[clang][sema] Add nonnull attribute to builtin format functions

[bozicrHT]

Annotate printf/scanf and related builtins with the nonnull attribute on their format string parameters. This enables diagnostics when NULL is passed, matching GCC behavior. Updated existing Sema tests and added new ones for coverage. Closes issue #33923

[llvmbot]

@llvm/pr-subscribers-clang

Author: Radovan Božić (bozicrHT)

Changes

Annotate printf/scanf and related builtins with the nonnull attribute on their format string parameters. This enables diagnostics when NULL is passed, matching GCC behavior. Updated existing Sema tests and added new one for coverage. Closes issue #33923

---

Full diff: https://github.com/llvm/llvm-project/pull/160988.diff

10 Files Affected:

• (modified) clang/include/clang/Basic/Builtins.h (+4)
• (modified) clang/include/clang/Basic/Builtins.td (+16-15)
• (modified) clang/include/clang/Basic/BuiltinsBase.td (+1-1)
• (modified) clang/lib/Basic/Builtins.cpp (+33-15)
• (modified) clang/lib/Sema/SemaDecl.cpp (+9)
• (added) clang/test/Sema/format-strings-nonnull.c (+74)
• (added) clang/test/Sema/format-strings-nonnull.cpp (+40)
• (modified) clang/test/Sema/format-strings.c (+2-4)
• (modified) clang/test/SemaCXX/format-strings-0x.cpp (+2)
• (modified) clang/test/SemaObjC/format-strings-objc.m (+2-1)

diff --git a/clang/include/clang/Basic/Builtins.h b/clang/include/clang/Basic/Builtins.h
index 3a5e31de2bc50..6d26a7b92a0fd 100644
--- a/clang/include/clang/Basic/Builtins.h
+++ b/clang/include/clang/Basic/Builtins.h
@@ -392,6 +392,10 @@ class Context {
   bool performsCallback(unsigned ID,
                         llvm::SmallVectorImpl<int> &Encoding) const;
 
+  /// Return true if this builtin has parameters that must be non-null.
+  /// The parameter indices are appended into 'Indxs'.
+  bool isNonNull(unsigned ID, llvm::SmallVectorImpl<int> &Indxs) const;
+
   /// Return true if this function has no side effects and doesn't
   /// read memory, except for possibly errno or raising FP exceptions.
   ///
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 9bc70ea5e5858..5f9843f482276 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -3095,104 +3095,105 @@ def StrLen : LibBuiltin<"string.h"> {
 // FIXME: This list is incomplete.
 def Printf : LibBuiltin<"stdio.h"> {
   let Spellings = ["printf"];
-  let Attributes = [PrintfFormat<0>];
+  let Attributes = [PrintfFormat<0>, NonNull<[0]>];
   let Prototype = "int(char const*, ...)";
 }
 
 // FIXME: The builtin and library function should have the same signature.
 def BuiltinPrintf : Builtin {
   let Spellings = ["__builtin_printf"];
-  let Attributes = [NoThrow, PrintfFormat<0>, FunctionWithBuiltinPrefix];
+  let Attributes = [NoThrow, PrintfFormat<0>, FunctionWithBuiltinPrefix,
+                    NonNull<[0]>];
   let Prototype = "int(char const* restrict, ...)";
 }
 
 def FPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["fprintf"];
-  let Attributes = [NoThrow, PrintfFormat<1>];
+  let Attributes = [NoThrow, PrintfFormat<1>, NonNull<[1]>];
   let Prototype = "int(FILE* restrict, char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def SnPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["snprintf"];
-  let Attributes = [NoThrow, PrintfFormat<2>];
+  let Attributes = [NoThrow, PrintfFormat<2>, NonNull<[2]>];
   let Prototype = "int(char* restrict, size_t, char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def SPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["sprintf"];
-  let Attributes = [NoThrow, PrintfFormat<1>];
+  let Attributes = [NoThrow, PrintfFormat<1>, NonNull<[1]>];
   let Prototype = "int(char* restrict, char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vprintf"];
-  let Attributes = [NoThrow, VPrintfFormat<0>];
+  let Attributes = [NoThrow, VPrintfFormat<0>, NonNull<[0]>];
   let Prototype = "int(char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VfPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vfprintf"];
-  let Attributes = [NoThrow, VPrintfFormat<1>];
+  let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<[1]>];
   let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VsnPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vsnprintf"];
-  let Attributes = [NoThrow, VPrintfFormat<2>];
+  let Attributes = [NoThrow, VPrintfFormat<2>, NonNull<[2]>];
   let Prototype = "int(char* restrict, size_t, char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VsPrintf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vsprintf"];
-  let Attributes = [NoThrow, VPrintfFormat<1>];
+  let Attributes = [NoThrow, VPrintfFormat<1>, NonNull<[1]>];
   let Prototype = "int(char* restrict, char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def Scanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["scanf"];
-  let Attributes = [ScanfFormat<0>];
+  let Attributes = [ScanfFormat<0>, NonNull<[0]>];
   let Prototype = "int(char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def FScanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["fscanf"];
-  let Attributes = [ScanfFormat<1>];
+  let Attributes = [ScanfFormat<1>, NonNull<[1]>];
   let Prototype = "int(FILE* restrict, char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def SScanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["sscanf"];
-  let Attributes = [ScanfFormat<1>];
+  let Attributes = [ScanfFormat<1>, NonNull<[1]>];
   let Prototype = "int(char const* restrict, char const* restrict, ...)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VScanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vscanf"];
-  let Attributes = [VScanfFormat<0>];
+  let Attributes = [VScanfFormat<0>, NonNull<[0]>];
   let Prototype = "int(char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VFScanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vfscanf"];
-  let Attributes = [VScanfFormat<1>];
+  let Attributes = [VScanfFormat<1>, NonNull<[1]>];
   let Prototype = "int(FILE* restrict, char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
 
 def VSScanf : LibBuiltin<"stdio.h"> {
   let Spellings = ["vsscanf"];
-  let Attributes = [VScanfFormat<1>];
+  let Attributes = [VScanfFormat<1>, NonNull<[1]>];
   let Prototype = "int(char const* restrict, char const* restrict, __builtin_va_list)";
   let AddBuiltinPrefixedAlias = 1;
 }
diff --git a/clang/include/clang/Basic/BuiltinsBase.td b/clang/include/clang/Basic/BuiltinsBase.td
index 09bc9f89059fe..73918ab167b8d 100644
--- a/clang/include/clang/Basic/BuiltinsBase.td
+++ b/clang/include/clang/Basic/BuiltinsBase.td
@@ -32,7 +32,6 @@ def Const : Attribute<"c">;
 def NoThrow : Attribute<"n">;
 def Pure : Attribute<"U">;
 def ReturnsTwice : Attribute<"j">;
-//  FIXME: gcc has nonnull
 
 // builtin-specific attributes
 // ---------------------------
@@ -85,6 +84,7 @@ def Consteval : Attribute<"EG">;
 // Callback behavior: the first index argument is called with the arguments
 // indicated by the remaining indices.
 class Callback<list<int> ArgIndices> : MultiIndexAttribute<"C", ArgIndices>;
+class NonNull<list<int> ArgIndices> : MultiIndexAttribute<"N", ArgIndices>;
 
 // Prefixes
 // ========
diff --git a/clang/lib/Basic/Builtins.cpp b/clang/lib/Basic/Builtins.cpp
index acd98fe84adf5..a813726b7b848 100644
--- a/clang/lib/Basic/Builtins.cpp
+++ b/clang/lib/Basic/Builtins.cpp
@@ -293,30 +293,48 @@ bool Builtin::Context::isScanfLike(unsigned ID, unsigned &FormatIdx,
   return isLike(ID, FormatIdx, HasVAListArg, "sS");
 }
 
-bool Builtin::Context::performsCallback(unsigned ID,
-                                        SmallVectorImpl<int> &Encoding) const {
-  const char *CalleePos = ::strchr(getAttributesString(ID), 'C');
-  if (!CalleePos)
-    return false;
-
-  ++CalleePos;
-  assert(*CalleePos == '<' &&
-         "Callback callee specifier must be followed by a '<'");
-  ++CalleePos;
+static void parseCommaSeparatedIndices(const char *CurrPos,
+                                       llvm::SmallVectorImpl<int> &Indxs) {
+  assert(*CurrPos == '<' && "Expected '<' to start index list");
+  ++CurrPos;
 
   char *EndPos;
-  int CalleeIdx = ::strtol(CalleePos, &EndPos, 10);
-  assert(CalleeIdx >= 0 && "Callee index is supposed to be positive!");
-  Encoding.push_back(CalleeIdx);
+  int PosIdx = ::strtol(CurrPos, &EndPos, 10);
+  assert(PosIdx >= 0 && "Index is supposed to be positive!");
+  Indxs.push_back(PosIdx);
 
   while (*EndPos == ',') {
     const char *PayloadPos = EndPos + 1;
 
     int PayloadIdx = ::strtol(PayloadPos, &EndPos, 10);
-    Encoding.push_back(PayloadIdx);
+    Indxs.push_back(PayloadIdx);
   }
 
