On Tuesday 01 July 2008 14:22, Daniel Cheng wrote:
> On Tue, Jul 1, 2008 at 8:33 PM, Matthew Toseland
> <toad at amphibian.dyndns.org> wrote:
> > On Sunday 29 June 2008 14:52, Daniel Cheng wrote:
> >> On Sat, Jun 28, 2008 at 10:49 PM, <toad at freenetproject.org> wrote:
> >> > Author: toad
> >> > Date: 2008-06-28 14:49:23 +0000 (Sat, 28 Jun 2008)
> >> > New Revision: 20845
> >> >
> >> > Modified:
> >> >
> >
branches/db4o/freenet/src/freenet/support/io/PaddedEphemerallyEncryptedBucket.java
> >> > Log:
> >> > Don't store the RNG here. Just store a seed created on startup.
> >> >
> >>
> >> If we restarted the node twice... The PRNG will be seeded with the
> >> same seed in the two times... -- Is this behavior expected?
> >
> > The PRNG is only instantiated when we need it for padding... what's the
> > problem?
> >>
> >> IMO, make Node a static field would make the code much simpler in most
> >> code -- we can refer to the node object when we need a prng, database,
> >> etc.... For "many node in a jvm", we can use some ClassLoader
> >> hacks....
> >
> > Yay God Object. :)
> >
> > When we have many node in a jvm, we want to be able to deal with them
> > easily... e.g. in the simulation code. Not to have to go through 10 layers
of
> > indirection because of the hacks we're using.
>
> The simulator is just ~1% of all source (in terms of line of code,
> exclude unit test).
> And we don't really need 10 layers of indirection...
>
> I saw you passing the ObjectContainer and ClientContext around recently,
hence
> the suggestion.
Actually, ClientContext could reasonably be static too... But not
ObjectContainer.
>
> > W.r.t. the database, it was a deliberate decision to pass it around, this
> > makes writing relatively non-buggy code easier at this point. It could be
> > changed later ... but the single-database-thread
> > single-simultaneous-transaction design really doesn't go well with having
> > static access to the database: it would be far too easy to just grab the
> > database handle and write something, when there might be 10 other threads
> > doing the same at the same time. So we'd have to use an embedded
> > client-server design, which gives some benefits, but a number of major
> > headaches too. Performance-wise I'm not sure it would be a benefit: what
we
> > gain in parallel reads/writes (which isn't always a gain especially on
> > commodity drives), we lose in caching and refreshing.
> >
> > Having said that, the weak and strong random sources could reasonably be
> > static. These are self-contained and can be common between many nodes.
That's
> > a sensible optimisation both to code quality and to performance.
> >>
> >> > Modified:
> >
branches/db4o/freenet/src/freenet/support/io/PaddedEphemerallyEncryptedBucket.java
> >> > ===================================================================
> >> > ---
> >
branches/db4o/freenet/src/freenet/support/io/PaddedEphemerallyEncryptedBucket.java
> > 2008-06-28 14:48:55 UTC (rev 20844)
> >> > +++
> >
branches/db4o/freenet/src/freenet/support/io/PaddedEphemerallyEncryptedBucket.java
> > 2008-06-28 14:49:23 UTC (rev 20845)
> >> > @@ -18,6 +18,8 @@
> >> > import freenet.support.api.Bucket;
> >> > import java.util.Random;
> >> >
> >> > +import org.spaceroots.mantissa.random.MersenneTwister;
> >> > +
> >> > import com.db4o.ObjectContainer;
> >> >
> >> > /**
> >> > @@ -29,10 +31,10 @@
> >> >
> >> > private final Bucket bucket;
> >> > private final int minPaddedSize;
> >> > - private final Random randomSource;
> >> > private transient SoftReference /* <Rijndael> */ aesRef;
> >> > /** The decryption key. */
