[llvm] 40d3949 - [CAS] Add llvm-cas-fuzzer for ObjectStore::validate() (#190635)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 6 13:31:56 PDT 2026
Author: Steven Wu
Date: 2026-04-06T13:31:51-07:00
New Revision: 40d394916264903073bccdf0b208f9565cad3ef8
URL: https://github.com/llvm/llvm-project/commit/40d394916264903073bccdf0b208f9565cad3ef8
DIFF: https://github.com/llvm/llvm-project/commit/40d394916264903073bccdf0b208f9565cad3ef8.diff
LOG: [CAS] Add llvm-cas-fuzzer for ObjectStore::validate() (#190635)
Add a fuzzer that creates an on-disk CAS database, stores objects, then
corrupts the on-disk data files using fuzzer-provided bytes and calls
validate(). The goal is that validate() should either succeed or return
an error, never crash.
The fuzzer supports 6 corruption modes: byte-level mutations, file
truncation, appending garbage, zeroing ranges, standalone file
corruption, and combined mutations with continued CAS operations.
Assisted-By: Claude
Added:
llvm/tools/llvm-cas-fuzzer/CMakeLists.txt
llvm/tools/llvm-cas-fuzzer/DummyCASFuzzer.cpp
llvm/tools/llvm-cas-fuzzer/cas-fuzzer.cpp
Modified:
Removed:
################################################################################
diff --git a/llvm/tools/llvm-cas-fuzzer/CMakeLists.txt b/llvm/tools/llvm-cas-fuzzer/CMakeLists.txt
new file mode 100644
index 0000000000000..bf4446f770208
--- /dev/null
+++ b/llvm/tools/llvm-cas-fuzzer/CMakeLists.txt
@@ -0,0 +1,10 @@
+set(LLVM_LINK_COMPONENTS
+ CAS
+ Support
+ FuzzerCLI
+)
+
+add_llvm_fuzzer(llvm-cas-fuzzer
+ cas-fuzzer.cpp
+ DUMMY_MAIN DummyCASFuzzer.cpp
+)
diff --git a/llvm/tools/llvm-cas-fuzzer/DummyCASFuzzer.cpp b/llvm/tools/llvm-cas-fuzzer/DummyCASFuzzer.cpp
new file mode 100644
index 0000000000000..2f61c9afc8da5
--- /dev/null
+++ b/llvm/tools/llvm-cas-fuzzer/DummyCASFuzzer.cpp
@@ -0,0 +1,14 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/FuzzMutate/FuzzerCLI.h"
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size);
+int main(int argc, char *argv[]) {
+ return llvm::runFuzzerOnInputs(argc, argv, LLVMFuzzerTestOneInput);
+}
diff --git a/llvm/tools/llvm-cas-fuzzer/cas-fuzzer.cpp b/llvm/tools/llvm-cas-fuzzer/cas-fuzzer.cpp
new file mode 100644
index 0000000000000..1508a1d57559a
--- /dev/null
+++ b/llvm/tools/llvm-cas-fuzzer/cas-fuzzer.cpp
@@ -0,0 +1,387 @@
+//===-- cas-fuzzer.cpp - Fuzzer for CAS ObjectStore::validate() -----------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Fuzzer for on-disk CAS validation. Creates a valid CAS database, stores
+// objects, corrupts the on-disk files using fuzzer-provided bytes, then calls
+// validate(). The invariant: validate() must either succeed or return an error,
+// never crash.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/ScopeExit.h"
+#include "llvm/CAS/ActionCache.h"
+#include "llvm/CAS/BuiltinUnifiedCASDatabases.h"
+#include "llvm/CAS/ObjectStore.h"
+#include "llvm/Support/Error.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/Path.h"
+#include "llvm/Support/ScopedPrinter.h"
+#include "llvm/Support/raw_ostream.h"
+#include <cstdint>
+#include <cstring>
+
+using namespace llvm;
+using namespace llvm::cas;
+
+namespace {
+
+/// Read a little-endian uint32 from Data, or 0 if not enough bytes.
+static uint32_t readU32(ArrayRef<uint8_t> Data, size_t Offset) {
+ if (Offset + sizeof(uint32_t) > Data.size())
+ return 0;
+ return support::endian::read32le(Data.data() + Offset);
+}
+
+/// Read a little-endian uint16 from Data, or 0 if not enough bytes.
