[libc] Adds fuzz test for strstr and alphabetizes string fuzz CMakeList.

[sivachandra]

libc/fuzzing/string/strstr_fuzz.cpp

23

I am not sure if we should use another implementation of strstr to test the real implementation. Instead, can we use strcmp in a brute force manner as strcmp has its own fuzz test.

auto result = __llvm_libc::strstr(haystack, needle);
// Because of the construction below, the input for which result is NULL can be tested separately.
// For the rest:
if (strcmp(result, needle) != 0) {
  __builtin_trap():
}
auto *haystack_ptr = haystack;
for (;haystack_ptr != result; ++haystack)
    if (strcmp(haystack_ptr, needle) == 0) {
      // An earlier occurance of needle was missed by strstr.
      __builtin_trap();
    }
}

This is brute force and calls strcmp for every byte, but avoids reimplementing strstr. WDYT?

[cgyurgyik]

libc/fuzzing/string/strstr_fuzz.cpp

23

PTAL and let me know if this is what you're looking for.

I had to make some slight modifications since the result returned will include the rest of the string, and thus strcmp() will not always work.

[cgyurgyik]

libc/fuzzing/string/strstr_fuzz.cpp

23

Hmm nevermind I need to make some changes to this.

[cgyurgyik]

libc/fuzzing/string/strstr_fuzz.cpp

23

As discussed, strcmp won't quite work here. Created a helper function that mimics memcmp by comparing each letter up to N.

[sivachandra]

Few nits and questions inline but overall LGTM.

libc/fuzzing/string/strstr_fuzz.cpp

21

LLVM style is to use static functions over anonymous namespaces: https://llvm.org/docs/CodingStandards.html#anonymous-namespaces

21

May be call it simple_memcmp.

64

data1 should be copied so that we can add the null terminator. But, why should we copy into data2? Can we not use the input directly?

[cgyurgyik]

libc/fuzzing/string/strstr_fuzz.cpp

64

We can, I mostly did it to show there is no intersection between the two. Fixed so that only one new container is created.

libc/fuzzing/string/CMakeLists.txt

 add_libc_fuzzer(
+  strcmp_fuzz
+  SRCS
+    strcmp_fuzz.cpp
+  DEPENDS
+    libc.src.string.strcmp
+)
+
+add_libc_fuzzer(
   strcpy_fuzz
   SRCS
     strcpy_fuzz.cpp
   DEPENDS
     libc.src.string.memcpy
     libc.src.string.strcpy
     libc.src.string.strlen
 )
 
 add_libc_fuzzer(
-  strcmp_fuzz
+  strstr_fuzz
   SRCS
-    strcmp_fuzz.cpp
+    strstr_fuzz.cpp
   DEPENDS
-    libc.src.string.strcmp
+    libc.src.string.strstr
+    libc.src.string.strlen
 )

libc/fuzzing/string/strstr_fuzz.cpp

+//===-- strstr_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 strstr implementation.
+///
+//===----------------------------------------------------------------------===//
+
+#include "src/string/strlen.h"
+#include "src/string/strstr.h"
+#include <stddef.h>
+#include <stdint.h>
+
+namespace {
+
+// Simple loop to compare two strings up to a size n.
+int raw_strcmp(const char *left, const char *right, size_t n) {
+  for (; n && *left == *right; ++left, ++right, --n)
+    ;
+  return n ? *left - *right : 0;
+}
+
+} // namespace
+
+// The general structure is to take the value of the first byte, set size1 to
+// that value, and add the null terminator. size2 will then contain the rest of
+// the bytes in data.
+// For example, with inputs (data={2, 6, 4, 8, 0}, size=5):
+//         size1: data[0] = 2
+//         data1: {2, 6} + '\0' = {2, 6, '\0'}
+//         size2: size - size1 = 3
+//         data2: {4, 8, '\0'}
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+  // Verify the size is at least 1 and the data is null terminated.
+  if (!size || data[size - 1] != '\0')
+    return 0;
+
+  const size_t size1 = (data[0] <= size ? data[0] : size);
+  const size_t size2 = size - size1;
+
+  // The first size will always be at least 1 since
+  // we need to append the null terminator. The second size
+  // needs to be checked since it must also contain the null
+  // terminator.
+  if (!size2)
+    return 0;
+
+  // Copy the data into new containers.
+  // Add one to data1 for null terminator.
+  uint8_t *data1 = new uint8_t[size1 + 1];
+  uint8_t *data2 = new uint8_t[size2];
+  if (!data1 || !data2)
+    __builtin_trap();
+
+  size_t i;
+  for (i = 0; i < size1; ++i)
+    data1[i] = data[i];
+  data1[size1] = '\0'; // Add null terminator to data1.
+
+  for (size_t j = 0; j < size2; ++j)
+    data2[j] = data[i++];
+
+  const char *needle = reinterpret_cast<const char *>(data1);
+  const char *haystack = reinterpret_cast<const char *>(data2);
+  const char *result = __llvm_libc::strstr(haystack, needle);
+
+  // A null terminator may exist earlier.
+  const size_t haystack_size = __llvm_libc::strlen(haystack);
+  const size_t needle_size = __llvm_libc::strlen(needle);
+
+  if (result) {
+    // The needle is in the haystack.
+    // 1. Verify that the result matches the needle.
+    if (raw_strcmp(needle, result, needle_size) != 0)
+      __builtin_trap();
+
+    const char *haystack_ptr = haystack;
+    // 2. Verify that the result is the first occurrence of the needle.
+    for (; haystack_ptr != result; ++haystack_ptr) {
+      if (raw_strcmp(needle, haystack_ptr, needle_size) == 0)
+        __builtin_trap(); // There was an earlier occurrence of the needle.
+    }
+  } else {
+    // No result was found. Verify that the needle doesn't exist within the
+    // haystack.
+    for (size_t i = 0; i + needle_size < haystack_size; ++i) {
+      if (raw_strcmp(needle, haystack + i, needle_size) == 0)
+        __builtin_trap(); // There was an earlier occurrence of the needle.
+    }
+  }
+  delete[] data1;
+  delete[] data2;
+  return 0;
+}