spdk_app_start() now only accepts a single context argument.
Added asynchronous probe support. New APIs spdk_nvme_probe_async(), spdk_nvme_connect_async() and spdk_nvme_probe_poll_async() were added to enable this feature, spdk_nvme_probe_async() and spdk_nvme_connect_async() return a context associated with the specified controllers. Users then call spdk_nvme_probe_poll_async() until it returns 0, indicating that the operation is completed with success.
A new qpair creation option, delay_pcie_doorbell, was added. This can be passed to spdk_nvme_alloc_io_qpair(). This makes the I/O submission functions, such as spdk_nvme_ns_writev(), skip ringing the submission queue doorbell. Instead the doorbell will be rung as necessary inside spdk_nvme_qpair_process_completions(). This can result in significantly fewer MMIO writes to the doorbell register under heavy load, greatly improving performance.
New API spdk_nvme_ctrlr_get_flags() was added.
NVMe hotplug poller is now able to detach devices hotremoved from the system via
Added new strip_size_kb rpc param on create to replace the more ambiguous strip_size. The strip_size rpc param is deprecated.
Added spdk_thread_has_pollers() function to verify if there are any registered pollers to be run on the thread.
Added spdk_thread_is_idle() function to check if there are any scheduled operations to be performed on the thread at given time.
An new API
spdk_bdev_get_data_block_size has been added to get size of data block except for metadata.
Support for per-device shared receive queues in the RDMA transport has been added. It is enabled by default for any device that supports it.
The size of a shared receive queue is defined by transport configuration file parameter
nvmf_create_transport RPC method parameter
max_srq_depth. Default size is 4096.
spdk_pci_device_is_removed() has been added to let the upper-layer SPDK drivers know that device has a pending external hotremove request.
Dropped support for DPDK 17.07 and earlier, which SPDK won't even compile with right now.
spdk_env_fini() and spdk_env_dpdk_post_fini() were added to release any resources allocated by spdk_env_init() or spdk_env_dpdk_post_init() respectively. It is expected that common usage of those functions is to call them just before terminating the process.
get_spdk_version RPC method is introduced to get version info of the running SPDK application.
start_nbd_disk RPC method now take nbd_device as an optional parameter. If nbd_device is specified, use that specified nbd device. If it's not specified, pick available one.
Add Opal scan support for NVMe to check whether it supports SED Opal and dump device info. nvme_manage tool can be used to invoke this.
New virtual bdev module based on Open CAS Framework has been added. This module allows for the use of one bdev to act as a high performance cache in front of another bdev. Please see documentation for more details. Only write through mode is currently supported and this feature is considered experimental.
spdk_app_parse_args, add return value to the callback which parses application specific command line parameters to protect SPDK applications from crashing by invalid values from user input.
By default, all SPDK applications will now reserve all hugepages at runtime. The pre-reserved memory size can be still set with
--mem-size option, although the default value was reduced down to 0.
A custom hugetlbfs directory can now be specified via spdk_app_opts. This can be used to configure hugepages with different sizes, a different size limit, or different access permissions than the system's default hugepage pool. SPDK applications can specify a custom hugetlbfs mount with the
spdk_vtophys() has been refactored to accept length of the translated region as a new parameter. The function will now update that parameter with the largest possible value for which the memory is contiguous in the physical memory address space.
The following functions were removed:
spdk_pci_hook_device() and spdk_pci_unhook_device() were added. Those allow adding a virtual spdk_pci_device into the SPDK PCI subsystem. A virtual device calls provided callbacks for each BAR mapping request or PCI config access. It's attachable with spdk_pci_device_attach() or spdk_pci_enumerate() like any other device.
A new spdk_pause() function was added to pause CPU execution for an implementation specific amount of time. Quoting from DPDK function this is based on: "This call is intended for tight loops which poll a shared resource or wait for an event. A short pause within the loop may reduce the power consumption."
A new public header file env_dpdk.h has been introduced, and function spdk_env_dpdk_post_init is added into it. If user is using DPDK, and already called rte_eal_init, then include include/spdk/env_dpdk.h, and call spdk_env_dpdk_post_init() instead of spdk_env_init.
ISA-L has been added as an SPDK submodule. ISA-L is enabled by default on x86 architecture to accelerate algorithms such as CRC for iSCSI and NVMe-oF. Users may still disable ISA-L by explicitly passing –without-isal to the configure script.
A new uuid API
spdk_uuid_copy was added to make a copy of the source uuid.
An new parameter
init_crc representing the initial CRC value was added to
spdk_crc16_t10dif. The parameter can be used to calculate a CRC value spanning multiple separate buffers.
