IPMI driver¶
Overview¶
The ipmi
hardware type manage nodes by using IPMI (Intelligent Platform
Management Interface) protocol versions 2.0 or 1.5. It uses the IPMItool
utility which is an open-source command-line interface (CLI) for controlling
IPMI-enabled devices.
Glossary¶
Enabling the IPMI hardware type¶
Please see Configuring IPMI support for the required dependencies.
The
ipmi
hardware type is enabled by default starting with the Ocata release. To enable it explicitly, add the following to yourironic.conf
:[DEFAULT] enabled_hardware_types = ipmi enabled_management_interfaces = ipmitool,noop enabled_power_interfaces = ipmitool
Optionally, enable the vendor passthru interface and either or both console interfaces:
[DEFAULT] enabled_hardware_types = ipmi enabled_console_interfaces = ipmitool-socat,ipmitool-shellinabox,no-console enabled_management_interfaces = ipmitool,noop enabled_power_interfaces = ipmitool enabled_vendor_interfaces = ipmitool,no-vendor
Restart the Ironic conductor service.
Please see Enabling drivers and hardware types for more details.
Registering a node with the IPMI driver¶
Nodes configured to use the IPMItool drivers should have the driver
field
set to ipmi
.
The following configuration value is required and has to be added to
the node’s driver_info
field:
ipmi_address
: The IP address or hostname of the BMC.
Other options may be needed to match the configuration of the BMC, the following options are optional, but in most cases, it’s considered a good practice to have them set:
ipmi_username
: The username to access the BMC; defaults to NULL user.ipmi_password
: The password to access the BMC; defaults to NULL.ipmi_port
: The remote IPMI RMCP port. By default ipmitool will use the port 623.
Note
It is highly recommend that you setup a username and password for your BMC.
The baremetal node create
command can be used to enroll a node
with an IPMItool-based driver. For example:
baremetal node create --driver ipmi \
--driver-info ipmi_address=<address> \
--driver-info ipmi_username=<username> \
--driver-info ipmi_password=<password>
Changing The Default IPMI Credential Persistence Method¶
store_cred_in_env
:ipmi.store_cred_in_env
.
The store_cred_in_env
configuration option allow users to switch
between file-based and environment variable persistence methods for
IPMI password.
For the temporary file option, long lived IPMI sessions, such as those for console support, leave files with credentials on the conductor disk for the duration of the session.
To switch to environment variable persistence, set the
store_cred_in_env
parameter to True
in the configuration file:
[ipmi]
store_cred_in_env = True
Advanced configuration¶
When a simple configuration such as providing the address
,
username
and password
is not enough, the IPMItool driver contains
many other options that can be used to address special usages.
Single/Double bridging functionality¶
Note
A version of IPMItool higher or equal to 1.8.12 is required to use the bridging functionality.
There are two different bridging functionalities supported by the IPMItool-based drivers: single bridge and dual bridge.
The following configuration values need to be added to the node’s
driver_info
field so bridging can be used:
ipmi_bridging
: The bridging type; default is no; other supported values are single for single bridge or dual for double bridge.ipmi_local_address
: The local IPMB address for bridged requests.Required only if
ipmi_bridging
is set to single or dual. This configuration is optional, if not specified it will be auto discovered by IPMItool.
ipmi_target_address
: The destination address for bridged requests. Required only ifipmi_bridging
is set to single or dual.ipmi_target_channel
: The destination channel for bridged requests. Required only ifipmi_bridging
is set to single or dual.
Double bridge specific options:
ipmi_transit_address
: The transit address for bridged requests. Required only ifipmi_bridging
is set to dual.ipmi_transit_channel
: The transit channel for bridged requests. Required only ifipmi_bridging
is set to dual.
The parameter ipmi_bridging
should specify the type of bridging
required: single or dual to access the bare metal node. If the
parameter is not specified, the default value will be set to no.
The baremetal node set
command can be used to set the required
bridging information to the Ironic node enrolled with the IPMItool
driver. For example:
Single Bridging:
baremetal node set <UUID or name> \ --driver-info ipmi_local_address=<address> \ --driver-info ipmi_bridging=single \ --driver-info ipmi_target_channel=<channel> \ --driver-info ipmi_target_address=<target address>
Double Bridging:
baremetal node set <UUID or name> \ --driver-info ipmi_local_address=<address> \ --driver-info ipmi_bridging=dual \ --driver-info ipmi_transit_channel=<transit channel> \ --driver-info ipmi_transit_address=<transit address> \ --driver-info ipmi_target_channel=<target channel> \ --driver-info ipmi_target_address=<target address>
Changing the version of the IPMI protocol¶
The IPMItool-based drivers works with the versions 2.0 and 1.5 of the IPMI protocol. By default, the version 2.0 is used.
In order to change the IPMI protocol version in the bare metal node,
the following option needs to be set to the node’s driver_info
field:
ipmi_protocol_version
: The version of the IPMI protocol; default is 2.0. Supported values are 1.5 or 2.0.
The baremetal node set
command can be used to set the desired
protocol version:
baremetal node set <UUID or name> --driver-info ipmi_protocol_version=<version>
Warning
Version 1.5 of the IPMI protocol does not support encryption. Therefore, it is highly recommended that version 2.0 is used.
Cipher suites¶
IPMI 2.0 introduces support for encryption and allows setting which cipher
suite to use. Traditionally, ipmitool
was using cipher suite 3 by default,
but since SHA1 no longer complies with modern security requirement, recent
versions (e.g. the one used in RHEL 8.2) are switching to suite 17.
Normally, the cipher suite to use is negotiated with the BMC using the special command. On some hardware the negotiation yields incorrect results and IPMI commands fail with
Error in open session response message : no matching cipher suite
Error: Unable to establish IPMI v2 / RMCP+ session
Another possible problem is ipmitool
commands taking very long (tens of
seconds or even minutes) because the BMC does not support cipher suite
negotiation. In both cases you can specify the required suite yourself, e.g.
baremetal node set <UUID or name> --driver-info ipmi_cipher_suite=3
In scenarios where the operator can’t specify the ipmi_cipher_suite
for
each node, the configuration parameter [ipmi]/cipher_suite_versions
can be
set to a list of cipher suites that will be used, Ironic will attempt to find
a value that can be used from the list provided (from last to first):
[ipmi]
cipher_suite_versions = 1,2,3,6,7,8,11,12
To find the suitable values for this configuration, you can check the field
RMCP+ Cipher Suites
after running an ipmitool
command, e.g:
$ ipmitool -I lanplus -H $HOST -U $USER -v -R 12 -N 5 lan print
# output
Set in Progress : Set Complete
Auth Type Support : NONE MD2 MD5 PASSWORD OEM
Auth Type Enable : Callback : NONE MD2 MD5 PASSWORD OEM
IP Address Source : Static Address
IP Address : <IP>
Subnet Mask : <Subnet>
MAC Address : <MAC>
RMCP+ Cipher Suites : 0,1,2,3,6,7,8,11,12
Warning
Only the cipher suites 3 and 17 are considered secure by the modern standards. Cipher suite 0 means “no security at all”.
Using a different privilege level¶
By default Ironic requests the ADMINISTRATOR
privilege level of all
commands. This is the easiest option, but if it’s not available for you, you
can change it to CALLBACK
, OPERATOR
or USER
this way:
baremetal node set <UUID or name> --driver-info ipmi_priv_level=OPERATOR
You must ensure that the user can still change power state and boot devices.
Static boot order configuration¶
Vendor Differences¶
While the Intelligent Platform Management Interface (IPMI) interface is based upon a defined standard, the Ironic community is aware of at least one vendor which utilizes a non-standard boot device selector. In essence, this could be something as simple as different interpretation of the standard.
As of October 2020, the known difference is with Supermicro hardware where
a selector of 0x24
, signifying a REMOTE boot device in the standard,
must be used when a boot operation from the local disk subsystem is requested
in UEFI mode. This is contrary to BIOS mode where the same BMC’s expect
the selector to be a value of 0x08
.
Because the BMC does not respond with any sort of error, nor do we want to
risk BMC connectivity issues by explicitly querying all BMCs what vendor it may
be before every operation, the vendor can automatically be recorded in the
properties
field vendor
. When this is set to a value of
supermicro
, Ironic will navigate the UEFI behavior difference enabling
the UEFI to be requested with boot to disk.
Example:
baremetal node set <UUID or name> \
--properties vendor="supermicro"
Luckily, Ironic will attempt to perform this detection in power synchronization process, and record this value if not already set.
While similar issues may exist when setting the boot mode and target
boot device in other vendors’ BMCs, we are not aware of them at present.
Should you encounter such an issue, please feel free to report this via
Launchpad, and be sure to include
the chassis bootparam get 5
output value along with the mc info
output from your BMC.
Example:
ipmitool -I lanplus -H <BMC ADDRESS> -U <Username> -P <Password> \
mc info
ipmitool -I lanplus -H <BMC ADDRESS> -U <Username> -P <Password> \
chassis bootparam get 5
send_raw clean/deploy step¶
The send_raw
vendor passthru method is available to be invoked as a
clean or deployment step should raw bytes need to be transmitted to the
remote BMC in order to facilitate some sort of action or specific state.
In this case, the raw bytes to be set are conveyed with a raw_bytes
argument on the requested clean or deploy step.
Example:
{
"interface": "vendor",
"step": "send_raw",
"args": {
"raw_bytes": "0x00 0x00 0x00 0x00"
}
}