On Thu, Aug 23, 2012 at 01:00:31PM +0200, Philipp Kern wrote: > On Thu, Aug 23, 2012 at 12:40:10PM +0300, Riku Saikkonen wrote: > > - Does this step write random data or only 0s to the disk? (I think > > messages #15 and #35 in this bug report show that this is not clear > > to the users.) > > It writes 0s to the encrypted container which results in random data on disk > (which is by design).
(sending also to #721360, should probably be merged with this bug report) Hi, sharing my experiences here, I recently noticed this slowness when erasing a 300GB volume. Erasing from debian-installer looked really slow. I did some tests with a 1GB volume and /dev/urandom, resulting in ~0.5GB/min, that means at least 10 hours expected. Later noticed that zeros are written to the encripted volume, as pointed out above, but unfortunately I did not test with debian-installer and the original 1GB test volume and I no longer have that box for testing, so I do not know the actual times for debian-installer in that 1GB reference volume. However, looking at faster ways to do this, I found this reference http://www.globallinuxsecurity.pro/quickly-fill-a-disk-with-random-bits-without-dev-urandom/ claiming to be a faster method when CPU has AES suport. Indeed, with this method and the script (uptime only used to know the times) ------------ 8< ------------------------------- #!/bin/sh set -x cryptsetup luksFormat $1 || exit 1 cryptsetup luksOpen $1 cryptodev || exit 1 cat /proc/uptime dd if=/dev/zero bs=1M of=/dev/mapper/cryptodev cryptsetup luksClose cryptodev dd if=/dev/urandom bs=1M count=2 of=$1 cat /proc/uptime ------------ 8< ------------------------------- I wrote the 300GB volume in ~1.5 hours, something I find reasonable. Do not know about the randomness quality, but this method seems to do something similar to what partman-crypto is doing, although in a way that is aware of CPU AES support. Note that I did this from a shell inside debian-installer, with no extra packages installed. Regards, -- Agustin -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/[email protected]
