Inspect Partition Tables Safely with partx

Before you reach for fdisk or a script that parses partition rows by hand, check whether partx already gives you exactly the columns you need. You'll finish with a repeatable way to read a disk or disk image's partition table, select one partition, and produce script-friendly output, plus a clear picture of what the administrative options actually change. Allow about fifteen minutes.

• Needs: a shell and the util-linux package. A test image is handy if you don't want to touch a real disk.
• Versions matter: this follows the installed manual for util-linux 2.39.3. On this machine, the Ubuntu package is util-linux 2.39.3-9ubuntu6.6, but partx itself resolves to a util-linux 2.41.3 binary in /home/linuxbrew/.linuxbrew/sbin. Check your own path and version before trusting a small output difference.

1. Confirm which partx you are running

Start with read-only checks, no elevated privileges needed:

$ command -v partx
/home/linuxbrew/.linuxbrew/sbin/partx
$ partx --version
partx from util-linux 2.41.3
$ partx --help

The command takes a disk, or a partition together with its parent disk. The main read-only mode is --show (short form -s). The older --list form is documented as deprecated, so use --show in anything new.

2. List a real disk or image

Swap in a path you have permission to read. Listing is normally unprivileged, though device permissions or host policy can still demand sudo:

$ partx --show /path/to/disk-image.img
NR START   END SECTORS SIZE NAME UUID
 1  2048 18431   16384   8M      45cfca7a-01
 2 18432 34815   16384   8M      45cfca7a-02

Your rows will differ. NR is the partition number, START and END are sector positions, SECTORS is the length. Every displayed number is a 512-byte sector except SIZE, which is human-readable and uses binary units, so it shows 8M rather than the longer 8 MiB form.

For a block device, use its exact path such as /dev/sdb, confirmed with lsblk rather than guessed, and check its mount points before running anything that changes kernel bookkeeping.

3. Select one partition with stable columns

Use --nr to pick a particular partition or range. A single number selects one; a colon range selects an inclusive interval:

$ partx --show --nr 2 /path/to/disk-image.img
NR START   END SECTORS SIZE NAME UUID
 2 18432 34815   16384   8M      45cfca7a-02
$ partx --show --nr 2:4 /path/to/disk-image.img

• Open-ended ranges work too: 2: means partition 2 onwards, :4 means through partition 4.
• Negative numbers count from the end: --nr -1:-1 selects the last partition.
• Quote a range if your shell might read a leading hyphen as an option, or use the long option exactly as shown.

For scripts, choose columns explicitly and drop the headings rather than depending on the default display:

$ partx --show --noheadings --output NR,START,SECTORS,SIZE --nr 2 /path/to/disk-image.img
2 18432 16384 8M

Reach for --bytes when a byte count beats a human-readable size, --pairs for KEY="value" records, or --raw for the raw format. Run partx --help on your actual binary to see the columns it supports.

4. Scan a partition as a nested disk

If a partition contains its own partition table, pass a hyphen as the disk argument. This tells partx to treat that partition as a whole disk rather than one entry inside its parent:

$ partx --show - /dev/sda3

It's a read-only scan, but the device path still matters, confirm /dev/sda3 is the intended partition first. If it's ordinary filesystem data, an empty result or a parse error is expected: don't initialise it just to make partx print rows.

5. Understand add, delete and update

--add, --delete and --update are not partitioning. They tell the kernel about partitions that already exist in the on-disk table; they never rewrite that table. They can still affect device nodes and processes using the device, so treat them as privileged and service-disrupting.

$ sudo partx --add --nr 3:5 /dev/sdd
$ sudo partx --delete --nr 5 /dev/sdd
$ sudo partx --update --nr 3:5 /dev/sdd

The first adds available partitions 3 through 5 to the kernel's view. The second removes partition 5 from that view. The third asks the kernel to update the specified partitions. None of these undoes a changed partition table, because none of them changes the table in the first place. To restore kernel bookkeeping after a mistaken delete, use --add for the partition that still exists on disk, once you've confirmed the device is safe to rescan.

Don't use --add as a substitute for fdisk, parted or another partition editor. If the table itself is wrong, take a backup before reaching for a tool that writes it.

6. Diagnose an unexpected result

Rerun the harmless checks first:

$ stat /path/to/disk-image.img
$ partx --show --verbose /path/to/disk-image.img
$ printf 'partx exit status: %s\n' "$?"

• Non-zero status: check the path is readable, is the intended disk or image, and has a table type the command recognises.
• Wrong table type or sector size: --type TYPE selects a partition table type when autodetection fails, and --sector-size SIZE overrides the assumed sector size. Use these only when the image format or device documentation justifies them, not as generic fixes for a wrong path.
• Nested partition confusion: revisit the hyphen form from step 4. For just partition 3 of /dev/sdb, use either partx --show --nr 3 /dev/sdb or partx --show /dev/sdb3 /dev/sdb, both identify the same partition.