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),
36 hasArgument(1, characterLiteral(equals(static_cast<unsigned>('0')))
37 .bind("char-zero-fill")),
38 unless(
39 eachOf(hasArgument(0, anyOf(hasType(pointsTo(isAnyCharacter())),
40 hasType(arrayType(hasElementType(
41 isAnyCharacter()))))),
42 isInTemplateInstantiation()))),
43 this);
44
45 // Look for memset with an integer literal in its fill_char argument.
46 // Will check if it gets truncated.
47 Finder->addMatcher(callExpr(callee(MemsetDecl),
48 hasArgument(1, integerLiteral().bind("num-fill")),
49 unless(isInTemplateInstantiation())),
50 this);
51
52 // Look for memset(x, y, 0) as that is most likely an argument swap.
53 Finder->addMatcher(
54 callExpr(callee(MemsetDecl),
55 unless(hasArgument(1, anyOf(characterLiteral(equals(
56 static_cast<unsigned>('0'))),
57 integerLiteral()))),
58 unless(isInTemplateInstantiation()))
59 .bind("call"),
60 this);
61 }
62
check(const MatchFinder::MatchResult & Result)63 void SuspiciousMemsetUsageCheck::check(const MatchFinder::MatchResult &Result) {
64 if (const auto *CharZeroFill =
65 Result.Nodes.getNodeAs<CharacterLiteral>("char-zero-fill")) {
66 // Case 1: fill_char of memset() is a character '0'. Probably an
67 // integer zero was intended.
68
69 SourceRange CharRange = CharZeroFill->getSourceRange();
70 auto Diag =
71 diag(CharZeroFill->getBeginLoc(), "memset fill value is char '0', "
72 "potentially mistaken for int 0");
73
74 // Only suggest a fix if no macros are involved.
75 if (CharRange.getBegin().isMacroID())
76 return;
77 Diag << FixItHint::CreateReplacement(
78 CharSourceRange::getTokenRange(CharRange), "0");
79 }
80
81 else if (const auto *NumFill =
82 Result.Nodes.getNodeAs<IntegerLiteral>("num-fill")) {
83 // Case 2: fill_char of memset() is larger in size than an unsigned char
84 // so it gets truncated during conversion.
85
86 const auto UCharMax = (1 << Result.Context->getCharWidth()) - 1;
87 Expr::EvalResult EVResult;
88 if (!NumFill->EvaluateAsInt(EVResult, *Result.Context))
89 return;
90
91 llvm::APSInt NumValue = EVResult.Val.getInt();
92 if (NumValue >= 0 && NumValue <= UCharMax)
93 return;
94
95 diag(NumFill->getBeginLoc(), "memset fill value is out of unsigned "
96 "character range, gets truncated");
97 }
98
99 else if (const auto *Call = Result.Nodes.getNodeAs<CallExpr>("call")) {
100 // Case 3: byte_count of memset() is zero. This is most likely an
101 // argument swap.
102
103 const Expr *FillChar = Call->getArg(1);
104 const Expr *ByteCount = Call->getArg(2);
105
106 // Return if `byte_count` is not zero at compile time.
107 Expr::EvalResult Value2;
108 if (ByteCount->isValueDependent() ||
109 !ByteCount->EvaluateAsInt(Value2, *Result.Context) ||
110 Value2.Val.getInt() != 0)
111 return;
112
113 // Return if `fill_char` is known to be zero or negative at compile
114 // time. In these cases, swapping the args would be a nop, or
115 // introduce a definite bug. The code is likely correct.
116 Expr::EvalResult EVResult;
117 if (!FillChar->isValueDependent() &&
118 FillChar->EvaluateAsInt(EVResult, *Result.Context)) {
119 llvm::APSInt Value1 = EVResult.Val.getInt();
120 if (Value1 == 0 || Value1.isNegative())
121 return;
122 }
123
124 // `byte_count` is known to be zero at compile time, and `fill_char` is
125 // either not known or known to be a positive integer. Emit a warning
126 // and fix-its to swap the arguments.
127 auto D = diag(Call->getBeginLoc(),
128 "memset of size zero, potentially swapped arguments");
129 StringRef RHSString = tooling::fixit::getText(*ByteCount, *Result.Context);
130 StringRef LHSString = tooling::fixit::getText(*FillChar, *Result.Context);
131 if (LHSString.empty() || RHSString.empty())
132 return;
133
134 D << tooling::fixit::createReplacement(*FillChar, RHSString)
135 << tooling::fixit::createReplacement(*ByteCount, LHSString);
136 }
137 }
138
139 } // namespace bugprone
140 } // namespace tidy
141 } // namespace clang
142