UPDATED Summary: I'm looking at Linux FDE options that are transparent to the user (my parents) in that the user doesn't need to enter 2 passwords. I found/thought of several options and tried to think through the security implications of each... but I am not great at infosec. Mostly I am wanting to verify that I have correctly understood the pros/cons of each option.
I have seen several people ITT and others mention an "Evil Maid" attack... But that seems to imply the maid is someone who doesn't know computers. I've heard it said that it is bad to underestimate one's opponents; it is also true that neither technology or tech-skills sit still... So I'm more interested in protecting from an "Evil Maid + Her Linux-Hobbyist Boyfriend" attack. Apologies for the length; hopefully my edits have trimmed it down a bit.
I am looking into using LUKS/FDE for a Linux
/home partition setup for my parents a precaution against physical theft. In an age where there are many tech savvy people, I don't think it at all unreasonable that even a computer illiterate thief might at least "know a guy" who dabbles in Linux. So I would like to protect to at least the level of making things too frustrating for a Linux hobbyist to pursue.
My parents absolutely will not bother with FDE if they have to type multiple passwords and have stated they would rather not be protected than have to type a second password (basically, this is a real-world example of AviD's Rule of Usability: "Security at the expense of usability comes at the expense of security.").
This will be on a mix of Fedora 33 (Cinnamon spin) and Linux Mint 20 systems. I am mostly trying to eliminate truly "bad" options and present them with a list of viable options to choose from. I realize that there is a subjective decision that has to be made based on security vs. convenience - that is for them to decide, I am just trying to make sure that I understand it well enough to pare it down for them. :-)
I think all of the options I have listed would produce a transparent (e.g. single-password entry) FDE system. But I don't think the security/maintenance of each is equal. I am especially interested in whether options 5 and 7 have any glaring security flaws which are easy for someone with a Kali livedisc to quickly exploit (say less than a week).
Local-file auto-unlock: LUKS can be auto-unlocked with an
/etc/crypttabentry pointing to a local keyfile per crypttab man pages. Pros: easy to setup/maintain/etc. Cons: False sense of security: anyone with a Linux livedisc could easily use the same keyfile to gain access to the unencrypted files, no special tools required. Some people consider putting the keyfile on a USB stick to be different; but my parents would either leave it plugged in or keep in next to the computer so I am lumping them together. I don't think I would really even consider this and am just listing it in case I have grossly misunderstood something.
Merged into option 1 for brevity but trying to keep same order so replies make sense.
Network-file auto-unlock: Same as option 1 but use the
_netdevoption in crypttab to get the keyfile from a network device. To prevent the key server device from being stolen along with the PC, best bet is probably to hide a RaspPi etc somewhere as the key server. The main downside is another device I have to learn/maintain. Another downside is that last I checked IOT security was garbage (I might be out-of-date though).
Tang-server auto-unlock: Basically option 3 but using tang + clevis instead of just hosting keyfile on fileserver/NFS/samba/whatever. Same cons as option 3 but probably more secure in the server software.
Linux password auto-unlock: Make LUKS passphrase same as Linux user password, then replace
/etc/mounttabmappings with something like pam_mount. Pros: sounds easy to setup, low maintenance, good usability. Cons: There is some risk here that someone could boot up a Kali (or other) livedisc, look at the hashes in
/etc/shadowand then run those hashes through some kind of cracking utility to help guess the password... something like a smarter brute-force that only targets matching hashes. But no clue how long this might take the "average" Linux hobbyist with "average" consumer hardware.... Is it days or weeks? Or something that only takes a few hours or less? If this is kind of attack is easy to pull off on plain hardware, then I'll pass; but if not, this could be a great compromise for setup/maintenance/usability/security vs setting up a key server.
UPDATE: This seems very dependent on the password used... trying simple tests using
john against the unshadowed hash for password
test took ~8s. The same test with password
example took ~8m. Trying with an old 5-digit numeric garage door pin I no longer use... it's been running for over 11.5 hours and still hasn't got it yet. That said, I'm using an older mid-tier consumer gaming rig, I'm running with defaults and no distributed cloud/custom dictionaries, and haven't tried
hashcat yet... I expect any/all of those things could yield better results (from the attacker's perspective).
However, the more I think about this from a computer science perspective, the less I like this approach. Unlike with a web-service or crypto-container like LUKS/VeraCrypt, attackers aren't just blindly trying to guess the password. They already have something that tells them what the correct password will be and it is just a matter of doing the legwork. So why couldn't the legwork be done up front and stored in a database/lookup table. I am very much a novice when it comes to this stuff. But what if there was a tool that pre-computed hashes into such lookup tables and combined that with other techniques to drastically ramp up speed? Google searches indicate things like this might already exist. Maybe the "Linux-Hobbyist Boyfriend" attacking the drive is ignorant of them like me, but considering that LUKS doesn't appear to expose its password hashes, this option would definitely make attacking a LUKS partition easier if such a tool does exist unless there is some way to have Linux not store the hash in
/etc/shadow (I am unaware of any such method/configuration shy of heavily modifying the kernel).
