On Wed, Jul 04, 2012 at 11:02:20PM +0300, Konstantin Belousov wrote: > On Wed, Jul 04, 2012 at 07:51:25PM +0000, Pawel Jakub Dawidek wrote: > > Author: pjd > > Date: Wed Jul 4 19:51:25 2012 > > New Revision: 238118 > > URL: http://svn.freebsd.org/changeset/base/238118 > > > > Log: > > Prefer sysctl to open/read/close for obtaining random data. > > This method is more sandbox-friendly and also should be faster as only > > one syscall is needed instead of three. > > In case of an error fall back to the old method. > > > > Reviewed by: simon, gleb > > MFC after: 2 weeks > IMO it is weird and against a purpose of sysctl that kern.arand sysctl > exists at all. I would prefer to not spread its usage more. We have > to keep it to preserve ABI compatibility, but I do not think that the > location for random data provider is right, not to mention higher > overhead of sysctl machinery.
1) /dev/urandom may not exist in jails/sandboxes while sysctls (or old way initialization) always exists. 2) Current NetBSD code uses KERN_URND we don't have. 3) Current OpenBSD code uses KERN_ARND as in the change committed. 4) Our KERN_ARND initially is initialized only from from the weak value at the boot stage. 5) I already provide two working patches (one with atomic and another one without) to bypass issue 4) but they are never committed by person who promise to handle them (markm@ CCed) for the reason unknown and I can't do it by myself due to stupid secteam@ 5 years old ban. 6) So, current KERN_ARND way gives weak randomness for just started after boot programs and should be fixed by either my patches mentioned in 5) or by implementing KERN_URND as in NetBSD. Note that the initial ARND seeding problem does not exist in OpenBSD since they use different KERN_ARND implementation. (besides usage in arc4random code I think KERN_ARND should be fixed in anycase) -- http://ache.vniz.net/
pgpyrYimDeGuF.pgp
Description: PGP signature
