ISC DHCP 3.0.2b1 pre-release testing.

[David W. Hankins]

This is the first BETA release of ISC DHCP 3.0.2.  I need you kind folks
to look it over, compile it on your favorite strange and unusual systems,
run it under unusual configurations, so forth.  You can fetch the tarball
and gpg signature from this location:

  ftp://ftp.isc.org/isc/dhcp/private/dc722e7036b27a22/dhcp-3.0.2b1.tar.gz
  ftp://ftp.isc.org/isc/dhcp/private/dc722e7036b27a22/dhcp-3.0.2b1.tar.gz.asc

I would like to hear your reports, both positive and negative.

If no problems evolve in the next week, 3.0.2b1 will be promoted to
release candidate 1 and distributed to the general population.

NOTE:  The 'mis-host patch' that I circulated following the discovery
of the bug which caused host references to become stale has not been
included in this release.  The problem was re-examined and repaired
in a completely different manner.  So, if you had this problem yourself
please be particularly cautious looking to see if the problem recurs.


                        Changes since 3.0.1

- A bug was fixed in the server's 'option overloading' implementation,
  where options loaded into the 'file' and 'sname' packet fields were
  not aligned precisely as rfc2131 dictates.

- The FreeBSD client script was changed to support the case where a domain
  name was not provided by the server.

- A memory leak in 'omshell' per each command line parsed was
  repaired, thanks to a patch from Jarkko Torppa.

- Log functions writing to stderr were adjusted to use the STDERR_FILENO
  system definition rather than '2'.  This is a no-op for 90% of platforms.

- One call to trace_write_packet_iov() counted the number of io vectors
  incorrectly, causing inconsistent tracefiles.  This was fixed.

- Some expression parse failure memory leaks were closed.

- A host byte order problem in tracefiles was repaired.

- Pools configured in DHCPD for failover possessing permission lists that
  previously were assumed to not include dyanmic bootp clients are now
  a little more pessimistic.  The result is, dhcpd will nag you about just
  about most pools that possess a 'allow' statement with no 'deny' that
  would definitely match a dynamic bootp client.

- The 'ddns-update-style' configuration warning bit now insists that
  the configuration be globally scoped.

- Two memory leaks in dhclient were closed thanks to a patch from Felix
  Farkas.

- Some minor but excellently pedantic documentation errors were fixed
  thanks to a patch from Thomas Klausner.

- Bugs in operator precedence in executable statements have been repaired
  once again.  More legal syntaxes should be parsed legally.

- Failing to initialize a tracefile for any reason if a tracefile was
  specified is now a fatal error.  Thanks to a patch from Albert Herranz.

- Corrected a bug in which the number of leases transferred as calculated
  by the failover primary and sent to peers in POOLRESP responses may be
  incorrect.  This value is not believed to be used by other failover
  implementations, excepting perhaps as logged information.

- Corrected a bug in which 'dhcp_failover_send_poolresp()' was in fact
  sending POOLREQ messages instead of POOLRESP mesasges.  This message
  was essentially ignored since failover secondaries effectively do not
  respond to POOLREQ messages.

- Type definitions for various bitwidths of integers in the sunos5-5
  build of ISC DHCP have been fixed.  It should compile and run more
  easily when built in 64-bit for this platform.

- "allow known-clients;" is now a legal syntax, to avoid confusion.

- If one dhcp server chooses to 'load balance' a request to its failover
  peer, it first checks to see if it believes said peer has a free
  lease to allocate before ignoring the DISCOVER.

- log() was logging a work buffer, rather than the value returned by
  executing the statements configured by the user.  In some cases,
  the work buffer and the intended results were the same.  In some other
  cases, they were not.  This was fixed thanks to a patch from Gunnar
  Fjone and directconnect.no.

- Compiler warnings for some string type conversions was fixed, thanks
  to Andreas Gustafsson.

- The netbsd build environments were simplified to one, in which
  -Wconversion is not used, thanks to Andreas Gustafsson.

- How randomness in the backoff-cutoff dhclient configuration variable
  is implemented was better documented in the manpage, and the behaviour
  of dhclient in REQUEST timeout handling was changed to match that of
  DISCOVER timeout handling.

- Omapi was hardened against clients that pass in null values, thanks
  to a patch from Mark Jason Dominus.

- A bug was fixed in dhclient that kept it from doing client-side
  ddns updates.  Thanks to a patch from Andreas Gustafsson, which
  underwent some modification after review by Jason Vas Dias.

- Failover implementations disconnected due to the network between
  them (rather than one of the two shutting down) will now try to
  re-establish the failover connection every 5 seconds, rather than
  to simply try once and give up until one of them is restarted.
  Thanks to a patch from Ulf Ekberg from Infoblox, and field testing
  by Greger V. Teigre which led to an enhancement to it.

- A problem that kept DHCP Failover secondaries from tearing down
  ddns records was repaired.  Thanks to a patch from Ulf Ekberg from
  Infoblox.

- 64bit pointer sizes are detected properly on FreeBSD now.

- A bug was repaired where the DHCP server would leave stale references
  to host records on leases it once thought about offering to certain
  clients.  The result would be to apply host and 'known' scopes to the
  wrong clients (possibly denying booting).  NOTE:  The 'mis-host' patch
  that was being circulated as a workaround is not the way this bug was
  fixed.  If you were a victim of this bug in 3.0.1, you are cautioned
  to proceed carefully and see if it fixes your problem.

- A bug was repaired in the server's DHCPINFORM handling, where it
  tried to divine the client's address from the source packet and
  would get it wrong.  Thanks to Anshuman Singh Rawat.

- A log message was introduced to help illuminate the case where the
  server was unable to find a lease to assign to any BOOTP client.
  Thanks to Daniel Baker.

- A minor dhcpd.conf.5 manpage error was fixed.


-- 
David W. Hankins		"If you don't do it right the first time,
Operations Engineer			you'll just have to do it again."
Internet Systems Consortium, Inc.		-- Jack T. Hankins