The SPDK NVMe over Fabrics target is a user space application that presents block devices over a fabrics such as Ethernet, Infiniband or Fibre Channel. SPDK currently supports RDMA and TCP transports.
The NVMe over Fabrics specification defines subsystems that can be exported over different transports. SPDK has chosen to call the software that exports these subsystems a "target", which is the term used for iSCSI. The specification refers to the "client" that connects to the target as a "host". Many people will also refer to the host as an "initiator", which is the equivalent thing in iSCSI parlance. SPDK will try to stick to the terms "target" and "host" to match the specification.
The Linux kernel also implements an NVMe-oF target and host, and SPDK is tested for interoperability with the Linux kernel implementations.
If you want to kill the application using signal, make sure use the SIGTERM, then the application will release all the share memory resource before exit, the SIGKILL will make the share memory resource have no chance to be released by application, you may need to release the resource manually.
It requires an RDMA-capable NIC with its corresponding OFED (OpenFabrics Enterprise Distribution) software package installed to run. Maybe OS distributions provide packages, but OFED is also available here.
To build nvmf_tgt with the RDMA transport, there are some additional dependencies, which can be install using pkgdep.sh script.
Then build SPDK with RDMA enabled:
Once built, the binary will be in build/bin.
Before starting our NVMe-oF target with the RDMA transport we must load the InfiniBand and RDMA modules that allow userspace processes to use InfiniBand/RDMA verbs directly.
Before starting our NVMe-oF target we must detect RDMA NICs and assign them IP addresses.
As RDMA NICs put a limitation on the number of memory regions registered, the SPDK NVMe-oF target application may eventually start failing to allocate more DMA-able memory. This is an imperfection of the DPDK dynamic memory management and is most likely to occur with too many 2MB hugepages reserved at runtime. One type of memory bottleneck is the number of NIC memory regions, e.g., some NICs report as many as 2048 for the maximum number of memory regions. This gives us a 4GB memory limit with 2MB hugepages for the total memory regions. It can be overcome by using 1GB hugepages or by pre-reserving memory at application startup with --mem-size or -s option. All pre-reserved memory will be registered as a single region, but won't be returned to the system until the SPDK application is terminated.
Another known issue occurs when using the E810 NICs in RoCE mode. Specifically, the NVMe-oF target sometimes cannot destroy a qpair, because its posted work requests don't get flushed. It can cause the NVMe-oF target application unable to terminate cleanly.
The transport is built into the nvmf_tgt by default, and it does not need any special libraries.
To build nvmf_tgt with the FC transport, there is an additional FC LLD (Low Level Driver) code dependency. Please contact your FC vendor for instructions to obtain FC driver module.
FC LLD driver for Broadcom FC NVMe capable adapters can be obtained from, https://github.com/ecdufcdrvr/bcmufctdrvr.
After cloning SPDK repo and initialize submodules, FC LLD library is built which then can be linked with the fc transport.
An NVMe over Fabrics target can be configured using JSON RPCs. The basic RPCs needed to configure the NVMe-oF subsystem are detailed below. More information about working with NVMe over Fabrics specific RPCs can be found on the NVMe-oF Target RPC page.
Start the nvmf_tgt application with elevated privileges. Once the target is started, the nvmf_create_transport rpc can be used to initialize a given transport. Below is an example where the target is started and configured with two different transports. The RDMA transport is configured with an I/O unit size of 8192 bytes, max I/O size 131072 and an in capsule data size of 8192 bytes. The TCP transport is configured with an I/O unit size of 16384 bytes, 8 max qpairs per controller, and an in capsule data size of 8192 bytes.
Below is an example of creating a malloc bdev and assigning it to a subsystem. Adjust the bdevs, NQN, serial number, and IP address with RDMA transport to your own circumstances. If you replace "rdma" with "TCP", then the subsystem will add a listener with TCP transport.
NVMe qualified names or NQNs are defined in section 7.9 of the NVMe specification. SPDK has attempted to formalize that definition using Extended Backus-Naur form. SPDK modules use this formal definition (provided below) when validating NQNs.
Please note that the following types from the definition above are defined elsewhere:
While not stated in the formal definition, SPDK enforces the requirement from the spec that the "maximum name is 223 bytes in length". SPDK does not include the null terminating character when defining the length of an nqn, and will accept an nqn containing up to 223 valid bytes with an additional null terminator. To be precise, SPDK follows the same conventions as the c standard library function strlen().
SPDK compares NQNs byte for byte without case matching or unicode normalization. This has specific implications for uuid based NQNs. The following pair of NQNs, for example, would not match when compared in the SPDK NVMe-oF Target:
nqn.2014-08.org.nvmexpress:uuid:11111111-aaaa-bbdd-ffee-123456789abc nqn.2014-08.org.nvmexpress:uuid:11111111-AAAA-BBDD-FFEE-123456789ABC
In order to ensure the consistency of uuid based NQNs while using SPDK, users should use lowercase when representing alphabetic hex digits in their NQNs.
