I noticed that using extensions headers along with IPsecv6 tunnel mode did
not work in 2.5.73 + patch-2.5.73-bk3. The following patch
checks "nexthdr" instead of "iph->nexthdr" which could be an
extension header. I tested this with tunnel mode and transport mode
with and without extension headers and it worked ok.

into kernel.bkbits.net:/home/jmorris/net-2.5

into intercode.com.au:/home/jmorris/bk/net/net-2.5

Split xfrm_state_replace into xfrm_state_add and xfrm_state_replace.

Fixes:
1. Only update update lifetime and encap options if the state is valid.
2. Disallow updates to states that do not exist.
3. Bail if afinfo cannot be found.
 
This brings SADB_UPDATE in line with what is required by RFC2367.
It is also needed by SFS NAT-T support as it needs to update valid
states when the encap ports move.

I've tweaked the logic slightly so that SADB_UPDATE will fail on a
larval state that hasn't undergone SADB_GETSPI.  This is what RFC2367
calls for and it simplifies the code in that we don't have to call
find_acq for SADB_UPDATE.

This doesn't affect any of the three KMs as they either don't use
SADB_UPDATE or call SADB_GETSPI before doing an update.

The source address of an SA is not necessarily equal to the source address
in the selector.  This patch addresses this problem.

into linux-ipv6.org:/home/jmorris/bk/net/work-2.5

Use macro IP6_MF for the M-Flag.
Clean-up for readability (commented by <aj@dungeon.inka.de>).

 ETH1394  : - Minor cleanups.
            - Fix incorrect error handling with no broadcast channel.
	    - Set max/default MTU to 1500.
	    - Fix broken spinlock handling on device open.
 IEEE1394 : Make bus-nodeid printk's of a consistent format.
 OHCI1394 : Change one printk to a verbose/debug only message.

into kernel.bkbits.net:/home/jmorris/net-2.5

nmclan_cs doesn't compile with older compilers

From: Nick Piggin <piggin@cyberone.com.au>

It converts writes to reads and we end up returning the request to the wrong
queue.

The ext3 version number hasn't been updated since ext3 was merged.

We track ext3 via the kernel release ID.  Remove the ext3 version
number.

From: Philippe Elie <phil.el@wanadoo.fr>

kmem_cache_destroy() can fail with the following error: slab error in
kmem_cache_destroy(): cache `xxx': Can't free all objects but the cache
user really free'd all objects

This is because drain_array_locked() only frees 80% of thge objects.

Fix it by adding a parameter to drain_array_locked() telling it to drain
100% of the objects.

From: Bob Miller <rem@osdl.org>

Removed the check_mem_region() call and replaced with request_mem_region().

Because of the way the driver is structured the first request_mem_region()
call gets the default memory area.  After probing the complete memory area
that is needed to communicate with the device is known, so the first memory
area is released and the complete area is requested.

From: Bob Miller <rem@osdl.org>

- Replace the call to MOD_INC_USE_COUNT with a __module_get() when
  forcing the module to not be unloadable.

- Remove the check_region() calls and restructured things to only use
  request_region().

From: Kouichi ONO <co2b@ceres.dti.ne.jp>

in 2.5.73, I can't read/write mount DVD-RAM via ide-scsi (only r/o mount).
Without ide-scsi, I can mount DVD-RAM read/write mode and works fine.

(acked by Jens).

From: Andi Kleen <ak@suse.de>

Add proper EFAULT reporting to sys_syslog.

This fixes some silly LTP test in the 32bit emulation of an AMD64 kernel.

From: Oleg Drokin <green@namesys.com>

Fix BUG_ON(!buffer_mapped(bh)) in submit_bh()

Reiserfs is submitting unmapped buffers into submit_bh(), which trips a
BUG.

Make sure that the address_space is capable of performing the readahead
before going in and allocating the pages.

This fixes the kernel crash in the case when we do an SIOCSIFNAME
ioctl on /proc/net/dev to rename a network interface, and
we supply a string such as "foo%sbar".

into home.osdl.org:/home/torvalds/v2.5/linux

into are.twiddle.net:/home/rth/BK/axp-2.5

Fix PPP over async regression that the PPPoE changes caused.
Basically, PPP puts a zero length skbuff in the receive queue
as an error token, and the last change caused that to get flushed
as bad data.

Thanks to Diego Calleja Garcia, Matthew Harrell for validating this.