lsscsi - Man Page

• List all SCSI devices: lsscsi
• List all SCSI devices with detailed attributes: lsscsi [-L|--list]
• List all SCSI devices with human-readable disk capacity: lsscsi [-s|--size]

tldr.sh

lsscsi [--brief] [--classic] [--controllers] [--device] [--generic] [--help] [--hosts] [--json[=JO]] [--js-file=JFN] [--kname] [--list] [--long] [--long-unit] [--lunhex] [--no-nvme] [--pdt] [--protection] [--protmode] [--scsi_id] [--size] [--sysfsroot=PATH] [--sysroot=AR_PT] [--sz-lbs] [--transport] [--unit] [--verbose] [--version] [--wwn] [H:C:T:L]

Uses information mainly found in sysfs to list SCSI devices (or hosts) that are currently attached to the system. In order to get a brief overview the default is to output one line per device. Many non-SCSI storage devices (but not all) use the SCSI subsystem in Linux. Support for listing NVMe devices and controllers was added in lsscsi version/release 0.30 .

Only parts of sysfs that don't require root permissions, are accessed. Thus normal users, without root permissions, can use this utility. The devfs pseudo file system (normally found under '/dev' ) is also accessed.

The term "device" is wonderfully vague with one possible definition being the "thing" at the other end of the wire (from the CPU or host computer). In SCSI the "thing" at the other end of the wire is called the target which contains one or more "logical units" (LUs). A LU is a sequence of equally sized blocks addressed by a "logical block address" (LBA) starting at 0. In NVMe the "thing" at the other end of the wire is a "NVMe device" which contains one or more namespaces.

By default, this utility is in single_line_per_LU/namespace mode, each line starts with a 4 element tuple surrounded by square brackets. For SCSI devices the first element ('H') is the host number, the second element ('C') is the controller number, the third element ('T') is the target number and the final element is the Logical Unit Number ('L' or LUN). All four are integers. For NVMe namespaces see two paragraphs down. When the --hosts option is given for SCSI devices the tuple is reduced to one element: the host number.

If a H:C:T:L tuple is given as an argument on the command line then it acts as a filter and only devices that match it are listed. The colons don't have to be present, and '-', '*', '?' or missing components at the end are interpreted as wildcards. The default is '*:*:*:*' which means to match all devices (i.e.  LU or namespaces) currently attached. Any filter string using '*' of '?' should be surrounded by single quotes to stop shell expansions.  If '-' is used as a wildcard then the whole filter tuple should be prefixed by '-- ' to tell this utility there are no more options (following the '--') on the command line to be interpreted. The use of '-- ' like that is standard Unix command line practice. A leading '[' and trailing ']' are permitted ( e.g. '[1:0:0]' matches all LUNs on 1:0:0). May also be used to filter --hosts in which case only the H is active and may be either a number or in the form "host<n>" where <n> is a host number.

For NVMe devices and controllers almost all of the previous paragraph applies. The main difference is that the letter "N" appears in the 'H' (first) position. The 'C' position for NVMe is the controller's Linux generated "char" device's minor number. That number is the first number that appears in a typical NVMe controller name, for example: "/dev/nvme2". The 'T' position for NVMe is the "CNTLID" value. The final 'L' position is the NVMe namespace identifier which is typically a sequential value starting at 1. The leading explicit "N" for NVMe devices is converted internally into a large value (32,767) that should not interfere with any Linux generated SCSI host number; it also means that the numeric sort used to show hosts (controllers) and devices (LUs or namespaces) will always place NVMe devices and controllers after those that use the SCSI subsystem. To filter using a H:C:T:L argument for NVMe controllers, "hostN", "hostN:<num>", "N" or "N:<num>" may be used; when no "<num>" is given, only NVMe controllers will be listed (i.e. no SCSI hosts (HBAs) are listed).

By default in this utility device node names (e.g. "/dev/sda" or "/dev/root_disk") are obtained by noting the major and minor numbers for the listed device obtained from sysfs (e.g. the contents of "/sys/block/sda/dev") and then looking for a match in the "/dev" directory. This "match by major and minor" will allow devices that have been given a different name by udev (for example) to be correctly reported by this utility.

In some situations it may be useful to see the device node name that Linux would produce by default, so the --kname option is provided. An example of where this may be useful is kernel error logs which tend to report disk error messages using the disk's default kernel name.

Arguments to long options are mandatory for short options as well. The options are arranged in alphabetical order based on the long option name. Hyphenated long options can also take underscore, and vice versa (e.g. so --scsi_id or --scsi-id are both acceptable).

-b,  --brief

reduces one line per device output to the tuple and the primary device name. This may simplify scripts that process the output of this utility. With the --generic option it will show on each line the tuple (from which the bsg pass-through device name can be deduced), the primary device name (which the block subsystem uses) and the sg device name (i.e. a name of the form /dev/sg<n> where <n> starts at 0).
When the --pdt option is used together with this option the SCSI Peripheral Device Type (PDT) is displayed in hex (with a leading "0x") between the tuple and the primary device name. For NVMe namespaces "0x0" is displayed (for a disk or direct access device).

-c,  --classic

The output is similar to that obtained from 'cat /proc/scsi/scsi' . There is no JSON rendering of this output, the output is always in plain text.

-C,  --controllers

Lists NVMe controllers and SCSI hosts. This is a synonym for the --hosts option.

-d,  --device

After outputting the (probable) SCSI device name the device node major and minor numbers are shown in brackets (e.g. "/dev/sda[8:0]").

-g,  --generic

Output the SCSI generic device file name. Note that if the sg driver is a module it may need to be loaded otherwise '-' may appear.
NVMe now has generic (char) devices that have the form /dev/ngCnN where the "C" is the controller id (cntlid) and the "N" is the namespace identifier (nsid). Note that NVMe administration commands (e.g. Identify) and MI (Management Interface (e.g. to an enclosure)) commands can use the /dev/nvmeC (where "C" is the cntlid) char devices since that don't need a namespace identifier. Conversely the NVMe Read and Write commands do need an (implicit) namespace identifier, so, for example, the /dev/nvme1 char device cannot process NVMe Read commands.
To unclutter the single line per device mode the --brief option combined with this option should help.

-h,  --help

Output the usage message then exit. Note the output is sent to stderr.

-H,  --hosts

List the SCSI hosts and NVMe controllers currently attached to the system. If this option is not given (and the --controllers option is not given) then SCSI devices (logical units (LUs)) followed by NVMe devices (namespaces) are listed.

-j[=JO], --json[=JO]

output is in JSON format instead of plain text form. Note that arguments to the short and long form are themselves optional and if present start with "=" and no whitespace is permitted either side of that "=".
See the lsscsi_json(8) manpage or use '?' for JO to get a summary (e.g. '-j=?').

-J,  --js-file=JFN

Send the JSON output to a file named JFN instead of stdout. When this option is given, the --json option is not required unless JSON settings in JO are needed.
See the accompanying lsscsi_json(8) manpage.

-k,  --kname

Use Linux default algorithm for naming devices (e.g. block major 8, minor 0 is "/dev/sda") rather than the "match by major and minor" in the "/dev" directory as discussed above.
Said another way: Linux kernel drivers will create device nodes (Unix special files under the "/dev" directory) of 'char' or 'block' type at startup. This allows user programs to access this storage. One such program called udev has the ability to change those special file names which is what the user will see henceforth apart from sysfs which still uses the original name (i.e. the name issued by the kernel when the host/device is detected during bootup).
This option should be seldom needed.

-L,  --list

Output additional information in <attribute_name>=<value> pairs, one pair per line preceded by two spaces. This option has the same effect as '-lll'.
Getting JSON output is a more structured way of obtaining parse-able data and is encouraged rather than using this option.

-l,  --long

Output additional information for each SCSI device (host). Can be used multiple times for more output in which case the shorter option form is more convenient (e.g. '-lll'). When used three times (i.e. '-lll') outputs SCSI device (host) attributes one per line; preceded by two spaces; in the form "<attribute_name>=<value>".

-U,  --long-unit

Output logical unit name in full, if available. It replaces the normal vendor, product and revision strings given in the single logical unit per line mode. If no logical unit name is found "none" is printed. If the logical unit name is long (e.g. a UUID) then following fields are pushed further to the right as required. This option is functionally equivalent to the '-uuu' option.
If the option is used twice (e.g. '-UU') then EUI, NAA, UUID and T10 vendor ID formats are prefixed by "eui.", "naa.", "uuid." and "t10." respectively. Note that SCSI name format used by iSCSI should already be prefixed by 'iqn.'. Using the '--unit' option 4 or more times (e.g. '-uuuu') will have the same action as '-UU'.
--long_unit is also an acceptable form when invoking this option.

-x,  --lunhex

when this option is used once the LUN in the tuple (at the start of each device line) is shown in "T10" format which is up to 16 hexadecimal digits. It is prefixed by "0x" to distinguish the LUN from the decimal value shown in the absence of this option. Also hierarchal LUNs are shown with a "_" character separating the levels. For example the two level LUN: 0x0355006600000000 will appear as 0x0355_0066. If this option is given twice (e.g. using the short form: '-xx') then the full 16 hexadecimal digits are shown for each LUN, prefixed by "0x".
For NVMe, the namespace identifier (nsid) is shown in the "L" position. The nsid is a 32 bit unsigned quantities with 0x0 and 0xffffffff reserved. Without this option, the nsid is shown in decimal. When this option is used once the nsid is output in hex with a lead 0x and with up to 3 leading zeros. When this option is used twice the nsid is output in hex with up to 7 leading zeros.

-N,  --no-nvme

this option excludes NVMe devices and controllers for the output. This option may be needed to stop NVMe device output interfering with specific format output like that produced when the --classic option is used.
To only show NVMe devices, use 'lsscsi N'. To only show NVMe controllers, use 'lsscsi -H N'.

-D,  --pdt

this option displays the SCSI Peripheral Device Type (PDT) in hex preceded by "0x". For NVME namespaces "0x0' is displayed which corresponds to a disk ("Direct Access Device" or SSD). In single line output this hex PDT replaces the device type abbreviation (e.g. "0x0     " replaces "disk    ") and appears after the tuple.

-p,  --protection

Output target (DIF) and initiator (DIX) protection types in that order to the right in one_line_per_device mode. If there is no DIF then '-' is output, if there is no DIX then 'none' is output.

-P,  --protmode

Outputs the effective protection information mode for each disk device. This is typically at the end (i.e. right hand side) of the one_line_per_device mode; '-' is output is there is no protection information.

-i,  --scsi_id

outputs the udev derived matching id found in /dev/disk/by-id/scsi* . This is only for disk (and disk like) devices. If no match is found then "dm-uuid-mpath*" and "usb*" are searched in the same directory. If there is still no match then the /sys/class/block/<disk>/holders directory is searched. The matching id is printed following the device name (e.g.  /dev/sdc) and if there is no match "-" is output. Note that only disk (like) devices are matched by this option; so, for example, a SCSI enclosure will have an identifier of '-'.
Prior to revision 164 the first match in /dev/disk/by-id/scsi* was printed. A change was added at that point to check for identifiers in a specific order as some are considered "stronger" than others. See the Order of Scsi Identifiers section below.
Note that the identifier output is prefixed by one character from this string: "328S10" to indicate what type of identifier is being shown. Those values are explained in the Order of Scsi Identifiers section below. To show the identifier without that leading value, use --scsi_id twice (or simply -ii).

-s,  --size

Print disk capacity in plain text form. When given once, normal base 10 SI units are used as a prefix for 'B' which is bytes (aka octets). For example MB, GB and TB stand for 10^6, 10^9 and 10^12 bytes respectively. When given twice, IEC 80000-3 prefixes for 'B' are used; for example MiB, GiB and TiB stand for 2^20, 2^30 and 2^40 bytes respectively. The output is rounded to 3 or less significant figures in order to fit on a single line. It will also output the size of RBC devices, CD/DVD media and host managed ZBC disks. Host aware ZBC disks have their "peripheral device type" set to 0 (the same as normal disks) so their size is output.
If given three times (short form is the more convenient: '-sss') then the disk capacity as a logical block count is given. This is an exact figure in decimal reported by the storage device at discovery. Discovery is typically just after boot time, or when it was last attached if the storage device is removable.
To unclutter the single line per device mode the --brief option combined with this option should help.

-y,  --sysfsroot=PATH

assumes sysfs is mounted at PATH instead of the default '/sys' . If this option is given PATH should be an absolute path (i.e. start with '/').
May be useful when another machine has its whole root file system network-mounted on this machine. Then the other machine can have its SCSI devices enumerated with "--sysfsroot=/mnt/other_machine/sys" .
The clone_pseudo_fs utility may be used to take a snapshot of a machine's /sys directory at some time of interest (or /sys from another machine) then have its SCSI devices (and other attributes) replayed with an option like this: "--sysfsroot=/tmp/sas" .
Since this utility also depends on devfs, this option may not be sufficient as it only redirects sysfs access. A solution to this issue is to use the --sysroot=AR_PT option instead.
This option is closely related to the '--sysroot=' option found in lsblk and lscpu (amongst others). Those utilities assume the "sys" part of the path given as their argument (and lsblk expects to find "proc" there as well). procfs may also be cloned with clone_pseudo_fs but lsblk relies on the symlink proc/self which points to the PID of clone_pseudo_fs instance when it ran, and that is not valid when in "replay" mode. lsblk is still helpful in replay mode, even without a "valid" procfs, as only the "mountpoint" column is lost.

-Y,  --sysroot=AR_PT

AR_PT is an alternate root path. The default root path is '/' and this utility will therefore access '/sys' and '/dev' for information used to build its output. If this option is given, say with AR_PT being '/tmp' then this utility will access '/tmp/sys' and '/tmp/dev' for its data. So this option is more powerful than the --sysfsroot=PATH option. Both this option and --sysfsroot=PATH option should not be invoked together, but if they were then AR_PT must be the parent directory of PATH.
This option was added for compatibility with other 'ls*' utilities that have a --sysroot option. Having such an option allows a 'ls*' utility to decode data from another machine, or a snapshot of the current machine at an early time. This could be useful if one of those 'ls*' utilities was misbehaving (e.g. failing to list hardware that was present).

-S,  --sz-lbs

Print disk capacity as a number of logical blocks (which is the same as '-sss'). When used twice a comma is added followed by the logical block size in bytes. It should be a number like 512 or 4096.
If the logical block size cannot be found (e.g. because the version of Linux predates the /sys/block/<dev_name>/queue directory) then the number of 512 byte blocks followed comma and then '512' is output irrespective of what the true logical block size of the device is. This special case action occurs whether this option is given one or more times.
To unclutter the single line per device mode the --brief option combined with this option should help.

-t,  --transport

Output transport information. This will be target related information or, if --hosts is given, initiator related information. When used without --list, a name or identifier (or both) are output on a single line, usually prefixed by the type of transport. For devices this information replaces the normal vendor, product and revision strings. When the --list option is also given then additionally multiple lines of attribute_name=value pairs are output, each indented by two spaces. See the section on transports below.

-u,  --unit

Output logical unit name, if available. If this option is given once or twice, then the 30 character field where the vendor, product and revision strings are usually placed is expanded to 32 characters and replaced by the logical unit name. If no logical unit name is found "none" is printed. The first found of the NAA, EUI-64 or SCSI name string is output unless a SCSI name string is found and the associated target port indicates the iSCSI protocol, in which case the SCSI name string is preferred. Finally if there is no match on the above and a T10 Vendor ID descriptor is found then it is used.
If the name cannot fit in the 32 character field then it is truncated to the right and a trailing '_' character is used to alert the reader to the truncation. The 32 character width is chosen since that is large enough to hold 16 byte NAA or EUI-64 identifiers. However SCSI name strings as used by iSCSI can be larger than that.
If this option is used twice then this field is also 32 character wide. If the logical unit name cannot fit then it will be truncated to the left and a leading '_' character is used to alert the reader to the truncation.
If this option is used three times the whole logical unit name is output, followed by several spaces.
In order for this option to work, it needs a Linux kernel from and including 3.15 . It accesses the sysfs vpd_pg83 file for the device in question. Old SCSI and ATA (SATA) equipment may not provide this information. If it is provided by ATA (SATA) then it will be the WWN.

-v,  --verbose

outputs directory names where information is found. Use multiple times for more output.

-V,  --version

outputs version information then exits. If used once outputs to stderr; if used twice outputs to stdout and shortens the date to yyyymmdd numeric format. The first number in the version string is the release number.

-w,  --wwn

additionally outputs the WWN for disks. The World Wide Name (WWN) is typically 64 bits long (16 hex digits) but could be up to 128 bits long. To indicate the WWN is hexadecimal, it is prefixed by "0x". Originally this option looked at the '/dev/disk/by-id/wwn-*' symlinks to scsi devices; but this was changed to '/dev/disk/by-id/scsi-*' as the latter were more stable (i.e. less likely to change from one boot to the next).
If this option is used twice then the original action takes place. In other words the '/dev/disk/by-id/wwn-*' symlinks to scsi devices are used.

This utility lists SCSI devices which are known as logical units (LU) in the SCSI Architecture Model (ref: SAM-5 at https://www.t10.org) or hosts when the --hosts option is given. A host is called an initiator in SAM-5. A SCSI command travels out via an initiator, across some transport to a target and then onwards to a logical unit. A target device may contain several logical units. A target device has one or more ports that can be viewed as transport end points. Each FC and SAS disk is a single target that has two ports and contains one logical unit. If both target ports on a FC or SAS disk are connected and visible to a machine, then lsscsi will show two entries. Initiators (i.e. hosts) also have one or more ports and some HBAs in Linux have a host entry per initiator port while others have a host entry per initiator device.

When the --transport option is given for devices (i.e. --hosts not given) then most of the information produced by lsscsi is associated with the target, or more precisely: the target port, through which SCSI commands pass that access a logical unit.

Typically this utility provides one line of output per "device" or host. Significantly more information can be obtained by adding the --list option. When used together with the --transport option, after the summary line, multiple lines of transport specific information in the form "<attribute_name>=<value>" are output, each indented by two spaces. Using a filter argument will reduce the volume of output if a lot of devices or hosts are present.

The transports that are currently recognized are: IEEE 1394, ATA, FC, iSCSI, SAS, SATA, SPI, SRP and USB.

For IEEE 1394 (a.k.a. Firewire and "SBP" when storage is involved), the EUI-64 based target port name is output when --transport is given, in the absence of the --hosts option. When the --hosts option is given then the EUI-64 initiator port name is output. Output on the summary line specific to the IEEE 1394 transport is prefixed by "sbp:".

To detect ATA and SATA devices a crude check is performed on the driver name (after the checks for other transports are exhausted). Based on the driver name either the ATA or SATA transport type is chosen. Output on the summary line is either "ata:" or "sata:". A search is made for an associated vpd_pg83 file in sysfs, if found it may contain the device's WWN which is output if present. The WWN will not appear in Linux kernels before 3.15 and with old PATA and SATA devices. Most device and hosts flagged as "ata:" will use the parallel ATA transport (PATA). For SATA devices that are attached via a SAS expander, see the SAS paragraph below.

For Fibre Channel (FC) the port name and port identifier are output when --transport is given. In the absence of the --hosts option these ids will be for the target port associated with the device (logical unit) being listed. When the --hosts option is given then the ids are for the initiator port used by the host. Output on the summary line specific to the FC transport is prefixed by "fc:". If FCoE (over Ethernet) is detected the prefix is changed to "fcoe:".

For iSCSI the target port name is output when --transport is given, in the absence of the --hosts option. This is made up of the iSCSI name and the target portal group tag. Since the iSCSI name starts with "iqn" no further prefix is used. When the --hosts option is given then only "iscsi:" is output on the summary line.

For Serial Attached SCSI the SAS address of the target port (or initiator port if --hosts option is also given) is output. This will be a naa-5 address. For SAS HBAs and SAS targets (such as SAS disks and tape drives) the SAS address will be world wide unique. For SATA disks attached to a SAS expander, the expander provides the SAS address by adding a non zero value to its (i.e. the expander's) SAS address (e.g. expander_sas_address + phy_id + 1). SATA disks directly attached to SAS HBAs seem to have an indeterminate SAS address. Output on the summary line specific to the SAS transport is prefixed by "sas:".

For SATA devices, see the paragraph above on ATA devices. As noted in the previous paragraph, SATA devices attached to SAS expanders will display a manufactured SAS transport address (manufactured by the expander) rather than the SATA device's WWN.

For the SCSI Parallel Interface (SPI) the target port identifier (usually a number between 0 and 15 inclusive) is output when --transport is given, in the absence of the --hosts option. When the --hosts option is given then only "spi:" is output on the summary line.

For the PCIe transport (a.k.a. PCI Express) there at two possible storage types: NVMe and SOP/PQI (SCSI over PCIe). There are very few examples of the latter currently so this utility concentrates on NVMe. NVMe uses its own command set and not SCSI but has many things in common. Rather than re-invent everything currently in use that SCSI has accumulated over nearly 40 years, NVMe is beginning to use some parts of SCSI. A recent example is the SES-3 standard for enclosure management which has been adopted by NVMe. In SCSI a SES device is a logical unit with a peripheral device type (PDT) of 0xd (for enclosure) so it will appear when the lsscsi utility is invoked without any options. In NVMe is seems that an enclosure with appear as attached to the management interface (MI) of a NVMe controller. This means it should appear when "lsscsi --hosts" is invoked. It is unclear whether such a NVMe controller can have any storage namespaces associated with it. The sg_ses utility (in the sg3_utils package) can then be given that NVMe controller's device name (e.g. /dev/nmve1).
When the --transport option is given, after "pcie" the NVMe controller's subsystem vendor id and device id are output, separated by a colon (e.g. "pcie 0x8086:0x390a").

For the SCSI RDMA Protocol (SRP) the IB (InfiniBand) port's GUID is given. As an example, it has a form like this: 0002:c903:00fa:abcd .

When a USB transport is detected, the summary line will contain "usb:" followed by a USB device name. The USB device name has the form "<b>-<p1>[.<p2>[.<p3>]]:<c>.<i>" where <b> is the USB bus number, <p1> is the port on the host. <p2> is a port on a host connected hub, if present. If needed <p3> is a USB hub port closer to the USB storage device. <c> refers to the configuration number while <i> is the interface number. There is a separate SCSI host for each USB (SCSI) target. A USB SCSI target may contain multiple logical units. Thus the same "usb: <device_name>" string appears for a USB SCSI host and all logical units that belong to the USB SCSI target associated with that USB SCSI host.

The scsi_debug module/driver in the Linux kernel simulates one or more SCSI hosts attached to one or more SCSI devices. Strictly speaking the scsi_debug driver does not have an associated SCSI transport. However sysfs does associate the driver with the "pseudo_0" device (i.e. /sys/devices/pseudo_0) so "pseudo_0" is used as the SCSI transport name. This helps the lsscsi output to be more consistent when several SCSI transports are present.

For historical reasons and as used by several other Unix based Operating Systems, Linux uses a tuple of integers to describe (a path to) a SCSI device (also know as a Logical Unit (LU)). The last element of that tuple is the so-called Logical Unit Number (LUN). And originally in SCSI a LUN was an integer, at first 3 bits long, then 8 then 16 bits. SCSI LUNs today (SAM-5 section 4.7) are 64 bits but SCSI standards now consider a LUN to be an array of 8 bytes.

Up until 2013, Linux mapped SCSI LUNs to a 32 bit integer by taking the first 4 bytes of the SCSI LUN and ignoring the last 4 bytes. Linux treated the first two bytes of the SCSI LUN as a unit (a word) and it became the least significant 16 bits in the Linux LUN integer. The next two bytes of the SCSI LUN became the upper 16 bits in the Linux LUN integer. The rationale for this was to keep commonly used LUNs small Linux LUN integers. The most common LUN (by far) in SCSI LUN (hex) notation is 00 00 00 00 00 00 00 00 and this becomes the Linux LUN integer 0. The next most common LUN is 00 01 00 00 00 00 00 00 and this becomes the Linux LUN integer 1.

In 2013 it is proposed to increase Linux LUNs to a 64 bit integer by extending the mapping outlined above. In this case all information that is possible to represent in a SCSI LUN is mapped a Linux LUN (64 bit) integer. And the mapping can be reversed without losing information.

This version of the utility supports both 32 and 64 bit Linux LUN integers. By default the LUN shown at the end of the tuple commencing each line is a Linux LUN as a decimal integer. When the --lunhex option is given then the LUN is in SCSI LUN format with the 8 bytes run together, with the output in hexadecimal and prefixed by '0x'. The LUN is decoded according to SAM-5's description and trailing zeros (i.e. digits to the right) are not shown. So LUN 0 (i.e. 00 00 00 00 00 00 00 00) is shown as 0x0000 and LUN 65 (i.e. 00 41 00 00 00 00 00 00) is shown as 0x0041. If the --lunhex option is given twice then the full 64 bits (i.e. 16 hexadecimal digits) are shown.

If the --lunhex option is not given on the command line then the environment variable LSSCSI_LUNHEX_OPT is checked. If LSSCSI_LUNHEX_OPT is present then its associated value becomes the number of times the --lunhex is set internally. So, for example, 'LSSCSI_LUNHEX_OPT=2  lsscsi' and 'lsscsi -xx' are equivalent.

This section only applies to the --scsi_id option and only for disk like devices (e.g. not tapes nor enclosures). There are potentially several SCSI identifiers and from revision 164 they are checked in the following order and only the first found is output.

The SCSI identifier preference order is:

3 : NAA based (Network address Authority)

2 : EUI-64 based

8 : SCSI name string (e.g. iSCSI: iqn.1998-01.com.zzware.iscsi:name1)

S : serial number from SCSI VPD page 0x80

1 : T10 Vendor Identifier

0 : Vendor Specific
Those numbers prefixing each entry are the SCSI 'Designator Types' found in the definition of the Device Identification VPD page (0x83) in SPC-3, SPC-4 and SPC-5. The 'S' of course if not a number and it refers to a different VPD page: the Serial Number VPD page (0x80).

There is a more general --wwn option that should apply to almost all devices. The term "WWN" (world-wide name) comes from the ATA and NVMe standards and corresponds to the "Logical Unit (LU) name" in SCSI. The LU name in SCSI tends to change by transport. For SAS the LU name is the LU's NAA identifier.

Plus there the --unit and the --long-unit options that may be helpful in uniquely identifying storage devices.

Most of the numerical identifiers (and iSCSI url type strings) come from the SCSI INQUIRY command's Device Identification VPD (vital product data) page. It is one of many VPD pages. These VPD pages are essentially constant (they can be changed in extraordinary situations) and are read by the SCSI subsystem when a SCSI device is attached to the system. Modern versions of Linux make copies of those vpd pages available in the sysfs pseudo file system. They are binary files with names like "vpd_pg83" where 83 is the numerical identifier of the Device Identification VPD page (in hex).

The sg_inq and sg_vpd utilities in the sg3_utils package can decode those pages. Currently VPD pages 00 (list of supported VPD pages), 80 (serial number), 83 (device identification (DI)) and 89 (ATA Information) are available.  More pages may appear in the future. Root permissions are not required to access these pages. Using sg_vpd with dev/sg3 as an example, the following invocation will decode sg3's DI page:

 sg_vpd --raw --inhex=/sys/class/scsi_generic/sg3/device/vpd_pg83

which can be shortened to:

 sg_vpd -rI /sys/class/scsi_generic/sg3/device/vpd_pg83

Note that the DI VPD page contains identifiers for both a device (a Logical Unit (LU) in SCSI jargon) and the target device that contains that LU. A target may contain one or more LUs and LUs are often disks. A target is the (far) endpoint of the transport protocol in use, while the initiator is the near end of that transport.

Information about this utility including examples can also be found at: https://sg.danny.cz/scsi/lsscsi.html or https://doug-gilbert.github.io/scsi/lsscsi.html .

An example of a type of storage device not listed by this utility is SD cards (with device names like "/dev/mmc*").

Information for this command is derived from the sysfs file system, which is assumed to be mounted at /sys unless specified otherwise by the user. SCSI (pseudo) devices that have been detected by the SCSI mid level will be listed even if the required upper level drivers (i.e. sd, sr, st, osst or ch) have not been loaded. If the appropriate upper level driver has not been loaded then the device file name will appear as '-' rather than something like '/dev/st0'. Note that some devices (e.g. scanners and medium changers) do not have a primary upper level driver and can only be accessed via a SCSI generic (sg) device name.

