On Wed, 23 Aug 2017 at 19:13:57 -0700, Chris Lamb wrote: > gfsplit SIGSEGV if /dev/urandom is not accessible as it doesn't fallback > to the (correctly named!) gfshare_bad_idea_but_fill_rand_using_random; it > prints a warning to stderr.
I can't help wondering whether this might be deliberate, or at least, not a bad idea. The upstream commit message that changed the ABI to make it mandatory to set a randomness-obtaining function says: """ Update to require gfshare_fill_rand to be initialised. Require explicit initialisation of gfshare_fill_rand so that numpties get a SEGV instead of bad randomness. """ — https://git.gitano.org.uk/libgfshare.git/commit/?id=da0566422af4e0ad5c9e17cfe21f563e4274338d and I think failure to read /dev/urandom is a similar situation? If you're using gfsplit/gfcombine for crypto, it had better be using crypto-quality randomness (I don't think the theoretical basis for libgfshare is sound if your randomness isn't random). If that's the intention, it would be better if it was clearly deliberate by calling abort() or fprintf() + exit(). smcv
