| 1 | //===-- DataflowAnalysisContext.cpp -----------------------------*- 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 DataflowAnalysisContext class that owns objects that |
| 10 | // encompass the state of a program and stores context that is used during |
| 11 | // dataflow analysis. |
| 12 | // |
| 13 | //===----------------------------------------------------------------------===// |
| 14 | |
| 15 | #include "clang/Analysis/FlowSensitive/DataflowAnalysisContext.h" |
| 16 | #include "clang/Analysis/FlowSensitive/ASTOps.h" |
| 17 | #include "clang/Analysis/FlowSensitive/Formula.h" |
| 18 | #include "clang/Analysis/FlowSensitive/Logger.h" |
| 19 | #include "clang/Analysis/FlowSensitive/SimplifyConstraints.h" |
| 20 | #include "clang/Analysis/FlowSensitive/Value.h" |
| 21 | #include "clang/Basic/LLVM.h" |
| 22 | #include "llvm/ADT/DenseSet.h" |
| 23 | #include "llvm/ADT/SetOperations.h" |
| 24 | #include "llvm/ADT/SetVector.h" |
| 25 | #include "llvm/Support/CommandLine.h" |
| 26 | #include "llvm/Support/Debug.h" |
| 27 | #include "llvm/Support/FileSystem.h" |
| 28 | #include "llvm/Support/Path.h" |
| 29 | #include "llvm/Support/raw_ostream.h" |
| 30 | #include <cassert> |
| 31 | #include <memory> |
| 32 | #include <stack> |
| 33 | #include <string> |
| 34 | #include <utility> |
| 35 | #include <vector> |
| 36 | |
| 37 | static llvm::cl::opt<std::string> DataflowLog( |
| 38 | "dataflow-log" , llvm::cl::Hidden, llvm::cl::ValueOptional, |
| 39 | llvm::cl::desc("Emit log of dataflow analysis. With no arg, writes textual " |
| 40 | "log to stderr. With an arg, writes HTML logs under the " |
| 41 | "specified directory (one per analyzed function)." )); |
| 42 | |
| 43 | namespace clang { |
| 44 | namespace dataflow { |
| 45 | |
| 46 | FieldSet DataflowAnalysisContext::getModeledFields(QualType Type) { |
| 47 | // During context-sensitive analysis, a struct may be allocated in one |
| 48 | // function, but its field accessed in a function lower in the stack than |
| 49 | // the allocation. Since we only collect fields used in the function where |
| 50 | // the allocation occurs, we can't apply that filter when performing |
| 51 | // context-sensitive analysis. But, this only applies to storage locations, |
| 52 | // since field access it not allowed to fail. In contrast, field *values* |
| 53 | // don't need this allowance, since the API allows for uninitialized fields. |
| 54 | if (Opts.ContextSensitiveOpts) |
| 55 | return getObjectFields(Type); |
| 56 | |
| 57 | return llvm::set_intersection(S1: getObjectFields(Type), S2: ModeledFields); |
| 58 | } |
| 59 | |
| 60 | void DataflowAnalysisContext::addModeledFields(const FieldSet &Fields) { |
| 61 | ModeledFields.set_union(Fields); |
| 62 | } |
| 63 | |
| 64 | StorageLocation &DataflowAnalysisContext::createStorageLocation(QualType Type) { |
| 65 | if (!Type.isNull() && Type->isRecordType()) { |
| 66 | llvm::DenseMap<const ValueDecl *, StorageLocation *> FieldLocs; |
| 67 | for (const FieldDecl *Field : getModeledFields(Type)) |
| 68 | if (Field->getType()->isReferenceType()) |
| 69 | FieldLocs.insert(KV: {Field, nullptr}); |
| 70 | else |
| 71 | FieldLocs.insert(KV: {Field, &createStorageLocation( |
| 72 | Type: Field->getType().getNonReferenceType())}); |