-  assert(*EndPos == '>' && "Callback callee specifier must end with a '>'");
+  assert(*EndPos == '>' && "Index list must end with '>'");
+}
+
+bool Builtin::Context::isNonNull(unsigned ID,
+                                 llvm::SmallVectorImpl<int> &Indxs) const {
+
+  const char *AttrPos = ::strchr(getAttributesString(ID), 'N');
+  if (!AttrPos)
+    return false;
+
+  ++AttrPos;
+  parseCommaSeparatedIndices(AttrPos, Indxs);
+
+  return true;
+}
+
+bool Builtin::Context::performsCallback(unsigned ID,
+                                        SmallVectorImpl<int> &Encoding) const {
+  const char *CalleePos = ::strchr(getAttributesString(ID), 'C');
+  if (!CalleePos)
+    return false;
+
+  ++CalleePos;
+  parseCommaSeparatedIndices(CalleePos, Encoding);
+
   return true;
 }
 
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 9ef7a2698913d..d9df60af114fb 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -17141,6 +17141,15 @@ void Sema::AddKnownFunctionAttributes(FunctionDecl *FD) {
       }
     }
 
+    SmallVector<int, 4> Indxs;
+    if (Context.BuiltinInfo.isNonNull(BuiltinID, Indxs) &&
+        !FD->hasAttr<NonNullAttr>()) {
+      llvm::SmallVector<ParamIdx, 4> ParamIndxs;
+      for (int I : Indxs)
+        ParamIndxs.push_back(ParamIdx(I + 1, FD));
+      FD->addAttr(NonNullAttr::CreateImplicit(Context, ParamIndxs.data(),
+                                              ParamIndxs.size()));
+    }
     if (Context.BuiltinInfo.isReturnsTwice(BuiltinID) &&
         !FD->hasAttr<ReturnsTwiceAttr>())
       FD->addAttr(ReturnsTwiceAttr::CreateImplicit(Context,
diff --git a/clang/test/Sema/format-strings-nonnull.c b/clang/test/Sema/format-strings-nonnull.c
new file mode 100644
index 0000000000000..86ce24a27ca11
--- /dev/null
+++ b/clang/test/Sema/format-strings-nonnull.c
@@ -0,0 +1,74 @@
+// RUN: %clang_cc1 -fsyntax-only --std=c23 -verify -Wnonnull -Wno-format-security %s
+
+#define NULL  (void*)0
+
+typedef struct _FILE FILE;
+typedef __SIZE_TYPE__ size_t;
+typedef __builtin_va_list va_list;
+int printf(char const* restrict, ...);
+int __builtin_printf(char const* restrict, ...);
+int fprintf(FILE* restrict, char const* restrict, ...);
+int snprintf(char* restrict, size_t, char const* restrict, ...);
+int sprintf(char* restrict, char const* restrict, ...);
+int vprintf(char const* restrict, __builtin_va_list);
+int vfprintf(FILE* restrict, char const* restrict, __builtin_va_list);
+int vsnprintf(char* restrict, size_t, char const* restrict, __builtin_va_list);
+int vsprintf(char* restrict, char const* restrict, __builtin_va_list);
+
+int scanf(char const* restrict, ...);
+int fscanf(FILE* restrict, char const* restrict, ...);
+int sscanf(char const* restrict, char const* restrict, ...);
+int vscanf(char const* restrict, __builtin_va_list);
+int vfscanf(FILE* restrict, char const* restrict, __builtin_va_list);
+int vsscanf(char const* restrict, char const* restrict, __builtin_va_list);
+
+
+void check_format_string(FILE *fp, va_list ap) {
+    char buf[256];
+    char* const fmt = NULL;
+
+    printf(fmt);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    __builtin_printf(NULL, "xxd");
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    fprintf(fp, NULL, 25);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    sprintf(buf, NULL, 42);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    snprintf(buf, 10, 0, 42);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vprintf(fmt, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vfprintf(fp, 0, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vsprintf(buf, nullptr, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vsnprintf(buf, 10, fmt, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    scanf(NULL);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    fscanf(fp, nullptr);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    sscanf(buf, fmt);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vscanf(NULL, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vfscanf(fp, fmt, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+
+    vsscanf(buf, NULL, ap);
+    // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
+}
\ No newline at end of file
diff --git a/clang/test/Sema/format-strings-nonnull.cpp b/clang/test/Sema/format-strings-nonnull.cpp
new file mode 100644
index 0000000000000..55a3ed1788c79
--- /dev/null
+++ b/clang/test/Sema/format-strings-nonnull.cpp
@@ -0,0 +1,40 @@
+// RUN: %clang_cc1 -fsyntax-only -verify -Wnonnull %s
+
+#ifdef __cplusplus
+# define EXTERN_C extern "C"
+#else
+# define EXTERN_C extern
+#endif
+
+typedef struct _FILE FILE;
+typedef __SIZE_TYPE__ size_t;
+typedef __builtin_va_list va_list;
+
+EXTERN_C int printf(const char *, ...);
+EXTERN_C int fprintf(FILE *, const char *restrict, ...);
+EXTERN_C int sprintf(char* restrict, char const* res, ...);
+
+EXTERN_C int scanf(char const *restrict, ...);
+EXTERN_C int fscanf(FILE* restrict, char const* res, ...);
+
+void test(FILE *fp) {
+  char buf[256];
+
+  __builtin_printf(__null, "x");
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+
+  printf(__null, "xxd");
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+
+  fprintf(fp, __null, 42);
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+
+  sprintf(buf, __null);
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+
+  scanf(__null);
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+
+  fscanf(fp, __null);
+  // expected-warning@-1 {{null passed to a callee that requires a non-null argument}}
+}
diff --git a/clang/test/Sema/format-strings.c b/clang/test/Sema/format-strings.c
index 103dd8ab5a85c..164db45fa2053 100644
--- a/clang/test/Sema/format-strings.c
+++ b/clang/test/Sema/format-strings.c
@@ -480,11 +480,9 @@ void pr7981(wint_t c, wchar_t c2) {
 #endif
 }
 
