8 6. Backup and Data Recovery
9 7. Interoperability with other Disk Encryption Tools
10 8. Issues with Specific Versions of cryptsetup
11 9. References and Further Reading
20 This is the FAQ (Frequently Asked Questions) for cryptsetup. It
21 covers Linux disk encryption with plain dm-crypt (one passphrase,
22 no management, no metadata on disk) and LUKS (multiple user keys
23 with one master key, anti-forensic features, metadata block at
24 start of device, ...). The latest version of this FAQ should
25 usually be available at
26 http://code.google.com/p/cryptsetup/wiki/FrequentlyAskedQuestions
31 ATTENTION: If you are going to read just one thing, make it the
32 section on Backup and Data Recovery. By far the most questions on
33 the cryptsetup mailing list are from people that managed to damage
34 the start of their LUKS partitions, i.e. the LUKS header. In
35 most cases, there is nothing that can be done to help these poor
36 souls recover their data. Make sure you understand the problem and
37 limitations imposed by the LUKS security model BEFORE you face
38 such a disaster! In particular, make sure you have a current header
39 backup before doing any potentially dangerous operations.
41 SSDs/FLASH DRIVES: SSDs and Flash are different. Currently it is
42 unclear how to get LUKS or plain dm-crypt to run on them with the
43 full set of security features intact. This may or may not be a
44 problem, depending on the attacher model. See Section 5.17.
46 BACKUP: Yes, encrypted disks die, just as normal ones do. A full
47 backup is mandatory, see Section "6. Backup and Data Recovery" on
48 options for doing encrypted backup.
50 CLONING/IMAGING: If you clone or image a LUKS container, you make a
51 copy of the LUKS header and the master key will stay the same!
52 That means that if you distribute an image to several machines, the
53 same master key will be used on all of them, regardless of whether
54 you change the passphrases. Do NOT do this! If you do, a root-user
55 on any of the machines with a mapped (decrypted) container or a
56 passphrase on that machine can decrypt all other copies, breaking
57 security. See also Item 6.15.
59 DISTRIBUTION INSTALLERS: Some distribution installers offer to
60 create LUKS containers in a way that can be mistaken as activation
61 of an existing container. Creating a new LUKS container on top of
62 an existing one leads to permanent, complete and irreversible data
63 loss. It is strongly recommended to only use distribution
64 installers after a complete backup of all LUKS containers has been
67 NO WARNING ON NON-INTERACTIVE FORMAT: If you feed cryptsetup from
68 STDIN (e.g. via GnuPG) on LUKS format, it does not give you the
69 warning that you are about to format (and e.g. will lose any
70 pre-existing LUKS container on the target), as it assumes it is
71 used from a script. In this scenario, the responsibility for
72 warning the user and possibly checking for an existing LUKS header
73 is shifted to the script. This is a more general form of the
76 LUKS PASSPHRASE IS NOT THE MASTER KEY: The LUKS passphrase is not
77 used in deriving the master key. It is used in decrypting a master
78 key that is randomly selected on header creation. This means that
79 if you create a new LUKS header on top of an old one with
80 exactly the same parameters and exactly the same passphrase as the
81 old one, it will still have a different master key and your data
82 will be permanently lost.
84 PASSPHRASE CHARACTER SET: Some people have had difficulties with
85 this when upgrading distributions. It is highly advisable to only
86 use the 95 printable characters from the first 128 characters of
87 the ASCII table, as they will always have the same binary
88 representation. Other characters may have different encoding
89 depending on system configuration and your passphrase will not
90 work with a different encoding. A table of the standardized first
91 128 ASCII characters can, e.g. be found on
92 http://en.wikipedia.org/wiki/ASCII
95 * 1.3 System specific warnings
97 - Ubuntu as of 4/2011: It seems the installer offers to create
98 LUKS partitions in a way that several people mistook for an offer
99 to activate their existing LUKS partition. The installer gives no
100 or an inadequate warning and will destroy your old LUKS header,
101 causing permanent data loss. See also the section on Backup and
104 This issue has been acknowledged by the Ubuntu dev team, see here:
105 http://launchpad.net/bugs/420080
107 Update 7/2012: I am unsure whether this has been fixed by now, best
111 * 1.4 My LUKS-device is broken! Help!
113 First: Do not panic! In many cases the data is still recoverable.
114 Do not do anything hasty! Steps:
116 - Take some deep breaths. Maybe add some relaxing music. This may
117 sound funny, but I am completely serious. Often, critical damage is
118 done only after the initial problem.
120 - Do not reboot. The keys mays still be in the kernel if the device
123 - Make sure others do not reboot the system.
125 - Do not write to your disk without a clear understanding why this
126 will not make matters worse. Do a sector-level backup before any
127 writes. Often you do not need to write at all to get enough access
128 to make a backup of the data.
132 - Read section 6 of this FAQ.
134 - Ask on the mailing-list if you need more help.
137 * 1.5 Who wrote this?
139 Current FAQ maintainer is Arno Wagner <arno@wagner.name>. Other
140 contributors are listed at the end. If you want to contribute, send
141 your article, including a descriptive headline, to the maintainer,
142 or the dm-crypt mailing list with something like "FAQ ..." in the
143 subject. You can also send more raw information and have me write
144 the section. Please note that by contributing to this FAQ, you
145 accept the license described below.
147 This work is under the "Attribution-Share Alike 3.0 Unported"
148 license, which means distribution is unlimited, you may create
149 derived works, but attributions to original authors and this
150 license statement must be retained and the derived work must be
151 under the same license. See
152 http://creativecommons.org/licenses/by-sa/3.0/ for more details of
155 Side note: I did text license research some time ago and I think
156 this license is best suited for the purpose at hand and creates the
160 * 1.5 Where is the project website?
162 There is the project website at http://code.google.com/p/cryptsetup/
163 Please do not post questions there, nobody will read them. Use
164 the mailing-list instead.
167 * 1.6 Is there a mailing-list?
169 Instructions on how to subscribe to the mailing-list are at on the
170 project website. People are generally helpful and friendly on the
173 The question of how to unsubscribe from the list does crop up
174 sometimes. For this you need your list management URL, which is
175 sent to you initially and once at the start of each month. Go to
176 the URL mentioned in the email and select "unsubscribe". This page
177 also allows you to request a password reminder.
179 Alternatively, you can send an Email to dm-crypt-request@saout.de
180 with just the word "help" in the subject or message body. Make sure
181 to send it from your list address.
183 The mailing list archive is here:
184 http://dir.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt
187 * 1.7 Unsubscribe from the mailing-list
189 Send mail to dm-crypt-unsubscribe@saout.de from the subscribed
190 account. You will get an email with instructions.
192 Basically, you just have to respond to it unmodified to get
193 unsubscribed. The listserver admin functions are not very fast. It
194 can take 15 minutes or longer for a reply to arrive (I suspect
195 greylisting is in use), so be patient.
197 Also note that nobody on the list can unsubscribe you, sending
198 demands to be unsubscribed to the list just annoys people that are
199 entirely blameless for you being subscribed.
201 If you are subscribed, a subscription confirmation email was sent
202 to your email account and it had to be answered before the
203 subscription went active. The confirmation emails from the
204 listserver have subjects like these (with other numbers):
206 Subject: confirm 9964cf10.....
208 and are sent from dm-crypt-request@saout.de. You should check
209 whether you have anything like it in your sent email folder. If
210 you find nothing and are sure you did not confirm, then you should
211 look into a possible compromise of your email account.
217 * 2.1 What is the difference between "plain" and LUKS format?
219 First, unless you happen to understand the cryptographic background
220 well, you should use LUKS. It does protect the user from a lot of
221 common mistakes. Plain dm-crypt is for experts.
223 Plain format is just that: It has no metadata on disk, reads all
224 parameters from the commandline (or the defaults), derives a
225 master-key from the passphrase and then uses that to de-/encrypt
226 the sectors of the device, with a direct 1:1 mapping between
227 encrypted and decrypted sectors.
229 Primary advantage is high resilience to damage, as one damaged
230 encrypted sector results in exactly one damaged decrypted sector.
231 Also, it is not readily apparent that there even is encrypted data
232 on the device, as an overwrite with crypto-grade randomness (e.g.
233 from /dev/urandom) looks exactly the same on disk.
235 Side-note: That has limited value against the authorities. In
236 civilized countries, they cannot force you to give up a crypto-key
237 anyways. In the US, the UK and dictatorships around the world,
238 they can force you to give up the keys (using imprisonment or worse
239 to pressure you), and in the worst case, they only need a
240 nebulous "suspicion" about the presence of encrypted data. My
241 advice is to either be ready to give up the keys or to not have
242 encrypted data when traveling to those countries, especially when
243 crossing the borders.
245 Disadvantages are that you do not have all the nice features that
246 the LUKS metadata offers, like multiple passphrases that can be
247 changed, the cipher being stored in the metadata, anti-forensic
248 properties like key-slot diffusion and salts, etc..
