1//=== LLVMConventionsChecker.cpp - Check LLVM codebase conventions ---*- 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 defines LLVMConventionsChecker, a bunch of small little checks
11// for checking specific coding conventions in the LLVM/Clang codebase.
12//
13//===----------------------------------------------------------------------===//
14
15#include "ClangSACheckers.h"
16#include "clang/StaticAnalyzer/Core/Checker.h"
17#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h"
18#include "clang/AST/DeclTemplate.h"
19#include "clang/AST/StmtVisitor.h"
20#include <string>
21#include "llvm/ADT/StringRef.h"
22
23using namespace clang;
24using namespace ento;
25
26//===----------------------------------------------------------------------===//
27// Generic type checking routines.
28//===----------------------------------------------------------------------===//
29
30static bool IsLLVMStringRef(QualType T) {
31  const RecordType *RT = T->getAs<RecordType>();
32  if (!RT)
33    return false;
34
35  return StringRef(QualType(RT, 0).getAsString()) ==
36          "class StringRef";
37}
38
39/// Check whether the declaration is semantically inside the top-level
40/// namespace named by ns.
41static bool InNamespace(const Decl *D, StringRef NS) {
42  const NamespaceDecl *ND = dyn_cast<NamespaceDecl>(D->getDeclContext());
43  if (!ND)
44    return false;
45  const IdentifierInfo *II = ND->getIdentifier();
46  if (!II || !II->getName().equals(NS))
47    return false;
48  return isa<TranslationUnitDecl>(ND->getDeclContext());
49}
50
51static bool IsStdString(QualType T) {
52  if (const ElaboratedType *QT = T->getAs<ElaboratedType>())
53    T = QT->getNamedType();
54
55  const TypedefType *TT = T->getAs<TypedefType>();
56  if (!TT)
57    return false;
58
59  const TypedefNameDecl *TD = TT->getDecl();
60
61  if (!InNamespace(TD, "std"))
62    return false;
63
64  return TD->getName() == "string";
65}
66
67static bool IsClangType(const RecordDecl *RD) {
68  return RD->getName() == "Type" && InNamespace(RD, "clang");
69}
70
71static bool IsClangDecl(const RecordDecl *RD) {
72  return RD->getName() == "Decl" && InNamespace(RD, "clang");
73}
74
75static bool IsClangStmt(const RecordDecl *RD) {
76  return RD->getName() == "Stmt" && InNamespace(RD, "clang");
77}
78
79static bool IsClangAttr(const RecordDecl *RD) {
80  return RD->getName() == "Attr" && InNamespace(RD, "clang");
81}
82
83static bool IsStdVector(QualType T) {
84  const TemplateSpecializationType *TS = T->getAs<TemplateSpecializationType>();
85  if (!TS)
86    return false;
87
88  TemplateName TM = TS->getTemplateName();
89  TemplateDecl *TD = TM.getAsTemplateDecl();
90
91  if (!TD || !InNamespace(TD, "std"))
92    return false;
93
94  return TD->getName() == "vector";
95}
96
97static bool IsSmallVector(QualType T) {
98  const TemplateSpecializationType *TS = T->getAs<TemplateSpecializationType>();
99  if (!TS)
100    return false;
101
102  TemplateName TM = TS->getTemplateName();
103  TemplateDecl *TD = TM.getAsTemplateDecl();
104
105  if (!TD || !InNamespace(TD, "llvm"))
106    return false;
107
108  return TD->getName() == "SmallVector";
109}
110
111//===----------------------------------------------------------------------===//
112// CHECK: a StringRef should not be bound to a temporary std::string whose
113// lifetime is shorter than the StringRef's.
