Re: [PATCH] rand.c updates from FreeBSD RELENG_5
On 2004-05-13, Kris Kennaway <kkenn@xor.obsecurity.org> wrote:
> On 2004-05-12, Matthew Dillon <dillon@apollo.backplane.com> wrote:
>>
>>:What problems? I did a grep for /dev/random on 4.x and only came up
>>:with a couple of uses in contrib/.
>>:
>>:> I think we should simply use /dev/urandom here,
>>:> which is guarenteed not to block and will certainly produce a
>>:> random enough seed for rand's tiny little seed.
>>:
>>:/dev/random is non-blocking in FreeBSD 5.x since it uses a different,
>>:non-blocking algorithm.
>>:
>>:Kris
>>
>> Search the FBSD mail archives for the subject line '/dev/random'. There
>> have been several threads. Most recently, a newfs after booting into
>> single user would lock up until enough randomness existed for its
>> inode seeding.
>
> That was driver breakage, not a misuse of /dev/random vs. /dev/urandom.
To clarify, /dev/random and /dev/urandom are identical in FreeBSD 5.x,
and the only purpose of keeping both around is for compatibility.
If you're porting code from 5.x to a 4.x-based OS like DF, you need to
take care because of the different semantics of the /dev/random driver
interface. OTOH it may make sense for you to port the 5.x /dev/random
driver itself, because it has much better properties including better
entropy gathering and better robustness against entropy starvation.
Kris
討論串 (同標題文章)
完整討論串 (本文為第 9 之 18 篇)：