Commit c7feb85d authored by Haoqian He's avatar Haoqian He Committed by Tomasz Zawadzki
Browse files

nvmf: set max discard size and max write_zeroes size 



For Linux kernel version >= 5.13, when host connects the admin
queue successfully, it will call nvme_init_non_mdts_limits() to
post a NVME_ID_CNS_CS_CTRL identify cmd and init max discard size
with dmrl, dmrsl and max write_zeroes size with wzsl in response.

So we add the NVME_ID_CNS_CS_CTRL identify cmd handler in the nvmf
target side and add two new parameters to RPC nvmf_create_subsystem.

Change-Id: Ib72d6caa34fced68d63df9c01760a58effa5ca34
Signed-off-by: default avatarHaoqian He <haoqian.he@smartx.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/19632


Reviewed-by: default avatarAleksey Marchuk <alexeymar@nvidia.com>
Community-CI: Mellanox Build Bot
Reviewed-by: default avatarJim Harris <jim.harris@samsung.com>
Reviewed-by: default avatarBen Walker <ben@nvidia.com>
Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
parent f3244a4f

CHANGELOG.md

+5 −0
Original line number Diff line number Diff line
@@ -31,6 +31,11 @@ A new transport option `rdma_max_cq_size` was added to limit indefinite growth o 


Added SPDK commandline parameter --no-huge, which enables SPDK to run without hugepages.



### nvmf



Added `max_discard_size_kib` and `max_write_zeroes_size_kib` to `nvmf_create_subsystem` to set the

maximum discard size and maximum write_zeroes size.



## v23.09



### accel

doc/jsonrpc.md

+13 −11
Original line number Diff line number Diff line
@@ -8050,7 +8050,7 @@ Construct an NVMe over Fabrics target subsystem. 
#### Parameters



Name                       | Optional | Type        | Description

----------------------- | -------- | ----------- | -----------

-------------------------- | -------- | ----------- | -----------

nqn                        | Required | string      | Subsystem NQN

tgt_name                   | Optional | string      | Parent NVMe-oF target name.

serial_number              | Optional | string      | Serial number of virtual controller
@@ -8060,6 +8060,8 @@ allow_any_host | Optional | boolean | Allow any host (`true`) or en 
ana_reporting              | Optional | boolean     | Enable ANA reporting feature (default: `false`).

min_cntlid                 | Optional | number      | Minimum controller ID. Default: 1

max_cntlid                 | Optional | number      | Maximum controller ID. Default: 0xffef

max_discard_size_kib       | Optional | number      | Maximum discard size (Kib). Default: 0

max_write_zeroes_size_kib  | Optional | number      | Maximum write_zeroes size (Kib). Default: 0



#### Example

include/spdk/nvme_spec.h

+23 −0
Original line number Diff line number Diff line
@@ -2960,6 +2960,29 @@ struct spdk_nvme_ns_data { 
};

SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_ns_data) == 4096, "Incorrect size");



struct spdk_nvme_nvm_ctrlr_data {

	/* verify size limit */

	uint8_t			vsl;



	/* write zeroes size limit */

	uint8_t			wzsl;



	/* write uncorrectable size limit */

	uint8_t			wusl;



	/* dataset management ranges limit */

	uint8_t			dmrl;



	/* dataset management range size limit */

	uint32_t		dmrsl;



	/* dataset management size limit */

	uint64_t		dmsl;



	uint8_t			rsvd16[4080];

};

SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_nvm_ctrlr_data) == 4096, "Incorrect size");



struct spdk_nvme_zns_ns_data {

	/** zone operation characteristics */

	struct {

lib/nvmf/ctrlr.c

+27 −0
Original line number Diff line number Diff line
@@ -2904,6 +2904,31 @@ spdk_nvmf_ns_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 
	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;

}



static int

nvmf_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ctrlr *ctrlr,

			     struct spdk_nvme_cmd *cmd,

			     struct spdk_nvme_cpl *rsp,

			     struct spdk_nvme_nvm_ctrlr_data *cdata_nvm)

{

	/* The unit of max_write_zeroes_size_kib is KiB.

	 * The unit of wzsl is the minimum memory page size(2 ^ (12 + CAP.MPSMIN) bytes)

	 * and is reported as a power of two (2^n).

	 */

	cdata_nvm->wzsl = spdk_u64log2(ctrlr->subsys->max_write_zeroes_size_kib >>

				       (2 + ctrlr->vcprop.cap.bits.mpsmin));



	/* The unit of max_discard_size_kib is KiB.

	 * The dmrsl indicates the maximum number of logical blocks for

	 * dataset management command.

	 */

	cdata_nvm->dmrsl = ctrlr->subsys->max_discard_size_kib << 1;

	cdata_nvm->dmrl = 1;



	rsp->status.sct = SPDK_NVME_SCT_GENERIC;

	rsp->status.sc = SPDK_NVME_SC_SUCCESS;

	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;

}



static int

nvmf_ctrlr_identify_iocs_zns(struct spdk_nvmf_ctrlr *ctrlr,

			     struct spdk_nvme_cmd *cmd,
@@ -2934,6 +2959,8 @@ spdk_nvmf_ctrlr_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 
	memset(cdata, 0, cdata_size);



	switch (csi) {

	case SPDK_NVME_CSI_NVM:

		return nvmf_ctrlr_identify_iocs_nvm(ctrlr, cmd, rsp, cdata);

	case SPDK_NVME_CSI_ZNS:

		return nvmf_ctrlr_identify_iocs_zns(ctrlr, cmd, rsp, cdata);

	default:

lib/nvmf/ctrlr_bdev.c

+16 −0
Original line number Diff line number Diff line
@@ -473,11 +473,19 @@ nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc * 
	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);

	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;

	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;

	uint64_t max_write_zeroes_size = req->qpair->ctrlr->subsys->max_write_zeroes_size_kib;

	uint64_t start_lba;

	uint64_t num_blocks;

	int rc;



	nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks);

	if (spdk_unlikely(max_write_zeroes_size > 0 &&

			  num_blocks > (max_write_zeroes_size << 10) / spdk_bdev_get_block_size(bdev))) {

		SPDK_ERRLOG("invalid write zeroes size, should not exceed %" PRIu64 "Kib\n", max_write_zeroes_size);

		rsp->status.sct = SPDK_NVME_SCT_GENERIC;

		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;

		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;

	}



	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {

		SPDK_ERRLOG("end of media\n");
@@ -590,6 +598,8 @@ nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 
	uint16_t nr, i;

	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;

	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;

	uint64_t max_discard_size = req->qpair->ctrlr->subsys->max_discard_size_kib;

	uint32_t block_size = spdk_bdev_get_block_size(bdev);

	struct spdk_iov_xfer ix;

	uint64_t lba;

	uint32_t lba_count;
@@ -629,6 +639,12 @@ nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 


		lba = dsm_range.starting_lba;

		lba_count = dsm_range.length;

		if (max_discard_size > 0 && lba_count > (max_discard_size << 10) / block_size) {

			SPDK_ERRLOG("invalid unmap size, should not exceed %" PRIu64 "Kib\n", max_discard_size);

			response->status.sct = SPDK_NVME_SCT_GENERIC;

			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;

			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;

		}



		unmap_ctx->count++;