The following patch against 2.6.0-test4 adds four more iptables modules.
They are adding the following functionality
- iprange: matching against an arbitrary contiguous range of ip addresses
- CLASSIFY: setting skb->priority from iptables (so you can skip tc filter)
- NETMAP: SNAT a whole network 1:1 to another network
- SAME: tries to keep the assigned ip per client the same within an SNAT
  pool

The following patch against 2.6.0-test4 (courtesy of Patrick McHardy)
optimizes the conntrack code.  In the old implementation, the hash function
was passed to the LIST_DELETE macro, which resulted in it being called
two times instead of one.

The following patch against 2.6.0-test4 (courtesy of Patrick McHardy)
optimizes the NAT code.  In the old implementation, the hash function
was passed to the LIST_DELETE macro, which resulted in it being called
two times instead of one.

- moves all MODULE_{AUTHOR,DESCRIPTION,LICENSE} statements to the same
  location
- adds some missing MODULE_LICENSE(GPL) tags
- adds MODULE_DESCRIPTION and AUTHOR to all modules

We have decided to remove the unclean match, since it is considered
a potentially dangerous function of the current iptables code.

The match is used by lots of users who don't really undestand what kind
of danger they are imposing on the future-compatibility of their
networks.  (just think of the ECN issue resulting from this kind of
filtering)

We'd rather keep it in patch-o-matic, where lots of other modules that
are only useful in experimental scenarios are kept.

Now that we don't have to keep it for compatibility reasons, we'd like
to remove it before 2.6.0 final is released.

We have decided to remove the MIRROR target, since it was considered a
stupid and potentially dangeroups example code of the early netfilter
days that should never be used on the internet anyway.

* correct mtt bits to indicate 1msec or more
* rename IRENABLE_IREN to IRENABLE_PHYANDCLOCK
* rearrange driver metadata and header
* driver version 0.5

* tx-path cleanup
* fix deadlock when setting speed in tx_interrupt, issue was introduced by
  previous interrupt locking cleanup
* don't let start_xmit return NET_XMIT_DROP if we drop and free the skb.
  This fixes an old bug in the error path leading to skb_slab corruption

* cleanup baud rate setting and mode switch
* locking and barrier review

* interrupt handler cleanup, focus on fast path and low latency
* rx-path cleanup
* add missing crc16 check of incoming SIR frames

* fix error path for ring entry alloc in case pci_map failed
* get rid of BUG() - it's mostly in interrupt and there's no need
  to kill the box on such issues
* correct endianess for the hardware view of ring descriptors

* don't fail without procfs - it's only needed for diagnostics
* get rid of printk in favour of IRDA_DEBUG and friends wherever possible
* reduce kernellog noise depending on irda debuglevel

* Kconfig: we depend on CONFIG_PCI
* update header compatibility stuff
* beautify C99-initializers for PCI IDs
* PCIDEV_NAME wrapper to abstract device name storage location
* cleanup of the pci shutdown path. Also fixing a possible NULL-pointer
  dereference when the driver is rmmod with the netdev still running.

The lower layers of ax25 hold a reference to the underlying device
but don't increment the ref count. This is safe because it does the
right thing when UNREGISTER notification comes in, but it is better
to do the right thing.

In 2.6, the TFTP conntrack helper returns -1 if the packet is too
short, but that is an invalid return code.  Return NF_ACCEPT
instead.

* use proc_net_fops_create to setup
* collapse two_line setup functions into the init routine
* proc_exit routine was never called and can go. 
* cleaner to refer to proc_net as base rather than net/rt_acct

sound/pcmcia/vx/vx_entry.c:

This patch removes the PCMCIA timer release functionality which is no
longer required. Without this the module does not compile.

Too much cut-and-paste, noticed by Dmitry Torokhov <dtor_core@ameritech.net>

into samba.org:/home/paulus/kernel/for-linus-ppc

On PPC32 we reorder the arguments so they fit into 6 registers.  Glibc will
need a two-line stub to change them from the standard order to the ordering
used by the system call: (fd, advice, offset, len).

into samba.org:/home/paulus/kernel/for-linus-ppc

into samba.org:/stuff/paulus/kernel/for-linus-ppc

into samba.org:/stuff/paulus/kernel/for-linus-ppc

into samba.org:/stuff/paulus/kernel/for-linus-ppc

into samba.org:/stuff/paulus/kernel/for-linus-ppc

In both 2.4 and in 2.6, error handling for bad cdrom media is
wrong.  And it is my fault I'm afraid, since I botched an earlier
fix for the problem by putting the fix in the wrong spot.

My kids have a "Jumpstart Toddlers" cd they have long since
completely killed, which makes a great test disc.  Without this
fix, the best time projection I can get for completing a dd type
sector copy is about 2 years...  Most of that is spent thrashing
about in kernel space trying to re-read sectors we already know
are not correctable....  After the fix, I was able to rip a copy
the CD (or rather muddle through it getting lots of EIO errors)
in about 15 minutes.

Attached is the fix for 2.6.x,