[clang][analyzer] Model getline/getdelim preconditions and evaluation (#83027)

According to POSIX 2018.

1. lineptr, n and stream can not be NULL.
2. If *n is non-zero, *lineptr must point to a region of at least *n
   bytes, or be a NULL pointer.

Additionally, if *lineptr is not NULL, *n must not be undefined.
This commit is contained in:
Alejandro Álvarez Ayllón 2024-03-22 11:50:34 +01:00 committed by Balazs Benics
parent a62441d4bb
commit 730ca47a0c
7 changed files with 554 additions and 13 deletions

View File

@ -15,7 +15,6 @@
#include "ProgramState_Fwd.h"
#include "SVals.h"
#include "clang/AST/OperationKinds.h"
#include "clang/AST/Stmt.h"
#include "clang/Basic/OperatorKinds.h"
@ -113,8 +112,7 @@ public:
OperatorKind operationKindFromOverloadedOperator(OverloadedOperatorKind OOK,
bool IsBinary);
std::optional<DefinedSVal> getPointeeDefVal(SVal PtrSVal,
ProgramStateRef State);
std::optional<SVal> getPointeeVal(SVal PtrSVal, ProgramStateRef State);
} // namespace ento

View File

@ -1441,7 +1441,7 @@ void MallocChecker::preGetdelim(const CallEvent &Call,
return;
ProgramStateRef State = C.getState();
const auto LinePtr = getPointeeDefVal(Call.getArgSVal(0), State);
const auto LinePtr = getPointeeVal(Call.getArgSVal(0), State);
if (!LinePtr)
return;
@ -1470,8 +1470,10 @@ void MallocChecker::checkGetdelim(const CallEvent &Call,
SValBuilder &SVB = C.getSValBuilder();
const auto LinePtr = getPointeeDefVal(Call.getArgSVal(0), State);
const auto Size = getPointeeDefVal(Call.getArgSVal(1), State);
const auto LinePtr =
getPointeeVal(Call.getArgSVal(0), State)->getAs<DefinedSVal>();
const auto Size =
getPointeeVal(Call.getArgSVal(1), State)->getAs<DefinedSVal>();
if (!LinePtr || !Size || !LinePtr->getAsRegion())
return;

View File

@ -1200,10 +1200,25 @@ void StreamChecker::evalGetdelim(const FnDescription *Desc,
// Add transition for the successful state.
NonLoc RetVal = makeRetVal(C, E.CE).castAs<NonLoc>();
ProgramStateRef StateNotFailed =
State->BindExpr(E.CE, C.getLocationContext(), RetVal);
ProgramStateRef StateNotFailed = E.bindReturnValue(State, C, RetVal);
StateNotFailed =
E.assumeBinOpNN(StateNotFailed, BO_GE, RetVal, E.getZeroVal(Call));
// On success, a buffer is allocated.
auto NewLinePtr = getPointeeVal(Call.getArgSVal(0), State);
if (NewLinePtr && isa<DefinedOrUnknownSVal>(*NewLinePtr))
StateNotFailed = StateNotFailed->assume(
NewLinePtr->castAs<DefinedOrUnknownSVal>(), true);
// The buffer size `*n` must be enough to hold the whole line, and
// greater than the return value, since it has to account for '\0'.
SVal SizePtrSval = Call.getArgSVal(1);
auto NVal = getPointeeVal(SizePtrSval, State);
if (NVal && isa<NonLoc>(*NVal)) {
StateNotFailed = E.assumeBinOpNN(StateNotFailed, BO_GT,
NVal->castAs<NonLoc>(), RetVal);
StateNotFailed = E.bindReturnValue(StateNotFailed, C, RetVal);
}
if (!StateNotFailed)
return;
C.addTransition(StateNotFailed);
@ -1217,6 +1232,10 @@ void StreamChecker::evalGetdelim(const FnDescription *Desc,
E.isStreamEof() ? ErrorFEof : ErrorFEof | ErrorFError;
StateFailed = E.setStreamState(
StateFailed, StreamState::getOpened(Desc, NewES, !NewES.isFEof()));
// On failure, the content of the buffer is undefined.
if (auto NewLinePtr = getPointeeVal(Call.getArgSVal(0), State))
StateFailed = StateFailed->bindLoc(*NewLinePtr, UndefinedVal(),
C.getLocationContext());
C.addTransition(StateFailed, E.getFailureNoteTag(this, C));
}

