1 Frequently Asked Questions Cryptsetup/LUKS
9 6. Backup and Data Recovery
10 7. Interoperability with other Disk Encryption Tools
11 8. Issues with Specific Versions of cryptsetup
12 9. The Initrd question
14 11. References and Further Reading
22 This is the FAQ (Frequently Asked Questions) for cryptsetup. It covers
23 Linux disk encryption with plain dm-crypt (one passphrase, no
24 management, no metadata on disk) and LUKS (multiple user keys with one
25 master key, anti-forensic features, metadata block at start of device,
26 ...). The latest version of this FAQ should usually be available at
27 https://gitlab.com/cryptsetup/cryptsetup/wikis/FrequentlyAskedQuestions
32 LUKS2 COMPATIBILITY: This FAQ was originally written for LUKS1, not
33 LUKS2. Hence regarding LUKS2, some of the answers found here may not
34 apply. Updates for LUKS2 have been done and anything not applying to
35 LUKS2 should clearly say LUKS1. However, this is a Frequently Asked
36 Questions, and questions for LUKS2 are limited at this time or at least
37 those that have reached me are. In the following, "LUKS" refers to both
40 The LUKS1 on-disk format specification is at
41 https://www.kernel.org/pub/linux/utils/cryptsetup/LUKS_docs/on-disk-format.pdf
42 The LUKS2 on-disk format specification is at
43 https://gitlab.com/cryptsetup/LUKS2-docs
45 ATTENTION: If you are going to read just one thing, make it the section
46 on Backup and Data Recovery. By far the most questions on the
47 cryptsetup mailing list are from people that managed to damage the start
48 of their LUKS partitions, i.e. the LUKS header. In most cases, there
49 is nothing that can be done to help these poor souls recover their data.
50 Make sure you understand the problem and limitations imposed by the LUKS
51 security model BEFORE you face such a disaster! In particular, make
52 sure you have a current header backup before doing any potentially
53 dangerous operations. The LUKS2 header should be a bit more resilient
54 as critical data starts later and is stored twice, but you can decidely
55 still destroy it or a keyslot permanently by accident.
57 DEBUG COMMANDS: While the --debug and --debug-json options should not
58 leak secret data, "strace" and the like can leak your full passphrase.
59 Do not post an strace output with the correct passphrase to a
60 mailing-list or online! See Item 4.5 for more explanation.
62 SSDs/FLASH DRIVES: SSDs and Flash are different. Currently it is
63 unclear how to get LUKS or plain dm-crypt to run on them with the full
64 set of security assurances intact. This may or may not be a problem,
65 depending on the attacker model. See Section 5.19.
67 BACKUP: Yes, encrypted disks die, just as normal ones do. A full backup
68 is mandatory, see Section "6. Backup and Data Recovery" on options for
69 doing encrypted backup.
71 CLONING/IMAGING: If you clone or image a LUKS container, you make a copy
72 of the LUKS header and the master key will stay the same! That means
73 that if you distribute an image to several machines, the same master key
74 will be used on all of them, regardless of whether you change the
75 passphrases. Do NOT do this! If you do, a root-user on any of the
76 machines with a mapped (decrypted) container or a passphrase on that
77 machine can decrypt all other copies, breaking security. See also Item
80 DISTRIBUTION INSTALLERS: Some distribution installers offer to create
81 LUKS containers in a way that can be mistaken as activation of an
82 existing container. Creating a new LUKS container on top of an existing
83 one leads to permanent, complete and irreversible data loss. It is
84 strongly recommended to only use distribution installers after a
85 complete backup of all LUKS containers has been made.
87 UBUNTU INSTALLER: In particular the Ubuntu installer seems to be quite
88 willing to kill LUKS containers in several different ways. Those
89 responsible at Ubuntu seem not to care very much (it is very easy to
90 recognize a LUKS container), so treat the process of installing Ubuntu
91 as a severe hazard to any LUKS container you may have.
93 NO WARNING ON NON-INTERACTIVE FORMAT: If you feed cryptsetup from STDIN
94 (e.g. via GnuPG) on LUKS format, it does not give you the warning that
95 you are about to format (and e.g. will lose any pre-existing LUKS
96 container on the target), as it assumes it is used from a script. In
97 this scenario, the responsibility for warning the user and possibly
98 checking for an existing LUKS header is shifted to the script. This is
99 a more general form of the previous item.
101 LUKS PASSPHRASE IS NOT THE MASTER KEY: The LUKS passphrase is not used
102 in deriving the master key. It is used in decrypting a master key that
103 is randomly selected on header creation. This means that if you create
104 a new LUKS header on top of an old one with exactly the same parameters
105 and exactly the same passphrase as the old one, it will still have a
106 different master key and your data will be permanently lost.
108 PASSPHRASE CHARACTER SET: Some people have had difficulties with this
109 when upgrading distributions. It is highly advisable to only use the 95
110 printable characters from the first 128 characters of the ASCII table,
111 as they will always have the same binary representation. Other
112 characters may have different encoding depending on system configuration
113 and your passphrase will not work with a different encoding. A table of
114 the standardized first 128 ASCII characters can, e.g. be found on
115 http://en.wikipedia.org/wiki/ASCII
117 KEYBOARD NUM-PAD: Apparently some pre-boot authentication environments
118 (these are done by the distro, not by cryptsetup, so complain there)
119 treat digits entered on the num-pad and ones entered regularly
120 different. This may be because the BIOS USB keyboard driver is used and
121 that one may have bugs on some computers. If you cannot open your
122 device in pre-boot, try entering the digits over the regular digit keys.
125 * 1.3 System specific warnings
127 - The Ubuntu Natty uinstaller has a "won't fix" defect that may destroy
128 LUKS containers. This is quite old an not relevant for most people.
130 https://bugs.launchpad.net/ubuntu/+source/partman-crypto/+bug/420080
133 * 1.4 My LUKS-device is broken! Help!
135 First: Do not panic! In many cases the data is still recoverable.
136 Do not do anything hasty! Steps:
138 - Take some deep breaths. Maybe add some relaxing music. This may
139 sound funny, but I am completely serious. Often, critical damage is
140 done only after the initial problem.
142 - Do not reboot. The keys may still be in the kernel if the device is
145 - Make sure others do not reboot the system.
147 - Do not write to your disk without a clear understanding why this will
148 not make matters worse. Do a sector-level backup before any writes.
149 Often you do not need to write at all to get enough access to make a
154 - Read section 6 of this FAQ.
156 - Ask on the mailing-list if you need more help.
159 * 1.5 Who wrote this?
161 Current FAQ maintainer is Arno Wagner <arno@wagner.name>. If you want
162 to send me encrypted email, my current PGP key is DSA key CB5D9718,
163 fingerprint 12D6 C03B 1B30 33BB 13CF B774 E35C 5FA1 CB5D 9718.
165 Other contributors are listed at the end. If you want to contribute,
166 send your article, including a descriptive headline, to the maintainer,
167 or the dm-crypt mailing list with something like "FAQ ..."
168 in the subject. You can also send more raw information and have
169 me write the section. Please note that by contributing to this FAQ,
170 you accept the license described below.
172 This work is under the "Attribution-Share Alike 3.0 Unported" license,
173 which means distribution is unlimited, you may create derived works, but
174 attributions to original authors and this license statement must be
175 retained and the derived work must be under the same license. See
176 http://creativecommons.org/licenses/by-sa/3.0/ for more details of the
179 Side note: I did text license research some time ago and I think this
180 license is best suited for the purpose at hand and creates the least
184 * 1.6 Where is the project website?
186 There is the project website at
187 https://gitlab.com/cryptsetup/cryptsetup/ Please do not post
188 questions there, nobody will read them. Use the mailing-list
192 * 1.7 Is there a mailing-list?
194 Instructions on how to subscribe to the mailing-list are at on the
195 project website. People are generally helpful and friendly on the
198 The question of how to unsubscribe from the list does crop up sometimes.
199 For this you need your list management URL, which is sent to you
200 initially and once at the start of each month. Go to the URL mentioned
201 in the email and select "unsubscribe". This page also allows you to
202 request a password reminder.
204 Alternatively, you can send an Email to dm-crypt-request@saout.de with
205 just the word "help" in the subject or message body. Make sure to send
206 it from your list address.
208 The mailing list archive is here:
209 https://marc.info/?l=dm-crypt
212 * 1.8 Unsubscribe from the mailing-list
214 Send mail to dm-crypt-unsubscribe@saout.de from the subscribed account.
215 You will get an email with instructions.
217 Basically, you just have to respond to it unmodified to get
218 unsubscribed. The listserver admin functions are not very fast. It can
219 take 15 minutes or longer for a reply to arrive (I suspect greylisting
220 is in use), so be patient.
222 Also note that nobody on the list can unsubscribe you, sending demands
223 to be unsubscribed to the list just annoys people that are entirely
224 blameless for you being subscribed.
226 If you are subscribed, a subscription confirmation email was sent to
227 your email account and it had to be answered before the subscription
228 went active. The confirmation emails from the listserver have subjects
229 like these (with other numbers):
231 Subject: confirm 9964cf10.....
233 and are sent from dm-crypt-request@saout.de. You should check whether
234 you have anything like it in your sent email folder. If you find
235 nothing and are sure you did not confirm, then you should look into a
236 possible compromise of your email account.
241 * 2.1 LUKS Container Setup mini-HOWTO
243 This item tries to give you a very brief list of all the steps you
244 should go though when creating a new LUKS encrypted container, i.e.
245 encrypted disk, partition or loop-file.
247 01) All data will be lost, if there is data on the target, make a
250 02) Make very sure you use the right target disk, partition or
253 03) If the target was in use previously, it is a good idea to wipe it
254 before creating the LUKS container in order to remove any trace of old
255 file systems and data. For example, some users have managed to run
256 e2fsck on a partition containing a LUKS container, possibly because of
257 residual ext2 superblocks from an earlier use. This can do arbitrary
258 damage up to complete and permanent loss of all data in the LUKS
261 To just quickly wipe file systems (old data may remain), use
263 wipefs -a <target device>
265 To wipe file system and data, use something like
267 cat /dev/zero > <target device>
269 This can take a while. To get a progress indicator, you can use the
270 tool dd_rescue (->google) instead or use my stream meter "wcs" (source
271 here: http://www.tansi.org/tools/index.html) in the following fashion:
273 cat /dev/zero | wcs > <target device>
275 Plain "dd" also gives you the progress on a SIGUSR1, see its man-page.
277 Be very sure you have the right target, all data will be lost!
279 Note that automatic wiping is on the TODO list for cryptsetup, so at
280 some time in the future this will become unnecessary.
282 Alternatively, plain dm-crypt can be used for a very fast wipe with
283 crypto-grade randomness, see Item 2.19
285 04) Create the LUKS container.
289 cryptsetup luksFormat --type luks1 <target device>
293 cryptsetup luksFormat --type luks2 <target device>
296 Just follow the on-screen instructions.
298 Note: Passprase iteration count is based on time and hence security
299 level depends on CPU power of the system the LUKS container is created
300 on. For example on a Raspberry Pi and LUKS1, I found some time ago that
301 the iteration count is 15 times lower than for a regular PC (well, for
302 my old one). Depending on security requirements, this may need
303 adjustment. For LUKS1, you can just look at the iteration count on
304 different systems and select one you like. You can also change the
305 benchmark time with the -i parameter to create a header for a slower
308 For LUKS2, the parameters are more complex. ARGON2 has iteration,
309 parallelism and memory parameter. cryptsetup actually may adjust the
310 memory parameter for time scaling. Hence to use -i is the easiest way
311 to get slower or faster opening (default: 2000 = 2sec). Just make sure
312 to not drop this too low or you may get a memory parameter that is to
313 small to be secure. The luksDump command lists the memory parameter of
314 a created LUKS2 keyslot in kB. That parameter should probably be not
315 much lower than 100000, i.e. 100MB, but don't take my word for it.
317 05) Map the container. Here it will be mapped to /dev/mapper/c1:
319 cryptsetup luksOpen <target device> c1
321 06) (Optionally) wipe the container (make sure you have the right
324 cat /dev/zero > /dev/mapper/c1
326 This will take a while. Note that this creates a small information
327 leak, as an attacker can determine whether a 512 byte block is zero if
328 the attacker has access to the encrypted container multiple times.
329 Typically a competent attacker that has access multiple times can
330 install a passphrase sniffer anyways, so this leakage is not very
331 significant. For getting a progress indicator, see step 03.
333 07) Create a file system in the mapped container, for example an
334 ext3 file system (any other file system is possible):
336 mke2fs -j /dev/mapper/c1
338 08) Mount your encrypted file system, here on /mnt:
340 mount /dev/mapper/c1 /mnt
342 09) Make a LUKS header backup and plan for a container backup.
343 See Section 6 for details.
345 Done. You can now use the encrypted file system to store data. Be sure
346 to read though the rest of the FAQ, these are just the very basics. In
347 particular, there are a number of mistakes that are easy to make, but
348 will compromise your security.
351 * 2.2 LUKS on partitions or raw disks? What about RAID?
354 This is a complicated question, and made more so by the availability of
355 RAID and LVM. I will try to give some scenarios and discuss advantages
356 and disadvantages. Note that I say LUKS for simplicity, but you can do
357 all the things described with plain dm-crypt as well. Also note that
358 your specific scenario may be so special that most or even all things I
359 say below do not apply.
361 Be aware that if you add LVM into the mix, things can get very
362 complicated. Same with RAID but less so. In particular, data recovery
363 can get exceedingly difficult. Only add LVM if you have a really good
364 reason and always remember KISS is what separates an engineer from an
365 amateur. Of course, if you really need the added complexity, KISS is
366 satisfied. But be very sure as there is a price to pay for it. In
367 engineering, complexity is always the enemy and needs to be fought
368 without mercy when encountered.
370 Also consider using RAID instead of LVM, as at least with the old
371 superblock format 0.90, the RAID superblock is in the place (end of
372 disk) where the risk of it damaging the LUKS header is smallest and you
373 can have your array assembled by the RAID controller (i.e. the kernel),
374 as it should be. Use partition type 0xfd for that. I recommend staying
375 away from superblock formats 1.0, 1.1 and 1.2 unless you really need
380 (1) Encrypted partition: Just make a partition to your liking, and put
381 LUKS on top of it and a filesystem into the LUKS container. This gives
382 you isolation of differently-tasked data areas, just as ordinary
383 partitioning does. You can have confidential data, non-confidential
384 data, data for some specific applications, user-homes, root, etc.
385 Advantages are simplicity as there is a 1:1 mapping between partitions
386 and filesystems, clear security functionality and the ability to
387 separate data into different, independent (!) containers.
389 Note that you cannot do this for encrypted root, that requires an
390 initrd. On the other hand, an initrd is about as vulnerable to a
391 competent attacker as a non-encrypted root, so there really is no
392 security advantage to doing it that way. An attacker that wants to
393 compromise your system will just compromise the initrd or the kernel
394 itself. The better way to deal with this is to make sure the root
395 partition does not store any critical data and to move that to
396 additional encrypted partitions. If you really are concerned your root
397 partition may be sabotaged by somebody with physical access (who would
398 however strangely not, say, sabotage your BIOS, keyboard, etc.), protect
399 it in some other way. The PC is just not set-up for a really secure
400 boot-chain (whatever some people may claim).
