[Discuss] TrueCrypt with SSD

[cra at WPI.EDU (Chuck Anderson)]

On Mon, Aug 15, 2011 at 07:45:26AM -0400, Edward Ned Harvey wrote:
> Incidentally, what *is* the problem with TrueCrypt anyway?  It seems to me,
> a hard drive looks like a hard drive whether it's a HDD or SSD.  I would
> expect it to be fine.  Do they have any details anywhere, what is the
> problem?

Sure, an SSD looks the same as an HDD to the application running on
it.  But unlike HDDs, SSDs perform (in some cases much) worse and wear
out faster when they are full vs. when they still have empty unused
space.  So the ATA TRIM (and SCSI Discard) command was invented as a
way for the application (i.e. filesystem, RAID subsystem, LVM,
encryption subsystem, etc.) to tell the SSD which sectors are free/no
longer in use.  In a normal, unencrypted scenario, a disk starts out
mostly empty--the sectors start out zeroed and are only written to
when they are actually needed.  In contrast, encrypted disks are
always completely "full" or "untrimmed", because from the time they
are initially formatted and encrypted every sector is written to with
non-zero data because the empty encrypted sectors (i.e. the unused,
zeroed blocks of a filesystem living on top of the encrypted mapping)
appear random once they are encrypted onto the underlying physical
device.

TrueCrypt and many other full disk encryption packages cannot tell the
drive which sectors are actually free (and hence maintain them as
zeroed sectors on the SSD) because they don't support TRIM.  Many of
the packages don't want to support TRIM because it would leak
information about the encrypted disk to a potential attacker.  That's
pretty much it in a nutshell.