LPIC-2 204.2 — Adjusting Storage Device Access

Exam topic 204.2 — Adjusting Storage Device Access. Covers hdparm/sdparm, tune2fs, sysctl, SSD/NVMe, iSCSI configuration, storage identifiers (WWID, WWN, LUN, IQN), and SAN protocols.

Disk Types and Interfaces

Note: Old PATA disks were named /dev/hda, /dev/hdb, etc. Modern SATA and SCSI disks share the /dev/sd* prefix because the kernel handles them through the common libata driver.

hdparm

hdparm reads, tests, and changes parameters of SATA/IDE disks. It operates through the kernel’s libata interface and supports PATA, ATAPI, SATA, and SCSI disks with SAT (SCSI-ATA Command Translation) support.

Warning: Most hdparm options are dangerous and can damage a disk. Test on non-production disks only and read man hdparm before use.

Syntax

hdparm [options] [device]

Key options

Performance test example

sudo hdparm -tT /dev/sde
# /dev/sde:
#  Timing cached reads:   11810 MB in  2.00 seconds = 5916.78 MB/sec
#  Timing buffered disk reads:  1024 MB in  2.69 seconds =  380.94 MB/sec

Tip: Run the test several times and average the results — a single measurement is not reliable.

Making settings persistent

hdparm changes are lost on reboot. To make them permanent, create or edit:

/etc/udev/rules.d/50-hdparm.rules

sdparm

sdparm interacts with SCSI devices through VPD (Vital Product Data) tables and mode pages. It is not “hdparm for SCSI” — it is a separate tool with a different approach.

sdparm can read serial numbers, part codes, and device descriptions. It can also control disk behavior: stop the spindle, change write-back caching parameters.

Key options

Important for the exam: Running sdparm without options shows all common mode parameters. The --all option shows ALL recognized fields — not the same thing.

tune2fs

tune2fs configures how the kernel interacts with ext2 and ext3 filesystems.

tune2fs [options] /dev/sdXY

Key options

Tip: On large disks, the default 5% reserved blocks is wasteful. On a 2 TB disk that is 100 GB unused. Set it to 1% with tune2fs -m 1 /dev/sda1.

Warning: After changing the sparse superblock flag (-s), run e2fsck to make the filesystem valid again.

sysctl and /proc

sysctl changes kernel parameters at runtime. All tunable parameters live in the /proc/sys/ tree.

sysctl options

sysctl -a
sysctl -w vm.swappiness=10
sysctl -p /etc/sysctl.d/99-sysctl.conf

# Read/write directly via /proc
cat /proc/sys/vm/swappiness
echo 10 > /proc/sys/vm/swappiness

# Enable IP forwarding
echo 1 > /proc/sys/net/ipv4/ip_forward

Warning: Changes made via echo to /proc/sys/ or via sysctl do not persist after reboot. For permanent changes, add a line to /etc/sysctl.conf or a file in /etc/sysctl.d/.

Interrupt monitoring

cat /proc/interrupts

Example output on a single-CPU system:

           CPU0
  0:   16313284   XT-PIC  timer
  1:     334558   XT-PIC  keyboard
 14:     123824   XT-PIC  ide0
 15:          7   XT-PIC  ide1

On multi-core systems additional CPU columns appear. By default all interrupts are handled by CPU0, which becomes a bottleneck when it is flooded with IRQs from network storage while other cores are idle.

Controlling smp_affinity

Each interrupt entry in /proc/interrupts has a subdirectory at /proc/irq/. The smp_affinity file inside contains a bitmask where each bit corresponds to a CPU core.

# Route interrupt #13 to CPU5
echo 32 > /proc/irq/13/smp_affinity

# Route interrupt #12 to CPU1
echo 2 > /proc/irq/12/smp_affinity

Tip: Proper interrupt distribution across cores can noticeably speed up heavily loaded systems with network storage — CPU0 is no longer a bottleneck.