402 (2) Fully encrypted raw block device: For this, put LUKS on the raw
403 device (e.g. /dev/sdb) and put a filesystem into the LUKS container, no
404 partitioning whatsoever involved. This is very suitable for things like
405 external USB disks used for backups or offline data-storage.
407 (3) Encrypted RAID: Create your RAID from partitions and/or full
408 devices. Put LUKS on top of the RAID device, just if it were an
409 ordinary block device. Applications are just the same as above, but you
410 get redundancy. (Side note as many people seem to be unaware of it: You
411 can do RAID1 with an arbitrary number of components in Linux.) See also
414 (4) Now, some people advocate doing the encryption below the RAID layer.
415 That has several serious problems. One is that suddenly debugging RAID
416 issues becomes much harder. You cannot do automatic RAID assembly
417 anymore. You need to keep the encryption keys for the different RAID
418 components in sync or manage them somehow. The only possible advantage
419 is that things may run a little faster as more CPUs do the encryption,
420 but if speed is a priority over security and simplicity, you are doing
421 this wrong anyways. A good way to mitigate a speed issue is to get a
422 CPU that does hardware AES as most do today.
425 * 2.3 How do I set up encrypted swap?
427 As things that are confidential can end up in swap (keys, passphrases,
428 etc. are usually protected against being swapped to disk, but other
429 things may not be), it may be advisable to do something about the issue.
430 One option is to run without swap, which generally works well in a
431 desktop-context. It may cause problems in a server-setting or under
432 special circumstances. The solution to that is to encrypt swap with a
433 random key at boot-time.
435 NOTE: This is for Debian, and should work for Debian-derived
436 distributions. For others you may have to write your own startup script
437 or use other mechanisms.
439 01) Add the swap partition to /etc/crypttab. A line like the
440 following should do it:
442 swap /dev/<partition> /dev/urandom swap,noearly
444 Warning: While Debian refuses to overwrite partitions with a filesystem
445 or RAID signature on it, as your disk IDs may change (adding or removing
446 disks, failure of disk during boot, etc.), you may want to take
447 additional precautions. Yes, this means that your kernel device names
448 like sda, sdb, ... can change between reboots! This is not a concern
449 if you have only one disk. One possibility is to make sure the
450 partition number is not present on additional disks or also swap there.
451 Another is to encapsulate the swap partition (by making it a 1-partition
452 RAID1 or by using LVM), as that gets a persistent identifier.
453 Specifying it directly by UUID does not work, unfortunately, as the UUID
454 is part of the swap signature and that is not visible from the outside
455 due to the encryption and in addition changes on each reboot with this
458 Note: Use /dev/random if you are paranoid or in a potential low-entropy
459 situation (embedded system, etc.). This may cause the operation to take
460 a long time during boot however. If you are in a "no entropy"
461 situation, you cannot encrypt swap securely. In this situation you
462 should find some entropy, also because nothing else using crypto will be
463 secure, like ssh, ssl or GnuPG.
465 Note: The "noearly" option makes sure things like LVM, RAID, etc. are
466 running. As swap is non-critical for boot, it is fine to start it late.
468 02) Add the swap partition to /etc/fstab. A line like the following
471 /dev/mapper/swap none swap sw 0 0
473 That is it. Reboot or start it manually to activate encrypted swap.
474 Manual start would look like this:
476 /etc/init.d/crypdisks start
477 swapon /dev/mapper/swap
480 * 2.4 What is the difference between "plain" and LUKS format?
482 First, unless you happen to understand the cryptographic background
483 well, you should use LUKS. It does protect the user from a lot of
484 common mistakes. Plain dm-crypt is for experts.
486 Plain format is just that: It has no metadata on disk, reads all
487 parameters from the commandline (or the defaults), derives a master-key
488 from the passphrase and then uses that to de-/encrypt the sectors of the
489 device, with a direct 1:1 mapping between encrypted and decrypted
492 Primary advantage is high resilience to damage, as one damaged encrypted
493 sector results in exactly one damaged decrypted sector. Also, it is not
494 readily apparent that there even is encrypted data on the device, as an
495 overwrite with crypto-grade randomness (e.g. from
496 /dev/urandom) looks exactly the same on disk.
498 Side-note: That has limited value against the authorities. In civilized
499 countries, they cannot force you to give up a crypto-key anyways. In
500 quite a few countries around the world, they can force you to give up
501 the keys (using imprisonment or worse to pressure you, sometimes without
502 due process), and in the worst case, they only need a nebulous
503 "suspicion" about the presence of encrypted data. Sometimes this
504 applies to everybody, sometimes only when you are suspected of having
505 "illicit data" (definition subject to change) and sometimes specifically
506 when crossing a border. Note that this is going on in countries like
507 the US and the UK to different degrees and sometimes with courts
508 restricting what the authorities can actually demand.
510 My advice is to either be ready to give up the keys or to not have
511 encrypted data when traveling to those countries, especially when
512 crossing the borders. The latter also means not having any high-entropy
513 (random) data areas on your disk, unless you can explain them and
514 demonstrate that explanation. Hence doing a zero-wipe of all free
515 space, including unused space, may be a good idea.
517 Disadvantages are that you do not have all the nice features that the
518 LUKS metadata offers, like multiple passphrases that can be changed, the
519 cipher being stored in the metadata, anti-forensic properties like
520 key-slot diffusion and salts, etc..
522 LUKS format uses a metadata header and 8 key-slot areas that are being
523 placed at the beginning of the disk, see below under "What does the LUKS
524 on-disk format looks like?". The passphrases are used to decrypt a
525 single master key that is stored in the anti-forensic stripes. LUKS2
526 adds some more flexibility.
528 Advantages are a higher usability, automatic configuration of
529 non-default crypto parameters, defenses against low-entropy passphrases
530 like salting and iterated PBKDF2 or ARGON 2 passphrase hashing, the
531 ability to change passphrases, and others.
533 Disadvantages are that it is readily obvious there is encrypted data on
534 disk (but see side note above) and that damage to the header or
535 key-slots usually results in permanent data-loss. See below under "6.
536 Backup and Data Recovery" on how to reduce that risk. Also the sector
537 numbers get shifted by the length of the header and key-slots and there
538 is a loss of that size in capacity. Unless you have a specific need,
542 * 2.5 Can I encrypt an existing, non-empty partition to use LUKS?
544 There is no converter, and it is not really needed. The way to do this
545 is to make a backup of the device in question, securely wipe the device
546 (as LUKS device initialization does not clear away old data), do a
547 luksFormat, optionally overwrite the encrypted device, create a new
548 filesystem and restore your backup on the now encrypted device. Also
549 refer to sections "Security Aspects" and "Backup and Data Recovery".
551 For backup, plain GNU tar works well and backs up anything likely to be
555 * 2.6 How do I use LUKS with a loop-device?
557 This can be very handy for experiments. Setup is just the same as with
558 any block device. If you want, for example, to use a 100MiB file as
559 LUKS container, do something like this:
561 head -c 100M /dev/zero > luksfile # create empty file
562 losetup /dev/loop0 luksfile # map file to /dev/loop0
563 cryptsetup luksFormat --type luks2 /dev/loop0 # create LUKS2 container
565 Afterwards just use /dev/loop0 as a you would use a LUKS partition.
566 To unmap the file when done, use "losetup -d /dev/loop0".
569 * 2.7 When I add a new key-slot to LUKS, it asks for a passphrase
570 but then complains about there not being a key-slot with that
573 That is as intended. You are asked a passphrase of an existing key-slot
574 first, before you can enter the passphrase for the new key-slot.
575 Otherwise you could break the encryption by just adding a new key-slot.
576 This way, you have to know the passphrase of one of the already
577 configured key-slots in order to be able to configure a new key-slot.
580 * 2.8 Encryption on top of RAID or the other way round?
583 Unless you have special needs, place encryption between RAID and
584 filesystem, i.e. encryption on top of RAID. You can do it the other
585 way round, but you have to be aware that you then need to give the
586 passphrase for each individual disk and RAID auto-detection will not
587 work anymore. Therefore it is better to encrypt the RAID device, e.g.
590 This means that the typical layering looks like this:
598 Raw partitions (optional)
602 The big advantage of this is that you can manage the RAID container just
603 like any other regular RAID container, it does not care that its content
604 is encrypted. This strongly cuts down on complexity, something very
605 valuable with storage encryption.
608 * 2.9 How do I read a dm-crypt key from file?
610 Use the --key-file option, like this:
612 cryptsetup create --key-file keyfile e1 /dev/loop0
614 This will read the binary key from file, i.e. no hashing or
615 transformation will be applied to the keyfile before its bits are used
616 as key. Extra bits (beyond the length of the key) at the end are
617 ignored. Note that if you read from STDIN, the data will be hashed,
618 just as a key read interactively from the terminal. See the man-page
619 sections "NOTES ON PASSPHRASE PROCESSING..." for more detail.
622 * 2.10 How do I read a LUKS slot key from file?
624 What you really do here is to read a passphrase from file, just as you
625 would with manual entry of a passphrase for a key-slot. You can add a
626 new passphrase to a free key-slot, set the passphrase of an specific
627 key-slot or put an already configured passphrase into a file. Make sure
628 no trailing newline (0x0a) is contained in the input key file, or the
629 passphrase will not work because the whole file is used as input.
631 To add a new passphrase to a free key slot from file, use something
634 cryptsetup luksAddKey /dev/loop0 keyfile
636 To add a new passphrase to a specific key-slot, use something
639 cryptsetup luksAddKey --key-slot 7 /dev/loop0 keyfile
641 To supply a key from file to any LUKS command, use the --key-file
642 option, e.g. like this:
644 cryptsetup luksOpen --key-file keyfile /dev/loop0 e1
648 * 2.11 How do I read the LUKS master key from file?
650 The question you should ask yourself first is why you would want to do
651 this. The only legitimate reason I can think of is if you want to have
652 two LUKS devices with the same master key. Even then, I think it would
653 be preferable to just use key-slots with the same passphrase, or to use
654 plain dm-crypt instead. If you really have a good reason, please tell
655 me. If I am convinced, I will add how to do this here.
658 * 2.12 What are the security requirements for a key read from file?
660 A file-stored key or passphrase has the same security requirements as
661 one entered interactively, however you can use random bytes and thereby
662 use bytes you cannot type on the keyboard. You can use any file you
663 like as key file, for example a plain text file with a human readable
664 passphrase. To generate a file with random bytes, use something like
667 head -c 256 /dev/random > keyfile
671 * 2.13 If I map a journaled file system using dm-crypt/LUKS, does
672 it still provide its usual transactional guarantees?
674 Yes, it does, unless a very old kernel is used. The required flags come
675 from the filesystem layer and are processed and passed onward by
676 dm-crypt (regardless of direct key management or LUKS key management).
677 A bit more information on the process by which transactional guarantees
678 are implemented can be found here:
680 http://lwn.net/Articles/400541/
682 Please note that these "guarantees" are weaker than they appear to be.
683 One problem is that quite a few disks lie to the OS about having flushed
684 their buffers. This is likely still true with SSDs. Some other things
685 can go wrong as well. The filesystem developers are aware of these
686 problems and typically can make it work anyways. That said,
687 dm-crypt/LUKS will not make things worse.
689 One specific problem you can run into is that you can get short freezes
690 and other slowdowns due to the encryption layer. Encryption takes time
691 and forced flushes will block for that time. For example, I did run
692 into frequent small freezes (1-2 sec) when putting a vmware image on
693 ext3 over dm-crypt. When I went back to ext2, the problem went away.
694 This seems to have gotten better with kernel 2.6.36 and the reworking of
695 filesystem flush locking mechanism (less blocking of CPU activity during
696 flushes). This should improve further and eventually the problem should
700 * 2.14 Can I use LUKS or cryptsetup with a more secure (external)
701 medium for key storage, e.g. TPM or a smartcard?
703 Yes, see the answers on using a file-supplied key. You do have to write
704 the glue-logic yourself though. Basically you can have cryptsetup read
705 the key from STDIN and write it there with your own tool that in turn
706 gets the key from the more secure key storage.
708 For TPM support, you may want to have a look at tpm-luks at
709 https://github.com/shpedoikal/tpm-luks. Note that tpm-luks is not
710 related to the cryptsetup project.
713 * 2.15 Can I resize a dm-crypt or LUKS container?
715 Yes, you can, as neither dm-crypt nor LUKS1 stores partition size and
716 LUKS2 uses a generic "whole device" size as default. Note that LUKS2
717 can use specified data-area sizes as a non-standard case and that these
718 may cause issues when resizing a LUKS2 container if set to a specific
721 Whether you should do this is a different question. Personally I
722 recommend backup, recreation of the dm-crypt or LUKS container with new
723 size, recreation of the filesystem and restore. This gets around the
724 tricky business of resizing the filesystem. Resizing a dm-crypt or LUKS
725 container does not resize the filesystem in it. A backup is really
726 non-optional here, as a lot can go wrong, resulting in partial or
727 complete data loss. But if you have that backup, you can also just
730 You also need to be aware of size-based limitations. The one currently
731 relevant is that aes-xts-plain should not be used for encrypted
732 container sizes larger than 2TiB. Use aes-xts-plain64 for that.
735 * 2.16 How do I Benchmark the Ciphers, Hashes and Modes?
737 Since version 1.60 cryptsetup supports the "benchmark" command.
742 You can get more than the default benchmarks, see the man-page for the
743 relevant parameters. Note that XTS mode takes two keys, hence the
744 listed key sizes are double that for other modes and half of it is the
745 cipher key, the other half is the XTS key.
748 * 2.17 How do I Verify I have an Authentic cryptsetup Source Package?
750 Current maintainer is Milan Broz and he signs the release packages with
751 his PGP key. The key he currently uses is the "RSA key ID D93E98FC",
752 fingerprint 2A29 1824 3FDE 4664 8D06 86F9 D9B0 577B D93E 98FC. While I
753 have every confidence this really is his key and that he is who he
754 claims to be, don't depend on it if your life is at stake. For that
755 matter, if your life is at stake, don't depend on me being who I claim
758 That said, as cryptsetup is under good version control and a malicious
759 change should be noticed sooner or later, but it may take a while.
760 Also, the attacker model makes compromising the sources in a non-obvious
761 way pretty hard. Sure, you could put the master-key somewhere on disk,
762 but that is rather obvious as soon as somebody looks as there would be
763 data in an empty LUKS container in a place it should not be. Doing this
764 in a more nefarious way, for example hiding the master-key in the salts,
765 would need a look at the sources to be discovered, but I think that
766 somebody would find that sooner or later as well.
768 That said, this discussion is really a lot more complicated and longer
769 as an FAQ can sustain. If in doubt, ask on the mailing list.
772 * 2.18 Is there a concern with 4k Sectors?
774 Not from dm-crypt itself. Encryption will be done in 512B blocks, but
775 if the partition and filesystem are aligned correctly and the filesystem
776 uses multiples of 4kiB as block size, the dm-crypt layer will just
777 process 8 x 512B = 4096B at a time with negligible overhead. LUKS does
778 place data at an offset, which is 2MiB per default and will not break
779 alignment. See also Item 6.12 of this FAQ for more details. Note that
780 if your partition or filesystem is misaligned, dm-crypt can make the
781 effect worse though. Also note that SSDs typically have much larger
782 blocks internally (e.g. 128kB or even larger).
785 * 2.19 How can I wipe a device with crypto-grade randomness?
787 The conventional recommendation if you want to do more than just a
788 zero-wipe is to use something like
790 cat /dev/urandom > <taget-device>
792 That used to very slow and painful at 10-20MB/s on a fast computer, but
793 newer kernels can give you > 200MB/s (depending on hardware). An
794 alternative is using cryptsetup and a plain dm-crypt device with a
795 random key, which is fast and on the same level of security. The
796 defaults are quite enough.
798 For device set-up, do the following:
800 cryptsetup open --type plain -d /dev/urandom /dev/<device> target
802 This maps the container as plain under /dev/mapper/target with a random
803 password. For the actual wipe you have several options. Basically, you
804 pipe zeroes into the opened container that then get encrypted. Simple
805 wipe without progress-indicator:
807 cat /dev/zero > /dev/mapper/to_be_wiped
809 Progress-indicator by dd_rescue:
811 dd_rescue -w /dev/zero /dev/mapper/to_be_wiped
813 Progress-indicator by my "wcs" stream meter (available from
814 http://www.tansi.org/tools/index.html ):
816 cat /dev/zero | wcs > /dev/mapper/to_be_wiped
818 Or use plain "dd", which gives you the progress when sent a SIGUSR1, see
821 Remove the mapping at the end and you are done.
824 * 2.20 How to I wipe only the LUKS header?
826 This does _not_ describe an emergency wipe procedure, see Item 5.4 for
827 that. This procedure here is intended to be used when the data should
828 stay intact, e.g. when you change your LUKS container to use a detached
829 header and want to remove the old one. Please only do this if you have
833 01) Determine header size in 512 Byte sectors with luksDump:
835 cryptsetup luksDump <device with LUKS container>
838 Payload offset: <number>
841 02) Take the result number, multiply by 512 zeros and write to
842 the start of the device, e.g. like this:
844 dd bs=512 count=<number> if=/dev/zero of=<device>
847 LUKS2: (warning, untested! Remember that backup?) This assumes the
848 LUKS2 container uses the defaults, in particular there is only one data
849 segment. 01) Determine the data-segment offset using luksDump, same
855 offset: <number> [bytes]
858 02) Overwrite the stated number of bytes from the start of the device.
859 Just to give yet another way to get a defined number of zeros:
861 head -c /dev/zero > /dev/<device>
867 * 3.1 My dm-crypt/LUKS mapping does not work! What general steps
868 are there to investigate the problem?
870 If you get a specific error message, investigate what it claims first.
871 If not, you may want to check the following things.
873 - Check that "/dev", including "/dev/mapper/control" is there. If it is
874 missing, you may have a problem with the "/dev" tree itself or you may
875 have broken udev rules.
877 - Check that you have the device mapper and the crypt target in your
878 kernel. The output of "dmsetup targets" should list a "crypt" target.
879 If it is not there or the command fails, add device mapper and
880 crypt-target to the kernel.
882 - Check that the hash-functions and ciphers you want to use are in the
883 kernel. The output of "cat /proc/crypto" needs to list them.
886 * 3.2 My dm-crypt mapping suddenly stopped when upgrading cryptsetup.
888 The default cipher, hash or mode may have changed (the mode changed from
889 1.0.x to 1.1.x). See under "Issues With Specific Versions of
893 * 3.3 When I call cryptsetup from cron/CGI, I get errors about
896 If you get errors about unknown parameters or the like that are not
897 present when cryptsetup is called from the shell, make sure you have no
898 older version of cryptsetup on your system that then gets called by
899 cron/CGI. For example some distributions install cryptsetup into
900 /usr/sbin, while a manual install could go to /usr/local/sbin. As a
901 debugging aid, call "cryptsetup --version" from cron/CGI or the
902 non-shell mechanism to be sure the right version gets called.
905 * 3.4 Unlocking a LUKS device takes very long. Why?
907 The unlock time for a key-slot (see Section 5 for an explanation what
908 iteration does) is calculated when setting a passphrase. By default it
909 is 1 second (2 seconds for LUKS2). If you set a passphrase on a fast
910 machine and then unlock it on a slow machine, the unlocking time can be
911 much longer. Also take into account that up to 8 key-slots (LUKS2: up
912 to 32 key-slots) have to be tried in order to find the right one.
914 If this is problem, you can add another key-slot using the slow machine
915 with the same passphrase and then remove the old key-slot. The new
916 key-slot will have the unlock time adjusted to the slow machine. Use
917 luksKeyAdd and then luksKillSlot or luksRemoveKey. You can also use
918 the -i option to reduce iteration time (and security level) when setting
919 a passphrase. Default is 1000 (1 sec) for LUKS1 and 2000 (2sec) for
922 However, this operation will not change volume key iteration count ("MK
923 iterations" for LUKS1, "Iterations" under "Digests" for LUKS2). In
924 order to change that, you will have to backup the data in the LUKS
925 container (i.e. your encrypted data), luksFormat on the slow machine
926 and restore the data. Note that MK iterations are not very security
930 * 3.5 "blkid" sees a LUKS UUID and an ext2/swap UUID on the same
931 device. What is wrong?
933 Some old versions of cryptsetup have a bug where the header does not get
934 completely wiped during LUKS format and an older ext2/swap signature
935 remains on the device. This confuses blkid.
937 Fix: Wipe the unused header areas by doing a backup and restore of
938 the header with cryptsetup 1.1.x or later:
940 cryptsetup luksHeaderBackup --header-backup-file <file> <device>
941 cryptsetup luksHeaderRestore --header-backup-file <file> <device>
948 * 4.1 I get the error "LUKS keyslot x is invalid." What does that mean?
950 For LUKS1, this means that the given keyslot has an offset that points
951 outside the valid keyslot area. Typically, the reason is a corrupted
952 LUKS1 header because something was written to the start of the device
953 the LUKS1 container is on. For LUKS2, I do not know when this error can
954 happen, but I expect it will be something similar. Refer to Section
955 "Backup and Data Recovery" and ask on the mailing list if you have
956 trouble diagnosing and (if still possible) repairing this.
959 * 4.2 I cannot unlock my LUKS container! What could be the problem?
961 First, make sure you have a correct passphrase. Then make sure you have
962 the correct key-map and correct keyboard. And then make sure you have
963 the correct character set and encoding, see also "PASSPHRASE CHARACTER
964 SET" under Section 1.2.
966 If you are sure you are entering the passphrase right, there is the
967 possibility that the respective key-slot has been damaged. There is no
968 way to recover a damaged key-slot, except from a header backup (see
969 Section 6). For security reasons, there is also no checksum in the
970 key-slots that could tell you whether a key-slot has been damaged. The
971 only checksum present allows recognition of a correct passphrase, but
972 that only works with that correct passphrase and a respective key-slot
975 In order to find out whether a key-slot is damaged one has to look for
976 "non-random looking" data in it. There is a tool that automatizes this
977 for LUKS1 in the cryptsetup distribution from version 1.6.0 onwards. It
978 is located in misc/keyslot_checker/. Instructions how to use and how to
979 interpret results are in the README file. Note that this tool requires
980 a libcryptsetup from cryptsetup 1.6.0 or later (which means
981 libcryptsetup.so.4.5.0 or later). If the tool complains about missing
982 functions in libcryptsetup, you likely have an earlier version from your
983 distribution still installed. You can either point the symbolic link(s)
984 from libcryptsetup.so.4 to the new version manually, or you can
985 uninstall the distribution version of cryptsetup and re-install that
986 from cryptsetup >= 1.6.0 again to fix this.
989 * 4.3 Can a bad RAM module cause problems?
991 LUKS and dm-crypt can give the RAM quite a workout, especially when
992 combined with software RAID. In particular the combination RAID5 +
993 LUKS1 + XFS seems to uncover RAM problems that do not cause obvious
994 problems otherwise. Symptoms vary, but often the problem manifest
995 itself when copying large amounts of data, typically several times
996 larger than your main memory.
998 Note: One thing you should always do on large data copying or movements
999 is to run a verify, for example with the "-d" option of "tar" or by
1000 doing a set of MD5 checksums on the source or target with
1002 find . -type f -exec md5sum \{\} \; > checksum-file
1004 and then a "md5sum -c checksum-file" on the other side. If you get
1005 mismatches here, RAM is the primary suspect. A lesser suspect is an
1006 overclocked CPU. I have found countless hardware problems in verify
1007 runs after copying data or making backups. Bit errors are much more
1008 common than most people think.
1010 Some RAM issues are even worse and corrupt structures in one of the
1011 layers. This typically results in lockups, CPU state dumps in the
1012 system logs, kernel panic or other things. It is quite possible to have
1013 a problem with an encrypted device, but not with an otherwise the same
1014 unencrypted device. The reason for that is that encryption has an error
1015 amplification property: If you flip one bit in an encrypted data block,
1016 the decrypted version has half of its bits flipped. This is actually an
1017 important security property for modern ciphers. With the usual modes in
1018 cryptsetup (CBC, ESSIV, XTS), you can get a completely changed 512 byte
1019 block for a bit error. A corrupt block causes a lot more havoc than the
1020 occasionally flipped single bit and can result in various obscure
1023 Note that a verify run on copying between encrypted or unencrypted
1024 devices will reliably detect corruption, even when the copying itself
1025 did not report any problems. If you find defect RAM, assume all backups
1026 and copied data to be suspect, unless you did a verify.
1029 * 4.4 How do I test RAM?
1031 First you should know that overclocking often makes memory problems
1032 worse. So if you overclock (which I strongly recommend against in a
1033 system holding data that has any worth), run the tests with the
1034 overclocking active.
1036 There are two good options. One is Memtest86+ and the other is
1037 "memtester" by Charles Cazabon. Memtest86+ requires a reboot and then
1038 takes over the machine, while memtester runs from a root-shell. Both
1039 use different testing methods and I have found problems fast with either
1040 one that the other needed long to find. I recommend running the
1041 following procedure until the first error is found:
1043 - Run Memtest86+ for one cycle
1045 - Run memtester for one cycle (shut down as many other applications
1046 as possible and use the largest memory area you can get)
1048 - Run Memtest86+ for 24h or more
1050 - Run memtester for 24h or more
1052 If all that does not produce error messages, your RAM may be sound,
1053 but I have had one weak bit in the past that Memtest86+ needed around
1054 60 hours to find. If you can reproduce the original problem reliably,
1055 a good additional test may be to remove half of the RAM (if you have
1056 more than one module) and try whether the problem is still there and if
1057 so, try with the other half. If you just have one module, get a
1058 different one and try with that. If you do overclocking, reduce the
1059 settings to the most conservative ones available and try with that.
1062 * 4.5 Is there a risk using debugging tools like strace?
1064 There most definitely is. A dump from strace and friends can contain
1065 all data entered, including the full passphrase. Example with strace
1066 and passphrase "test":
1068 > strace cryptsetup luksOpen /dev/sda10 c1
1070 read(6, "test\n", 512) = 5
1073 Depending on different factors and the tool used, the passphrase may
1074 also be encoded and not plainly visible. Hence it is never a good idea
1075 to give such a trace from a live container to anybody. Recreate the
1076 problem with a test container or set a temporary passphrase like "test"
1077 and use that for the trace generation. Item 2.6 explains how to create
1078 a loop-file backed LUKS container that may come in handy for this
1081 See also Item 6.10 for another set of data you should not give to
1088 * 5.1 How long is a secure passphrase ?
1090 This is just the short answer. For more info and explanation of some of
1091 the terms used in this item, read the rest of Section 5. The actual
1092 recommendation is at the end of this item.
1094 First, passphrase length is not really the right measure, passphrase
1095 entropy is. If your passphrase is 200 times the letter "a", it is long
1096 but has very low entropy and is pretty insecure.
1098 For example, a random lowercase letter (a-z) gives you 4.7 bit of
1099 entropy, one element of a-z0-9 gives you 5.2 bits of entropy, an element
1100 of a-zA-Z0-9 gives you 5.9 bits and a-zA-Z0-9!@#$%\^&:-+ gives you 6.2
1101 bits. On the other hand, a random English word only gives you 0.6...1.3
1102 bits of entropy per character. Using sentences that make sense gives
1103 lower entropy, series of random words gives higher entropy. Do not use
1104 sentences that can be tied to you or found on your computer. This type
1105 of attack is done routinely today.
1107 That said, it does not matter too much what scheme you use, but it does
1108 matter how much entropy your passphrase contains, because an attacker
1109 has to try on average
1111 1/2 * 2^(bits of entropy in passphrase)
1113 different passphrases to guess correctly.
1115 Historically, estimations tended to use computing time estimates, but
1116 more modern approaches try to estimate cost of guessing a passphrase.
1118 As an example, I will try to get an estimate from the numbers in
1119 https://gist.github.com/epixoip/a83d38f412b4737e99bbef804a270c40 This
1120 thing costs 23kUSD and does 68Ghashes/sec for SHA1. This is in 2017.
1122 Incidentally, my older calculation for a machine around 1000 times
1123 slower was off by a factor of about 1000, but in the right direction,
1124 i.e. I estimated the attack to be too easy. Nobody noticed ;-) On the
1125 plus side, the tables are now (2017) pretty much accurate.
1127 More references can be found a the end of this document. Note that
1128 these are estimates from the defender side, so assuming something is
1129 easier than it actually is is fine. An attacker may still have
1130 significantly higher cost than estimated here.
1132 LUKS1 used SHA1 (since version 1.7.0 it uses SHA256) for hashing per
1133 default. We will leave aside the check whether a try actually decrypts
1134 a key-slot. I will assume a useful lifetime of the hardware of 2 years.
1135 (This is on the low side.) Disregarding downtime, the machine can then
1138 N = 68*10^9 * 3600 * 24 * 365 * 2 ~ 4*10^18
1140 passphrases for EUR/USD 23k. That is one 62 bit passphrase hashed once
1141 with SHA1 for EUR/USD 23k. This can be parallelized, it can be done
1142 faster than 2 years with several of these machines.
1144 For LUKS2, things look a bit better, as the advantage of using graphics
1145 cards is massively reduced. Using the recommendations below should
1146 hence be fine for LUKS2 as well and give a better security margin.
1148 For plain dm-crypt (no hash iteration) this is it. This gives (with
1149 SHA1, plain dm-crypt default is ripemd160 which seems to be slightly
1152 Passphrase entropy Cost to break
1162 For LUKS1, you have to take into account hash iteration in PBKDF2.
1163 For a current CPU, there are about 100k iterations (as can be queried
1164 with ''cryptsetup luksDump''.
1166 The table above then becomes:
1168 Passphrase entropy Cost to break
1180 To get reasonable security for the next 10 years, it is a good idea
1181 to overestimate by a factor of at least 1000.
1183 Then there is the question of how much the attacker is willing to spend.
1184 That is up to your own security evaluation. For general use, I will
1185 assume the attacker is willing to spend up to 1 million EUR/USD. Then
1186 we get the following recommendations:
1188 Plain dm-crypt: Use > 80 bit. That is e.g. 17 random chars from a-z
1189 or a random English sentence of > 135 characters length.
1191 LUKS1 and LUKS2: Use > 65 bit. That is e.g. 14 random chars from a-z
1192 or a random English sentence of > 108 characters length.
1194 If paranoid, add at least 20 bit. That is roughly four additional
1195 characters for random passphrases and roughly 32 characters for a
1196 random English sentence.
1199 * 5.2 Is LUKS insecure? Everybody can see I have encrypted data!
1201 In practice it does not really matter. In most civilized countries you
1202 can just refuse to hand over the keys, no harm done. In some countries
1203 they can force you to hand over the keys if they suspect encryption.
1204 The suspicion is enough, they do not have to prove anything. This is
1205 for practical reasons, as even the presence of a header (like the LUKS
1206 header) is not enough to prove that you have any keys. It might have
1207 been an experiment, for example. Or it was used as encrypted swap with
1208 a key from /dev/random. So they make you prove you do not have
1209 encrypted data. Of course, if true, that is impossible and hence the
1210 whole idea is not compatible with fair laws. Note that in this context,
1211 countries like the US or the UK are not civilized and do not have fair
1214 This means that if you have a large set of random-looking data, they can
1215 already lock you up. Hidden containers (encryption hidden within
1216 encryption), as possible with Truecrypt, do not help either. They will
1217 just assume the hidden container is there and unless you hand over the
1218 key, you will stay locked up. Don't have a hidden container? Though
1219 luck. Anybody could claim that.
1221 Still, if you are concerned about the LUKS header, use plain dm-crypt
1222 with a good passphrase. See also Section 2, "What is the difference
1223 between "plain" and LUKS format?"
1226 * 5.3 Should I initialize (overwrite) a new LUKS/dm-crypt partition?
1228 If you just create a filesystem on it, most of the old data will still
1229 be there. If the old data is sensitive, you should overwrite it before
1230 encrypting. In any case, not initializing will leave the old data there
1231 until the specific sector gets written. That may enable an attacker to
1232 determine how much and where on the partition data was written. If you
1233 think this is a risk, you can prevent this by overwriting the encrypted
1234 device (here assumed to be named "e1") with zeros like this:
1236 dd_rescue -w /dev/zero /dev/mapper/e1
1238 or alternatively with one of the following more standard commands:
1240 cat /dev/zero > /dev/mapper/e1
1241 dd if=/dev/zero of=/dev/mapper/e1
1245 * 5.4 How do I securely erase a LUKS container?
1247 For LUKS, if you are in a desperate hurry, overwrite the LUKS header and
1248 key-slot area. For LUKS1 and LUKS2, just be generous and overwrite the
1249 first 100MB. A single overwrite with zeros should be enough. If you
1250 anticipate being in a desperate hurry, prepare the command beforehand.
1251 Example with /dev/sde1 as the LUKS partition and default parameters:
1253 head -c 100000000 /dev/zero > /dev/sde1; sync
1255 A LUKS header backup or full backup will still grant access to most or
1256 all data, so make sure that an attacker does not have access to backups
1257 or destroy them as well.
1259 Also note that SSDs and also some HDDs (SMR and hybrid HDDs, for
1260 example) may not actually overwrite the header and only do that an
1261 unspecified and possibly very long time later. The only way to be sure
1262 there is physical destruction. If the situation permits, do both
1263 overwrite and physical destruction.
1265 If you have time, overwrite the whole drive with a single pass of random
1266 data. This is enough for most HDDs. For SSDs or FLASH (USB sticks) or
1267 SMR or hybrid drives, you may want to overwrite the whole drive several
1268 times to be sure data is not retained. This is possibly still insecure
1269 as the respective technologies are not fully understood in this regard.
1270 Still, due to the anti-forensic properties of the LUKS key-slots, a
1271 single overwrite could be enough. If in doubt, use physical destruction
1272 in addition. Here is a link to some current research results on erasing
1273 SSDs and FLASH drives:
1274 http://www.usenix.org/events/fast11/tech/full_papers/Wei.pdf
1276 Keep in mind to also erase all backups.
1278 Example for a random-overwrite erase of partition sde1 done with
1281 dd_rescue -w /dev/urandom /dev/sde1
1285 * 5.5 How do I securely erase a backup of a LUKS partition or header?
1287 That depends on the medium it is stored on. For HDD and SSD, use
1288 overwrite with random data. For an SSD, FLASH drive (USB stick) hybrid
1289 HDD or SMR HDD, you may want to overwrite the complete drive several
1290 times and use physical destruction in addition, see last item. For
1291 re-writable CD/DVD, a single overwrite should be enough, due to the
1292 anti-forensic properties of the LUKS keyslots. For write-once media,
1293 use physical destruction. For low security requirements, just cut the
1294 CD/DVD into several parts. For high security needs, shred or burn the
1297 If your backup is on magnetic tape, I advise physical destruction by
1298 shredding or burning, after (!) overwriting . The problem with magnetic
1299 tape is that it has a higher dynamic range than HDDs and older data may
1300 well be recoverable after overwrites. Also write-head alignment issues
1301 can lead to data not actually being deleted during overwrites.
1303 The best option is to actually encrypt the backup, for example with
1304 PGP/GnuPG and then just destroy all copies of the encryption key if
1305 needed. Best keep them on paper, as that has excellent durability and
1306 secure destruction is easy, for example by burning and then crushing the
1307 ashes to a fine powder. A blender and water also works nicely.
1310 * 5.6 What about backup? Does it compromise security?
1312 That depends. See item 6.7.
1315 * 5.7 Why is all my data permanently gone if I overwrite the LUKS header?
1317 Overwriting the LUKS header in part or in full is the most common reason
1318 why access to LUKS containers is lost permanently. Overwriting can be
1319 done in a number of fashions, like creating a new filesystem on the raw
1320 LUKS partition, making the raw partition part of a raid array and just
1321 writing to the raw partition.
1323 The LUKS1 header contains a 256 bit "salt" per key-slot and without that
1324 no decryption is possible. While the salts are not secret, they are
1325 key-grade material and cannot be reconstructed. This is a
1326 cryptographically strong "cannot". From observations on the cryptsetup
1327 mailing-list, people typically go though the usual stages of grief
1328 (Denial, Anger, Bargaining, Depression, Acceptance) when this happens to
1329 them. Observed times vary between 1 day and 2 weeks to complete the
1330 cycle. Seeking help on the mailing-list is fine. Even if we usually
1331 cannot help with getting back your data, most people found the feedback
1334 If your header does not contain an intact key-slot salt, best go
1335 directly to the last stage ("Acceptance") and think about what to do
1336 now. There is one exception that I know of: If your LUKS1 container is
1337 still open, then it may be possible to extract the master key from the
1338 running system. See Item "How do I recover the master key from a mapped
1339 LUKS1 container?" in Section "Backup and Data Recovery".
1341 For LUKS2, things are both better and worse. First, the salts are in a
1342 less vulnerable position now. But, on the other hand, the keys of a
1343 mapped (open) container are now stored in the kernel key-store, and
1344 while there probably is some way to get them out of there, I am not sure
1345 how much effort that needs.
1348 * 5.8 What is a "salt"?
1350 A salt is a random key-grade value added to the passphrase before it is
1351 processed. It is not kept secret. The reason for using salts is as
1352 follows: If an attacker wants to crack the password for a single LUKS
1353 container, then every possible passphrase has to be tried. Typically an
1354 attacker will not try every binary value, but will try words and
1355 sentences from a dictionary.
1357 If an attacker wants to attack several LUKS containers with the same
1358 dictionary, then a different approach makes sense: Compute the resulting
1359 slot-key for each dictionary element and store it on disk. Then the
1360 test for each entry is just the slow unlocking with the slot key (say
1361 0.00001 sec) instead of calculating the slot-key first (1 sec). For a
1362 single attack, this does not help. But if you have more than one
1363 container to attack, this helps tremendously, also because you can
1364 prepare your table before you even have the container to attack! The
1365 calculation is also very simple to parallelize. You could, for example,
1366 use the night-time unused CPU power of your desktop PCs for this.
1368 This is where the salt comes in. If the salt is combined with the
1369 passphrase (in the simplest form, just appended to it), you suddenly
1370 need a separate table for each salt value. With a reasonably-sized salt
1371 value (256 bit, e.g.) this is quite infeasible.
1374 * 5.9 Is LUKS secure with a low-entropy (bad) passphrase?
1376 Short answer: yes. Do not use a low-entropy passphrase.
1378 Note: For LUKS2, protection for bad passphrases is a bit better
1379 due to the use of Argon2, but that is only a gradual improvement.
1382 This needs a bit of theory. The quality of your passphrase is directly
1383 related to its entropy (information theoretic, not thermodynamic). The
1384 entropy says how many bits of "uncertainty" or "randomness" are in you
1385 passphrase. In other words, that is how difficult guessing the
1388 Example: A random English sentence has about 1 bit of entropy per
1389 character. A random lowercase (or uppercase) character has about 4.7
1392 Now, if n is the number of bits of entropy in your passphrase and t
1393 is the time it takes to process a passphrase in order to open the
1394 LUKS container, then an attacker has to spend at maximum
1396 attack_time_max = 2^n * t
1398 time for a successful attack and on average half that. There is no way
1399 getting around that relationship. However, there is one thing that does
1400 help, namely increasing t, the time it takes to use a passphrase, see
1403 Still, if you want good security, a high-entropy passphrase is the only
1404 option. For example, a low-entropy passphrase can never be considered
1405 secure against a TLA-level (Three Letter Agency level, i.e.
1406 government-level) attacker, no matter what tricks are used in the
1407 key-derivation function. Use at least 64 bits for secret stuff. That
1408 is 64 characters of English text (but only if randomly chosen) or a
1409 combination of 12 truly random letters and digits.
1411 For passphrase generation, do not use lines from very well-known texts
1412 (religious texts, Harry potter, etc.) as they are too easy to guess.
1413 For example, the total Harry Potter has about 1'500'000 words (my
1414 estimation). Trying every 64 character sequence starting and ending at
1415 a word boundary would take only something like 20 days on a single CPU
1416 and is entirely feasible. To put that into perspective, using a number
1417 of Amazon EC2 High-CPU Extra Large instances (each gives about 8 real
1418 cores), this test costs currently about 50USD/EUR, but can be made to
1419 run arbitrarily fast.
1421 On the other hand, choosing 1.5 lines from, say, the Wheel of Time, is
1422 in itself not more secure, but the book selection adds quite a bit of
1423 entropy. (Now that I have mentioned it here, don't use tWoT either!) If
1424 you add 2 or 3 typos and switch some words around, then this is good
1425 passphrase material.
1428 * 5.10 What is "iteration count" and why is decreasing it a bad idea?
1431 Iteration count is the number of PBKDF2 iterations a passphrase is put
1432 through before it is used to unlock a key-slot. Iterations are done
1433 with the explicit purpose to increase the time that it takes to unlock a
1434 key-slot. This provides some protection against use of low-entropy
1437 The idea is that an attacker has to try all possible passphrases. Even
1438 if the attacker knows the passphrase is low-entropy (see last item), it
1439 is possible to make each individual try take longer. The way to do this
1440 is to repeatedly hash the passphrase for a certain time. The attacker
1441 then has to spend the same time (given the same computing power) as the
1442 user per try. With LUKS1, the default is 1 second of PBKDF2 hashing.
1444 Example 1: Lets assume we have a really bad passphrase (e.g. a
1445 girlfriends name) with 10 bits of entropy. With the same CPU, an
1446 attacker would need to spend around 500 seconds on average to break that
1447 passphrase. Without iteration, it would be more like 0.0001 seconds on
1450 Example 2: The user did a bit better and has 32 chars of English text.
1451 That would be about 32 bits of entropy. With 1 second iteration, that
1452 means an attacker on the same CPU needs around 136 years. That is
1453 pretty impressive for such a weak passphrase. Without the iterations,
1454 it would be more like 50 days on a modern CPU, and possibly far less.
1456 In addition, the attacker can both parallelize and use special hardware
1457 like GPUs or FPGAs to speed up the attack. The attack can also happen
1458 quite some time after the luksFormat operation and CPUs can have become
1459 faster and cheaper. For that reason you want a bit of extra security.
1460 Anyways, in Example 1 your are screwed. In example 2, not necessarily.
1461 Even if the attack is faster, it still has a certain cost associated
1462 with it, say 10000 EUR/USD with iteration and 1 EUR/USD without
1463 iteration. The first can be prohibitively expensive, while the second
1464 is something you try even without solid proof that the decryption will
1465 yield something useful.
1467 The numbers above are mostly made up, but show the idea. Of course the
1468 best thing is to have a high-entropy passphrase.
1470 Would a 100 sec iteration time be even better? Yes and no.
1471 Cryptographically it would be a lot better, namely 100 times better.
1472 However, usability is a very important factor for security technology
1473 and one that gets overlooked surprisingly often. For LUKS, if you have
1474 to wait 2 minutes to unlock the LUKS container, most people will not
1475 bother and use less secure storage instead. It is better to have less
1476 protection against low-entropy passphrases and people actually use LUKS,
1477 than having them do without encryption altogether.
1479 Now, what about decreasing the iteration time? This is generally a very
1480 bad idea, unless you know and can enforce that the users only use
1481 high-entropy passphrases. If you decrease the iteration time without
1482 ensuring that, then you put your users at increased risk, and
1483 considering how rarely LUKS containers are unlocked in a typical
1484 work-flow, you do so without a good reason. Don't do it. The iteration
1485 time is already low enough that users with low entropy passphrases are
1486 vulnerable. Lowering it even further increases this danger
1489 LUKS2: Pretty much the same reasoning applies. The advantages of using
1490 GPUs or FPGAs in an attack have been significantly reduced, but that
1491 is the only main difference.
1494 * 5.11 Some people say PBKDF2 is insecure?
1496 There is some discussion that a hash-function should have a "large
1497 memory" property, i.e. that it should require a lot of memory to be
1498 computed. This serves to prevent attacks using special programmable
1499 circuits, like FPGAs, and attacks using graphics cards. PBKDF2 does not
1500 need a lot of memory and is vulnerable to these attacks. However, the
1501 publication usually referred in these discussions is not very convincing
1502 in proving that the presented hash really is "large memory" (that may
1503 change, email the FAQ maintainer when it does) and it is of limited
1504 usefulness anyways. Attackers that use clusters of normal PCs will not
1505 be affected at all by a "large memory" property. For example the US
1506 Secret Service is known to use the off-hour time of all the office PCs
1507 of the Treasury for password breaking. The Treasury has about 110'000
1508 employees. Assuming every one has an office PC, that is significant
1509 computing power, all of it with plenty of memory for computing "large
1510 memory" hashes. Bot-net operators also have all the memory they want.