| 73 | |
| 74 | RecordStorageLocation::SyntheticFieldMap SyntheticFields; |
| 75 | for (const auto &Entry : getSyntheticFields(Type)) |
| 76 | SyntheticFields.insert( |
| 77 | KV: {Entry.getKey(), |
| 78 | &createStorageLocation(Type: Entry.getValue().getNonReferenceType())}); |
| 79 | |
| 80 | return createRecordStorageLocation(Type, FieldLocs: std::move(FieldLocs), |
| 81 | SyntheticFields: std::move(SyntheticFields)); |
| 82 | } |
| 83 | return arena().create<ScalarStorageLocation>(args&: Type); |
| 84 | } |
| 85 | |
| 86 | // Returns the keys for a given `StringMap`. |
| 87 | // Can't use `StringSet` as the return type as it doesn't support `operator==`. |
| 88 | template <typename T> |
| 89 | static llvm::DenseSet<llvm::StringRef> getKeys(const llvm::StringMap<T> &Map) { |
| 90 | return llvm::DenseSet<llvm::StringRef>(llvm::from_range, Map.keys()); |
| 91 | } |
| 92 | |
| 93 | RecordStorageLocation &DataflowAnalysisContext::createRecordStorageLocation( |
| 94 | QualType Type, RecordStorageLocation::FieldToLoc FieldLocs, |
| 95 | RecordStorageLocation::SyntheticFieldMap SyntheticFields) { |
| 96 | assert(Type->isRecordType()); |
| 97 | assert(containsSameFields(getModeledFields(Type), FieldLocs)); |
| 98 | assert(getKeys(getSyntheticFields(Type)) == getKeys(SyntheticFields)); |
| 99 | |
| 100 | RecordStorageLocationCreated = true; |
| 101 | return arena().create<RecordStorageLocation>(args&: Type, args: std::move(FieldLocs), |
| 102 | args: std::move(SyntheticFields)); |
| 103 | } |
| 104 | |
| 105 | StorageLocation & |
| 106 | DataflowAnalysisContext::getStableStorageLocation(const ValueDecl &D) { |
| 107 | if (auto *Loc = DeclToLoc.lookup(Val: &D)) |
| 108 | return *Loc; |
| 109 | auto &Loc = createStorageLocation(Type: D.getType().getNonReferenceType()); |
| 110 | DeclToLoc[&D] = &Loc; |
| 111 | return Loc; |
| 112 | } |
| 113 | |
| 114 | StorageLocation & |
| 115 | DataflowAnalysisContext::getStableStorageLocation(const Expr &E) { |
| 116 | const Expr &CanonE = ignoreCFGOmittedNodes(E); |
| 117 | |
| 118 | if (auto *Loc = ExprToLoc.lookup(Val: &CanonE)) |
| 119 | return *Loc; |
| 120 | auto &Loc = createStorageLocation(Type: CanonE.getType()); |
| 121 | ExprToLoc[&CanonE] = &Loc; |
| 122 | return Loc; |
| 123 | } |
| 124 | |
| 125 | PointerValue & |
| 126 | DataflowAnalysisContext::getOrCreateNullPointerValue(QualType PointeeType) { |
| 127 | auto CanonicalPointeeType = |
| 128 | PointeeType.isNull() ? PointeeType : PointeeType.getCanonicalType(); |
| 129 | auto Res = NullPointerVals.try_emplace(Key: CanonicalPointeeType, Args: nullptr); |
| 130 | if (Res.second) { |
| 131 | auto &PointeeLoc = createStorageLocation(Type: CanonicalPointeeType); |
| 132 | Res.first->second = &arena().create<PointerValue>(args&: PointeeLoc); |
| 133 | } |
| 134 | return *Res.first->second; |
| 135 | } |
| 136 | |
| 137 | void DataflowAnalysisContext::addInvariant(const Formula &Constraint) { |
| 138 | if (Invariant == nullptr) |
| 139 | Invariant = &Constraint; |
| 140 | else |
| 141 | Invariant = &arena().makeAnd(LHS: *Invariant, RHS: Constraint); |
| 142 | } |
| 143 | |
| 144 | void DataflowAnalysisContext::addFlowConditionConstraint( |
| 145 | Atom Token, const Formula &Constraint) { |
| 146 | auto Res = FlowConditionConstraints.try_emplace(Key: Token, Args: &Constraint); |
| 147 | if (!Res.second) { |
| 148 | Res.first->second = |
| 149 | &arena().makeAnd(LHS: *Res.first->second, RHS: Constraint); |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | Atom DataflowAnalysisContext::forkFlowCondition(Atom Token) { |
| 154 | Atom ForkToken = arena().makeFlowConditionToken(); |
| 155 | FlowConditionDeps[ForkToken].insert(V: Token); |
| 156 | addFlowConditionConstraint(Token: ForkToken, Constraint: arena().makeAtomRef(A: Token)); |
| 157 | return ForkToken; |
| 158 | } |
| 159 | |
| 160 | Atom |
| 161 | DataflowAnalysisContext::joinFlowConditions(Atom FirstToken, |
| 162 | Atom SecondToken) { |
| 163 | Atom Token = arena().makeFlowConditionToken(); |
| 164 | auto &TokenDeps = FlowConditionDeps[Token]; |
| 165 | TokenDeps.insert(V: FirstToken); |
| 166 | TokenDeps.insert(V: SecondToken); |
| 167 | addFlowConditionConstraint(Token, |
| 168 | Constraint: arena().makeOr(LHS: arena().makeAtomRef(A: FirstToken), |
| 169 | RHS: arena().makeAtomRef(A: SecondToken))); |
| 170 | return Token; |
| 171 | } |
| 172 | |
| 173 | Solver::Result DataflowAnalysisContext::querySolver( |
| 174 | llvm::SetVector<const Formula *> Constraints) { |
| 175 | return S.solve(Vals: Constraints.getArrayRef()); |
| 176 | } |
| 177 | |
| 178 | bool DataflowAnalysisContext::flowConditionImplies(Atom Token, |
| 179 | const Formula &F) { |
| 180 | if (F.isLiteral(b: true)) |
| 181 | return true; |
| 182 | |
| 183 | // Returns true if and only if truth assignment of the flow condition implies |
| 184 | // that `F` is also true. We prove whether or not this property holds by |
| 185 | // reducing the problem to satisfiability checking. In other words, we attempt |
| 186 | // to show that assuming `F` is false makes the constraints induced by the |
| 187 | // flow condition unsatisfiable. |
| 188 | llvm::SetVector<const Formula *> Constraints; |
| 189 | Constraints.insert(X: &arena().makeAtomRef(A: Token)); |
| 190 | Constraints.insert(X: &arena().makeNot(Val: F)); |
| 191 | addTransitiveFlowConditionConstraints(Token, Out&: Constraints); |
| 192 | return isUnsatisfiable(Constraints: std::move(Constraints)); |
| 193 | } |
| 194 | |
| 195 | bool DataflowAnalysisContext::flowConditionAllows(Atom Token, |
| 196 | const Formula &F) { |
| 197 | if (F.isLiteral(b: false)) |
| 198 | return false; |
| 199 | |
| 200 | llvm::SetVector<const Formula *> Constraints; |
| 201 | Constraints.insert(X: &arena().makeAtomRef(A: Token)); |
| 202 | Constraints.insert(X: &F); |
| 203 | addTransitiveFlowConditionConstraints(Token, Out&: Constraints); |
| 204 | return isSatisfiable(Constraints: std::move(Constraints)); |
| 205 | } |
| 206 | |
| 207 | bool DataflowAnalysisContext::equivalentFormulas(const Formula &Val1, |
| 208 | const Formula &Val2) { |
| 209 | llvm::SetVector<const Formula *> Constraints; |
| 210 | Constraints.insert(X: &arena().makeNot(Val: arena().makeEquals(LHS: Val1, RHS: Val2))); |
| 211 | return isUnsatisfiable(Constraints: std::move(Constraints)); |
| 212 | } |
| 213 | |
| 214 | llvm::DenseSet<Atom> DataflowAnalysisContext::collectDependencies( |
| 215 | llvm::DenseSet<Atom> Tokens) const { |
| 216 | // Use a worklist algorithm, with `Remaining` holding the worklist and |
| 217 | // `Tokens` tracking which atoms have already been added to the worklist. |
| 218 | std::vector<Atom> Remaining(Tokens.begin(), Tokens.end()); |
| 219 | while (!Remaining.empty()) { |
| 220 | Atom CurrentToken = Remaining.back(); |
| 221 | Remaining.pop_back(); |
| 222 | if (auto DepsIt = FlowConditionDeps.find(Val: CurrentToken); |
| 223 | DepsIt != FlowConditionDeps.end()) |
| 224 | for (Atom A : DepsIt->second) |
| 225 | if (Tokens.insert(V: A).second) |
| 226 | Remaining.push_back(x: A); |
| 227 | } |
| 228 | |
| 229 | return Tokens; |
| 230 | } |
| 231 | |
| 232 | void DataflowAnalysisContext::addTransitiveFlowConditionConstraints( |
| 233 | Atom Token, llvm::SetVector<const Formula *> &Constraints) { |
| 234 | llvm::DenseSet<Atom> AddedTokens; |
| 235 | std::vector<Atom> Remaining = {Token}; |
| 236 | |
| 237 | if (Invariant) |
| 238 | Constraints.insert(X: Invariant); |
| 239 | // Define all the flow conditions that might be referenced in constraints. |
| 240 | while (!Remaining.empty()) { |
| 241 | auto Token = Remaining.back(); |
| 242 | Remaining.pop_back(); |
| 243 | if (!AddedTokens.insert(V: Token).second) |
| 244 | continue; |
| 245 | |
| 246 | auto ConstraintsIt = FlowConditionConstraints.find(Val: Token); |
| 247 | if (ConstraintsIt == FlowConditionConstraints.end()) { |
| 248 | // The flow condition is unconstrained. Just add the atom directly, which |
| 249 | // is equivalent to asserting it is true. |
| 250 | Constraints.insert(X: &arena().makeAtomRef(A: Token)); |
| 251 | } else { |
| 252 | // Bind flow condition token via `iff` to its set of constraints: |
| 253 | // FC <=> (C1 ^ C2 ^ ...), where Ci are constraints |
| 254 | Constraints.insert(X: &arena().makeEquals(LHS: arena().makeAtomRef(A: Token), |
| 255 | RHS: *ConstraintsIt->second)); |
| 256 | } |
| 257 | |
| 258 | if (auto DepsIt = FlowConditionDeps.find(Val: Token); |
| 259 | DepsIt != FlowConditionDeps.end()) |
| 260 | for (Atom A : DepsIt->second) |
| 261 | Remaining.push_back(x: A); |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | static void getReferencedAtoms(const Formula &F, |
| 266 | llvm::DenseSet<dataflow::Atom> &Refs) { |
| 267 | // Avoid recursion to avoid stack overflows from very large formulas. |
| 268 | // The shape of the tree structure for very large formulas is such that there |
| 269 | // are at most 2 children from any node, but there may be many generations. |
| 270 | std::stack<const Formula *> WorkList; |
| 271 | WorkList.push(x: &F); |
| 272 | |
| 273 | while (!WorkList.empty()) { |
| 274 | const Formula *Current = WorkList.top(); |
| 275 | WorkList.pop(); |
| 276 | switch (Current->kind()) { |
| 277 | case Formula::AtomRef: |
| 278 | Refs.insert(V: Current->getAtom()); |
| 279 | break; |
| 280 | case Formula::Literal: |
| 281 | break; |
| 282 | case Formula::Not: |
| 283 | WorkList.push(x: Current->operands()[0]); |
| 284 | break; |
| 285 | case Formula::And: |
| 286 | case Formula::Or: |
| 287 | case Formula::Implies: |
| 288 | case Formula::Equal: |
| 289 | ArrayRef<const Formula *> Operands = Current->operands(); |
| 290 | WorkList.push(x: Operands[0]); |
| 291 | WorkList.push(x: Operands[1]); |
| 292 | break; |
