Calculate memory usage for resume process

The OOM killer issue that's mentioned in #19 (closed) most likely is due to the fact that unlocking LUKS2 devices with argon2 as KDF eats a lot of memory. According to @guilhem, argon2 memory usage is capped at 1G. @guilhem, do you have a reference?

In any case, we should check whether one of the to-be-processed devices is a LUKS2 device with argon2 as KDF and if there's enough memory available.

According to @guilhem, argon2 memory usage is capped at 1G. @guilhem, do you have a reference?

AFAICT it's the DEFAULT_LUKS2_MEMORY_KB macro which defined at build time.:

$ cryptsetup --help | grep -A1 "^Default PBKDF for LUKS2:"
Default PBKDF for LUKS2: argon2i
	Iteration time: 2000, Memory required: 1048576kB, Parallel threads: 4

I've not seen it documented, but the memory cost will be also be lowered not to exceed half of the physical memory, otherwise devices couldn't be unlocked on systems with <1GiB RAM. (See also #924560#15.) So AFAICT the actual default memory cost is

min(phys_memory/2, 1GiB)

. Problems arise when the second half of the memory is not enough for the rest of the system; or when multiple LUKS2 devices are being unlocked in parallel, as Dimitri reported.

mentioned in merge request !4 (merged)

mentioned in issue #19 (closed)

I went with a first implementation that takes the default memory usage for argon2i keyslots: e303e291

A cleaner approach would be to iterate over all LUKS devices, try to get the actual memory usage values for all keyslots and take the maximum amount of them. But that's cumbersome and also further increases the runtime of cryptsetup-suspend.c.

TODO: We have to check what happens if there's less than the detected amount of memory available.

changed title from Calculate memory usage to Calculate memory usage for resume process

mentioned in commit 7054eb97

mentioned in commit cf116aab

mentioned in commit 6524dcfc

mentioned in commit 15d03181

mentioned in commit e52fad86

mentioned in commit f840a17b

The c implementation to calculate the memory usage for LUKS2 devices with argon2i PBKDF is now functional and activated.

I also decided to allocate 2MB extra memory just to be on the save side: https://salsa.debian.org/mejo/cryptsetup-suspend/blob/436c57786c8e444c96df9b54fe034214dd6388bf/debian/scripts/suspend/cryptsetup-suspend.c#L151-153

Our assumption is that non-argon2i hashing algorithms (like PBKDF2) also need some memory. Though, 2MB is a randomly picked value and we definitely need to find a sensible value here.

mentioned in commit 423e7c77

mentioned in commit f98ae88e

mentioned in commit e8142fcd

mentioned in commit bea29be4

mentioned in commit 10a7539f

mentioned in commit c98678fe

mentioned in commit 31c2a2a6

mentioned in commit b5fdc1fd

mentioned in commit 285b97e5

mentioned in commit 0d734698