1 files changed, 34 insertions, 0 deletions
diff --git a/doc/os-random-seed-comparison.md b/doc/os-random-seed-comparison.md
new file mode 100644
index 0000000..71ff0c3
--- /dev/null
+++ b/doc/os-random-seed-comparison.md
@@ -0,0 +1,34 @@
+This is an attempt to document the random seed behavior of different operating
+systems. This is based mostly on Goog^WInternet searches. If you believe this
+information is incorrect, please submit patches.
+
+## Linux
+
+The random seed behavior of Linux is well documented, but we will rehash it
+(heh) here for completeness. Linux has three interfaces for random access:
+/dev/random, /dev/urandom, and getrandom. /dev/random attempts to keep track of
+the entropy count and blocks when it reaches zero. /dev/urandom never blocks.
+getrandom blocks during early startup until the entropy count becomes "full".
+
+## OpenBSD
+
+OpenBSD has one central RNG for all its randomness. The bootloader seeds the
+RNG using random data from installation plus random data obtained from the
+OpenBSD servers. Therefore, none of the random interfaces ever block.
+
+## FreeBSD
+
+On FreeBSD, /dev/random and /dev/urandom both block until the random seed is
+installed. This is defined as the time when a FD opened read-write on
+/dev/random is closed. Thereafter, they do not block.
+
+## Windows
+
+The exact behavior of the Windows RNG is not publicly documented. It is,
+however, known to be seeded in part by a registry value.
+
+## Mac OS
+
+Dunno. https://github.com/jedisct1/libsodium/issues/594 says the PRNG is
+terrible, then says it's "totally fine". I don't have Mac, and the Mac man
+pages are shamefully not accessible online, so I cannot check for myself.