v16.12: NVMe-oF initiator, NVMe hotplug and multiprocess, Ceph RBD

@benlwalker benlwalker released this Dec 19, 2016 · 101 commits to master since this release

NVMe library

The NVMe library has been changed to create its own request memory pool rather than
requiring the user to initialize the global request_mempool variable. Apps can be
updated by simply removing the initialization of request_mempool. Since the NVMe
library user no longer needs to know the size of the internal NVMe request
structure to create the pool, the spdk_nvme_request_size() function was also removed.

The spdk_nvme_ns_cmd_deallocate() function was renamed and extended to become
spdk_nvme_ns_cmd_dataset_management(), which allows access to all of the NVMe
Dataset Management command's parameters. Existing callers can be updated to use
spdk_nvme_ns_cmd_dataset_management() with SPDK_NVME_DSM_ATTR_DEALLOCATE as the
type parameter.

The NVMe library SGL callback prototype has been changed to return virtual addresses
rather than physical addresses. Callers of spdk_nvme_ns_cmd_readv() and
spdk_nvme_ns_cmd_writev() must update their next_sge_fn callbacks to match.

The NVMe library now supports NVMe over Fabrics devices in addition to the existing
support for local PCIe-attached NVMe devices. For an example of how to enable
NVMe over Fabrics support in an application, see examples/nvme/identify and

Hot insert/remove support for NVMe devices has been added. To enable NVMe hotplug
support, an application should call the spdk_nvme_probe() function on a regular
basis to probe for new devices (reported via the existing probe_cb callback) and
removed devices (reported via a new remove_cb callback). Hotplug is currently
only supported on Linux with the uio_pci_generic driver, and newly-added NVMe
devices must be bound to uio_pci_generic by an external script or tool.

Multiple processes may now coordinate and use a single NVMe device simultaneously
using DPDK Multi-process Support.

NVMe over Fabrics target (nvmf_tgt)

The nvmf_tgt configuration file format has been updated significantly to enable
new features. See the example configuration file etc/spdk/nvmf.conf.in for
more details on the new and changed options.

The NVMe over Fabrics target now supports virtual mode subsystems, which allow the
user to export devices from the SPDK block device abstraction layer as NVMe over
Fabrics subsystems. Direct mode (raw NVMe device access) is also still supported,
and a single nvmf_tgt may export both types of subsystems simultaneously.

Block device abstraction layer (bdev)

The bdev layer now supports scatter/gather read and write I/O APIs, and the NVMe
blockdev driver has been updated to support scatter/gather. Apps can use the
new scatter/gather support via the spdk_bdev_readv() and spdk_bdev_writev()

The bdev status returned from each I/O has been extended to pass through NVMe
or SCSI status codes directly in cases where the underlying device can provide
a more specific status code.

A Ceph RBD (RADOS Block Device) blockdev driver has been added. This allows the
iscsi_tgt and nvmf_tgt apps to export Ceph RBD volumes as iSCSI LUNs or
NVMe namespaces.

General changes

libpciaccess has been removed as a dependency and DPDK PCI enumeration is
used instead. Prior to DPDK 16.07 enumeration by class code was not supported,
so for earlier DPDK versions, only Intel SSD DC P3x00 devices will be discovered
by the NVMe library.

The env environment abstraction library has been introduced, and a default
DPDK-based implementation is provided as part of SPDK. The goal of the env
layer is to enable use of alternate user-mode memory allocation and PCI access
libraries. See PORTING.md for more details.

The build process has been modified to produce all of the library files in the
build/lib directory. This is intended to simplify the use of SPDK from external
projects, which can now link to SPDK libraries by adding the build/lib directory
to the library path via -L and linking the SPDK libraries by name (for example,
-lspdk_nvme -lspdk_log -lspdk_util).

nvmf_tgt and iscsi_tgt now have a JSON-RPC interface, which allows the user
to query and modify the configuration at runtime. The RPC service is disabled by
default, since it currently does not provide any authentication or security
mechanisms; it should only be enabled on systems with controlled user access
behind a firewall. An example RPC client implemented in Python is provided in


Full documentation for this release is available at http://www.spdk.io/spdk/v16.12/doc/.


v16.08: iSCSI target, NVMe over Fabrics maturity

@benlwalker benlwalker released this Aug 8, 2016 · 772 commits to master since this release

This release adds a userspace iSCSI target. The iSCSI target is capable of exporting
NVMe devices over a network using the iSCSI protocol. The application is located
in app/iscsi_tgt and a documented configuration file can be found at etc/spdk/spdk.conf.in.

This release also significantly improves the existing NVMe over Fabrics target.

  • The configuration file format was changed, which will require updates to
    any existing nvmf.conf files (see etc/spdk/nvmf.conf.in):
    • SubsystemGroup was renamed to Subsystem.
    • AuthFile was removed (it was unimplemented).
    • nvmf_tgt was updated to correctly recognize NQN (NVMe Qualified Names)
      when naming subsystems. The default node name was changed to reflect this;
      it is now "nqn.2016-06.io.spdk".
    • Port and Host sections were merged into the Subsystem section
    • Global options to control max queue depth, number of queues, max I/O
      size, and max in-capsule data size were added.
    • The Nvme section was removed. Now a list of devices is specified by
      bus/device/function directly in the Subsystem section.
    • Subsystems now have a Mode, which can be Direct or Virtual. This is an attempt
      to future-proof the interface, so the only mode supported by this release
      is "Direct".
  • Many bug fixes and cleanups were applied to the nvmf_tgt app and library.
  • The target now supports discovery.

This release also adds one new feature and provides some better examples and tools
for the NVMe driver.

  • The Weighted Round Robin arbitration method is now supported. This allows
    the user to specify different priorities on a per-I/O-queue basis. To
    enable WRR, set the arb_mechanism field during spdk_nvme_probe().
  • A simplified "Hello World" example was added to show the proper way to use
    the NVMe library API; see examples/nvme/hello_world/hello_world.c.
  • A test for measuring software overhead was added. See test/lib/nvme/overhead.


v16.06: NVMf userspace target

@benlwalker benlwalker released this Jun 8, 2016 · 1154 commits to master since this release

This release adds a userspace NVMf (NVMe over Fabrics) target, conforming to the
newly-released NVMf 1.0/NVMe 1.2.1 specification. The NVMf target exports NVMe
devices from a host machine over the network via RDMA. Currently, the target is
limited to directly exporting physical NVMe devices, and the discovery subsystem
is not supported.

This release includes a general API cleanup, including renaming all declarations
in public headers to include a spdk prefix to prevent namespace clashes with
user code.

  • NVMe
    • The nvme_attach() API was reworked into a new probe/attach model, which
      moves device detection into the NVMe library. The new API also allows
      parallel initialization of NVMe controllers, providing a major reduction in
      startup time when using multiple controllers.
    • I/O queue allocation was changed to be explicit in the API. Each function
      that generates I/O requests now takes a queue pair (spdk_nvme_qpair *)
      argument, and I/O queues may be allocated using
      spdk_nvme_ctrlr_alloc_io_qpair(). This allows more flexible assignment of
      queue pairs than the previous model, which only allowed a single queue
      per thread and limited the total number of I/O queues to the lowest number
      supported on any attached controller.
    • Added support for the Write Zeroes command.
    • examples/nvme/perf can now report I/O command latency from the
      the controller's viewpoint using the Intel vendor-specific read/write latency
      log page.
    • Added namespace reservation command support, which can be used to coordinate
      sharing of a namespace between multiple hosts.
    • Added hardware SGL support, which enables use of scattered buffers that
      don't conform to the PRP list alignment and length requirements on supported
      NVMe controllers.
    • Added end-to-end data protection support, including the ability to write and
      read metadata in extended LBA (metadata appended to each block of data in the
      buffer) and separate metadata buffer modes.
      See spdk_nvme_ns_cmd_write_with_md() and spdk_nvme_ns_cmd_read_with_md()
      for details.
  • IOAT
    • The DMA block fill feature is now exposed via the ioat_submit_fill()
      function. This is functionally similar to memset(), except the memory is
      filled with an 8-byte repeating pattern instead of a single byte like memset.
  • PCI
    • Added support for using DPDK for PCI device mapping in addition to the
      existing libpciaccess option. Using the DPDK PCI support also allows use of
      the Linux VFIO driver model, which means that SPDK userspace drivers will work
      with the IOMMU enabled. Additionally, SPDK applications may be run as an
      unprivileged user with access restricted to a specific set of PCIe devices.
    • The PCI library API was made more generic to abstract away differences
      between the underlying PCI access implementations.


v1.2.0: IOAT user-space driver

@dverkamp-intel dverkamp-intel released this Jan 28, 2016 · 1395 commits to master since this release

This release adds a user-space driver with support for the Intel I/O Acceleration Technology (I/OAT, also known as "Crystal Beach") DMA offload engine.

  • IOAT
    • New user-space driver supporting DMA memory copy offload
    • Example programs ioat/perf and ioat/verify
    • 0e6463d: Kernel-mode DMA engine test driver kperf for performance comparison
  • NVMe
    • 81f4046: per-I/O flags for Force Unit Access (FUA) and Limited Retry
    • 9945c00: public API for retrieving log pages
    • 92fa3ec: reservation register/acquire/release/report command support
    • d3d6d19: scattered payload support - an alternate API to provide I/O buffers via a sequence of callbacks
    • b7b8dea, ffa548d, 3f7cfd2: declarations and nvme/identify support for Intel SSD DC P3700 series vendor-specific log pages and features
  • Updated to support DPDK 2.2.0


v1.0.0: NVMe user-space driver

@dverkamp-intel dverkamp-intel released this Dec 7, 2015 · 1461 commits to master since this release

This is the initial open source release of the Storage Performance Development Kit (SPDK).


  • NVMe user-space driver
  • NVMe example programs
    • examples/nvme/perf tests performance (IOPS) using the NVMe user-space driver
    • examples/nvme/identify displays NVMe controller information in a human-readable format
  • Linux and FreeBSD support