VMworld 2010 Recap: SSD Security

Themes I picked up at VMworld this year:

  • Difference between security and compliance
  • Mixed-mode compliance (burden of proof and segmentation)
  • Guidelines for compliance practices in cloud
  • Encryption with key management
  • Tokens for cloud to resolve latency between web in cloud and DB in private
  • Automation risks (HAL9000)
  • Forensics of cloud

Each one of those is a fun topic in and of itself. I could certainly write a book about securing the cloud, or the virtual environment, at this point.

One very interesting detail also caught my attention. I walked into the Kingston booth and saw them promoting SSD technology.

Never one to shy away from the opportunity to ask about security I found the most technical staff person in the booth and peppered him with questions. Specifically I asked about the difference in forensics, given the lack of a spinning disk.

It very quickly became apparent that a secure wipe of a SSD was IMPOSSIBLE. The Kingston tech told me point-blank that as pages in the disk where data is written become unstable, the disk marks them and then stops writing to them. The data is simply left behind.

I asked Kingston if they would release a tool to wipe those tombstoned pages and they said no. I asked if they would release a tool to read them. No. They will simply sit there with data; without manufacturer support there is no way for the disk owner to get at those dead pages of data.

Then I asked if it would be possible for me to connect directly to flash chips to audit for data or measure the success of a secure-wipe. The answer, as expected, was no. And for good measure the Kingston rep told me they OEM chips (commodity-based or market-based storage on the boards) so not only would I be unable to get help for bypassing a controller and direct access to chips…the chips would probably be different from SSD to SSD.

OK, then.

The good news for privacy is that when an SSD dies it is really dead and very difficult to recover. It may be nearly impossible without expensive tools to recover the data, unlike recovery from spinning disks.

The bottom line is that flash-based “disks” such as SSD use a completely new internal architecture. Our old drives used magnetic storage “platters” and we have a long history of reading and recovering memory locations even if we have to pull out the platters and play them like records.

SSD disks of non-moving flash chips soldered to a board, however, have a whole new setup and aren’t susceptible to our old magnetic tricks (don’t even try degaussing).

Instead of an arm moving along the platters, an SSD has a FTL (flash translation layer) to put data down, move it around and keep track of it. This layer is proprietary and means directly writing or modifying storage is impossible, for example, without first having the FTL erase a space. If the FTL decides a “page” isn’t worth writing to anymore because of wear and tear, then that page is set aside despite still having data on it.

Interesting to note that wear and tear is a serious problem with SSD. The pages have only so much life to them, unlike our old magnetic platters that seem to last infinitely. So running a “wipe” command actually can reduce the life of the disk. Trying to defrag a SSD (totally useless because there is no platter/arm speed issue) therefore just reduces the life of the disk by increasing wear and tear on it.

What this all means, really, is that future storage systems will have to rely more and more on encryption for data privacy. Traditional methods such as multi-pass erase/wipe (meant to ensure all paths of the head hit the platter) are now irrelevant. The consumer/owner of the storage simply will not be able to secure erase based on guidance from the past. Instead, erasing a key will make all the encrypted data on the disks no longer readable.

From the smallest SSD to the largest array of storage, encryption is the future with the focus of secure-wipe procedures now left only for the key.

One thought on “VMworld 2010 Recap: SSD Security”

  1. [quote]
    The good news for privacy is that when an SSD dies it is really dead and very difficult to recover. It may be nearly impossible without expensive tools to recover the data, unlike recovery from spinning disks.
    [/quote]

    I wouldn’t trust on this difficulty at all. An SSD may also fail due to typical electronics problems such as bad capacitors, a non-working crystal oscillator, bad soldering or (maybe) a blown SMD fuse. Errors like this can be repaired without expensive lab tools. It is also possible that a failing SATA transmission circuit only works with some SATA controllers, reduced transmission speed (1.5 GBPS instead of 6 GPBS) and short cables.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.