New DIF APIs were added to generate and verify DIF by byte granularity for both DIF and DIX formats. Among them, DIF with copy APIs will be usable to emulate DIF operations such as DIF insert and strip.
spdk_strtoll to provide additional error checking around
spdk_vsprintf_append_realloc for appending a string with automatic buffer re-allocation.
admin_timeout_ms was added to NVMe controller initialization options, users can change the default value when probing a controller.
Add two new fields "header_digest" and "data_digest" in struct spdk_nvme_ctrlr_opts, it will be used to enable the digest support for the NVMe/TCP transport.
Add a new TCP/IP transport(located in lib/nvme/nvme_tcp.c) in nvme driver. With this new transport, it can be used to connect the NVMe-oF target with the same TCP/IP support.
Added API, spdk_nvme_ctrlr_is_discovery(), to indicate whether the ctrlr arg refers to a Discovery Controller or not.
Added an API function
spdk_nvme_host_id_parse and corresponding object
spdk_nvme_host_id for parsing host address and host service ID arguments on a per connection basis.
construct_nvme_bdev now allows a user to specify a source address and service id for the host to use when connecting to the controller backing the NVMe bdev.
spdk_nvmf_tgt_opts struct has been deprecated in favor of
spdk_nvmf_transport_opts. Users will no longer be able to specify target wide I/O parameters.
spdk_nvmf_tgt_listen will also no longer implicitly initialize a transport with the default target options (since there are none). Instead, a user must manually instantiate the transport with
spdk_nvmf_transport_create prior to calling
Related to the previous change, the rpc
set_nvmf_target_options has been renamed to
set_nvmf_target_max_subsystems to indicate that this is the only target option available for the user to edit.
buf_cache_size in struct spdk_nvmf_transport_opts, and also updated the related rpc function nvmf_create_transport, to make this configurable parameter available to users. The
num_shared_buffers is used to configure the shared buffer numbers of the transport used by RDMA or TCP transport.
buf_cache_size configures number of shared buffers to cache per poll group.
Add a new TCP/IP transport (located in lib/nvmf/tcp.c). With this tranport, the SPDK NVMe-oF target can have a new transport, and can serve the NVMe-oF protocol via TCP/IP from the host.
Added optional mechanism to modify the RDMA transport's behavior when creating protection domains and registering memory. By default, the RDMA transport will use the ibverbs library to create protection domains and register memory. Using
spdk_nvme_rdma_init_hooks will subvert that and use an existing registration.
get_bdev_histogram RPC commands to allow gathering latency data for specified bdev. Please see documentation for more details.
required_alignment field to
spdk_bdev, that specifies an alignment requirement for data buffers associated with an spdk_bdev_io. Bdev layer will automatically double buffer any spdk_bdev_io that violates this alignment, before the spdk_bdev_io is submitted to the bdev module.
On shutdown, bdev unregister now proceeds in top-down fashion, with claimed bdevs skipped (these will be unregistered later, when virtual bdev built on top of the respective base bdev unclaims it). This allows virtual bdevs to be shut down cleanly as opposed to the previous behavior that didn't differentiate between hotremove and planned shutdown.
delete_bdev RPC is now deprecated. Users should instead use the specific deletion RPC for the bdev type to be removed (i.e. delete_malloc_bdev).
Added support for separate bandwidth rate limits for read and write to QoS in bdev layer.
Bdev I/O statistics now track unmap opertations.
Logical volume bdev can now be marked as read only using
set_read_only_lvol_bdev RPC. This allows for basing clones on top of lvol_bdev without first creating a snapshot.
Added option to change method for data erasure when deleting lvol or resizing down. Default of unmapping clusters can now be changed to writing zeroes or no operation.
Added option to change method for erasing data region on lvol store creation. Default of unmapping can now be changed to writing zeroes or no operation.
"trace flags" are now referred to as "log flags" in the SPDK log API. The set_trace_flag, clear_trace_flag and get_trace_flags RPCs are now deprecated, and set_log_flag, clear_log_flag and get_log_flags RPCs have been added.
trace_record application was added. It can be used to poll spdk trace shm file and append any new trace entries into another specified file. This can help retain those entries that would otherwise be overwritten in the shm file. See Capturing sufficient trace events for more details.
Number of trace entries in circular buffer per lcore can now be assigned by starting SPDK app with argument "--num-trace-entries <NUM>" provided.
get_tpoint_group_mask RPC was added to get current tpoint_group_mask, and each tpoint group status. New
disable_tpoint_group RPC were added to enable or disable a specific tpoint group.
EXPERIMENTAL: Added basic flash translation layer module allowing for using Open Channel SSDs as block devices. The module is split into the library (located in lib/ftl) and bdev_ftl (lib/bdev/ftl). See the documentation for more details.
A security vulnerability has been identified and fixed in the SPDK vhost target. A malicious vhost client (i.e. virtual machine) could carefully construct a circular descriptor chain which would result in a partial denial of service in the SPDK vhost target. These types of descriptor chains are now properly detected by the vhost target. All SPDK vhost users serving untrusted vhost clients are strongly recommended to upgrade. (Reported by Dima Stepanov and Evgeny Yakovlev.)
Vhost SCSI and Vhost Block devices can now accept multiple connections on the same socket file. Each connection (internally called a vhost session) will have access to the same storage, but will use different virtqueues, different features and possibly different memory.
SCSI target hotremove can now be performed even without the VIRTIO_SCSI_F_HOTPLUG feature negotiated. Regardless of VIRTIO_SCSI_F_HOTPLUG support, the hotremoval will be still reported through SCSI sense codes.
DPDK submodule was updated to DPDK 18.11. Note that SPDK does not fully leverage the new multi-process device hotplug yet and continues to work the same way it always did.
Dropped support for DPDK 16.07 and earlier, which SPDK won't even compile with right now.
The following RPC commands deprecated in the previous release are now removed:
The configure options
--without-raid that were deprecated in the previous release are now removed.
Starting nbd using
spdk_nbd_start is now performed asynchronously.
Net framework initialization and finish is now done asynchronously.
spdk_rpc_is_method_allowed function for checking whether method is permitted in a given state. Added
spdk_rpc_get_state to check current state of RPC server. RPC
wait_subsystem_init has been added to allow clients to block untill all subsystems are initialized.
JSON RPC client is now running in non-blocking mode. Requests are sent and received during spdk_jsonrpc_client_poll. JSON RPC server can now recieve a callback on connection termination or server shutdown using
spdk_nvme_ctrlr_cmd_security_send() and spdk_nvme_ctrlr_cmd_security_receive() were added to support sending or receiving security protocol data to or from nvme controller.
spdk_nvme_ns_get_extended_sector_size() was added. This function includes the metadata size per sector (if any). spdk_nvme_ns_get_sector_size() still returns only the data size per sector, not including metadata.
send_nvme_cmd RPC was added to allow sending NVMe commands directly to NVMe controller. See the send_nvme_cmd documentation for more details.
--without-shared [default], provide the capability to build, or not, SPDK shared libraries. This includes the single SPDK shared lib encompassing all of the SPDK static libs as well as individual SPDK shared libs corresponding to each of the SPDK static ones. Although the production of the shared libs conforms with conventional version naming practices, such naming does not at this time confer any SPDK ABI compatibility claims.
spdk_bdev_alias_del_all() was added to delete all alias from block device.
A new virtual bdev module has been added to perform at rest data encryption using the DPDK CryptoDev Framework. The module initially uses a software AESNI CBC cipher with experimental support for the Intel QAT hardware accelerator also currently implemented with support for CBC cipher. Future work may include additional ciphers as well as consideration for authentication.
The RAID virtual bdev module is now always enabled by default. The configure –with-raid and –without-raid options are now ignored and deprecated and will be removed in the next release.
Enforcement of bandwidth limits for quality of service (QoS) has been added to the bdev layer. See the new set_bdev_qos_limit documentation for more details. The previous set_bdev_qos_limit_iops RPC method introduced at 18.04 release has been deprecated. The new set_bdev_qos_limit RPC method can support both bandwidth and IOPS limits.
spdk_bdev_config_json() and corresponding
get_bdevs_config RPC was removed.
The size parameter of spdk_mem_map_translate is now a pointer. This allows the function to report back the actual size of the translation relative to the original request made by the user.
A new structure spdk_mem_map_ops has been introduced to hold memory map related callbacks. This structure is now passed as the second argument of spdk_mem_map_alloc in lieu of the notify callback.
The DPDK submodule has been updated to the DPDK 18.08 release. SPDK will now automatically utilize DPDK's dynamic memory management with DPDK versions >= 18.05.1.
Hugepages can be still reserved with
[-s|--mem-size <size>] option at application startup, but once we use them all up, instead of failing user allocations with -ENOMEM, we'll try to dynamically reserve even more. This allows starting SPDK with
--mem-size 0 and using only as many hugepages as it is really needed.
Due to this change, the memory buffers returned by
spdk_*malloc() are no longer guaranteed to be physically contiguous.
I/OAT driver can now reinitialize I/OAT channels after encountering DMA errors.
Parameter names of
get_iscsi_global_params RPC method for CHAP authentication in discovery sessions have been changed to align with
construct_target_node RPC method. Old names are still usable but will be removed in future release.
set_iscsi_target_node_auth RPC methods have been added to set CHAP authentication for discovery sessions and existing target nodes, respectively.
The SPDK iSCSI target supports an AuthFile which can be used to load CHAP shared secrets when the iSCSI target starts. SPDK previously provided a default location for this file (
/usr/local/etc/spdk/auth.conf) if none was specified. This default has been removed. Users must now explicitly specify the location of this file to load CHAP shared secrets from a file, or use the related iSCSI RPC methods to add them at runtime.
The SPDK iSCSI initiator is no longer considered experimental and becomes a first-class citizen among bdev modules. The basic usage has been briefly described in the bdev user guide: iSCSI bdev
The SPDK application framework can now parse long name command line parameters. Most single-character parameters have a long name equivalent now. See the Command Line Parameters documentation for details or use the
--help command line parameter to list all available params.
-s param (io size) was renamed to
-s had been already used by existing apps for memory size.
bdevio can now accept all SPDK command line parameters. The config now has to be provided with
The following ioat/perf and nvme/perf parameters were renamed as well:
-s (io size) to
-d (mem size) to
The ReactorMask config file parameter has been deprecated. Users should use the -m or –cpumask command line option to specify the CPU core mask for the application.
Default config file pathnames have been removed from iscsi_tgt, nvmf_tgt and vhost. Config file pathnames may now only be specified using the -c command line option.
Users may no longer set DPDK_DIR in their environment to specify the location of the DPDK installation used to build SPDK. Using DPDK_DIR has not been the documented nor recommended way to specify the DPDK location for several releases, but removing it ensures no unexpected surprises for users who may have DPDK_DIR defined for other reasons. Users should just use the "configure" script to specify the DPDK location before building SPDK.
Although we know that many developers still use Python 2 we are officially switching to Python3 with requirement that all new code must be valid also for Python 2 up to the EOL which is year 2020.
Invoking interpreter explicitly is forbidden for executable scripts. There is no need to use syntax like "python ./scripts/rpc.py". All executable scripts must contain proper shebang pointing to the right interpreter. Scripts without shebang musn't be executable.
A Python script has been added to enable conversion of old INI config file to new JSON-RPC config file format. This script can be found at scripts/config_converter.py. Example how this script can be used:
Two additional parameters were added to spdk_sock_get_addr() for the server port and client port. These parameters are named "sport" and "cport" respectively.
The following RPC commands have been deprecated:
construct_virtio_* ones were replaced with a single
construct_virtio_dev command that can create any type of Virtio bdev(s).
remove_virtio_scsi_bdev was replaced with
remove_virtio_bdev that can delete both Virtio Block and SCSI devices.
spdk_file_get_id() returning unique ID for the file was added.
Added jsonrpc-client C library intended for issuing RPC commands from applications.
Added API enabling iteration over JSON object:
Blobstore I/O operations are now based on io_units, instead of blobstore page size. The io_unit size is now the same as the underlying block device's block size. Logical volumes built on a block device with 512B block size can now be used as boot devices in QEMU.
The SPDKCLI interactive command tool for managing SPDK is no longer considered experimental. Support for the iSCSI and NVMe-oF targets has been added.
A new public header file bdev_module.h has been introduced to facilitate the development of new bdev modules. This header includes an interface for the spdk_bdev_part and spdk_bdev_part_base objects to enable the creation of multiple virtual bdevs on top of a single base bdev and should act as the primary API for module authors.
spdk_bdev_get_opts() and spdk_bdev_set_opts() were added to set bdev-wide options.
A mechanism for handling out of memory condition errors (ENOMEM) returned from I/O submission requests at the bdev layer has been added. See spdk_bdev_queue_io_wait().
The spdk_bdev_get_io_stat() function now returns cumulative totals instead of resetting on each call. This allows multiple callers to query I/O statistics without conflicting with each other. Existing users will need to adjust their code to record the previous I/O statistics to calculate the delta between calls.
A new bdev module called "raid" has been added as experimental module which aggregates underlying NVMe bdevs and exposes a single raid bdev. Please note that vhost will not work with this module because it does not yet have support for multi-element io vectors.
The debug log component flag available on several SPDK applications has been renamed from
-L to prevent confusion with tracepoints and to allow the option to be added to tools that already use
-t to mean something else.
A new function, spdk_bs_dump(), has been added that dumps all of the contents of a blobstore to a file pointer. This includes the metadata and is very useful for debugging.
