TIL busybox generates password salts from current time in usec. Does this matter? Probably not, but seems mildly sus.
12
11
123
Conversation
Replying to
I expect BusyBox machines have very little entropy from attached hardware, etc., possibly not enough for standard syscalls? But yeah, still sus
2
4
No, the kernel CSPRNG works properly everywhere on modern Linux kernels. Software should only have a fallback code path for legacy kernels if at all. Should never be doing things wrong as the first choice regardless.
1
9
What’s “modern”? Plenty of microcontrollers don’t go past 2.6…
1
Oldest Linux kernel branch that's still supported is Linux 4.4 and it will be dropped soon.
A new LTS is added every year and they receive 6 years of support. Products with a longer lifetime are expected to move to new LTS branches. Most projects shouldn't support older kernels.
1
2
It's up to distributions freezing package versions to provide support for their ancient frankenstein software versions. It makes sense to support the oldest LTS branches that are still maintained but anything older is strictly the realm of insecure embedded / enterprise nonsense.
1
3
github.com/torvalds/linux fixed this upstream for 5.4 (mainline). It was quickly backported to the 4.19 LTS from mainline.
That means it works properly in the 2020 LTS (5.10), 2019 LTS (5.4) and 2018 LTS (4.19). There's going to be a new 2021 LTS in the next couple months too.
1
3
I'm glad to hear Linux has improved the entropy situation. All my knowledge in this arena was from FIPS work in the 2.6 and 3.x days, where a minimal headless setup could actually be a risk
1
getrandom(buf, size, GRND_RANDOM) and /dev/random also no longer have the cargo cult blocking based on input entropy. They only block at most once during early boot and if they do the kernel quickly produces entropy. Main remaining API issue is that /dev/urandom doesn't do that.