View File

@ -20,6 +20,7 @@
#include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerHelpers.h"
#include "clang/StaticAnalyzer/Core/PathSensitive/DynamicExtent.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/StringExtras.h"
@ -44,10 +45,23 @@ namespace {
class UnixAPIMisuseChecker
: public Checker<check::PreCall, check::ASTDecl<TranslationUnitDecl>> {
const BugType BT_open{this, "Improper use of 'open'", categories::UnixAPI};
const BugType BT_getline{this, "Improper use of getdelim",
categories::UnixAPI};
const BugType BT_pthreadOnce{this, "Improper use of 'pthread_once'",
categories::UnixAPI};
const BugType BT_ArgumentNull{this, "NULL pointer", categories::UnixAPI};
mutable std::optional<uint64_t> Val_O_CREAT;
ProgramStateRef
EnsurePtrNotNull(SVal PtrVal, const Expr *PtrExpr, CheckerContext &C,
ProgramStateRef State, const StringRef PtrDescr,
std::optional<std::reference_wrapper<const BugType>> BT =
std::nullopt) const;
ProgramStateRef EnsureGetdelimBufferAndSizeCorrect(
SVal LinePtrPtrSVal, SVal SizePtrSVal, const Expr *LinePtrPtrExpr,
const Expr *SizePtrExpr, CheckerContext &C, ProgramStateRef State) const;
public:
void checkASTDecl(const TranslationUnitDecl *TU, AnalysisManager &Mgr,
BugReporter &BR) const;
@ -56,6 +70,7 @@ public:
void CheckOpen(CheckerContext &C, const CallEvent &Call) const;
void CheckOpenAt(CheckerContext &C, const CallEvent &Call) const;
void CheckGetDelim(CheckerContext &C, const CallEvent &Call) const;
void CheckPthreadOnce(CheckerContext &C, const CallEvent &Call) const;
void CheckOpenVariant(CheckerContext &C, const CallEvent &Call,
@ -95,6 +110,30 @@ private:
} // end anonymous namespace
ProgramStateRef UnixAPIMisuseChecker::EnsurePtrNotNull(
SVal PtrVal, const Expr *PtrExpr, CheckerContext &C, ProgramStateRef State,
const StringRef PtrDescr,
std::optional<std::reference_wrapper<const BugType>> BT) const {
const auto Ptr = PtrVal.getAs<DefinedSVal>();
if (!Ptr)
return State;
const auto [PtrNotNull, PtrNull] = State->assume(*Ptr);
if (!PtrNotNull && PtrNull) {
if (ExplodedNode *N = C.generateErrorNode(PtrNull)) {
auto R = std::make_unique<PathSensitiveBugReport>(
BT.value_or(std::cref(BT_ArgumentNull)),
(PtrDescr + " pointer might be NULL.").str(), N);
if (PtrExpr)
bugreporter::trackExpressionValue(N, PtrExpr, *R);
C.emitReport(std::move(R));
}
return nullptr;
}
return PtrNotNull;
}
void UnixAPIMisuseChecker::checkASTDecl(const TranslationUnitDecl *TU,
AnalysisManager &Mgr,
BugReporter &) const {
@ -137,6 +176,9 @@ void UnixAPIMisuseChecker::checkPreCall(const CallEvent &Call,
else if (FName == "pthread_once")
CheckPthreadOnce(C, Call);
else if (is_contained({"getdelim", "getline"}, FName))
CheckGetDelim(C, Call);
}
void UnixAPIMisuseChecker::ReportOpenBug(CheckerContext &C,
ProgramStateRef State,
@ -215,8 +257,7 @@ void UnixAPIMisuseChecker::CheckOpenVariant(CheckerContext &C,
OS << "Call to '" << VariantName << "' with more than " << MaxArgCount
<< " arguments";
ReportOpenBug(C, state,
SBuf.c_str(),
ReportOpenBug(C, state, SBuf.c_str(),
Call.getArgExpr(MaxArgCount)->getSourceRange());
return;
}
@ -266,6 +307,93 @@ void UnixAPIMisuseChecker::CheckOpenVariant(CheckerContext &C,
