How does bit rot affect a LUKS container and the filesystem inside?
Suppose you have a filesystem that is well suited to deal with bit rot. Now put it inside a LUKS container. In case bit rot corrupted the container, I assume the decrypted filesystem will suffer huge amounts of corrupted raw bytes / blocks.
How does LUKS protect against this?
Bitrot in the LUKS header (key and otherwise critical material): it's *poof* gone.
(There is a bit of redundancy and checksum for the LUKS2 header but it doesn't cover much, so chances are... it's still gone).
Bitrot in encrypted data: it depends on the encryption mode, but in general, a single bit flip will result in 16 wrong bytes.
Set up encryption:
# truncate -s 32M bitrottest.img
# cryptsetup luksFormat bitrottest.img
# cryptsetup luksOpen bitrottest.img bitrottest
Make it all zero:
# shred -n 0 -z /dev/mapper/bitrottest
# hexdump -C /dev/mapper/bitrottest
00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
01000000
Flip a bit:
# losetup
NAME SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE DIO LOG-SEC
/dev/loop0 0 0 1 0 bitrottest.img 0 4096
# dd bs=1 count=1 skip=30M if=/dev/loop0 | hexdump -C
00000000 a2 |.|
00000001
# printf "\xa3" | dd bs=1 count=1 seek=30M of=/dev/loop0
# dd bs=1 count=1 skip=30M if=/dev/loop0 | hexdump -C
00000000 a3 |.|
00000001
Result:
# hexdump -C /dev/mapper/bitrottest
00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00e00000 eb d1 bd b0 2a f5 77 73 35 df 82 40 1e a7 27 11 |....*.ws5..@..'.|
00e00010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
01000000
One flipped bit, 16 whacky bytes.
Protection? None whatsoever. For that, you'd have to add integrity (just to report errors, redundancy is still a separate issue from that).
You are not supposed to deliberately write corrupt data to your storage.
Storage is supposed to report read errors instead of returning bogus data. In that case your data is still gone, but at least, it's not silent bitrot.
LUKS itself doesn't protect against bit rot. If you want protection against bit rot on the block device level you need DM integrity. LUKS 2 can be combined with integrity to get authenticated encryption (cryptsetup luksFormat with --integrity) and with that system will be able to detect that the data changed, but that isn't really helpful against bit rot -- the sector will simply give you IO error if you try to read it and doesn't know how to fix it. (The point of AEAD is to detect that someone changed the encrypted data.)
To actually fix the data you'll need RAID 1 and the standalone integrity device under the RAID legs to make self healing possible (this works automatically, RAID gets IO error from one leg, reads the data from the other one and fixes (tries to) the first one). The new systemd 250 added support for assembling integrity devices during boot via /etc/integritytab (for non-root filesystems) so you can try that. But this all needs to be done "under" the LUKS level to make it work as @frostschutz showed in his answer because the error needs to be fixed in the encrypted data.