+static uint16_t readU16(ArrayRef<uint8_t> Data, size_t Offset) {
+ if (Offset + sizeof(uint16_t) > Data.size())
+ return 0;
+ return support::endian::read16le(Data.data() + Offset);
+}
+
+/// Find the versioned subdirectory (v1.N) inside the CAS root.
+static std::string findVersionedSubdir(StringRef CASDir) {
+ std::error_code EC;
+ std::string Best;
+ uint64_t BestOrder = 0;
+ for (sys::fs::directory_iterator DirI(CASDir, EC), DirE; !EC && DirI != DirE;
+ DirI.increment(EC)) {
+ if (DirI->type() != sys::fs::file_type::directory_file)
+ continue;
+ StringRef Name = sys::path::filename(DirI->path());
+ if (!Name.starts_with("v1."))
+ continue;
+ uint64_t Order;
+ if (Name.substr(3).getAsInteger(10, Order))
+ continue;
+ if (Best.empty() || Order > BestOrder) {
+ Best = DirI->path();
+ BestOrder = Order;
+ }
+ }
+ return Best;
+}
+
+/// Collect paths of files matching a prefix in a directory.
+static void collectFilesWithPrefix(StringRef Dir, StringRef Prefix,
+ SmallVectorImpl<std::string> &Results) {
+ std::error_code EC;
+ for (sys::fs::directory_iterator DirI(Dir, EC), DirE; !EC && DirI != DirE;
+ DirI.increment(EC)) {
+ StringRef Name = sys::path::filename(DirI->path());
+ if (Name.starts_with(Prefix))
+ Results.push_back(DirI->path());
+ }
+}
+
+/// Read an entire file into a buffer.
+static bool readFileBytes(StringRef Path, SmallVectorImpl<char> &Buf) {
+ auto MBOrErr = MemoryBuffer::getFile(Path, /*IsText=*/false,
+ /*RequiresNullTerminator=*/false);
+ if (!MBOrErr)
+ return false;
+ Buf.assign((*MBOrErr)->getBufferStart(), (*MBOrErr)->getBufferEnd());
+ return true;
+}
+
+/// Write buffer contents to a file, replacing it entirely.
+static bool writeFileBytes(StringRef Path, ArrayRef<char> Buf) {
+ std::error_code EC;
+ raw_fd_ostream OS(Path, EC, sys::fs::OF_None);
+ if (EC)
+ return false;
+ OS.write(Buf.data(), Buf.size());
+ return !OS.has_error();
+}
+
+/// Create a CAS database and store some baseline objects.
+/// Returns true on success, populating CAS and AC via output parameters.
+static bool createAndPopulateCAS(StringRef TmpDir,
+ std::unique_ptr<ObjectStore> &CAS,
+ std::unique_ptr<ActionCache> &AC) {
+ auto Result = createOnDiskUnifiedCASDatabases(TmpDir);
+ if (!Result) {
+ consumeError(Result.takeError());
+ return false;
+ }
+ CAS = std::move(Result->first);
+ AC = std::move(Result->second);
+
+ // Store a leaf node (no refs, small data).
+ const char LeafData[] = "hello-cas-fuzzer-leaf-data";
+ auto Leaf = CAS->store({}, arrayRefFromStringRef<char>(LeafData));
+ if (!Leaf) {
+ consumeError(Leaf.takeError());
+ return false;
+ }
+
+ // Store a node with a ref to the leaf.
+ const char NodeData[] = "node-with-one-ref";
+ auto Node1 = CAS->store({*Leaf}, arrayRefFromStringRef<char>(NodeData));
+ if (!Node1) {
+ consumeError(Node1.takeError());
+ return false;
+ }
+
+ // Store a node referencing both previous nodes.
+ const char Node2Data[] = "node-with-two-refs";
+ auto Node2 =
+ CAS->store({*Leaf, *Node1}, arrayRefFromStringRef<char>(Node2Data));
+ if (!Node2) {
+ consumeError(Node2.takeError());
+ return false;
+ }
+
+ // Store a larger data node to potentially exercise
diff erent size encodings.
+ std::string LargeData(4096, 'X');
+ auto LargeNode =
+ CAS->store({}, arrayRefFromStringRef<char>(StringRef(LargeData)));
+ if (!LargeNode) {
+ consumeError(LargeNode.takeError());
+ return false;
+ }
+
+ return true;
+}
+
+/// Apply byte-level mutations to a file.
