Signed-off by: James Lamanna <jlamanna@gmail.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Marcel Sebek <sebek64@post.cz>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

Anand Parthasarathy pointed out that draft-ietf-ipoib-ip-over-infiniband-09.txt says:

            [DISC] specifies the length of source/target option in
            number of 8-octets as indicated by a length of '3' above.
            Since the IPoIB link-layer address is only 20-octet long,
            two octets of zero MUST be prepended to fill the total
            option length of 24 octets.

The current Linux neighbour discovery code puts the padding after the
link address.  This patch fixes up ndisc.c to put the padding in the
correct place by adding a general ndisc_addr_option_pad() function
(which could be used in the future if someone ever implements RFC 3831
IPv6-over-FC or some other encapsulation that requires padding).
Signed-off-by: Roland Dreier <roland@topspin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

into nuts.davemloft.net:/disk1/BK/net-2.6

1) move ip_nat_protocol_register/unregister since lock is now static.
2) fix ip_nat_snmp_basic: may not work, but now compiles.
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Rusty removed ip_nat_helper_{register,unregister} and
{__,}ip_nat_find_helper but left the symbol exports in
by mistake.
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

IMHO the TSO stuff has mostly settled down now.  As a result
tcp_pcount_t is now more of a liability in that it reduces the
readability of the code.

So here is a patch to remove it.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch contains the following cleanups:
- make some needlessly global code static
- net/802/hippi.c: remove the unused global function hippi_net_init
- net/8021q/vlan.c: remove the global variable vlan_default_dev_flags
                    that was never changed
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Anton points out that cacheline aligning conntrack entries is a wank.
He's right: there's lots of them, and they're currently ~200 bytes.
Same with the cargo-cult programming in ipt_hashlimit.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fairly simple patch to move the 'initialized' NAT bitfield to bits in
the 'status' word.  This saves the size of a pointer from the
connection tracking structure.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Rather than calling the sequence adjustment code on every connection
which has a helper, we can set a status bit on the conntrack when we
change the length of a TCP packet, and use that to indicate that we
should call the routine.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

We keep a pointer from the hash table entry into the connection
tracking entry it's a part of.  However, there's a spare byte in the
hash entry anyway, which we can use to indicate which of the two
tuples it is, and the simply use container_of() to access the
conntrack.

This saves two pointers per connection tracking entry.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Original patch and multo bugfixes by Krisztian Kovacs.

Now NAT has been simplified, there is only one place to NAT each
packet.  That means we can intuit what to do by looking at the
difference between this packet and the reply we expect, getting rid of
the manips[] array in the connection tracking structure, which is 72
bytes.  Rework NAT to be based on 'change this packet to make src/dst
look like this tuple'.

1) Each protocol's manip_pkt takes a 'struct ip_conntrack_manip',
   which is half (the source half) of a tuple.  Hand the whole desired
   tuple to the NAT code and have it use the 'maniptype' arg to decide
   what part to copy.

2) Krisztian points out that we don't need the NAT lock to read the
   NAT information (or the tuples) as they never change once set, and
   while being set we have exclusive access.  A lock is only needed to
   deal with only remaining NAT list: the bysource hash.

3) We don't need to rehash for the bysource hash: it depends on the
   incoming packet, which we can't change.

4) Many NAT functions only need the maniptype they are to perform, not
   the actual hook, which makes the code clearer.

5) New status bits to indicate what NAT needs to be done.  We can
   always figure it out by inverting the tuple we expect in the other
   direction and comparing it, but this is faster.

6) Rename 'do_bindings' to 'nat_packet'.

7) ICMP handing is vastly simplified: we unconditionally change to
   look the way we want.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Adrian Bunk's cleanup patch, updated for after all the Rusty patches.
The ip_nat_protocol_register/unregister EXPORT_SYMBOLs() stay, as they
are used by future patches.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (modified)
Signed-off-by: David S. Miller <davem@davemloft.net>

From: KOVACS Krisztian <hidden@sch.bme.hu>

  Hi Rusty,

Your recent patch which removed the byipsproto hash left some unused
code around. The following patch cleans up that. I'm not sure it's
correct, but please take a look at it.
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch simplifies the code by always having expectation timeouts.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Now we've changed expect handling, we can simplify it significantly.

1) struct ip_conntrack_expect only exists until the connection
   matching it is created.  Now NAT is done directly at the time the
   expectation is matched, we don't need to keep this information
   around.

2) The term 'master' is used everywhere to mean the connection that
   expected this connection.  The 'master' field in the new connection
   points straight to the master connection, and holds a reference.

3) There is no direct link from the connection to the expectations it
   has created: we walk the global list to find them if we need to
   clean them up.  Each expectation holds a reference.

4) The ip_conntrack_expect_tuple_lock is now a proper subset of
   ip_conntrack_lock, so we can eliminate it.

5) Remove flags from helper: the policy of evicting the oldest
   expectation seems to be appropriate for everyone.

6) ip_conntrack_expect_find_get() and ip_conntrack_expect_put() are no
   longer required.