1511 The only protection against a resourceful attacker is a high-entropy
1512 passphrase, see items 5.9 and 5.10.
1514 That said, LUKS2 defaults to Argon2, which has a large-memory property
1515 and massively reduces the advantages of GPUs and FPGAs.
1518 * 5.12 What about iteration count with plain dm-crypt?
1520 Simple: There is none. There is also no salting. If you use plain
1521 dm-crypt, the only way to be secure is to use a high entropy passphrase.
1522 If in doubt, use LUKS instead.
1525 * 5.13 Is LUKS with default parameters less secure on a slow CPU?
1527 Unfortunately, yes. However the only aspect affected is the protection
1528 for low-entropy passphrase or master-key. All other security aspects
1529 are independent of CPU speed.
1531 The master key is less critical, as you really have to work at it to
1532 give it low entropy. One possibility to mess this up is to supply the
1533 master key yourself. If that key is low-entropy, then you get what you
1534 deserve. The other known possibility to create a LUKS container with a
1535 bad master key is to use /dev/urandom for key generation in an
1536 entropy-starved situation (e.g. automatic installation on an embedded
1537 device without network and other entropy sources or installation in a VM
1538 under certain circumstances).
1540 For the passphrase, don't use a low-entropy passphrase. If your
1541 passphrase is good, then a slow CPU will not matter. If you insist on a
1542 low-entropy passphrase on a slow CPU, use something like
1543 "--iter-time=10000" or higher and wait a long time on each LUKS unlock
1544 and pray that the attacker does not find out in which way exactly your
1545 passphrase is low entropy. This also applies to low-entropy passphrases
1546 on fast CPUs. Technology can do only so much to compensate for problems
1547 in front of the keyboard.
1549 Also note that power-saving modes will make your CPU slower. This will
1550 reduce iteration count on LUKS container creation. It will keep unlock
1551 times at the expected values though at this CPU speed.
1554 * 5.14 Why was the default aes-cbc-plain replaced with aes-cbc-essiv?
1556 Note: This item applies both to plain dm-crypt and to LUKS
1558 The problem is that cbc-plain has a fingerprint vulnerability, where a
1559 specially crafted file placed into the crypto-container can be
1560 recognized from the outside. The issue here is that for cbc-plain the
1561 initialization vector (IV) is the sector number. The IV gets XORed to
1562 the first data chunk of the sector to be encrypted. If you make sure
1563 that the first data block to be stored in a sector contains the sector
1564 number as well, the first data block to be encrypted is all zeros and
1565 always encrypted to the same ciphertext. This also works if the first
1566 data chunk just has a constant XOR with the sector number. By having
1567 several shifted patterns you can take care of the case of a
1568 non-power-of-two start sector number of the file.
1570 This mechanism allows you to create a pattern of sectors that have the
1571 same first ciphertext block and signal one bit per sector to the
1572 outside, allowing you to e.g. mark media files that way for recognition
1573 without decryption. For large files this is a practical attack. For
1574 small ones, you do not have enough blocks to signal and take care of
1575 different file starting offsets.
1577 In order to prevent this attack, the default was changed to cbc-essiv.
1578 ESSIV uses a keyed hash of the sector number, with the encryption key as
1579 key. This makes the IV unpredictable without knowing the encryption key
1580 and the watermarking attack fails.
1583 * 5.15 Are there any problems with "plain" IV? What is "plain64"?
1585 First, "plain" and "plain64" are both not secure to use with CBC, see
1588 However there are modes, like XTS, that are secure with "plain" IV. The
1589 next limit is that "plain" is 64 bit, with the upper 32 bit set to zero.
1590 This means that on volumes larger than 2TiB, the IV repeats, creating a
1591 vulnerability that potentially leaks some data. To avoid this, use
1592 "plain64", which uses the full sector number up to 64 bit. Note that
1593 "plain64" requires a kernel 2.6.33 or more recent. Also note that
1594 "plain64" is backwards compatible for volume sizes of maximum size 2TiB,
1595 but not for those > 2TiB. Finally, "plain64" does not cause any
1596 performance penalty compared to "plain".
1599 * 5.16 What about XTS mode?
1601 XTS mode is potentially even more secure than cbc-essiv (but only if
1602 cbc-essiv is insecure in your scenario). It is a NIST standard and
1603 used, e.g. in Truecrypt. From version 1.6.0 of cryptsetup onwards,
1604 aes-xts-plain64 is the default for LUKS. If you want to use it with a
1605 cryptsetup before version 1.6.0 or with plain dm-crypt, you have to
1606 specify it manually as "aes-xts-plain", i.e.
1608 cryptsetup -c aes-xts-plain luksFormat <device>
1610 For volumes >2TiB and kernels >= 2.6.33 use "plain64" (see FAQ item
1611 on "plain" and "plain64"):
1613 cryptsetup -c aes-xts-plain64 luksFormat <device>
1615 There is a potential security issue with XTS mode and large blocks.
1616 LUKS and dm-crypt always use 512B blocks and the issue does not apply.
1619 * 5.17 Is LUKS FIPS-140-2 certified?
1621 No. But that is more a problem of FIPS-140-2 than of LUKS. From a
1622 technical point-of-view, LUKS with the right parameters would be
1623 FIPS-140-2 compliant, but in order to make it certified, somebody has to
1624 pay real money for that. And then, whenever cryptsetup is changed or
1625 extended, the certification lapses and has to be obtained again.
1627 From the aspect of actual security, LUKS with default parameters should
1628 be as good as most things that are FIPS-140-2 certified, although you
1629 may want to make sure to use /dev/random (by specifying --use-random on
1630 luksFormat) as randomness source for the master key to avoid being
1631 potentially insecure in an entropy-starved situation.
1634 * 5.18 What about Plausible Deniability?
1636 First let me attempt a definition for the case of encrypted filesystems:
1637 Plausible deniability is when you store data inside an encrypted
1638 container and it is not possible to prove it is there without having a
1639 special passphrase. And at the same time it must be "plausible" that
1640 there actually is no hidden data there.
1642 As a simple entropy-analysis will show that here may be data there, the
1643 second part is what makes it tricky.
1645 There seem to be a lot of misunderstandings about this idea, so let me
1646 make it clear that this refers to the situation where the attackers can
1647 prove that there is data that either may be random or may be part of a
1648 plausible-deniability scheme, they just cannot prove which one it is.
1649 Hence a plausible-deniability scheme must hold up when the attackers
1650 know there is something potentially fishy. If you just hide data and
1651 rely on it not being found, that is just simple deniability, not
1652 "plausible" deniability and I am not talking about that in the
1653 following. Simple deniability against a low-competence attacker may be
1654 as simple as renaming a file or putting data into an unused part of a
1655 disk. Simple deniability against a high-skill attacker with time to
1656 invest is usually pointless unless you go for advanced steganographic
1657 techniques, which have their own drawbacks, such as low data capacity.
1659 Now, the idea of plausible deniability is compelling and on a first
1660 glance it seems possible to do it. And from a cryptographic point of
1661 view, it actually is possible.
1663 So, does the idea work in practice? No, unfortunately. The reasoning
1664 used by its proponents is fundamentally flawed in several ways and the
1665 cryptographic properties fail fatally when colliding with the real
1668 First, why should "I do not have a hidden partition" be any more
1669 plausible than "I forgot my crypto key" or "I wiped that partition with
1670 random data, nothing in there"? I do not see any reason.
1672 Second, there are two types of situations: Either they cannot force you
1673 to give them the key (then you simply do not) or they can. In the
1674 second case, they can always do bad things to you, because they cannot
1675 prove that you have the key in the first place! This means they do not
1676 have to prove you have the key, or that this random looking data on your
1677 disk is actually encrypted data. So the situation will allow them to
1678 waterboard/lock-up/deport you anyways, regardless of how "plausible"
1679 your deniability is. Do not have a hidden partition you could show to
1680 them, but there are indications you may? Too bad for you.
1681 Unfortunately "plausible deniability" also means you cannot prove there
1684 Third, hidden partitions are not that hidden. There are basically just
1685 two possibilities: a) Make a large crypto container, but put a smaller
1686 filesystem in there and put the hidden partition into the free space.
1687 Unfortunately this is glaringly obvious and can be detected in an
1688 automated fashion. This means that the initial suspicion to put you
1689 under duress in order to make you reveal your hidden data is given. b)
1690 Make a filesystem that spans the whole encrypted partition, and put the
1691 hidden partition into space not currently used by that filesystem.
1692 Unfortunately that is also glaringly obvious, as you then cannot write
1693 to the filesystem without a high risk of destroying data in the hidden
1694 container. Have not written anything to the encrypted filesystem in a
1695 while? Too bad, they have the suspicion they need to do unpleasant
1698 To be fair, if you prepare option b) carefully and directly before going
1699 into danger, it may work. But then, the mere presence of encrypted data
1700 may already be enough to get you into trouble in those places were they
1701 can demand encryption keys.
1703 Here is an additional reference for some problems with plausible
1704 deniability: http://www.schneier.com/paper-truecrypt-dfs.pdf I strongly
1705 suggest you read it.
1707 So, no, I will not provide any instructions on how to do it with plain
1708 dm-crypt or LUKS. If you insist on shooting yourself in the foot, you
1709 can figure out how to do it yourself.
1712 * 5.19 What about SSDs, Flash, Hybrid and SMR Drives?
1714 The problem is that you cannot reliably erase parts of these devices,
1715 mainly due to wear-leveling and possibly defect management and delayed
1716 writes to the main data area.
1718 For example for SSDs, when overwriting a sector, what the device does is
1719 to move an internal sector (may be 128kB or even larger) to some pool of
1720 discarded, not-yet erased unused sectors, take a fresh empty sector from
1721 the empty-sector pool and copy the old sector over with the changes to
1722 the small part you wrote. This is done in some fashion so that larger
1723 writes do not cause a lot of small internal updates.
1725 The thing is that the mappings between outside-addressable sectors and
1726 inside sectors is arbitrary (and the vendors are not talking). Also the
1727 discarded sectors are not necessarily erased immediately. They may
1730 For plain dm-crypt, the consequences are that older encrypted data may
1731 be lying around in some internal pools of the device. Thus may or may
1732 not be a problem and depends on the application. Remember the same can
1733 happen with a filesystem if consecutive writes to the same area of a
1734 file can go to different sectors.
1736 However, for LUKS, the worst case is that key-slots and LUKS header may
1737 end up in these internal pools. This means that password management
1738 functionality is compromised (the old passwords may still be around,
1739 potentially for a very long time) and that fast erase by overwriting the
1740 header and key-slot area is insecure.
1742 Also keep in mind that the discarded/used pool may be large. For
1743 example, a 240GB SSD has about 16GB of spare area in the chips that it
1744 is free to do with as it likes. You would need to make each individual
1745 key-slot larger than that to allow reliable overwriting. And that
1746 assumes the disk thinks all other space is in use. Reading the internal
1747 pools using forensic tools is not that hard, but may involve some
1752 If you trust the device vendor (you probably should not...) you can try
1753 an ATA "secure erase" command. That is not present in USB keys though
1754 and may or may not be secure for a hybrid drive.
1756 If you can do without password management and are fine with doing
1757 physical destruction for permanently deleting data (always after one or
1758 several full overwrites!), you can use plain dm-crypt.
1760 If you want or need all the original LUKS security features to work, you
1761 can use a detached LUKS header and put that on a conventional, magnetic
1762 disk. That leaves potentially old encrypted data in the pools on the
1763 main disk, but otherwise you get LUKS with the same security as on a
1764 traditional magnetic disk. Note however that storage vendors are prone
1765 to lying to their customers. For example, it recently came out that
1766 HDDs sold without any warning or mentioning in the data-sheets were
1767 actually using SMR and that will write data first to a faster area and
1768 only overwrite the original data area some time later when things are
1771 If you are concerned about your laptop being stolen, you are likely fine
1772 using LUKS on an SSD or hybrid drive. An attacker would need to have
1773 access to an old passphrase (and the key-slot for this old passphrase
1774 would actually need to still be somewhere in the SSD) for your data to
1775 be at risk. So unless you pasted your old passphrase all over the
1776 Internet or the attacker has knowledge of it from some other source and
1777 does a targeted laptop theft to get at your data, you should be fine.
1780 * 5.20 LUKS1 is broken! It uses SHA-1!
1782 No, it is not. SHA-1 is (academically) broken for finding collisions,
1783 but not for using it in a key-derivation function. And that collision
1784 vulnerability is for non-iterated use only. And you need the hash-value
1787 This basically means that if you already have a slot-key, and you have
1788 set the PBKDF2 iteration count to 1 (it is > 10'000 normally), you could
1789 (maybe) derive a different passphrase that gives you the the same
1790 slot-key. But if you have the slot-key, you can already unlock the
1791 key-slot and get the master key, breaking everything. So basically,
1792 this SHA-1 vulnerability allows you to open a LUKS1 container with high
1793 effort when you already have it open.
1795 The real problem here is people that do not understand crypto and claim
1796 things are broken just because some mechanism is used that has been
1797 broken for a specific different use. The way the mechanism is used
1798 matters very much. A hash that is broken for one use can be completely
1799 secure for other uses and here it is.
1801 Since version 1.7.0, cryptsetup uses SHA-256 as default to ensure that
1802 it will be compatible in the future. There are already some systems
1803 where SHA-1 is completely phased out or disabled by a security policy.
1806 * 5.21 Why is there no "Nuke-Option"?
1808 A "Nuke-Option" or "Kill-switch" is a password that when entered upon
1809 unlocking instead wipes the header and all passwords. So when somebody
1810 forces you to enter your password, you can destroy the data instead.
1812 While this sounds attractive at first glance, it does not make sense
1813 once a real security analysis is done. One problem is that you have to
1814 have some kind of HSM (Hardware Security Module) in order to implement
1815 it securely. In the movies, a HSM starts to smoke and melt once the
1816 Nuke-Option has been activated. In actual reality, it just wipes some
1817 battery-backed RAM cells. A proper HSM costs something like
1818 20'000...100'000 EUR/USD and there a Nuke-Option may make some sense.
1819 BTW, a chipcard or a TPM is not a HSM, although some vendors are
1820 promoting that myth.
1822 Now, a proper HSMs will have a wipe option but not a Nuke-Option, i.e.
1823 you can explicitly wipe the HSM, but by a different process than
1824 unlocking it takes. Why is that? Simple: If somebody can force you to
1825 reveal passwords, then they can also do bad things to you if you do not
1826 or if you enter a nuke password instead. Think locking you up for a few
1827 years for "destroying evidence" or for far longer and without trial for
1828 being a "terrorist suspect". No HSM maker will want to expose its
1829 customers to that risk.