Skip FDE and just use VeraCrypt container: Still a few usability cons here and I'm not sure they'd like this option. There are also a few security cons (some files will probably be left in the open due to human nature/human error, browser profiles unprotected, and unencrypted copies of moved files might be viewable via recovery software). Pros: easy to setup, easy to maintain, and any data that is stored correctly is very secure. Cons: usability (still have another password to manage), some security gaps.
Cheat - FDE + auto-login w non-sudo user: If they just don't want to enter 2 passwords, I could probably set up a LUKS container normally (requires passphrase during boot via crypttab) and then setup a normal/non-root/non-sudoer user that does not require any password and just automatically logs right in. After looking into this more seriously, and assuming it was only the auto-login on the desktop manager that was used (e.g. user-account was still password-protected), then this seems like a fairly viable and secure option... as long as we don't get any more prodigy hackers attacking lockscreens. Pros: easy to setup/maintain/FDE. Cons: ??? Can't find any...
Systemd-homed?: While looking into the security of option 5, I came across this article on systemd-homed which it sounds like that is a more secure version of option 5 where you get the LUKS encrypted container for the user's
$HOMEdir (sounds like it is per user rather than all of
/hometho) and possibly without the risk of someone being able to simply fire up a cracking tool against password hashes from your unencrypted
/etc/shadowfile. It appears to be released and there are notes on the arch wiki. However, I am not yet clear on: a) how stable is it in common distros like fedora or mint, b) is the json file that is replacing
/etc/shadowsimilarly vulnerable to live disc snooping and cracking tools.
I was considering removing this question as I think I have done a poor job of constructing it in a manner than is easily answerable on stackexchange and I also feel like I did a bad job conveying what I was looking for when I initially posted it... I think that the real problem boils down to the fact that I have asked about a topic that requires answer multiple questions instead of asking about a single specific question.
However, I've been revising it and adding updates from my own research and now I am considering leaving it up in case it is useful to others that might stumble in from google... Assuming stackexchange mods don't mind?
semi-relevant remaining research questions I'm still working on:
For option #5 (and maybe 8?): Are there any estimates about how long it would take to crack non-leaked passwords (based mostly on haveibeenpwned offline db) using low-to-mid range consumer gaming hardware without distributed/cloud-based attacks or crypto-mining rigs? e.g. commonly available hardware that someone a thief might know could reasonably have. Since length/entropy matters, the examples I had planned to test with (subpar to the actual passwords) have been 5- and 10-digit random numeric "passwords", then 8-, 10-, and 12- digit random alphanumeric "passwords".
For option #5 (and maybe 8?): Need to determine what "good" parameters or dictionaries to use with
hashcatare and/or if there are better (faster) tools for cracking from a known password hash. If I am able to get the time for cracking non-leaked random 10- or 12- digit alphanumeric "passwords" down to a week or less on my older gaming hw, I will probably consider that option to be completely and utterly busted.
For option 8: how stable is
homed? Does the json file have the same vulnerabilities as option 5 does in the
Currently, as of 2021-Jun-11, my opinion of the options is as follows:
- Local-file auto-unlock: NOT SECURE ON-SITE: Vulnerable to any live discs
- USB auto-unlock: NOT SECURE ON-SITE FOR US: Vulnerable to poorly hidden usb + live discs
- Network-file auto-unlock: DON'T BOTHER; OPTION 4 IS BETTER
- Tang-server auto-unlock: LOT OF WORK, POSSIBLY INSECURE DUE TO IOT, ADDED COSTS (RASP-PI)
- Linux password auto-unlock: NOT SECURE ON-SITE. Depending on tech level of the thieves, it is potentially vulnerable to live discs. It will likely take longer to break than option 1 but it definitely can be broken if someone has a spare box to run the password cracking on for awhile. In my limited testing, I also tried 3 different pam_mount configurations without success on Fedora 33 so possibly also harder to setup than I thought.
- Skip FDE and just use VeraCrypt container: SECURE BUT USABILITY ISSUES
- FDE + auto-login w non-sudo user: FAIRLY SECURE, HIGH USABILITY
- Systemd-homed?: UNKNOWN SECURITY, UNKNOWN USABILITY
I am leaning strongly towards option 7 (LUKS + autologin). If it doesn't work, I might look into option 8 (homed) or option 4 (hidden rasp-pi running tang server) as fallbacks.