In old kernels, NET_SKB_PAD was defined to 16.

Then commit d6301d3d (net: Increase default NET_SKB_PAD to 32), and
commit 18e8c134 (net: Increase NET_SKB_PAD to 64 bytes) increased it
to 64.

While first patch was governed by network stack needs, second was more
driven by performance issues on current hardware. Real intent was to
align data on a cache line boundary.

So use max(32, L1_CACHE_BYTES) instead of 64, to be more generic.

Remove microblaze and powerpc own NET_SKB_PAD definitions.

Thanks to Alexander Duyck and David Miller for their comments.
Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Third param (work) is unused, remove it.

Remove __inline__ and inline qualifiers.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Instead of doing one atomic operation per frag, we can factorize them.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When syncookies are in effect, req->iif is left uninitialized.
In case of e.g. link-local addresses the route lookup then fails
and no syn-ack is sent.

Rearrange things so ->iif is also initialized in the syncookie case.

want_cookie can only be true when the isn was zero, thus move the want_cookie
check into the "!isn" branch.

Cc: Glenn Griffin <ggriffin.kernel@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

ndo_get_stats still returns struct net_device_stats *; there is
no struct net_device_stats64.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

SKBs hold onto resources that can't be held indefinitely, such as TCP
socket references and netfilter conntrack state.  So if a packet is left
in TX ring for a long time, there might be a TCP socket that cannot be
closed and freed up.

Current blackfin EMAC driver always reclaim and free used tx skbs in future
transfers. The problem is that future transfer may not come as soon as
possible. This patch start a timer after transfer to reclaim and free skb.
There is nearly no performance drop with this patch.

TX interrupt is not enabled because of a strange behavior of the Blackfin EMAC.
If EMAC TX transfer control is turned on, endless TX interrupts are triggered
no matter if TX DMA is enabled or not. Since DMA walks down the ring automatically,
TX transfer control can't be turned off in the middle. The only way is to disable
TX interrupt completely.
Signed-off-by: Sonic Zhang <sonic.zhang@analog.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

inetpeer currently uses an AVL tree protected by an rwlock.

It's possible to make most lookups use RCU

1) Add a struct rcu_head to struct inet_peer

2) add a lookup_rcu_bh() helper to perform lockless and opportunistic
lookup. This is a normal function, not a macro like lookup().

3) Add a limit to number of links followed by lookup_rcu_bh(). This is
needed in case we fall in a loop.

4) add an smp_wmb() in link_to_pool() right before node insert.

5) make unlink_from_pool() use atomic_cmpxchg() to make sure it can take
last reference to an inet_peer, since lockless readers could increase
refcount, even while we hold peers.lock.

6) Delay struct inet_peer freeing after rcu grace period so that
lookup_rcu_bh() cannot crash.

7) inet_getpeer() first attempts lockless lookup.
   Note this lookup can fail even if target is in AVL tree, but a
concurrent writer can let tree in a non correct form.
   If this attemps fails, lock is taken a regular lookup is performed
again.

8) convert peers.lock from rwlock to a spinlock

9) Remove SLAB_HWCACHE_ALIGN when peer_cachep is created, because
rcu_head adds 16 bytes on 64bit arches, doubling effective size (64 ->
128 bytes)
In a future patch, this is probably possible to revert this part, if rcu
field is put in an union to share space with rid, ip_id_count, tcp_ts &
tcp_ts_stamp. These fields being manipulated only with refcnt > 0.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix the code that handles the error case when cnic_cm_abort() cannot
proceed normally.  We cannot just set the csk->state and we must
go through cnic_ready_to_close() to handle all the conditions.  We
also add error return code in cnic_cm_abort().
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Eddie Wai <waie@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Combine RESET_RECEIVED and RESET_COMP logic and fix race condition
between these 2 events and cnic_cm_close().  In particular, we need
to (test_and_clear_bit(SK_F_OFFLD_COMPLETE, &csk->flags)) before we
update csk->state.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Eddie Wai <waie@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Move chip-specific code to the respective chip's ->close_conn() functions
for better code organization.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Eddie Wai <waie@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

So that bnx2i can handle the error condition immediately and not have to
wait for timeout.

Signed-off-by: Michael Chan <mchan@broadcom.com.
Signed-off-by: Eddie Wai <waie@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This change corrects issues where macvlan was not correctly triggering
promiscuous mode on ixgbe due to the filters not being correctly set.  It
also corrects the fact that VF rar filters were being overwritten when the
PF was reset.

CC: Shirley Ma <xma@us.ibm.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

unify tcp flag macros: TCPHDR_FIN, TCPHDR_SYN, TCPHDR_RST, TCPHDR_PSH,
TCPHDR_ACK, TCPHDR_URG, TCPHDR_ECE and TCPHDR_CWR. TCBCB_FLAG_* are replaced
with the corresponding TCPHDR_*.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
 include/net/tcp.h                      |   24 ++++++-------
 net/ipv4/tcp.c                         |    8 ++--
 net/ipv4/tcp_input.c                   |    2 -
 net/ipv4/tcp_output.c                  |   59 ++++++++++++++++-----------------
 net/netfilter/nf_conntrack_proto_tcp.c |   32 ++++++-----------
 net/netfilter/xt_TCPMSS.c              |    4 --
 6 files changed, 58 insertions(+), 71 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>

Register net_bridge_port pointer as rx_handler data pointer. As br_port is
removed from struct net_device, another netdev priv_flag is added to indicate
the device serves as a bridge port. Also rcuized pointers are now correctly
dereferenced in br_fdb.c and in netfilter parts.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Register macvlan_port pointer as rx_handler data pointer. As macvlan_port is
removed from struct net_device, another netdev priv_flag is added to indicate
the device serves as a macvlan port.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add possibility to register rx_handler data pointer along with a rx_handler.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There are multiple problems with the newly added netpoll support:

1) Use-after-free on each netpoll packet.
2) Invoking unsafe code on netpoll/IRQ path.
3) Breaks when netpoll is enabled on the underlying device.

This patch fixes all of these problems.  In particular, we now
allocate proper netpoll structures for each underlying device.

We only allow netpoll to be enabled on the bridge when all the
devices underneath it support netpoll.  Once it is enabled, we
do not allow non-netpoll devices to join the bridge (until netpoll
is disabled again).

This allows us to do away with the npinfo juggling that caused
problem number 1.

Incidentally this patch fixes number 2 by bypassing unsafe code
such as multicast snooping and netfilter.
Reported-by: Qianfeng Zhang <frzhang@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds the helper netpoll_tx_running for use within
ndo_start_xmit.  It returns non-zero if ndo_start_xmit is being
invoked by netpoll, and zero otherwise.

This is currently implemented by simply looking at the hardirq
count.  This is because for all non-netpoll uses of ndo_start_xmit,
IRQs must be enabled while netpoll always disables IRQs before
calling ndo_start_xmit.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds the functions __netpoll_setup/__netpoll_cleanup
which is designed to be called recursively through ndo_netpoll_seutp.

They must be called with RTNL held, and the caller must initialise
np->dev and ensure that it has a valid reference count.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds ndo_netpoll_setup as the initialisation primitive
to complement ndo_netpoll_cleanup.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

As it stands, netpoll_setup and netpoll_cleanup have no locking
protection whatsoever.  So chaos ensures if two entities try to
perform them on the same device.

This patch adds RTNL to the equation.  The code has been rearranged so
that bits that do not need RTNL protection are now moved to the top of
netpoll_setup.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

The use of RCU in netpoll is incorrect in a number of places:

1) The initial setting is lacking a write barrier.
2) The synchronize_rcu is in the wrong place.
3) Read barriers are missing.
4) Some places are even missing rcu_read_lock.
5) npinfo is zeroed after freeing.

This patch fixes those issues.  As most users are in BH context,
this also converts the RCU usage to the BH variant.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Now that netpoll always zaps npinfo we no longer need to do it
in bridge.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since we have to NULL npinfo regardless of whether there is a
ndo_netpoll_cleanup, it makes sense to do this unconditionally
in netpoll_cleanup rather than having every driver do it by
themselves.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Conflicts:
	include/net/netfilter/xt_rateest.h
	net/bridge/br_netfilter.c
	net/netfilter/nf_conntrack_core.c
Signed-off-by: Patrick McHardy <kaber@trash.net>

This patch implements an idletimer Xtables target that can be used to
identify when interfaces have been idle for a certain period of time.

Timers are identified by labels and are created when a rule is set with a new
label.  The rules also take a timeout value (in seconds) as an option.  If
more than one rule uses the same timer label, the timer will be restarted
whenever any of the rules get a hit.

One entry for each timer is created in sysfs.  This attribute contains the
timer remaining for the timer to expire.  The attributes are located under
the xt_idletimer class:

/sys/class/xt_idletimer/timers/<label>

When the timer expires, the target module sends a sysfs notification to the
userspace, which can then decide what to do (eg. disconnect to save power).

Cc: Timo Teras <timo.teras@iki.fi>
Signed-off-by: Luciano Coelho <luciano.coelho@nokia.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

- clusterip_lock becomes a spinlock
- lockless lookups
- kfree() deferred after RCU grace period
- rcu_barrier_bh() inserted in clusterip_tg_exit()

v2)
- As Patrick pointed out, we use atomic_inc_not_zero() in
clusterip_config_find_get().
- list_add_rcu() and list_del_rcu() variants are used.
- atomic_dec_and_lock() used in clusterip_config_entry_put()
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

Change 2wire transfer rate of SFP+ module EEPROM from 400Khz to 100Khz
since some DACs(direct attached cables) do not work at 400Khz.
Reported-by: Krzysztof Oldzki <ole@ans.pl>
Signed-off-by: Yaniv Rosner <yanivr@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Try to reduce cache line contentions in peer management, to reduce IP
defragmentation overhead.

- peer_fake_node is marked 'const' to make sure its not modified.
  (tested with CONFIG_DEBUG_RODATA=y)

- Group variables in two structures to reduce number of dirtied cache
lines. One named "peers" for avl tree root, its number of entries, and
associated lock. (candidate for RCU conversion)

- A second one named "unused_peers" for unused list and its lock

- Add a !list_empty() test in unlink_from_unused() to avoid taking lock
when entry is not unused.

- Use atomic_dec_and_lock() in inet_putpeer() to avoid taking lock in
some cases.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Uses a seqcount_t to synchronize stat producer and consumer, for packets
and bytes counter, now u64 types.

(dropped counter being rarely used, stay a native "unsigned long" type)

No noticeable performance impact on x86, as it only adds two increments
per frame. It might be more expensive on arches where smp_wmb() is not
free.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use RCU to avoid atomic ops on idev refcnt in ipv6_get_mtu()
and ip6_dst_hoplimit()
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use __in6_dev_get() instead of in6_dev_get()/in6_dev_put()
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

commit 97f8aefb "net: fix ethtool
coding style errors and warnings" changed the indentation of several
macro definitions in ethtool.h.  These definitions line up in the diff
where there is an extra character at the start of each line, but not
in the resulting file.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Conflicts:
	drivers/net/ixgbe/ixgbe_ethtool.c

With merge conflict help from Stephen Rothwell.
Signed-off-by: David S. Miller <davem@davemloft.net>

The parameter (work) is unused, remove it.
Reported from Eric Dumazet.
Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

Instead of doing one atomic operation per frag, we can factorize them.
Reported from Eric Dumazet.
Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

If the returned csum value is 0, We has set ip_summed with
CHECKSUM_UNNECESSARY flag in __skb_checksum_complete_head().

So this patch kills the check and changes to return to upper
caller directly.
Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

- must use atomic_inc_not_zero() in instance_lookup_get()

- must use hlist_add_head_rcu() instead of hlist_add_head()

- must use hlist_del_rcu() instead of hlist_del()

- Introduce NFULNL_COPY_DISABLED to stop lockless reader from using an
instance, before we do final instance_put() on it.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

This patch added to 2.6.34:

	commit f8d1dcaf
	Author: Jesse Brandeburg <jesse.brandeburg@intel.com>
	Date:   Tue Apr 27 01:37:20 2010 +0000

	    ixgbe: enable extremely low latency

introduced a feature where LRO (called RSC on the hardware) was disabled
automatically when setting rx-usecs to 0 via ethtool.  Some might not
like the fact that LRO was disabled automatically, but I'm fine with
that.  What I don't like is that LRO/RSC is automatically enabled when
rx-usecs is set >0 via ethtool.

This would certainly be a problem if the device was used for forwarding
and it was determined that the low latency wasn't needed after the
device was already forwarding.  I played around with saving the state of
LRO in the driver, but it just didn't seem worthwhile and would require
a small change to dev_disable_lro() that I did not like.

This patch simply leaves LRO disabled when setting rx-usecs >0 and
requires that the user enable it again.  An extra informational message
will also now appear in the log so users can understand why LRO isn't
being enabled as they expect.

Inconsistency of LRO setting first noticed by Stanislaw Gruszka.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
CC: Stanislaw Gruszka <sgruszka@redhat.com>
CC: stable@kernel.org
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>