1831 Now think of the typical LUKS application scenario, i.e. disk
1832 encryption. Usually the ones forcing you to hand over your password
1833 will have access to the disk as well, and, if they have any real
1834 suspicion, they will mirror your disk before entering anything supplied
1835 by you. This neatly negates any Nuke-Option. If they have no suspicion
1836 (just harassing people that cross some border for example), the
1837 Nuke-Option would work, but see above about likely negative consequences
1838 and remember that a Nuke-Option may not work reliably on SSD and hybrid
1841 Hence my advice is to never take data that you do not want to reveal
1842 into any such situation in the first place. There is no need to
1843 transfer data on physical carriers today. The Internet makes it quite
1844 possible to transfer data between arbitrary places and modern encryption
1845 makes it secure. If you do it right, nobody will even be able to
1846 identify source or destination. (How to do that is out of scope of this
1847 document. It does require advanced skills in this age of pervasive
1850 Hence, LUKS has not kill option because it would do much more harm than
1853 Still, if you have a good use-case (i.e. non-abstract real-world
1854 situation) where a Nuke-Option would actually be beneficial, please let
1858 * 5.22 Does cryptsetup open network connections to websites, etc. ?
1860 This question seems not to make much sense at first glance, but here is
1861 an example form the real world: The TrueCrypt GUI has a "Donation"
1862 button. Press it, and a web-connection to the TrueCrypt website is
1863 opened via the default browser, telling everybody that listens that you
1864 use TrueCrypt. In the worst case, things like this can get people
1867 So: Cryptsetup will never open any network connections except the
1868 local netlink socket it needs to talk to the kernel crypto API.
1870 In addition, the installation package should contain all documentation,
1871 including this FAQ, so that you do not have to go to a web-site to read
1872 it. (If your distro cuts the docu, please complain to them.) In
1873 security software, any connection initiated to anywhere outside your
1874 machine should always be the result of an explicit request for such a
1875 connection by the user and cryptsetup will stay true to that principle.
1878 6. Backup and Data Recovery
1881 * 6.1 Why do I need Backup?
1883 First, disks die. The rate for well-treated (!) disk is about 5% per
1884 year, which is high enough to worry about. There is some indication
1885 that this may be even worse for some SSDs. This applies both to LUKS
1886 and plain dm-crypt partitions.
1888 Second, for LUKS, if anything damages the LUKS header or the key-stripe
1889 area then decrypting the LUKS device can become impossible. This is a
1890 frequent occurrence. For example an accidental format as FAT or some
1891 software overwriting the first sector where it suspects a partition boot
1892 sector typically makes a LUKS1 partition permanently inaccessible. See
1893 more below on LUKS header damage.
1895 So, data-backup in some form is non-optional. For LUKS, you may also
1896 want to store a header backup in some secure location. This only needs
1897 an update if you change passphrases.
1900 * 6.2 How do I backup a LUKS header?
1902 While you could just copy the appropriate number of bytes from the start
1903 of the LUKS partition, the best way is to use command option
1904 "luksHeaderBackup" of cryptsetup. This protects also against errors
1905 when non-standard parameters have been used in LUKS partition creation.
1908 cryptsetup luksHeaderBackup --header-backup-file <file> <device>
1910 To restore, use the inverse command, i.e.
1912 cryptsetup luksHeaderRestore --header-backup-file <file> <device>
1914 If you are unsure about a header to be restored, make a backup of the
1915 current one first! You can also test the header-file without restoring
1916 it by using the --header option for a detached header like this:
1918 cryptsetup --header <file> luksOpen <device> </dev/mapper/name>
1920 If that unlocks your keys-lot, you are good. Do not forget to close
1923 Under some circumstances (damaged header), this fails. Then use the
1924 following steps in case it is LUKS1:
1926 First determine the master-key size:
1928 cryptsetup luksDump <device>
1930 gives a line of the form
1934 with bits equal to 256 for the old defaults and 512 for the new
1935 defaults. 256 bits equals a total header size of 1'052'672 Bytes and
1936 512 bits one of 2MiB. (See also Item 6.12) If luksDump fails, assume
1937 2MiB, but be aware that if you restore that, you may also restore the
1938 first 1M or so of the filesystem. Do not change the filesystem if you
1939 were unable to determine the header size! With that, restoring a
1940 too-large header backup is still safe.
1942 Second, dump the header to file. There are many ways to do it, I
1943 prefer the following:
1945 head -c 1052672 <device> > header_backup.dmp
1949 head -c 2M <device> > header_backup.dmp
1951 for a 2MiB header. Verify the size of the dump-file to be sure.
1953 To restore such a backup, you can try luksHeaderRestore or do a more
1956 cat header_backup.dmp > <device>
1960 * 6.3 How do I test for a LUKS header?
1964 cryptsetup -v isLuks <device>
1966 on the device. Without the "-v" it just signals its result via
1967 exit-status. You can also use the more general test
1971 which will also detect other types and give some more info. Omit
1972 "-p" for old versions of blkid that do not support it.
1975 * 6.4 How do I backup a LUKS or dm-crypt partition?
1977 There are two options, a sector-image and a plain file or filesystem
1978 backup of the contents of the partition. The sector image is already
1979 encrypted, but cannot be compressed and contains all empty space. The
1980 filesystem backup can be compressed, can contain only part of the
1981 encrypted device, but needs to be encrypted separately if so desired.
1983 A sector-image will contain the whole partition in encrypted form, for
1984 LUKS the LUKS header, the keys-slots and the data area. It can be done
1985 under Linux e.g. with dd_rescue (for a direct image copy) and with
1986 "cat" or "dd". Examples:
1988 cat /dev/sda10 > sda10.img
1989 dd_rescue /dev/sda10 sda10.img
1991 You can also use any other backup software that is capable of making a
1992 sector image of a partition. Note that compression is ineffective for
1993 encrypted data, hence it does not make sense to use it.
1995 For a filesystem backup, you decrypt and mount the encrypted partition
1996 and back it up as you would a normal filesystem. In this case the
1997 backup is not encrypted, unless your encryption method does that. For
1998 example you can encrypt a backup with "tar" as follows with GnuPG:
2000 tar cjf - <path> | gpg --cipher-algo AES -c - > backup.tbz2.gpg
2002 And verify the backup like this if you are at "path":
2004 cat backup.tbz2.gpg | gpg - | tar djf -
2006 Note: Always verify backups, especially encrypted ones!
2008 There is one problem with verifying like this: The kernel may still have
2009 some files cached and in fact verify them against RAM or may even verify
2010 RAM against RAM, which defeats the purpose of the exercise. The
2011 following command empties the kernel caches:
2013 echo 3 > /proc/sys/vm/drop_caches
2015 Run it after backup and before verify.
2017 In both cases GnuPG will ask you interactively for your symmetric key.
2018 The verify will only output errors. Use "tar dvjf -" to get all
2019 comparison results. To make sure no data is written to disk
2020 unencrypted, turn off swap if it is not encrypted before doing the
2023 Restore works like certification with the 'd' ('difference') replaced
2024 by 'x' ('eXtract'). Refer to the man-page of tar for more explanations
2025 and instructions. Note that with default options tar will overwrite
2026 already existing files without warning. If you are unsure about how
2027 to use tar, experiment with it in a location where you cannot do damage.
2029 You can of course use different or no compression and you can use an
2030 asymmetric key if you have one and have a backup of the secret key that
2033 A second option for a filesystem-level backup that can be used when the
2034 backup is also on local disk (e.g. an external USB drive) is to use a
2035 LUKS container there and copy the files to be backed up between both
2036 mounted containers. Also see next item.
2039 * 6.5 Do I need a backup of the full partition? Would the header
2040 and key-slots not be enough?
2042 Backup protects you against two things: Disk loss or corruption and user
2043 error. By far the most questions on the dm-crypt mailing list about how
2044 to recover a damaged LUKS partition are related to user error. For
2045 example, if you create a new filesystem on a non-mapped LUKS container,
2046 chances are good that all data is lost permanently.
2048 For this case, a header+key-slot backup would often be enough. But keep
2049 in mind that a well-treated (!) HDD has roughly a failure risk of 5% per
2050 year. It is highly advisable to have a complete backup to protect
2054 * 6.6 What do I need to backup if I use "decrypt_derived"?
2056 This is a script in Debian, intended for mounting /tmp or swap with a
2057 key derived from the master key of an already decrypted device. If you
2058 use this for an device with data that should be persistent, you need to
2059 make sure you either do not lose access to that master key or have a
2060 backup of the data. If you derive from a LUKS device, a header backup
2061 of that device would cover backing up the master key. Keep in mind that
2062 this does not protect against disk loss.
2064 Note: If you recreate the LUKS header of the device you derive from
2065 (using luksFormat), the master key changes even if you use the same
2066 passphrase(s) and you will not be able to decrypt the derived device
2067 with the new LUKS header.
2070 * 6.7 Does a backup compromise security?
2072 Depends on how you do it. However if you do not have one, you are going
2073 to eventually lose your encrypted data.
2075 There are risks introduced by backups. For example if you
2076 change/disable a key-slot in LUKS, a binary backup of the partition will
2077 still have the old key-slot. To deal with this, you have to be able to
2078 change the key-slot on the backup as well, securely erase the backup or
2079 do a filesystem-level backup instead of a binary one.
2081 If you use dm-crypt, backup is simpler: As there is no key management,
2082 the main risk is that you cannot wipe the backup when wiping the
2083 original. However wiping the original for dm-crypt should consist of
2084 forgetting the passphrase and that you can do without actual access to
2087 In both cases, there is an additional (usually small) risk with binary
2088 backups: An attacker can see how many sectors and which ones have been
2089 changed since the backup. To prevent this, use a filesystem level
2090 backup method that encrypts the whole backup in one go, e.g. as
2091 described above with tar and GnuPG.
2093 My personal advice is to use one USB disk (low value data) or three
2094 disks (high value data) in rotating order for backups, and either use
2095 independent LUKS partitions on them, or use encrypted backup with tar
2098 If you do network-backup or tape-backup, I strongly recommend to go
2099 the filesystem backup path with independent encryption, as you
2100 typically cannot reliably delete data in these scenarios, especially
2101 in a cloud setting. (Well, you can burn the tape if it is under your
2105 * 6.8 What happens if I overwrite the start of a LUKS partition or
2106 damage the LUKS header or key-slots?
2108 There are two critical components for decryption: The salt values in the
2109 key-slot descriptors of the header and the key-slots. For LUKS2 they
2110 are a bit better protected. but for LUKS1, these are right in the first
2111 sector. If the salt values are overwritten or changed, nothing (in the
2112 cryptographically strong sense) can be done to access the data, unless
2113 there is a backup of the LUKS header. If a key-slot is damaged, the
2114 data can still be read with a different key-slot, if there is a
2115 remaining undamaged and used key-slot. Note that in order to make a
2116 key-slot completely unrecoverable, changing about 4-6 bits in random
2117 locations of its 128kiB size is quite enough.
2120 * 6.9 What happens if I (quick) format a LUKS partition?
2122 I have not tried the different ways to do this, but very likely you will
2123 have written a new boot-sector, which in turn overwrites the LUKS
2124 header, including the salts, making your data permanently irretrievable,
2125 unless you have a LUKS header backup. For LUKS2 this may still be
2126 recoverable without that header backup, for LUKS1 it is not. You may
2127 also damage the key-slots in part or in full. See also last item.
2130 * 6.10 How do I recover the master key from a mapped LUKS1 container?
2132 Note: LUKS2 uses the kernel keyring to store keys and hence this
2133 procedure does not work unless you have explicitly disabled the use of
2134 the keyring with "--disable-keyring" on opening.
2136 This is typically only needed if you managed to damage your LUKS1
2137 header, but the container is still mapped, i.e. "luksOpen"ed. It also
2138 helps if you have a mapped container that you forgot or do not know a
2139 passphrase for (e.g. on a long running server.)
2141 WARNING: Things go wrong, do a full backup before trying this!
2143 WARNING: This exposes the master key of the LUKS1 container. Note that
2144 both ways to recreate a LUKS header with the old master key described
2145 below will write the master key to disk. Unless you are sure you have
2146 securely erased it afterwards, e.g. by writing it to an encrypted
2147 partition, RAM disk or by erasing the filesystem you wrote it to by a
2148 complete overwrite, you should change the master key afterwards.
2149 Changing the master key requires a full data backup, luksFormat and then
2150 restore of the backup. Alternatively the tool cryptsetup-reencrypt from
2151 the cryptsetup package can be used to change the master key (see its
2152 man-page), but a full backup is still highly recommended.
2154 First, there is a script by Milan that automates the whole process,
2155 except generating a new LUKS1 header with the old master key (it prints
2156 the command for that though):
2158 https://gitlab.com/cryptsetup/cryptsetup/blob/master/misc/luks-header-from-active
2160 You can also do this manually. Here is how:
2162 - Get the master key from the device mapper. This is done by the
2163 following command. Substitute c5 for whatever you mapped to:
2165 # dmsetup table --target crypt --showkey /dev/mapper/c5
2168 0 200704 crypt aes-cbc-essiv:sha256
2169 a1704d9715f73a1bb4db581dcacadaf405e700d591e93e2eaade13ba653d0d09
2172 The result is actually one line, wrapped here for clarity. The long
2173 hex string is the master key.
2175 - Convert the master key to a binary file representation. You can do
2176 this manually, e.g. with hexedit. You can also use the tool "xxd"
2179 echo "a1704d9....53d0d09" | xxd -r -p > <master-key-file>
2182 - Do a luksFormat to create a new LUKS1 header.
2184 NOTE: If your header is intact and you just forgot the passphrase,
2185 you can just set a new passphrase, see next sub-item.
2187 Unmap the device before you do that (luksClose). Then do
2189 cryptsetup luksFormat --master-key-file=<master-key-file> <luks device>
2191 Note that if the container was created with other than the default
2192 settings of the cryptsetup version you are using, you need to give
2193 additional parameters specifying the deviations. If in doubt, try the
2194 script by Milan. It does recover the other parameters as well.
2196 Side note: This is the way the decrypt_derived script gets at the master
2197 key. It just omits the conversion and hashes the master key string.
2199 - If the header is intact and you just forgot the passphrase, just
2200 set a new passphrase like this:
2202 cryptsetup luksAddKey --master-key-file=<master-key-file> <luks device>
2204 You may want to disable the old one afterwards.
2207 * 6.11 What does the on-disk structure of dm-crypt look like?
2209 There is none. dm-crypt takes a block device and gives encrypted access
2210 to each of its blocks with a key derived from the passphrase given. If
2211 you use a cipher different than the default, you have to specify that as
2212 a parameter to cryptsetup too. If you want to change the password, you
2213 basically have to create a second encrypted device with the new
2214 passphrase and copy your data over. On the plus side, if you
2215 accidentally overwrite any part of a dm-crypt device, the damage will be
2216 limited to the area you overwrote.
2219 * 6.12 What does the on-disk structure of LUKS1 look like?
2221 Note: For LUKS2, refer to the LUKS2 document referenced in Item 1.2
2223 A LUKS1 partition consists of a header, followed by 8 key-slot
2224 descriptors, followed by 8 key slots, followed by the encrypted data
2227 Header and key-slot descriptors fill the first 592 bytes. The key-slot
2228 size depends on the creation parameters, namely on the number of
2229 anti-forensic stripes, key material offset and master key size.
