Commit a9a86a14 authored by Shuhei Matsumoto's avatar Shuhei Matsumoto Committed by Tomasz Zawadzki
Browse files

bdev/nvme: Retry admin passthru immediately if it got ctrlr path error



This patch supports admin passthrough retry when we get ctrlr path
error at completion.

Signed-off-by: default avatarShuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ice0045b84054ec66a9db9ef23e21786d2c082b1d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9943


Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
Reviewed-by: default avatarAleksey Marchuk <alexeymar@mellanox.com>
parent 35a2f4e2
Loading
Loading
Loading
Loading
+29 −2
Original line number Diff line number Diff line
@@ -71,8 +71,8 @@ struct nvme_bdev_io {
	/** Offset in current iovec. */
	uint32_t iov_offset;

	/** I/O path the current I/O is submitted on, or the I/O path being reset
	 *  in a reset I/O.
	/** I/O path the current I/O or admin passthrough is submitted on, or the I/O path
	 *  being reset in a reset I/O.
	 */
	struct nvme_io_path *io_path;

@@ -4154,7 +4154,33 @@ bdev_nvme_admin_passthru_complete_nvme_status(void *ctx)
	struct nvme_bdev_io *bio = ctx;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	const struct spdk_nvme_cpl *cpl = &bio->cpl;
	struct nvme_bdev_channel *nbdev_ch;
	struct nvme_ctrlr *nvme_ctrlr;

	assert(bdev_nvme_io_type_is_admin(bdev_io->type));

	if (spdk_likely(spdk_nvme_cpl_is_success(cpl))) {
		goto complete;
	}

	if (cpl->status.dnr != 0) {
		goto complete;
	}

	nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
	nvme_ctrlr = nvme_ctrlr_channel_get_ctrlr(bio->io_path->ctrlr_ch);

	if (spdk_nvme_cpl_is_path_error(cpl) ||
	    spdk_nvme_cpl_is_aborted_sq_deletion(cpl) ||
	    spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr) ||
	    nvme_ctrlr->resetting) {
		if (any_ctrlr_may_become_available(nbdev_ch)) {
			bdev_nvme_queue_retry_io(nbdev_ch, bio, 0);
			return;
		}
	}

complete:
	spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc);
}

@@ -4674,6 +4700,7 @@ bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io
			goto err;
		}

		bio->io_path = io_path;
		bio->orig_thread = spdk_get_thread();

		rc = spdk_nvme_ctrlr_cmd_admin_raw(nvme_ctrlr->ctrlr, cmd, buf, (uint32_t)nbytes,
+179 −0
Original line number Diff line number Diff line
@@ -4779,6 +4779,184 @@ test_retry_admin_passthru_if_ctrlr_is_resetting(void)
	CU_ASSERT(nvme_bdev_ctrlr_get_by_name("nvme0") == NULL);
}

static void
test_retry_admin_passthru_for_path_error(void)
{
	struct nvme_path_id path1 = {}, path2 = {};
	struct spdk_nvme_ctrlr *ctrlr1, *ctrlr2;
	struct nvme_bdev_ctrlr *nbdev_ctrlr;
	struct nvme_ctrlr *nvme_ctrlr1, *nvme_ctrlr2;
	const int STRING_SIZE = 32;
	const char *attached_names[STRING_SIZE];
	struct nvme_bdev *bdev;
	struct spdk_bdev_io *admin_io;
	struct spdk_io_channel *ch;
	struct ut_nvme_req *req;
	int rc;

	memset(attached_names, 0, sizeof(char *) * STRING_SIZE);
	ut_init_trid(&path1.trid);
	ut_init_trid2(&path2.trid);

	g_opts.bdev_retry_count = 1;

	set_thread(0);

	g_ut_attach_ctrlr_status = 0;
	g_ut_attach_bdev_count = 1;

	ctrlr1 = ut_attach_ctrlr(&path1.trid, 1, true, true);
	SPDK_CU_ASSERT_FATAL(ctrlr1 != NULL);

	memset(&ctrlr1->ns[0].uuid, 1, sizeof(struct spdk_uuid));

	rc = bdev_nvme_create(&path1.trid, "nvme0", attached_names, STRING_SIZE, 0,
			      attach_ctrlr_done, NULL, NULL, true);
	CU_ASSERT(rc == 0);

	spdk_delay_us(1000);
	poll_threads();

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_threads();

	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name("nvme0");
	SPDK_CU_ASSERT_FATAL(nbdev_ctrlr != NULL);

	nvme_ctrlr1 = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, &path1.trid);
	CU_ASSERT(nvme_ctrlr1 != NULL);

	bdev = nvme_bdev_ctrlr_get_bdev(nbdev_ctrlr, 1);
	CU_ASSERT(bdev != NULL);

	admin_io = ut_alloc_bdev_io(SPDK_BDEV_IO_TYPE_NVME_ADMIN, bdev, NULL);
	admin_io->u.nvme_passthru.cmd.opc = SPDK_NVME_OPC_GET_FEATURES;

	ch = spdk_get_io_channel(bdev);
	SPDK_CU_ASSERT_FATAL(ch != NULL);

	admin_io->internal.ch = (struct spdk_bdev_channel *)ch;