250 LUKS format uses a metadata header and 8 key-slot areas that are
251 being placed at the beginning of the disk, see below under "What
252 does the LUKS on-disk format looks like?". The passphrases are used
253 to decrypt a single master key that is stored in the anti-forensic
256 Advantages are a higher usability, automatic configuration of
257 non-default crypto parameters, defenses against low-entropy
258 passphrases like salting and iterated PBKDF2 passphrase hashing,
259 the ability to change passphrases, and others.
261 Disadvantages are that it is readily obvious there is encrypted
262 data on disk (but see side note above) and that damage to the
263 header or key-slots usually results in permanent data-loss. See
264 below under "6. Backup and Data Recovery" on how to reduce that
265 risk. Also the sector numbers get shifted by the length of the
266 header and key-slots and there is a loss of that size in capacity
267 (1MB+4096B for defaults and 2MB for the most commonly used
268 non-default XTS mode).
271 * 2.2 Can I encrypt an already existing, non-empty partition to use
274 There is no converter, and it is not really needed. The way to do
275 this is to make a backup of the device in question, securely wipe
276 the device (as LUKS device initialization does not clear away old
277 data), do a luksFormat, optionally overwrite the encrypted device,
278 create a new filesystem and restore your backup on the now
279 encrypted device. Also refer to sections "Security Aspects" and
280 "Backup and Data Recovery".
282 For backup, plain GNU tar works well and backs up anything likely
283 to be in a filesystem.
286 * 2.3 How do I use LUKS with a loop-device?
288 This can be very handy for experiments. Setup is just the same as
289 with any block device. If you want, for example, to use a 100MiB
290 file as LUKS container, do something like this:
292 head -c 100M /dev/zero > luksfile # create empty file
293 losetup /dev/loop0 luksfile # map luksfile to /dev/loop0
294 cryptsetup luksFormat /dev/loop0 # create LUKS on loop device
296 Afterwards just use /dev/loop0 as a you would use a LUKS partition.
297 To unmap the file when done, use "losetup -d /dev/loop0".
300 * 2.4 When I add a new key-slot to LUKS, it asks for a passphrase but
301 then complains about there not being a key-slot with that
304 That is as intended. You are asked a passphrase of an existing
305 key-slot first, before you can enter the passphrase for the new
306 key-slot. Otherwise you could break the encryption by just adding a
307 new key-slot. This way, you have to know the passphrase of one of
308 the already configured key-slots in order to be able to configure a
312 * 2.5 Encryption on top of RAID or the other way round?
314 Unless you have special needs, place encryption between RAID and
315 filesystem, i.e. encryption on top of RAID. You can do it the other
316 way round, but you have to be aware that you then need to give the
317 passphrase for each individual disk and RAID autodetection will
318 not work anymore. Therefore it is better to encrypt the RAID
319 device, e.g. /dev/dm0 .
322 * 2.6 How do I read a dm-crypt key from file?
324 Note that the file will still be hashed first, just like keyboard
325 input. Use the --key-file option, like this:
327 cryptsetup create --key-file keyfile e1 /dev/loop0
330 * 2.7 How do I read a LUKS slot key from file?
332 What you really do here is to read a passphrase from file, just as
333 you would with manual entry of a passphrase for a key-slot. You can
334 add a new passphrase to a free key-slot, set the passphrase of an
335 specific key-slot or put an already configured passphrase into a
336 file. In the last case make sure no trailing newline (0x0a) is
337 contained in the key file, or the passphrase will not work because
338 the whole file is used as input.
340 To add a new passphrase to a free key slot from file, use something
343 cryptsetup luksAddKey /dev/loop0 keyfile
345 To add a new passphrase to a specific key-slot, use something like
348 cryptsetup luksAddKey --key-slot 7 /dev/loop0 keyfile
350 To supply a key from file to any LUKS command, use the --key-file
351 option, e.g. like this:
353 cryptsetup luksOpen --key-file keyfile /dev/loop0 e1
356 * 2.8 How do I read the LUKS master key from file?
358 The question you should ask yourself first is why you would want to
359 do this. The only legitimate reason I can think of is if you want
360 to have two LUKS devices with the same master key. Even then, I
361 think it would be preferable to just use key-slots with the same
362 passphrase, or to use plain dm-crypt instead. If you really have a
363 good reason, please tell me. If I am convinced, I will add how to
367 * 2.9 What are the security requirements for a key read from file?
369 A file-stored key or passphrase has the same security requirements
370 as one entered interactively, however you can use random bytes and
371 thereby use bytes you cannot type on the keyboard. You can use any
372 file you like as key file, for example a plain text file with a
373 human readable passphrase. To generate a file with random bytes,
374 use something like this:
376 head -c 256 /dev/random > keyfile
379 * 2.10 If I map a journaled file system using dm-crypt/LUKS, does it
380 still provide its usual transactional guarantees?
382 Yes, it does, unless a very old kernel is used. The required flags
383 come from the filesystem layer and are processed and passed onwards
384 by dm-crypt. A bit more information on the process by which
385 transactional guarantees are implemented can be found here:
387 http://lwn.net/Articles/400541/
389 Please note that these "guarantees" are weaker than they appear to
390 be. One problem is that quite a few disks lie to the OS about
391 having flushed their buffers. Some other things can go wrong as
392 well. The filesystem developers are aware of these problems and
393 typically can make it work anyways. That said, dm-crypt/LUKS will
394 not make things worse.
396 One specific problem you can run into though is that you can get
397 short freezes and other slowdowns due to the encryption layer.
398 Encryption takes time and forced flushes will block for that time.
399 For example, I did run into frequent small freezes (1-2 sec) when
400 putting a vmware image on ext3 over dm-crypt. When I went back to
401 ext2, the problem went away. This seems to have gotten better with
402 kernel 2.6.36 and the reworking of filesystem flush locking
403 mechanism (less blocking of CPU activity during flushes). It
404 should improve further and eventually the problem should go away.
407 * 2.11 Can I use LUKS or cryptsetup with a more secure (external)
408 medium for key storage, e.g. TPM or a smartcard?
410 Yes, see the answers on using a file-supplied key. You do have to
411 write the glue-logic yourself though. Basically you can have
412 cryptsetup read the key from STDIN and write it there with your
413 own tool that in turn gets the key from the more secure key
417 * 2.12 Can I resize a dm-crypt or LUKS partition?
419 Yes, you can, as neither dm-crypt nor LUKS stores partition size.
420 Whether you should is a different question. Personally I recommend
421 backup, recreation of the encrypted partition with new size,
422 recreation of the filesystem and restore. This gets around the
423 tricky business of resizing the filesystem. Resizing a dm-crypt or
424 LUKS container does not resize the filesystem in it. The backup is
425 really non-optional here, as a lot can go wrong, resulting in
426 partial or complete data loss. Using something like gparted to
427 resize an encrypted partition is slow, but typically works. This
428 will not change the size of the filesystem hidden under the
431 You also need to be aware of size-based limitations. The one
432 currently relevant is that aes-xts-plain should not be used for
433 encrypted container sizes larger than 2TiB. Use aes-xts-plain64
440 * 3.1 My dm-crypt/LUKS mapping does not work! What general steps are
441 there to investigate the problem?
443 If you get a specific error message, investigate what it claims
444 first. If not, you may want to check the following things.
446 - Check that "/dev", including "/dev/mapper/control" is there. If it
447 is missing, you may have a problem with the "/dev" tree itself or
448 you may have broken udev rules.
450 - Check that you have the device mapper and the crypt target in your
451 kernel. The output of "dmsetup targets" should list a "crypt"
452 target. If it is not there or the command fails, add device mapper
453 and crypt-target to the kernel.
455 - Check that the hash-functions and ciphers you want to use are in
456 the kernel. The output of "cat /proc/crypto" needs to list them.
459 * 3.2 My dm-crypt mapping suddenly stopped when upgrading cryptsetup.
461 The default cipher, hash or mode may have changed (the mode changed
462 from 1.0.x to 1.1.x). See under "Issues With Specific Versions of
466 * 3.3 When I call cryptsetup from cron/CGI, I get errors about
469 If you get errors about unknown parameters or the like that are not
470 present when cryptsetup is called from the shell, make sure you
471 have no older version of cryptsetup on your system that then gets
472 called by cron/CGI. For example some distributions install
473 cryptsetup into /usr/sbin, while a manual install could go to
474 /usr/local/sbin. As a debugging aid, call "cryptsetup --version"
475 from cron/CGI or the non-shell mechanism to be sure the right
479 * 3.4 Unlocking a LUKS device takes very long. Why?
481 The iteration time for a key-slot (see Section 5 for an explanation
482 what iteration does) is calculated when setting a passphrase. By
483 default it is 1 second on the machine where the passphrase is set.
484 If you set a passphrase on a fast machine and then unlock it on a
485 slow machine, the unlocking time can be much longer. Also take into
486 account that up to 8 key-slots have to be tried in order to find the
489 If this is problem, you can add another key-slot using the slow
490 machine with the same passphrase and then remove the old key-slot.
491 The new key-slot will have an iteration count adjusted to 1 second
492 on the slow machine. Use luksKeyAdd and then luksKillSlot or
495 However, this operation will not change volume key iteration count
496 (MK iterations in output of "cryptsetup luksDump"). In order to
497 change that, you will have to backup the data in the LUKS
498 container (i.e. your encrypted data), luksFormat on the slow
499 machine and restore the data. Note that in the original LUKS
500 specification this value was fixed to 10, but it is now derived
501 from the PBKDF2 benchmark as well and set to iterations in 0.125
502 sec or 1000, whichever is larger. Also note that MK iterations
503 are not very security relevant. But as each key-slot already takes
504 1 second, spending the additional 0.125 seconds really does not
508 * 3.5 "blkid" sees a LUKS UUID and an ext2/swap UUID on the same
509 device. What is wrong?
511 Some old versions of cryptsetup have a bug where the header does
512 not get completely wiped during LUKS format and an older ext2/swap
513 signature remains on the device. This confuses blkid.
515 Fix: Wipe the unused header areas by doing a backup and restore of
516 the header with cryptsetup 1.1.x:
518 cryptsetup luksHeaderBackup --header-backup-file <file> <device>
519 cryptsetup luksHeaderRestore --header-backup-file <file> <device>
522 * 3.6 cryptsetup segfaults on Gentoo amd64 hardened ...
524 There seems to be some interference between the hardening and and
525 the way cryptsetup benchmarks PBKDF2. The solution to this is
526 currently not quite clear for an encrypted root filesystem. For
527 other uses, you can apparently specify USE="dynamic" as compile
528 flag, see http://bugs.gentoo.org/show_bug.cgi?id=283470
534 * 4.1 I get the error "LUKS keyslot x is invalid." What does that
537 This means that the given keyslot has an offset that points
538 outside the valid keyslot area. Typically, the reason is a
539 corrupted LUKS header because something was written to the start of
540 the device the LUKS container is on. Refer to Section "Backup and
541 Data Recovery" and ask on the mailing list if you have trouble
542 diagnosing and (if still possible) repairing this.
545 * 4.2 Can a bad RAM module cause problems?
547 LUKS and dm-crypt can give the RAM quite a workout, especially when
548 combined with software RAID. In particular the combination RAID5 +
549 LUKS + XFS seems to uncover RAM problems that never caused obvious
550 problems before. Symptoms vary, but often the problem manifest
551 itself when copying large amounts of data, typically several times
552 larger than your main memory.
554 Side note: One thing you should always do on large data
555 copy/movements is to run a verify, for example with the "-d"
556 option of "tar" or by doing a set of MD5 checksums on the source
559 find . -type f -exec md5sum \{\} \; > checksum-file
561 and then a "md5sum -c checksum-file" on the other side. If you get
562 mismatches here, RAM is the primary suspect. A lesser suspect is
563 an overclocked CPU. I have found countless hardware problems in
564 verify runs after copying or making backups. Bit errors are much
565 more common than most people think.
567 Some RAM issues are even worse and corrupt structures in one of the
568 layers. This typically results in lockups, CPU state dumps in the
569 system logs, kernel panic or other things. It is quite possible to
570 have the problem with an encrypted device, but not with an
571 otherwise the same unencrypted device. The reason for that is that
572 encryption has an error amplification property: You flip one bit
573 in an encrypted data block, and the decrypted version has half of
574 its bits flipped. This is an important security property for modern
575 ciphers. With the usual modes in cryptsetup (CBC, ESSIV, XTS), you
576 get up to a completely changed 512 byte block per bit error. A
577 corrupt block causes a lot more havoc than the occasionally
578 flipped single bit and can result in various obscure errors.
580 Note, that a verify run on copying between encrypted or
581 unencrypted devices will reliably detect corruption, even when the
582 copying itself did not report any problems. If you find defect
583 RAM, assume all backups and copied data to be suspect, unless you
587 * 4.3 How do I test RAM?
589 First you should know that overclocking often makes memory
590 problems worse. So if you overclock (which I strongly recommend
591 against in a system holding data that has some worth), run the
592 tests with the overclocking active.
594 There are two good options. One is Memtest86+ and the other is
595 "memtester" by Charles Cazabon. Memtest86+ requires a reboot and
596 then takes over the machine, while memtester runs from a
597 root-shell. Both use different testing methods and I have found
598 problems fast with each one that the other needed long to find. I
599 recommend running the following procedure until the first error is
602 - Run Memtest86+ for one cycle
604 - Run memtester for one cycle (shut down as many other applications
607 - Run Memtest86+ for 24h or more
609 - Run memtester for 24h or more
611 If all that does not produce error messages, your RAM may be sound,
612 but I have had one weak bit that Memtest86+ needed around 60 hours
613 to find. If you can reproduce the original problem reliably, a good
614 additional test may be to remove half of the RAM (if you have more
615 than one module) and try whether the problem is still there and if
616 so, try with the other half. If you just have one module, get a
617 different one and try with that. If you do overclocking, reduce
618 the settings to the most conservative ones available and try with
625 * 5.1 How long is a secure passphrase ?
627 This is just the short answer. For more info and explanation of
628 some of the terms used in this item, read the rest of Section 5.
629 The actual recommendation is at the end of this item.
631 First, passphrase length is not really the right measure,
632 passphrase entropy is. For example, a random lowercase letter (a-z)
633 gives you 4.7 bit of entropy, one element of a-z0-9 gives you 5.2
634 bits of entropy, an element of a-zA-Z0-9 gives you 5.9 bits and
635 a-zA-Z0-9!@#$%^&:-+ gives you 6.2 bits. On the other hand, a random
636 English word only gives you 0.6...1.3 bits of entropy per
637 character. Using sentences that make sense gives lower entropy,
638 series of random words gives higher entropy. Do not use sentences
639 that can be tied to you or found on your computer. This type of
640 attack is done routinely today.
642 That said, it does not matter too much what scheme you use, but it
643 does matter how much entropy your passphrase contains, because an
644 attacker has to try on average
646 1/2 * 2^(bits of entropy in passphrase)
648 different passphrases to guess correctly.
650 Historically, estimations tended to use computing time estimates,
651 but more modern approaches try to estimate cost of guessing a
654 As an example, I will try to get an estimate from the numbers in
655 http://it.slashdot.org/story/12/12/05/0623215/new-25-gpu-monster-devours-strong-passwords-in-minutes
656 More references can be found a the end of this document. Note that
657 these are estimates from the defender side, so assuming something
658 is easier than it actually is is fine. An attacker may still have
659 vastly higher cost than estimated here.
661 LUKS uses SHA1 for hasing per default. The claim in the reference is
662 63 billion tries/second for SHA1. We will leave aside the check
663 whether a try actually decrypts a key-slot. Now, the machine has 25
664 GPUs, which I will estimate at an overall lifetime cost of USD/EUR
665 1000 each, and an useful lifetime of 2 years. (This is on the low
666 side.) Disregarding downtime, the machine can then break
668 N = 63*10^9 * 3600 * 24 * 365 * 2 ~ 4*10^18
670 passphrases for EUR/USD 25k. That is one 62 bit passphrase hashed
671 once with SHA1 for EUR/USD 25k. Note that as this can be
672 parallelized, it can be done faster than 2 years with several of
675 For plain dm-crypt (no hash iteration) this is it. This gives (with
676 SHA1, plain dm-crypt default is ripemd160 which seems to be
677 slightly slower than SHA1):
679 Passphrase entropy Cost to break
688 For LUKS, you have to take into account hash iteration in PBKDF2.
689 For a current CPU, there are about 100k iterations (as can be
690 queried with ''cryptsetup luksDump''.
692 The table above then becomes:
694 Passphrase entropy Cost to break
705 To get reasonable security for the next 10 years, it is a good idea
706 to overestimate by a factor of at least 1000.
708 Then there is the question of how much the attacker is willing to
709 spend. That is up to your own security evaluation. For general use,
710 I will assume the attacker is willing to spend up to 1 million
711 EUR/USD. Then we get the following recommendations:
713 Plain dm-crypt: Use > 80 bit. That is e.g. 17 random chars from a-z
714 or a random English sentence of > 135 characters length.
716 LUKS: Use > 65 bit. That is e.g. 14 random chars from a-z or a
717 random English sentence of > 108 characters length.
719 If paranoid, add at least 20 bit. That is roughly four additional
720 characters for random passphrases and roughly 32 characters for a
721 random English sentence.
724 * 5.2 Is LUKS insecure? Everybody can see I have encrypted data!
726 In practice it does not really matter. In most civilized countries
727 you can just refuse to hand over the keys, no harm done. In some
728 countries they can force you to hand over the keys, if they suspect
729 encryption. However the suspicion is enough, they do not have to
730 prove anything. This is for practical reasons, as even the presence
731 of a header (like the LUKS header) is not enough to prove that you
732 have any keys. It might have been an experiment, for example. Or it
733 was used as encrypted swap with a key from /dev/random. So they
734 make you prove you do not have encrypted data. Of course that is
735 just as impossible as the other way round.
737 This means that if you have a large set of random-looking data,
738 they can already lock you up. Hidden containers (encryption hidden
739 within encryption), as possible with Truecrypt, do not help
740 either. They will just assume the hidden container is there and
741 unless you hand over the key, you will stay locked up. Don't have
742 a hidden container? Though luck. Anybody could claim that.
