60

How do you delete all partitions on a device from the command line on Linux (specifically Ubuntu)? I tried looking at fdisk, but it presents an interactive prompt. I'm looking for a single command, which I can give a device path (e.g. /dev/sda) and it'll delete the ext4, linux-swap, and whatever other partitions it finds. Essentially, this would be the same thing as if I were to open GParted, and manually select and delete all partitions. This seems fairly simple, but unfortunately, I haven't been able to find anything through Google.

Cerin
  • 3,497
  • 17
  • 57
  • 72

9 Answers9

59

Would this suffice?

dd if=/dev/zero of=/dev/sda bs=512 count=1 conv=notrunc
pk.
  • 6,413
  • 1
  • 41
  • 63
  • 6
    This will not delete the partitions. By deleting the partitions he meant to preserve the MBR and just empty the partition table. – Mircea Vutcovici Mar 23 '11 at 19:22
  • 3
    No, this appears to do exactly what I need. I don't really care if the data is still there. GParted shows that the partitions are gone after running this, and that's what I wanted. – Cerin Mar 24 '11 at 11:53
  • Mircea Vutcovici wasn't talking about your data, but about the bootstrap code in your MBR. That's now gone, because you've erased it along with the 4 primary entries from the MBR-style partition table. – JdeBP May 17 '11 at 11:22
  • 2
    Don't forget to unmount the driver, otherwise it won't work. – OrangeTux Dec 06 '13 at 14:47
  • This also works great if you've used ZFS on a drive and are repurposing it for something else. Neither a standard Windows or Linux partitioning and format will get rid of the ZFS labels on it which can cause you major problems. – Tony Maro Nov 06 '18 at 16:50
  • 1
    Is the `conv=notrunc` option really needed? See [here](https://stackoverflow.com/a/20531600/1582182). – Mmmh mmh May 18 '19 at 21:30
  • Does this try and fill the entirety of the /dev/sda disk with zeros? That sounds pretty time consuming if /dev/sda is large, I'm just looking to put a new filesystem on it, there's nothing secure on it that I want to erase. – jrh Apr 07 '20 at 17:43
  • 1
    To be fair, he said delete the partitions, not the partition table or data, so it wasn't really clear. But I think we knew he just wanted the parts not to register, which makes this the correct answer. To delete the data would be uneccesary. IF you want to zero it, thats a different story, that's deleting the data too. – Brian Thomas Aug 10 '20 at 19:04
  • 1
    @jrh : No. This only zeroes (`count=1`) 1 block of (`bs=512`) 512 bytes (starting at the first block of the device). (It also won't truncate the output file, but the output target is a block device, so this seems [to do nothing](https://stackoverflow.com/a/20531600/478891)). – Eric Towers Oct 09 '20 at 21:38
54

The wipefs program lets you easily delete the partition-table signature:

wipefs -a /dev/sda

From man wipefs

wipefs can erase filesystem, raid or partition-table signatures (magic strings) from the specified device to make the signatures invisible for libblkid.

wipefs does not erase the filesystem itself nor any other data from the device. When used without any options, wipefs lists all visible filesystems and the offsets of their basic signatures.

wipefs calls the BLKRRPART ioctl when it has erased a partition-table signature to inform the kernel about the change.

user144437
  • 805
  • 7
  • 6
21

Quick and Dirty: use gparted to delete the partitions, or if you’re in a hurry:

dd if=/dev/zero of=/dev/[disk device] bs=512 count=1

This will zap the MBR of the drive (Data is still intact).

Alternatively:

dd if=/dev/zero of=/dev/[disk device]

to wipe the whole drive (write a single pass of zeros over everything. Not "secure" but usually good enough), or use a "disk shredder" tool for a secure wipe.

Giacomo1968
  • 3,522
  • 25
  • 38
voretaq7
  • 79,345
  • 17
  • 128
  • 213
  • 1
    why is the second "not secure"? – Herdsman Jul 11 '20 at 08:54
  • Not secure because it's possible that the disk's contents could be partially recovered by someone with physical access to it. A secure wipe writes multiple passes of random data over the whole disk. – Alex Grounds Mar 16 '21 at 20:51
11

Use improved non-interactive version of fdisk, which is sfdisk

To erase partition table use this command:

sfdisk --delete /dev/sda

Note: no confirmation will be thrown, the partitions will be deleted instantly and forever!