Generic SCSI devices can also be accessed via the bsg driver in Linux. By default, the bsg driver's device node names are of the form '/dev/bsg/H:C:T:L'. So, for example, the SCSI device shown by this utility on a line starting with the tuple '6:0:1:2' could be accessed via the bsg driver with the '/dev/bsg/6:0:1:2' device node name.

lsscsi version 0.21 or later is required to correctly display SCSI devices in Linux kernel 2.6.26 (and possibly later) when the CONFIG_SYSFS_DEPRECATED_V2 kernel option is not defined.

In Unix, device nodes (e.g. /dev/sdb) are "special" devices that are either "block" or "char" devices. They also have a "major" and "minor" numbers and a file name (e.g. "sdb"). The following partial listing should clarify how these are shown by the ls command in Linux:

 /dev$ ls -l n*
 crw------- 1 root root 247,   0 Apr 24 16:56 ng0n1
 crw-rw-rw- 1 root root   1,   3 Apr 24 16:56 null
 crw------- 1 root root 248,   0 Apr 29 15:25 nvme0
 brw-rw---- 1 root disk 259,   0 Apr 24 16:56 nvme0n1
 brw-rw---- 1 root disk 259,   1 Apr 24 16:56 nvme0n1p1
 brw-rw---- 1 root disk 259,   2 Apr 24 16:56 nvme0n1p2

The permissions string (e.g. "crw-rw-rw-") starts with a "c" for a char device and "b" for a block device. The two numbers separated by comma (and a few whitespace) are the major and minor numbers respectively. As a general rule the major number identifies the Linux driver which will handle all devices that share the same mode (i.e. char or block) and the same major number. Traditionally Unix major numbers were fixed (e.g. the sg driver is char device 21) but as Linux grew it was obvious that the "fixed major number" scheme would not scale. So now device node major numbers are allocated dynamically (i.e. at bootup) and the mapping between the driver name and its major number can be found in the output of "/proc/devices".

Why explain this? The reason is that all SCSI (and ATA) device nodes in Linux have fixed (i.e. well-known) device major numbers. However NVMe, which is much newer, has dynamically allocated major numbers. So when a utility like lsscsi is dealing with NVMe devices and controllers, often an extra step is required: looking at /proc/devices to determine the mapping between a driver name and its major number.

Also the device node name (and udev can change these to override the kernel's initial settings) may not always suggest the corresponding driver name. For the SCSI generic driver, just remove the trailing number (e.g. device node: /dev/sg37, driver: sg); for SCSI disks the first two letters are the driver name (e.g. device node: /dev/sdabc, driver: sd) and for NVMe generic devices the "driver" name is "nvme-generic" (e.g. device node: /dev/ng2n1, driver: nvme-generic). Trying to find the nvme-generic driver itself is difficult as it is a built-in part of the core nvme driver.

When a system has been sensibly configured and working properly, this utility should be able to hide the details in the above paragraphs. However when things go wrong, the above information may be useful.

The ls_name_value utility is a general purpose tool for listing name:value pairs, especially from pseudo file systems like sysfs in Linux. In this case the "name" is the filename of a regular file and the corresponding "value" is the contents of that file. Only the first 256 characters of the value are output, and only if all its characters are (7 bit) ASCII characters. This guards against large files (not supposed to be in sysfs but ...) and files containing binary. This utility can scan up to two levels of directories from its starting point downward. A hierarchical file system (forgetting about symlinks) can be viewed as an inverted tree with its root '/' at the top and regular files as its leaves, at the bottom of the inverted tree.

How does this relate to the lsscsi utility? The lsscsi utility scans sysfs which is typically mounted under /sys to obtain the information it presents. Over time sysfs has changed, with many new attributes being added. In rare cases some attributes have been removed (e.g. because the hardware that needs them has not been sold for 15 to 20 years). So in maintaining lsscsi it is useful to periodically check the directories lsscsi scans in sysfs for changes. And this is what the ls_name_value utility (a bash script) was written to help with. The ls_name_value utility has its own manpage.

Written by Doug Gilbert

Report bugs to <dgilbert at interlog dot com>.

Copyright © 2003-2026 Douglas Gilbert
This software is distributed under the GPL version 2. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

lsscsi_json(8) ls_name_value(8) lscpu lsusb lsblk sg_inq, sg_vpd (both in sg3_utils package), clone_pseudo_fs(clone_pseudo_fs)