> >> > private final byte[] key;
> >> > + private final byte[] randomSeed;
> >> > private long dataLength;
> >> > private boolean readOnly;
> >> > private int lastOutputStream;
> >> > @@ -49,10 +51,11 @@
> >> > * @throws UnsupportedCipherException
> >> > */
> >> > public PaddedEphemerallyEncryptedBucket(Bucket bucket, int
minSize,
> > RandomSource strongPRNG, Random weakPRNG) {
> >> > - this.randomSource = weakPRNG;
> >> > this.bucket = bucket;
> >> > if(bucket.size() != 0) throw new
> > IllegalArgumentException("Bucket must be empty");
> >> > byte[] tempKey = new byte[32];
> >> > + randomSeed = new byte[32];
> >> > + weakPRNG.nextBytes(randomSeed);
> >> > strongPRNG.nextBytes(tempKey);
> >> > this.key = tempKey;
> >> > this.minPaddedSize = minSize;
> >> > @@ -76,9 +79,10 @@
> >> > if(bucket.size() < knownSize)
> >> > throw new IOException("Bucket "+bucket+" is too
> > small on disk - knownSize="+knownSize+" but bucket.size="+bucket.size()+"
> > for "+bucket);
> >> > this.dataLength = knownSize;
> >> > - this.randomSource = origRandom;
> >> > this.bucket = bucket;
> >> > if(key.length != 32) throw new
> > IllegalArgumentException("Key wrong length: "+key.length);
> >> > + randomSeed = new byte[32];
> >> > + origRandom.nextBytes(randomSeed);
> >> > this.key = key;
> >> > this.minPaddedSize = minSize;
> >> > readOnly = false;
> >> > @@ -86,7 +90,6 @@
> >> > }
> >> >
> >> > public PaddedEphemerallyEncryptedBucket(SimpleFieldSet fs,
> > RandomSource origRandom, PersistentFileTracker f) throws
> > CannotCreateFromFieldSetException {
> >> > - this.randomSource = origRandom;
> >> > String tmp = fs.get("DataLength");
> >> > if(tmp == null)
> >> > throw new CannotCreateFromFieldSetException("No
> > DataLength");
> >> > @@ -116,6 +119,8 @@
> >> > }
> >> > if(dataLength > bucket.size())
> >> > throw new
> > CannotCreateFromFieldSetException("Underlying bucket "+bucket+" is too
small:
> > should be "+dataLength+" actually "+bucket.size());
> >> > + randomSeed = new byte[32];
> >> > + origRandom.nextBytes(randomSeed);
> >> > }
> >> >
> >> > public OutputStream getOutputStream() throws IOException {
> >> > @@ -184,6 +189,7 @@
> >> > Logger.normal(this, "Not padding
> > out to length because have been superceded: "+getName());
> >> > return;
> >> > }
> >> > + Random random = new
> > MersenneTwister(randomSeed);
> >> >
> > synchronized(PaddedEphemerallyEncryptedBucket.this) {
> >> > long finalLength =
paddedLength();
> >> > long padding = finalLength -
> > dataLength;
> >> > @@ -191,7 +197,7 @@
> >> > long writtenPadding = 0;
> >> > while(writtenPadding < padding)
{
> >> > int left = (int)
> > Math.min((padding - writtenPadding), (long)buf.length);
> >> > -
> > randomSource.nextBytes(buf);
> >> > + random.nextBytes(buf);
> >> > out.write(buf, 0, left);
> >> > writtenPadding += left;
> >> > }
> >> >
> >> > _______________________________________________
> >> > cvs mailing list
> >> > cvs at freenetproject.org
> >> > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/cvs
> >> >
> >> _______________________________________________
> >> Devl mailing list
> >> Devl at freenetproject.org
> >> http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
> >>
> >>
> >
> > _______________________________________________
> > Devl mailing list
> > Devl at freenetproject.org
> > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
> >
> _______________________________________________
> Devl mailing list
> Devl at freenetproject.org
> http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL:
<https://emu.freenetproject.org/pipermail/devl/attachments/20080701/83e131b6/attachment.pgp>
