freenetwork at web.de wrote:
> Thanks to both Davids, that cleared it up, I vaguely thought it would work
> like this.
>
> But isn't the "border" between too new and too old builds creating an event
> horizon, a network separation as the old nodes behind the border don't know
> that they are outdated (unless a direct neighbor updates his node and places
> a new "seed" for the current
> lastGoodBuild)? If the border nodes don't upgrade (any supposing freenet is
> run all day long as a daemon in the background without user intervention and
> attention) the old nodes won't see the newer lastGoodBuild.
>
> I would not like to have the lastGoodBuild to be propagated as that would
> open a can of worms, but how to address above?
>
lastGoodBuild propagation basically works by "peer pressure" (pun sorta
intended). The more peers you are connected to, the more likely you are
to know about a new version. However, you don't know until one of your
peers upgrades or you read about it, maybe out of band. The better the
network interconnections, the faster the word spreads (seven degrees of
separation and such ideas help).
>> freenetwork at web.de wrote:
>>
>>> What if someone forges his reference and sets lastGoodBuild=9999 ? How will
>>> that affect this?
>>> How ist lastGoodBuild propagated anyway? Is it propagated? What are the
>>> implications for the network if one or more nodes are forging lastGoodBuild?
>>>
>>>
>> A node running r8755 or higher simply won't connect to such a rogue node
>> since they are too new for our node. Likewise, unless the rogue node
>> changes some code in the process, his node won't connect with anybody
>> else's either. There are much more sinister things such a rogue node
>> could try to do than fake a lastGoodBuild.
>>
>> lastGoodBuild is "propagated" by people upgrading their nodes to new
>> software. A node does not start advertising a new lastGoodVersion (the
>> noderef version of lastGoodBuild) by any means other than getting
>> upgraded to a newer version of software that declares a different
>> lastGoodBuild. A node does not "adopt" a lastGoodVersion advertised by
>> another node as it's own advertised lastGoodVersion, nor does it treat
>> it internally as it's lastGoodBuild (it would immediately stop
>> connecting to anybody if it did anyway... :).
>>
>> If there are any existing problems with rogue nodes advertising fake
>> lastGoodVersions, my recent looking at the code suggests that they're
>> very subtle and hard to notice as the current mechanism looks solid to me.
>>
>>>> Author: zothar
>>>> Date: 2006-05-18 03:57:03 +0000 (Thu, 18 May 2006)
>>>> New Revision: 8755
>>>>
>>>> Modified:
>>>> trunk/freenet/src/freenet/node/PeerNode.java
>>>> Log:
>>>> Activate the user build too old alerting commited in r8754. We only aler=
>>>> t if we get a handshake request from a peer that's still advertising a la=
>>>> stGoodVersion that's newer than our version. This allows the lastGoodVer=
>>>> sion to be lowered and thus older nodes would potentially be able to conn=
>>>> ect again. This commit also removes the handshake timer renew when we re=
>>>> ceive a handshake from a peer that's too old because, otherwise, that mig=
>>>> ht prevent us from sending them our ref (and the latest lastGoodVersion i=
>>>> nformation) in a handshake request sent to them. Persons running builds =
>>>> after this that are too old should get the idea pretty quick with a user =
>>>> alert on the welcome page and one or more peers in "TOO NEW" status on th=
>>>> e Darknet Connections page.
>>>>
