1 //== SymbolManager.h - Management of Symbolic Values ------------*- 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 SymbolManager, a class that manages symbolic values
11 // created for use by ExprEngine and related classes.
12 //
13 //===----------------------------------------------------------------------===//
14
15 #include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h"
16 #include "clang/Analysis/Analyses/LiveVariables.h"
17 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
18 #include "clang/StaticAnalyzer/Core/PathSensitive/Store.h"
19 #include "llvm/Support/raw_ostream.h"
20
21 using namespace clang;
22 using namespace ento;
23
anchor()24 void SymExpr::anchor() { }
25
dump() const26 void SymExpr::dump() const {
27 dumpToStream(llvm::errs());
28 }
29
dumpToStream(raw_ostream & os) const30 void SymIntExpr::dumpToStream(raw_ostream &os) const {
31 os << '(';
32 getLHS()->dumpToStream(os);
33 os << ") "
34 << BinaryOperator::getOpcodeStr(getOpcode()) << ' '
35 << getRHS().getZExtValue();
36 if (getRHS().isUnsigned())
37 os << 'U';
38 }
39
dumpToStream(raw_ostream & os) const40 void IntSymExpr::dumpToStream(raw_ostream &os) const {
41 os << getLHS().getZExtValue();
42 if (getLHS().isUnsigned())
43 os << 'U';
44 os << ' '
45 << BinaryOperator::getOpcodeStr(getOpcode())
46 << " (";
47 getRHS()->dumpToStream(os);
48 os << ')';
49 }
50
dumpToStream(raw_ostream & os) const51 void SymSymExpr::dumpToStream(raw_ostream &os) const {
52 os << '(';
53 getLHS()->dumpToStream(os);
54 os << ") "
55 << BinaryOperator::getOpcodeStr(getOpcode())
56 << " (";
57 getRHS()->dumpToStream(os);
58 os << ')';
59 }
60
dumpToStream(raw_ostream & os) const61 void SymbolCast::dumpToStream(raw_ostream &os) const {
62 os << '(' << ToTy.getAsString() << ") (";
63 Operand->dumpToStream(os);
64 os << ')';
65 }
66
dumpToStream(raw_ostream & os) const67 void SymbolConjured::dumpToStream(raw_ostream &os) const {
68 os << "conj_$" << getSymbolID() << '{' << T.getAsString() << '}';
69 }
70
dumpToStream(raw_ostream & os) const71 void SymbolDerived::dumpToStream(raw_ostream &os) const {
72 os << "derived_$" << getSymbolID() << '{'
73 << getParentSymbol() << ',' << getRegion() << '}';
74 }
75
dumpToStream(raw_ostream & os) const76 void SymbolExtent::dumpToStream(raw_ostream &os) const {
77 os << "extent_$" << getSymbolID() << '{' << getRegion() << '}';
78 }
79
dumpToStream(raw_ostream & os) const80 void SymbolMetadata::dumpToStream(raw_ostream &os) const {
81 os << "meta_$" << getSymbolID() << '{'
82 << getRegion() << ',' << T.getAsString() << '}';
83 }
84
anchor()85 void SymbolData::anchor() { }
86
dumpToStream(raw_ostream & os) const87 void SymbolRegionValue::dumpToStream(raw_ostream &os) const {
88 os << "reg_$" << getSymbolID() << "<" << R << ">";
89 }
90
operator ==(const symbol_iterator & X) const91 bool SymExpr::symbol_iterator::operator==(const symbol_iterator &X) const {
92 return itr == X.itr;
93 }
94
operator !=(const symbol_iterator & X) const95 bool SymExpr::symbol_iterator::operator!=(const symbol_iterator &X) const {
96 return itr != X.itr;
97 }
98
symbol_iterator(const SymExpr * SE)99 SymExpr::symbol_iterator::symbol_iterator(const SymExpr *SE) {
100 itr.push_back(SE);
101 }
102
operator ++()103 SymExpr::symbol_iterator &SymExpr::symbol_iterator::operator++() {
104 assert(!itr.empty() && "attempting to iterate on an 'end' iterator");
105 expand();
106 return *this;
107 }
108
operator *()109 SymbolRef SymExpr::symbol_iterator::operator*() {
