nvme: add iovec passthru (04a428f5) · Commits · Public Repositories / spdk

include/spdk/bdev.h

+34 −0

Original line number	Diff line number	Diff line
		@@ -121,6 +121,7 @@ enum spdk_bdev_io_type {
		SPDK_BDEV_IO_TYPE_SEEK_HOLE,
		SPDK_BDEV_IO_TYPE_SEEK_DATA,
		SPDK_BDEV_IO_TYPE_COPY,
		SPDK_BDEV_IO_TYPE_NVME_IOV_MD,
		SPDK_BDEV_NUM_IO_TYPES /* Keep last */
		};

		@@ -1791,6 +1792,39 @@ int spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *bdev_desc,
		void buf, size_t nbytes, void md_buf, size_t md_len,
		spdk_bdev_io_completion_cb cb, void *cb_arg);

		/**
		* Submit an NVMe I/O command to the bdev. This passes directly through
		* the block layer to the device. Support for NVMe passthru is optional,
		* indicated by calling spdk_bdev_io_type_supported().
		*
		* \ingroup bdev_io_submit_functions
		*
		* The namespace id (nsid) will be populated automatically.
		*
		* \param desc Block device descriptor
		* \param ch I/O channel. Obtained by calling spdk_bdev_get_io_channel().
		* \param cmd The raw NVMe command. Must be in the NVM command set.
		* \param iov A scatter gather list of buffers for the command to use.
		* \param iovcnt The number of elements in iov.
		* \param nbytes The number of bytes to transfer. The total size of the buffers in iov must be greater than or equal to this size.
		* \param md_buf Meta data buffer to written from.
		* \param md_len md_buf size to transfer. md_buf must be greater than or equal to this size.
		* \param cb Called when the request is complete.
		* \param cb_arg Argument passed to cb.
		*
		* \return 0 on success. On success, the callback will always
		* be called (even if the request ultimately failed). Return
		* negated errno on failure, in which case the callback will not be called.
		* * -ENOMEM - spdk_bdev_io buffer cannot be allocated
		* * -EBADF - desc not open for writing
		*/
		int spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
		struct spdk_io_channel *ch,
		const struct spdk_nvme_cmd *cmd,
		struct iovec *iov, int iovcnt,
		size_t nbytes, void *md_buf, size_t md_len,
		spdk_bdev_io_completion_cb cb, void *cb_arg);

		/**
		* Submit a copy request to the block device.
		*

include/spdk/bdev_module.h

+6 −0

Original line number	Diff line number	Diff line
		@@ -873,6 +873,12 @@ struct spdk_bdev_io {
		/* The NVMe command to execute */
		struct spdk_nvme_cmd cmd;

		/* For SG buffer cases, array of iovecs to transfer. */
		struct iovec *iovs;

		/* For SG buffer cases, number of iovecs in iovec array. */
		int iovcnt;

		/* The data buffer to transfer */
		void *buf;

include/spdk/nvme.h

+57 −21

Original line number	Diff line number	Diff line
		@@ -1851,6 +1851,63 @@ int spdk_nvme_ctrlr_cmd_io_raw_with_md(struct spdk_nvme_ctrlr *ctrlr,
		void buf, uint32_t len, void md_buf,
		spdk_nvme_cmd_cb cb_fn, void *cb_arg);

		/**
		* Restart the SGL walk to the specified offset when the command has scattered
		* payloads.
		*
		* \param cb_arg Argument passed to readv/writev.
		* \param offset Offset for SGL.
		*/
		typedef void (spdk_nvme_req_reset_sgl_cb)(void cb_arg, uint32_t offset);

		/**
		* Fill out address and length with the current SGL entry and advance to the
		* next entry for the next time the callback is invoked.
		*
		* The described segment must be physically contiguous.
		*
		* \param cb_arg Argument passed to readv/writev.
		* \param address Virtual address of this segment, a value of UINT64_MAX
		* means the segment should be described via Bit Bucket SGL.
		* \param length Length of this physical segment.
		*/
		typedef int (spdk_nvme_req_next_sge_cb)(void cb_arg, void **address,
		uint32_t *length);

		/**
		* Send the given NVM I/O command with metadata to the NVMe controller.
		*
		* This is a low level interface for submitting I/O commands directly. Prefer
		* the spdk_nvme_ns_cmd_* functions instead. The validity of the command will
		* not be checked!
		*
		* The command is submitted to a qpair allocated by spdk_nvme_ctrlr_alloc_io_qpair().
		* The user must ensure that only one thread submits I/O on a given qpair at any
		* given time.
		*
		* \param ctrlr Opaque handle to NVMe controller.
		* \param qpair I/O qpair to submit command.
		* \param cmd NVM I/O command to submit.
		* \param len Size of buffer.
		* \param md_buf Virtual memory address of a single physically contiguous metadata buffer.
		* \param cb_fn Callback function invoked when the I/O command completes.
		* \param cb_arg Argument passed to callback function.
		* \param reset_sgl_fn Callback function to reset scattered payload.
		* \param next_sge_fn Callback function to iterate each scattered payload memory segment.
		*
		* \return 0 if successfully submitted, negated errnos on the following error
		conditions:
		* -ENOMEM: The request cannot be allocated.
		* -ENXIO: The qpair is failed at the transport level.
		*/
		int spdk_nvme_ctrlr_cmd_iov_raw_with_md(struct spdk_nvme_ctrlr *ctrlr,
		struct spdk_nvme_qpair *qpair,
		struct spdk_nvme_cmd *cmd, uint32_t len,
		void *md_buf, spdk_nvme_cmd_cb cb_fn,
		void *cb_arg,
		spdk_nvme_req_reset_sgl_cb reset_sgl_fn,
		spdk_nvme_req_next_sge_cb next_sge_fn);

		/**
		* Process any outstanding completions for I/O submitted on a queue pair.
		*
		@@ -3054,27 +3111,6 @@ uint32_t spdk_nvme_ns_get_ana_group_id(const struct spdk_nvme_ns *ns);
		*/
		enum spdk_nvme_ana_state spdk_nvme_ns_get_ana_state(const struct spdk_nvme_ns *ns);

		/**
		* Restart the SGL walk to the specified offset when the command has scattered payloads.
		*
		* \param cb_arg Argument passed to readv/writev.
		* \param offset Offset for SGL.
		*/
		typedef void (spdk_nvme_req_reset_sgl_cb)(void cb_arg, uint32_t offset);

		/**
		* Fill out address and length with the current SGL entry and advance to the next
		* entry for the next time the callback is invoked.
		*
		* The described segment must be physically contiguous.
		*
		* \param cb_arg Argument passed to readv/writev.
		* \param address Virtual address of this segment, a value of UINT64_MAX
		* means the segment should be described via Bit Bucket SGL.
		* \param length Length of this physical segment.
		*/
		typedef int (spdk_nvme_req_next_sge_cb)(void cb_arg, void *address, uint32_t length);

		/**
		* Submit a write I/O to the specified NVMe namespace.
		*

lib/bdev/bdev.c

+48 −0

Original line number	Diff line number	Diff line
		@@ -6690,6 +6690,54 @@ spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channe
		return 0;
		}

		int
		spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
		struct spdk_io_channel *ch,
		const struct spdk_nvme_cmd *cmd,
		struct iovec *iov, int iovcnt, size_t nbytes,
		void *md_buf, size_t md_len,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
		{
		struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
		struct spdk_bdev_io *bdev_io;
		struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);

		if (!desc->write) {
		/*
		* Do not try to parse the NVMe command - we could maybe use bits in the opcode
		* to easily determine if the command is a read or write, but for now just
		* do not allow io_passthru with a read-only descriptor.
		*/
		return -EBADF;
		}

		if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
		return -ENOTSUP;
		} else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
		return -ENOTSUP;
		}

		bdev_io = bdev_channel_get_io(channel);
		if (!bdev_io) {
		return -ENOMEM;
		}

		bdev_io->internal.ch = channel;
		bdev_io->internal.desc = desc;
		bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
		bdev_io->u.nvme_passthru.cmd = *cmd;
		bdev_io->u.nvme_passthru.iovs = iov;
		bdev_io->u.nvme_passthru.iovcnt = iovcnt;
		bdev_io->u.nvme_passthru.nbytes = nbytes;
		bdev_io->u.nvme_passthru.md_buf = md_buf;
		bdev_io->u.nvme_passthru.md_len = md_len;

		bdev_io_init(bdev_io, bdev, cb_arg, cb);

		bdev_io_submit(bdev_io);
		return 0;
		}

		static void bdev_abort_retry(void *ctx);
		static void bdev_abort(struct spdk_bdev_io *parent_io);

lib/bdev/spdk_bdev.map

+1 −0

Original line number	Diff line number	Diff line
		@@ -86,6 +86,7 @@
		spdk_bdev_nvme_admin_passthru;
		spdk_bdev_nvme_io_passthru;
		spdk_bdev_nvme_io_passthru_md;
		spdk_bdev_nvme_iov_passthru_md;
		spdk_bdev_free_io;
		spdk_bdev_queue_io_wait;
		spdk_bdev_get_io_stat;