SPDK uses the DPDK Environment Abstraction Layer to gain access to hardware resources such as huge memory pages and CPU core(s). DPDK EAL provides functions to assign threads to specific cores. To ensure the SPDK NVMe-oF target has the best performance, configure the NICs and NVMe devices to be located on the same NUMA node.
The -m core mask option specifies a bit mask of the CPU cores that SPDK is allowed to execute work items on. For example, to allow SPDK to use cores 24, 25, 26 and 27:
Both the Linux kernel and SPDK implement an NVMe over Fabrics host. The Linux kernel NVMe-oF RDMA host support is provided by the nvme-rdma driver (to support RDMA transport) and nvme-tcp (to support TCP transport). And the following shows two different commands for loading the driver.
The nvme-cli tool may be used to interface with the Linux kernel NVMe over Fabrics host. See below for examples of the discover, connect and disconnect commands. In all three instances, the transport can be changed to TCP by interchanging 'rdma' for 'tcp'.
Discovery:
Connect:
Disconnect:
SPDK has a tracing framework for capturing low-level event information at runtime. Tracing Framework enable analysis of both performance and application crashes.
The SPDK NVMe-oF target and initiator support multiple independent paths to the same NVMe-oF subsystem. For step-by-step instructions for configuring and switching between paths, see NVMe-oF Multipath HOWTO .
The SPDK NVMe-oF target and initiator support establishing a secure TCP connection using Transport Layer Security (TLS) protocol in compliance with NVMe TCP transport specification. Only version 1.3 of the TLS protocol is supported. This feature is considered experimental.
Currently, it is only possible to establish a fabric secure channel using TLS. The channel is protected by a symmetric pre-shared key (PSK) using either TLS_AES_256_GCM_SHA384 (recommended) or TLS_AES_128_GCM_SHA256 cipher suite. The cipher suite is selected based on the hash function associated with a key. During configuration, the keys are expected to be in the PSK interchange format (see NVMe TCP transport specification 1.0c, section 3.6.1.5).
The target supports assigning different keys for each host connecting to a given subsystem. It is also possible for a single host to use different keys for different subsystems. The keys are expected to be placed in separate files (with permissions configured only to allow read/write access to the owner) and can be configured using the --psk option in the nvmf_subsystem_add_host RPC. Additionally, to allow establishing TLS connections on a given listener, it must be created with --secure-channel option enabled. It's also worth noting that this option is mutually exclusive with --allow-any-host subsystem option and trying to add a listener to such a subsystem will result in an error.
On the initiator side, the key can be specified using --psk option in the bdev_nvme_attach_controller RPC.
Recommendations on the pre-shared keys:
Additionally, it is recommended to follow: RFC 9257 'Guidance for External Pre-Shared Key (PSK) Usage in TLS'
For SPDK initiator example, bdevperf application may be used, because it depends on SPDK's NVMe TCP driver.
First of the two commands will launch bdevperf, the second one will attempt to construct NVMe bdev and establish TLS connection. Of course, the same PSK must be used on both the target and the initiator side.
The NVMe-oF driver and NVMe-oF target both support in-band authentication using the DH-HMAC-CHAP protocol. It allows the target to authenticate the host and the host to authenticate the target (the latter part is optional).
The authentication will be performed if a subsystem is configured to allow a host with a set of DH-HMAC-CHAP keys. Each host is allowed to use different keys to connect to different subsystems and each subsystem might use different keys for different hosts. For instance, the following configures three hosts, two of which can request bidirectional authentication:
Additionally, it's possible to change the keys while preserving existing connections to a subsystem via nvmf_subsystem_set_keys. After that's done, new connections and reauthentication requests will be required to use the new keys.
On the host side, the keys are specified when attaching controllers, e.g.:
All hash functions/Diffie-Hellman groups defined in the NVMe Base Specification 2.0d are supported and the algorithms used for a given DH-HMAC-CHAP transaction are negotiated at the beginning. The SPDK NVMe-oF target selects the strongest available hash/group depending on its configuration and the capabilities of a peer. Users can limit the allowed hash functions and/or Diffie-Hellman groups via RPCs. For example, the following limits the target (nvmf_set_config) and the driver (bdev_nvme_set_options) to use sha384, sha512 and ffdhe6114, ffdhe8192:
The NVMe specification describes the method for using in-band authentication in conjunction with establishing a secure channel (e.g. TLS). However, that isn't supported currently, so in order to perform in-band authentication, hosts must connect over regular listeners (i.e. those that weren't created with the --secure-channel option).