NVMECONTROL(8) FreeBSD System Manager's Manual NVMECONTROL(8)
NAME
nvmecontrol - NVM Express control utility
SYNOPSIS
nvmecontrol devlist [-h]
nvmecontrol identify [-v] [-x] [-n nsid] <device-id | namespace-id>
nvmecontrol perftest <-n num_threads> <-o read|write> [-p]
<-s size_in_bytes> <-t time_in_sec> <namespace-id>
nvmecontrol reset <device-id>
nvmecontrol logpage <-p page_id> [-x] [-v vendor-string] [-b] [-f LSP]
[-i LSI] [-r] <device-id | namespace-id>
nvmecontrol ns active <device-id>
nvmecontrol ns allocated <device-id>
nvmecontrol ns attach <-n nsid> <-c cntid> <device-id>
nvmecontrol ns attached <-n nsid> <device-id>
nvmecontrol ns controllers <device-id>
nvmecontrol ns create <-s nsze> [-c ncap] [-f lbaf] [-m mset] [-n nmic]
[-p pi] [-l pil] [-L flbas] [-d dps] <device-id>
nvmecontrol ns delete <-n nsid> <device-id>
nvmecontrol ns detach <-n nsid> <-c cntid> <device-id>
nvmecontrol ns identify [-v] [-x] <-n nsid> <device-id>
nvmecontrol nsid <device-id | namespace-id>
nvmecontrol resv acquire <-c crkey> [-p prkey] <-t rtype> <-a racqa>
<namespace-id>
nvmecontrol resv register [-i] [-c crkey] <-k nrkey> <-r rrega>
[-p cptpl] <namespace-id>
nvmecontrol resv release <-c crkey> <-t rtype> <-a rrela> <namespace-id>
nvmecontrol resv report [-e] [-v] [-x] <namespace-id>
nvmecontrol firmware [-s slot] [-f path_to_firmware] [-a] <device-id>
nvmecontrol format [-f fmt] [-m mset] [-p pi] [-l pil] [-E] [-C]
<device-id | namespace-id>
nvmecontrol sanitize <-a sanact> [-c owpass] [-d] [-p ovrpat] [-r] [-I]
[-U] <device-id>
nvmecontrol power [-l] [-p -power_state] [-w -workload_hint]
nvmecontrol selftest <-c code> <device-id | namespace-id>
nvmecontrol wdc cap-diag [-o -path_template] <device-id>
nvmecontrol wdc drive-log [-o -path_template] <device-id>
nvmecontrol wdc get-crash-dump [-o -path_template] <device-id>
nvmecontrol admin-passthru [args] <device-id>
nvmecontrol io-passthru [args] <namespace-id>
nvmecontrol discover [-v] [-t transport] [-q HostNQN]
nvmecontrol connect [-FGg] [-c cntl-id] [-i queues] [-k seconds]
[-l seconds] [-r seconds] [-t transport] [-q HostNQN]
[-Q entries] <address> <SubNQN>
nvmecontrol connect-all [-FGg] [-i queues] [-k seconds] [-l seconds]
[-r seconds] [-t transport] [-q HostNQN] [-Q entries]
<address>
nvmecontrol disconnect <device-id | namespace-id | SubNQN>
nvmecontrol reconnect <device-id>
nvmecontrol reconnect [-FGg] [-i queues] [-k seconds] [-l seconds]
[-r seconds] [-t transport] [-q HostNQN] [-Q entries]
<device-id> <address>
nvmecontrol telemetry-log -O output-file [-d data-area] <device-id>
DESCRIPTION
NVM Express (NVMe) is a storage protocol standard for SSDs and other
high-speed storage devices over PCI Express as well as remote storage
devices accessed via a network fabric.
devlist
List all NVMe controllers and namespaces along with their device nodes.
With the -h argument, use unit suffixes: Byte, Kibibyte, Mebibyte,
Gibibyte, Tebibyte and Pebibyte (based on powers of 1024) when showing
the disk space. By default, uses Mebibyte.
identify
The identify commands reports information from the drive's
IDENTIFY_CONTROLLER if a device-id is specified. It reports
IDENTIFY_NAMESPACE data if a namespace-id is specified. When used with
disk names, the IDENTIFY_NAMESPACE data is reported, unless the namespace
nsid is overridden with the -n flag. Then that namespace's data is
reported, if it exists. The command accepts the following parameters:
-n The namespace <nsid> to use instead of the namespace associated
with the device. A nsid of "0" is used to retrieve the
IDENTIFY_CONTROLLER data associated with that drive.
logpage
The logpage command knows how to print log pages of various types. It
also knows about vendor specific log pages from HGST/WDC, Samsung, Micron
and Intel. Note that some vendors use the same log page numbers for
different data.
Page 0x01 Drive Error Log
Page 0x02 Health/SMART Data
Page 0x03 Firmware Information
Page 0x04 Changed Namespace List
Page 0x05 Commands Supported and Effects
Page 0x06 Device Self-test
Page 0x80 Reservation Notification
Page 0x81 Sanitize Status
Page 0xc1 Advanced SMART information (WDC/HGST)
Page 0xc1 Read latency stats (Intel)
Page 0xc2 Write latency stats (Intel)
Page 0xc5 Temperature stats (Intel)
Page 0xca Advanced SMART information (Intel)
Page 0xca Extended SMART information (Samsung)
Page 0xca Vendor Unique SMART information (Micron)
Specifying -v help will list all valid vendors and pages. -x will print
the page as hex. -b will print the binary data for the page. -s will
set Log Specific Field. -i will set Log Specific Identifier. -r will
set Retain Asynchronous Event.
ns
Various namespace management commands. If namespace management is
supported by device, allow list, create and delete namespaces, list,
attach and detach controllers to namespaces. Each NVM device consists of
one or more NVM subsystems. Each NVM subsystem has one or more NVM
ports. Each NVM port is attached to one or more NVM controllers (though
typically 1). Each NVM controller is attached to one or more namespaces.
After a namespace is created, it is considered "allocated". All
namespaces that have not been created are unallocated. An allocated
namespace may be active or inactive. An active namespace is attached to
the controller and may be interacted with. A namespace can move from
active to inactive when detached. An allocated namespace may be deleted
to become unallocated. For more details on the nuances of NVM
namespaces, please see section 2 Theory of Operation and section 3 NVM
Express Architecture of the latest NVM standard.
ns active
Provide a list of active namespace identifiers for the givne NVM
controller.
ns allocated
Provide a list of allocated namespace identifiers for the givne NVM
controller.
ns attach
Attach an nsid to a controller. The primary controller is used if one is
not specified.
ns attached
Provide a list of controllers attached to a nsid. If only a nvme
controller argument is provided, a nsid must also be specified.
ns controllers
Provide a list of all controllers in the NVM subsystem.
ns create
Creates a new namespace.
ns delete
Delete a namespace. It must be currently inactive.
ns detach
Detach a namespace from a controller. The namespace will become
inaccessible, but its contents will remain if it is activated again.
ns identify
Print detailed information about the namespace.
nsid
Reports the namespace id and controller device associated with the
<namespace-id> or <device-id> argument.
resv acquire
Acquire or preempt namespace reservation, using specified parameters:
-a Acquire action:
0 Acquire
1 Preempt
2 Preempt and abort
-c Current reservation key.
-p Preempt reservation key.
-t Reservation type:
1 Write Exclusive
2 Exclusive Access
3 Write Exclusive - Registrants Only
4 Exclusive Access - Registrants Only
5 Write Exclusive - All Registrants
6 Exclusive Access - All Registrants
resv register
Register, unregister or replace reservation key, using specified
parameters:
-c Current reservation key.
-k New reservation key.
-r Register action:
0 Register
1 Unregister
2 Replace
-i Ignore Existing Key
-p Change Persist Through Power Loss State:
0 No change to PTPL state
2 Set PTPL state to `0'. Reservations are released and
registrants are cleared on a power on.
3 Set PTPL state to `1'. Reservations and registrants
persist across a power loss.
resv release
Release or clear reservation, using specified parameters:
-c Current reservation key.
-t Reservation type.
-a Release action:
0 Release
1 Clean
resv report
Print reservation status, using specified parameters:
-x Print reservation status in hex.
-e Use Extended Data Structure.
format
Format either specified namespace, or all namespaces of specified
controller, using specified parameters:
-f fmt The index fmt of the parameters to use. LBA Format #, as
specified in the identification of the namespace using
"nvmecontrol identify" command with a namespace specified maps
this index into these parameters.
-m mset Metadata Setting. mset
0 do not transfer metadata with LBA information
1 Transfer the metadata as part of the extended LBA
information.
-p pi Protection Information.
0 Protection Information not enabled.
1 Type 1 information protection enabled.
2 Type 2 information protection enabled.
3 Type 3 information protection enabled.
-l pil Protection Information Location.
0 Transfer the protection metadata as the last N bytes of
the transfer.
1 Transfer the protection metadata as the first N bytes
of the transfer.
-E Enables User Data Erase during format. All users data is
erased and subsequent reads are indeterminate. The drive may
implement this as a cryptographic erase or it may physically
erase the underlying media.
-C Enables Cryptographic Erase during format. All user data is
erased cryptographically by deleting the encryption key,
rendering it unintelligible.
When formatting specific namespace, existing values are used as defaults.
When formatting all namespaces, all parameters should be specified. Some
controllers may not support formatting or erasing specific or all
namespaces. The nvme(4) driver does not currently support metadata and
protection information transfers.
sanitize
Sanitize NVM subsystem of specified controller, using specified
parameters:
-a operation
Specify the sanitize operation to perform.
overwrite Perform an overwrite operation by writing a
user supplied data pattern to the device one or
more times. The pattern is given by the -p
argument. The number of times is given by the
-c argument.
block Perform a block erase operation. All the
device's blocks are set to a vendor defined
value, typically zero.
crypto Perform a cryptographic erase operation. The
encryption keys are changed to prevent the
decryption of the data.
exitfailure Exits a previously failed sanitize operation.
A failed sanitize operation can only be exited
if it was run in the unrestricted completion
mode, as provided by the -U argument.
1, 2, 3, 4 nvme-cli compatible -a values for
"exitfailure", "block", "overwrite", and
"crypto" respectively.
-c passes
The number of passes when performing an `overwrite' operation.
Valid values are between 1 and 16. The default is 1.
-d No Deallocate After Sanitize.
-I When performing an `overwrite' operation, the pattern is inverted
between consecutive passes.
-p pattern
32 bits of pattern to use when performing an `overwrite'
operation. The pattern is repeated as needed to fill each block.
-U Perform the sanitize in the unrestricted completion mode. If the
operation fails, it can later be exited with the `exitfailure'
operation.
-r Run in "report only" mode. This will report status on a sanitize
that is already running on the drive.
power
Manage the power modes of the NVMe controller.
-l List all supported power modes.
-p mode
Set the power mode to mode. This must be a mode listed with the
nvmecontrol power -l
command.
-w hint
Set the workload hint for automatic power mode control.
0 No workload hint is provided.
1 Extended idle period workload. The device is often idle
for minutes at a time. A burst of write commands comes
in over a period of seconds. Then the device returns to
being idle.
2 Heavy sequential writes. A huge number of sequential
writes will be submitted, filling the submission queues.
Other All other values are reserved and have no standard
meaning.
Please see the "NVM Subsystem Workloads" section of the relevant
NVM Express Base Standard for details.
selftest
Start the specified device self-test:
-c code
Specify the device self-test command code. Common codes are:
0x1 Start a short device self-test operation
0x2 Start an extended device self-test operation
0xe Start a vendor specific device self-test operation
0xf Abort the device self-test operation
wdc
The various wdc command retrieve log data from the wdc/hgst drives. The
-o flag specifies a path template to use to output the files. Each file
takes the path template (which defaults to nothing), appends the drive's
serial number and the type of dump it is followed by .bin. These logs
must be sent to the vendor for analysis. This tool only provides a way
to extract them.
passthru
The "admin-passthru" and "io-passthru" commands send NVMe commands to
either the administrative or the data part of the device. These commands
are expected to be compatible with nvme-cli. Please see the NVM Express
Base Standard for details.
-o --opcode opcode
Opcode to send.
-2 --cdw2 value 32-bit value for CDW2.
-3 --cdw3 value 32-bit value for CDW3.
-4 --cdw10 value 32-bit value for CDW10.
-5 --cdw11 value 32-bit value for CDW11.
-6 --cdw12 value 32-bit value for CDW12.
-7 --cdw13 value 32-bit value for CDW13.
-8 --cdw14 value 32-bit value for CDW14.
-9 --cdw15 value 32-bit value for CDW15.
-l --data-len Length of the data for I/O (bytes).
-m --metadata-len
Length of the metadata segment for command (bytes).
This is ignored and not implemented in nvme(4).
-f --flags Nvme command flags.
-n --namespace-id
Namespace ID for command (Ignored).
-p --prefill Value to prefill payload with.
-b --raw-binary Output in binary format (otherwise a hex dump is
produced).
-d --dry-run Do not actually execute the command, but perform sanity
checks on it.
-r --read Command reads data from the device.
-s --show-command
Show all the command values on stdout.
-w --write Command writes data to the device.
Send arbitrary commands to the device. Can be used to extract vendor
specific logs. Transfers to/from the device possible, but limited to
MAXPHYS bytes. Commands either read data or write it, but not both.
Commands needing metadata are not supported by the nvme(4) drive.
discover
List the remote controllers advertised by a remote Discovery Controller:
-t transport
Transport to use. The default is tcp.
-q HostNQN
NVMe Qualified Name to use for this host. By default an NQN is
auto-generated from the current host's UUID.
-v Display the IDENTIFY_CONTROLLER data for the Discovery
Controller.
connect
Establish an association with the I/O controller named SubNQN at address.
The address must include a port.
An admin queue pair and one or more I/O queue pairs are created and
handed off to the kernel to create a new controller device.
-c cntl-id
Remote controller ID to request:
dynamic
Request a dynamic controller ID for controllers using the
dynamic controller model. This is the default.
static Request a dynamic controller ID for controllers using the
static controller model.
number Request a specific controller ID for controllers using
the static controller model.
-F Request submission queue flow control. By default submission
queue flow control is disabled unless the remote controller
requires it.
-g Enable TCP PDU header digests.
-G Enable TCP PDU data digests.
-i queues
Number of I/O queue pairs to create. The default is 1.
-k seconds
Keep Alive timer duration in seconds. The default is 120.
-l seconds
Controller Loss timer duration in seconds. The default is 600.
This timer starts when an association is lost with a remote I/O
controller and is cancelled when a new association is
established. If the timer expires, the controller device is
deleted. A setting of zero disables this timer.
-r seconds
Reconnect timer duration in seconds. The default is 10.
When an association is lost with a remote I/O controller, the
controller device will request reconnection via periodic
devctl(4) notifications until either a new association is
established or the controller device is deleted. This timer sets
the interval between each devctl(4) notification. Note that the
first notification is triggered immediately after an association
is lost. A setting of zero disables this timer.
-t transport
Transport to use. The default is tcp.
-q HostNQN
NVMe Qualified Name to use for this host. By default an NQN is
auto-generated from the current host's UUID.
-Q entries
Number of entries in each I/O queue. The default is the minimum
of 128 or the maximum queue size reported by the MQES field of
the remote host's CAP property.
connect-all
Query the Discovery Controller at address and establish an association
for each advertised I/O controller. The -t flag determines the transport
used for the initial association with the Discovery Controller and
defaults to tcp. All other flags are used to control properties of each
I/O assocation as described above for the connect command.
disconnect
Delete the controller device associated with a remote I/O controller
including any active association and open queues.
reconnect
Reestablish an association for the remote I/O controller associated with
device-id. If an address is not provided, the resolved address and
settings from the previous association are used to establish a new
association. If an address is provided, the supplied address and command
line flags are used to establish a new association. In this case, the
address must include a port and the flags have the same meaning for the
new association as described above for the connect command.
telemetry-log
Extract the telemetry log associated with device-id, using the specified
parameters:
-O output-file
Output file for the data. This parameter is mandatory.
-d data-area
The data area is either 1, 2 or 3.
DEVICE NAMES
Where <namespace-id> is required, you can use either the nvmeXnsY device,
or the disk device such as ndaZ or nvdZ. The leading /dev/ may be
omitted. Where <device-id> is required, you can use either the nvmeX
device, or the disk device such as ndaZ or nvdZ. For commands that take
an optional <nsid> you can use it to get information on other namespaces,
or to query the drive itself. A <nsid> of "0" means query the drive
itself.
FABRICS TRANSPORTS
The following NVM Express over Fabrics transports are supported for
accessing remote controllers:
tcp TCP transport
NETWORK ADDRESSES
Network addresses for remote controllers can use one of the following
formats:
o [IPv6 address]:port
o IPv4 address:port
o hostname:port
o [IPv6 address]
o IPv6 address
o IPv4 address
o hostname
If a port is not provided, a default value is used if possible.