}
}
//===----------------------------------------------------------------------===//
// getdelim and getline
//===----------------------------------------------------------------------===//
ProgramStateRef UnixAPIMisuseChecker::EnsureGetdelimBufferAndSizeCorrect(
SVal LinePtrPtrSVal, SVal SizePtrSVal, const Expr *LinePtrPtrExpr,
const Expr *SizePtrExpr, CheckerContext &C, ProgramStateRef State) const {
static constexpr llvm::StringLiteral SizeGreaterThanBufferSize =
"The buffer from the first argument is smaller than the size "
"specified by the second parameter";
static constexpr llvm::StringLiteral SizeUndef =
"The buffer from the first argument is not NULL, but the size specified "
"by the second parameter is undefined.";
auto EmitBugReport = [this, &C, SizePtrExpr, LinePtrPtrExpr](
ProgramStateRef BugState, StringRef ErrMsg) {
if (ExplodedNode *N = C.generateErrorNode(BugState)) {
auto R = std::make_unique<PathSensitiveBugReport>(BT_getline, ErrMsg, N);
bugreporter::trackExpressionValue(N, SizePtrExpr, *R);
bugreporter::trackExpressionValue(N, LinePtrPtrExpr, *R);
C.emitReport(std::move(R));
}
};
// We have a pointer to a pointer to the buffer, and a pointer to the size.
// We want what they point at.
auto LinePtrSVal = getPointeeVal(LinePtrPtrSVal, State)->getAs<DefinedSVal>();
auto NSVal = getPointeeVal(SizePtrSVal, State);
if (!LinePtrSVal || !NSVal || NSVal->isUnknown())
return nullptr;
assert(LinePtrPtrExpr && SizePtrExpr);
const auto [LinePtrNotNull, LinePtrNull] = State->assume(*LinePtrSVal);
if (LinePtrNotNull && !LinePtrNull) {
// If `*lineptr` is not null, but `*n` is undefined, there is UB.
if (NSVal->isUndef()) {
EmitBugReport(LinePtrNotNull, SizeUndef);
return nullptr;
}
// If it is defined, and known, its size must be less than or equal to
// the buffer size.
auto NDefSVal = NSVal->getAs<DefinedSVal>();
auto &SVB = C.getSValBuilder();
auto LineBufSize =
getDynamicExtent(LinePtrNotNull, LinePtrSVal->getAsRegion(), SVB);
auto LineBufSizeGtN = SVB.evalBinOp(LinePtrNotNull, BO_GE, LineBufSize,
*NDefSVal, SVB.getConditionType())
.getAs<DefinedOrUnknownSVal>();
if (!LineBufSizeGtN)
return LinePtrNotNull;
if (auto LineBufSizeOk = LinePtrNotNull->assume(*LineBufSizeGtN, true))
return LineBufSizeOk;
EmitBugReport(LinePtrNotNull, SizeGreaterThanBufferSize);
return nullptr;
}
return State;
}
void UnixAPIMisuseChecker::CheckGetDelim(CheckerContext &C,
const CallEvent &Call) const {
ProgramStateRef State = C.getState();
// The parameter `n` must not be NULL.
SVal SizePtrSval = Call.getArgSVal(1);
State = EnsurePtrNotNull(SizePtrSval, Call.getArgExpr(1), C, State, "Size");
if (!State)
return;
// The parameter `lineptr` must not be NULL.
SVal LinePtrPtrSVal = Call.getArgSVal(0);
State =
EnsurePtrNotNull(LinePtrPtrSVal, Call.getArgExpr(0), C, State, "Line");
if (!State)
return;
State = EnsureGetdelimBufferAndSizeCorrect(LinePtrPtrSVal, SizePtrSval,
Call.getArgExpr(0),
Call.getArgExpr(1), C, State);
if (!State)
return;
C.addTransition(State);
}
//===----------------------------------------------------------------------===//
// pthread_once
//===----------------------------------------------------------------------===//

