@cryptotaxi247 / netdata-1 / commits / fc83a2e9e

Release memory on shutdown - detect invalid extent in journal files (#19861)

* Release all memory for metasync thread * Cleanup * Validate extent index in journal entries to prevent invalid access * Proper free * Improve error logging for invalid extent index in journal file

Stelios Fragkakis committed Mar 14, 2025 at 23:53 UTC fc83a2e9e7503afa4758d64315862a5cd6f8d34a

2 files changed +21 -6

src/database/engine/pagecache.c

@@ -530,6 +530,7 @@ static ALWAYS_INLINE_HOT size_t get_page_list_from_journal_v2(struct rrdengine_i

530 struct journal_page_header *page_list_header = (struct journal_page_header *) ((uint8_t *) j2_header + uuid_entry->page_offset);

531 struct journal_page_list *page_list = (struct journal_page_list *)((uint8_t *) page_list_header + sizeof(*page_list_header));

532 struct journal_extent_list *extent_list = (void *)((uint8_t *)j2_header + j2_header->extent_offset);

533 + uint32_t extent_entries = j2_header->extent_count;

534 uint32_t uuid_page_entries = page_list_header->entries;

535

536 for (uint32_t index = 0; index < uuid_page_entries; index++) {

@@ -545,6 +546,13 @@ static ALWAYS_INLINE_HOT size_t get_page_list_from_journal_v2(struct rrdengine_i

546 if(prc == PAGE_IS_IN_THE_FUTURE)

547 break;

548

549 + // Make sure index is valid for this file

550 + if (page_entry_in_journal->extent_index > extent_entries) {

551 + nd_log_limit_static_thread_var(erl, 60, 0);

552 + nd_log_limit(&erl, NDLS_DAEMON, NDLP_ERR, "DBENGINE: Invalid extent index in journalfile %u", datafile->fileno);

553 + break;

554 + }

555 +

556 uint32_t page_update_every_s = page_entry_in_journal->update_every_s;

557 size_t page_length = page_entry_in_journal->page_length;

558

src/database/sqlite/sqlite_metadata.c

+13 -6

@@ -2417,6 +2417,15 @@ static void start_metadata_hosts(uv_work_t *req)

2417 worker_is_idle();

2418 }

2419

2420 +static void close_callback(uv_handle_t *handle, void *data __maybe_unused)

2421 +{

2422 + if (handle->type == UV_TIMER) {

2423 + uv_timer_stop((uv_timer_t *)handle);

2424 + }

2425 +

2426 + uv_close(handle, NULL); // Automatically close and free the handle

2427 +}

2428 +

2429 static void metadata_event_loop(void *arg)

2430 {

2431 worker_register("METASYNC");

@@ -2430,7 +2439,6 @@ static void metadata_event_loop(void *arg)

2439 worker_register_job_name(METADATA_DEL_HOST_AE, "delete host alert entry");

2440

2441 int ret;

2433 - uv_loop_t *loop;

2442 unsigned cmd_batch_size;

2443 struct metadata_wc *wc = arg;

2444 enum metadata_opcode opcode;

@@ -2438,12 +2446,13 @@ static void metadata_event_loop(void *arg)

2446 wc->ar = aral_by_size_acquire(sizeof(struct metadata_cmd));

2447

2448 uv_thread_set_name_np("METASYNC");

2441 - loop = wc->loop = mallocz(sizeof(uv_loop_t));

2449 + uv_loop_t *loop = wc->loop = mallocz(sizeof(uv_loop_t));

2450 ret = uv_loop_init(loop);

2451 if (ret) {

2452 netdata_log_error("uv_loop_init(): %s", uv_strerror(ret));

2453 goto error_after_loop_init;

2454 }

2455 +

2456 loop->data = wc;

2457

2458 ret = uv_async_init(wc->loop, &wc->async, async_cb);

@@ -2619,17 +2628,15 @@ static void metadata_event_loop(void *arg)

2628 } while (opcode != METADATA_DATABASE_NOOP);

2629 }

2630

2622 - if (!uv_timer_stop(&wc->timer_req))

2623 - uv_close((uv_handle_t *)&wc->timer_req, NULL);

2631 + uv_walk(loop, (uv_walk_cb) close_callback, NULL);

2632 + uv_run(loop, UV_RUN_NOWAIT);

2633

2625 - uv_close((uv_handle_t *)&wc->async, NULL);

2634 int rc;

2635 do {

2636 rc = uv_loop_close(loop);

2637 } while (rc != UV_EBUSY);

2638

2639 buffer_free(work_buffer);

2632 - freez(loop);

2640 worker_unregister();

2641

2642 nd_log(NDLS_DAEMON, NDLP_DEBUG, "Shutting down metadata thread");