114//===----------------------------------------------------------------------===//
115
116namespace {
117class StringRefCheckerVisitor : public StmtVisitor<StringRefCheckerVisitor> {
118  BugReporter &BR;
119public:
120  StringRefCheckerVisitor(BugReporter &br) : BR(br) {}
121  void VisitChildren(Stmt *S) {
122    for (Stmt::child_iterator I = S->child_begin(), E = S->child_end() ;
123      I != E; ++I)
124      if (Stmt *child = *I)
125        Visit(child);
126  }
127  void VisitStmt(Stmt *S) { VisitChildren(S); }
128  void VisitDeclStmt(DeclStmt *DS);
129private:
130  void VisitVarDecl(VarDecl *VD);
131};
132} // end anonymous namespace
133
134static void CheckStringRefAssignedTemporary(const Decl *D, BugReporter &BR) {
135  StringRefCheckerVisitor walker(BR);
136  walker.Visit(D->getBody());
137}
138
139void StringRefCheckerVisitor::VisitDeclStmt(DeclStmt *S) {
140  VisitChildren(S);
141
142  for (DeclStmt::decl_iterator I = S->decl_begin(), E = S->decl_end();I!=E; ++I)
143    if (VarDecl *VD = dyn_cast<VarDecl>(*I))
144      VisitVarDecl(VD);
145}
146
147void StringRefCheckerVisitor::VisitVarDecl(VarDecl *VD) {
148  Expr *Init = VD->getInit();
149  if (!Init)
150    return;
151
152  // Pattern match for:
153  // StringRef x = call() (where call returns std::string)
154  if (!IsLLVMStringRef(VD->getType()))
155    return;
156  ExprWithCleanups *Ex1 = dyn_cast<ExprWithCleanups>(Init);
157  if (!Ex1)
158    return;
159  CXXConstructExpr *Ex2 = dyn_cast<CXXConstructExpr>(Ex1->getSubExpr());
160  if (!Ex2 || Ex2->getNumArgs() != 1)
161    return;
162  ImplicitCastExpr *Ex3 = dyn_cast<ImplicitCastExpr>(Ex2->getArg(0));
163  if (!Ex3)
164    return;
165  CXXConstructExpr *Ex4 = dyn_cast<CXXConstructExpr>(Ex3->getSubExpr());
166  if (!Ex4 || Ex4->getNumArgs() != 1)
167    return;
168  ImplicitCastExpr *Ex5 = dyn_cast<ImplicitCastExpr>(Ex4->getArg(0));
169  if (!Ex5)
170    return;
171  CXXBindTemporaryExpr *Ex6 = dyn_cast<CXXBindTemporaryExpr>(Ex5->getSubExpr());
172  if (!Ex6 || !IsStdString(Ex6->getType()))
173    return;
174
175  // Okay, badness!  Report an error.
176  const char *desc = "StringRef should not be bound to temporary "
177                     "std::string that it outlives";
178  PathDiagnosticLocation VDLoc =
179    PathDiagnosticLocation::createBegin(VD, BR.getSourceManager());
180  BR.EmitBasicReport(desc, "LLVM Conventions", desc,
181                     VDLoc, Init->getSourceRange());
182}
183
184//===----------------------------------------------------------------------===//
185// CHECK: Clang AST nodes should not have fields that can allocate
186//   memory.
187//===----------------------------------------------------------------------===//
188
189static bool AllocatesMemory(QualType T) {
190  return IsStdVector(T) || IsStdString(T) || IsSmallVector(T);
191}
192
193// This type checking could be sped up via dynamic programming.