-// -Wformat-security says NULL is not a string literal
 void rdar8269537(void) {
-  // This is likely to crash in most cases, but -Wformat-nonliteral technically
-  // doesn't warn in this case.
-  printf(0); // no-warning
+  printf(0);
+  // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
 }
 
 // Handle functions with multiple format attributes.
diff --git a/clang/test/SemaCXX/format-strings-0x.cpp b/clang/test/SemaCXX/format-strings-0x.cpp
index 7d37f8276f29f..e0ca7a270c993 100644
--- a/clang/test/SemaCXX/format-strings-0x.cpp
+++ b/clang/test/SemaCXX/format-strings-0x.cpp
@@ -14,6 +14,7 @@ void f(char **sp, float *fp) {
   printf("%a", 1.0);
   scanf("%afoobar", fp);
   printf(nullptr);
+  // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
   printf(*sp); // expected-warning {{not a string literal}}
   // expected-note@-1{{treat the string as an argument to avoid this}}
 
@@ -32,4 +33,5 @@ void f(char **sp, float *fp) {
   printf("init list: %d", { 0 }); // expected-error {{cannot pass initializer list to variadic function; expected type from format string was 'int'}}
   printf("void: %d", f(sp, fp)); // expected-error {{cannot pass expression of type 'void' to variadic function; expected type from format string was 'int'}}
   printf(0, { 0 }); // expected-error {{cannot pass initializer list to variadic function}}
+  // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
 }
diff --git a/clang/test/SemaObjC/format-strings-objc.m b/clang/test/SemaObjC/format-strings-objc.m
index 40c1d31b1fd4c..babbb40394267 100644
--- a/clang/test/SemaObjC/format-strings-objc.m
+++ b/clang/test/SemaObjC/format-strings-objc.m
@@ -130,7 +130,7 @@ void rdar10743758(id x) {
   printf(s2); // expected-warning {{more '%' conversions than data arguments}}
 
   const char * const s3 = (const char *)0;
-  printf(s3); // no-warning (NULL is a valid format string)
+  printf(s3); // expected-warning {{null passed to a callee that requires a non-null argument}}
 
   NSString * const ns1 = @"constant string %s"; // expected-note {{format string is defined here}}
   NSLog(ns1); // expected-warning {{more '%' conversions than data arguments}}
@@ -259,6 +259,7 @@ void testByValueObjectInFormat(Foo *obj) {
   printf("%d %d %d", 1L, *obj, 1L); // expected-error {{cannot pass object with interface type 'Foo' by value to variadic function; expected type from format string was 'int'}} expected-warning 2 {{format specifies type 'int' but the argument has type 'long'}}
   printf("%!", *obj); // expected-error {{cannot pass object with interface type 'Foo' by value through variadic function}} expected-warning {{invalid conversion specifier}}
   printf(0, *obj); // expected-error {{cannot pass object with interface type 'Foo' by value through variadic function}}
+  // expected-warning@-1{{null passed to a callee that requires a non-null argument}}
 
   [Bar log2:@"%d", *obj]; // expected-error {{cannot pass object with interface type 'Foo' by value to variadic method; expected type from format string was 'int'}}
 }

[bozicrHT]

@Sirraide @erichkeane

[erichkeane]

This is very much in @AaronBallman 's range here, particularly with the C stuff. I don't see anything of question, so it is just a matter of whether he thinks this is a good idea. he's supposed to come back any day now (maybe next week?) so please ping him in another week+ if you don't hear from him.

[philnik777]

For reference, this was #158626 before.