Two new operations have been added for thin-provisioned blobs. spdk_bs_inflate_blob() will allocate clusters for all thinly provisioned regions of the blob and populate them with the correct data by reading from the backing blob(s). spdk_bs_blob_decouple_parent() works similarly, but will only allocate clusters that correspond to data in the blob's immediate parent. Clusters allocated to grandparents or that aren't allocated at all will remain thin-provisioned.
Changed the return type of spdk_file_truncate() from void to int to allow the propagation of
The new API functions spdk_nvme_qpair_add_cmd_error_injection() and spdk_nvme_qpair_remove_cmd_error_injection() have been added for NVMe error emulation. Users can set a specified command to fail with a particular error status.
Changed the name
timeout_sec parameter to
timeout_us in spdk_nvme_ctrlr_register_timeout_callback(), and also changed the type from uint32_t to uint64_t. This will give users more fine-grained control over the timeout period.
Basic support for Open Channel SSDs was added. See nvme_ocssd.h
The spdk_nvmf_tgt_destroy() function is now asynchronous and takes a callback as a parameter.
spdk_nvmf_qpair_disconnect() was added to allow the user to disconnect qpairs.
spdk_nvmf_subsystem_get_max_namespaces() was added to query the maximum allowed number of namespaces for a given subsystem.
The build system now generates a combined shared library (libspdk.so) that may be used in place of the individual static libraries (libspdk_*.a). The combined library includes all components of SPDK and is intended to make linking against SPDK easier. The static libraries are also still provided for users that prefer to link only the minimal set of components required.
The pre-commit hook will run
scripts/check_format.sh and verify there are no formating errors before allowing
git commit to run. The pre-push hook runs
make CONFIG_WERROR=y with and without
CONFIG_DEBUG=y using both the gcc and clang compiler before allowing
git push to run. Following each DEBUG build
test/unit/unittest.sh is run and verified. Results are recorded in the
To enable type: 'git config core.hooksPath .githooks'. To override after configuration use the
git --no-verify flag.
start_nbd_disk RPC method now returns the path to the kernel NBD device node rather than always returning
The DPDK submodule has been rebased on the DPDK 18.05 release. DPDK 18.05 supports dynamic memory allocation, but due to some issues found after the DPDK 18.05 release, that support is not enabled for SPDK 18.07. Therefore, SPDK 18.07 will continue to use the legacy memory allocation model. The plan is to enable dynamic memory allocation after the DPDK 18.08 release which should fix these issues.
The spdk_mem_map_translate() function now takes a size parameter to indicate the size of the memory region. This can be used by environment implementations to validate the requested translation.
The I/O Channel implementation has been moved to its own library - lib/thread. The public API that was previously in spdk/io_channel.h is now in spdk/thread.h The file spdk/io_channel.h remains and includes spdk/thread.h.
spdk_reactor_get_tsc_stats was added to return interesting statistics for each reactor.
IOAT for copy engine is disabled by default. It can be enabled by specifying the Enable option with "Yes" in
[Ioat] section of the configuration file. The Disable option is now deprecated and will be removed in a future release.
The SPDK vhost-scsi, vhost-blk and vhost-nvme applications have fixes to address the DPDK rte_vhost vulnerability CVE-2018-1059. Please see this security advisory for additional information on the DPDK vulnerability.
Workarounds have been added to ensure vhost compatibility with QEMU 2.12.
EXPERIMENTAL: Support for vhost-nvme has been added to the SPDK vhost target. See the vhost documentation for more details.
A new unified SPDK target application,
spdk_tgt, has been added. This application combines the functionality of several existing SPDK applications, including the iSCSI target, NVMe-oF target, and vhost target. The new application can be managed through the existing configuration file and JSON-RPC methods.
spdk_mempool_get_bulk() has been added to wrap DPDK rte_mempool_get_bulk().
New memory management functions spdk_malloc(), spdk_zmalloc(), and spdk_free() have been added. These new functions have a
flags parameter that allows the user to specify whether the allocated memory needs to be suitable for DMA and whether it should be shared across processes with the same shm_id. The new functions are intended to replace spdk_dma_malloc() and related functions, which will eventually be deprecated and removed.
A new optional bdev module interface function,
init_complete, has been added to notify bdev modules when the bdev subsystem initialization is complete. This may be useful for virtual bdevs that require notification that the set of initialization examine() calls is complete.
The bdev layer now allows modules to provide an optional per-bdev UUID, which can be retrieved with the spdk_bdev_get_uuid() function.
Enforcement of IOPS limits for quality of service (QoS) has been added to the bdev layer. See the set_bdev_qos_limit_iops documentation for more details.
[Rpc] configuration file section, which was deprecated in v18.01, has been removed. Users should switch to the
-r command-line parameter instead.
