bdev/zone: Zone append handling added to vbdev

Changed write function to be able to work in zone append mode.

Signed-off-by: Mateusz Kozlowski <mateusz.kozlowski@intel.com>
Change-Id: I3af98f1d8e69eac6922e19addebabbc6ce847c63
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/468042
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>

module/bdev/zone_block/vbdev_zone_block.c

@@ -351,6 +351,10 @@ _zone_block_complete_write(struct spdk_bdev_io *bdev_io, bool success, void *cb_
 	struct spdk_bdev_io *orig_io = cb_arg;
 	int status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
 
+	if (success && orig_io->type == SPDK_BDEV_IO_TYPE_ZONE_APPEND) {
+		orig_io->u.bdev.offset_blocks = bdev_io->u.bdev.offset_blocks;
+	}
+
 	/* Complete the original IO and then free the one that we created here
 	 * as a result of issuing an IO via submit_reqeust.
 	 */
@@ -367,8 +371,13 @@ zone_block_write(struct bdev_zone_block *bdev_node, struct zone_block_io_channel
 	uint64_t lba = bdev_io->u.bdev.offset_blocks;
 	uint64_t num_blocks_left, wp;
 	int rc = 0;
+	bool is_append = bdev_io->type == SPDK_BDEV_IO_TYPE_ZONE_APPEND;
 
+	if (is_append) {
+		zone = zone_block_get_zone_by_slba(bdev_node, lba);
+	} else {
 		zone = zone_block_get_zone_containing_lba(bdev_node, lba);
+	}
 	if (!zone) {
 		SPDK_ERRLOG("Trying to write to invalid zone (lba 0x%lx)\n", lba);
 		return -EINVAL;
@@ -389,12 +398,15 @@ zone_block_write(struct bdev_zone_block *bdev_node, struct zone_block_io_channel
 	}
 
 	wp = zone->zone_info.write_pointer;
-
+	if (is_append) {
+		lba = wp;
+	} else {
 		if (lba != wp) {
 			SPDK_ERRLOG("Trying to write to zone with invalid address (lba 0x%lx, wp 0x%lx)\n", lba, wp);
 			rc = -EINVAL;
 			goto write_fail;
 		}
+	}
 
 	num_blocks_left = zone->zone_info.zone_id + zone->zone_info.capacity - wp;
 	if (len > num_blocks_left) {
@@ -494,6 +506,7 @@ zone_block_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_
 		rc = zone_block_zone_management(bdev_node, dev_ch, bdev_io);
 		break;
 	case SPDK_BDEV_IO_TYPE_WRITE:
+	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
 		rc = zone_block_write(bdev_node, dev_ch, bdev_io);
 		break;
 	case SPDK_BDEV_IO_TYPE_READ:
@@ -523,6 +536,7 @@ zone_block_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
 	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
 	case SPDK_BDEV_IO_TYPE_WRITE:
 	case SPDK_BDEV_IO_TYPE_READ:
+	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
 		return true;
 	default:
 		return false;

test/unit/lib/bdev/vbdev_zone_block.c/vbdev_zone_block_ut.c

@@ -782,7 +782,7 @@ test_supported_io_types(void)
 	bdev = create_and_get_vbdev("zone_dev1", name, num_zones, 1, true);
 
 	CU_ASSERT(zone_block_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT) == true);
-	CU_ASSERT(zone_block_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZONE_APPEND) == false);
+	CU_ASSERT(zone_block_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZONE_APPEND) == true);
 	CU_ASSERT(zone_block_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ) == true);
 	CU_ASSERT(zone_block_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE) == true);
 
@@ -1044,6 +1044,28 @@ send_read_zone(struct bdev_zone_block *bdev, struct spdk_io_channel *ch, uint64_
 	bdev_io_cleanup(bdev_io);
 }
 
+static void
+send_append_zone(struct bdev_zone_block *bdev, struct spdk_io_channel *ch, uint64_t lba,
+		 uint64_t blocks, uint32_t output_index, bool success, uint64_t wp)
+{
+	struct spdk_bdev_io *bdev_io;
+
+	bdev_io = calloc(1, sizeof(struct spdk_bdev_io) + sizeof(struct zone_block_io));
+	SPDK_CU_ASSERT_FATAL(bdev_io != NULL);
+	bdev_io_initialize(bdev_io, &bdev->bdev, lba, blocks, SPDK_BDEV_IO_TYPE_ZONE_APPEND);
+	memset(g_io_output, 0, (g_max_io_size * sizeof(struct io_output)));
+	g_io_output_index = output_index;
+
+	g_io_comp_status = !success;
+	zone_block_submit_request(ch, bdev_io);
+
+	CU_ASSERT(g_io_comp_status == success);
+	if (success) {
+		CU_ASSERT(bdev_io->u.bdev.offset_blocks == wp);
+	}
+	bdev_io_cleanup(bdev_io);
+}
+
 static void
 test_open_zone(void)
 {
@@ -1366,6 +1388,78 @@ test_finish_zone(void)
 	test_cleanup();
 }
 
+static void
+test_append_zone(void)
+{
+	struct spdk_io_channel *ch;
+	struct bdev_zone_block *bdev;
+	char *name = "Nvme0n1";
+	uint32_t num_zones = 20;
+	uint64_t zone_id, block_len, i;
+	uint32_t output_index = 0;
+
+	init_test_globals(20 * 1024ul);
+	CU_ASSERT(zone_block_init() == 0);
+
+	/* Create zone dev */
+	bdev = create_and_get_vbdev("zone_dev1", name, num_zones, 1, true);
+
+	ch = calloc(1, sizeof(struct spdk_io_channel) + sizeof(struct zone_block_io_channel));
+	SPDK_CU_ASSERT_FATAL(ch != NULL);
+
+	/* Append to full zone */
+	zone_id = 0;
+	send_append_zone(bdev, ch, zone_id, 1, output_index, false, 0);
+
+	/* Append out of device range */
+	zone_id = g_block_cnt;
+	send_append_zone(bdev, ch, zone_id, 1, output_index, false, 0);
+
+	/* Append 1 sector to zone 0 */
+	zone_id = 0;
+	send_reset_zone(bdev, ch, zone_id, output_index, true);
+	send_append_zone(bdev, ch, zone_id, 1, output_index, true, zone_id);
+	send_zone_info(bdev, ch, zone_id, 1, SPDK_BDEV_ZONE_STATE_OPEN, output_index, true);
+
+	/* Append to another zone */
+	zone_id = bdev->bdev.zone_size;
+	send_reset_zone(bdev, ch, zone_id, output_index, true);
+	send_append_zone(bdev, ch, zone_id, 5, output_index, true, zone_id);
+	send_zone_info(bdev, ch, zone_id, zone_id + 5, SPDK_BDEV_ZONE_STATE_OPEN, output_index, true);
+
+	/* Fill zone 0 and verify zone state change */
+	zone_id = 0;
+	block_len = 15;
+	send_append_zone(bdev, ch, zone_id, block_len, output_index, true, 1);
+	block_len++;
+	for (i = block_len; i < bdev->zone_capacity; i += block_len) {
+		send_append_zone(bdev, ch, zone_id, block_len, output_index, true, i);
+	}
+	send_zone_info(bdev, ch, zone_id, bdev->bdev.zone_size, SPDK_BDEV_ZONE_STATE_FULL, output_index,
+		       true);
+
+	/* Append to two zones at once */
+	for (i = 0; i < num_zones; i++) {
+		zone_id = i * bdev->bdev.zone_size;
+		send_reset_zone(bdev, ch, zone_id, output_index, true);
+		send_zone_info(bdev, ch, zone_id, zone_id, SPDK_BDEV_ZONE_STATE_EMPTY, output_index, true);
+	}
+
+	zone_id = 0;
+	block_len = 16;
+	for (i = 0; i < bdev->zone_capacity - block_len; i += block_len) {
+		send_append_zone(bdev, ch, zone_id, block_len, output_index, true, zone_id + i);
+	}
+	send_append_zone(bdev, ch, zone_id, 32, output_index, false, 0);
+	/* Delete zone dev */
+	send_delete_vbdev("zone_dev1", true);
+
+	while (spdk_thread_poll(g_thread, 0, 0) > 0) {}
+	free(ch);
+
+	test_cleanup();
+}
+
 int main(int argc, char **argv)
 {
 	CU_pSuite       suite = NULL;
@@ -1391,7 +1485,8 @@ int main(int argc, char **argv)
 		CU_add_test(suite, "test_zone_write", test_zone_write) == NULL ||
 		CU_add_test(suite, "test_zone_read", test_zone_read) == NULL ||
 		CU_add_test(suite, "test_close_zone", test_close_zone) == NULL ||
-		CU_add_test(suite, "test_finish_zone", test_finish_zone) == NULL
+		CU_add_test(suite, "test_finish_zone", test_finish_zone) == NULL ||
+		CU_add_test(suite, "test_append_zone", test_append_zone) == NULL
 	) {
 		CU_cleanup_registry();
 		return CU_get_error();