Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
Hello,
Here are the first set of changes I'd like to contribute to the
protocol-dns subproject. Here's a list of things I've done.
1) Converted the enumeration types ServiceType, ProtocolType,
RecordClass, OpCode, RecordType, MessageType and ResponseCode into
actual Java 5 enums. Because we need to encode and decode the actual
values we can't rely on the ordinals the compiler will assign to them,
so I've used a technique I found on the Java Specialists site
(http://www.javaspecialists.co.za/archive/newsletter.do?issue=113), to
efficiently do a conversion between enums and values with minimal code
duplication.
2) Removed QuestionRecords and ResourceRecords classes since they were
really just type-safe wrappers around Lists and replaced their usage
with generic Lists.
3) Moved any encoding and decoding code that was in the MINA
ProtocolEncoder and ProtocolDecoder to classes in the
o.a.d.s.dns.io.encoder and io.decoder packages called DnsMessageEncoder
and DnsMessageDecoder, respectively.
4) Removed the dependency by the encoders and decoders in the
o.a.d.s.dns.io packages on MINA. The only thing they really used were
the ByteBuffers from MINA and it was a pretty simple task to just switch
to java.nio.ByteBuffers instead. The only hurdle I had to overcome was
in the allocation of temporary buffers for things like enecoding domain
names, which leads me to...
5) Instead of creating (or having to pull them from the MINA ByteBuffer
pool) several ByteBuffers during encoding, I simply wrote directly to
the buffer that was passed in. From what I could tell, the main reason
for allocating the temporary buffers was to do the actual encoding of
the domain name, get the byte array that was written to the buffer,
write the length of the data to the actual ByteBuffer, then write the
byte array from the temporary ByteBuffer. Instead, what I did was to
record the position of the ByteBuffer, increment it by one, write the
data (the domain name for instance), find the difference between the
starting position + 1 and the current position, then go back to the
starting position and write the length of the data and finally jump back
to the end of the data and go onto the next bit of encoding. This
should be more efficient because we're not constantly having to get
ByteBuffers from a pool or have new ones allocated, and we also don't
have to then create byte arrays and have the data from those temporary
ByteBuffers read into the byte arrays.
6) Moved all the (get|put)Unsigned(Byte|Short|Int) methods into a
utility class called ByteBufferUtil.
7) Changed the DnsMessage.transactionId property to be an int because it
actually needs to be an unsigned short.
8) Fixed a bug with decoding domain names when they are compressed. The
new position to jump to wasn't being calculated right all the time.
There also seemed to be an assumption that the first thing to do after
jumping was to read a label, but it could be the case that the new
position is another jump so we need to call recurseDomainName() right
after doing the jump in position.
9) Updated existing tests to actually check that what is encoded and
decoded are correct. For this I also had to add equals() and hashCode()
methods to classes that didn't already have them, like DnsMessage,
ResourceRecordImpl, and others. I simply used the commons-lang equals
and hashCode builders for this because I've found them to be
tremendously useful for this in the past.
10) Added additional test for MX query and response parsing. I used the
MX records from apache.org.
11) Added decoders to make the tests added in (10) pass. these are the
IPv6RecordDecoder, MailExchangeRecordDecoder, AddressRecordDecoder, and
NameServerRecordDecoder.
That should cover it all. I'd like to add a few more decoders for
things like TXT and SOA records. That should cover about 90% of the
records in use on most DNS servers.
Then I'd like to get things separated a little bit more and create
subprojects within protocol-dns. I'm thinking something like
protocol-dns/
core/
mina-shared/
server/
shared/
mina/
store/
client/
core would contain all the DNS message and record types as well as the
decoders and encoders in dns.io right now. mina-shared would contain
the ProtocolCodec stuff in dns.protocol.
server/shared would contain the DnsConfiguration and other stuff that
would be shard among server implementations, whether they use MINA or
some other IO framework and whether they use ApacheDS for the store or
some other backend like Bind zone files.
server/mina would contain the DnsProtocolHandler from dns.protocol as
well as the DnsServer class itself.
server/store would contain the the backend store type stuff. It would
probably wind up being separate subprojects itself, one for LDAP, JDBC,
or Bind zone files.
client would obviously contain any code for developers to generate and
send queries to dns servers, and maybe an implementation of the dig
command line tool.
I wouldn't mind actually seeing this becoming a separate project of it's
own, like dns.apache.org. That would be really cool.
What do you think of these changes?