The JSON-RPC server implementation now allows up to 32 megabyte responses, growing as needed; previously, the response was limited to 32 kilobytes.
EXPERIMENTAL: New SPDKCLI interactive command tool for managing SPDK is available. See the SPDKCLI documentation for more details.
EXPERIMENTAL: Support for WDS and RDS capable CMBs in NVMe controllers has been added. This support is experimental pending a functional allocator to free and reallocate CMB buffers.
spdk_nvme_ns_get_uuid() has been added to allow retrieval of per-namespace UUIDs when available.
New API functions spdk_nvme_ctrlr_get_first_active_ns() and spdk_nvme_ctrlr_get_next_active_ns() have been added to iterate active namespaces, as well as spdk_nvme_ctrlr_is_active_ns() to check if a namespace ID is active.
Namespaces may now be assigned unique identifiers via new optional
nguid parameters to the
nvmf_subsystem_add_ns RPC method. Additionally, the NVMe-oF target automatically exposes the backing bdev's UUID as the namespace UUID when available.
spdk_nvmf_subsystem_remove_ns() is now asynchronous and requires a callback to indicate completion.
A number of functions have been renamed:
The old names still exist but are deprecated. They will be removed in the v18.07 release.
spdk_blob_resize() is now an asynchronous operation to enable resizing a blob while I/O are in progress to that blob on other threads. An explicit spdk_blob_sync_md() is still required to sync the updated metadata to disk.
destroy_lvol_bdev RPC method to delete logical volumes has been added.
Lvols now have their own UUIDs which replace previous LvolStoreUUID_BlobID combination.
New Snapshot and Clone functionalities have been added. User may create Snapshots of existing Lvols and Clones of existing Snapshots. See the lvol snapshots documentation for more details.
Resizing logical volumes is now supported via the
resize_lvol_bdev RPC method.
A set of changes were made in the SPDK's lib code altering instances of calls to
abort() to return a failure instead wherever reasonably possible.
spdk_app_start() no longer exit()'s on an internal failure, but instead returns a non-zero error status.
spdk_app_parse_args() no longer exit()'s on help, '-h', or an invalid option, but instead returns SPDK_APP_PARSE_ARGS_HELP and SPDK_APP_PARSE_ARGS_FAIL, respectively, and SPDK_APP_PARSE_ARGS_SUCCESS on success.
spdk_pci_get_device() has been deprecated and will be removed in SPDK v18.07.
The prototype for spdk_poller_fn() has been modified; it now returns a value indicating whether or not the poller did any work. Existing pollers will need to be updated to return a value.
The SPDK iSCSI target now supports the fd.io Vector Packet Processing (VPP) framework userspace TCP/IP stack. See the iSCSI VPP documentation for more details.
An iSCSI initiator bdev module has been added to SPDK. This module should be considered experimental pending additional features and tests. More details can be found in lib/bdev/iscsi/README.
The persistent memory (PMDK) bdev module is now enabled using
--with-pmdk instead of
--with-nvml. This reflects the renaming of the persistent memory library from NVML to PMDK.
A userspace driver for Virtio Block devices has been added. It was built on top of the Virtio library and can be managed similarly to the Virtio SCSI driver. See the Virtio Block reference for more information.
The previous 1GB hugepage limitation has now been lifted. A new
-g command-line option enables SPDK Virtio to work with 2MB hugepages. See 2MB hugepages for details.
The build system now includes a
make install rule, including support for the common
prefix variables as used in other build systems. Additionally, the prefix may be set via the configure
--prefix option. Example:
make install prefix=/usr.
A JSON RPC listener is now enabled by default using a UNIX domain socket at /var/run/spdk.sock. A -r option command line option has been added to enable an alternative UNIX domain socket location, or a TCP port in the format ip_addr:tcp_port (i.e. 127.0.0.1:5260). The Rpc configuration file section is now deprecated and will be removed in the v18.04 release.
spdk_poller_register() and spdk_poller_unregister() were moved from the event framework (include/spdk/event.h) to the I/O channel library (include/spdk/io_channel.h). This allows code that doesn't depend on the event framework to request registration and unregistration of pollers.
spdk_for_each_channel() now allows asynchronous operations during iteration. Instead of immediately continuing the interation upon returning from the iteration callback, the user must call spdk_for_each_channel_continue() to resume iteration.
The poller abstraction was removed from the bdev layer. There is now a general purpose abstraction for pollers available in include/spdk/io_channel.h
A set of changes were made in the SPDK's lib code altering, instances of calls to
abort() to return a failure instead wherever reasonably possible. This has resulted in return type changes of the API for:
Applications making use of these APIs should be modified to check for a non-zero return value instead of relying on them to fail without return.