EXAMPLES
nvmecontrol devlist
Display a list of NVMe controllers and namespaces along with their device
nodes.
nvmecontrol identify nvme0
nvmecontrol identify -n 0 nvd0
Display a human-readable summary of the nvme0 IDENTIFY_CONTROLLER data.
In this example, nvd0 is connected to nvme0.
nvmecontrol identify -x -v nvme0ns1
nvmecontrol identify -x -v -n 1 nvme0
Display an hexadecimal dump of the nvme0 IDENTIFY_NAMESPACE data for
namespace 1.
nvmecontrol perftest -n 32 -o read -s 512 -t 30 nvme0ns1
Run a performance test on nvme0ns1 using 32 kernel threads for 30
seconds. Each thread will issue a single 512 byte read command. Results
are printed to stdout when 30 seconds expires.
nvmecontrol reset nvme0
nvmecontrol reset nda4
Perform a controller-level reset of the nvme0 controller. In this
example, nda4 is wired to nvme0.
nvmecontrol logpage -p 1 nvme0
Display a human-readable summary of the nvme0 controller's Error
Information Log. Log pages defined by the NVMe specification include
Error Information Log (ID=1), SMART/Health Information Log (ID=2), and
Firmware Slot Log (ID=3).
nvmecontrol logpage -p 0xc1 -v wdc nvme0
Display a human-readable summary of the nvme0's wdc-specific advanced
SMART data.
nvmecontrol logpage -p 1 -x nvme0
Display a hexadecimal dump of the nvme0 controller's Error Information
Log.
nvmecontrol logpage -p 0xcb -b nvme0 > /tmp/page-cb.bin
Print the contents of vendor specific page 0xcb as binary data on
standard out. Redirect it to a temporary file.
nvmecontrol firmware -s 2 -f /tmp/nvme_firmware nvme0
Download the firmware image contained in "/tmp/nvme_firmware" to slot 2
of the nvme0 controller, but do not activate the image.
nvmecontrol firmware -s 4 -a nvme0
Activate the firmware in slot 4 of the nvme0 controller on the next
reset.
nvmecontrol firmware -s 7 -f /tmp/nvme_firmware -a nvme0
Download the firmware image contained in "/tmp/nvme_firmware" to slot 7
of the nvme0 controller and activate it on the next reset.
nvmecontrol power -l nvme0
List all the current power modes.
nvmecontrol power -p 3 nvme0
Set the current power mode.
nvmecontrol power nvme0
Get the current power mode.
nvmecontrol identify -n 0 nda0
Identify the drive data associated with the nda0 device. The
corresponding nvmeX devices is used automatically.
nvmecontrol identify nda0
Get the namespace parameters associated with the nda0 device. The
corresponding nvmeXnsY device is used automatically.
nvmecontrol format -f 2 -m 0 -p 0 -l 0 -C nvme2
Format all the name spaces on nvme2 using parameters from "LBA Format #2"
with no metadata or protection data using cryptographic erase. If the
"nvmecontrol identify -n 1 nvme2" command ended with
LBA Format #00: Data Size: 512 Metadata Size: 0 Performance: Good
LBA Format #01: Data Size: 512 Metadata Size: 8 Performance: Good
LBA Format #02: Data Size: 4096 Metadata Size: 0 Performance: Good
LBA Format #03: Data Size: 4096 Metadata Size: 8 Performance: Good
LBA Format #04: Data Size: 4096 Metadata Size: 64 Performance: Good
then this would give a 4k data format for at least namespace 1, with no
metadata.
DYNAMIC LOADING
The directories /lib/nvmecontrol and /usr/local/lib/nvmecontrol are
scanned for any .so files. These files are loaded. The members of the
top linker set are added to the top-level commands. The members of the
logpage linker set are added to the logpage parsers.
SEE ALSO
The NVM Express Base Specification,
https://nvmexpress.org/wp-content/uploads/NVM-Express-1_4-2019.06.10-Ratified.pdf,
June 10, 2019.
HISTORY
The nvmecontrol utility appeared in FreeBSD 9.2.
AUTHORS
nvmecontrol was developed by Intel and originally written by Jim Harris
<jimharris@FreeBSD.org>.
This man page was written by Jim Harris <jimharris@FreeBSD.org>.
FreeBSD 15.1-STABLE-HBSD February 20, 2026 NVMECONTROL(8)