+static void applyByteMutations(StringRef Path, ArrayRef<uint8_t> Data) {
+ SmallVector<char> Buf;
+ if (!readFileBytes(Path, Buf) || Buf.empty())
+ return;
+
+ // Parse as 7-byte chunks: [offset(4)][op(1)][value(1)][unused(1)]
+ for (size_t I = 0; I + 6 <= Data.size(); I += 7) {
+ uint32_t Offset = readU32(Data, I) % Buf.size();
+ uint8_t Op = Data[I + 4] % 3;
+ uint8_t Value = Data[I + 5];
+ switch (Op) {
+ case 0: // XOR
+ Buf[Offset] ^= Value;
+ break;
+ case 1: // SET
+ Buf[Offset] = Value;
+ break;
+ case 2: // Zero
+ Buf[Offset] = 0;
+ break;
+ }
+ }
+ writeFileBytes(Path, Buf);
+}
+
+/// Truncate a file to a given fraction of its size.
+static void truncateFile(StringRef Path, uint8_t Fraction) {
+ SmallVector<char> Buf;
+ if (!readFileBytes(Path, Buf) || Buf.empty())
+ return;
+ // Fraction is 0-255, map to 0-100% of file size.
+ size_t NewSize =
+ static_cast<size_t>(static_cast<uint64_t>(Buf.size()) * Fraction / 255);
+ // Don't zero out the size.
+ if (NewSize == 0)
+ NewSize = 1;
+ Buf.resize(NewSize);
+ writeFileBytes(Path, Buf);
+}
+
+/// Append garbage bytes to a file.
+static void appendGarbage(StringRef Path, ArrayRef<uint8_t> Data) {
+ SmallVector<char> Buf;
+ if (!readFileBytes(Path, Buf))
+ return;
+ Buf.append(Data.begin(), Data.end());
+ writeFileBytes(Path, Buf);
+}
+
+/// Zero out a range in a file.
+static void zeroRange(StringRef Path, uint32_t Offset, uint16_t Length) {
+ SmallVector<char> Buf;
+ if (!readFileBytes(Path, Buf) || Buf.empty())
+ return;
+ size_t Start = Offset % Buf.size();
+ size_t End = std::min(Start + static_cast<size_t>(Length), Buf.size());
+ std::memset(Buf.data() + Start, 0, End - Start);
+ writeFileBytes(Path, Buf);
+}
+
+/// Corrupt standalone files (obj.*, leaf.*, leaf+0.*).
+static void corruptStandaloneFiles(StringRef SubDir, ArrayRef<uint8_t> Data) {
+ SmallVector<std::string> StandaloneFiles;
+ collectFilesWithPrefix(SubDir, "obj.", StandaloneFiles);
+ collectFilesWithPrefix(SubDir, "leaf.", StandaloneFiles);
+ collectFilesWithPrefix(SubDir, "leaf+0.", StandaloneFiles);
+
+ if (StandaloneFiles.empty())
+ return;
+
+ for (size_t I = 0; I < Data.size() && !StandaloneFiles.empty(); I += 3) {
+ size_t FileIdx = Data[I] % StandaloneFiles.size();
+ uint8_t Action = (I + 1 < Data.size()) ? Data[I + 1] % 4 : 0;
+ uint8_t Param = (I + 2 < Data.size()) ? Data[I + 2] : 128;
+
+ StringRef FilePath = StandaloneFiles[FileIdx];
+ switch (Action) {
+ case 0: // Delete the file
+ sys::fs::remove(FilePath);
+ break;
+ case 1: // Truncate
+ truncateFile(FilePath, Param);
+ break;
+ case 2: // Corrupt bytes
+ if (I + 3 < Data.size())
+ applyByteMutations(
+ FilePath, Data.slice(I + 3, std::min(Data.size() - I - 3,
+ static_cast<size_t>(21))));
+ break;
+ case 3: // Zero out beginning
+ zeroRange(FilePath, 0, Param);
+ break;
+ }
+ }
+}
+
+/// Select which data file to target (index.v1 or data.v1).
+static std::string selectTargetFile(StringRef SubDir, uint8_t Selector) {
+ SmallString<256> Path(SubDir);
+ if (Selector % 2 == 0)
+ sys::path::append(Path, "index.v1");
+ else
+ sys::path::append(Path, "data.v1");
+ return std::string(Path);
+}
+
+/// Try to exercise the CAS after corruption: store and load.
+static void exerciseCAS(ObjectStore &CAS) {
+ // Try storing a new object.
