SCSI RDMA Protocol

The Linux SCSI Target Wiki

(Difference between revisions)
Jump to: navigation, search
m (See also)
m
 
(123 intermediate revisions not shown)
Line 1: Line 1:
{{Infobox software
{{Infobox software
-
| name                  = LIO Target
+
| name                  = {{Target}}
-
| logo                  = [[Image:Logo_Mellanox.png|180px|Logo]]
+
| logo                  = [[Image:Corp_Logo.png|180px|Logo]]
-
| screenshot            = Mellanox Technologies, Ltd.
+
| screenshot            = {{RTS screenshot|iSER}}
-
| caption                = Mellanox Infiniband SRP fabric module
+
| caption                = [[InfiniBand]] SRP fabric module
| collapsible            =  
| collapsible            =  
-
| author                = [http://www.linkedin.com/pub/vu-pham/6/93a/b40 Vu Pham]
+
| author                = {{Vu Pham}}<br/>{{Bart Van Assche}}<br/>{{Nicholas Bellinger}}
-
| developer              = Mellanox Technologies, Ltd.
+
| developer              = Mellanox Technologies, Ltd.<br/>{{RTS legal}}
| released              = {{RTS releases|SRP-Mellanox|initial_date}}
| released              = {{RTS releases|SRP-Mellanox|initial_date}}
| discontinued          = {{RTS releases|SRP-Mellanox|eol_date}}
| discontinued          = {{RTS releases|SRP-Mellanox|eol_date}}
Line 13: Line 13:
| latest preview version = {{RTS releases|SRP-Mellanox|preview_ver}}
| latest preview version = {{RTS releases|SRP-Mellanox|preview_ver}}
| latest preview date    = {{RTS releases|SRP-Mellanox|preview_date}}
| latest preview date    = {{RTS releases|SRP-Mellanox|preview_date}}
 +
| status                = {{RTS releases|SRP-Mellanox|status}}
| frequently updated    =
| frequently updated    =
| programming language  = C
| programming language  = C
Line 19: Line 20:
| size                  =
| size                  =
| language              =
| language              =
-
| status                = Beta
 
| genre                  = Fabric module
| genre                  = Fabric module
-
| license                = GNU General Public License
+
| license                = {{GPLv2}}
-
| website                = [http://www.mellanox.com/ www.mellanox.com]
+
| website                = {{RTS website}}
}}
}}
-
:''See [[Target]] for a complete overview over all fabric modules.''
+
:''See [[LIO]] for a complete overview over all fabric modules.''
-
An '''SCSI RDMA Protocol''' ('''SRP''', aka '''SCSI Remote Protocol''') fabric module (tcm_ib_srpt) for [http://www.mellanox.com Mellanox] HCAs is available upon request. [http://en.wikipedia.org/wiki/SCSI_RDMA_Protocol SRP] defines a SCSI protocol mapping onto the [http://en.wikipedia.org/wiki/InfiniBand Infiniband] architecture and/or functionally similar cluster protocols.
+
The '''SCSI RDMA Protocol''' ('''SRP''') is a network protocol that allows one computer system to access [[SCSI]] devices attached to another computer system via [[RDMA]].<ref name="SRP">[http://www.t10.org/drafts.htm#SCSI3_SRP ANSI T10 SRPr16a, www.t10.org].</ref><ref name="SRP-alternative">[http://web.archive.org/web/20080422013122/http://www.t10.org/ftp/t10/drafts/srp/srp-r16a.pdf ANSI T10 SRPr16a, web.archive.org]</ref>
-
The use of RDMA generally allows higher throughput and lower latency than TCP/IP based communication. RDMA is only possible with network adapters that support RDMA in hardware. Examples of such network adapters are InfiniBand HCAs and 10&nbsp;GbE network adapters with [[iWARP]] support. While the SRP protocol has been designed to use RDMA networks efficiently, it is also possible to implement the SRP protocol over networks that do not support RDMA.
+
== Overview ==
-
SRP is easier to implement than iSER, but iSER offers more management functionality, e.g. the target discovery infrastructure enabled by the iSCSI protocol.
+
SRP was designed to use RDMA networks efficiently. RDMA allows lower latencies and higher throughput than TCP/IP protocols, but requires network adapters with native RDMA support, e.g., [[HCA]]s for [[InfiniBand]], [[RNIC]]s on "lossless" ([[DCB]]) Ethernet, or [[TOE]] [[NIC]]s with [[iWARP]] for standard Ethernet.
-
== Status ==
+
SRP is based on the [[SCSI]] protocol, which is a point-to-point protocol with corresponding design limitations. In contrast, [[iSER]] is based on [[iSCSI]], and thus better accommodates modern network requirements, including complex topologies, multipathing, target discovery, etc. Hence, iSER is most likely the best choice for [[InfiniBand]] networks going forward.
-
The SRP driver is currently being tested and optimized.
+
SRP never became an official standard: the latest draft of the SRP protocol, revision 16a, dates from July 3, 2002.<ref name="SRP"/>
 +
 
 +
The InfiniBand/SRP fabric module ({{RTS releases|SRP-Mellanox|module_repo}}, {{RTS releases|SRP-Mellanox|module_info}}) for the {{Target}} was released with Linux kernel {{RTS releases|SRP-Mellanox|kernel_ver}} on {{RTS releases|SRP-Mellanox|initial_date}}.<ref>{{RTS releases|SRP-Mellanox|kernel_rel}}</ref>
 +
 
 +
== targetcli ==
 +
 
 +
''[[targetcli]]'' from {{RTS full}} is used to configure InfiniBand targets. ''targetcli'' aggregates {{T}} service modules via a core library, and exports them through an API, to provide a unified single-node SAN configuration shell, independently of the underlying fabric(s).
 +
 
 +
{{Ambox| type=info| head=[[LIO]] Admin Manual| text=The {{LIO Admin Manual}} provides comprehensive background and many examples on using ''targetcli'' and on programming the [[Datera]] library.}}
 +
 
 +
=== Cheat sheet ===
 +
 
 +
{| class="table-left"
 +
|- bgcolor=#a0c0f0
 +
! Command
 +
! Comment
 +
|-
 +
| ''/backstores/iblock create my_disk /dev/sdb''
 +
| Create the LUN ''my_disk'' on the block device ''/dev/sdb''
 +
|-
 +
| ''/ib_srpt create <WWPN>''
 +
| Create an SRP target
 +
|-
 +
| In /ib_srpt/<WWPN>:<br/>''luns/ create /backstores/iblock/my_disk''
 +
| Export the LUN ''my_disk''
 +
|-
 +
| In /ib_srpt/<WWPN>:<br/>''acls/ create <Initiator WWPN>''
 +
| Allow access for the initiator at ''<WWPN>''
 +
|-
 +
| ''/saveconfig''
 +
| Commit the configuration
 +
|-
 +
|}
 +
 
 +
=== Startup ===
 +
 
 +
[[targetcli]] is invoked by running ''targetcli'' as root from the command prompt of the underlying {{OS}} shell.
 +
 
 +
<pre>
 +
# targetcli
 +
Welcome to targetcli:
 +
 
 +
Copyright (c) 2014 by Datera, Inc.
 +
All rights reserved.
 +
 
 +
Visit us at http://www.datera.io.
 +
 
 +
Using ib_srpt fabric module.
 +
Using qla2xxx fabric module.
 +
Using iscsi fabric module.
 +
Using loopback fabric module.
 +
 
 +
/> ib_srpt/ info
 +
Fabric module name: ib_srpt
 +
ConfigFS path: /sys/kernel/config/target/srpt
 +
Allowed WWN list type: free
 +
Fabric module specfile: /var/target/fabric/ib_srpt.spec
 +
Fabric module features: acls
 +
Corresponding kernel module: ib_srpt
 +
/>
 +
</pre>
 +
 
 +
Upon targetcli initialization, the underlying RTSlib loads the installed fabric modules, and creates the corresponding [[ConfigFS]] mount points (at ''/sys/kernel/config/target/<fabric>''), as specified by the associated spec files (located in ''/var/target/fabric/fabric.spec'').
 +
 
 +
=== Display the object tree ===
 +
 
 +
Use ''ls'' to list the object hierarchy, which is initially empty:
 +
 
 +
<pre>
 +
/> ls
 +
o- / ..................................................................... [...]
 +
  o- backstores .......................................................... [...]
 +
  | o- fileio ............................................... [0 Storage Object]
 +
  | o- iblock ............................................... [0 Storage Object]
 +
  | o- pscsi ................................................ [0 Storage Object]
 +
  | o- rd_dr ................................................ [0 Storage Object]
 +
  | o- rd_mcp ............................................... [0 Storage Object]
 +
  o- ib_srpt ........................................................ [0 Target]
 +
  o- iscsi .......................................................... [0 Target]
 +
  o- loopback ....................................................... [0 Target]
 +
  o- qla2xxx ........................................................ [0 Target]
 +
/>
 +
</pre>
 +
 
 +
{{Ambox| type=info| head=Global parameter ''auto_cd_after_create''| text=After the creation of a new object, automatically enter its object context.}}
 +
 
 +
Per default, ''auto_cd_after_create'' is set to ''true'', which automatically enters an object context (or working directory) after its creation. The examples here are modeled after this behavior.
 +
 
 +
Optionally, set ''auto_cd_after_create=false'' to prevent ''targetcli'' from automatically entering new object contexts after their creation:
 +
 
 +
<pre>
 +
/> set global auto_cd_after_create=false
 +
Parameter auto_cd_after_create is now 'false'.
 +
/>
 +
</pre>
 +
 
 +
=== Create a backstore ===
 +
 
 +
Create a backstore using the IBLOCK or FILEIO type devices.
 +
 
 +
For instance, enter the top-level ''backstore'' context and create an IBLOCK backstore from a ''/dev/sdb'' block device:
 +
 
 +
<pre>
 +
/> cd backstores/
 +
/backstores> iblock/ create name=my_disk dev=/dev/sdb
 +
Generating a wwn serial.
 +
Created iblock storage object my_disk using /dev/sdb.
 +
Entering new node /backstores/iblock/my_disk.
 +
/backstores/iblock/my_disk>
 +
</pre>
 +
 
 +
''targetcli'' automatically creates a WWN serial ID for the backstore device and then changes the working context to it.
 +
 
 +
The resulting object hierarchy looks as follows (displayed from the root object):
 +
 
 +
<pre>
 +
/> ls
 +
o- / ..................................................................... [...]
 +
  o- backstores .......................................................... [...]
 +
  | o- fileio ............................................... [0 Storage Object]
 +
  | o- iblock ............................................... [1 Storage Object]
 +
  | | o- my_disk .......................................... [/dev/sdb activated]
 +
  | o- pscsi ................................................ [0 Storage Object]
 +
  | o- rd_dr ................................................ [0 Storage Object]
 +
  | o- rd_mcp ............................................... [0 Storage Object]
 +
  o- ib_srpt ........................................................ [0 Target]
 +
  o- iscsi .......................................................... [0 Target]
 +
  o- loopback ....................................................... [0 Target]
 +
  o- qla2xxx ........................................................ [0 Target]
 +
/>
 +
</pre>
 +
 
 +
Alternatively, any LVM logical volume can be used as a backstore, please refer to the {{LIO Admin Manual}} on how to create them properly.
 +
 
 +
For instance, create an IBLOCK backstore on a logical volume (under ''/dev/<volume_group_name>/<logical_volume_name>''):
 +
 
 +
<pre>
 +
/backstores> iblock/ create name=my_disk dev=/dev/vg0/lv1
 +
Generating a wwn serial.
 +
Created iblock storage object my_disk using /dev/vg0/lv1.
 +
Entering new node /backstores/iblock/my_disk.
 +
/backstores/iblock/my_disk>
 +
</pre>
 +
 
 +
Again, ''targetcli'' automatically creates a WWN serial ID for the backstore device and then changes the working context to it.
 +
 
 +
{{Ambox| type=info| head=More backstore examples| text=More examples on creating backstores can be found in ''[[targetcli#Create a backstore|targetcli]]''.}}
 +
 
 +
=== Instantiate a target ===
 +
 
 +
The InfiniBand ports that are available on the storage array are presented in the [[WWN]] context with the following WWPNs, for instance:
 +
 
 +
* 0x00000000000000000002c903000e8acd
 +
* 0x00000000000000000002c903000e8ace
 +
 
 +
Instantiate an InfiniBand target, in this example for SRP over Mellanox Connect-X HCAs, on the existing IBLOCK backstore device ''my_disk'' (as set up in [[targetcli]]):
 +
 
 +
<pre>
 +
/backstores/iblock/my_disk> /ip_srpt create 0x00000000000000000002c903000e8acd
 +
Created target 0x00000000000000000002c903000e8acd.
 +
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd.
 +
/ib_srpt/0x00...2c903000e8acd>
 +
</pre>
 +
 
 +
''targetcli'' automatically changes the working context to the resulting tagged Endpoint.
 +
 
 +
=== Export LUNs ===
 +
 
 +
Declare a LUN for the backstore device, to form a valid SAN storage object:
 +
 
 +
<pre>
 +
/ib_srpt/0x00...2c903000e8acd> luns/ create /backstores/iblock/my_disk
 +
Selected LUN 0.
 +
Successfully created LUN 0.
 +
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd/luns/lun0.
 +
/ib_srpt/0x00...acd/luns/lun0>
 +
</pre>
 +
 
 +
''targetcli'' per default automatically assigns the default ID '0' to the LUN, and then changes the working context to the SAN storage object. The target is now created, and exports ''/dev/sdb'' as LUN 0.
 +
 
 +
Return to the underlying Endpoint as the working context, as no attributes need to be set or modified for standard LUNs:
 +
 
 +
<pre>
 +
/ib_srpt/0x00...act/luns/lun0> cd <
 +
Taking you back to /ib_srpt/0x00000000000000000002c903000e8acd.
 +
/ib_srpt/0x00...2c903000e8acd>
 +
</pre>
 +
 
 +
=== Define access rights ===
 +
 
 +
Configure the access rights to allow logins from initiators. This requires setting up individual access rights for each initiator, based on its WWPN.
 +
 
 +
Determine the WWPN for the respective InfiniBand initiator. For instance, for Linux initiator systems, use:
 +
 
 +
<pre>
 +
# cat /sys/class/infiniband/*/ports/*/gids/0 | sed -e s/fe80/0x0000/ -e 's/\://g'
 +
</pre>
 +
 
 +
For a simple setup, allow access to the initiator with the WWPN as determined above:
 +
 
 +
<pre>
 +
/ib_srpt/0x00...2c903000e8acd> acls/ create 0x00000000000000000002c903000e8be9
 +
Successfully created Node ACL for 0x00000000000000000002c903000e8be9.
 +
Created mapped LUN 0.
 +
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd/acls/
 +
0x00000000000000000002c903000e8be9.
 +
/ib_srpt/0x00...2c903000e8be9> cd /
 +
</pre>
 +
 
 +
''targetcli'' per default automatically adds the appropriate mapped LUNs.
 +
 
 +
=== Display the object tree ===
 +
 
 +
The resulting InfiniBand SAN object hierarchy looks as follows (displayed from the root object):
 +
 
 +
<pre>
 +
/> ls
 +
o- / ..................................................................... [...]
 +
  o- backstores .......................................................... [...]
 +
  | o- fileio ............................................... [0 Storage Object]
 +
  | o- iblock ............................................... [1 Storage Object]
 +
  | | o- my_disk .......................................... [/dev/sdb activated]
 +
  | o- pscsi ................................................ [0 Storage Object]
 +
  | o- rd_dr ................................................ [0 Storage Object]
 +
  | o- rd_mcp ............................................... [0 Storage Object]
 +
  o- ib_srpt ........................................................ [1 Target]
 +
  | o- 0x00000000000000000002c903000e8acd ............................ [enabled]
 +
  |  o- acls .......................................................... [1 ACL]
 +
  |  | o- 0x00000000000000000002c903000e8be9 ................... [1 Mapped LUN]
 +
  |  |  o- mapped_lun0 ........................................... [lun0 (rw)]
 +
  |  o- luns .......................................................... [1 LUN]
 +
  |    o- lun0 .................................... [iblock/my_disk (/dev/sdb)]
 +
  o- iscsi .......................................................... [0 Target]
 +
  o- loopback ....................................................... [0 Target]
 +
  o- qla2xxx ........................................................ [0 Target]
 +
/>
 +
</pre>
 +
 
 +
=== Persist the configuration ===
 +
 
 +
{{Ambox| type=warning| head=Don't forget to use ''saveconfig''!| text=Without ''saveconfig'', the {{T}} configuration will be lost upon rebooting or unloading the target service, as the target configuration will revert back to the last saved one.}}
 +
 
 +
Use ''saveconfig'' from the root context to persist the {{T}} configuration across {{OS}} reboots:
 +
 
 +
<pre>
 +
/> saveconfig
 +
WARNING: Saving rtsnode1 current configuration to disk will overwrite your boot settings.
 +
The current target configuration will become the default boot config.
 +
Are you sure? Type 'yes': yes
 +
Making backup of srpt/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
 +
Successfully updated default config /etc/target/srpt_start.sh
 +
Making backup of qla2xxx/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
 +
Successfully updated default config /etc/target/qla2xxx_start.sh
 +
Making backup of loopback/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
 +
Successfully updated default config /etc/target/loopback_start.sh
 +
Making backup of LIO-Target/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
 +
Successfully updated default config /etc/target/lio_backup-2012-02-27_23:19:37.660264.sh
 +
Making backup of Target_Core_Mod/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
 +
Successfully updated default config /etc/target/tcm_backup-2012-02-27_23:19:37.660264.sh
 +
Generated Target_Core_Mod config: /etc/target/backup/tcm_backup-2012-02-27_23:19:37.660264.sh
 +
Successfully updated default config /etc/target/lio_start.sh
 +
Successfully updated default config /etc/target/tcm_start.sh
 +
/>
 +
</pre>
 +
 
 +
=== Spec file ===
 +
 
 +
[[Datera]] spec files define the fabric-dependent feature set, capabilities and available target ports of the specific underlying fabric.
 +
 
 +
In particular, the InfiniBand spec file ''/var/target/fabric/ib_srpt.spec'' is included via RTSlib. WWN values are extracted via ''/sys/class/infiniband/*/ports/*/gids/0'' in ''wwn_from_files_filter'' below, and are presented in the [[targetcli]] [[WWN]] context to register individual InfiniBand port GUIDs.
 +
 
 +
<pre>
 +
# WARNING: This is a draft specfile supplied for demo purposes only.
 +
 
 +
# The ib_srpt fabric module uses the default feature set.
 +
features = acls
 +
 
 +
# The module uses hardware addresses from there
 +
wwn_from_files = /sys/class/infiniband/*/ports/*/gids/0
 +
 
 +
# Transform 'fe80:0000:0000:0000:0002:1903:000e:8acd' WWN notation to
 +
# '0x00000000000000000002c903000e8acd'
 +
wwn_from_files_filter = "sed -e s/fe80/0x0000/ -e 's/\://g'"
 +
 
 +
# Non-standard module naming scheme
 +
kernel_module = ib_srpt
 +
 
 +
# The configfs group is standard
 +
configfs_group = srpt
 +
</pre>
 +
 
 +
== Scripting with RTSlib ==
 +
 
 +
=== Setup script ===
 +
 
 +
The following Python code illustrates how to setup a basic SRP target and export a mapped LUN:
 +
 
 +
<pre>
 +
#!/usr/bin/python
 +
# InfiniBand setup script example with RTSlib
 +
from rtslib import *
 +
 
 +
# Setup an IBLOCK backstore
 +
backstore = IBlockBackstore(3, mode='create')
 +
try:
 +
    so = IBlockStorageObject(backstore, "fioa", "/dev/fioa", gen_wwn=True)
 +
except:
 +
    backstore.delete()
 +
    raise
 +
 
 +
# Create an IB target endpoint using an ib_srpt WWPN
 +
fabric = FabricModule('ib_srpt')
 +
target = Target(fabric, '0x00000000000000000002c903000e8acd')
 +
tpg = TPG(target, 1)
 +
 
 +
# Export LUN 0 via the 'so' StorageObject class
 +
lun0 = tpg.lun(0, so, "my_lun")
 +
 
 +
# Setup the NodeACL for an IB initiator, and create MappedLUN 0
 +
node_acl = tpg.node_acl('0x00000000000000000002c903000e8acd')
 +
mapped_lun = node_acl.mapped_lun(0, 0, False)
 +
</pre>
 +
 
