enclosing = nullptr)
392 : NodeBuilder(SrcNode, DstSet, Ctx), EnclosingBldr(Enclosing) {
393 if (EnclosingBldr)
394 EnclosingBldr->takeNodes(N: SrcNode);
395 }
396
397 StmtNodeBuilder(ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
398 const NodeBuilderContext &Ctx,
399 NodeBuilder *Enclosing = nullptr)
400 : NodeBuilder(SrcSet, DstSet, Ctx), EnclosingBldr(Enclosing) {
401 if (EnclosingBldr)
402 for (const auto I : SrcSet)
403 EnclosingBldr->takeNodes(N: I);
404 }
405
406 ~StmtNodeBuilder() override;
407
408 using NodeBuilder::generateNode;
409 using NodeBuilder::generateSink;
410
411 ExplodedNode *generateNode(const Stmt *S,
412 ExplodedNode *Pred,
413 ProgramStateRef St,
414 const ProgramPointTag *tag = nullptr,
415 ProgramPoint::Kind K = ProgramPoint::PostStmtKind){
416 const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K,
417 LC: Pred->getLocationContext(), tag);
418 return NodeBuilder::generateNode(PP: L, State: St, Pred);
419 }
420
421 ExplodedNode *generateSink(const Stmt *S,
422 ExplodedNode *Pred,
423 ProgramStateRef St,
424 const ProgramPointTag *tag = nullptr,
425 ProgramPoint::Kind K = ProgramPoint::PostStmtKind){
426 const ProgramPoint &L = ProgramPoint::getProgramPoint(S, K,
427 LC: Pred->getLocationContext(), tag);
428 return NodeBuilder::generateSink(PP: L, State: St, Pred);
429 }
430};
431
432/// BranchNodeBuilder is responsible for constructing the nodes
433/// corresponding to the two branches of the if statement - true and false.
434class BranchNodeBuilder: public NodeBuilder {
435 const CFGBlock *DstT;
436 const CFGBlock *DstF;
437
438 bool InFeasibleTrue;
439 bool InFeasibleFalse;
440
441 void anchor() override;
442
443public:
444 BranchNodeBuilder(ExplodedNode *SrcNode, ExplodedNodeSet &DstSet,
445 const NodeBuilderContext &C,
446 const CFGBlock *dstT, const CFGBlock *dstF)
447 : NodeBuilder(SrcNode, DstSet, C), DstT(dstT), DstF(dstF),
448 InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) {
449 // The branch node builder does not generate autotransitions.
450 // If there are no successors it means that both branches are infeasible.
451 takeNodes(N: SrcNode);
452 }
453
454 BranchNodeBuilder(const ExplodedNodeSet &SrcSet, ExplodedNodeSet &DstSet,
455 const NodeBuilderContext &C,
456 const CFGBlock *dstT, const CFGBlock *dstF)
457 : NodeBuilder(SrcSet, DstSet, C), DstT(dstT), DstF(dstF),
458 InFeasibleTrue(!DstT), InFeasibleFalse(!DstF) {
459 takeNodes(S: SrcSet);
460 }
461
462 ExplodedNode *generateNode(ProgramStateRef State, bool branch,
463 ExplodedNode *Pred);
464
465 const CFGBlock *getTargetBlock(bool branch) const {
466 return branch ? DstT : DstF;
467 }
468
469 void markInfeasible(bool branch) {
470 if (branch)
471 InFeasibleTrue = true;
472 else
473 InFeasibleFalse = true;
474 }
475
476 bool isFeasible(bool branch) {
477 return branch ? !InFeasibleTrue : !InFeasibleFalse;
478 }
479};
480
481class IndirectGotoNodeBuilder {
482 CoreEngine& Eng;
483 const CFGBlock *Src;
484 const CFGBlock &DispatchBlock;
485 const Expr *E;
486 ExplodedNode *Pred;
487
488public:
489 IndirectGotoNodeBuilder(ExplodedNode *pred, const CFGBlock *src,
490 const Expr *e, const CFGBlock *dispatch, CoreEngine* eng)
491 : Eng(*eng), Src(src), DispatchBlock(*dispatch), E(e), Pred(pred) {}
492
493 class iterator {
494 friend class IndirectGotoNodeBuilder;
495
496 CFGBlock::const_succ_iterator I;
497
498 iterator(CFGBlock::const_succ_iterator i) : I(i) {}
499
500 public:
501 // This isn't really a conventional iterator.
502 // We just implement the deref as a no-op for now to make range-based for
503 // loops work.
504 const iterator &operator*() const { return *this; }
505
506 iterator &operator++() { ++I; return *this; }
507 bool operator!=(const iterator &X) const { return I != X.I; }
508
509 const LabelDecl *getLabel() const {
510 return cast<LabelStmt>(Val: (*I)->getLabel())->getDecl();
511 }
512
513 const CFGBlock *getBlock() const {
514 return *I;
515 }
516 };
517
518 iterator begin() { return iterator(DispatchBlock.succ_begin()); }
519 iterator end() { return iterator(DispatchBlock.succ_end()); }
520
521 ExplodedNode *generateNode(const iterator &I,
522 ProgramStateRef State,
523 bool isSink = false);
524
525 const Expr *getTarget() const { return E; }
526
527 ProgramStateRef getState() const { return Pred->State; }
528
529 const LocationContext *getLocationContext() const {
530 return Pred->getLocationContext();
531 }
532};
533
534class SwitchNodeBuilder {
535 CoreEngine& Eng;
536 const CFGBlock *Src;
537 const Expr *Condition;
538 ExplodedNode *Pred;
539
540public:
541 SwitchNodeBuilder(ExplodedNode *pred, const CFGBlock *src,
542 const Expr *condition, CoreEngine* eng)
543 : Eng(*eng), Src(src), Condition(condition), Pred(pred) {}
544
545 class iterator {
546 friend class SwitchNodeBuilder;
547
548 CFGBlock::const_succ_reverse_iterator I;
549
550 iterator(CFGBlock::const_succ_reverse_iterator i) : I(i) {}
551
552 public:
553 iterator &operator++() { ++I; return *this; }
554 bool operator!=(const iterator &X) const { return I != X.I; }
555 bool operator==(const iterator &X) const { return I == X.I; }
556
557 const CaseStmt *getCase() const {
558 return cast<CaseStmt>(Val: (*I)->getLabel());
559 }
560
561 const CFGBlock *getBlock() const {
562 return *I;
563 }
564 };
565
566 iterator begin() { return iterator(Src->succ_rbegin()+1); }
567 iterator end() { return iterator(Src->succ_rend()); }
568
569 const SwitchStmt *getSwitch() const {
570 return cast<SwitchStmt>(Val: Src->getTerminator());
571 }
572
573 ExplodedNode *generateCaseStmtNode(const iterator &I,
574 ProgramStateRef State);
575
576 ExplodedNode *generateDefaultCaseNode(ProgramStateRef State,
577 bool isSink = false);
578
579 const Expr *getCondition() const { return Condition; }
580
581 ProgramStateRef getState() const { return Pred->State; }
582
583 const LocationContext *getLocationContext() const {
584 return Pred->getLocationContext();
585 }
586};
587
588} // namespace ento
589
590} // namespace clang
591
592#endif // LLVM_CLANG_STATICANALYZER_CORE_PATHSENSITIVE_COREENGINE_H
593

source code of clang/include/clang/StaticAnalyzer/Core/PathSensitive/CoreEngine.h