194static bool IsPartOfAST(const CXXRecordDecl *R) {
195  if (IsClangStmt(R) || IsClangType(R) || IsClangDecl(R) || IsClangAttr(R))
196    return true;
197
198  for (CXXRecordDecl::base_class_const_iterator I = R->bases_begin(),
199                                                E = R->bases_end(); I!=E; ++I) {
200    CXXBaseSpecifier BS = *I;
201    QualType T = BS.getType();
202    if (const RecordType *baseT = T->getAs<RecordType>()) {
203      CXXRecordDecl *baseD = cast<CXXRecordDecl>(baseT->getDecl());
204      if (IsPartOfAST(baseD))
205        return true;
206    }
207  }
208
209  return false;
210}
211
212namespace {
213class ASTFieldVisitor {
214  SmallVector<FieldDecl*, 10> FieldChain;
215  const CXXRecordDecl *Root;
216  BugReporter &BR;
217public:
218  ASTFieldVisitor(const CXXRecordDecl *root, BugReporter &br)
219    : Root(root), BR(br) {}
220
221  void Visit(FieldDecl *D);
222  void ReportError(QualType T);
223};
224} // end anonymous namespace
225
226static void CheckASTMemory(const CXXRecordDecl *R, BugReporter &BR) {
227  if (!IsPartOfAST(R))
228    return;
229
230  for (RecordDecl::field_iterator I = R->field_begin(), E = R->field_end();
231       I != E; ++I) {
232    ASTFieldVisitor walker(R, BR);
233    walker.Visit(*I);
234  }
235}
236
237void ASTFieldVisitor::Visit(FieldDecl *D) {
238  FieldChain.push_back(D);
239
240  QualType T = D->getType();
241
242  if (AllocatesMemory(T))
243    ReportError(T);
244
245  if (const RecordType *RT = T->getAs<RecordType>()) {
246    const RecordDecl *RD = RT->getDecl()->getDefinition();
247    for (RecordDecl::field_iterator I = RD->field_begin(), E = RD->field_end();
248         I != E; ++I)
249      Visit(*I);
250  }
251
252  FieldChain.pop_back();
253}
254
255void ASTFieldVisitor::ReportError(QualType T) {
256  llvm::SmallString<1024> buf;
257  llvm::raw_svector_ostream os(buf);
258
259  os << "AST class '" << Root->getName() << "' has a field '"
260     << FieldChain.front()->getName() << "' that allocates heap memory";
261  if (FieldChain.size() > 1) {
262    os << " via the following chain: ";
263    bool isFirst = true;
264    for (SmallVectorImpl<FieldDecl*>::iterator I=FieldChain.begin(),
265         E=FieldChain.end(); I!=E; ++I) {
266      if (!isFirst)
267        os << '.';
268      else
269        isFirst = false;
270      os << (*I)->getName();
271    }
272  }
273  os << " (type " << FieldChain.back()->getType().getAsString() << ")";
274  os.flush();
275
276  // Note that this will fire for every translation unit that uses this
277  // class.  This is suboptimal, but at least scan-build will merge
278  // duplicate HTML reports.  In the future we need a unified way of merging
279  // duplicate reports across translation units.  For C++ classes we cannot
280  // just report warnings when we see an out-of-line method definition for a
281  // class, as that heuristic doesn't always work (the complete definition of
282  // the class may be in the header file, for example).
283  PathDiagnosticLocation L = PathDiagnosticLocation::createBegin(
284                               FieldChain.front(), BR.getSourceManager());
285  BR.EmitBasicReport("AST node allocates heap memory", "LLVM Conventions",
286                     os.str(), L);
287}
288
289//===----------------------------------------------------------------------===//
290// LLVMConventionsChecker
291//===----------------------------------------------------------------------===//
292
293namespace {
294class LLVMConventionsChecker : public Checker<
295                                                check::ASTDecl<CXXRecordDecl>,
296                                                check::ASTCodeBody > {
297public:
298  void checkASTDecl(const CXXRecordDecl *R, AnalysisManager& mgr,
299                    BugReporter &BR) const {
300    if (R->isCompleteDefinition())
301      CheckASTMemory(R, BR);
302  }
303
304  void checkASTCodeBody(const Decl *D, AnalysisManager& mgr,
305                        BugReporter &BR) const {
306    CheckStringRefAssignedTemporary(D, BR);
307  }
308};
309}
310
311void ento::registerLLVMConventionsChecker(CheckerManager &mgr) {
312  mgr.registerChecker<LLVMConventionsChecker>();
313}
314