View File

@ -183,10 +183,9 @@ OperatorKind operationKindFromOverloadedOperator(OverloadedOperatorKind OOK,
}
}
std::optional<DefinedSVal> getPointeeDefVal(SVal PtrSVal,
ProgramStateRef State) {
std::optional<SVal> getPointeeVal(SVal PtrSVal, ProgramStateRef State) {
if (const auto *Ptr = PtrSVal.getAsRegion()) {
return State->getSVal(Ptr).getAs<DefinedSVal>();
return State->getSVal(Ptr);
}
return std::nullopt;
}

View File

@ -0,0 +1,322 @@
// RUN: %clang_analyze_cc1 -analyzer-checker=core,unix,debug.ExprInspection -verify %s
#include "Inputs/system-header-simulator.h"
#include "Inputs/system-header-simulator-for-malloc.h"
#include "Inputs/system-header-simulator-for-valist.h"
void clang_analyzer_eval(int);
void clang_analyzer_dump_int(int);
void clang_analyzer_dump_ptr(void*);
void clang_analyzer_warnIfReached();
void test_getline_null_lineptr() {
FILE *F1 = tmpfile();
if (!F1)
return;
char **buffer = NULL;
size_t n = 0;
getline(buffer, &n, F1); // expected-warning {{Line pointer might be NULL}}
fclose(F1);
}
void test_getline_null_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
getline(&buffer, NULL, F1); // expected-warning {{Size pointer might be NULL}}
fclose(F1);
}
void test_getline_null_buffer_size_gt0() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
size_t n = 8;
getline(&buffer, &n, F1); // ok since posix 2018
free(buffer);
fclose(F1);
}
void test_getline_null_buffer_size_gt0_2(size_t n) {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
if (n > 0) {
getline(&buffer, &n, F1); // ok since posix 2018
}
free(buffer);
fclose(F1);
}
void test_getline_null_buffer_unknown_size(size_t n) {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
getline(&buffer, &n, F1); // ok
fclose(F1);
free(buffer);
}
void test_getline_null_buffer_undef_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
size_t n;
getline(&buffer, &n, F1); // ok since posix 2018
fclose(F1);
free(buffer);
}
void test_getline_buffer_size_0() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = malloc(10);
size_t n = 0;
if (buffer != NULL)
getline(&buffer, &n, F1); // ok, the buffer is enough for 0 character
fclose(F1);
free(buffer);
}
void test_getline_buffer_bad_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = malloc(10);
size_t n = 100;
if (buffer != NULL)
getline(&buffer, &n, F1); // expected-warning {{The buffer from the first argument is smaller than the size specified by the second parameter}}
fclose(F1);
free(buffer);
}
void test_getline_buffer_smaller_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = malloc(100);
size_t n = 10;
if (buffer != NULL)
getline(&buffer, &n, F1); // ok, there is enough space for 10 characters
fclose(F1);
free(buffer);
}
void test_getline_buffer_undef_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = malloc(100);
size_t n;
if (buffer != NULL)
getline(&buffer, &n, F1); // expected-warning {{The buffer from the first argument is not NULL, but the size specified by the second parameter is undefined}}
fclose(F1);
free(buffer);
}
void test_getline_null_buffer() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
size_t n = 0;
ssize_t r = getline(&buffer, &n, F1);
// getline returns -1 on failure, number of char reads on success (>= 0)
if (r < -1) {
clang_analyzer_warnIfReached(); // must not happen
} else {
// The buffer could be allocated both on failure and success
clang_analyzer_dump_int(n); // expected-warning {{conj_$}}
clang_analyzer_dump_ptr(buffer); // expected-warning {{conj_$}}
}
free(buffer);
fclose(F1);
}
void test_getdelim_null_size() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
getdelim(&buffer, NULL, ',', F1); // expected-warning {{Size pointer might be NULL}}
fclose(F1);
}
void test_getdelim_null_buffer_size_gt0() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
size_t n = 8;
getdelim(&buffer, &n, ';', F1); // ok since posix 2018
free(buffer);
fclose(F1);
}
void test_getdelim_null_buffer_size_gt0_2(size_t n) {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
if (n > 0) {
getdelim(&buffer, &n, ' ', F1); // ok since posix 2018
}
free(buffer);
fclose(F1);
}
void test_getdelim_null_buffer_unknown_size(size_t n) {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
getdelim(&buffer, &n, '-', F1); // ok
fclose(F1);
free(buffer);
}
void test_getdelim_null_buffer() {
FILE *F1 = tmpfile();
if (!F1)
return;
char *buffer = NULL;
size_t n = 0;
ssize_t r = getdelim(&buffer, &n, '\r', F1);
// getdelim returns -1 on failure, number of char reads on success (>= 0)
if (r < -1) {
clang_analyzer_warnIfReached(); // must not happen
}
else {
// The buffer could be allocated both on failure and success
clang_analyzer_dump_int(n); // expected-warning {{conj_$}}
clang_analyzer_dump_ptr(buffer); // expected-warning {{conj_$}}
}
free(buffer);
fclose(F1);
}
void test_getline_while() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
char *line = NULL;
size_t len = 0;
ssize_t read;
while ((read = getline(&line, &len, file)) != -1) {
printf("%s\n", line);
}
free(line);
fclose(file);
}
void test_getline_return_check() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
char *line = NULL;
size_t len = 0;
ssize_t r = getline(&line, &len, file);
if (r != -1) {
if (line[0] == '\0') {} // ok
}
free(line);
fclose(file);
}
void test_getline_clear_eof() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
size_t n = 10;
char *buffer = malloc(n);
ssize_t read = fread(buffer, n, 1, file);
if (feof(file)) {
clearerr(file);
getline(&buffer, &n, file); // ok
}
fclose(file);
free(buffer);
}
void test_getline_not_null(char **buffer, size_t *size) {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
getline(buffer, size, file);
fclose(file);
if (size == NULL || buffer == NULL) {
clang_analyzer_warnIfReached(); // must not happen
}
}
void test_getline_size_constraint(size_t size) {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
size_t old_size = size;
char *buffer = malloc(10);
if (buffer != NULL) {
ssize_t r = getline(&buffer, &size, file);
if (r >= 0) {
// Since buffer has a size of 10, old_size must be less than or equal to 10.
// Otherwise, there would be UB.
clang_analyzer_eval(old_size <= 10); // expected-warning{{TRUE}}
}
}
fclose(file);
free(buffer);
}
void test_getline_negative_buffer() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
char *buffer = NULL;
size_t n = -1;
getline(&buffer, &n, file); // ok since posix 2018
free(buffer);
fclose(file);
}
void test_getline_negative_buffer_2(char *buffer) {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
size_t n = -1;
(void)getline(&buffer, &n, file); // ok
free(buffer);
fclose(file);
}