2231 With the default parameters, each key-slot is a bit less than 128kiB in
2232 size. Due to sector alignment of the key-slot start, that means the key
2233 block 0 is at offset 0x1000-0x20400, key block 1 at offset
2234 0x21000-0x40400, and key block 7 at offset 0xc1000-0xe0400. The space
2235 to the next full sector address is padded with zeros. Never used
2236 key-slots are filled with what the disk originally contained there, a
2237 key-slot removed with "luksRemoveKey" or "luksKillSlot" gets filled with
2238 0xff. Due to 2MiB default alignment, start of the data area for
2239 cryptsetup 1.3 and later is at 2MiB, i.e. at 0x200000. For older
2240 versions, it is at 0x101000, i.e. at 1'052'672 bytes, i.e. at 1MiB +
2241 4096 bytes from the start of the partition. Incidentally,
2242 "luksHeaderBackup" for a LUKS container created with default parameters
2243 dumps exactly the first 2MiB (or 1'052'672 bytes for headers created
2244 with cryptsetup versions < 1.3) to file and "luksHeaderRestore" restores
2247 For non-default parameters, you have to figure out placement yourself.
2248 "luksDump" helps. See also next item. For the most common non-default
2249 settings, namely aes-xts-plain with 512 bit key, the offsets are: 1st
2250 keyslot 0x1000-0x3f800, 2nd keyslot 0x40000-0x7e000, 3rd keyslot
2251 0x7e000-0xbd800, ..., and start of bulk data at 0x200000.
2253 The exact specification of the format is here:
2254 https://gitlab.com/cryptsetup/cryptsetup/wikis/Specification
2256 For your convenience, here is the LUKS1 header with hex offsets.
2258 The spec counts key-slots from 1 to 8, but the cryptsetup tool counts
2259 from 0 to 7. The numbers here refer to the cryptsetup numbers.
2262 Refers to LUKS1 On-Disk Format Specification Version 1.2.3
2266 offset length name data type description
2267 -----------------------------------------------------------------------
2268 0x0000 0x06 magic byte[] 'L','U','K','S', 0xba, 0xbe
2270 0x0006 0x02 version uint16_t LUKS version
2272 0x0008 0x20 cipher-name char[] cipher name spec.
2274 0x0028 0x20 cipher-mode char[] cipher mode spec.
2276 0x0048 0x20 hash-spec char[] hash spec.
2278 0x0068 0x04 payload-offset uint32_t bulk data offset in sectors
2279 104 4 (512 bytes per sector)
2280 0x006c 0x04 key-bytes uint32_t number of bytes in key
2282 0x0070 0x14 mk-digest byte[] master key checksum
2283 112 20 calculated with PBKDF2
2284 0x0084 0x20 mk-digest-salt byte[] salt for PBKDF2 when
2285 132 32 calculating mk-digest
2286 0x00a4 0x04 mk-digest-iter uint32_t iteration count for PBKDF2
2287 164 4 when calculating mk-digest
2288 0x00a8 0x28 uuid char[] partition UUID
2290 0x00d0 0x30 key-slot-0 key slot key slot 0
2292 0x0100 0x30 key-slot-1 key slot key slot 1
2294 0x0130 0x30 key-slot-2 key slot key slot 2
2296 0x0160 0x30 key-slot-3 key slot key slot 3
2298 0x0190 0x30 key-slot-4 key slot key slot 4
2300 0x01c0 0x30 key-slot-5 key slot key slot 5
2302 0x01f0 0x30 key-slot-6 key slot key slot 6
2304 0x0220 0x30 key-slot-7 key slot key slot 7
2310 offset length name data type description
2311 -------------------------------------------------------------------------
2312 0x0000 0x04 active uint32_t key slot enabled/disabled
2314 0x0004 0x04 iterations uint32_t PBKDF2 iteration count
2316 0x0008 0x20 salt byte[] PBKDF2 salt
2318 0x0028 0x04 key-material-offset uint32_t key start sector
2319 40 4 (512 bytes/sector)
2320 0x002c 0x04 stripes uint32_t number of anti-forensic
2325 * 6.13 What is the smallest possible LUKS1 container?
2327 Note: From cryptsetup 1.3 onwards, alignment is set to 1MB. With modern
2328 Linux partitioning tools that also align to 1MB, this will result in
2329 alignment to 2k sectors and typical Flash/SSD sectors, which is highly
2330 desirable for a number of reasons. Changing the alignment is not
2333 That said, with default parameters, the data area starts at exactly 2MB
2334 offset (at 0x101000 for cryptsetup versions before 1.3). The smallest
2335 data area you can have is one sector of 512 bytes. Data areas of 0
2336 bytes can be created, but fail on mapping.
2338 While you cannot put a filesystem into something this small, it may
2339 still be used to contain, for example, key. Note that with current
2340 formatting tools, a partition for a container this size will be 3MiB
2341 anyways. If you put the LUKS container into a file (via losetup and a
2342 loopback device), the file needs to be 2097664 bytes in size, i.e. 2MiB
2345 The two ways to influence the start of the data area are key-size and
2348 For alignment, you can go down to 1 on the parameter. This will still
2349 leave you with a data-area starting at 0x101000, i.e. 1MiB+4096B
2350 (default parameters) as alignment will be rounded up to the next
2351 multiple of 8 (i.e. 4096 bytes) If in doubt, do a dry-run on a larger
2352 file and dump the LUKS header to get actual information.
2354 For key-size, you can use 128 bit (e.g. AES-128 with CBC), 256 bit
2355 (e.g. AES-256 with CBC) or 512 bit (e.g. AES-256 with XTS mode). You
2356 can do 64 bit (e.g. blowfish-64 with CBC), but anything below 128 bit
2357 has to be considered insecure today.
2359 Example 1 - AES 128 bit with CBC:
2361 cryptsetup luksFormat -s 128 --align-payload=8 <device>
2363 This results in a data offset of 0x81000, i.e. 516KiB or 528384
2364 bytes. Add one 512 byte sector and the smallest LUKS container size
2365 with these parameters is 516KiB + 512B or 528896 bytes.
2367 Example 2 - Blowfish 64 bit with CBC (WARNING: insecure):
2369 cryptsetup luksFormat -c blowfish -s 64 --align-payload=8 /dev/loop0
2371 This results in a data offset of 0x41000, i.e. 260kiB or 266240
2372 bytes, with a minimal LUKS1 container size of 260kiB + 512B or 266752
2376 * 6.14 I think this is overly complicated. Is there an alternative?
2378 Not really. Encryption comes at a price. You can use plain dm-crypt to
2379 simplify things a bit. It does not allow multiple passphrases, but on
2380 the plus side, it has zero on disk description and if you overwrite some
2381 part of a plain dm-crypt partition, exactly the overwritten parts are
2382 lost (rounded up to full sectors).
2384 * 6.15 Can I clone a LUKS container?
2386 You can, but it breaks security, because the cloned container has the
2387 same header and hence the same master key. Even if you change the
2388 passphrase(s), the master key stays the same. That means whoever has
2389 access to one of the clones can decrypt them all, completely bypassing
2392 While you can use cryptsetup-reencrypt to change the master key,
2393 this is probably more effort than to create separate LUKS containers
2396 The right way to do this is to first luksFormat the target container,
2397 then to clone the contents of the source container, with both containers
2398 mapped, i.e. decrypted. You can clone the decrypted contents of a LUKS
2399 container in binary mode, although you may run into secondary issues
2400 with GUIDs in filesystems, partition tables, RAID-components and the
2401 like. These are just the normal problems binary cloning causes.
2403 Note that if you need to ship (e.g.) cloned LUKS containers with a
2404 default passphrase, that is fine as long as each container was
2405 individually created (and hence has its own master key). In this case,
2406 changing the default passphrase will make it secure again.
2409 7. Interoperability with other Disk Encryption Tools
2412 * 7.1 What is this section about?
2414 Cryptsetup for plain dm-crypt can be used to access a number of on-disk
2415 formats created by tools like loop-aes patched into losetup. This
2416 sometimes works and sometimes does not. This section collects insights
2417 into what works, what does not and where more information is required.
2419 Additional information may be found in the mailing-list archives,
2420 mentioned at the start of this FAQ document. If you have a solution
2421 working that is not yet documented here and think a wider audience may
2422 be interested, please email the FAQ maintainer.
2425 * 7.2 loop-aes: General observations.
2427 One problem is that there are different versions of losetup around.
2428 loop-aes is a patch for losetup. Possible problems and deviations
2429 from cryptsetup option syntax include:
2431 - Offsets specified in bytes (cryptsetup: 512 byte sectors)
2433 - The need to specify an IV offset
2435 - Encryption mode needs specifying (e.g. "-c twofish-cbc-plain")
2437 - Key size needs specifying (e.g. "-s 128" for 128 bit keys)
2439 - Passphrase hash algorithm needs specifying
2441 Also note that because plain dm-crypt and loop-aes format does not have
2442 metadata, and while the loopAES extension for cryptsetup tries
2443 autodetection (see command loopaesOpen), it may not always work. If you
2444 still have the old set-up, using a verbosity option (-v) on mapping with
2445 the old tool or having a look into the system logs after setup could
2446 give you the information you need. Below, there are also some things
2447 that worked for somebody.
2450 * 7.3 loop-aes patched into losetup on Debian 5.x, kernel 2.6.32
2452 In this case, the main problem seems to be that this variant of
2453 losetup takes the offset (-o option) in bytes, while cryptsetup takes
2454 it in sectors of 512 bytes each.
2456 Example: The losetup command
2458 losetup -e twofish -o 2560 /dev/loop0 /dev/sdb1
2459 mount /dev/loop0 mount-point
2463 cryptsetup create -c twofish -o 5 --skip 5 e1 /dev/sdb1
2464 mount /dev/mapper/e1 mount-point
2468 * 7.4 loop-aes with 160 bit key
2470 This seems to be sometimes used with twofish and blowfish and represents
2471 a 160 bit ripemed160 hash output padded to 196 bit key length. It seems
2472 the corresponding options for cryptsetup are
2474 --cipher twofish-cbc-null -s 192 -h ripemd160:20
2478 * 7.5 loop-aes v1 format OpenSUSE
2480 Apparently this is done by older OpenSUSE distros and stopped working
2481 from OpenSUSE 12.1 to 12.2. One user had success with the following:
2483 cryptsetup create <target> <device> -c aes -s 128 -h sha256
2487 * 7.6 Kernel encrypted loop device (cryptoloop)
2489 There are a number of different losetup implementations for using
2490 encrypted loop devices so getting this to work may need a bit of
2493 NOTE: Do NOT use this for new containers! Some of the existing
2494 implementations are insecure and future support is uncertain.
2496 Example for a compatible mapping:
2498 losetup -e twofish -N /dev/loop0 /image.img
2502 cryptsetup create image_plain /image.img -c twofish-cbc-plain -H plain
2504 with the mapping being done to /dev/mapper/image_plain instead of
2509 Cipher, mode and password hash (or no hash):
2511 -e cipher [-N] => -c cipher-cbc-plain -H plain [-s 256]
2512 -e cipher => -c cipher-cbc-plain -H ripemd160 [-s 256]
2515 Key size and offsets (losetup: bytes, cryptsetuop: sectors of 512 bytes):
2518 -o 2560 => -o 5 -p 5 # 2560/512 = 5
2521 There is no replacement for --pass-fd, it has to be emulated using
2522 keyfiles, see the cryptsetup man-page.
2525 8. Issues with Specific Versions of cryptsetup
2528 * 8.1 When using the create command for plain dm-crypt with
2529 cryptsetup 1.1.x, the mapping is incompatible and my data is not
2532 With cryptsetup 1.1.x, the distro maintainer can define different
2533 default encryption modes. You can check the compiled-in defaults using
2534 "cryptsetup --help". Moreover, the plain device default changed because
2535 the old IV mode was vulnerable to a watermarking attack.
2537 If you are using a plain device and you need a compatible mode, just
2538 specify cipher, key size and hash algorithm explicitly. For
2539 compatibility with cryptsetup 1.0.x defaults, simple use the following:
2541 cryptsetup create -c aes-cbc-plain -s 256 -h ripemd160 <name> <dev>
2543 LUKS stores cipher and mode in the metadata on disk, avoiding this
2547 * 8.2 cryptsetup on SLED 10 has problems...
2549 SLED 10 is missing an essential kernel patch for dm-crypt, which is
2550 broken in its kernel as a result. There may be a very old version of
2551 cryptsetup (1.0.x) provided by SLED, which should also not be used
2552 anymore as well. My advice would be to drop SLED 10.
2555 * 8.3 Gcrypt 1.6.x and later break Whirlpool
2557 It is the other way round: In gcrypt 1.5.x, Whirlpool is broken and it
2558 was fixed in 1.6.0 and later. If you selected whirlpool as hash on
2559 creation of a LUKS container, it does not work anymore with the fixed
2560 library. This shows one serious risk of using rarely used settings.
2562 Note that at the time this FAQ item was written, 1.5.4 was the latest
2563 1.5.x version and it has the flaw, i.e. works with the old Whirlpool
2564 version. Possibly later 1.5.x versions will work as well. If not,
2567 The only two ways to access older LUKS containers created with Whirlpool
2568 are to either decrypt with an old gcrypt version that has the flaw or to
2569 use a compatibility feature introduced in cryptsetup 1.6.4 and gcrypt
2570 1.6.1 or later. Version 1.6.0 cannot be used.
2574 - Make at least a header backup or better, refresh your full backup.
2575 (You have a full backup, right? See Item 6.1 and following.)
2577 - Make sure you have cryptsetup 1.6.4 or later and check the gcrypt
2580 cryptsetup luksDump <your luks device> --debug | grep backend
2582 If gcrypt is at version 1.5.x or before:
2584 - Reencrypt the LUKS header with a different hash. (Requires entering
2585 all keyslot passphrases. If you do not have all, remove the ones you
2586 do not have before.):
2588 cryptsetup-reencrypt --keep-key --hash sha256 <your luks device>
2590 If gcrypt is at version 1.6.1 or later:
2592 - Patch the hash name in the LUKS header from "whirlpool" to
2593 "whirlpool_gcryptbug". This activates the broken implementation.
2594 The detailed header layout is in Item 6.12 of this FAQ and in the
2595 LUKS on-disk format specification. One way to change the hash is
2596 with the following command:
2598 echo -n -e 'whirlpool_gcryptbug\0' | dd of=<luks device> bs=1 seek=72 conv=notrunc
2600 - You can now open the device again. It is highly advisable to change
2601 the hash now with cryptsetup-reencrypt as described above. While you
2602 can reencrypt to use the fixed whirlpool, that may not be a good idea
2603 as almost nobody seems to use it and hence the long time until the
2607 9. The Initrd question
2610 * 9.1 My initrd is broken with cryptsetup
2612 That is not nice! However the initrd is supplied by your distribution,
2613 not by the cryptsetup project and hence you should complain to them. We
2614 cannot really do anything about it.
2617 * 9.2 CVE-2016-4484 says cryptsetup is broken!