SSD, AHCI, and NVMe

AHCI

AHCI (Advanced Host Controller Interface) is the interface for SATA disks. Designed for mechanical drives: 1 queue of 32 commands. SATA SSDs use the same AHCI, creating a bottleneck.

Important: If an SSD is not recognized, check BIOS/UEFI settings. The SATA mode must be set to AHCI, not IDE.

NVMe

NVMe (Non-Volatile Memory Express) connects via PCIe and uses the NVMHCI protocol. Instead of 1 queue of 32 commands, NVMe supports 65,000 queues of 65,000 commands each. Up to 7× faster than AHCI SSDs.

NVMe support was added in Linux kernel 3.3.

NVMe device naming

/dev/nvme0n1p1
       ^  ^ ^
       |  | └── p1 = partition 1
       |  └──── n1 = namespace 1
       └─────── nvme0 = first NVMe disk (numbered from 0)

Disks are numbered from 0; namespaces and partitions from 1. Third disk, fourth namespace, second partition = /dev/nvme2n4p2.

SSD TRIM

SSDs store data in blocks, erase whole blocks, and write page by page. This leads to write amplification: changing one page requires rewriting an entire block. The TRIM command tells the SSD controller which blocks are free, allowing internal defragmentation.

# Automatic TRIM (via mount option)
/dev/sda1 / ext4 discard,noatime 0 1

# Manual TRIM (for all free blocks in the FS)
fstrim /

nvme-cli

Managing NVMe disks requires the nvme utility from the nvme-cli package. Standard SMART tools (smartctl) do not work with NVMe directly because there is no AHCI or SCSI interface.

nvme help
nvme smart-log /dev/nvme0

iSCSI

iSCSI (Internet Small Computer System Interface) is a network implementation of the SCSI protocol over IP. The client sends SCSI commands (CDB, Command Descriptor Blocks) over the network; the server receives and executes them. Unlike Fibre Channel, iSCSI works on ordinary network infrastructure over LAN, WAN, or the Internet.

Initiator (client) configuration

Configuration file (RHEL and derivatives):

/etc/iscsi/iscsid.conf

Service management:

/etc/init.d/iscsi status
/etc/init.d/iscsi start
/etc/init.d/iscsi restart

Discovering targets

iscsiadm -m discovery -t sendtargets -p 192.168.1.100

After discovery, log in to a volume or restart the service to mount all discovered volumes automatically.

/etc/fstab for iSCSI

/dev/sdb1 /mnt/iscsi ext4 _netdev 0 0

Important: The _netdev option is mandatory for iSCSI volumes in fstab. It ensures the network comes up before the system attempts to mount the device.

/etc/iscsi/iscsid.conf example

# iSNS settings
isns.address = 192.168.1.200
isns.port = 3260

# CHAP authentication
node.session.auth.authmethod = CHAP
node.session.auth.username = myuser
node.session.auth.password = mypassword

# CHAP for discovery sessions
discovery.sendtargets.auth.authmethod = CHAP
discovery.sendtargets.auth.username = discuser
discovery.sendtargets.auth.password = discpassword

iSCSI target (server)

Before kernel 2.6.38 the standard was STGT/TGT. From 2.6.38 the standard Linux target is LIO (linux-iscsi.org), which supports Fibre Channel, FCoE, iSCSI, iSER, SRP, USB, and other fabric protocols.

Legacy: tgtd / tgt

Target configuration file:

/etc/tgt/targets.conf

Example with two LUNs:

<target iqn.2008-09.com.example:server.target1>
    backing-store /srv/images/iscsi-share.img
    direct-store /dev/sdd
</target>

backing-store points to a file or block device; direct-store passes local SCSI device parameters (serial number, vendor) directly into the iSCSI LUN.