7) Remove reference count from expectations, and don't free when we
   fail ip_conntrack_expect_related(): have user call
   ip_conntrack_expect_free().
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes up the other helpers for direct conntrack->NAT helper calling.
SNMP doesn't really need a conntrack helper, but under this new model,
the NAT helper will register at that point anyway: NAT helpers
themselves are removed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently connection tracking and NAT helper modules for a protocol
interact only indirectly (the conntrack module places information in
the conntrack structure, which the NAT module pulls out).

This leads to several issues:
1) Both modules must know what port to watch, and must match.
2) Identifying the particular packet which created the connection
   is cumbersome (TCP) or impossible (UDP).
3) The connection tracking code sets up an expectation which the
   NAT code then has to change.
4) The lack of direct symbol dependencies means we have to contrive
   one, since they are functionally dependent.

Here is the current code flow:
FTP CONTROL PACKET:
NF_IP_PRE_ROUTING:
   ip_conntrack_in
      resolve_normal_ct
         init_conntrack: sets ct->helper to ip_conntrack_ftp.c:help()
   ct->help(): if PORT/PASV command:
      Sets exp->help.exp_ftp_info to tcp seq number of data.
      ip_conntrack_expect(): expects the connection

   ip_nat_setup_info: sets ct->nat.info->helper to ip_nat_ftp.c:help()
   ip_nat_fn:
      proto->exp_matches_pkt: if packet matches expectation
      ct->nat.info->helper(): If packet going client->server,
            and packet data is one in ct_ftp_info:
         ftp_data_fixup():
            ip_conntrack_change_expect(): change the expectation
            Modify packet contents with new address.

NF_IP_POST_ROUTING:
   ip_nat_fn
      ct->nat.info->helper(): If packet going server->client,
            and packet data is one in ct_ftp_info:
         ftp_data_fixup():
            ip_conntrack_change_expect(): change the expectation
            Modify packet contents with new address.

FTP DATA (EXPECTED) CONNECTION FIRST PACKET:
NF_IP_PRE_ROUTING:
   ip_conntrack_in
      resolve_normal_ct
         init_conntrack: set ct->master.
   ip_nat_fn:
      master->nat.info.helper->expect()
         Set up source NAT mapping to match FTP control connection.

NF_IP_PRE_ROUTING:
   ip_nat_fn:
      master->nat.info.helper->expect()
         Set up dest NAT mapping to match FTP control connection.


The new flow looks like this:
FTP CONTROL PACKET:
NF_IP_PRE_ROUTING:
   ip_conntrack_in
      resolve_normal_ct
         init_conntrack: sets ct->helper to ip_conntrack_ftp.c:help()

NF_IP_POST_ROUTING:
   ip_confirm:
      ct->helper->help:
         If !ip_nat_ftp_hook: ip_conntrack_expect().
         ip_nat_ftp: 
            set exp->oldproto to old port.
            ip_conntrack_change_expect(): change the expectation
            set exp->expectfn to ftp_nat_expected.
            Modify packet contents with new address.

FTP DATA (EXPECTED) CONNECTION FIRST PACKET:
NF_IP_PRE_ROUTING:
   ip_conntrack_in
      resolve_normal_ct
         init_conntrack: set ct->master.
         call exp->expectfn (ftp_nat_expected):
             call ip_nat_follow_master().

The big changes are that the ip_nat_ftp module sets ip_conntrack_ftp's
ip_nat_ftp_hook when it initializes, so it calls the NAT code directly
when a packet containing the expect information is found by the
conntrack helper: and this interface can carry all the information
these two want to share.  Also, that conntrack helper is called as the
packet leaves the box, so there are no issues with expectations being
set up before the packet has been filtered.  The NAT helper doesn't
need to register and duplicate the conntrack ports.

The other trick is ip_nat_follow_master(), which does the NAT setup
all at once (source and destination NAT as required) such that the
expected connection is NATed the same way the master connection
was.

We also call ip_conntrack_tcp_update() (which I incidentally neatened)
after mangling a TCP packet; ip_nat_seq_adjust() does this, but now
mangling is done at the last possible moment, after
ip_nat_seq_adjust() was already called.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Change kmem_cache_free() calls in ip_conntrack_expect_related() to
ip_conntrack_expect_put(): they should be equivalent but allows a hack
in next patch (caller can keep expect).

More importantly, a previous expectation should only be refreshed and return
EEXIST if it's owned by the same connection (nfsim found this bug).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Arthur Kepner <akepner@sgi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

 - put a dev_id field in struct net_device, so that it uses space that
   would be wasted by padding otherwise.
 - if this fields is non-null let ipv6_generate_eui64 use the algorithm
   from the QETH code to generate an EUI that's different for each
   OS instance.  See code comments for details.
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

This makes the skb->truesize modifications always OK.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

into nuts.davemloft.net:/disk1/BK/net-2.6

into nuts.davemloft.net:/disk1/BK/net-2.6

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

RPC: Fix a module refcount leak in RPCSEC_GSS
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>