SPDK now supports hotplug for vfio-attached devices. But there is one thing keep in mind: Only physical removal events are supported; removing devices via the sysfs
remove file will not work.
Subsystems are no longer tied explicitly to CPU cores. Instead, connections are handed out to the available cores round-robin. The "Core" option in the configuration file has been removed.
A number of functions have been renamed:
The function signature of spdk_blob_close() has changed. It now takes a struct spdk_blob * argument rather than struct spdk_blob **.
The function signature of spdk_bs_iter_next() has changed. It now takes a struct spdk_blob * argument rather than struct spdk_blob **.
The NBD application (test/lib/bdev/nbd) has been removed; Same functionality can now be achieved by using the test/app/bdev_svc application and start_nbd_disk RPC method. See the GPT documentation for more details.
fio_plugin now supports FIO 3.3. The support for previous FIO 2.21 has been dropped, although it still remains to work for now. The new FIO contains huge amount of bugfixes and it's recommended to do an update.
Previously a part of the bdev_virtio module, now a separate library. Virtio is now available via
spdk_internal/virtio.h file. This is an internal interface to be used when implementing new Virtio backends, namely Virtio-BLK.
The MinConnectionIdleInterval parameter has been removed, and connections are no longer migrated to an epoll/kqueue descriptor on the master core when idle.
libuuid was added as new dependency for logical volumes.
libnuma is now required unconditionally now that the DPDK submodule has been updated to DPDK 17.08.
spdk_bdev_unmap() was modified to take an offset and a length in bytes as arguments instead of requiring the user to provide an array of SCSI unmap descriptors. This limits unmaps to a single contiguous range.
spdk_bdev_write_zeroes() was introduced. It ensures that all specified blocks will be zeroed out. If a block device doesn't natively support a write zeroes command, the bdev layer emulates it using write commands.
New API functions that accept I/O parameters in units of blocks instead of bytes have been added:
The bdev layer now handles temporary out-of-memory I/O failures internally by queueing the I/O to be retried later.
The AIO bdev now allows the user to override the auto-detected block size.
The NVMe driver now recognizes the NVMe 1.3 Namespace Optimal I/O Boundary field. NVMe 1.3 devices may report an optimal I/O boundary, which the driver will take into account when splitting I/O requests.
The HotplugEnable option in
[Nvme] sections of the configuration file is now "No" by default. It was previously "Yes".
The NVMe library now includes a spdk_nvme_ns_get_ctrlr() function which returns the NVMe Controller associated with a given namespace.
The NVMe library now allows the user to specify a host identifier when attaching to a controller. The host identifier is used as part of the Reservations feature, as well as in the NVMe-oF Connect command. The default host ID is also now a randomly-generated UUID, and the default host NQN uses the host ID to generate a UUID-based NQN.
spdk_nvme_connect() was added to allow the user to connect directly to a single NVMe or NVMe-oF controller.
The NVMe-oF target no longer requires any in-capsule data buffers to run, and the feature is now entirely optional. Previously, at least 4 KiB in-capsule data buffers were required.
NVMe-oF subsytems have a new configuration option, AllowAnyHost, to control whether the host NQN whitelist is enforced when accepting new connections. If no Host options have been specified and AllowAnyHost is disabled, the connection will be denied; this is a behavior change from previous releases, which allowed any host NQN to connect if the Host list was empty. AllowAnyHost is disabled by default.
NVMe-oF namespaces may now be assigned arbitrary namespace IDs, and the number of namespaces per subsystem is no longer limited.
The NVMe-oF target now supports the Write Zeroes command.
A new default value, SPDK_MEMPOOL_DEFAULT_CACHE_SIZE, was added to provide additional clarity when constructing spdk_mempools. Previously, -1 could be passed and the library would choose a reasonable default, but this new value makes it explicit that the default is being used.
spdk_bs_destroy() was added to allow destroying blobstore on device with an initialized blobstore.
spdk_bs_io_readv_blob() and spdk_bs_io_writev_blob() were added to enable scattered payloads.
A CLI tool for blobstore has been added, allowing basic operations through either command line or shell interface. See the blobcli documentation for more details.
The ability to set a thread name, previously only used by the reactor code, is now part of the spdk_thread_allocate() API. Users may specify a thread name which will show up in tools like
The spdk_trace_dump() function now takes a new parameter to allow the caller to specify an output file handle (stdout or stderr, for example).
Logical volumes library built on top of SPDK blobstore has been added. It is possible to create logical volumes on top of other devices using RPC.
See the logical volumes documentation for more information.
A new persistent memory bdev type has been added. The persistent memory block device is built on top of libpmemblk. It is possible to create pmem devices on top of pmem pool files using RPC.
