26

I want to wipe an SSD clean of all its partitions and data, so I can repartition it (this is not for security purposes).

I've looked at sudo dd if=/dev/zero of=/dev/sdb bs=1M but if this just fills each partition with zeros, I'm not sure this what I want to do.

I plan to run this command quite often, and I don't want to exhaust my SSD's writes. Could anyone give me advice on tackling this problem?

10
  • 1
    Study this: ATA Secure Erase. I have no experience with the process, so this is not an answer, just a hint. Anyone feel free to write your own answer about this if you know this is the right way. Jan 11, 2018 at 8:54
  • 19
    For security purposes, or merely because you need to repartition? Jan 11, 2018 at 8:55
  • mostly just to repartition
    – IgDV
    Jan 11, 2018 at 9:07
  • 14
    12 gauge pump-action. And you knew someone was going to say it sooner or later!
    – ssimm
    Jan 11, 2018 at 13:40
  • 1
    @ssimm - My preference is a charcoal barbecue. Jan 12, 2018 at 2:08

3 Answers 3

43

On an SSD: You can TRIM whole disks or partitions using blkdiscard. It's not very secure, but practically instant (the disk merely marks all cells as unused).

For SATA SSDs, the ATA "Secure Erase" command (available through hdparm) is also very fast. There is also "Enhanced Secure Erase", which (at least on the SSDs I've used it on) takes a few seconds longer and appears to physically erase all cells.

For security: Use full-disk encryption. Don't bother wiping the entire disk if it's encrypted – you only need to wipe the area containing your keys (e.g. the first 1–2 MiB of every encrypted partition).

For repartitioning: Again, don't bother erasing all data. You only need to destroy the filesystems using wipefs, then scrub the first 1 MiB of your disk to purge leftover bootloaders. After you format a partition using mkfs, the OS will simply assume it is completely empty.

(In fact, on Linux, mkfs.ext4 will automatically TRIM the entire partition before formatting it.)

8
  • 6
    For repartitioning, why not just delete old partitions and create new ones? Jan 11, 2018 at 10:41
  • 5
    Because if you create new partitions at the same location (start position), you'll just find the same old data in them! (Of course that's usually not a problem, because reformatting with mkfs will trash it anyway. But various odd situations do happen – wipefs exists because it was needed.) Jan 11, 2018 at 10:45
  • 2
    What command will I need to use to apply this to /dev/sdb just wipefs /dev/sdb ?
    – IgDV
    Jan 11, 2018 at 12:35
  • 4
    @IgDV - what does man wipefs say? (Hint: look at the first example in the man page's examples section)
    – Timo
    Jan 11, 2018 at 15:51
  • 2
    "you only need to wipe the area containing your keys" - on SSD the keys are likely to be recoverable as well. TPM or strong password protecting keys might be better option. blkdiscard would be nice too as it would create a puzzle out of rest of the disk even if attacker knew key (assuming the old mapping cannot be recovered as well). "then scrub the first 1 MiB of your disk to purge leftover bootloaders" - the extra section is used only as overflow of first 512B. If you wipe first 512B you're not harmed by the rest of 1M. Jan 11, 2018 at 21:50
14

As Kamil Maciorowski mentions, the best way, with the least write-wear, of deleting an entire disk, is to use the ATA 'secure erase' command. This will instruct the hardware to do a single full wipe, rather than overwriting the cells repeatedly as with tools like shred. This can only be done for the entire disk, if you need to selectively wipe partitions, see grawity's answer (blkdiscard)

The exact implementation of the command depends on the hardware.

  • Most SSDs will use a bulk electric signal to wipe entire chips in an all-or-nothing fashion. This does incur (normal) write-wear, but only to the minimum extent possible (~ single write cycle).

  • Self-encrypting SSDs will usually just wipe the encryption key inside the controller chip (really instantaneous). Self-encrypting drives always encrypt, even out-of-the-box (with a factory-default key). So wiping the key leaves only un-decryptable jumble on the flash chips, even if no user-key was set.

  • Spinning-Rust hard-disks will do a hardware-based zero-write of all sectors, which is equivalent (and as time-consuming) as doing dd if=/dev/zero.

The process is documented quite well here: https://www.thomas-krenn.com/en/wiki/SSD_Secure_Erase (I have personally used this process repeatedly on my own SSDs when re-installing OS'es)

Edit: if you're interested in the security implications: check this Security.SE's question

7
  • Is Secure Erase supported by most ATA disks? I was under the impression that its support is uncommon.
    – Ruslan
    Jan 11, 2018 at 10:21
  • 1
    I haven't found a (modern, SATA) disk yet that didn't support it, but I admit my experience is limited to harddrives I have personally owned. There also seems to be a difference between "secure erase" and "enhanced secure erase", with the latter one being less supported. Jan 11, 2018 at 12:17
  • 3
    @Ruslan I have never seen (or heard of) a SATA-2 specification disk that didn't support it. Just about any SATA disk made in the last 10 years or so should have it.
    – Tonny
    Jan 11, 2018 at 14:56
  • Does this also mean that on self-encrypting SSDs the TRIM command is a no-op? (Otherwise it's not always encrypting, right?)
    – user541686
    Jan 11, 2018 at 22:18
  • 1
    @Mehrdad: Yes, using TRIM on an encrypted SSD leaks information about which blocks are free. Whether that is acceptable depends on your security needs. See e.g. askubuntu.com/questions/399211/… for a discussion.
    – sleske
    Jan 12, 2018 at 13:55
4

Keeping in mind you've asked for a solution on how to quickly wipe disks Replace /dev/sdx with your disk, most likly /dev/sda

This will wipe the partition table.

dd if=/dev/zero of=/dev/sdx bs=1024 count=50

This will wipe the entire disk, it will take a while.

cat /dev/zero > /dev/sdx
3
  • 4
    Do not use cat. Use dd to improve performance. Jan 11, 2018 at 19:57
  • 6
    @Micheal You say that, but by default dd uses 512-byte buffers, which will ruin performance... Coreutils (cp and presumably also cat) have a more sensible buffer size. There's nothing about dd that makes it somehow inherently faster, anyway. Jan 11, 2018 at 22:21
  • 4
    @grawity You can change the buffer size with dd. Last time I tried to use cat to write an image to a disk, it was significantly slower than using dd with an appropriate buffer size. Jan 12, 2018 at 22:04

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .