Document RAND_DRBG fork-safety locking model

Add some more exposition on why unlocked access to the global rand_fork_count
is safe, and provide a comment for the struct rand_drbg_st fork_count field.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/4110)
This commit is contained in:
Benjamin Kaduk 2017-08-07 09:55:48 -05:00 committed by Ben Kaduk
parent fffb1c5496
commit f2633200eb

View File

@ -116,6 +116,12 @@ struct rand_drbg_st {
RAND_DRBG *parent;
int secure; /* 1: allocated on the secure heap, 0: otherwise */
int type; /* the nid of the underlying algorithm */
/*
* Stores the value of the rand_fork_count global as of when we last
* reseeded. The DRG reseeds automatically whenever drbg->fork_count !=
* rand_fork_count. Used to provide fork-safety and reseed this DRBG in
* the child process.
*/
int fork_count;
unsigned short flags; /* various external flags */
@ -202,7 +208,17 @@ struct rand_drbg_st {
/* The global RAND method, and the global buffer and DRBG instance. */
extern RAND_METHOD rand_meth;
/* How often we've forked (only incremented in child). */
/*
* A "generation count" of forks. Incremented in the child process after a
* fork. Since rand_fork_count is increment-only, and only ever written to in
* the child process of the fork, which is guaranteed to be single-threaded, no
* locking is needed for normal (read) accesses; the rest of pthread fork
* processing is assumed to introduce the necessary memory barriers. Sibling
* children of a given parent will produce duplicate values, but this is not
* problematic because the reseeding process pulls input from the system CSPRNG
* and/or other global sources, so the siblings will end up generating
* different output streams.
*/
extern int rand_fork_count;
/* DRBG helpers */