2619 Not really. It says the initrd in some Debian versions have a behavior
2620 that under some very special and unusual conditions may be considered
2623 What happens is that you can trick the initrd to go to a rescue-shell if
2624 you enter the LUKS password wrongly in a specific way. But falling back
2625 to a rescue shell on initrd errors is a sensible default behavior in the
2626 first place. It gives you about as much access as booting a rescue
2627 system from CD or USB-Stick or as removing the disk would give you. So
2628 this only applies when an attacker has physical access, but cannot boot
2629 anything else or remove the disk. These will be rare circumstances
2630 indeed, and if you rely on the default distribution initrd to keep you
2631 safe under these circumstances, then you have bigger problems than this
2632 somewhat expected behavior.
2634 The CVE was exagerrated and should not be assigned to upstream
2635 cryptsetup in the first place (it is a distro specific initrd issue).
2636 It was driven more by a try to make a splash for self-aggrandizement,
2637 than by any actual security concerns. Ignore it.
2640 * 9.3 How do I do my own initrd with cryptsetup?
2642 Note: The instructions here apply to an initrd in initramfs format, not
2643 to an initrd in initrd format. The latter is a filesystem image, not a
2644 cpio-archive, and seems to not be widely used anymore.
2646 It depends on the distribution. Below, I give a very simple example and
2647 step-by-step instructions for Debian. With a bit of work, it should be
2648 possible to adapt this to other distributions. Note that the
2649 description is pretty general, so if you want to do other things with an
2650 initrd it provides a useful starting point for that too.
2652 01) Unpacking an existing initrd to use as template
2654 A Linux initrd is in gzip'ed cpio format. To unpack it, use something
2657 md tmp; cd tmp; cat ../initrd | gunzip | cpio -id
2659 After this, you have the full initrd content in tmp/
2661 02) Inspecting the init-script
2663 The init-script is the only thing the kernel cares about. All activity
2664 starts there. Its traditional location is /sbin/init on disk, but /init
2665 in an initrd. In an initrd unpacked as above it is tmp/init.
2667 While init can be a binary despite usually being called "init script",
2668 in Debian the main init on the root partition is a binary, but the init
2669 in the initrd (and only that one is called by the kernel) is a script
2670 and starts like this:
2675 The "sh" used here is in tmp/bin/sh as just unpacked, and in Debian it
2676 currently is a busybox.
2678 03) Creating your own initrd
2680 The two examples below should give you most of what is needed. This is
2681 tested with LUKS1 and should work with LUKS2 as well. If not, please
2684 Here is a really minimal example. It does nothing but set up some
2685 things and then drop to an interactive shell. It is perfect to try out
2686 things that you want to go into the init-script.
2689 export PATH=/sbin:/bin
2690 [ -d /sys ] || mkdir /sys
2691 [ -d /proc ] || mkdir /proc
2692 [ -d /tmp ] || mkdir /tmp
2693 mount -t sysfs -o nodev,noexec,nosuid sysfs /sys
2694 mount -t proc -o nodev,noexec,nosuid proc /proc
2695 echo "initrd is running, starting BusyBox..."
2696 exec /bin/sh --login
2699 Here is an example that opens the first LUKS-partition it finds with the
2700 hard-coded password "test2" and then mounts it as root-filesystem. This
2701 is intended to be used on an USB-stick that after boot goes into a safe,
2702 as it contains the LUKS-passphrase in plain text and is not secure to be
2703 left in the system. The script contains debug-output that should make it
2704 easier to see what is going on. Note that the final hand-over to the init
2705 on the encrypted root-partition is done by "exec switch_root /mnt/root
2706 /sbin/init", after mounting the decrypted LUKS container with "mount
2707 /dev/mapper/c1 /mnt/root". The second argument of switch_root is relative
2708 to the first argument, i.e. the init started with this command is really
2709 /mnt/sbin/init before switch_root runs.
2712 export PATH=/sbin:/bin
2713 [ -d /sys ] || mkdir /sys
2714 [ -d /proc ] || mkdir /proc
2715 [ -d /tmp ] || mkdir /tmp
2716 mount -t sysfs -o nodev,noexec,nosuid sysfs /sys
2717 mount -t proc -o nodev,noexec,nosuid proc /proc
2718 echo "detecting LUKS containers in sda1-10, sdb1-10"; sleep 1
2721 for j in 1 2 3 4 5 6 7 8 9 10
2726 cryptsetup isLuks $d >/dev/null 2>&1
2728 echo -n " result: "$r""
2729 # 0 = is LUKS, 1 = is not LUKS, 4 = other error
2730 if expr $r = 0 > /dev/null
2732 echo " is LUKS, attempting unlock"
2733 echo -n "test2" | cryptsetup luksOpen --key-file=- $d c1
2735 echo " result of unlock attempt: "$r""
2737 if expr $r = 0 > /dev/null
2739 echo "*** LUKS partition unlocked, switching root ***
2740 echo " (waiting 30 seconds before doing that)"
2741 mount /dev/mapper/c1 /mnt/root
2743 exec switch_root /mnt/root /sbin/init
2750 echo "FAIL finding root on LUKS, loading BusyBox..."; sleep 5
2751 exec /bin/sh --login
2754 04) What if I want a binary in the initrd, but libraries are missing?
2756 That is a bit tricky. One option is to compile statically, but that
2757 does not work for everything. Debian puts some libraries into lib/ and
2758 lib64/ which are usually enough. If you need more, you can add the
2759 libraries you need there. That may or may not need a configuration
2760 change for the dynamic linker "ld" as well. Refer to standard Linux
2761 documentation on how to add a library to a Linux system. A running
2762 initrd is just a running Linux system after all, it is not special in
2765 05) How do I repack the initrd?
2767 Simply repack the changed directory. While in tmp/, do
2770 find . | cpio --create --format='newc' | gzip > ../new_initrd
2772 Rename "new_initrd" to however you want it called (the name of
2773 the initrd is a kernel-parameter) and move to /boot. That is it.
2779 * 10.1 Is the cryptography of LUKS2 different?
2781 Mostly not. The header has changed in its structure, but the
2782 crytpgraphy is the same. The one exception is that PBKDF2 has been
2783 replaced by Argon2 to give better resilience against attacks attacks by
2784 graphics cards and other hardware with lots of computing power but
2785 limited local memory per computing element.
2788 * 10.2 What new features does LUKS2 have?
2790 There are quite a few. I recommend reading the man-page and the on-disk
2791 format specification, see Item 1.2.
2794 - A lot of the metadata is JSON, allowing for easier extension
2795 - Max 32 key-slots per default
2796 - Better protection for bad passphrases now available with Argon2
2797 - Authenticated encryption
2798 - The LUKS2 header is less vulnerable to corruption and has a 2nd copy
2801 * 10.3 Why does LUKS2 need so much memory?
2803 LUKS2 uses Argon2 instead of PBKDF2. That causes the increase in memory.
2807 * 10.4 Why use Argon2 in LUKS 2 instead of PBKDF2?
2809 LUKS tries to be secure with not-so-good passwords. Bad passwords need to
2810 be protected in some way against an attacker that just tries all possible
2811 combinations. (For good passwords, you can just wait for the attacker to
2812 die of old age...) The situation with LUKS is not quite the same as with a
2813 password stored in a database, but there are similarities.
2815 LUKS does not store passwords on disk. Instead, the passwords are used to
2816 decrypt the master-key with it and that one is stored on disk in encrypted
2817 form. If you have a good password, with, say, more than 80 bits of
2818 entropy, you could just put the password through a single crypto-hash (to
2819 turn it into something that can be used as a key) and that would be secure.
2820 This is what plain dm-crypt does.
2822 If the password has lower entropy, you want to make this process cost some
2823 effort, so that each try takes time and resources and slows the attacker
2824 down. LUKS1 uses PBKDF2 for that, adding an iteration count and a salt.
2825 The iteration count is per default set to that it takes 1 second per try on
2826 the CPU of the device where the respective passphrase was set. The salt is
2827 there to prevent precomputation.
2829 The problem with that is that if you use a graphics card, you can massively
2830 speed up these computations as PBKDF2 needs very little memeory to compute
2831 it. A graphics card is (grossly simplified) a mass of small CPUs with some
2832 small very fast local memory per CPU and a large slow memory (the 4/6/8 GB
2833 a current card may have). If you can keep a computation in the small,
2834 CPU-local memory, you can gain a speed factor of 1000 or more when trying
2835 passwords with PBKDF2.
2837 Argon2 was created to address this problem. It adds a "large memory
2838 property" where computing the result with less memory than the memory
2839 parameter requires is massively (exponentially) slowed down. That means,
2840 if you set, for example, 4GB of memory, computing Argon2 on a graphics card
2841 with around 100kB of memory per "CPU" makes no sense at all because it is
2842 far too slow. An attacker has hence to use real CPUs and furthermore is
2843 limited by main memory bandwith.
2845 Hence the large amount of memory used is a security feature and should not
2846 be turned off or reduced. If you really (!) understand what you are doing
2847 and can assure good passwords, you can either go back to PBKDF2 or set a
2848 low amount of memory used for Argon2 when creating the header.
2851 * 10.5 LUKS2 is insecure! It uses less memory than the Argon2 RFC say!
2853 Well, not really. The RFC recommends 6GiB of memory for use with disk
2854 encryption. That is a bit insane and something clearly went wrong in the
2855 standardization process here. First, that makes Argon2 unusable on any 32
2856 bit Linux and that is clearly a bad thing. Second, there are many small
2857 Linux devices around that do not have 6GiB of RAM in the first place. For
2858 example, the current Raspberry Pi has 1GB, 2GB or 4GB of RAM, and with the
2859 RFC recommendations, none of these could compute Argon2 hashes.
2861 Hence LUKS2 uses a more real-world approach. Iteration is set to a
2862 minimum of 4 because there are some theoretical attacks that work up to an
2863 iteration count of 3. The thread parameter is set to 4. To achieve 2
2864 second/slot unlock time, LUKS2 adjusts the memory parameter down if
2865 needed. In the other direction, it will respect available memory and not
2866 exceed it. On a current PC, the memory parameter will be somewhere around
2867 1GB, which should quite generous. The minimum I was able to set in an
2868 experiment with "-i 1" was 400kB of memory and that is too low to be
2869 secure. A Raspberry Pi would probably end up somewhere around 50MB (have
2870 not tried it) and that should still be plenty.
2872 That said, if you have a good, high-entropy passphrase, LUKS2 is secure
2873 with any memory parameter.
2876 * 10.6 How does re-encryption store data while it is running?
2878 All metadata necessary to perform a recovery of said segment (in case of
2879 crash) is stored in the LUKS2 metadata area. No matter if the LUKS2
2880 reencryption was run in online or offline mode.
2883 * 10.7 What do I do if re-encryption crashes?
2885 In case of a reencryption application crash, try to close the original
2886 device via following command first:
2888 cryptsetup close <my_crypt_device>.
2890 Cryptsetup assesses if it's safe to teardown the reencryption device stack
2891 or not. It will also cut off I/O (via dm-error mapping) to current
2892 hotzone segment (to make later recovery possible). If it can't be torn
2893 down, i.e. due to a mounted fs, you must unmount the filesystem first.
2894 Never try to tear down reencryption dm devices manually using e.g.
2895 dmsetup tool, at least not unless cryptsetup says it's safe to do so. It
2896 could damage the data beyond repair.
2899 * 10.8 Do I need to enter two passphrases to recover a crashed
2902 Cryptsetup (command line utility) expects the passphrases to be identical
2903 for the keyslot containing old volume key and for the keyslot containing
2904 new one. So the recovery happens during normal the "cryptsetup open"
2905 operation or the equivalent during boot.
2907 Re-encryption recovery can be also performed in offline mode by
2908 the "cryptsetup repair" command.
2911 * 10.9 What is an unbound keyslot and what is it used for?
2913 Quite simply, an 'unbound key' is an independent 'key' stored in a luks2
2914 keyslot that cannot be used to unlock a LUKS2 data device. More specifically,
2915 an 'unbound key' or 'unbound luks2 keyslot' contains a secret that is not
2916 currently associated with any data/crypt segment (encrypted area) in the
2917 LUKS2 'Segments' section (displayed by luksDump).
2919 This is a bit of a more general idea. It basically allows to use a keyslot
2920 as a container for a key to be used in other things than decrypting a
2923 As of April 2020, the following uses are defined:
2925 1) LUKS2 re-encryption. The new volume key is stored in an unbound keyslot
2926 which becomes a regular LUKS2 keyslot later when re-encryption is
2929 2) Somewhat similar is the use with a wrapped key scheme (e.g. with the
2930 paes cipher). In this case, the VK (Volume Key) stored in a keyslot
2931 is an encrypted binary binary blob. The KEK (Key Encryption Key) for
2932 that blob may be refreshed (Note that this KEK is not managed by
2933 cryptsetup!) and the binary blob gets changed. The KEK refresh process
2934 uses an 'unbound keyslot'. First the future effective VK is placed
2935 in the unbound keyslot and later it gets turned into the new real VK
2936 (and bound to the respective crypt segment).
2939 * 10.10 What about the size of the LUKS2 header?
2941 While the LUKS1 header has a fixed size that is determined by the cipher
2942 spec (see Item 6.12), LUKS2 is more variable. The default size is 16MB,
2943 but it can be adjusted on creation by using the --luks2-metadata-size
2944 and --luks2-keyslots-size options. Refer to the man-page for details.
2945 While adjusting the size in an existing LUKS2 container is possible,
2946 it is somewhat complicated and risky. My advice is to do a backup,
2947 recreate the container with changed parameters and restore that backup.
2950 * 10.11 Does LUKS2 store metadata anywhere except in the header?
2952 It does not. But note that if you use the experimental integrity support,
2953 there will be an integrity header as well at the start of the data area
2954 and things get a bit more complicated. All metadata will still be at the
2955 start of the device, nothing gets stored somewhere in the middle or at
2959 11. References and Further Reading
2961 * Purpose of this Section
2963 The purpose of this section is to collect references to all materials
2964 that do not fit the FAQ but are relevant in some fashion. This can be
2965 core topics like the LUKS spec or disk encryption, but it can also be
2966 more tangential, like secure storage management or cryptography used in
2967 LUKS. It should still have relevance to cryptsetup and its
2970 If you want to see something added here, send email to the maintainer
2971 (or the cryptsetup mailing list) giving an URL, a description (1-3 lines
2972 preferred) and a section to put it in. You can also propose new
2975 At this time I would like to limit the references to things that are
2976 available on the web.
2980 - LUKS on-disk format spec: See Item 1.2
2982 * Other Documentation
2984 - Arch Linux on LUKS, LVM and full-disk encryption:
2985 https://wiki.archlinux.org/index.php/Dm-crypt/Encrypting_an_entire_system
2989 - Some code examples are in the source package under docs/examples
2991 - LUKS AF Splitter in Ruby by John Lane: https://rubygems.org/gems/afsplitter
2993 * Brute-forcing passphrases
2995 - http://news.electricalchemy.net/2009/10/password-cracking-in-cloud-part-5.html
2997 - http://it.slashdot.org/story/12/12/05/0623215/new-25-gpu-monster-devours-strong-passwords-in-minutes
3001 * SSD and Flash Disk Related
3005 * Attacks Against Disk Encryption
3007 * Risk Management as Relevant for Disk Encryption
3015 In no particular order: