[libc] Add fuzzing for printf floats

[michaelrj]

Summary

To guarantee accuracy for all potential float values, this patch adds a
fuzzer to compare the results for float conversions from our printf
against MPFR's.

[lntue]

libc/fuzzing/stdio/printf_float_conv_fuzz.cpp

38

Will using enums for the error codes be better here?

[michaelrj]

use an enum instead of int for the return value of test_vals

[lntue]

libc/fuzzing/stdio/printf_float_conv_fuzz.cpp

2–3

Nit: line format.

[michaelrj]

Fix the header and also several bugs found with the fuzzer

[michaelrj]

fix another fuzz-found bug. Given this I'm going to let the fuzzer run over the weekend before landing

[michaelrj]

fix another bug

[michaelrj]

more bugfixes and a new patch in the chain

[michaelrj]

A bunch more bugfixes, but it seems to be stable now. The fuzzer has ran for ~12 hours with no failure.

[sivachandra]

It would be helpful to actually leave a trail of descriptions of the mutually exclusive bugs that were identified. Also, prefer landing their fixes in separate patches (unless the fixes are trivial).

[michaelrj]

split out the fixes into separate changes:

https://reviews.llvm.org/D157534
https://reviews.llvm.org/D157535
https://reviews.llvm.org/D157536

libc/fuzzing/stdio/CMakeLists.txt

 add_libc_fuzzer(
   printf_parser_fuzz
   SRCS
     printf_parser_fuzz.cpp
   DEPENDS
     libc.src.stdio.printf_core.mock_parser
   COMPILE_OPTIONS
     -DLIBC_COPT_MOCK_ARG_LIST
 )
+
+add_libc_fuzzer(
+  printf_float_conv_fuzz
+  NEED_MPFR
+  SRCS
+    printf_float_conv_fuzz.cpp
+  DEPENDS
+    libc.src.stdio.snprintf
+    libc.src.__support.FPUtil.fp_bits
+)

libc/fuzzing/stdio/printf_float_conv_fuzz.cpp

+//===-- printf_float_conv_fuzz.cpp
+//-----------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+///
+/// Fuzzing test for llvm-libc printf %f/e/g/a implementations.
+///
+//===----------------------------------------------------------------------===//
+#include "src/stdio/snprintf.h"
+
+#include "src/__support/FPUtil/FPBits.h"
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "utils/MPFRWrapper/mpfr_inc.h"
+
+constexpr int MAX_SIZE = 10000;
+
+inline bool simple_streq(char *first, char *second, int length) {
+  for (int i = 0; i < length; ++i) {
+    if (first[i] != second[i]) {
+      return false;
+    }
+  }
+  return true;
+}
+
+inline int test_vals(const char *fmt, double num, int prec, int width) {
+  // Call snprintf on a nullptr to get the buffer size.
+  int buffer_size = snprintf(nullptr, 0, fmt, width, prec, num);
+
+  if (buffer_size < 0) {
+    return -1;
+  }
+
+  char *test_buff = new char[buffer_size + 1];
+  char *reference_buff = new char[buffer_size + 1];
+
+  int test_result = 0;
+  int reference_result = 0;
+
+  test_result = snprintf(test_buff, buffer_size + 1, fmt, width, prec, num);
+  reference_result =
+      mpfr_snprintf(reference_buff, buffer_size + 1, fmt, width, prec, num);
+
+  // All of these calls should return that they wrote the same amount.
+  if (test_result != reference_result || test_result != buffer_size) {
+    return -2;
+  }
+
+  if (!simple_streq(test_buff, reference_buff, buffer_size)) {
+    return -3;
+  }
+
+  delete[] test_buff;
+  delete[] reference_buff;
+  return 0;
+}
+
+constexpr char const *fmt_arr[] = {
+    "%*.*f",
+    "%*.*e",
+    "%*.*g",
+    "%*.*a",
+};
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+  double num = 0.0;
+  int prec = 0;
+  int width = 0;
+
+  __llvm_libc::fputil::FPBits<double>::UIntType raw_num = 0;
+
+  // Copy as many bytes of data as will fit into num, prec, and with. Any extras
+  // are ignored.
+  for (size_t cur = 0; cur < size; ++cur) {
+    if (cur < sizeof(raw_num)) {
+      raw_num = (raw_num << 8) + data[cur];
+    } else if (cur < sizeof(raw_num) + sizeof(prec)) {
+      prec = (prec << 8) + data[cur];
+    } else if (cur < sizeof(raw_num) + sizeof(prec) + sizeof(width)) {
+      width = (width << 8) + data[cur];
+    }
+  }
+
+  num = __llvm_libc::fputil::FPBits<double>(raw_num).get_val();
+
+  if (width > MAX_SIZE) {
+    width = MAX_SIZE;
+  } else if (width < -MAX_SIZE) {
+    width = -MAX_SIZE;
+  }
+
+  if (prec > MAX_SIZE) {
+    prec = MAX_SIZE;
+  } else if (prec < -MAX_SIZE) {
+    prec = -MAX_SIZE;
+  }
+
+  int result = 0;
+
+  for (size_t cur_fmt = 0; cur_fmt < sizeof(fmt_arr) / sizeof(char *);
+       ++cur_fmt) {
+    result = test_vals(fmt_arr[cur_fmt], num, prec, width);
+    if (result != 0) {
+      __builtin_trap();
+    }
+  }
+  return 0;
+}