See the Pmem Block Device documentation for more information.
A userspace driver for Virtio SCSI devices has been added. The driver is capable of creating block devices on top of LUNs exposed by another SPDK vhost-scsi application.
The vhost target application now supports live migration between QEMU instances.
configure script has been added to simplify the build configuration process. The existing CONFIG file and
make CONFIG_... options are also still supported. Run
./configure --help for information about available configuration options.
A DPDK submodule has been added to make building SPDK easier. If no
--with-dpdk option is specified to configure, the SPDK build system will automatically build a known-good configuration of DPDK with the minimal options enabled. See the Building section of README.md for more information.
A Vagrant setup has been added to make it easier to develop and use SPDK on systems without suitable NVMe hardware. See the Vagrant section of README.md for more information.
The vhost library and example app have been updated to support the vhost-blk protocol in addition to the existing vhost-scsi protocol. See the vhost documentation for more details.
A GPT virtual block device has been added, which automatically exposes GPT partitions with a special SPDK-specific partition type as bdevs. See the GPT bdev documentation for more information.
The NVMe driver has been updated to support recent Intel SSDs, including the Intel® Optane™ SSD DC P4800X series.
A workaround has been added for devices that failed to recognize register writes during controller reset.
The NVMe driver now allocates request tracking objects on a per-queue basis. The number of requests allowed on an I/O queue may be set during
spdk_nvme_probe() by modifying
io_queue_requests in the opts structure.
The SPDK NVMe
fio_plugin has been updated to support multiple threads (
spdk_nvme_ctrlr_alloc_io_qpair() has been modified to allow the user to override controller-level options for each individual I/O queue pair. Existing callers with qprio == 0 can be updated to:
Callers that need to specify a non-default qprio should be updated to:
The environment abstraction layer has been updated to include several new functions in order to wrap additional DPDK functionality. See
include/spdk/env.h for the current set of functions.
Support for SPDK performance analysis has been added to Intel® VTune™ Amplifier 2018.
This analysis provides:
See the VTune Amplifier documentation for more information.
The blobstore is a persistent, power-fail safe block allocator designed to be used as the local storage system backing a higher-level storage service. See the blobstore documentation for more details.
BlobFS adds basic filesystem functionality like filenames on top of the blobstore. This release also includes a RocksDB Env implementation using BlobFS in place of the kernel filesystem. See the BlobFS documentation for more details.
A userspace implementation of the QEMU vhost-scsi protocol has been added. The vhost target is capable of exporting SPDK bdevs to QEMU-based VMs as virtio devices. See the vhost documentation for more details.
The overhead of the main reactor event loop was reduced by optimizing the number of calls to spdk_get_ticks() per iteration.
The NVMe library will now automatically split readv/writev requests with scatter-gather lists that do not map to valid PRP lists when the NVMe controller does not natively support SGLs.
perf NVMe examples were modified to add a consistent format for specifying remote NVMe over Fabrics devices via the
-r option. This is implemented using the new
The [Nvme] section of the configuration file was modified to remove the
BDF directive and replace it with a
TransportID directive. Both local (PCIe) and remote (NVMe-oF) devices can now be specified as the backing block device. A script to generate an entire [Nvme] section based on the local NVMe devices attached was added at
The [Nvme] section of the configuration file was modified to remove the
BDF directive and replace it with a
TransportID directive. Both local (PCIe) and remote (NVMe-oF) devices can now be specified as the backing block device. A script to generate an entire [Nvme] section based on the local NVMe devices attached was added at
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.
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_DSM_ATTR_DEALLOCATE as the
The NVMe library SGL callback prototype has been changed to return virtual addresses rather than physical addresses. Callers of
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
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.
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.
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
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
nvmf_tgt apps to export Ceph RBD volumes as iSCSI LUNs or NVMe namespaces.
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.
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
doc/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).
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
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.
SubsystemGroupwas renamed to
AuthFilewas removed (it was unimplemented).
nvmf_tgtwas 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".
Hostsections were merged into the
nvmf_tgtapp and library.
This release also adds one new feature and provides some better examples and tools for the NVMe driver.
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_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.
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.
examples/nvme/perfcan now report I/O command latency from the the controller's viewpoint using the Intel vendor-specific read/write latency log page.
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.
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.
kperffor performance comparison
nvme/identifysupport for Intel SSD DC P3700 series vendor-specific log pages and features
This is the initial open source release of the Storage Performance Development Kit (SPDK).
examples/nvme/perftests performance (IOPS) using the NVMe user-space driver
examples/nvme/identifydisplays NVMe controller information in a human-readable format