Solarflare NVMeoF over TCP Linux kernel drivers

From Define Wiki
Revision as of 11:01, 8 August 2018 by Matt hole (talk | contribs)
Jump to navigation Jump to search

Version: v18.1-rev3


Introduction

This is a pre-NVM Express specification release from Solarflare of a TCP transport for NVMe over Fabrics (NVMeoF). The package includes Linux kernel drivers which enable NVMeoF over TCP.


Supported platforms

This release is for Red Hat Enterprise Linux or Centos 7.4 only. Kernel and Devel of 3.10.0-693.el7.x86_64

Prerequisites

When using with Solarflare adapters, we recommend installing the latest Solarflare Linux NET driver for best performance.

The latest Solarflare driver packages are available from https://support.solarflare.com.

The installation of the NVMeoF over TCP drivers requires the kernel-devel package for the target kernel version to be installed on the system.

If you have installed Centos Minimum then you will need to install gcc rpmbuild

Installation

To build for the currently running kernel, use 

 rpmbuild --rebuild <source rpm file>

To install the resulting binary rpm: 

 Locate the file that was created in the preceding step, prefixed with the
 annotation "Wrote:", then type:

 rpm -Uvh <binary rpm filename>

Repeat the above installation procedure for both the host and target systems. The 'target' system is where the NVMe block device is located. The 'host' system is the system connecting to the device.

On the *host* system, load the nvme-tcp module 

 modprobe nvme-tcp

Check it has loaded 

 lsmod | grep nvme_tcp

You should see nvme_tcp, nvme_fabrics and nvme_core modules in use.

On the *target* system, load the nvmet-tcp module 

 modprobe nvmet-tcp

Check it has loaded 

 lsmod | grep nvmet_tcp

You should see nvmet_tcp and nvmet modules in use.


Configuration

On the *target* system, perform the steps below. The example commands assume a target IP of 10.0.0.1, a subsytem name of ramdisk and an underlying block device /dev/ram0. Users should specify their NVMe block device in place of this. (If you do wish to test with a ramdisk and /dev/ram0 is not present, then on RHEL 7 you may need to load the brd module. 'modprobe brd rd_size=<x>') 

 # Set up storage subsystem. 'ramdisk' is used below as an arbitrary name
 # for the NVMe subsystem.
 mkdir /sys/kernel/config/nvmet/subsystems/ramdisk
 echo 1 > /sys/kernel/config/nvmet/subsystems/ramdisk/attr_allow_any_host

 mkdir /sys/kernel/config/nvmet/subsystems/ramdisk/namespaces/1
 echo -n /dev/ram0 > /sys/kernel/config/nvmet/subsystems/ramdisk/namespaces/1/device_path
 echo 1 > /sys/kernel/config/nvmet/subsystems/ramdisk/namespaces/1/enable

 # Set up port
 mkdir /sys/kernel/config/nvmet/ports/1
 echo "ipv4" > /sys/kernel/config/nvmet/ports/1/addr_adrfam
 echo "tcp" > /sys/kernel/config/nvmet/ports/1/addr_trtype
 echo "11345" > /sys/kernel/config/nvmet/ports/1/addr_trsvcid
  

 # If a firewall is active on the system, ensure it allows TCP traffic
 # to pass using the TCP port number configured in the above commands
 # (11345 in this example).

 # Associate subsystem with port
 ln -s /sys/kernel/config/nvmet/subsystems/ramdisk /sys/kernel/config/nvmet/ports/1/subsystems/ramdisk

My version of the configuration when setting for the NVMe drive if showing in /dev/ as "nvme0n1"

touch /var/lock/subsys/local
mkdir /sys/kernel/config/nvmet/subsystems/nvme0n1
echo 1 > /sys/kernel/config/nvmet/subsystems/nvme0n1/attr_allow_any_host
mkdir /sys/kernel/config/nvmet/subsystems/nvme0n1/namespaces/1
echo -n /dev/nvme0n1 > /sys/kernel/config/nvmet/subsystems/nvme0n1/namespaces/1/device_path
echo 1 > /sys/kernel/config/nvmet/subsystems/nvme0n1/namespaces/1/enable
mkdir /sys/kernel/config/nvmet/ports/1
echo "ipv4" > /sys/kernel/config/nvmet/ports/1/addr_adrfam
echo "tcp" > /sys/kernel/config/nvmet/ports/1/addr_trtype
echo "11345" > /sys/kernel/config/nvmet/ports/1/addr_trsvcid
echo "192.168.0.10" > /sys/kernel/config/nvmet/ports/1/addr_traddr
ln -s /sys/kernel/config/nvmet/subsystems/nvme0n1 /sys/kernel/config/nvmet/ports/1/subsystems/nvme0n1

Install the latest Solarflare NET driver which itself needed to be compiled from source using rpmbuild.

Once installed setup the host and clients with static IPv4 addresses. In my example i made the target (storage server) 192.168.0.10 and the client 192.168.0.11 since it was a closed network.

On the *host* system, perform the steps below. The example commands assume a target IP of 192.168.0.10and port 11345. 

 # Install a version of the nvme-cli utility that supports NVMeoF over TCP.
 git clone https://github.com/solarflarecommunications/nvme-cli
 cd nvme-cli
 make
 make install

 # Connect to the target
 nvme connect -t tcp -a 192.168.0.10 -s 11345 -n nvme0n1

 # Confirm NVMe device is present
 lsblk | grep nvme

Make everything above stateful

Everything set above for the kernel is stateless having been set in /sys/ meaning once you reboot then your NVMe drive will go missing, the modules wont be loaded and none of the configuration above will be set anymore.

Target System:

Create a new service unit file at /etc/systemd/system/nvmet-tcp.service: 

[Unit]
Description=Service unit file for NVMe over Fabrics TCP target system
After=network.target
[Service]
Type=simple
ExecStart=/usr/local/sbin/nvmet-tcp.sh
TimeoutStartSec=0
[Install]
WantedBy=default.target

Create the ExecStart script outlined above: 

vim /usr/local/sbin/nvmet-tcp.sh

Make script executable: 

chmod a+x /usr/local/sbin/nvmet-tcp.sh

To get around this all of the configuration shown for nvme0n1 above in /etc/rc.d/rc.local then enabled/started the rc-local service.

However this failed since the nvmet kernel module thats required does not get loaded either and this will be need to be imported into the kernel before rc.local is run on boot.

In comes the following (this will load the module in on every boot): 

echo modprobe nvmet-tcp > /etc/rc.modules
chmod +x /etc/rc.modules
Retrieved from "http://wiki.define-technology.com/mediawiki-1.35.0/index.php?title=Solarflare_NVMeoF_over_TCP_Linux_kernel_drivers&oldid=27374"