crypt_r(3): fix reentrancy problems with DES

[trasz]

Summary

This code was originally written for non-reentrant crypt(3).
In 5f521d7ba72, a thread-safe crypt_r(3) was introduced. However,
it looks like the DES implementation is still not re-entrant;
routines like setup_salt() or des_setkey() still use global
variables.

Instead of something drastic, eg removing DES support altogether,
just mark those variables as thread-local.

Given that this only applies to DES, I think the impact is minimal.

Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.

[markj]

secure/lib/libcrypt/crypt-des.c

77

It looks like this and some of the other tables below should be marked const. Can we do that as a part of this change?

83–84

Doesn't this need to be thread-local as well?

[trasz]

Add const, move 'static' first, and add __thread for stuff touched
by des_init().

[trasz]

secure/lib/libcrypt/crypt-des.c

77

Good idea, done.

83–84

These only seem to be written to by des_init(), so they should be equal for all threads, but... yeah, you're right.

[markj]

I don't see any problems with the change, but note that this adds something like 30KB of data per thread. Probably not the end of the world but not ideal either.

[trasz]

Yeah; I think it's acceptable, given it's a separate library and not libc.

(Tinderboxed.)