Suncatcher
  • 552
  • 2
  • 7
  • 22
8

See man sfdisk, which is a non-interactive variant of fdisk. Other than that, you can delete the whole partition table with dd, as pk wrote.

Sven
  • 97,248
  • 13
  • 177
  • 225
7

You should be able to use parted for this aswell, although that may involve some scripting to loop through the partitions.

Decado
  • 1,949
  • 11
  • 17
  • It may be worth to mention that e.g. `parted /dev/mmcblk0 --script mklabel gpt` deletes all existing partitions. For me it was `parted /dev/mmcblk0 --script mklabel msdos` – grenix Jan 28 '22 at 11:30
6

If we're talking about MBR-style partitions...

dd if=/dev/zero of=/dev/[disk device] bs=1 count=64 seek=446 conv=notrunc

Explanation:

dd

This standard command copies bytes from a source and writes them to a destination. It's the simplest flexible tool for this job.

if=/dev/zero

Here, we specify that we're reading from /dev/zero, which is a special device which emits NUL bytes--zeros.

of=/dev/[disk device]

Here, we specify which device we're writing to.

bs=1

dd thinks in terms of blocks. The default block size may be 512 bytes, 1024 bytes or 4096 bytes, depending on your system. However, we need to address things more precisely than that, so we tell dd to use a block size of 1 byte.

count=64

Here, we tell dd to write 64 blocks (or bytes, because of our bs=1 parameter), since the primary partition table consists of 4 16-byte partition entries, for a total of 64 bytes.

seek=446

The primary partition table within the MBR (so, not talking about GPT here) is located 446 bytes in, so we instruct dd to seek 446 bytes in before writing.

Extended partitions are generally created by using a primary partition slot to point at the extended partition table, so if we erase the 4 primary partitions, we effectively wipe the extended partition table as well; the OS won't be able to find it, so it won't be able to read and interpret it. (If you want to wipe the extended partition table, you'll need to know more about the operating system; different operating systems do extended partitions in different ways.)

Michael Mol
  • 292
  • 2
  • 8
3

I wanted to do the same thing (except in Slackware 14.2) but found I could not effect most of the solutions proposed here, with the most elaborate and well-documented solution creating new problems for making replacement partitions. That deleted the partition but some partitioning software apparently found the partition backups automatically.

I found f3probe (http://oss.digirati.com.br/f3) solved the problem of deleting all the partitions, quickly and easily, working with large capacity drives, and created exactly 1 partition spanning the whole drive, which was easy to delete.

It was also easy, from there to create new partitions, in a straight-forward way.

i.e.

f3probe --destructive --time-ops /dev/sdb
# Now we know only 1 partition exists on /dev/sdb
#    which is /dev/sdb1
#
# Unmount that partition
umount /dev/sdb1

#
# Delete that single partition
parted /dev/sdb rm 1

#
# Now you can create new partitions
# i.e. parted /dev/sdb mkpart primary fat32 1049K 15.8G
# 
# Update /etc/fstab before rebooting....
Thomas
  • 4,155
  • 5
  • 21
  • 28
John
  • 31
  • 1
0

Using gparted from petalinux on EMMC for me it was

parted /dev/mmcblk0 --script mklabel msdos

Here the corresponding help output

# parted /dev/mmcblk0 --script help mklabel      
  mklabel,mktable LABEL-TYPE               create a new disklabel[ 6897.473870]  mmcblk0:
 (partition table)

        LABEL-TYPE is one of: aix, amiga, bsd, dvh, gpt, mac, msdos, pc98, sun, loop

BTW: concerning wipefs -a /dev/sda (and maybe some other solutions here)

Using this made commands like parted /dev/mmcblk0 --script mkpart primary fat32 1MiB 2049MiB for me to fail with Error: /dev/mmcblk0: unrecognised disk label. The mklabel command of gparted is 'repairing' this condition resp. seems to be a good preparation for follwing steps.

See also https://unix.stackexchange.com/questions/200582/scripteable-gpt-partitions-using-parted

grenix
  • 101
  • 3