110 assert(!itr.empty() && "attempting to dereference an 'end' iterator");
111 return itr.back();
112 }
113
expand()114 void SymExpr::symbol_iterator::expand() {
115 const SymExpr *SE = itr.pop_back_val();
116
117 switch (SE->getKind()) {
118 case SymExpr::RegionValueKind:
119 case SymExpr::ConjuredKind:
120 case SymExpr::DerivedKind:
121 case SymExpr::ExtentKind:
122 case SymExpr::MetadataKind:
123 return;
124 case SymExpr::CastSymbolKind:
125 itr.push_back(cast<SymbolCast>(SE)->getOperand());
126 return;
127 case SymExpr::SymIntKind:
128 itr.push_back(cast<SymIntExpr>(SE)->getLHS());
129 return;
130 case SymExpr::IntSymKind:
131 itr.push_back(cast<IntSymExpr>(SE)->getRHS());
132 return;
133 case SymExpr::SymSymKind: {
134 const SymSymExpr *x = cast<SymSymExpr>(SE);
135 itr.push_back(x->getLHS());
136 itr.push_back(x->getRHS());
137 return;
138 }
139 }
140 llvm_unreachable("unhandled expansion case");
141 }
142
computeComplexity() const143 unsigned SymExpr::computeComplexity() const {
144 unsigned R = 0;
145 for (symbol_iterator I = symbol_begin(), E = symbol_end(); I != E; ++I)
146 R++;
147 return R;
148 }
149
150 const SymbolRegionValue*
getRegionValueSymbol(const TypedValueRegion * R)151 SymbolManager::getRegionValueSymbol(const TypedValueRegion* R) {
152 llvm::FoldingSetNodeID profile;
153 SymbolRegionValue::Profile(profile, R);
154 void *InsertPos;
155 SymExpr *SD = DataSet.FindNodeOrInsertPos(profile, InsertPos);
156 if (!SD) {
157 SD = (SymExpr*) BPAlloc.Allocate<SymbolRegionValue>();
158 new (SD) SymbolRegionValue(SymbolCounter, R);
159 DataSet.InsertNode(SD, InsertPos);
160 ++SymbolCounter;
161 }
162
163 return cast<SymbolRegionValue>(SD);
164 }
165
conjureSymbol(const Stmt * E,const LocationContext * LCtx,QualType T,unsigned Count,const void * SymbolTag)166 const SymbolConjured* SymbolManager::conjureSymbol(const Stmt *E,
167 const LocationContext *LCtx,
168 QualType T,
169 unsigned Count,
170 const void *SymbolTag) {
171 llvm::FoldingSetNodeID profile;
172 SymbolConjured::Profile(profile, E, T, Count, LCtx, SymbolTag);
173 void *InsertPos;
174 SymExpr *SD = DataSet.FindNodeOrInsertPos(profile, InsertPos);
175 if (!SD) {
176 SD = (SymExpr*) BPAlloc.Allocate<SymbolConjured>();
177 new (SD) SymbolConjured(SymbolCounter, E, LCtx, T, Count, SymbolTag);
178 DataSet.InsertNode(SD, InsertPos);
179 ++SymbolCounter;
180 }
181
182 return cast<SymbolConjured>(SD);
183 }
184
185 const SymbolDerived*
getDerivedSymbol(SymbolRef parentSymbol,const TypedValueRegion * R)186 SymbolManager::getDerivedSymbol(SymbolRef parentSymbol,
187 const TypedValueRegion *R) {
188
189 llvm::FoldingSetNodeID profile;
190 SymbolDerived::Profile(profile, parentSymbol, R);
191 void *InsertPos;
192 SymExpr *SD = DataSet.FindNodeOrInsertPos(profile, InsertPos);
193 if (!SD) {
194 SD = (SymExpr*) BPAlloc.Allocate<SymbolDerived>();
195 new (SD) SymbolDerived(SymbolCounter, parentSymbol, R);
196 DataSet.InsertNode(SD, InsertPos);
197 ++SymbolCounter;
198 }
199
200 return cast<SymbolDerived>(SD);
201 }
202
203 const SymbolExtent*
getExtentSymbol(const SubRegion * R)204 SymbolManager::getExtentSymbol(const SubRegion *R) {
205 llvm::FoldingSetNodeID profile;
206 SymbolExtent::Profile(profile, R);
207 void *InsertPos;
208 SymExpr *SD = DataSet.FindNodeOrInsertPos(profile, InsertPos);
209 if (!SD) {
210 SD = (SymExpr*) BPAlloc.Allocate<SymbolExtent>();
211 new (SD) SymbolExtent(SymbolCounter, R);
212 DataSet.InsertNode(SD, InsertPos);
213 ++SymbolCounter;
214 }
215
216 return cast<SymbolExtent>(SD);
217 }
218
219 const SymbolMetadata*
getMetadataSymbol(const MemRegion * R,const Stmt * S,QualType T,unsigned Count,const void * SymbolTag)220 SymbolManager::getMetadataSymbol(const MemRegion* R, const Stmt *S, QualType T,
221 unsigned Count, const void *SymbolTag) {
222
223 llvm::FoldingSetNodeID profile;
224 SymbolMetadata::Profile(profile, R, S, T, Count, SymbolTag);
225 void *InsertPos;
226 SymExpr *SD = DataSet.FindNodeOrInsertPos(profile, InsertPos);
227 if (!SD) {
228 SD = (SymExpr*) BPAlloc.Allocate<SymbolMetadata>();
229 new (SD) SymbolMetadata(SymbolCounter, R, S, T, Count, SymbolTag);
230 DataSet.InsertNode(SD, InsertPos);
231 ++SymbolCounter;
232 }
233
234 return cast<SymbolMetadata>(SD);
235 }
236
237 const SymbolCast*
getCastSymbol(const SymExpr * Op,QualType From,QualType To)238 SymbolManager::getCastSymbol(const SymExpr *Op,
239 QualType From, QualType To) {
240 llvm::FoldingSetNodeID ID;
241 SymbolCast::Profile(ID, Op, From, To);
242 void *InsertPos;
243 SymExpr *data = DataSet.FindNodeOrInsertPos(ID, InsertPos);
244 if (!data) {
245 data = (SymbolCast*) BPAlloc.Allocate<SymbolCast>();
246 new (data) SymbolCast(Op, From, To);
247 DataSet.InsertNode(data, InsertPos);
248 }
249
250 return cast<SymbolCast>(data);
251 }
252
getSymIntExpr(const SymExpr * lhs,BinaryOperator::Opcode op,const llvm::APSInt & v,QualType t)253 const SymIntExpr *SymbolManager::getSymIntExpr(const SymExpr *lhs,
254 BinaryOperator::Opcode op,
255 const llvm::APSInt& v,
256 QualType t) {
257 llvm::FoldingSetNodeID ID;
258 SymIntExpr::Profile(ID, lhs, op, v, t);
259 void *InsertPos;
260 SymExpr *data = DataSet.FindNodeOrInsertPos(ID, InsertPos);
261
262 if (!data) {
263 data = (SymIntExpr*) BPAlloc.Allocate<SymIntExpr>();
264 new (data) SymIntExpr(lhs, op, v, t);
265 DataSet.InsertNode(data, InsertPos);
266 }
267
268 return cast<SymIntExpr>(data);
269 }
270
getIntSymExpr(const llvm::APSInt & lhs,BinaryOperator::Opcode op,const SymExpr * rhs,QualType t)271 const IntSymExpr *SymbolManager::getIntSymExpr(const llvm::APSInt& lhs,
272 BinaryOperator::Opcode op,
273 const SymExpr *rhs,
274 QualType t) {
275 llvm::FoldingSetNodeID ID;
276 IntSymExpr::Profile(ID, lhs, op, rhs, t);
277 void *InsertPos;
278 SymExpr *data = DataSet.FindNodeOrInsertPos(ID, InsertPos);
279
280 if (!data) {
281 data = (IntSymExpr*) BPAlloc.Allocate<IntSymExpr>();
282 new (data) IntSymExpr(lhs, op, rhs, t);
283 DataSet.InsertNode(data, InsertPos);
284 }
285
286 return cast<IntSymExpr>(data);
287 }
288
getSymSymExpr(const SymExpr * lhs,BinaryOperator::Opcode op,const SymExpr * rhs,QualType t)289 const SymSymExpr *SymbolManager::getSymSymExpr(const SymExpr *lhs,
290 BinaryOperator::Opcode op,
291 const SymExpr *rhs,
292 QualType t) {
293 llvm::FoldingSetNodeID ID;
294 SymSymExpr::Profile(ID, lhs, op, rhs, t);
295 void *InsertPos;
296 SymExpr *data = DataSet.FindNodeOrInsertPos(ID, InsertPos);
297
298 if (!data) {
299 data = (SymSymExpr*) BPAlloc.Allocate<SymSymExpr>();
300 new (data) SymSymExpr(lhs, op, rhs, t);
301 DataSet.InsertNode(data, InsertPos);
302 }
303
304 return cast<SymSymExpr>(data);
305 }
306
getType() const307 QualType SymbolConjured::getType() const {
308 return T;
309 }
310
getType() const311 QualType SymbolDerived::getType() const {
312 return R->getValueType();
313 }
314
getType() const315 QualType SymbolExtent::getType() const {
316 ASTContext &Ctx = R->getMemRegionManager()->getContext();
317 return Ctx.getSizeType();
318 }
319
getType() const320 QualType SymbolMetadata::getType() const {
321 return T;
322 }
323
getType() const324 QualType SymbolRegionValue::getType() const {
325 return R->getValueType();
326 }
327
~SymbolManager()328 SymbolManager::~SymbolManager() {
329 llvm::DeleteContainerSeconds(SymbolDependencies);
330 }
331
canSymbolicate(QualType T)332 bool SymbolManager::canSymbolicate(QualType T) {
333 T = T.getCanonicalType();
334
335 if (Loc::isLocType(T))
336 return true;
337
338 if (T->isIntegralOrEnumerationType())
339 return true;
340
341 if (T->isRecordType() && !T->isUnionType())
342 return true;
343
344 return false;
345 }
346
addSymbolDependency(const SymbolRef Primary,const SymbolRef Dependent)347 void SymbolManager::addSymbolDependency(const SymbolRef Primary,
348 const SymbolRef Dependent) {
349 SymbolDependTy::iterator I = SymbolDependencies.find(Primary);
350 SymbolRefSmallVectorTy *dependencies = nullptr;
351 if (I == SymbolDependencies.end()) {
352 dependencies = new SymbolRefSmallVectorTy();
353 SymbolDependencies[Primary] = dependencies;
354 } else {
355 dependencies = I->second;
356 }
357 dependencies->push_back(Dependent);
358 }
359
getDependentSymbols(const SymbolRef Primary)360 const SymbolRefSmallVectorTy *SymbolManager::getDependentSymbols(
361 const SymbolRef Primary) {
362 SymbolDependTy::const_iterator I = SymbolDependencies.find(Primary);
363 if (I == SymbolDependencies.end())
364 return nullptr;
365 return I->second;
366 }
367
markDependentsLive(SymbolRef sym)368 void SymbolReaper::markDependentsLive(SymbolRef sym) {
369 // Do not mark dependents more then once.
370 SymbolMapTy::iterator LI = TheLiving.find(sym);
371 assert(LI != TheLiving.end() && "The primary symbol is not live.");
372 if (LI->second == HaveMarkedDependents)
373 return;
374 LI->second = HaveMarkedDependents;
375
376 if (const SymbolRefSmallVectorTy *Deps = SymMgr.getDependentSymbols(sym)) {
377 for (SymbolRefSmallVectorTy::const_iterator I = Deps->begin(),
378 E = Deps->end(); I != E; ++I) {
379 if (TheLiving.find(*I) != TheLiving.end())
380 continue;
381 markLive(*I);
382 }
383 }
384 }
385
markLive(SymbolRef sym)386 void SymbolReaper::markLive(SymbolRef sym) {
387 TheLiving[sym] = NotProcessed;
388 TheDead.erase(sym);
389 markDependentsLive(sym);
390 }
391
markLive(const MemRegion * region)392 void SymbolReaper::markLive(const MemRegion *region) {
393 RegionRoots.insert(region);
394 }
395
markInUse(SymbolRef sym)396 void SymbolReaper::markInUse(SymbolRef sym) {
397 if (isa<SymbolMetadata>(sym))
398 MetadataInUse.insert(sym);
399 }
400
maybeDead(SymbolRef sym)401 bool SymbolReaper::maybeDead(SymbolRef sym) {
402 if (isLive(sym))
403 return false;
404
405 TheDead.insert(sym);
406 return true;
407 }
408
isLiveRegion(const MemRegion * MR)409 bool SymbolReaper::isLiveRegion(const MemRegion *MR) {
410 if (RegionRoots.count(MR))
411 return true;
412
413 MR = MR->getBaseRegion();
414
415 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(MR))
416 return isLive(SR->getSymbol());
417
418 if (const VarRegion *VR = dyn_cast<VarRegion>(MR))
419 return isLive(VR, true);
420
421 // FIXME: This is a gross over-approximation. What we really need is a way to
422 // tell if anything still refers to this region. Unlike SymbolicRegions,
423 // AllocaRegions don't have associated symbols, though, so we don't actually
424 // have a way to track their liveness.
425 if (isa<AllocaRegion>(MR))
426 return true;
427
428 if (isa<CXXThisRegion>(MR))
429 return true;
430
431 if (isa<MemSpaceRegion>(MR))
432 return true;
433
434 if (isa<CodeTextRegion>(MR))
435 return true;
436
437 return false;
438 }
439
isLive(SymbolRef sym)440 bool SymbolReaper::isLive(SymbolRef sym) {
441 if (TheLiving.count(sym)) {
442 markDependentsLive(sym);
443 return true;
444 }
445
446 bool KnownLive;
447
448 switch (sym->getKind()) {
449 case SymExpr::RegionValueKind:
450 KnownLive = isLiveRegion(cast<SymbolRegionValue>(sym)->getRegion());
451 break;
452 case SymExpr::ConjuredKind:
453 KnownLive = false;
454 break;
455 case SymExpr::DerivedKind:
456 KnownLive = isLive(cast<SymbolDerived>(sym)->getParentSymbol());
457 break;
458 case SymExpr::ExtentKind:
459 KnownLive = isLiveRegion(cast<SymbolExtent>(sym)->getRegion());
460 break;
461 case SymExpr::MetadataKind:
462 KnownLive = MetadataInUse.count(sym) &&
463 isLiveRegion(cast<SymbolMetadata>(sym)->getRegion());
464 if (KnownLive)
465 MetadataInUse.erase(sym);
466 break;
467 case SymExpr::SymIntKind:
468 KnownLive = isLive(cast<SymIntExpr>(sym)->getLHS());
469 break;
470 case SymExpr::IntSymKind:
471 KnownLive = isLive(cast<IntSymExpr>(sym)->getRHS());
472 break;
473 case SymExpr::SymSymKind:
474 KnownLive = isLive(cast<SymSymExpr>(sym)->getLHS()) &&
475 isLive(cast<SymSymExpr>(sym)->getRHS());
476 break;
477 case SymExpr::CastSymbolKind:
478 KnownLive = isLive(cast<SymbolCast>(sym)->getOperand());
479 break;
480 }
481
482 if (KnownLive)
483 markLive(sym);
484
485 return KnownLive;
486 }
487
488 bool
isLive(const Stmt * ExprVal,const LocationContext * ELCtx) const489 SymbolReaper::isLive(const Stmt *ExprVal, const LocationContext *ELCtx) const {
490 if (LCtx == nullptr)
491 return false;
492
493 if (LCtx != ELCtx) {
494 // If the reaper's location context is a parent of the expression's
495 // location context, then the expression value is now "out of scope".
496 if (LCtx->isParentOf(ELCtx))
497 return false;
498 return true;
499 }
500
501 // If no statement is provided, everything is this and parent contexts is live.
502 if (!Loc)
503 return true;
504
505 return LCtx->getAnalysis<RelaxedLiveVariables>()->isLive(Loc, ExprVal);
506 }
507
isLive(const VarRegion * VR,bool includeStoreBindings) const508 bool SymbolReaper::isLive(const VarRegion *VR, bool includeStoreBindings) const{
509 const StackFrameContext *VarContext = VR->getStackFrame();
510
511 if (!VarContext)
512 return true;
513
514 if (!LCtx)
515 return false;
516 const StackFrameContext *CurrentContext = LCtx->getCurrentStackFrame();
517
518 if (VarContext == CurrentContext) {
519 // If no statement is provided, everything is live.
520 if (!Loc)
521 return true;
522
523 if (LCtx->getAnalysis<RelaxedLiveVariables>()->isLive(Loc, VR->getDecl()))
524 return true;
525
526 if (!includeStoreBindings)
527 return false;
528
529 unsigned &cachedQuery =
530 const_cast<SymbolReaper*>(this)->includedRegionCache[VR];
531
532 if (cachedQuery) {
533 return cachedQuery == 1;
534 }
535
536 // Query the store to see if the region occurs in any live bindings.
537 if (Store store = reapedStore.getStore()) {
538 bool hasRegion =
539 reapedStore.getStoreManager().includedInBindings(store, VR);
540 cachedQuery = hasRegion ? 1 : 2;
541 return hasRegion;
542 }
543
544 return false;
545 }
546
547 return VarContext->isParentOf(CurrentContext);
548 }
549
~SymbolVisitor()550 SymbolVisitor::~SymbolVisitor() {}
551