 +
Note that while SRP [[TPG]]s are masked by [[targetcli]], they are not masked by RTSlib.
 +
 
 +
=== Object tree ===
 +
 
 +
The resulting object tree looks as follows:
 +
 
 +
<pre>
 +
o- / ..................................................................... [...]
 +
  o- backstores .......................................................... [...]
 +
  | o- fileio ............................................... [0 Storage Object]
 +
  | o- iblock ............................................... [1 Storage Object]
 +
  | | o- my_disk .......................................... [/dev/sdb activated]
 +
  | o- pscsi ................................................ [0 Storage Object]
 +
  | o- rd_dr ................................................ [0 Storage Object]
 +
  | o- rd_mcp ............................................... [0 Storage Object]
 +
  o- ib_srpt ........................................................ [1 Target]
 +
    o- 0x00000000000000000002c903000e8acd ............................ [enabled]
 +
      o- acls .......................................................... [1 ACL]
 +
      | o- 0x00000000000000000002c903000e8be9 ................... [1 Mapped LUN]
 +
      |  o- mapped_lun0 ........................................... [lun0 (rw)]
 +
      o- luns .......................................................... [1 LUN]
 +
        o- lun0 .................................... [iblock/my_disk (/dev/sdb)]
 +
  o- iscsi .......................................................... [0 Target]
 +
  o- loopback ....................................................... [0 Target]
 +
  o- qla2xxx ........................................................ [0 Target]
 +
</pre>
== Specifications ==
== Specifications ==
Line 40: Line 388:
SRP was not approved as an official standard. The following specifications are available as available as [http://www.t10.org/drafts.htm T10 Working Drafts]:  
SRP was not approved as an official standard. The following specifications are available as available as [http://www.t10.org/drafts.htm T10 Working Drafts]:  
-
* '''SCSI RDMA Protocol''' ('''SRP'''): SRP defines a SCSI protocol mapping onto the InfiniBand (tm) Architecture and/or functionally similar cluster protocols. ANSI INCITS 365-2002. Status: Final Draft. 7/3/2002
+
* '''SCSI RDMA Protocol''' ('''SRP'''): SRP defines a SCSI protocol mapping onto the InfiniBand (tm) Architecture and/or functionally similar cluster protocols. ANSI INCITS 365-2002. Status: Final Draft. 2002-07-03
-
== Timeline ==
+
== RFCs ==
-
{{LIO Timeline}}
+
* {{RFC|4297|Remote Direct Memory Access (RDMA) over IP Problem Statement}}
 +
* {{RFC|4390|Dynamic Host Configuration Protocol (DHCP) over InfiniBand}}
 +
* {{RFC|4391|Transmission of IP over InfiniBand (IPoIB)}}
 +
* {{RFC|4392|IP over InfiniBand (IPoIB) Architecture}}
 +
* {{RFC|4755|IP over InfiniBand: Connected Mode}}
 +
* {{RFC|5040|A Remote Direct Memory Access Protocol Specification}}
 +
* {{RFC|5045|Applicability of Remote Direct Memory Access Protocol (RDMA) and Direct Data Placement Protocol (DDP)}}
 +
* {{RFC|5046|Internet Small Computer System Interface (iSCSI) Extensions for Remote Direct Memory Access (RDMA)}}
 +
* {{RFC|5047|DA: Datamover Architecture for the Internet Small Computer System Interface (iSCSI)}}
== See also ==
== See also ==
-
* [[ConfigFS]] low level kernel API: [[SCSI RDMA Protocol/configFS|SRP/configFS]]
+
* [[{{OS}}]]
-
* [[InfiniBand]], [[Target]]/ [[SCSI]]
+
* {{Target}}, [[targetcli]]
-
* Other fabric modules: [[iSCSI]], [[Fibre Channel]], [[Fibre Channel over Ethernet|FCoE]], [[tcm_loop]]
+
* [[FCoE]], [[Fibre Channel]], [[iSCSI]], [[iSER]], [[tcm_loop]], [[vHost]]
== Notes ==
== Notes ==
{{Reflist}}
{{Reflist}}
 +
 +
== Wikipedia entries ==
 +
* [http://en.wikipedia.org/wiki/InfiniBand InfiniBand] (IB)
 +
* [http://en.wikipedia.org/wiki/IWARP Internet Wide Area RDMA Protocol] (iWARP)
 +
* [http://en.wikipedia.org/wiki/Remote_direct_memory_access Remote direct memory access] (RDMA)
 +
* [http://en.wikipedia.org/wiki/SCSI_RDMA_Protocol SCSI RDMA Protocol] (SRP)
 +
* [http://en.wikipedia.org/wiki/Virtual_Interface_Architecture Virtual Interface Architecture] (VIA)
== External links ==
== External links ==
 +
* {{LIO Admin Manual}}
 +
* RTSlib Reference Guide {{Lib Ref Guide HTML}}{{Lib Ref Guide PDF}}
* [http://en.wikipedia.org/wiki/SCSI_RDMA_Protocol SRP] Wikipedia entry
* [http://en.wikipedia.org/wiki/SCSI_RDMA_Protocol SRP] Wikipedia entry
-
* [http://en.wikipedia.org/wiki/Infiniband InfiniBand] Wikipedia entry
+
* {{cite web| url=http://www.youtube.com/watch?v=4UVRuq-lgKo| title=SRP Update and Directions| date=2012-05-27| publisher=openfabrics.org}}
 +
* [http://marc.info/?l=linux-rdma&r=1&w=2 linux-rdma] mailing list
 +
* [http://www.openfabrics.org/ OpenFabrics Alliance]
* [http://www.mellanox.com/ Mellanox] website
* [http://www.mellanox.com/ Mellanox] website
* [http://www.t10.org/index.html T10] home page
* [http://www.t10.org/index.html T10] home page
 +
 +
{{LIO Timeline}}
[[Category:Fabric modules]]
[[Category:Fabric modules]]
[[Category:InfiniBand]]
[[Category:InfiniBand]]
[[Category:Network protocols]]
[[Category:Network protocols]]