744 Still, if you are concerned about the LUKS header, use plain
745 dm-crypt with a good passphrase. See also Section 2, "What is the
746 difference between "plain" and LUKS format?"
749 * 5.3 Should I initialize (overwrite) a new LUKS/dm-crypt partition?
751 If you just create a filesystem on it, most of the old data will
752 still be there. If the old data is sensitive, you should overwrite
753 it before encrypting. In any case, not initializing will leave the
754 old data there until the specific sector gets written. That may
755 enable an attacker to determine how much and where on the
756 partition data was written. If you think this is a risk, you can
757 prevent this by overwriting the encrypted device (here assumed to
758 be named "e1") with zeros like this:
760 dd_rescue -w /dev/zero /dev/mapper/e1
762 or alternatively with one of the following more standard commands:
764 cat /dev/zero > /dev/mapper/e1
765 dd if=/dev/zero of=/dev/mapper/e1
768 * 5.4 How do I securely erase a LUKS (or other) partition?
770 For LUKS, if you are in a desperate hurry, overwrite the LUKS
771 header and key-slot area. This means overwriting the first
772 (keyslots x stripes x keysize) + offset bytes. For the default
773 parameters, this is the 1'052'672 bytes, i.e. 1MiB + 4096 of the
774 LUKS partition. For 512 bit key length (e.g. for aes-xts-plain with
775 512 bit key) this is 2MiB. (The different offset stems from
776 differences in the sector alignment of the key-slots.) If in doubt,
777 just be generous and overwrite the first 10MB or so, it will likely
778 still be fast enough. A single overwrite with zeros should be
779 enough. If you anticipate being in a desperate hurry, prepare the
780 command beforehand. Example with /dev/sde1 as the LUKS partition
781 and default parameters:
783 head -c 1052672 /dev/zero > /dev/sde1; sync
785 A LUKS header backup or full backup will still grant access to
786 most or all data, so make sure that an attacker does not have
787 access to backups or destroy them as well.
789 If you have time, overwrite the whole LUKS partition with a single
790 pass of zeros. This is enough for current HDDs. For SSDs or FLASH
791 (USB sticks) you may want to overwrite the whole drive several
792 times to be sure data is not retained by wear leveling. This is
793 possibly still insecure as SSD technology is not fully understood
794 in this regard. Still, due to the anti-forensic properties of the
795 LUKS key-slots, a single overwrite of an SSD or FLASH drive could
796 be enough. If in doubt, use physical destruction in addition. Here
797 is a link to some current research results on erasing SSDs and
799 http://www.usenix.org/events/fast11/tech/full_papers/Wei.pdf
801 Keep in mind to also erase all backups.
803 Example for a zero-overwrite erase of partition sde1 done with
806 dd_rescue -w /dev/zero /dev/sde1
809 * 5.5 How do I securely erase a backup of a LUKS partition or header?
811 That depends on the medium it is stored on. For HDD and SSD, use
812 overwrite with zeros. For an SSD or FLASH drive (USB stick), you
813 may want to overwrite the complete SSD several times and use
814 physical destruction in addition, see last item. For re-writable
815 CD/DVD, a single overwrite should also be enough, due to the
816 anti-forensic properties of the LUKS keyslots. For write-once
817 media, use physical destruction. For low security requirements,
818 just cut the CD/DVD into several parts. For high security needs,
819 shred or burn the medium. If your backup is on magnetic tape, I
820 advise physical destruction by shredding or burning, after
821 overwriting . The problem with magnetic tape is that it has a
822 higher dynamic range than HDDs and older data may well be
823 recoverable after overwrites. Also write-head alignment issues can
824 lead to data not actually being deleted at all during overwrites.
827 * 5.6 What about backup? Does it compromise security?
829 That depends. See item 6.7.
832 * 5.7 Why is all my data permanently gone if I overwrite the LUKS
835 Overwriting the LUKS header in part or in full is the most common
836 reason why access to LUKS containers is lost permanently.
837 Overwriting can be done in a number of fashions, like creating a
838 new filesystem on the raw LUKS partition, making the raw partition
839 part of a raid array and just writing to the raw partition.
841 The LUKS header contains a 256 bit "salt" value and without that no
842 decryption is possible. While the salt is not secret, it is
843 key-grade material and cannot be reconstructed. This is a
844 cryptographically strong "cannot". From observations on the
845 cryptsetup mailing-list, people typically go though the usual
846 stages of grief (Denial, Anger, Bargaining, Depression, Acceptance)
847 when this happens to them. Observed times vary between 1 day and 2
848 weeks to complete the cycle. Seeking help on the mailing-list is
849 fine. Even if we usually cannot help with getting back your data,
850 most people found the feedback comforting.
852 If your header does not contain an intact salt, best go directly
853 to the last stage ("Acceptance") and think about what to do now.
854 There is one exception that I know of: If your LUKS container is
855 still open, then it may be possible to extract the master key from
856 the running system. See Item "How do I recover the master key from
857 a mapped LUKS container?" in Section "Backup and Data Recovery".
860 * 5.8 What is a "salt"?
862 A salt is a random key-grade value added to the passphrase before
863 it is processed. It is not kept secret. The reason for using salts
864 is as follows: If an attacker wants to crack the password for a
865 single LUKS container, then every possible passphrase has to be
866 tried. Typically an attacker will not try every binary value, but
867 will try words and sentences from a dictionary.
869 If an attacker wants to attack several LUKS containers with the
870 same dictionary, then a different approach makes sense: Compute the
871 resulting slot-key for each dictionary element and store it on
872 disk. Then the test for each entry is just the slow unlocking with
873 the slot key (say 0.00001 sec) instead of calculating the slot-key
874 first (1 sec). For a single attack, this does not help. But if you
875 have more than one container to attack, this helps tremendously,
876 also because you can prepare your table before you even have the
877 container to attack! The calculation is also very simple to
878 parallelize. You could, for example, use the night-time unused CPU
879 power of your desktop PCs for this.
881 This is where the salt comes in. If the salt is combined with the
882 passphrase (in the simplest form, just appended to it), you
883 suddenly need a separate table for each salt value. With a
884 reasonably-sized salt value (256 bit, e.g.) this is quite
888 * 5.9 Is LUKS secure with a low-entropy (bad) passphrase?
890 Note: You should only use the 94 printable characters from 7 bit
891 ASCII code to prevent your passphrase from failing when the
892 character encoding changes, e.g. because of a system upgrade, see
893 also the note at the very start of this FAQ under "WARNINGS".
895 This needs a bit of theory. The quality of your passphrase is
896 directly related to its entropy (information theoretic, not
897 thermodynamic). The entropy says how many bits of "uncertainty" or
898 "randomness" are in you passphrase. In other words, that is how
899 difficult guessing the passphrase is.
901 Example: A random English sentence has about 1 bit of entropy per
902 character. A random lowercase (or uppercase) character has about
905 Now, if n is the number of bits of entropy in your passphrase and t
906 is the time it takes to process a passphrase in order to open the
907 LUKS container, then an attacker has to spend at maximum
909 attack_time_max = 2^n * t
911 time for a successful attack and on average half that. There is no
912 way getting around that relationship. However, there is one thing
913 that does help, namely increasing t, the time it takes to use a
914 passphrase, see next FAQ item.
916 Still, if you want good security, a high-entropy passphrase is the
917 only option. For example, a low-entropy passphrase can never be
918 considered secure against a TLA-level (Three Letter Agency level,
919 i.e. government-level) attacker, no matter what tricks are used in
920 the key-derivation function. Use at least 64 bits for secret stuff.
921 That is 64 characters of English text (but only if randomly chosen)
922 or a combination of 12 truly random letters and digits.
924 For passphrase generation, do not use lines from very well-known
925 texts (religious texts, Harry potter, etc.) as they are to easy to
926 guess. For example, the total Harry Potter has about 1'500'000
927 words (my estimation). Trying every 64 character sequence starting
928 and ending at a word boundary would take only something like 20
929 days on a single CPU and is entirely feasible. To put that into
930 perspective, using a number of Amazon EC2 High-CPU Extra Large
931 instances (each gives about 8 real cores), this test costs
932 currently about 50USD/EUR, but can be made to run arbitrarily fast.
934 On the other hand, choosing 1.5 lines from, say, the Wheel of Time
935 is in itself not more secure, but the book selection adds quite a
936 bit of entropy. (Now that I have mentioned it here, don't use tWoT
937 either!) If you add 2 or 3 typos or switch some words around, then
938 this is good passphrase material.
941 * 5.10 What is "iteration count" and why is decreasing it a bad idea?
943 Iteration count is the number of PBKDF2 iterations a passphrase is
944 put through before it is used to unlock a key-slot. Iterations are
945 done with the explicit purpose to increase the time that it takes
946 to unlock a key-slot. This provides some protection against use of
947 low-entropy passphrases.
949 The idea is that an attacker has to try all possible passphrases.
950 Even if the attacker knows the passphrase is low-entropy (see last
951 item), it is possible to make each individual try take longer. The
952 way to do this is to repeatedly hash the passphrase for a certain
953 time. The attacker then has to spend the same time (given the same
954 computing power) as the user per try. With LUKS, the default is 1
955 second of PBKDF2 hashing.