	/* Admin passthrough got a path error, but it should not retry if DNR is set. */
	admin_io->internal.in_submit_request = true;

	bdev_nvme_submit_request(ch, admin_io);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 1);
	CU_ASSERT(admin_io->internal.in_submit_request == true);

	req = ut_get_outstanding_nvme_request(&ctrlr1->adminq, admin_io->driver_ctx);
	SPDK_CU_ASSERT_FATAL(req != NULL);

	req->cpl.status.sc = SPDK_NVME_SC_INTERNAL_PATH_ERROR;
	req->cpl.status.sct = SPDK_NVME_SCT_PATH;
	req->cpl.status.dnr = 1;

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_thread_times(0, 2);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 0);
	CU_ASSERT(admin_io->internal.in_submit_request == false);
	CU_ASSERT(admin_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR);

	/* Admin passthrough got a path error, but it should succeed after retry. */
	admin_io->internal.in_submit_request = true;

	bdev_nvme_submit_request(ch, admin_io);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 1);
	CU_ASSERT(admin_io->internal.in_submit_request == true);

	req = ut_get_outstanding_nvme_request(&ctrlr1->adminq, admin_io->driver_ctx);
	SPDK_CU_ASSERT_FATAL(req != NULL);

	req->cpl.status.sc = SPDK_NVME_SC_INTERNAL_PATH_ERROR;
	req->cpl.status.sct = SPDK_NVME_SCT_PATH;

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_thread_times(0, 2);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 1);
	CU_ASSERT(admin_io->internal.in_submit_request == true);

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_threads();

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 0);
	CU_ASSERT(admin_io->internal.in_submit_request == false);
	CU_ASSERT(admin_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);

	/* Add ctrlr2 dynamically, and create a multipath configuration. */
	ctrlr2 = ut_attach_ctrlr(&path2.trid, 1, true, true);
	SPDK_CU_ASSERT_FATAL(ctrlr2 != NULL);

	memset(&ctrlr2->ns[0].uuid, 1, sizeof(struct spdk_uuid));

	rc = bdev_nvme_create(&path2.trid, "nvme0", attached_names, STRING_SIZE, 0,
			      attach_ctrlr_done, NULL, NULL, true);
	CU_ASSERT(rc == 0);

	spdk_delay_us(1000);
	poll_threads();

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_threads();

	nvme_ctrlr2 = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, &path2.trid);
	CU_ASSERT(nvme_ctrlr2 != NULL);

	/* Admin passthrough was submitted to ctrlr1, but ctrlr1 was failed.
	 * Hence the admin passthrough was aborted. But ctrlr2 is avaialble.
	 * So after a retry, the admin passthrough is submitted to ctrlr2 and
	 * should succeed.
	 */
	admin_io->internal.in_submit_request = true;

	bdev_nvme_submit_request(ch, admin_io);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 1);
	CU_ASSERT(ctrlr2->adminq.num_outstanding_reqs == 0);
	CU_ASSERT(admin_io->internal.in_submit_request == true);

	req = ut_get_outstanding_nvme_request(&ctrlr1->adminq, admin_io->driver_ctx);
	SPDK_CU_ASSERT_FATAL(req != NULL);

	req->cpl.status.sc = SPDK_NVME_SC_ABORTED_SQ_DELETION;
	req->cpl.status.sct = SPDK_NVME_SCT_GENERIC;
	ctrlr1->is_failed = true;

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_thread_times(0, 2);

	CU_ASSERT(ctrlr1->adminq.num_outstanding_reqs == 0);
	CU_ASSERT(ctrlr2->adminq.num_outstanding_reqs == 1);
	CU_ASSERT(admin_io->internal.in_submit_request == true);

	spdk_delay_us(g_opts.nvme_adminq_poll_period_us);
	poll_threads();

	CU_ASSERT(ctrlr2->adminq.num_outstanding_reqs == 0);
	CU_ASSERT(admin_io->internal.in_submit_request == false);
	CU_ASSERT(admin_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);

	free(admin_io);

	spdk_put_io_channel(ch);

	poll_threads();

	rc = bdev_nvme_delete("nvme0", &g_any_path);
	CU_ASSERT(rc == 0);

	poll_threads();
	spdk_delay_us(1000);
	poll_threads();

	CU_ASSERT(nvme_bdev_ctrlr_get_by_name("nvme0") == NULL);

	g_opts.bdev_retry_count = 0;
}

int
main(int argc, const char **argv)
{
@@ -4818,6 +4996,7 @@ main(int argc, const char **argv)
	CU_ADD_TEST(suite, test_concurrent_read_ana_log_page);
	CU_ADD_TEST(suite, test_retry_io_for_ana_error);
	CU_ADD_TEST(suite, test_retry_admin_passthru_if_ctrlr_is_resetting);
	CU_ADD_TEST(suite, test_retry_admin_passthru_for_path_error);

	CU_basic_set_mode(CU_BRM_VERBOSE);