View File

@ -4,6 +4,7 @@
// RUN: %clang_analyze_cc1 -triple=hexagon -analyzer-checker=core,alpha.unix.Stream,debug.ExprInspection -verify %s
#include "Inputs/system-header-simulator.h"
#include "Inputs/system-header-simulator-for-malloc.h"
#include "Inputs/system-header-simulator-for-valist.h"
void clang_analyzer_eval(int);
@ -376,3 +377,75 @@ void fflush_on_open_failed_stream(void) {
}
fclose(F);
}
void getline_null_file() {
char *buffer = NULL;
size_t n = 0;
getline(&buffer, &n, NULL); // expected-warning {{Stream pointer might be NULL}}
}
void getdelim_null_file() {
char *buffer = NULL;
size_t n = 0;
getdelim(&buffer, &n, '\n', NULL); // expected-warning {{Stream pointer might be NULL}}
}
void getline_buffer_on_error() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
char *line = NULL;
size_t len = 0;
if (getline(&line, &len, file) == -1) {
if (line[0] == '\0') {} // expected-warning {{The left operand of '==' is a garbage value}}
} else {
if (line[0] == '\0') {} // no warning
}
free(line);
fclose(file);
}
void getline_ret_value() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
size_t n = 0;
char *buffer = NULL;
ssize_t r = getline(&buffer, &n, file);
if (r > -1) {
// The return value does *not* include the terminating null byte.
// The buffer must be large enough to include it.
clang_analyzer_eval(n > r); // expected-warning{{TRUE}}
clang_analyzer_eval(buffer != NULL); // expected-warning{{TRUE}}
}
fclose(file);
free(buffer);
}
void getline_buffer_size_negative() {
FILE *file = fopen("file.txt", "r");
if (file == NULL) {
return;
}
size_t n = -1;
clang_analyzer_eval((ssize_t)n >= 0); // expected-warning{{FALSE}}
char *buffer = NULL;
ssize_t r = getline(&buffer, &n, file);
if (r > -1) {
clang_analyzer_eval((ssize_t)n > r); // expected-warning{{TRUE}}
clang_analyzer_eval(buffer != NULL); // expected-warning{{TRUE}}
}
free(buffer);
fclose(file);
}