test: add unit test for sds_check (#1)

+ minor adjustments to error/warning messages

Signed-off-by: Oksana Salyk <oksana.salyk@hpe.com>

Author: osalyk

src/common/pmemcommon.inc

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright 2017-2023, Intel Corporation
+# Copyright 2025, Hewlett Packard Enterprise Development LP
 #
 # src/pmemcommon.inc -- common SOURCE definitions for PMDK libraries
 #
@@ -21,6 +22,7 @@ SOURCE +=\
 	$(COMMON)/rand.c\
 	$(COMMON)/set.c\
 	$(COMMON)/shutdown_state.c\
+	$(COMMON)/shutdown_state_helper.c\
 	$(COMMON)/uuid.c\
 	$(COMMON)/uuid_linux.c\
 	$(PMEM2)/pmem2_utils.c\

src/common/shutdown_state.c

@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: BSD-3-Clause
 /* Copyright 2017-2024, Intel Corporation */
+/* Copyright 2025, Hewlett Packard Enterprise Development LP */
 
 /*
  * shutdown_state.c -- unsafe shudown detection
@@ -17,13 +18,10 @@
 #include "bad_blocks.h"
 #include "../libpmem2/pmem2_utils.h"
 
-#define FLUSH_SDS(sds, rep) \
-	if ((rep) != NULL) os_part_deep_common(rep, 0, sds, sizeof(*(sds)), 1)
-
 /*
- * shutdown_state_checksum -- (internal) counts SDS checksum and flush it
+ * shutdown_state_checksum -- counts SDS checksum and flush it
  */
-static void
+void
 shutdown_state_checksum(struct shutdown_state *sds, struct pool_replica *rep)
 {
 	LOG(3, "sds %p", sds);
@@ -159,24 +157,6 @@ shutdown_state_clear_dirty(struct shutdown_state *sds, struct pool_replica *rep)
 	shutdown_state_checksum(sds, rep);
 }
 
-/*
- * shutdown_state_reinit -- (internal) reinitializes shutdown_state struct
- */
-static void
-shutdown_state_reinit(struct shutdown_state *curr_sds,
-	struct shutdown_state *pool_sds, struct pool_replica *rep)
-{
-	LOG(3, "curr_sds %p, pool_sds %p", curr_sds, pool_sds);
-	shutdown_state_init(pool_sds, rep);
-	pool_sds->uuid = htole64(curr_sds->uuid);
-	pool_sds->usc = htole64(curr_sds->usc);
-	pool_sds->dirty = 0;
-
-	FLUSH_SDS(pool_sds, rep);
-
-	shutdown_state_checksum(pool_sds, rep);
-}
-
 /*
  * shutdown_state_check -- compares and fixes shutdown state
  */
@@ -186,15 +166,26 @@ shutdown_state_check(struct shutdown_state *curr_sds,
 {
 	LOG(3, "curr_sds %p, pool_sds %p", curr_sds, pool_sds);
 
+	/*
+	 * This is likely to occur only when the pool is being opened for
+	 * the first time after the SHUTDOWN_STATE feature has been enabled on
+	 * the pool, for example, via (lib)pmempool.
+	 * Please do not confuse this with establishing SDS during creation.
+	 */
 	if (util_is_zeroed(pool_sds, sizeof(*pool_sds)) &&
 			!util_is_zeroed(curr_sds, sizeof(*curr_sds))) {
+		CORE_LOG_WARNING(
+			"Enabling ADR failure detection, assuming pool consistency up to this point.");
 		shutdown_state_reinit(curr_sds, pool_sds, rep);
 		return 0;
 	}
 
+	bool is_uuid_correct =
+		le64toh(pool_sds->uuid) == le64toh(curr_sds->uuid);
+
 	bool is_uuid_usc_correct =
 		le64toh(pool_sds->usc) == le64toh(curr_sds->usc) &&
-		le64toh(pool_sds->uuid) == le64toh(curr_sds->uuid);
+		is_uuid_correct;
 
 	bool is_checksum_correct = util_checksum(pool_sds,
 		sizeof(*pool_sds), &pool_sds->checksum, 0, 0);
@@ -204,7 +195,7 @@ shutdown_state_check(struct shutdown_state *curr_sds,
 	if (!is_checksum_correct) {
 		/* the program was killed during opening or closing the pool */
 		CORE_LOG_WARNING(
-			"incorrect checksum - SDS will be reinitialized");
+			"The pool was not opened/closed properly - reinitializing ADR failure detection.");
 		shutdown_state_reinit(curr_sds, pool_sds, rep);
 		return 0;
 	}
@@ -217,19 +208,23 @@ shutdown_state_check(struct shutdown_state *curr_sds,
 		 * but there wasn't an ADR failure
 		 */
 		CORE_LOG_WARNING(
-			"the pool was not closed - SDS will be reinitialized");
+			"The ADR failure was detected but the pool was closed properly - reinitializing ADR failure detection.");
 		shutdown_state_reinit(curr_sds, pool_sds, rep);
 		return 0;
 	}
 	if (dirty == 0) {
-		/* an ADR failure but the pool was closed */
-		CORE_LOG_WARNING(
-			"an ADR failure was detected but the pool was closed - SDS will be reinitialized");
+		if (is_uuid_correct)
+			CORE_LOG_WARNING(
+				"The ADR failure was detected but the pool was closed properly - reinitializing ADR failure detection.");
+		else
+			CORE_LOG_HARK(
+				"The pool has moved to a new location but it was closed properly - reinitializing ADR failure detection.");
 		shutdown_state_reinit(curr_sds, pool_sds, rep);
 		return 0;
 	}
-	/* an ADR failure - the pool might be corrupted */
-	ERR_WO_ERRNO(
-		"an ADR failure was detected, the pool might be corrupted");
+
+	ERR_WO_ERRNO("%s, the pool might be corrupted.", is_uuid_correct ?
+		"The ADR failure was detected" :
+		"The pool has moved to a new location while it was not closed properly");
 	return 1;
 }

src/common/shutdown_state.h

@@ -1,5 +1,6 @@
 /* SPDX-License-Identifier: BSD-3-Clause */
 /* Copyright 2017-2020, Intel Corporation */
+/* Copyright 2025, Hewlett Packard Enterprise Development LP */
 
 /*
  * shutdown_state.h -- unsafe shudown detection
@@ -23,6 +24,9 @@ struct shutdown_state {
 	uint64_t checksum;
 };
 
+#define FLUSH_SDS(sds, rep) \
+	if ((rep) != NULL) os_part_deep_common(rep, 0, sds, sizeof(*(sds)), 1)
+
 int shutdown_state_init(struct shutdown_state *sds, struct pool_replica *rep);
 int shutdown_state_add_part(struct shutdown_state *sds, int fd,
 	struct pool_replica *rep);
@@ -34,6 +38,12 @@ void shutdown_state_clear_dirty(struct shutdown_state *sds,
 int shutdown_state_check(struct shutdown_state *curr_sds,
 	struct shutdown_state *pool_sds, struct pool_replica *rep);
 
+void shutdown_state_reinit(struct shutdown_state *curr_sds,
+	struct shutdown_state *pool_sds, struct pool_replica *rep);
+
+void shutdown_state_checksum(struct shutdown_state *sds,
+	struct pool_replica *rep);
+
 #ifdef __cplusplus
 }
 #endif

src/common/shutdown_state_helper.c

@@ -0,0 +1,28 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/* Copyright 2025, Hewlett Packard Enterprise Development LP */
+
+/*
+ * shutdown_state_helper.c -- implementation of shutdown_state_reinit
+ */
+
+#include "out.h"
+#include "os_deep.h"
+#include "set.h"
+
+/*
+ * shutdown_state_reinit -- reinitializes shutdown_state struct
+ */
+void
+shutdown_state_reinit(struct shutdown_state *curr_sds,
+	struct shutdown_state *pool_sds, struct pool_replica *rep)
+{
+	LOG(3, "curr_sds %p, pool_sds %p", curr_sds, pool_sds);
+	shutdown_state_init(pool_sds, rep);
+	pool_sds->uuid = htole64(curr_sds->uuid);
+	pool_sds->usc = htole64(curr_sds->usc);
+	pool_sds->dirty = 0;
+
+	FLUSH_SDS(pool_sds, rep);
+
+	shutdown_state_checksum(pool_sds, rep);
+}

src/test/Makefile

@@ -148,6 +148,7 @@ OTHER_TESTS = \
 	util_poolset_size\
 	util_ravl\
 	util_sds\
+	util_sds_check\
 	util_uuid_generate\
 	util_vec\
 	util_vecq\

