add fuzzer for padded_string (#1312)

This also fixes an overflow problem.

.github/workflows/fuzzers.yml

@@ -15,9 +15,9 @@ jobs:
     runs-on: ubuntu-latest
     env:
       # fuzzers that change behaviour with SIMDJSON_FORCE_IMPLEMENTATION
-      defaultimplfuzzers: atpointer dump dump_raw_tape element minify  parser print_json
+      defaultimplfuzzers: atpointer dump dump_raw_tape element minify parser print_json
       # fuzzers that loop over the implementations themselves, or don't need to switch.
-      implfuzzers: implementations minifyimpl ondemand utf8
+      implfuzzers: implementations minifyimpl ondemand padded utf8
       implementations: haswell westmere fallback
       UBSAN_OPTIONS: halt_on_error=1
       MAXLEN: -max_len=4000

fuzz/CMakeLists.txt

@@ -58,6 +58,7 @@ if(ENABLE_FUZZING)
   implement_fuzzer(fuzz_minify)          # minify *with* parsing
   implement_fuzzer(fuzz_minifyimpl)      # minify *without* parsing, plus compare implementations
   implement_fuzzer(fuzz_ondemand)
+  implement_fuzzer(fuzz_padded)
   implement_fuzzer(fuzz_parser)
   implement_fuzzer(fuzz_print_json)
   implement_fuzzer(fuzz_utf8)            # utf8 verification, compares across implementations

fuzz/build_fuzzer_variants.sh

@@ -24,7 +24,8 @@ fi
 set -u
 
 # common options
-COMMON="-GNinja -DCMAKE_CXX_COMPILER=clang++$CLANGSUFFIX -DCMAKE_C_COMPILER=clang$CLANGSUFFIX -DSIMDJSON_BUILD_STATIC=Off -DENABLE_FUZZING=On -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_GIT=Off -DSIMDJSON_DISABLE_DEPRECATED_API=On"
+CXX_CLAGS_COMMON=-DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+COMMON="-GNinja -DCMAKE_CXX_COMPILER=clang++$CLANGSUFFIX -DCMAKE_C_COMPILER=clang$CLANGSUFFIX -DSIMDJSON_BUILD_STATIC=Off -DENABLE_FUZZING=On -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_DISABLE_DEPRECATED_API=On"
 
 # A replay build, as plain as it gets. For use with valgrind/gdb.
 variant=replay
@@ -63,7 +64,7 @@ variant=sanitizers-O3
 	cd build-$variant
 	cmake .. \
 	      $COMMON \
-	      -DCMAKE_CXX_FLAGS="-O3 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined" \
+	      -DCMAKE_CXX_FLAGS="-O3 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined $CXX_CLAGS_COMMON" \
 	      -DCMAKE_C_FLAGS="-O3 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined" \
 	      -DCMAKE_BUILD_TYPE=Debug \
 	      -DSIMDJSON_FUZZ_LINKMAIN=Off \
@@ -81,7 +82,7 @@ variant=sanitizers-O0
 	cd build-$variant
 	cmake .. \
 	      $COMMON \
-	      -DCMAKE_CXX_FLAGS="-O0 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined" \
+	      -DCMAKE_CXX_FLAGS="-O0 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined $CXX_CLAGS_COMMON" \
 	      -DCMAKE_C_FLAGS="-O0 -fsanitize=fuzzer-no-link,address,undefined -fno-sanitize-recover=undefined" \
 	      -DCMAKE_BUILD_TYPE=Debug \
 	      -DSIMDJSON_FUZZ_LINKMAIN=Off \
@@ -101,7 +102,7 @@ variant=fast
 
 	cmake .. \
 	      $COMMON \
-	      -DCMAKE_CXX_FLAGS="-fsanitize=fuzzer-no-link" \
+	      -DCMAKE_CXX_FLAGS="-fsanitize=fuzzer-no-link $CXX_CLAGS_COMMON" \
 	      -DCMAKE_C_FLAGS="-fsanitize=fuzzer-no-link" \
 	      -DCMAKE_BUILD_TYPE=Release \
 	      -DSIMDJSON_FUZZ_LINKMAIN=Off \

fuzz/build_like_ossfuzz.sh

@@ -14,7 +14,7 @@ export OUT=$(pwd)/ossfuzz-out
 export CC=clang
 export CXX="clang++"
 export CFLAGS="-fsanitize=fuzzer-no-link"
-export CXXFLAGS="-fsanitize=fuzzer-no-link,address,undefined -O1"
+export CXXFLAGS="-fsanitize=fuzzer-no-link,address,undefined -O1 -DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION"
 export LIB_FUZZING_ENGINE="-fsanitize=fuzzer"
 
 $ossfuzz

fuzz/fuzz_padded.cpp

@@ -0,0 +1,54 @@
+#include "FuzzUtils.h"
+#include "simdjson.h"
+#include <cstddef>
+#include <cstdint>
+#include <fstream>
+#include <string>
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
+  FuzzData fd(Data, Size);
+
+  const int action = fd.getInt<0, 10>();
+
+  const auto s1 = fd.get<size_t>();
+  const auto s2 = fd.get<size_t>();
+  switch (action) {
+  case 0: {
+    simdjson_unused simdjson::padded_string p(s1);
+  } break;
+  case 1: {
+    // operator== with temp value
+    simdjson_unused simdjson::padded_string p1(s1);
+    simdjson_unused simdjson::padded_string p2(s2);
+    p1 = std::move(p2);
+  } break;
+  case 2: {
+    // swap
+    simdjson_unused simdjson::padded_string p1(s1);
+    simdjson_unused simdjson::padded_string p2(s2);
+    p1.swap(p2);
+  } break;
+  case 3: {
+    simdjson_unused simdjson::padded_string p(s1);
+    p.data();
+  } break;
+  case 4: {
+    simdjson::padded_string p(s1);
+    simdjson_unused auto sv = static_cast<std::string_view>(p);
+  } break;
+  case 5: {
+    // load from file.
+    const std::string filename = "/dev/shm/fuzz_padded.tmp";
+    {
+      std::ofstream file(filename);
+      assert(file);
+      const long ssize = static_cast<long>(fd.Size);
+      file.write(fd.chardata(), ssize);
+      assert(file.tellp() == ssize);
+    }
+    simdjson_unused auto data = simdjson::padded_string::load(filename);
+  } break;
+  default:;
+  }
+  return 0;
+}

include/simdjson/padded_string-inl.h

@@ -19,7 +19,18 @@ namespace internal {
 // The length parameter is the maximum size in bytes of the string.
 // The caller is responsible to free the memory (e.g., delete[] (...)).
 inline char *allocate_padded_buffer(size_t length) noexcept {
-  size_t totalpaddedlength = length + SIMDJSON_PADDING;
+  const size_t totalpaddedlength = length + SIMDJSON_PADDING;
+  if(totalpaddedlength<length) {
+    // overflow
+    return nullptr;
+  }
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+  // avoid getting out of memory
+  if (totalpaddedlength>(1UL<<20)) {
+    return nullptr;
+  }
+#endif
+
   char *padded_buffer = new (std::nothrow) char[totalpaddedlength];
   if (padded_buffer == nullptr) {
     return nullptr;