Latest revision as of 02:42, 7 August 2015

LinuxIO
Logo
LIO 150513.png
InfiniBand SRP fabric module
Original author(s) Vu Pham
Bart Van Assche
Nicholas Bellinger
Developer(s) Mellanox Technologies, Ltd.
Datera, Inc.
Initial release March 18, 2012 (2012-03-18)
Stable release 4.1.0 / June 20, 2012;
7 years ago
 (2012-06-20)
Preview release 4.2.0-rc5 / June 28, 2012;
7 years ago
 (2012-06-28)
Development status Production
Written in C
Operating system Linux
Type Fabric module
License GNU General Public License, version 2 (GPLv2)
Website datera.io
See LIO for a complete overview over all fabric modules.

The SCSI RDMA Protocol (SRP) is a network protocol that allows one computer system to access SCSI devices attached to another computer system via RDMA.[1][2]

Contents

Overview

SRP was designed to use RDMA networks efficiently. RDMA allows lower latencies and higher throughput than TCP/IP protocols, but requires network adapters with native RDMA support, e.g., HCAs for InfiniBand, RNICs on "lossless" (DCB) Ethernet, or TOE NICs with iWARP for standard Ethernet.

SRP is based on the SCSI protocol, which is a point-to-point protocol with corresponding design limitations. In contrast, iSER is based on iSCSI, and thus better accommodates modern network requirements, including complex topologies, multipathing, target discovery, etc. Hence, iSER is most likely the best choice for InfiniBand networks going forward.

SRP never became an official standard: the latest draft of the SRP protocol, revision 16a, dates from July 3, 2002.[1]

The InfiniBand/SRP fabric module (srpt.ko, Linux kernel driver database) for the LinuxIO was released with Linux kernel 3.3 on March 18, 2012 (2012-03-18).[3]

targetcli

targetcli from Datera, Inc. is used to configure InfiniBand targets. targetcli aggregates LIO service modules via a core library, and exports them through an API, to provide a unified single-node SAN configuration shell, independently of the underlying fabric(s).

Cheat sheet

Command Comment
/backstores/iblock create my_disk /dev/sdb Create the LUN my_disk on the block device /dev/sdb
/ib_srpt create <WWPN> Create an SRP target
In /ib_srpt/<WWPN>:
luns/ create /backstores/iblock/my_disk
Export the LUN my_disk
In /ib_srpt/<WWPN>:
acls/ create <Initiator WWPN>
Allow access for the initiator at <WWPN>
/saveconfig Commit the configuration

Startup

targetcli is invoked by running targetcli as root from the command prompt of the underlying LIO shell.

# targetcli
Welcome to targetcli:

 Copyright (c) 2014 by Datera, Inc.
 All rights reserved.

Visit us at http://www.datera.io.

Using ib_srpt fabric module.
Using qla2xxx fabric module.
Using iscsi fabric module.
Using loopback fabric module.

/> ib_srpt/ info
Fabric module name: ib_srpt
ConfigFS path: /sys/kernel/config/target/srpt
Allowed WWN list type: free
Fabric module specfile: /var/target/fabric/ib_srpt.spec
Fabric module features: acls
Corresponding kernel module: ib_srpt
/>

Upon targetcli initialization, the underlying RTSlib loads the installed fabric modules, and creates the corresponding ConfigFS mount points (at /sys/kernel/config/target/<fabric>), as specified by the associated spec files (located in /var/target/fabric/fabric.spec).

Display the object tree

Use ls to list the object hierarchy, which is initially empty:

/> ls
o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- fileio ............................................... [0 Storage Object]
  | o- iblock ............................................... [0 Storage Object]
  | o- pscsi ................................................ [0 Storage Object]
  | o- rd_dr ................................................ [0 Storage Object]
  | o- rd_mcp ............................................... [0 Storage Object]
  o- ib_srpt ........................................................ [0 Target]
  o- iscsi .......................................................... [0 Target]
  o- loopback ....................................................... [0 Target]
  o- qla2xxx ........................................................ [0 Target]
/>

Per default, auto_cd_after_create is set to true, which automatically enters an object context (or working directory) after its creation. The examples here are modeled after this behavior.

Optionally, set auto_cd_after_create=false to prevent targetcli from automatically entering new object contexts after their creation:

/> set global auto_cd_after_create=false
Parameter auto_cd_after_create is now 'false'.
/>

Create a backstore

Create a backstore using the IBLOCK or FILEIO type devices.