| 293 | } |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | SimpleLogicalContext DataflowAnalysisContext::exportLogicalContext( |
| 298 | llvm::DenseSet<dataflow::Atom> TargetTokens) const { |
| 299 | SimpleLogicalContext LC; |
| 300 | |
| 301 | // Copy `Invariant` even if it is null, to initialize the field. |
| 302 | LC.Invariant = Invariant; |
| 303 | if (Invariant != nullptr) |
| 304 | getReferencedAtoms(F: *Invariant, Refs&: TargetTokens); |
| 305 | |
| 306 | llvm::DenseSet<dataflow::Atom> Dependencies = |
| 307 | collectDependencies(Tokens: std::move(TargetTokens)); |
| 308 | |
| 309 | for (dataflow::Atom Token : Dependencies) { |
| 310 | // Only process the token if it is constrained. Unconstrained tokens don't |
| 311 | // have dependencies. |
| 312 | const Formula *Constraints = FlowConditionConstraints.lookup(Val: Token); |
| 313 | if (Constraints == nullptr) |
| 314 | continue; |
| 315 | LC.TokenDefs[Token] = Constraints; |
| 316 | |
| 317 | if (auto DepsIt = FlowConditionDeps.find(Val: Token); |
| 318 | DepsIt != FlowConditionDeps.end()) |
| 319 | LC.TokenDeps[Token] = DepsIt->second; |
| 320 | } |
| 321 | |
| 322 | return LC; |
| 323 | } |
| 324 | |
| 325 | void DataflowAnalysisContext::initLogicalContext(SimpleLogicalContext LC) { |
| 326 | Invariant = LC.Invariant; |
| 327 | FlowConditionConstraints = std::move(LC.TokenDefs); |
| 328 | // TODO: The dependencies in `LC.TokenDeps` can be reconstructed from |
| 329 | // `LC.TokenDefs`. Give the caller the option to reconstruct, rather than |
| 330 | // providing them directly, to save caller space (memory/disk). |
| 331 | FlowConditionDeps = std::move(LC.TokenDeps); |
| 332 | } |
| 333 | |
| 334 | static void printAtomList(const llvm::SmallVector<Atom> &Atoms, |
| 335 | llvm::raw_ostream &OS) { |
| 336 | OS << "(" ; |
| 337 | for (size_t i = 0; i < Atoms.size(); ++i) { |
| 338 | OS << Atoms[i]; |
| 339 | if (i + 1 < Atoms.size()) |
| 340 | OS << ", " ; |
| 341 | } |
| 342 | OS << ")\n" ; |
| 343 | } |
| 344 | |
| 345 | void DataflowAnalysisContext::dumpFlowCondition(Atom Token, |
| 346 | llvm::raw_ostream &OS) { |
| 347 | llvm::SetVector<const Formula *> Constraints; |
| 348 | Constraints.insert(X: &arena().makeAtomRef(A: Token)); |
| 349 | addTransitiveFlowConditionConstraints(Token, Constraints); |
| 350 | |
| 351 | OS << "Flow condition token: " << Token << "\n" ; |
| 352 | SimplifyConstraintsInfo Info; |
| 353 | llvm::SetVector<const Formula *> OriginalConstraints = Constraints; |
| 354 | simplifyConstraints(Constraints, arena&: arena(), Info: &Info); |
| 355 | if (!Constraints.empty()) { |
| 356 | OS << "Constraints:\n" ; |
| 357 | for (const auto *Constraint : Constraints) { |
| 358 | Constraint->print(OS); |
| 359 | OS << "\n" ; |
| 360 | } |
| 361 | } |
| 362 | if (!Info.TrueAtoms.empty()) { |
| 363 | OS << "True atoms: " ; |
| 364 | printAtomList(Atoms: Info.TrueAtoms, OS); |
| 365 | } |
| 366 | if (!Info.FalseAtoms.empty()) { |
| 367 | OS << "False atoms: " ; |
| 368 | printAtomList(Atoms: Info.FalseAtoms, OS); |
| 369 | } |
| 370 | if (!Info.EquivalentAtoms.empty()) { |
| 371 | OS << "Equivalent atoms:\n" ; |
| 372 | for (const llvm::SmallVector<Atom> &Class : Info.EquivalentAtoms) |
| 373 | printAtomList(Atoms: Class, OS); |
| 374 | } |
| 375 | |
| 376 | OS << "\nFlow condition constraints before simplification:\n" ; |
| 377 | for (const auto *Constraint : OriginalConstraints) { |
| 378 | Constraint->print(OS); |
| 379 | OS << "\n" ; |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | const AdornedCFG * |
| 384 | DataflowAnalysisContext::getAdornedCFG(const FunctionDecl *F) { |
| 385 | // Canonicalize the key: |
| 386 | F = F->getDefinition(); |
| 387 | if (F == nullptr) |
| 388 | return nullptr; |
| 389 | auto It = FunctionContexts.find(Val: F); |
| 390 | if (It != FunctionContexts.end()) |
| 391 | return &It->second; |
| 392 | |
| 393 | if (F->doesThisDeclarationHaveABody()) { |
| 394 | auto ACFG = AdornedCFG::build(Func: *F); |
| 395 | // FIXME: Handle errors. |
| 396 | assert(ACFG); |
| 397 | auto Result = FunctionContexts.insert(KV: {F, std::move(*ACFG)}); |
| 398 | return &Result.first->second; |
| 399 | } |
| 400 | |
| 401 | return nullptr; |
| 402 | } |
| 403 | |
| 404 | static std::unique_ptr<Logger> makeLoggerFromCommandLine() { |
| 405 | if (DataflowLog.empty()) |
| 406 | return Logger::textual(llvm::errs()); |
| 407 | |
| 408 | llvm::StringRef Dir = DataflowLog; |
| 409 | if (auto EC = llvm::sys::fs::create_directories(path: Dir)) |
| 410 | llvm::errs() << "Failed to create log dir: " << EC.message() << "\n" ; |
| 411 | // All analysis runs within a process will log to the same directory. |
| 412 | // Share a counter so they don't all overwrite each other's 0.html. |
| 413 | // (Don't share a logger, it's not threadsafe). |
| 414 | static std::atomic<unsigned> Counter = {0}; |
| 415 | auto StreamFactory = |
| 416 | [Dir(Dir.str())]() mutable -> std::unique_ptr<llvm::raw_ostream> { |
| 417 | llvm::SmallString<256> File(Dir); |
| 418 | llvm::sys::path::append(path&: File, |
| 419 | a: std::to_string(val: Counter.fetch_add(i: 1)) + ".html" ); |
| 420 | std::error_code EC; |
| 421 | auto OS = std::make_unique<llvm::raw_fd_ostream>(args&: File, args&: EC); |
| 422 | if (EC) { |
| 423 | llvm::errs() << "Failed to create log " << File << ": " << EC.message() |
| 424 | << "\n" ; |
| 425 | return std::make_unique<llvm::raw_null_ostream>(); |
| 426 | } |
| 427 | return OS; |
| 428 | }; |
| 429 | return Logger::html(std::move(StreamFactory)); |
| 430 | } |
| 431 | |
| 432 | DataflowAnalysisContext::DataflowAnalysisContext( |
| 433 | Solver &S, std::unique_ptr<Solver> &&OwnedSolver, Options Opts) |
| 434 | : S(S), OwnedSolver(std::move(OwnedSolver)), A(std::make_unique<Arena>()), |
| 435 | Opts(Opts) { |
| 436 | // If the -dataflow-log command-line flag was set, synthesize a logger. |
| 437 | // This is ugly but provides a uniform method for ad-hoc debugging dataflow- |
| 438 | // based tools. |
| 439 | if (Opts.Log == nullptr) { |
| 440 | if (DataflowLog.getNumOccurrences() > 0) { |
| 441 | LogOwner = makeLoggerFromCommandLine(); |
| 442 | this->Opts.Log = LogOwner.get(); |
| 443 | // FIXME: if the flag is given a value, write an HTML log to a file. |
| 444 | } else { |
| 445 | this->Opts.Log = &Logger::null(); |
| 446 | } |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | DataflowAnalysisContext::~DataflowAnalysisContext() = default; |
| 451 | |
| 452 | } // namespace dataflow |
| 453 | } // namespace clang |
| 454 | |