src/test/Makefile.inc

@@ -246,6 +246,7 @@ OBJS +=\
 	$(TOP)/src/nondebug/common/pool_hdr.o\
 	$(TOP)/src/nondebug/common/set.o\
 	$(TOP)/src/nondebug/common/shutdown_state.o\
+	$(TOP)/src/nondebug/common/shutdown_state_helper.o\
 	$(TOP)/src/nondebug/common/util.o\
 	$(TOP)/src/nondebug/common/util_posix.o\
 	$(TOP)/src/nondebug/common/uuid.o\
@@ -287,6 +288,7 @@ OBJS +=\
 	$(TOP)/src/debug/common/pool_hdr.o\
 	$(TOP)/src/debug/common/set.o\
 	$(TOP)/src/debug/common/shutdown_state.o\
+	$(TOP)/src/debug/common/shutdown_state_helper.o\
 	$(TOP)/src/debug/common/uuid.o\
 	$(TOP)/src/debug/common/uuid_linux.o\
 	$(TOP)/src/debug/libpmem2/pmem2_utils.o\

src/test/core_log_max/call_all.c.generated

@@ -181,13 +181,13 @@ call_all_ERR_WO_ERRNO(void)
 	// src/common/set_badblocks.c
 	ERR_WO_ERRNO("part file contains bad blocks -- '%s'", _s);
 	// src/common/shutdown_state.c
-	ERR_WO_ERRNO("cannot read uuid of %d", _d);
-	// src/common/shutdown_state.c
-	ERR_WO_ERRNO("an ADR failure was detected, the pool might be corrupted");
+	ERR_WO_ERRNO("%s, the pool might be corrupted.", _s);
 	// src/common/shutdown_state.c
 	ERR_WO_ERRNO("Cannot read unsafe shutdown count. For more information please check https://github.com/pmem/pmdk/issues/4207");
 	// src/common/shutdown_state.c
 	ERR_WO_ERRNO("cannot read uuid of %d", _d);
+	// src/common/shutdown_state.c
+	ERR_WO_ERRNO("cannot read uuid of %d", _d);
 	// src/libpmem/libpmem.c
 	ERR_WO_ERRNO("libpmem major version mismatch (need %u, found %u)", _u, _u);
 	// src/libpmem/libpmem.c
@@ -462,11 +462,13 @@ call_all_CORE_LOG_WARNING(void)
 	// src/common/set.c
 	CORE_LOG_WARNING("file permissions changed during pool initialization, file: %s (%o)", _s, _u);
 	// src/common/shutdown_state.c
-	CORE_LOG_WARNING("incorrect checksum - SDS will be reinitialized");
+	CORE_LOG_WARNING("Enabling ADR failure detection, assuming pool consistency up to this point.");
+	// src/common/shutdown_state.c
+	CORE_LOG_WARNING("The pool was not opened/closed properly - reinitializing ADR failure detection.");
 	// src/common/shutdown_state.c
-	CORE_LOG_WARNING("the pool was not closed - SDS will be reinitialized");
+	CORE_LOG_WARNING("The ADR failure was detected but the pool was closed properly - reinitializing ADR failure detection.");
 	// src/common/shutdown_state.c
-	CORE_LOG_WARNING("an ADR failure was detected but the pool was closed - SDS will be reinitialized");
+	CORE_LOG_WARNING("The ADR failure was detected but the pool was closed properly - reinitializing ADR failure detection.");
 	// src/libpmemobj/heap.c
 	CORE_LOG_WARNING("failed to allocate memory block runtime tracking info");
 	// src/libpmemobj/heap.c

src/test/core_log_max/core_log_max.c

@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: BSD-3-Clause
 /* Copyright 2024, Intel Corporation */
+/* Copyright 2025, Hewlett Packard Enterprise Development LP */
 
 /*
  * core_log_max.c -- unit test to verify max size of log buffers
@@ -119,7 +120,7 @@ test_ERR_W_ERRNO(const struct test_case *tc, int argc, char *argv[])
 	return NO_ARGS_CONSUMED;
 }
 
-#define TOTAL_MESSAGE_NUM_EXPECTED 213
+#define TOTAL_MESSAGE_NUM_EXPECTED 214
 static int Max_message_len = 0;
 static int Total_message_num = 0;
 static char The_longest_message[BIG_BUF_SIZE];

src/test/util_sds/Makefile

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright 2015-2020, Intel Corporation
+# Copyright 2025, Hewlett Packard Enterprise Development LP
 
 #
 # src/test/util_sds/Makefile -- shutdown_tests unit test
@@ -11,6 +12,7 @@ vpath %.c $(TOP)/src/common/
 TARGET = util_sds
 OBJS = util_sds.o\
 	shutdown_state.o\
+	shutdown_state_helper.o\
 	ut_pmem2_config.o\
 	ut_pmem2_source.o\
 	ut_pmem2_utils.o

src/test/util_sds/grep1.log.match

@@ -0,0 +1 @@
+<libpmem2>: <2> [shutdown_state.c:$(N) shutdown_state_check] Enabling ADR failure detection, assuming pool consistency up to this point.