[clang] Warning Libc functions (PR #101583)

[Artem Dergachev via cfe-commits]

================
@@ -443,6 +448,368 @@ AST_MATCHER(ArraySubscriptExpr, isSafeArraySubscript) {
   return false;
 }
 
+namespace libc_fun_disjoint_inner_matchers {
+// `libc_fun_disjoint_inner_matchers` covers a set of matchers that match
+// disjoint node sets.   They all take a `CoreName`, which is the substring of a
+// function name after `ignoreLibcPrefixAndSuffix`.  They are suppose to be used
+// as an inner matcher of `ignoreLibcPrefixAndSuffix` to deal with different
+// libc function calls.
+
+// Matches a function call node such that
+//  1. It's name, after stripping off predefined prefix and suffix, is
+//     `CoreName`; and
+//  2. `CoreName` or `CoreName[str/wcs]` is one of the `PredefinedNames`, which
+//     is a set of libc function names.
+//
+//  Note: For predefined prefix and suffix, see `ignoreLibcPrefixAndSuffix`.
+//  And, the notation `CoreName[str/wcs]` means a new name obtained from replace
+//  string "wcs" with "str" in `CoreName`.
+//
+//  Also note, the set of predefined function names does not include `printf`
+//  functions, they are checked exclusively with other matchers below.
+//  Maintaining the invariant that all matchers under
+//  `libc_fun_disjoint_inner_matchers` are disjoint.
+AST_MATCHER_P(CallExpr, predefinedUnsafeLibcFunCall, StringRef, CoreName) {
+  static const std::set<StringRef> PredefinedNames{
+      // numeric conversion:
+      "atof",
+      "atoi",
+      "atol",
+      "atoll",
+      "strtol",
+      "strtoll",
+      "strtoul",
+      "strtoull",
+      "strtof",
+      "strtod",
+      "strtold",
+      "strtoimax",
+      "strtoumax",
+      // "strfromf",  "strfromd", "strfroml", // C23?
+      // string manipulation:
+      "strcpy",
+      "strncpy",
+      "strlcpy",
+      "strcat",
+      "strncat",
+      "strlcat",
+      "strxfrm",
+      "strdup",
+      "strndup",
+      // string examination:
+      "strlen",
+      "strnlen",
+      "strcmp",
+      "strncmp",
+      "stricmp",
+      "strcasecmp",
+      "strcoll",
+      "strchr",
+      "strrchr",
+      "strspn",
+      "strcspn",
+      "strpbrk",
+      "strstr",
+      "strtok",
+      // "mem-" functions
+      "memchr",
+      "wmemchr",
+      "memcmp",
+      "wmemcmp",
+      "memcpy",
+      "memccpy",
+      "mempcpy",
+      "wmemcpy",
+      "memmove",
+      "wmemmove",
+      "memset",
+      "wmemset",
+      // IO:
+      "fread",
+      "fwrite",
+      "fgets",
+      "fgetws",
+      "gets",
+      "fputs",
+      "fputws",
+      "puts",
+      // others
+      "strerror_s",
+      "strerror_r",
+      "bcopy",
+      "bzero",
+      "bsearch",
+      "qsort",
+  };
+  // This is safe: strlen("hello").  We don't want to be noisy on this case.
+  auto isSafeStrlen = [&Node](StringRef Name) -> bool {
+    return Name == "strlen" && Node.getNumArgs() == 1 &&
+           isa<StringLiteral>(Node.getArg(0)->IgnoreParenImpCasts());
+  };
+
+  // Match predefined names:
+  if (PredefinedNames.find(CoreName.str()) != PredefinedNames.end())
+    return !isSafeStrlen(CoreName);
+
+  std::string NameWCS = CoreName.str();
+  size_t WcsPos = NameWCS.find("wcs");
+
+  while (WcsPos != std::string::npos) {
+    NameWCS[WcsPos++] = 's';
+    NameWCS[WcsPos++] = 't';
+    NameWCS[WcsPos++] = 'r';
+    WcsPos = NameWCS.find("wcs", WcsPos);
+  }
+  if (PredefinedNames.find(NameWCS) != PredefinedNames.end())
+    return !isSafeStrlen(NameWCS);
+  // All `scanf` functions are unsafe (including `sscanf`, `vsscanf`, etc.. They
+  // all should end with "scanf"):
+  return CoreName.ends_with("scanf");
+}
+
+// Match a call to one of the `-printf` functions taking `va_list`.  We cannot
+// check safety for these functions so they should be changed to their
+// non-va_list versions.
+AST_MATCHER_P(CallExpr, unsafeVaListPrintfs, StringRef, CoreName) {
+  StringRef Name = CoreName;
+
+  if (!Name.ends_with("printf"))
+    return false; // neither printf nor scanf
+  return Name.starts_with("v");
+}
+
+// Matches a call to one of the `-sprintf` functions (excluding the ones with
+// va_list) as they are always unsafe and should be changed to corresponding
+// `-snprintf`s.
+AST_MATCHER_P(CallExpr, unsafeSprintfs, StringRef, CoreName) {
+  StringRef Name = CoreName;
+
+  if (!Name.ends_with("printf") || Name.starts_with("v"))
+    return false;
+
+  StringRef Prefix = Name.drop_back(6);
+
+  if (Prefix.ends_with("w"))
+    Prefix = Prefix.drop_back(1);
+  return Prefix == "s";
+}
+
+// A pointer type expression is known to be null-terminated, if it has the
+// form: E.c_str(), for any expression E of `std::string` type.
+static bool isNullTermPointer(const Expr *Ptr) {
+  if (isa<StringLiteral>(Ptr->IgnoreParenImpCasts()))
+    return true;
+  if (isa<PredefinedExpr>(Ptr->IgnoreParenImpCasts()))
----------------
haoNoQ wrote:

I just looked up `PredefinedExpr` on Doxygen and was very confused about what did they mean by "A predefined identifier such as __func__". Turns out they meant `__func__` but it got markdown'd 😄

Yeah nice one, this has to be safe.

https://github.com/llvm/llvm-project/pull/101583