Reset a Zoned SCSI Write Pointer Safely with sg_reset_wp

sg_reset_wp sends the SCSI RESET WRITE POINTER command to a zoned storage device, for one zone or for every open and full zone at once. This is a state-changing operation: use it only once you understand the device's zone state and have stopped writers that could race with it. Allow about 15 minutes for a prepared maintenance task, longer if you first need to identify the correct device.

The examples use sg_reset_wp from the installed sg3-utils package, version 1.46-3ubuntu4. The utility reports its own version as 1.14 20191220. The local manual page is dated May 2018, while the package is newer, so the installed help output is included below wherever it gives a useful confirmation.

1. Confirm the utility and its syntax

Start with an unprivileged check. It does not contact a storage device or change any zone:

$ command -v sg_reset_wp
/usr/bin/sg_reset_wp
$ dpkg-query -W -f='${Package} ${Version}\n' sg3-utils
sg3-utils 1.46-3ubuntu4
$ sg_reset_wp --version
version: 1.14 20191220

Read the usage text before you prepare a command:

$ sg_reset_wp --help
Usage: sg_reset_wp  [--all] [--count=ZC] [--help] [--verbose]
                    [--version] [--zone=ID] DEVICE
...

The required shape is sg_reset_wp --zone=ZONE_START_LBA DEVICE or sg_reset_wp --all DEVICE. The device argument is normally a SCSI generic pass-through path such as /dev/sg3, but the correct path is host-specific: do not guess it from an example.

2. Identify the target before touching it

Map the device path to the storage system's own inventory. Check the path and permissions without opening it for a write-pointer reset:

$ DEVICE='/dev/sg3'
$ ls -l "$DEVICE"
crw-rw---- 1 root disk 21, 3 ... /dev/sg3

Replace /dev/sg3 with the path you have verified from your SCSI host, enclosure or multipath documentation. A generic device number is not an identity. Check the serial number, enclosure slot and logical unit with your normal inventory tools, and make sure the path has not become a different disk after a reboot or device rescan.

3. Stop competing writers

A reset write pointer changes the device's zoned write state. Stop the application, filesystem workload, RAID operation or other process that can write to the target. Follow the storage vendor's maintenance procedure for flushing or quiescing outstanding I/O. This utility offers no rollback, and the manpage makes no promise to preserve data written after the pointer state becomes inconsistent.

Elevated privileges may be required to open the device. Use them only after the target and maintenance window are confirmed:

$ sudo sg_reset_wp --help
Usage: sg_reset_wp  [--all] [--count=ZC] [--help] [--verbose]
                    [--version] [--zone=ID] DEVICE
...

This is still only a help check. It does not establish that the chosen account can safely operate the device, nor does it prove the device supports RESET WRITE POINTER at all.

4. Reset one zone by its starting LBA

For a single zone, pass the zone's starting logical block address to --zone. The option calls the value a zone ID, but it is specifically a zone start LBA, not any arbitrary block inside the zone:

$ sudo sg_reset_wp --zone=1048576 "$DEVICE"
$ status=$?
$ printf 'sg_reset_wp exit status: %s\n' "$status"
sg_reset_wp exit status: 0

A zero exit status means sg_reset_wp completed successfully according to its documented exit status. It does not tell you the LBA was the zone you intended, or that an application has actually stopped writing. The device's response and your subsequent zone report are the operational checks that matter.

Hexadecimal is also accepted: prefix the value with 0x or add a trailing h. Use one notation consistently in a runbook:

$ sudo sg_reset_wp --zone=0x100000 "$DEVICE"
$ printf 'sg_reset_wp exit status: %s\n' "$?"
sg_reset_wp exit status: 0

Do not run both examples for the same maintenance event. Each successful command sends another reset request.

5. Reset all open and full zones only when required

The --all option sets the command's ALL field. It requests a reset operation for all open zones and full zones, which is a much wider change than resetting one identified zone:

$ sudo sg_reset_wp --all "$DEVICE"
$ printf 'sg_reset_wp exit status: %s\n' "$?"
sg_reset_wp exit status: 0

When --all is present, any --zone=ID value is ignored. Keep the command simple so a review cannot miss its scope. Do not use --all as a diagnostic probe, and do not put it in an automated recovery loop unless the storage design explicitly requires that behaviour.

6. Understand the zone count and diagnostic options

--count=ZC places a zone count in the command. The accepted range is 0 through 65535 inclusive, and the installed help reports a default of 0. The local manual describes where the value is placed, but does not define a safe policy for choosing a non-zero count for your device, so do not add it merely because it is available:

$ sudo sg_reset_wp --zone=1048576 --count=1 "$DEVICE"
$ printf 'sg_reset_wp exit status: %s\n' "$?"
sg_reset_wp exit status: 0

Use a non-zero count only when the device or storage procedure documents the required command semantics. --verbose adds diagnostic output and can help when investigating a failure, but it does not make a failed command safe to repeat:

$ sudo sg_reset_wp --verbose --zone=1048576 "$DEVICE"
$ printf 'sg_reset_wp exit status: %s\n' "$?"
sg_reset_wp exit status: 0

7. Verify the result and recover from failure

After the command returns, use the device's zone-reporting procedure to inspect the affected zone or zones. Confirm the write pointer and zone condition against the storage standard, vendor documentation, or the output of the appropriate sg3-utils reporting utility. A successful process exit alone is not enough for a production recovery record.

If the command returns non-zero, preserve the diagnostic output and stop competing writers. Do not immediately retry with --all, a different LBA, or a larger count. Check that the device is the intended zoned target, that the account can access its SCSI pass-through interface, and that the device supports the command. The manpage directs readers to sg3_utils(8) for the detailed exit-status conventions; the exact sense data from the device is relevant to diagnosis too.

There is no undo command in sg_reset_wp. Recovery means following the storage device's documented procedure, restoring the application or filesystem's consistent state from its own recovery method, and checking the zone report before allowing writers to resume. Keep the original diagnostics with the maintenance record.