Service management:

service tgtd start
service tgtd stop
service tgtd force-stop   # immediately drops all sessions

Modern: LIO + targetcli

systemctl enable target
targetcli

Basic workflow in targetcli:

# 1. Create a backstore
/> cd /backstores/block
/backstores/block> create iscsidisk1 dev=/dev/sde

# 2. Create an IQN for the target
/backstores/block> cd /iscsi
/iscsi> create iqn.2016-02.com.example.server07:iscsidisk1

# 3. Restrict portal to a specific IP (remove 0.0.0.0)
/iscsi> cd iqn.2016-02.com.example.server07:iscsidisk1/tpg1/portals
/iscsi/.../portals> delete 0.0.0.0 3260
/iscsi/.../portals> create 192.168.56.103

# 4. Create a LUN
/iscsi/.../portals> cd ../luns
/iscsi/.../luns> create /backstores/block/iscsidisk1

# 5. Exit (config is saved automatically)
/> exit

Tip: targetcli saves configuration to /etc/target/saveconfig.json on exit. The last 10 config backups are stored in /etc/target/backup/.

Warning: Use only lowercase in backstore names — targetcli behaves unpredictably with uppercase.

After setting up the target, connect from the initiator:

iscsiadm -m discovery -t sendtargets -p 192.168.56.103
iscsiadm -m node -T iqn.2016-02.com.example.server07:iscsidisk1 \
         -p 192.168.56.103 -l
iscsiadm -m session -P3

Discovered iSCSI entries are stored in:

/var/lib/iscsi/nodes/
/var/lib/iscsi/send_targets/

Initiator IQN is set in:

/etc/iscsi/initiatorname.iscsi

scsi_id and LUN Persistence

Linux device names (/dev/sda, /dev/sdb) change between reboots if the disk discovery order changes. LUN persistence solves this: you always reference a disk by a stable name regardless of connection order.

WWID via /dev/disk/by-id/

Linux automatically creates symlinks in /dev/disk/by-id/ where the link name contains the device’s WWID:

ls -l /dev/disk/by-id/
# scsi-3600508b400105e210000900000490000 -> ../../sda   (page 0x83)
# scsi-SSEAGATE_ST373453LW_3HW1RHM6     -> ../../sda   (page 0x80)

There are two WWID types: page 0x83 and page 0x80. Any SCSI device has one of them.

scsi_id

The scsi_id utility retrieves the WWID directly from the device via an inquiry command.

By default scsi_id does not show devices not in its whitelist. To get the WWID of any device, use the -g flag.

# Get WWID (modern syntax)
/sbin/scsi_id -g -u -d /dev/sdc
# 3200049454505080f

# Old sysfs syntax (RHEL5 and older)
scsi_id -g -u -s /block/sdc

To avoid specifying -g every time, edit /etc/scsi_id.conf: remove the options=-b line and add options=-g.

LUN persistence via udev

After obtaining the WWID, create a rule in /etc/udev/rules.d/20-names.rules:

KERNEL="sd*", BUS="scsi", PROGRAM="/sbin/scsi_id -g -u -d /dev/$parent", RESULT="3200049454505080f", NAME="bookone"

Now every time this disk connects, udev compares the scsi_id result to the specified RESULT and creates a device named bookone, regardless of which /dev/sd* the kernel assigns.

Note: On RHEL6 and newer use -d /dev/$parent instead of -s /block/$parent, as new scsi_id versions do not accept sysfs paths.

Multipath

If a device has multiple paths (different NICs or HBAs), device-mapper-multipath uses the WWID to merge them into a single pseudo-device:

/dev/mapper/3600508b400105df70000e00000ac0000

With user_friendly_names enabled, the device gets a name like /dev/mapper/mpath0, and the WWID mapping is stored in:

/etc/multipath/bindings

Show all paths to a device:

multipath -l

Persistent alias in /etc/multipath.conf

Assign a human-readable name via an alias. This works even with a single path — just add the WWID and alias:

multipaths {
    multipath {
        wwid    3600a0b80001327510000015427b625e
        alias   backupdisk
    }
}

The device is always accessible as /dev/mpath/backupdisk.

WWID, WWN, LUN, IQN

Important: IQN identifies both the iSCSI target server and the disk it offers. WWID and WWN uniquely identify only the disk, not the server.

Example IQN:

iqn.2003-01.org.linux-iscsi.server1.x8664:sn.abc123

SAN

SAN (Storage Area Network) is a dedicated high-speed network for access to disk storage. Disks in a SAN appear to the server as locally attached.

Note: AoE and FCoE work only on local networks because they do not support routing. iSCSI can work over WAN and the Internet.

Exam Cheat Sheet

Files and Paths

Key Commands

# hdparm
hdparm -I /dev/sda             # detailed disk info
hdparm -tT /dev/sda            # speed test (disk + cache)
hdparm -d 1 /dev/sda           # enable DMA
hdparm -W 1 /dev/sda           # enable write-caching

# sdparm
sdparm /dev/sda                # common mode parameters
sdparm --all /dev/sda          # all recognized fields
sdparm --command stop /dev/sda # stop the spindle

# tune2fs
tune2fs -m 1 /dev/sda1         # reserve 1% of blocks
tune2fs -e remount-ro /dev/sda1  # behavior on FS error

# sysctl
sysctl -a                      # all kernel parameters
sysctl -w vm.swappiness=10     # set a parameter
sysctl -p /etc/sysctl.conf     # load from file
echo 32 > /proc/irq/13/smp_affinity  # route IRQ13 to CPU5

# iSCSI
iscsiadm -m discovery -t sendtargets -p 192.168.1.100
iscsiadm -m node -T iqn... -p 192.168.1.100 -l
iscsiadm -m session -P3
/etc/init.d/iscsi restart

# targetcli (LIO target)
targetcli
systemctl enable target

# tgtd (legacy TGT target)
service tgtd start
service tgtd force-stop

# scsi_id
/sbin/scsi_id -g -u -d /dev/sdc

# multipath
multipath -l

# NVMe
nvme help
nvme smart-log /dev/nvme0

# TRIM
fstrim /

Common Exam Traps

• sdparm without options shows common mode parameters. --all shows all fields — not the same.
• hdparm changes are temporary. Without /etc/udev/rules.d/50-hdparm.rules they disappear on reboot.
• iSCSI entries in /etc/fstab must have the _netdev option. Without it the system hangs at boot trying to mount the volume before the network is up.
• WWID and WWN identify the disk; IQN identifies both the target server and the disk. Different exam answers.
• NVMe naming: disk from 0, namespace and partition from 1. Second disk, first namespace, first partition = /dev/nvme1n1p1.

Practice Questions

Q1. Which command shows the disk read speed (not cache) for /dev/sdb?

Answer: hdparm -t /dev/sdb — lowercase -t tests buffered disk reads. Uppercase -T tests cache reads.

Q2. An administrator runs sdparm /dev/sdc without parameters. What is displayed?

Answer: All common mode parameters of the device. --all would show all recognized fields (different output).

Q3. Which file must be created to make hdparm settings persist across reboots?

Answer: /etc/udev/rules.d/50-hdparm.rules — hdparm settings are applied at boot via udev rules.

Q4. What does /dev/nvme1n3p2 mean?

Answer: Second NVMe disk (nvme1, numbered from 0), third namespace (n3, from 1), second partition (p2, from 1).

Q5. Which identifier simultaneously points to both the iSCSI target server and the disk it offers?

Answer: IQN (iSCSI Qualified Name). WWID and WWN uniquely identify only the disk.

Q6. An admin added an iSCSI disk to /etc/fstab. The system hangs at boot before the disk is mounted. What is most likely wrong?

Answer: The _netdev option is missing. Without it the system tries to mount the device before the network is up. Adding _netdev delays the mount until network initialization is complete.

Q7. Which utility manages NVMe disks and retrieves SMART logs from them?

Answer: nvme (from the nvme-cli package). smartctl does not work with NVMe because there is no AHCI or SCSI interface.

Q8. What percentage of blocks does tune2fs reserve by default, and how do you change it to 1%?

Answer: 5% by default. Change with tune2fs -m 1 /dev/sda1.