Fix IndexSVSVamana null-deref in deserialize_impl (#5029)

Summary:
Pull Request resolved: https://github.com/facebookresearch/faiss/pull/5029

1. **Null pointer dereference in `deserialize_impl`**:
   `IndexSVSVamana::deserialize_impl()` and `IndexSVSFlat::deserialize_impl()`
   called `impl->load(...)` as a member function on a null `impl` pointer,
   causing a SIGSEGV. The `load` function is actually a static factory method
   (like `build` and `destroy`), so it should be called as
   `ClassName::load(&impl, ...)` instead. This matches the pattern used in
   `IndexSVSVamanaLeanVec::deserialize_training_data()` which correctly calls
   `svs_runtime::LeanVecTrainingData::load(&tdata, in)`.

2. **Added post-load validation**:
   Added `FAISS_THROW_IF_NOT_MSG(impl, ...)` after `load()` returns to ensure
   the impl pointer was actually populated.

3. **Added tests**:
   Added deserialization tests that provide invalid SVS stream data and verify
   the load fails with a FaissException rather than crashing.

Reviewed By: mnorris11

Differential Revision: D99117347

fbshipit-source-id: 3165048d06dc5ab66fcf83ecba3a144605c5732f

faiss/svs/IndexSVSFlat.cpp

@@ -107,11 +107,11 @@ void IndexSVSFlat::serialize_impl(std::ostream& out) const {
 void IndexSVSFlat::deserialize_impl(std::istream& in) {
     FAISS_THROW_IF_MSG(impl, "Cannot deserialize: SVS index already loaded.");
     auto metric = to_svs_metric(metric_type);
-    auto status = impl->load(&impl, in, metric);
+    auto status = svs_runtime::FlatIndex::load(&impl, in, metric);
     if (!status.ok()) {
         FAISS_THROW_MSG(status.message());
     }
-    FAISS_THROW_IF_NOT(impl);
+    FAISS_THROW_IF_NOT_MSG(impl, "Failed to load SVS Flat index.");
 }
 
 } // namespace faiss

faiss/svs/IndexSVSVamana.cpp

@@ -236,10 +236,12 @@ void IndexSVSVamana::deserialize_impl(std::istream& in) {
     FAISS_THROW_IF_MSG(impl, "Cannot deserialize: SVS index already loaded.");
     auto svs_metric = to_svs_metric(metric_type);
     auto svs_storage_kind = to_svs_storage_kind(storage_kind);
-    auto status = impl->load(&impl, in, svs_metric, svs_storage_kind);
+    auto status = svs_runtime::DynamicVamanaIndex::load(
+            &impl, in, svs_metric, svs_storage_kind);
     if (!status.ok()) {
         FAISS_THROW_MSG(status.message());
     }
+    FAISS_THROW_IF_NOT_MSG(impl, "Failed to load SVS Vamana index.");
 }
 
 } // namespace faiss

tests/test_read_index_deserialize.cpp

@@ -2966,3 +2966,80 @@ TEST(ReadIndexDeserialize, IndexRQFastScanAQDimensionMismatch) {
 
     expect_read_throws_with(buf, "does not match index d");
 }
+
+// ============================================================
+// SVS fourcc rejection / deserialization safety (Group F: T262015608)
+// ============================================================
+
+#ifdef FAISS_ENABLE_SVS
+
+// When SVS is enabled, deserializing an SVS Vamana index with invalid SVS
+// stream data should throw a FaissException (from the SVS runtime load
+// failure) rather than crashing with a null-pointer dereference.
+// Previously, deserialize_impl called impl->load() on a null impl pointer.
+TEST(ReadIndexDeserialize, SVSVamanaInvalidStreamThrows) {
+    std::vector<uint8_t> buf;
+    push_fourcc(buf, "ISVD");
+    push_index_header(buf, 8, 0);
+    // SVS Vamana deserialization fields:
+    push_val<size_t>(buf, 32);  // graph_max_degree
+    push_val<float>(buf, 1.2f); // alpha
+    push_val<size_t>(buf, 10);  // search_window_size
+    push_val<size_t>(buf, 10);  // search_buffer_capacity
+    push_val<size_t>(buf, 64);  // construction_window_size
+    push_val<size_t>(buf, 750); // max_candidate_pool_size
+    push_val<size_t>(buf, 28);  // prune_to
+    push_val<bool>(buf, false); // use_full_search_history
+    push_val<int>(buf, 0);      // storage_kind (SVS_Float16)
+    push_val<bool>(buf, true); // initialized = true → triggers deserialize_impl
+    // Provide garbage SVS stream data — load should fail gracefully.
+    for (int i = 0; i < 256; i++) {
+        push_val<uint8_t>(buf, 0);
+    }
+
+    // Should throw from SVS runtime load failure, NOT crash with SIGSEGV.
+    EXPECT_THROW(
+            {
+                auto reader = faiss::VectorIOReader();
+                reader.data = buf;
+                faiss::read_index(&reader);
+            },
+            faiss::FaissException);
+}
+
+// Same test for SVS Flat index.
+TEST(ReadIndexDeserialize, SVSFlatInvalidStreamThrows) {
+    std::vector<uint8_t> buf;
+    push_fourcc(buf, "ISVF");
+    push_index_header(buf, 8, 0);
+    push_val<bool>(buf, true); // initialized = true → triggers deserialize_impl
+    // Provide garbage SVS stream data.
+    for (int i = 0; i < 256; i++) {
+        push_val<uint8_t>(buf, 0);
+    }
+
+    EXPECT_THROW(
+            {
+                auto reader = faiss::VectorIOReader();
+                reader.data = buf;
+                faiss::read_index(&reader);
+            },
+            faiss::FaissException);
+}
+
+#else // !FAISS_ENABLE_SVS
+
+// When SVS is not enabled, attempting to read an index with an SVS fourcc
+// should fail with an "unknown fourcc" error rather than crashing.
+TEST(ReadIndexDeserialize, SVSVamanaFourccRejected) {
+    std::vector<uint8_t> buf;
+    push_fourcc(buf, "ISVD");
+    push_index_header(buf, 8, 0);
+    for (int i = 0; i < 128; i++) {
+        push_val<uint8_t>(buf, 0);
+    }
+
+    expect_read_throws_with(buf, "fourcc");
+}
+
+#endif // FAISS_ENABLE_SVS