1 //===--- SuspiciousMemsetUsageCheck.cpp - clang-tidy-----------------------===//
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 #include "SuspiciousMemsetUsageCheck.h"
10 #include "clang/AST/ASTContext.h"
11 #include "clang/ASTMatchers/ASTMatchFinder.h"
12 #include "clang/ASTMatchers/ASTMatchers.h"
13 #include "clang/Lex/Lexer.h"
14 #include "clang/Tooling/FixIt.h"
15 
16 using namespace clang::ast_matchers;
17 
18 namespace clang {
19 namespace tidy {
20 namespace bugprone {
21 
registerMatchers(MatchFinder * Finder)22 void SuspiciousMemsetUsageCheck::registerMatchers(MatchFinder *Finder) {
23   // Match the standard memset:
24   // void *memset(void *buffer, int fill_char, size_t byte_count);
25   auto MemsetDecl =
26       functionDecl(hasName("::memset"),
27                    parameterCountIs(3),
28                    hasParameter(0, hasType(pointerType(pointee(voidType())))),
29                    hasParameter(1, hasType(isInteger())),
30                    hasParameter(2, hasType(isInteger())));
31 
32   // Look for memset(x, '0', z). Probably memset(x, 0, z) was intended.
33   Finder->addMatcher(
34       callExpr(
35           callee(MemsetDecl), argumentCountIs(3),
36           hasArgument(1, characterLiteral(equals(static_cast<unsigned>('0')))
37                              .bind("char-zero-fill")),
38           unless(hasArgument(
39               0, anyOf(hasType(pointsTo(isAnyCharacter())),
40                        hasType(arrayType(hasElementType(isAnyCharacter()))))))),
41       this);
42 
43   // Look for memset with an integer literal in its fill_char argument.
44   // Will check if it gets truncated.
45   Finder->addMatcher(
46       callExpr(callee(MemsetDecl), argumentCountIs(3),
47                hasArgument(1, integerLiteral().bind("num-fill"))),
48       this);
49 
50   // Look for memset(x, y, 0) as that is most likely an argument swap.
51   Finder->addMatcher(
52       callExpr(callee(MemsetDecl), argumentCountIs(3),
53                unless(hasArgument(1, anyOf(characterLiteral(equals(
54                                                static_cast<unsigned>('0'))),
55                                            integerLiteral()))))
56           .bind("call"),
57       this);
58 }
59 
check(const MatchFinder::MatchResult & Result)60 void SuspiciousMemsetUsageCheck::check(const MatchFinder::MatchResult &Result) {
61   if (const auto *CharZeroFill =
62           Result.Nodes.getNodeAs<CharacterLiteral>("char-zero-fill")) {
63     // Case 1: fill_char of memset() is a character '0'. Probably an
64     // integer zero was intended.
65 
66     SourceRange CharRange = CharZeroFill->getSourceRange();
67     auto Diag =
68         diag(CharZeroFill->getBeginLoc(), "memset fill value is char '0', "
69                                           "potentially mistaken for int 0");
70 
71     // Only suggest a fix if no macros are involved.
72     if (CharRange.getBegin().isMacroID())
73       return;
74     Diag << FixItHint::CreateReplacement(
75         CharSourceRange::getTokenRange(CharRange), "0");
76   }
77 
78   else if (const auto *NumFill =
79                Result.Nodes.getNodeAs<IntegerLiteral>("num-fill")) {
80     // Case 2: fill_char of memset() is larger in size than an unsigned char
81     // so it gets truncated during conversion.
82 
83     const auto UCharMax = (1 << Result.Context->getCharWidth()) - 1;
84     Expr::EvalResult EVResult;
85     if (!NumFill->EvaluateAsInt(EVResult, *Result.Context))
86       return;
87 
88     llvm::APSInt NumValue = EVResult.Val.getInt();
89     if (NumValue >= 0 && NumValue <= UCharMax)
90       return;
91 
92     diag(NumFill->getBeginLoc(), "memset fill value is out of unsigned "
93                                  "character range, gets truncated");
94   }
95 
96   else if (const auto *Call = Result.Nodes.getNodeAs<CallExpr>("call")) {
97     // Case 3: byte_count of memset() is zero. This is most likely an
98     // argument swap.
99 
100     const Expr *FillChar = Call->getArg(1);
101     const Expr *ByteCount = Call->getArg(2);
102 
103     // Return if `byte_count` is not zero at compile time.
104     Expr::EvalResult Value2;
105     if (ByteCount->isValueDependent() ||
106         !ByteCount->EvaluateAsInt(Value2, *Result.Context) ||
107         Value2.Val.getInt() != 0)
108       return;
109 
110     // Return if `fill_char` is known to be zero or negative at compile
111     // time. In these cases, swapping the args would be a nop, or
112     // introduce a definite bug. The code is likely correct.
113     Expr::EvalResult EVResult;
114     if (!FillChar->isValueDependent() &&
115         FillChar->EvaluateAsInt(EVResult, *Result.Context)) {
116       llvm::APSInt Value1 = EVResult.Val.getInt();
117       if (Value1 == 0 || Value1.isNegative())
118         return;
119     }
120 
121     // `byte_count` is known to be zero at compile time, and `fill_char` is
122     // either not known or known to be a positive integer. Emit a warning
123     // and fix-its to swap the arguments.
124     auto D = diag(Call->getBeginLoc(),
125                   "memset of size zero, potentially swapped arguments");
126     StringRef RHSString = tooling::fixit::getText(*ByteCount, *Result.Context);
127     StringRef LHSString = tooling::fixit::getText(*FillChar, *Result.Context);
128     if (LHSString.empty() || RHSString.empty())
129       return;
130 
131     D << tooling::fixit::createReplacement(*FillChar, RHSString)
132       << tooling::fixit::createReplacement(*ByteCount, LHSString);
133   }
134 }
135 
136 } // namespace bugprone
137 } // namespace tidy
138 } // namespace clang
139