+ const char NewData[] = "post-corruption-data";
+ auto NewObj = CAS.store({}, arrayRefFromStringRef<char>(NewData));
+ if (!NewObj)
+ consumeError(NewObj.takeError());
+
+ // Try validate again with CheckHash=false.
+ if (auto E = CAS.validate(/*CheckHash=*/false))
+ consumeError(std::move(E));
+}
+
+} // end anonymous namespace
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
+ if (Size == 0)
+ return 0;
+
+ // Create a unique temp directory for this fuzzer run.
+ SmallString<256> TmpDir;
+ if (sys::fs::createUniqueDirectory("cas-fuzzer", TmpDir))
+ return 0;
+
+ // Ensure cleanup on exit.
+ auto Cleanup = scope_exit([&]() { sys::fs::remove_directories(TmpDir); });
+
+ // Step 1: Create and populate a valid CAS.
+ std::unique_ptr<ObjectStore> CAS;
+ std::unique_ptr<ActionCache> AC;
+ if (!createAndPopulateCAS(TmpDir, CAS, AC))
+ return 0;
+
+ // Step 2: Validate baseline - should succeed.
+ if (auto E = CAS->validate(/*CheckHash=*/true)) {
+ // If baseline validation fails, something is wrong with the setup.
+ consumeError(std::move(E));
+ return 0;
+ }
+
+ // Step 3: Close the CAS so files are unmapped.
+ CAS.reset();
+ AC.reset();
+
+ // Step 4: Find the versioned subdirectory.
+ std::string SubDir = findVersionedSubdir(TmpDir);
+ if (SubDir.empty())
+ return 0;
+
+ // Step 5: Apply corruption based on mode selector (first byte).
+ ArrayRef<uint8_t> Input(Data, Size);
+ uint8_t Mode = Input[0] % 6;
+ ArrayRef<uint8_t> Rest = Input.drop_front(1);
+
+ switch (Mode) {
+ case 0: { // Byte-level mutations
+ if (Rest.empty())
+ break;
+ std::string Target = selectTargetFile(SubDir, Rest[0]);
+ if (Rest.size() > 1)
+ applyByteMutations(Target, Rest.drop_front(1));
+ break;
+ }
+ case 1: { // File truncation
+ if (Rest.size() < 2)
+ break;
+ std::string Target = selectTargetFile(SubDir, Rest[0]);
+ truncateFile(Target, Rest[1]);
+ break;
+ }
+ case 2: { // Append garbage
+ if (Rest.empty())
+ break;
+ std::string Target = selectTargetFile(SubDir, Rest[0]);
+ if (Rest.size() > 1)
+ appendGarbage(Target, Rest.drop_front(1));
+ break;
+ }
+ case 3: { // Zero out a range
+ if (Rest.size() < 7)
+ break;
+ std::string Target = selectTargetFile(SubDir, Rest[0]);
+ uint32_t Offset = readU32(Rest, 1);
+ uint16_t Length = readU16(Rest, 5);
+ zeroRange(Target, Offset, Length);
+ break;
+ }
+ case 4: { // Standalone file corruption
+ corruptStandaloneFiles(SubDir, Rest);
+ break;
+ }
+ case 5: { // Combined: byte mutations + exercise CAS
+ if (Rest.empty())
+ break;
+ std::string Target = selectTargetFile(SubDir, Rest[0]);
+ if (Rest.size() > 1)
+ applyByteMutations(Target, Rest.drop_front(1));
+ break;
+ }
+ }
+
+ // Step 6: Reopen the CAS after corruption.
+ auto Reopened = createOnDiskUnifiedCASDatabases(TmpDir);
+ if (!Reopened) {
+ // Reopen failing is acceptable — corruption may have broken the format.
+ consumeError(Reopened.takeError());
+ return 0;
+ }
+ CAS = std::move(Reopened->first);
+ AC = std::move(Reopened->second);
+
+ // Step 7: Validate — must not crash.
+ bool ValidationFailed = false;
+ if (auto E = CAS->validate(/*CheckHash=*/true)) {
+ consumeError(std::move(E));
+ ValidationFailed = true;
+ }
+ if (auto E = CAS->validate(/*CheckHash=*/false)) {
+ consumeError(std::move(E));
+ ValidationFailed = true;
+ }
+
+ // Step 8: For mode 5, exercise the CAS only if validation passed.
+ if (Mode == 5 && !ValidationFailed)
+ exerciseCAS(*CAS);
+
+ return 0;
+}
More information about the llvm-commits
mailing list