For instance, enter the top-level backstore context and create an IBLOCK backstore from a /dev/sdb block device:

/> cd backstores/
/backstores> iblock/ create name=my_disk dev=/dev/sdb
Generating a wwn serial.
Created iblock storage object my_disk using /dev/sdb.
Entering new node /backstores/iblock/my_disk.
/backstores/iblock/my_disk>

targetcli automatically creates a WWN serial ID for the backstore device and then changes the working context to it.

The resulting object hierarchy looks as follows (displayed from the root object):

/> ls
o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- fileio ............................................... [0 Storage Object]
  | o- iblock ............................................... [1 Storage Object]
  | | o- my_disk .......................................... [/dev/sdb activated]
  | o- pscsi ................................................ [0 Storage Object]
  | o- rd_dr ................................................ [0 Storage Object]
  | o- rd_mcp ............................................... [0 Storage Object]
  o- ib_srpt ........................................................ [0 Target]
  o- iscsi .......................................................... [0 Target]
  o- loopback ....................................................... [0 Target]
  o- qla2xxx ........................................................ [0 Target]
/>

Alternatively, any LVM logical volume can be used as a backstore, please refer to the LIO Admin Manual on how to create them properly.

For instance, create an IBLOCK backstore on a logical volume (under /dev/<volume_group_name>/<logical_volume_name>):

/backstores> iblock/ create name=my_disk dev=/dev/vg0/lv1
Generating a wwn serial.
Created iblock storage object my_disk using /dev/vg0/lv1.
Entering new node /backstores/iblock/my_disk.
/backstores/iblock/my_disk>

Again, targetcli automatically creates a WWN serial ID for the backstore device and then changes the working context to it.

Instantiate a target

The InfiniBand ports that are available on the storage array are presented in the WWN context with the following WWPNs, for instance:

Instantiate an InfiniBand target, in this example for SRP over Mellanox Connect-X HCAs, on the existing IBLOCK backstore device my_disk (as set up in targetcli):

/backstores/iblock/my_disk> /ip_srpt create 0x00000000000000000002c903000e8acd
Created target 0x00000000000000000002c903000e8acd.
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd.
/ib_srpt/0x00...2c903000e8acd>

targetcli automatically changes the working context to the resulting tagged Endpoint.

Export LUNs

Declare a LUN for the backstore device, to form a valid SAN storage object:

/ib_srpt/0x00...2c903000e8acd> luns/ create /backstores/iblock/my_disk
Selected LUN 0.
Successfully created LUN 0.
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd/luns/lun0.
/ib_srpt/0x00...acd/luns/lun0>

targetcli per default automatically assigns the default ID '0' to the LUN, and then changes the working context to the SAN storage object. The target is now created, and exports /dev/sdb as LUN 0.

Return to the underlying Endpoint as the working context, as no attributes need to be set or modified for standard LUNs:

/ib_srpt/0x00...act/luns/lun0> cd <
Taking you back to /ib_srpt/0x00000000000000000002c903000e8acd.
/ib_srpt/0x00...2c903000e8acd>

Define access rights

Configure the access rights to allow logins from initiators. This requires setting up individual access rights for each initiator, based on its WWPN.

Determine the WWPN for the respective InfiniBand initiator. For instance, for Linux initiator systems, use:

# cat /sys/class/infiniband/*/ports/*/gids/0 | sed -e s/fe80/0x0000/ -e 's/\://g'

For a simple setup, allow access to the initiator with the WWPN as determined above:

/ib_srpt/0x00...2c903000e8acd> acls/ create 0x00000000000000000002c903000e8be9
Successfully created Node ACL for 0x00000000000000000002c903000e8be9.
Created mapped LUN 0.
Entering new node /ib_srpt/0x00000000000000000002c903000e8acd/acls/
0x00000000000000000002c903000e8be9.
/ib_srpt/0x00...2c903000e8be9> cd /

targetcli per default automatically adds the appropriate mapped LUNs.

Display the object tree

The resulting InfiniBand SAN object hierarchy looks as follows (displayed from the root object):

/> ls
o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- fileio ............................................... [0 Storage Object]
  | o- iblock ............................................... [1 Storage Object]
  | | o- my_disk .......................................... [/dev/sdb activated]
  | o- pscsi ................................................ [0 Storage Object]
  | o- rd_dr ................................................ [0 Storage Object]
  | o- rd_mcp ............................................... [0 Storage Object]
  o- ib_srpt ........................................................ [1 Target]
  | o- 0x00000000000000000002c903000e8acd ............................ [enabled]
  |   o- acls .......................................................... [1 ACL]
  |   | o- 0x00000000000000000002c903000e8be9 ................... [1 Mapped LUN]
  |   |   o- mapped_lun0 ........................................... [lun0 (rw)]
  |   o- luns .......................................................... [1 LUN]
  |     o- lun0 .................................... [iblock/my_disk (/dev/sdb)]
  o- iscsi .......................................................... [0 Target]
  o- loopback ....................................................... [0 Target]
  o- qla2xxx ........................................................ [0 Target]
/>

Persist the configuration

Use saveconfig from the root context to persist the LIO configuration across LIO reboots:

/> saveconfig
WARNING: Saving rtsnode1 current configuration to disk will overwrite your boot settings.
The current target configuration will become the default boot config.
Are you sure? Type 'yes': yes
Making backup of srpt/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
Successfully updated default config /etc/target/srpt_start.sh
Making backup of qla2xxx/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
Successfully updated default config /etc/target/qla2xxx_start.sh
Making backup of loopback/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
Successfully updated default config /etc/target/loopback_start.sh
Making backup of LIO-Target/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
Successfully updated default config /etc/target/lio_backup-2012-02-27_23:19:37.660264.sh
Making backup of Target_Core_Mod/ConfigFS with timestamp: 2012-02-27_23:19:37.660264
Successfully updated default config /etc/target/tcm_backup-2012-02-27_23:19:37.660264.sh
Generated Target_Core_Mod config: /etc/target/backup/tcm_backup-2012-02-27_23:19:37.660264.sh
Successfully updated default config /etc/target/lio_start.sh
Successfully updated default config /etc/target/tcm_start.sh
/>

Spec file

Datera spec files define the fabric-dependent feature set, capabilities and available target ports of the specific underlying fabric.

In particular, the InfiniBand spec file /var/target/fabric/ib_srpt.spec is included via RTSlib. WWN values are extracted via /sys/class/infiniband/*/ports/*/gids/0 in wwn_from_files_filter below, and are presented in the targetcli WWN context to register individual InfiniBand port GUIDs.

# WARNING: This is a draft specfile supplied for demo purposes only.

# The ib_srpt fabric module uses the default feature set.
features = acls

# The module uses hardware addresses from there
wwn_from_files = /sys/class/infiniband/*/ports/*/gids/0

# Transform 'fe80:0000:0000:0000:0002:1903:000e:8acd' WWN notation to
# '0x00000000000000000002c903000e8acd'
wwn_from_files_filter = "sed -e s/fe80/0x0000/ -e 's/\://g'"

# Non-standard module naming scheme
kernel_module = ib_srpt

# The configfs group is standard
configfs_group = srpt

Scripting with RTSlib

Setup script

The following Python code illustrates how to setup a basic SRP target and export a mapped LUN:

#!/usr/bin/python
# InfiniBand setup script example with RTSlib
from rtslib import *

# Setup an IBLOCK backstore
backstore = IBlockBackstore(3, mode='create')
try:
    so = IBlockStorageObject(backstore, "fioa", "/dev/fioa", gen_wwn=True)
except:
    backstore.delete()
    raise

# Create an IB target endpoint using an ib_srpt WWPN
fabric = FabricModule('ib_srpt')
target = Target(fabric, '0x00000000000000000002c903000e8acd')
tpg = TPG(target, 1)

# Export LUN 0 via the 'so' StorageObject class
lun0 = tpg.lun(0, so, "my_lun")

# Setup the NodeACL for an IB initiator, and create MappedLUN 0
node_acl = tpg.node_acl('0x00000000000000000002c903000e8acd')
mapped_lun = node_acl.mapped_lun(0, 0, False)

Note that while SRP TPGs are masked by targetcli, they are not masked by RTSlib.

Object tree

The resulting object tree looks as follows:

o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- fileio ............................................... [0 Storage Object]
  | o- iblock ............................................... [1 Storage Object]
  | | o- my_disk .......................................... [/dev/sdb activated]
  | o- pscsi ................................................ [0 Storage Object]
  | o- rd_dr ................................................ [0 Storage Object]
  | o- rd_mcp ............................................... [0 Storage Object]
  o- ib_srpt ........................................................ [1 Target]
    o- 0x00000000000000000002c903000e8acd ............................ [enabled]
      o- acls .......................................................... [1 ACL]
      | o- 0x00000000000000000002c903000e8be9 ................... [1 Mapped LUN]
      |   o- mapped_lun0 ........................................... [lun0 (rw)]
      o- luns .......................................................... [1 LUN]
        o- lun0 .................................... [iblock/my_disk (/dev/sdb)]
  o- iscsi .......................................................... [0 Target]
  o- loopback ....................................................... [0 Target]
  o- qla2xxx ........................................................ [0 Target]

Specifications

SRP was not approved as an official standard. The following specifications are available as available as T10 Working Drafts:

RFCs

See also

Notes

  1. a b ANSI T10 SRPr16a, www.t10.org.
  2. ANSI T10 SRPr16a, web.archive.org
  3. Linus Torvalds (2012-03-18). "Linux 3.3". lkml.org. 

Wikipedia entries

External links

Timeline of the LinuxIO
Release Details 2011 2012 2013 2014 2015
123456789101112 123456789101112 123456789101112 123456789101112 123456789101112
4.x Version 4.0 4.1
Feature LIO Core Loop back FCoE iSCSI Perf SRP
CM WQ FC
USB
1394
vHost Perf Misc 16 GFC iSER Misc VAAI Misc DIF Core
NPIV
DIF iSER DIF FC vhost TCMU Xen Misc Misc virtio 1.0 Misc NVMe OF
Linux 2.6.38 2.6.39 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 3.20 3.21 3.22
Personal tools
Namespaces
Variants
Actions
Navigation
Toolbox
Google AdSense