957 Example 1: Lets assume we have a really bad passphrase (e.g. a
958 girlfriends name) with 10 bits of entropy. With the same CPU, an
959 attacker would need to spend around 500 seconds on average to
960 break that passphrase. Without iteration, it would be more like
961 0.0001 seconds on a modern CPU.
963 Example 2: The user did a bit better and has 32 chars of English
964 text. That would be about 32 bits of entropy. With 1 second
965 iteration, that means an attacker on the same CPU needs around 136
966 years. That is pretty impressive for such a weak passphrase.
967 Without the iterations, it would be more like 50 days on a modern
968 CPU, and possibly far less.
970 In addition, the attacker can both parallelize and use special
971 hardware like GPUs or FPGAs to speed up the attack. The attack can
972 also happen quite some time after the luksFormat operation and CPUs
973 can have become faster and cheaper. For that reason you want a
974 bit of extra security. Anyways, in Example 1 your are screwed.
975 In example 2, not necessarily. Even if the attack is faster, it
976 still has a certain cost associated with it, say 10000 EUR/USD
977 with iteration and 1 EUR/USD without iteration. The first can be
978 prohibitively expensive, while the second is something you try
979 even without solid proof that the decryption will yield something
982 The numbers above are mostly made up, but show the idea. Of course
983 the best thing is to have a high-entropy passphrase.
985 Would a 100 sec iteration time be even better? Yes and no.
986 Cryptographically it would be a lot better, namely 100 times better.
987 However, usability is a very important factor for security
988 technology and one that gets overlooked surprisingly often. For
989 LUKS, if you have to wait 2 minutes to unlock the LUKS container,
990 most people will not bother and use less secure storage instead. It
991 is better to have less protection against low-entropy passphrases
992 and people actually use LUKS, than having them do without
993 encryption altogether.
995 Now, what about decreasing the iteration time? This is generally a
996 very bad idea, unless you know and can enforce that the users only
997 use high-entropy passphrases. If you decrease the iteration time
998 without ensuring that, then you put your users at increased risk,
999 and considering how rarely LUKS containers are unlocked in a
1000 typical work-flow, you do so without a good reason. Don't do it.
1001 The iteration time is already low enough that users with entropy
1002 low passphrases are vulnerable. Lowering it even further increases
1003 this danger significantly.
1006 * 5.11 Some people say PBKDF2 is insecure?
1008 There is some discussion that a hash-function should have a "large
1009 memory" property, i.e. that it should require a lot of memory to be
1010 computed. This serves to prevent attacks using special programmable
1011 circuits, like FPGAs, and attacks using graphics cards. PBKDF2
1012 does not need a lot of memory and is vulnerable to these attacks.
1013 However, the publication usually referred in these discussions is
1014 not very convincing in proving that the presented hash really is
1015 "large memory" (that may change, email the FAQ maintainer when it
1016 does) and it is of limited usefulness anyways. Attackers that use
1017 clusters of normal PCs will not be affected at all by a "large
1018 memory" property. For example the US Secret Service is known to
1019 use the off-hour time of all the office PCs of the Treasury for
1020 password breaking. The Treasury has about 110'000 employees.
1021 Assuming every one has an office PC, that is significant computing
1022 power, all of it with plenty of memory for computing "large
1023 memory" hashes. Bot-net operators also have all the memory they
1024 want. The only protection against a resourceful attacker is a
1025 high-entropy passphrase, see items 5.8 and 5.9.
1028 * 5.12 What about iteration count with plain dm-crypt?
1030 Simple: There is none. There is also no salting. If you use plain
1031 dm-crypt, the only way to be secure is to use a high entropy
1032 passphrase. If in doubt, use LUKS instead.
1035 * 5.13 Is LUKS with default parameters less secure on a slow CPU?
1037 Unfortunately, yes. However the only aspect affected is the
1038 protection for low-entropy passphrase or master-key. All other
1039 security aspects are independent of CPU speed.
1041 The master key is less critical, as you really have to work at it
1042 to give it low entropy. One possibility is to supply the master key
1043 yourself. If that key is low-entropy, then you get what you
1044 deserve. The other known possibility is to use /dev/urandom for
1045 key generation in an entropy-starved situation (e.g. automatic
1046 installation on an embedded device without network and other entropy
1049 For the passphrase, don't use a low-entropy passphrase. If your
1050 passphrase is good, then a slow CPU will not matter. If you insist
1051 on a low-entropy passphrase on a slow CPU, use something like
1052 "--iter-time=10" or higher and wait a long time on each LUKS unlock
1053 and pray that the attacker does not find out in which way exactly
1054 your passphrase is low entropy. This also applies to low-entropy
1055 passphrases on fast CPUs. Technology can do only so much to
1056 compensate for problems in front of the keyboard.
1059 * 5.14 Why was the default aes-cbc-plain replaced with aes-cbc-essiv?
1061 Note: This item applies both to plain dm-crypt and to LUKS
1063 The problem is that cbc-plain has a fingerprint vulnerability, where
1064 a specially crafted file placed into the crypto-container can be
1065 recognized from the outside. The issue here is that for cbc-plain
1066 the initialization vector (IV) is the sector number. The IV gets
1067 XORed to the first data chunk of the sector to be encrypted. If you
1068 make sure that the first data block to be stored in a sector
1069 contains the sector number as well, the first data block to be
1070 encrypted is all zeros and always encrypted to the same ciphertext.
1071 This also works if the first data chunk just has a constant XOR
1072 with the sector number. By having several shifted patterns you can
1073 take care of the case of a non-power-of-two start sector number of
1076 This mechanism allows you to create a pattern of sectors that have
1077 the same first ciphertext block and signal one bit per sector to the
1078 outside, allowing you to e.g. mark media files that way for
1079 recognition without decryption. For large files this is a
1080 practical attack. For small ones, you do not have enough blocks to
1081 signal and take care of different file starting offsets.
1083 In order to prevent this attack, the default was changed to
1084 cbc-essiv. ESSIV uses a keyed hash of the sector number, with the
1085 encryption key as key. This makes the IV unpredictable without
1086 knowing the encryption key and the watermarking attack fails.
1089 * 5.15 Are there any problems with "plain" IV? What is "plain64"?
1091 First, "plain" and "plain64" are both not secure to use with CBC,
1092 see previous FAQ item.
1094 However there are modes, like XTS, that are secure with "plain" IV.
1095 The next limit is that "plain" is 64 bit, with the upper 32 bit set
1096 to zero. This means that on volumes larger than 2TiB, the IV
1097 repeats, creating a vulnerability that potentially leaks some
1098 data. To avoid this, use "plain64", which uses the full sector
1099 number up to 64 bit. Note that "plain64" requires a kernel >=
1100 2.6.33. Also note that "plain64" is backwards compatible for
1101 volume sizes <= 2TiB, but not for those > 2TiB. Finally, "plain64"
1102 does not cause any performance penalty compared to "plain".
1105 * 5.16 What about XTS mode?
1107 XTS mode is potentially even more secure than cbc-essiv (but only if
1108 cbc-essiv is insecure in your scenario). It is a NIST standard and
1109 used, e.g. in Truecrypt. At the moment, if you want to use it, you
1110 have to specify it manually as "aes-xts-plain", i.e.
1112 cryptsetup -c aes-xts-plain luksFormat <device>
1114 For volumes >2TiB and kernels >= 2.6.33 use "plain64" (see FAQ
1115 item on "plain" and "plain64"):
1117 cryptsetup -c aes-xts-plain64 luksFormat <device>
1119 There is a potential security issue with XTS mode and large blocks.
1120 LUKS and dm-crypt always use 512B blocks and the issue does not
1124 * 5.17 Is LUKS FIPS-140-2 certified?
1126 No. But that is more a problem of FIPS-140-2 than of LUKS. From a
1127 technical point-of-view, LUKS with the right parameters would be
1128 FIPS-140-2 compliant, but in order to make it certified, somebody
1129 has to pay real money for that. And then, whenever cryptsetup is
1130 changed or extended, the certification lapses and has to be
1133 From the aspect of actual security, LUKS with default parameters
1134 should be as good as most things that are FIPS-140-2 certified,
1135 although you may want to make sure to use /dev/random (by
1136 specifying --use-random on luksFormat) as randomness source for
1137 the master key to avoid being potentially insecure in an
1138 entropy-starved situation.
1141 * 5.18 What about Plausible Deniability?
1143 First let me attempt a definition for the case of encrypted
1144 filesystems: Plausible deniability is when you hide encrypted data
1145 inside an encrypted container and it is not possible to prove it is
1146 there. The idea is compelling and on first glance it seems
1147 possible to do it. And from a cryptographic point of view, it
1148 actually is possible.
1150 So, does it work in practice? No, unfortunately. The reasoning used
1151 by its proponents is fundamentally flawed in several ways and the
1152 cryptographic properties fail fatally when colliding with the real
1155 First, why should "I do not have a hidden partition" be any more
1156 plausible than "I forgot my crypto key" or "I wiped that partition
1157 with random data, nothing in there"? I do not see any reason.
1159 Second, there are two types of situations: Either they cannot force
1160 you to give them the key (then you simply do not) or the can. In
1161 the second case, they can always do bad things to you, because they
1162 cannot prove that you have the key in the first place! This means
1163 they do not have to prove you have the key, or that this random
1164 looking data on your disk is actually encrypted data. So the
1165 situation will allow them to waterboard/lock-up/deport you
1166 anyways, regardless of how "plausible" your deniability is. Do not
1167 have a hidden partition you could show to them, but there are
1168 indications you may? Too bad for you. Unfortunately "plausible
1169 deniability" also means you cannot prove there is no hidden data.
1171 Third, hidden partitions are not that hidden. There are basically
1172 just two possibilities: a) Make a large crypto container, but put a
1173 smaller filesystem in there and put the hidden partition into the
1174 free space. Unfortunately this is glaringly obvious and can be
1175 detected in an automated fashion. This means that the initial
1176 suspicion to put you under duress in order to make you reveal you
1177 hidden data is given. b) Make a filesystem that spans the whole
1178 encrypted partition, and put the hidden partition into space not
1179 currently used by that filesystem. Unfortunately that is also
1180 glaringly obvious, as you then cannot write to the filesystem
1181 without a high risk of destroying data in the hidden container.
1182 Have not written anything to the encrypted filesystem in a while?
1183 Too bad, they have the suspicion they need to do unpleasant things
1186 To be fair, if you prepare option b) carefully and directly before
1187 going into danger, it may work. But then, the mere presence of
1188 encrypted data may already be enough to get you into trouble in
1189 those places were they can demand encryption keys.
1191 Here is an additional reference for some problems with plausible
1192 deniability: http://www.schneier.com/paper-truecrypt-dfs.pdf I
1193 strongly suggest you read it.
1195 So, no, I will not provide any instructions on how to do it with
1196 plain dm-crypt or LUKS. If you insist on shooting yourself in the
1197 foot, you can figure out how to do it yourself.
1200 * 5.19 What about SSDs or Flash Drives?
1202 The problem is that you cannot reliably erase parts of these
1203 devices, mainly due to wear-leveling and possibly defect
1206 Basically, when overwriting a sector (of 512B), what the device
1207 does is to move an internal sector (may be 128kB or even larger) to
1208 some pool of discarded, not-yet erased unused sectors, take a
1209 fresh empty sector from the empty-sector pool and copy the old
1210 sector over with the changes to the small part you wrote. This is
1211 done in some fashion so that larger writes do not cause a lot of
1212 small internal updates.
1214 The thing is that the mappings between outside-adressable sectors
1215 and inside sectors is arbitrary (and the vendors are not talking).
1216 Also the discarded sectors are not necessarily erased immediately.
1217 They may linger a long time.
1219 For plain dm-crypt, the consequences are that older encrypted data
1220 may be lying around in some internal pools of the device. Thus may
1221 or may not be a problem and depends on the application. Remember
1222 the same can happen with a filesystem if consecutive writes to the
1223 same area of a file can go to different sectors.
1225 However, for LUKS, the worst case is that key-slots and LUKS
1226 header may end up in these internal pools. This means that password
1227 management functionality is compromised (the old passwords may
1228 still be around, potentially for a very long time) and that fast
1229 erase by overwriting the header and key-slot area is insecure.
1231 Also keep in mind that the discarded/used pool may be large. For
1232 example, a 240GB SSD has about 16GB of spare area in the chips that
1233 it is free to do with as it likes. You would need to make each
1234 individual key-slot larger than that to allow reliable overwriting.
1235 And that assumes the disk thinks all other space is in use.
1236 Reading the internal pools using forensic tools is not that hard,
1237 but may involve some soldering.
1241 If you trust the device vendor (you probably should not...) you can
1242 try an ATA "secure erase" command for SSDs. That does not work for
1243 USB keys though. And if it finishes after a few seconds, it was
1244 possibly faked by the SSD.
1246 If you can do without password management and are fine with doing
1247 physical destruction for permenently deleting data (allways after
1248 one or several full overwrites!), you can use plain dm-crypt or
1251 If you want or need the original LUKS security features to work,
1252 you can use a detached LUKS header and put that on a conventional,
1253 magnetic disk. That leaves potentially old encrypted data in the
1254 pools on the disk, but otherwise you get LUKS with the same
1255 security as on a magnetic disk.
1257 If you are concerned about your laptop being stolen, you are likely
1258 fine using LUKS on an SSD. An attacker would need to have access
1259 to an old passphrase (and the key-slot for this old passphrase
1260 would actually need to still be somewhere in the SSD) for your
1261 data to be at risk. So unless you pasted your old passphrase all
1262 over the Internet or the attacker has knowledge of it from some
1263 other source and does a targetted laptop theft to get at your
1264 data, you should be fine.
1267 6. Backup and Data Recovery
1270 * 6.1 Why do I need Backup?
1272 First, disks die. The rate for well-treated (!) disk is about 5%
1273 per year, which is high enough to worry about. There is some
1274 indication that this may be even worse for some SSDs. This applies
1275 both to LUKS and plain dm-crypt partitions.
1277 Second, for LUKS, if anything damages the LUKS header or the
1278 key-stripe area then decrypting the LUKS device can become
1279 impossible. This is a frequent occurrence. For example an
1280 accidental format as FAT or some software overwriting the first
1281 sector where it suspects a partition boot sector typically makes a
1282 LUKS partition permanently inaccessible. See more below on LUKS
1285 So, data-backup in some form is non-optional. For LUKS, you may
1286 also want to store a header backup in some secure location. This
1287 only needs an update if you change passphrases.
1290 * 6.2 How do I backup a LUKS header?
1292 While you could just copy the appropriate number of bytes from the
1293 start of the LUKS partition, the best way is to use command option
1294 "luksHeaderBackup" of cryptsetup. This protects also against
1295 errors when non-standard parameters have been used in LUKS
1296 partition creation. Example:
1299 cryptsetup luksHeaderBackup --header-backup-file <file> <device>
1301 To restore, use the inverse command, i.e.
1303 cryptsetup luksHeaderRestore --header-backup-file <file> <device>
1306 * 6.3 How do I test a LUKS header?
1310 cryptsetup -v isLuks <device>
1312 on the device. Without the "-v" it just signals its result via
1313 exit-status. You can also use the more general test
1317 which will also detect other types and give some more info. Omit
1318 "-p" for old versions of blkid that do not support it.
1321 * 6.4 How do I backup a LUKS or dm-crypt partition?
1323 There are two options, a sector-image and a plain file or
1324 filesystem backup of the contents of the partition. The sector
1325 image is already encrypted, but cannot be compressed and contains
1326 all empty space. The filesystem backup can be compressed, can
1327 contain only part of the encrypted device, but needs to be
1328 encrypted separately if so desired.
1330 A sector-image will contain the whole partition in encrypted form,
1331 for LUKS the LUKS header, the keys-slots and the data area. It can
1332 be done under Linux e.g. with dd_rescue (for a direct image copy)
1333 and with "cat" or "dd". Example:
1335 cat /dev/sda10 > sda10.img
1336 dd_rescue /dev/sda10 sda10.img
1338 You can also use any other backup software that is capable of making
1339 a sector image of a partition. Note that compression is
1340 ineffective for encrypted data, hence it does not make sense to
1343 For a filesystem backup, you decrypt and mount the encrypted
1344 partition and back it up as you would a normal filesystem. In this
1345 case the backup is not encrypted, unless your encryption method
1346 does that. For example you can encrypt a backup with "tar" as
1349 tar cjf - <path> | gpg --cipher-algo AES -c - > backup.tbz2.gpg
1351 And verify the backup like this if you are at "path":
1353 cat backup.tbz2.gpg | gpg - | tar djf -
1355 Note: Always verify backups, especially encrypted ones.
1357 In both cases GnuPG will ask you interactively for your symmetric
1358 key. The verify will only output errors. Use "tar dvjf -" to get
1359 all comparison results. To make sure no data is written to disk
1360 unencrypted, turn off swap if it is not encrypted before doing the
1363 You can of course use different or no compression and you can use
1364 an asymmetric key if you have one and have a backup of the secret
1365 key that belongs to it.
1367 A second option for a filesystem-level backup that can be used when
1368 the backup is also on local disk (e.g. an external USB drive) is
1369 to use a LUKS container there and copy the files to be backed up
1370 between both mounted containers. Also see next item.
1373 * 6.5 Do I need a backup of the full partition? Would the header and
1374 key-slots not be enough?
1376 Backup protects you against two things: Disk loss or corruption
1377 and user error. By far the most questions on the dm-crypt mailing
1378 list about how to recover a damaged LUKS partition are related
1379 to user error. For example, if you create a new filesystem on a
1380 LUKS partition, chances are good that all data is lost
1383 For this case, a header+key-slot backup would often be enough. But
1384 keep in mind that a well-treated (!) HDD has roughly a failure
1385 risk of 5% per year. It is highly advisable to have a complete
1386 backup to protect against this case.
1389 * *6.6 What do I need to backup if I use "decrypt_derived"?
1391 This is a script in Debian, intended for mounting /tmp or swap with
1392 a key derived from the master key of an already decrypted device.
1393 If you use this for an device with data that should be persistent,
1394 you need to make sure you either do not lose access to that master
1395 key or have a backup of the data. If you derive from a LUKS
1396 device, a header backup of that device would cover backing up the
1397 master key. Keep in mind that this does not protect against disk
1400 Note: If you recreate the LUKS header of the device you derive from
1401 (using luksFormat), the master key changes even if you use the same
1402 passphrase(s) and you will not be able to decrypt the derived
1403 device with the new LUKS header.
1406 * 6.7 Does a backup compromise security?
1408 Depends on how you do it. However if you do not have one, you are
1409 going to eventually lose your encrypted data.
1411 There are risks introduced by backups. For example if you
1412 change/disable a key-slot in LUKS, a binary backup of the partition
1413 will still have the old key-slot. To deal with this, you have to
1414 be able to change the key-slot on the backup as well, securely
1415 erase the backup or do a filesystem-level backup instead of a binary
1418 If you use dm-crypt, backup is simpler: As there is no key
1419 management, the main risk is that you cannot wipe the backup when
1420 wiping the original. However wiping the original for dm-crypt
1421 should consist of forgetting the passphrase and that you can do
1422 without actual access to the backup.
1424 In both cases, there is an additional (usually small) risk with
1425 binary backups: An attacker can see how many sectors and which
1426 ones have been changed since the backup. To prevent this, use a
1427 filesystem level backup method that encrypts the whole backup in
1428 one go, e.g. as described above with tar and GnuPG.
1430 My personal advice is to use one USB disk (low value data) or
1431 three disks (high value data) in rotating order for backups, and
1432 either use independent LUKS partitions on them, or use encrypted
1433 backup with tar and GnuPG.
1435 If you do network-backup or tape-backup, I strongly recommend to
1436 go the filesystem backup path with independent encryption, as you
1437 typically cannot reliably delete data in these scenarios,
1438 especially in a cloud setting. (Well, you can burn the tape if it
1439 is under your control...)
1442 * 6.8 What happens if I overwrite the start of a LUKS partition or
1443 damage the LUKS header or key-slots?
1445 There are two critical components for decryption: The salt values
1446 in the header itself and the key-slots. If the salt values are
1447 overwritten or changed, nothing (in the cryptographically strong
1448 sense) can be done to access the data, unless there is a backup
1449 of the LUKS header. If a key-slot is damaged, the data can still
1450 be read with a different key-slot, if there is a remaining
1451 undamaged and used key-slot. Note that in order to make a key-slot
1452 unrecoverable in a cryptographically strong sense, changing about
1453 4-6 bits in random locations of its 128kiB size is quite enough.
1456 * 6.9 What happens if I (quick) format a LUKS partition?
1458 I have not tried the different ways to do this, but very likely you
1459 will have written a new boot-sector, which in turn overwrites the
1460 LUKS header, including the salts, making your data permanently
1461 irretrievable, unless you have a LUKS header backup. You may also
1462 damage the key-slots in part or in full. See also last item.
1465 * 6.10 How do I recover the master key from a mapped LUKS container?
1467 This is typically only needed if you managed to damage your LUKS
1468 header, but the container is still mapped, i.e. "luksOpen"ed. It
1469 also helps if you have a mapped container that you forgot or do not
1470 know a passphrase for (e.g. on a long running server.)
1472 WARNING: Things go wrong, do a full backup before trying this!
1474 WARNING: This exposes the master key of the LUKS container. Note
1475 that both ways to recreate a LUKS header with the old master key
1476 described below will write the master key to disk. Unless you are
1477 sure you have securely erased it afterwards, e.g. by writing it to
1478 an encrypted partition, RAM disk or by erasing the filesystem you
1479 wrote it to by a complete overwrite, you should change the master
1480 key afterwards. Changing the master key requires a full data
1481 backup, luksFormat and then restore of the backup.
1483 First, there is a script by Milan that automates the whole
1484 process, except generating a new LUKS header with the old master
1485 key (it prints the command for that though):
1487 http://code.google.com/p/cryptsetup/source/browse/misc/luks-header-from-active
1489 You can also do this manually. Here is how:
1491 - Get the master key from the device mapper. This is done by the
1492 following command. Substitute c5 for whatever you mapped to:
1494 # dmsetup table --target crypt --showkey /dev/mapper/c5
1496 0 200704 crypt aes-cbc-essiv:sha256
1497 a1704d9715f73a1bb4db581dcacadaf405e700d591e93e2eaade13ba653d0d09
1500 The result is actually one line, wrapped here for clarity. The long
1501 hex string is the master key.
1503 - Convert the master key to a binary file representation. You can
1504 do this manually, e.g. with hexedit. You can also use the tool
1505 "xxd" from vim like this:
1507 echo "a1704d9....53d0d09" | xxd -r -p > <master-key-file>
1509 - Do a luksFormat to create a new LUKS header.
1511 NOTE: If your header is intact and you just forgot the
1512 passphrase, you can just set a new passphrase, see next
1515 Unmap the device before you do that (luksClose). Then do
1517 cryptsetup luksFormat --master-key-file=<master-key-file> <luks device>
1519 Note that if the container was created with other than the default
1520 settings of the cryptsetup version you are using, you need to give
1521 additional parameters specifying the deviations. If in doubt, try
1522 the script by Milan. It does recover the other parameters as well.
1524 Side note: This is the way the decrypt_derived script gets at the
1525 master key. It just omits the conversion and hashes the master key
1528 - If the header is intact and you just forgot the passphrase, just
1529 set a new passphrase like this:
1531 cryptsetup luksAddKey --master-key-file=<master-key-file> <luks device>
1533 You may want to disable the old one afterwards.
1536 * 6.11 What does the on-disk structure of dm-crypt look like?
1538 There is none. dm-crypt takes a block device and gives encrypted
1539 access to each of its blocks with a key derived from the passphrase
1540 given. If you use a cipher different than the default, you have to
1541 specify that as a parameter to cryptsetup too. If you want to
1542 change the password, you basically have to create a second
1543 encrypted device with the new passphrase and copy your data over.
1544 On the plus side, if you accidentally overwrite any part of a
1545 dm-crypt device, the damage will be limited to the are you
1549 * 6.12 What does the on-disk structure of LUKS look like?
1551 A LUKS partition consists of a header, followed by 8 key-slot
1552 descriptors, followed by 8 key slots, followed by the encrypted
1555 Header and key-slot descriptors fill the first 592 bytes. The
1556 key-slot size depends on the creation parameters, namely on the
1557 number of anti-forensic stripes, key material offset and master
1560 With the default parameters, each key-slot is a bit less than
1561 128kiB in size. Due to sector alignment of the key-slot start,
1562 that means the key block 0 is at offset 0x1000-0x20400, key
1563 block 1 at offset 0x21000-0x40400, and key block 7 at offset
1564 0xc1000-0xe0400. The space to the next full sector address is
1565 padded with zeros. Never used key-slots are filled with what the
1566 disk originally contained there, a key-slot removed with
1567 "luksRemoveKey" or "luksKillSlot" gets filled with 0xff. Due to
1568 2MiB default alignment, start of the data area for cryptsetup 1.3
1569 and later is at 2MiB, i.e. at 0x200000. For older versions, it is
1570 at 0x101000, i.e. at 1'052'672 bytes, i.e. at 1MiB + 4096 bytes
1571 from the start of the partition. Incidentally, "luksHeaderBackup"
1572 for a LUKS container created with default parameters dumps exactly
1573 the first 2MiB (or 1'052'672 bytes for headers created with
1574 cryptsetup versions < 1.3) to file and "luksHeaderRestore" restores
1577 For non-default parameters, you have to figure out placement
1578 yourself. "luksDump" helps. See also next item. For the most common
1579 non-default settings, namely aes-xts-plain with 512 bit key, the
1580 offsets are: 1st keyslot 0x1000-0x3f800, 2nd keyslot
1581 0x40000-0x7e000, 3rd keyslot 0x7e000-0xbd800, ..., and start of
1582 bulk data at 0x200000.
1584 The exact specification of the format is here:
1585 http://code.google.com/p/cryptsetup/wiki/Specification
1588 * 6.13 What is the smallest possible LUKS container?
1590 Note: From cryptsetup 1.3 onwards, alignment is set to 1MB. With
1591 modern Linux partitioning tools that also align to 1MB, this will
1592 result in alignment to 2k sectors and typical Flash/SSD sectors,
1593 which is highly desirable for a number of reasons. Changing the
1594 alignment is not recommended.
1596 That said, with default parameters, the data area starts at
1597 exactly 2MB offset (at 0x101000 for cryptsetup versions before
1598 1.3). The smallest data area you can have is one sector of 512
1599 bytes. Data areas of 0 bytes can be created, but fail on mapping.
1601 While you cannot put a filesystem into something this small, it may
1602 still be used to contain, for example, key. Note that with current
1603 formatting tools, a partition for a container this size will be
1604 3MiB anyways. If you put the LUKS container into a file (via
1605 losetup and a loopback device), the file needs to be 2097664 bytes
1606 in size, i.e. 2MiB + 512B.
1608 There two ways to influence the start of the data area are key-size
1611 For alignment, you can go down to 1 on the parameter. This will
1612 still leave you with a data-area starting at 0x101000, i.e.
1613 1MiB+4096B (default parameters) as alignment will be rounded up to
1614 the next multiple of 8 (i.e. 4096 bytes) If in doubt, do a dry-run
1615 on a larger file and dump the LUKS header to get actual
1618 For key-size, you can use 128 bit (e.g. AES-128 with CBC), 256 bit
1619 (e.g. AES-256 with CBC) or 512 bit (e.g. AES-256 with XTS mode).
1620 You can do 64 bit (e.g. blowfish-64 with CBC), but anything below
1621 128 bit has to be considered insecure today.
1623 Example 1 - AES 128 bit with CBC:
1625 cryptsetup luksFormat -s 128 --align-payload=8 <device>
1627 This results in a data offset of 0x81000, i.e. 516KiB or 528384
1628 bytes. Add one 512 byte sector and the smallest LUKS container size
1629 with these parameters is 516KiB + 512B or 528896 bytes.
1631 Example 2 - Blowfish 64 bit with CBC (WARNING: insecure):
1633 cryptsetup luksFormat -c blowfish -s 64 --align-payload=8 /dev/loop0
1635 This results in a data offset of 0x41000, i.e. 260kiB or 266240
1636 bytes, with a minimal LUKS container size of 260kiB + 512B or
1640 * 6.14 I think this is overly complicated. Is there an alternative?
1642 Not really. Encryption comes at a price. You can use plain
1643 dm-crypt to simplify things a bit. It does not allow multiple
1644 passphrases, but on the plus side, it has zero on disk description
1645 and if you overwrite some part of a plain dm-crypt partition,
1646 exactly the overwritten parts are lost (rounded up to sector
1650 * 6.15 Can I clone a LUKS container?
1652 You can, but it breaks security, because the cloned container has
1653 the same header and hence the same master key. You cannot change
1654 the master key on a LUKS container, even if you change the
1655 passphrase(s), the master key stays the same. That means whoever
1656 has access to one of the clones can decrypt them all, completely
1657 bypassing the passphrases.
1659 The right way to do this is to first luksFormat the target
1660 container, then to clone the contents of the source container, with
1661 both containers mapped, i.e. decrypted. You can clone the decrypted
1662 contents of a LUKS container in binary mode, although you may run
1663 into secondary issues with GUIDs in filesystems, partition tables,
1664 RAID-components and the like. These are just the normal problems
1665 binary cloning causes.
1667 Note that if you need to ship (e.g.) cloned LUKS containers with a
1668 default passphrase, that is fine as long as each container was
1669 individually created (and hence has its own master key). In this
1670 case, changing the default passphrase will make it secure again.
1673 7. Interoperability with other Disk Encryption Tools
1676 * 7.1 What is this section about?
1678 Cryptsetup for plain dm-crypt can be used to access a number of
1679 on-disk formats created by tools like loop-aes patched into
1680 losetup. This sometimes works and sometimes does not. This
1681 section collects insights into what works, what does not and where
1682 more information is required.
1684 Additional information may be found in the mailing-list archives,
1685 mentioned at the start of this FAQ document. If you have a
1686 solution working that is not yet documented here and think a wider
1687 audience may be interested, please email the FAQ maintainer.
1690 * 7.2 loop-aes: General observations.
1692 One problem is that there are different versions of losetup around.
1693 loop-aes is a patch for losetup. Possible problems and deviations
1694 from cryptsetup option syntax include:
1696 - Offsets specified in bytes (cryptsetup: 512 byte sectors)
1698 - The need to specify an IV offset
1700 - Encryption mode needs specifying (e.g. "-c twofish-cbc-plain")
1702 - Key size needs specifying (e.g. "-s 128" for 128 bit keys)
1704 - Passphrase hash algorithm needs specifying
1706 Also note that because plain dm-crypt and loop-aes format does not
1707 have metadata, and while the loopAES extension for cryptsetup tries
1708 autodetection (see command loopaesOpen), it may not always work.
1709 If you still have the old set-up, using a verbosity option (-v)
1710 on mapping with the old tool or having a look into the system logs
1711 after setup could give you the information you need. Below, there
1712 are also some things that worked for somebody.
1715 * 7.3 loop-aes patched into losetup on Debian 5.x, kernel 2.6.32
1717 In this case, the main problem seems to be that this variant of
1718 losetup takes the offset (-o option) in bytes, while cryptsetup
1719 takes it in sectors of 512 bytes each. Example: The losetup command
1721 losetup -e twofish -o 2560 /dev/loop0 /dev/sdb1
1722 mount /dev/loop0 mount-point
1726 cryptsetup create -c twofish -o 5 --skip 5 e1 /dev/sdb1
1727 mount /dev/mapper/e1 mount-point
1730 * 7.4 loop-aes with 160 bit key
1732 This seems to be sometimes used with twofish and blowfish and
1733 represents a 160 bit ripemed160 hash output padded to 196 bit key
1734 length. It seems the corresponding options for cryptsetup are
1736 --cipher twofish-cbc-null -s 192 -h ripemd160:20
1739 * 7.5 loop-aes v1 format OpenSUSE
1741 Apparently this is done by older OpenSUSE distros and stopped
1742 working from OpenSUSE 12.1 to 12.2. One user had success with the
1745 cryptsetup create <target> <device> -c aes -s 128 -h sha256
1748 * 7.6 Kernel encrypted loop device (cryptoloop)
1750 There are a number of different losetup implementations for using
1751 encrypted loop devices so getting this to work may need a bit of
1754 NOTE: Do NOT use this for new containers! Some of the existing
1755 implementations are insecure and future support is uncertain.
1757 Example for a compatible mapping:
1759 losetup -e twofish -N /dev/loop0 /image.img
1763 cryptsetup create image_plain /image.img -c twofish-cbc-plain -H plain
1765 with the mapping being done to /dev/mapper/image_plain instead of
1770 Cipher, mode and pasword hash (or no hash):
1772 -e cipher [-N] => -c cipher-cbc-plain -H plain [-s 256]
1773 -e cipher => -c cipher-cbc-plain -H ripemd160 [-s 256]
1775 Key size and offsets (losetup: bytes, cryptsetuop: sectors of 512
1779 -o 2560 => -o 5 -p 5 # 2560/512 = 5
1781 There is no replacement for --pass-fd, it has to be emulated using
1782 keyfiles, see the cryptsetup man-page.
1785 8. Issues with Specific Versions of cryptsetup
1788 * 8.1 When using the create command for plain dm-crypt with
1789 cryptsetup 1.1.x, the mapping is incompatible and my data is not
1792 With cryptsetup 1.1.x, the distro maintainer can define different
1793 default encryption modes for LUKS and plain devices. You can check
1794 these compiled-in defaults using "cryptsetup --help". Moreover, the
1795 plain device default changed because the old IV mode was
1796 vulnerable to a watermarking attack.
1798 If you are using a plain device and you need a compatible mode, just
1799 specify cipher, key size and hash algorithm explicitly. For
1800 compatibility with cryptsetup 1.0.x defaults, simple use the
1803 cryptsetup create -c aes-cbc-plain -s 256 -h ripemd160 <name> <dev>
1805 LUKS stores cipher and mode in the metadata on disk, avoiding this
1809 * 8.2 cryptsetup on SLED 10 has problems...
1811 SLED 10 is missing an essential kernel patch for dm-crypt, which
1812 is broken in its kernel as a result. There may be a very old
1813 version of cryptsetup (1.0.x) provided by SLED, which should also
1814 not be used anymore as well. My advice would be to drop SLED 10.
1817 9. References and Further Reading
1820 * Purpose of this Section
1822 The purpose of this section is to collect references to all
1823 materials that do not fit the FAQ but are relevant in some fashion.
1824 This can be core topics like the LUKS spec or disk encryption, but
1825 it can also be more tangential, like secure storage management or
1826 cryptography used in LUKS. It should still have relevance to
1827 cryptsetup and its applications.
1829 If you wan to see something added here, send email to the
1830 maintainer (or the cryptsetup mailing list) giving an URL, a
1831 description (1-3 lines preferred) and a section to put it in. You
1832 can also propose new sections.
1834 At this time I would like to limit the references to things that
1835 are available on the web.
1840 - LUKS on-disk format spec:
1841 http://code.google.com/p/cryptsetup/wiki/Specification
1846 - Some code examples are in the source package under docs/examples
1849 * *Brute-forciong passphrases
1852 http://news.electricalchemy.net/2009/10/password-cracking-in-cloud-part-5.html
1855 http://it.slashdot.org/story/12/12/05/0623215/new-25-gpu-monster-devours-strong-passwords-in-minutes
1861 * SSD and Flash Disk Related
1867 * Attacks Against Disk Encryption
1870 * Risk Management as Relevant for Disk Encryption
1878 A. Contributors In no particular order: