[IPV6]: Fix ICMP relookup error path dst leak

When we encounter an error while looking up the dst the second
time we need to drop the first dst.  This patch is pretty much
the same as the one for IPv4.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM] drivers/atm/iphase.c: compilation warning fix

Removed the warning messages:
drivers/atm/iphase.c:961: warning: 'tcnter' defined but not used
drivers/atm/iphase.c:963: warning: 'xdump' defined but not used

tcnter and xdump() are used only in debug build

Signed-off-by: Leonardo Potenza <lpotenza@inwind.it>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: do not create temporary adresses with too short preferred lifetime

From RFC341:
A temporary address is created only if this calculated Preferred
Lifetime is greater than REGEN_ADVANCE time units.  In particular, an
implementation must not create a temporary address with a zero
Preferred Lifetime.

Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: only update the lifetime of the relevant temporary address

When receiving a prefix information from a routeur, only update the
lifetimes of the temporary address associated with that prefix.

Otherwise if one deprecated prefix is advertized, all your temporary
addresses will become deprecated.

Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

bluetooth : __rfcomm_dlc_close lock fix

Lockdep warning will be trigged while rfcomm connection closing.

The locks taken in rfcomm_dev_add:
rfcomm_dev_lock --> d->lock

In __rfcomm_dlc_close:
d->lock --> rfcomm_dev_lock (in rfcomm_dev_state_change)

There's two way to fix it, one is in rfcomm_dev_add we first locking
d->lock then the rfcomm_dev_lock

The other (in this patch), remove the locking of d->lock for
rfcomm_dev_state_change because just locking "d->state = BT_CLOSED;"
is enough.

[  295.002046] =======================================================
[  295.002046] [ INFO: possible circular locking dependency detected ]
[  295.002046] 2.6.25-rc7 #1
[  295.002046] -------------------------------------------------------
[  295.002046] krfcommd/2705 is trying to acquire lock:
[  295.002046]  (rfcomm_dev_lock){-.--}, at: [<f89a090a>] rfcomm_dev_state_change+0x6a/0xd0 [rfcomm]
[  295.002046]
[  295.002046] but task is already holding lock:
[  295.002046]  (&d->lock){--..}, at: [<f899c533>] __rfcomm_dlc_close+0x43/0xd0 [rfcomm]
[  295.002046]
[  295.002046] which lock already depends on the new lock.
[  295.002046]
[  295.002046]
[  295.002046] the existing dependency chain (in reverse order) is:
[  295.002046]
[  295.002046] -> #1 (&d->lock){--..}:
[  295.002046]        [<c0149b23>] check_prev_add+0xd3/0x200
[  295.002046]        [<c0149ce5>] check_prevs_add+0x95/0xe0
[  295.002046]        [<c0149f6f>] validate_chain+0x23f/0x320
[  295.002046]        [<c014b7b1>] __lock_acquire+0x1c1/0x760
[  295.002046]        [<c014c349>] lock_acquire+0x79/0xb0
[  295.002046]        [<c03d6b99>] _spin_lock+0x39/0x80
[  295.002046]        [<f89a01c0>] rfcomm_dev_add+0x240/0x360 [rfcomm]
[  295.002046]        [<f89a047e>] rfcomm_create_dev+0x6e/0xe0 [rfcomm]
[  295.002046]        [<f89a0823>] rfcomm_dev_ioctl+0x33/0x60 [rfcomm]
[  295.002046]        [<f899facc>] rfcomm_sock_ioctl+0x2c/0x50 [rfcomm]
[  295.002046]        [<c0363d38>] sock_ioctl+0x118/0x240
[  295.002046]        [<c0194196>] vfs_ioctl+0x76/0x90
[  295.002046]        [<c0194446>] do_vfs_ioctl+0x56/0x140
[  295.002046]        [<c0194569>] sys_ioctl+0x39/0x60
[  295.002046]        [<c0104faa>] syscall_call+0x7/0xb
[  295.002046]        [<ffffffff>] 0xffffffff
[  295.002046]
[  295.002046] -> #0 (rfcomm_dev_lock){-.--}:
[  295.002046]        [<c0149a84>] check_prev_add+0x34/0x200
[  295.002046]        [<c0149ce5>] check_prevs_add+0x95/0xe0
[  295.002046]        [<c0149f6f>] validate_chain+0x23f/0x320
[  295.002046]        [<c014b7b1>] __lock_acquire+0x1c1/0x760
[  295.002046]        [<c014c349>] lock_acquire+0x79/0xb0
[  295.002046]        [<c03d6639>] _read_lock+0x39/0x80
[  295.002046]        [<f89a090a>] rfcomm_dev_state_change+0x6a/0xd0 [rfcomm]
[  295.002046]        [<f899c548>] __rfcomm_dlc_close+0x58/0xd0 [rfcomm]
[  295.002046]        [<f899d44f>] rfcomm_recv_ua+0x6f/0x120 [rfcomm]
[  295.002046]        [<f899e061>] rfcomm_recv_frame+0x171/0x1e0 [rfcomm]
[  295.002046]        [<f899e357>] rfcomm_run+0xe7/0x550 [rfcomm]
[  295.002046]        [<c013c18c>] kthread+0x5c/0xa0
[  295.002046]        [<c0105c07>] kernel_thread_helper+0x7/0x10
[  295.002046]        [<ffffffff>] 0xffffffff
[  295.002046]
[  295.002046] other info that might help us debug this:
[  295.002046]
[  295.002046] 2 locks held by krfcommd/2705:
[  295.002046]  #0:  (rfcomm_mutex){--..}, at: [<f899e2eb>] rfcomm_run+0x7b/0x550 [rfcomm]
[  295.002046]  #1:  (&d->lock){--..}, at: [<f899c533>] __rfcomm_dlc_close+0x43/0xd0 [rfcomm]
[  295.002046]
[  295.002046] stack backtrace:
[  295.002046] Pid: 2705, comm: krfcommd Not tainted 2.6.25-rc7 #1
[  295.002046]  [<c0128a38>] ? printk+0x18/0x20
[  295.002046]  [<c014927f>] print_circular_bug_tail+0x6f/0x80
[  295.002046]  [<c0149a84>] check_prev_add+0x34/0x200
[  295.002046]  [<c0149ce5>] check_prevs_add+0x95/0xe0
[  295.002046]  [<c0149f6f>] validate_chain+0x23f/0x320
[  295.002046]  [<c014b7b1>] __lock_acquire+0x1c1/0x760
[  295.002046]  [<c014c349>] lock_acquire+0x79/0xb0
[  295.002046]  [<f89a090a>] ? rfcomm_dev_state_change+0x6a/0xd0 [rfcomm]
[  295.002046]  [<c03d6639>] _read_lock+0x39/0x80
[  295.002046]  [<f89a090a>] ? rfcomm_dev_state_change+0x6a/0xd0 [rfcomm]
[  295.002046]  [<f89a090a>] rfcomm_dev_state_change+0x6a/0xd0 [rfcomm]
[  295.002046]  [<f899c548>] __rfcomm_dlc_close+0x58/0xd0 [rfcomm]
[  295.002046]  [<f899d44f>] rfcomm_recv_ua+0x6f/0x120 [rfcomm]
[  295.002046]  [<f899e061>] rfcomm_recv_frame+0x171/0x1e0 [rfcomm]
[  295.002046]  [<c014abd9>] ? trace_hardirqs_on+0xb9/0x130
[  295.002046]  [<c03d6e89>] ? _spin_unlock_irqrestore+0x39/0x70
[  295.002046]  [<f899e357>] rfcomm_run+0xe7/0x550 [rfcomm]
[  295.002046]  [<c03d4559>] ? __sched_text_start+0x229/0x4c0
[  295.002046]  [<c0120000>] ? cpu_avg_load_per_task+0x20/0x30
[  295.002046]  [<f899e270>] ? rfcomm_run+0x0/0x550 [rfcomm]
[  295.002046]  [<c013c18c>] kthread+0x5c/0xa0
[  295.002046]  [<c013c130>] ? kthread+0x0/0xa0
[  295.002046]  [<c0105c07>] kernel_thread_helper+0x7/0x10
[  295.002046]  =======================

Signed-off-by: Dave Young <hidave.darkstar@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bluetooth : use lockdep sub-classes for diffrent bluetooth protocol

'rfcomm connect' will trigger lockdep warnings which is caused by
locking diffrent kinds of bluetooth sockets at the same time.

So using sub-classes per AF_BLUETOOTH sub-type for lockdep.

Thanks for the hints from dave jones.

---
> From: Dave Jones <davej@codemonkey.org.uk>
> Date: Thu, 27 Mar 2008 12:21:56 -0400
>
> > Mar 27 08:10:57 localhost kernel: Pid: 3611, comm: obex-data-serve Not tainted 2.6.25-0.121.rc5.git4.fc9 #1
> > Mar 27 08:10:57 localhost kernel:  [__lock_acquire+2287/3089] __lock_acquire+0x8ef/0xc11
> > Mar 27 08:10:57 localhost kernel:  [sched_clock+8/11] ? sched_clock+0x8/0xb
> > Mar 27 08:10:57 localhost kernel:  [lock_acquire+106/144] lock_acquire+0x6a/0x90
> > Mar 27 08:10:57 localhost kernel:  [<f8bd9321>] ? l2cap_sock_bind+0x29/0x108 [l2cap]
> > Mar 27 08:10:57 localhost kernel:  [lock_sock_nested+182/198] lock_sock_nested+0xb6/0xc6
> > Mar 27 08:10:57 localhost kernel:  [<f8bd9321>] ? l2cap_sock_bind+0x29/0x108 [l2cap]
> > Mar 27 08:10:57 localhost kernel:  [security_socket_post_create+22/27] ? security_socket_post_create+0x16/0x1b
> > Mar 27 08:10:57 localhost kernel:  [__sock_create+388/472] ? __sock_create+0x184/0x1d8
> > Mar 27 08:10:57 localhost kernel:  [<f8bd9321>] l2cap_sock_bind+0x29/0x108 [l2cap]
> > Mar 27 08:10:57 localhost kernel:  [kernel_bind+10/13] kernel_bind+0xa/0xd
> > Mar 27 08:10:57 localhost kernel:  [<f8dad3d7>] rfcomm_dlc_open+0xc8/0x294 [rfcomm]
> > Mar 27 08:10:57 localhost kernel:  [lock_sock_nested+187/198] ? lock_sock_nested+0xbb/0xc6
> > Mar 27 08:10:57 localhost kernel:  [<f8dae18c>] rfcomm_sock_connect+0x8b/0xc2 [rfcomm]
> > Mar 27 08:10:57 localhost kernel:  [sys_connect+96/125] sys_connect+0x60/0x7d
> > Mar 27 08:10:57 localhost kernel:  [__lock_acquire+1370/3089] ? __lock_acquire+0x55a/0xc11
> > Mar 27 08:10:57 localhost kernel:  [sys_socketcall+140/392] sys_socketcall+0x8c/0x188
> > Mar 27 08:10:57 localhost kernel:  [syscall_call+7/11] syscall_call+0x7/0xb
---

Signed-off-by: Dave Young <hidave.darkstar@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ROSE/AX25] af_rose: rose_release() fix

rose_release() doesn't release sockets properly, e.g. it skips
sock_orphan(), so OOPSes are triggered in sock_def_write_space(),
which was observed especially while ROSE skbs were kfreed from
ax25_frames_acked(). There is also sock_hold() and lock_sock() added -
similarly to ax25_release(). Thanks to Bernard Pidoux for substantial
help in debugging this problem.

Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Reported-and-tested-by: Bernard Pidoux <bpidoux@free.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: correct use_short_preamble handling

ERP IE bit for preamble mode is 0 for short and 1 for long, not the other
way around. This fixes the value reported to the driver via
bss_conf->use_short_preamble field.

Signed-off-by: Vladimir Koutny <vlado@ksp.sk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

b43: Fix PCMCIA IRQ routing

This fixes the IRQ routing on PCMCIA devices.
With this patch the card will finally be able to receive IRQs.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

b43: Add DMA mapping failure messages

This adds messages for some DMA mapping failures.
These are useful for debugging DMA address problems, as they appear
on x86_64 machines with IOMMU enabled.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: trigger ieee80211_sta_work after opening interface

ieee80211_sta_work is disabled while network interface
is down. Therefore, if you configure wireless parameters
before bringing the interface up, these configurations are
not yet effective and association fails.

A workaround from userspace is calling a command like
'iwconfig wlan0 ap any' after the interface is brought up.

To fix this behaviour, trigger execution of ieee80211_sta_work from
ieee80211_open when in STA or IBSS mode.

Signed-off-by: Jan Niehusmann <jan@gondor.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

[LLC]: skb allocation size for responses

Allocate the skb for llc responses with the received packet size by
using the size adjustable llc_frame_alloc.
Don't allocate useless extra payload.
Cleanup magic numbers.

So, this fixes oops.
Reported by Jim Westfall:
kernel: skb_over_panic: text:c0541fc7 len:1000 put:997 head:c166ac00 data:c166ac2f tail:0xc166b017 end:0xc166ac80 dev:eth0
kernel: ------------[ cut here ]------------
kernel: kernel BUG at net/core/skbuff.c:95!

Signed-off-by: Joonwoo Park <joonwpark81@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IP] UDP: Use SEQ_START_TOKEN.

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

[NET]: Remove Documentation/networking/sk98lin.txt

Since the driver is gone there's no point in keeping the documentation.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM] atm/idt77252.c: Make 2 functions static

This patch makes the following needlessly global functions static:
- idt77252_send()
- idt77252_dev_close()

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

[ATM]: Make atm/he.c:read_prom_byte() static

This patch makes the needlessly global read_prom_byte() static.

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

[IPV6] MCAST: Ensure to check multicast listener(s).

In ip6_mc_input(), we need to check whether we have listener(s) for
the packet.

After commit ae7bf20a6316272acfcaef5d265b18aaa54b41e4, all packets
for multicast destinations are delivered to upper layer if
IFF_PROMISC or IFF_ALLMULTI is set.

In fact, bug was rather ancient; the original (before the commit)
intent of the dev->flags check was to skip the ipv6_chk_mcast_addr()
call, assuming L2 filters packets appropriately, but it was even not
true.

Let's explicitly check our multicast list.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Acked-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[LLC]: Kill llc_station_mac_sa symbol export.

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

Merge branch 'upstream-davem' of /linux/kernel/git/jgarzik/netdev-2.6

forcedeth: fix locking bug with netconsole

While using netconsole on forcedeth, lockdep noticed the following locking
bug:

=================================
[ INFO: inconsistent lock state ]
2.6.24-rc6 #6
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---------------------------------
inconsistent {softirq-on-W} -> {in-softirq-W} usage.
udevd/719 [HC0[0]:SC1[1]:HE1:SE0] takes:
 (_xmit_ETHER){-+..}, at: [<c043062e>] dev_watchdog+0x1c/0xb9
{softirq-on-W} state was registered at:
  [<c0147f67>] mark_held_locks+0x4e/0x66
  [<c014810e>] trace_hardirqs_on+0xfe/0x136
  [<c048ae63>] _spin_unlock_irq+0x22/0x42
  [<c02ec617>] nv_start_xmit_optimized+0x347/0x37a
  [<c042c80d>] netpoll_send_skb+0xa4/0x147
  [<c042d4a6>] netpoll_send_udp+0x238/0x242
  [<c02f44f6>] write_msg+0x6d/0x9b
  [<c012c129>] __call_console_drivers+0x4e/0x5a
  [<c012c18c>] _call_console_drivers+0x57/0x5b
  [<c012c2dd>] release_console_sem+0x11c/0x1b9
  [<c012caeb>] register_console+0x1eb/0x1f3
  [<c06ae673>] init_netconsole+0x119/0x15f
  [<c069149b>] kernel_init+0x147/0x294
  [<c01058cb>] kernel_thread_helper+0x7/0x10
  [<ffffffff>] 0xffffffff
irq event stamp: 950
hardirqs last  enabled at (950): [<c048ae63>] _spin_unlock_irq+0x22/0x42
hardirqs last disabled at (949): [<c048aaf7>] _spin_lock_irq+0xc/0x38
softirqs last  enabled at (0): [<c012a29c>] copy_process+0x375/0x126d
softirqs last disabled at (947): [<c0106d43>] do_softirq+0x61/0xc6

other info that might help us debug this:
no locks held by udevd/719.

stack backtrace:
Pid: 719, comm: udevd Not tainted 2.6.24-rc6 #6
 [<c0105c46>] show_trace_log_lvl+0x12/0x25
 [<c01063ec>] show_trace+0xd/0x10
 [<c010670c>] dump_stack+0x57/0x5f
 [<c0147505>] print_usage_bug+0x10a/0x117
 [<c0147c38>] mark_lock+0x121/0x402
 [<c01488b6>] __lock_acquire+0x3d1/0xb64
 [<c0149405>] lock_acquire+0x4e/0x6a
 [<c048a99b>] _spin_lock+0x23/0x32
 [<c043062e>] dev_watchdog+0x1c/0xb9
 [<c0133e4a>] run_timer_softirq+0x133/0x193
 [<c0130907>] __do_softirq+0x78/0xed
 [<c0106d43>] do_softirq+0x61/0xc6
 =======================
eth1: link down

The fix is to disable/restore irqs instead of disable/enable.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Ayaz Abdulla <aabdulla@nvidia.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

usb net: asix does not really need 10/100mbit

The asix usb driver currently depends on NET_ETHERNET which means you
cannot enable this driver if you only have 1000mbit enabled in your kernel.
Since there is no real dependency between the NET_ETHERNET portion and the
asix driver, simply drop it.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Greg KH <greg@kroah.com>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

Blackfin EMAC Driver: delete unused variables to fixup gcc warnings

Signed-off-by: Mike Frysinger <vapier.adi@gmail.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

ibm_newemac: emac_tx_csum typo fix

Move the "&& skb->ip_summed == CHECKSUM_PARTIAL" part out of
emac_has_feature parameters.

Signed-off-by: Valentine Barshak <vbarshak@ru.mvista.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

Make pasemi_mac.c depend on PPC_PASEMI to prevent link errors

drivers/net/pasemi_mac.c is enabled by CONFIG_PASEMI_MAC, which depends on
PPC64 && PCI. However pasemi_mac.c uses several routines that are only
built when PPC_PASEMI is selected. This can lead to an unbuildable config:

 ERROR: ".pasemi_dma_start_chan" [drivers/net/pasemi_mac.ko] undefined!

So make CONFIG_PASEMI_MAC depend on PPC_PASEMI instead of PPC64.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

S2io: Version update for Tx completion patch

- Updated version number.
- Resubmitting with correct version update.
- this patch to be applied for upstream-davem branch

Signed-off-by: Surjit Reang <surjit.reang@neterion.com>
Signed-off-by: Sreenivasa Honnur <sreenivasa.honnur@neterion.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

drivers/net/usb/pegasus.c: remove in_atomic() check

Remove superfluous in-atomic() check; ethtool MII ops are called from task
context.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

dm9601: Fix multicast hash table handling

The loop forgot to walk the net->mc_list list, so only the first
multicast address was programmed into the hash table.

Signed-off-by: Jeff Garzik <jeff@garzik.org>

[netdrvr] tulip_read_eeprom fixes for BUG 4420

If "location" is > "addr_len" bits, the high bits of location would interfere
with the READ_CMD sent to the eeprom controller.

A patch was submitted to bug:
    http://bugzilla.kernel.org/show_bug.cgi?id=4420

which simply truncated the "location", read whatever was in "location
modulo addr_len", and returned that value. That avoids confusing the
eeprom but seems like the wrong solution to me.

Correct would be to not read beyond "1 << addr_len" address of the eeprom.
I am submitting two changes to implement this:
1) tulip_read_eeprom will return zero (since we can't return -EINVAL)
   if this is attempted (defensive programming).
2) In tulip_core.c, fix the tulip_read_eeprom caller so they don't
   iterate past addr_len bits and make sure the entire tp->eeprom[]
   array is cleared.

I konw we don't strictly need both. I would prefer both in the tree
since it documents the issue and provides a second "defense" from
the bug from creeping back in.

Signed-off-by: Grant Grundler <grundler@parisc-linux.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

[INET]: inet_frag_evictor() must run with BH disabled

Based upon a lockdep trace from Dave Jones.

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

net: Comment dev_kfree_skb_irq and dev_kfree_skb_any better

Comment dev_kfree_skb_irq and dev_kfree_skb_any better.

Signed-off-by: Matti Linnanvuori <mattilinnanvuori@yahoo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[LLC]: station source mac address

kill unnecessary llc_station_mac_sa.

Signed-off-by: Joonwoo Park <joonwpark81@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[LLC]: bogus llc packet length

discard llc packet which has bogus packet length.

Signed-off-by: Joonwoo Park <joonwpark81@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Add preemption point in qdisc_run

The qdisc_run loop is currently unbounded and runs entirely in a
softirq.  This is bad as it may create an unbounded softirq run.

This patch fixes this by calling need_resched and breaking out if
necessary.

It also adds a break out if the jiffies value changes since that would
indicate we've been transmitting for too long which starves other
softirqs.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Don't send ICMP_FRAG_NEEDED for GSO packets

Commit 9af3912ec9e30509b76cb376abb65a4d8af27df3 ("[NET] Move DF check
to ip_forward") added a new check to send ICMP fragmentation needed
for large packets.

Unlike the check in ip_finish_output(), it doesn't check for GSO.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM] firestream: Fix uninitialized var warning.

All code paths set tmc0 in some way, but GCC can't
see that for some reason.  Explicitly initialize
to zero.

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

bluetooth: replace deprecated RW_LOCK_UNLOCKED macros

The older RW_LOCK_UNLOCKED macros defeat lockdep state tracing so
replace them with the newer __RW_LOCK_UNLOCKED macros.

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

lapb: use the shorter LIST_HEAD form for brevity

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[BOND]: Fix warning in bond_sysfs.c

original_mtu is only used if we end up with a non-NULL
dev, and it is assigned in all such cases, but GCC can't
see that.

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

[LLC]: Restrict LLC sockets to root

LLC currently allows users to inject raw frames, including IP packets
encapsulated in SNAP. While Linux doesn't handle IP over SNAP, other
systems do. Restrict LLC sockets to root similar to packet sockets.

[ Modified Patrick's patch to use CAP_NEW_RAW --DaveM ]

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

[NET]: Use local_irq_{save,restore}() in napi_complete().

Based upon a lockdep report.

Since ->poll() can be invoked from netpoll with interrupts
disabled, we must not unconditionally enable interrupts
in napi_complete().

Instead we must use local_irq_{save,restore}().

Noticed by Peter Zijlstra:

<irqs disabled>

  netpoll_poll()
    poll_napi()
      spin_trylock(&napi->poll_lock)
      poll_one_napi()
        napi->poll() := sky2_poll()
          napi_complete()
            local_irq_disable()
            local_irq_enable() <--- *BUG*

  <irq>
    irq_exit()
      do_softirq()
        net_rx_action()
          spin_lock(&napi->poll_lock) <--- Deadlock!

Because we still hold the lock....

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

[NETFILTER]: Replate direct proc_fops assignment with proc_create call.

This elliminates infamous race during module loading when one could lookup
proc entry without proc_fops assigned.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ESP]: Ensure IV is in linear part of the skb to avoid BUG() due to OOB access

ESP does not account for the IV size when calling pskb_may_pull() to
ensure everything it accesses directly is within the linear part of a
potential fragment. This results in a BUG() being triggered when the
both the IPv4 and IPv6 ESP stack is fed with an skb where the first
fragment ends between the end of the esp header and the end of the IV.

This bug was found by Dirk Nehring <dnehring@gmx.net> .

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

rt2x00: Ignore set_state(STATE_SLEEP) failure

Some hardware never seem to accept the "goto sleep" command, since the legacy
drivers don't have suspend and resume handlers the entire code for it was
basically a educated guess (based on the "enable radio" code).
This patch will only print a warning when the "goto sleep" command fails, and
just continues as usual. Perhaps that means the device will not reach a sleep
state and consumes more power then it should, but it is equally possible it
simply needs some seconds longer to sleep. Anyway, by making the command
non-fatal it will not block the rest of the suspend procedure.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

drivers/net/wireless/iwlwifi/iwl-4965.c: correct use of ! and &

In commit e6bafba5b4765a5a252f1b8d31cbf6d2459da337, a bug was fixed that
involved converting !x & y to !(x & y).  The code below shows the same
pattern, and thus should perhaps be fixed in the same way.

This is not tested and clearly changes the semantics, so it is only
something to consider.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@ expression E1,E2; @@
(
  !E1 & !E2
|
- !E1 & E2
+ !(E1 & E2)
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Cc: Tomas Winkler <tomas.winkler@intel.com>
Cc: Guy Cohen <guy.cohen@intel.com>
Cc: Reinette Chatre <reinette.chatre@intel.com>
Cc: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

libertas: fix spinlock recursion bug

This fixes a bug detected by CONFIG_DEBUG_SPINLOCK:

if_cs_get_int_status() is only called from lbs_thread(), via
priv->hw_get_int_status. However, lbs_thread() has already taken the
priv->driver_lock. So it's a fault to take the same lock again here.

Signed-off-by: Holger Schurig <hs4233@mail.mn-solutions.de>
Acked-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

[IPSEC]: Fix BEET output

The IPv6 BEET output function is incorrectly including the inner
header in the payload to be protected.  This causes a crash as
the packet doesn't actually have that many bytes for a second
header.

The IPv4 BEET output on the other hand is broken when it comes
to handling an inner IPv6 header since it always assumes an
inner IPv4 header.

This patch fixes both by making sure that neither BEET output
function touches the inner header at all.  All access is now
done through the protocol-independent cb structure.  Two new
attributes are added to make this work, the IP header length
and the IPv4 option length.  They're filled in by the inner
mode's output function.

Thanks to Joakim Koskela for finding this problem.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

[ICMP]: Dst entry leak in icmp_send host re-lookup code (v2).

Commit 8b7817f3a959ed99d7443afc12f78a7e1fcc2063 ([IPSEC]: Add ICMP host
relookup support) introduced some dst leaks on error paths: the rt
pointer can be forgotten to be put. Fix it bu going to a proper label.

Found after net namespace's lo refused to unregister :) Many thanks to
Den for valuable help during debugging.

Herbert pointed out, that xfrm_lookup() will put the rtable in case
of error itself, so the first goto fix is redundant.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[AX25]: Remove obsolete references to BKL from TODO file.

Given that there are no apparent calls to lock_kernel() or
unlock_kernel() under net/ax25, delete the TODO reference related to
that.

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Fix multicast device ioctl checks

SIOCADDMULTI/SIOCDELMULTI check whether the driver has a set_multicast_list
method to determine whether it supports multicast. Drivers implementing
secondary unicast support use set_rx_mode however.

Check for both dev->set_multicast_mode and dev->set_rx_mode to determine
multicast capabilities.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IRDA]: Store irnet_socket termios properly.

It should be a "struct ktermios" not a "struct termios".

Based upon a build warning reported by Stephen Rothwell.

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

[UML]: uml-net: don't set IFF_ALLMULTI in set_multicast_list

IFF_ALLMULTI is an indication from the network stack to the driver
to disable multicast filters, drivers should never set it directly.

Since the UML networking device doesn't have any filtering capabilites,
it doesn't the set_multicast_list function at all, it is kept so userspace
can still issue SIOCADDMULTI/SIOCDELMULTI ioctls however.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[VLAN]: Don't copy ALLMULTI/PROMISC flags from underlying device

Changing these flags requires to use dev_set_allmulti/dev_set_promiscuity
or dev_change_flags. Setting it directly causes two unwanted effects:

- the next dev_change_flags call will notice a difference between
  dev->gflags and the actual flags, enable promisc/allmulti
  mode and incorrectly update dev->gflags

- this keeps the underlying device in promisc/allmulti mode until
  the VLAN device is deleted

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

netxen, phy/marvell, skge: minor checkpatch fixes

Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

S2io: Handle TX completions on the same CPU as the sender for MIS-X interrupts

- Handling TX completions on the same cpu as the sender.

Signed-off-by: Surjit Reang <surjit.reang@neterion.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

b44: Truncate PHY address

Some ROMs on embedded devices store incorrect values for
the PHY address of the ethernet device.
It looks like the number is sign-extended.
Truncate the value by applying the PHY-address mask to it.
The patch was tested on a bcm47xx embedded system (where the bug
triggers) and a bcm4400 PCI card.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

skge napi->poll() locking bug

According to: Documentation/networking/netdevices.txt:

<cite>
napi->poll:
..........
Context: softirq
         will be called with interrupts disabled by netconsole.
</cite>

napi->poll() could be called either with interrupts enabled
(in softirq context) or disabled (by netconsole), so the irq flag
should be preserved.

Inspired by Ingo's resent forcedeth patch :-)

Signed-off-by: Marin Mitov <mitov@issp.bas.bg>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

rndis_host: fix oops when query for OID_GEN_PHYSICAL_MEDIUM fails

When query for OID_GEN_PHYSICAL_MEDIUM fails, uninitialized pointer
'phym' is being accessed in generic_rndis_bind(), resulting OOPS.
Patch fixes phym to be initialized and setup correctly when
rndis_query() for physical medium fails.

Bug was introduced by following commit:
commit 039ee17d1baabaa21783a0d5ab3e8c6d8c794bdf
Author: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Date:   Sun Jan 27 23:34:33 2008 +0200

Reported-by: Dmitri Monakhov <dmonakhov@openvz.org>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

cxgb3: Fix lockdep problems with sge.reg_lock

Using iWARP with a Chelsio T3 NIC generates the following lockdep warning:

    =================================
    [ INFO: inconsistent lock state ]
    2.6.25-rc6 #50
    ---------------------------------
    inconsistent {softirq-on-W} -> {in-softirq-W} usage.
    swapper/0 [HC0[0]:SC1[1]:HE0:SE0] takes:
     (&adap->sge.reg_lock){-+..}, at: [<ffffffff880e5ee2>] cxgb_offload_ctl+0x3af/0x507 [cxgb3]

The problem is that reg_lock is used with plain spin_lock() in
drivers/net/cxgb3/sge.c but is used with spin_lock_irqsave() in
drivers/net/cxgb3/cxgb3_offload.c.  This is technically a false
positive, since the uses in sge.c are only in the initialization and
cleanup paths and cannot overlap with any use in interrupt context.

The best fix is probably just to use spin_lock_irq() with reg_lock in
sge.c.  Even though it's not strictly required for correctness, it
avoids triggering lockdep and the extra overhead of disabling
interrupts is not important at all in the initialization and cleanup
slow paths.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

ehea: Fix IPv6 support

Indicate that HEA calculates IPv4 checksums only

Signed-off-by: Thomas Klein <tklein@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

dm9000: Support promisc and all-multi modes

Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

dm9601: configure MAC to drop invalid (crc/length) packets

Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

dm9601: add Hirose USB-100 device ID

The Hirose USB-100 adapter uses a dm9601 chip.
Reported by Robert Brockway.

Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

Marvell PHY m88e1111 driver fix

Marvell PHY m88e1111 (not sure about other models, but think they too)
works in two modes: fiber and copper. In Marvell PHY driver (that we
have in current community kernels) code supported only copper mode,
and this is not configurable, bits for copper mode are simply written
in registers during PHY initialization.

This patch adds support for both modes.

Signed-off-by: Alexandr Smirnov <asmirnov@ru.mvista.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

netxen: fix rx dropped stats

Don't count rx dropped packets based on return value of netif_receive_skb(),
which is misleading.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Tested-by: Vernon Mauery <mauery@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

netxen: remove low level tx lock

o eliminate tx lock in netxen adapter struct, instead pound on netdev
  tx lock appropriately.
o remove old "concurrent transmit" code that unnecessarily drops and
  reacquires tx lock in hard_xmit_frame(), this is already serialized
  the netdev xmit lock.
o reduce scope of tx lock in tx cleanup. tx cleanup operates on
  different section of the ring than transmitting cpus and is
  guarded by producer and consumer indices. This fixes a race
  caused by rx softirq preemption on realtime kernels.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Tested-by: Vernon Mauery <mauery@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

netxen: napi and irq cleanup

o separate and simpler irq handler for msi interrupts, avoids few checks
  than legacy mode.
o avoid redudant tx_has_work() and rx_has_work() checks in interrupt
  and napi, which can uncork irq based on racy (lockless) access to tx
  and rx ring indices. If we get interrupt, there's sufficient reason to
  schedule napi.
o replenish rx ring more often, remove self-imposed threshold rcv_free
  that prevents posting rx desc to card. This improves performance in
  low memory.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Tested-by: Vernon Mauery <mauery@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

netxen: improve msi support

Recent netxen firmware has new scheme of generating MSI interrupts, it
raises interrupt and blocks itself, waiting for driver to unmask. This
reduces chance of spurious interrupts.

The driver will be able to deal with older firmware as well.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Tested-by: Vernon Mauery <mauery@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

smc91x: fix build breakage from the SMC_GET_MAC_ADDR API upgrade

Cc: Jeff Garzik <jeff@garzik.org>
Cc: Deepak Saxena <dsaxena@plexity.net>
Cc: Nicolas Pitre <nico@cam.org>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

ixgb: remove unused variable

The variable num_group_tail_writes is initialized but never used otherwise.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T;
identifier i;
constant C;
@@

(
extern T i;
|
- T i;
  <+... when != i
- i = C;
  ...+>
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

e100: endianness annotations

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

igb trivial annotations

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

igb: endianness fix

le16_to_cpu() should be done before mask and shift...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

bonding: update version

Update version to 3.2.5.

Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

bonding: Fix sysfs attribute handling

For bonding interfaces any attempt to read the sysfs directory contents after
module removal results in an oops.  The fix is to release sysfs attributes
for the interfaces upon module unload.

Signed-off-by: Libor Pechacek <lpechacek@suse.cz>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

bonding: fix two compiler warnings

Fix two compiler warnings that are new with recent versions of gcc
(apparently 4.2 and up).  One is fixed by refactoring; this change was
supplied by Stephen Hemminger.  The other was fixed by labelling the
variable as uninitialized_var() after confirming via inspection that it
cannot actually be used uninitialized.

Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

bonding: Fix locking in 802.3ad mode

The 802.3ad state machine lock can be acquired in both softirq and
not softirq context, but was not held at _bh to prevent a deadlock (which
could occur if a LACPDU arrived and was processed while the lock was
held).

Corrected this, now hold the state machine lock at _bh to prevent
deadlock.

Bug reported by Todd Fleisher <todd@fleish.org>.

Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

smctr.c: fix logical-bitwise-or confusion

This patch to drivers/net/tokenring/smctr.c fixes a "bitwise vs
logical" or error.

Signed-off-by: Jay Schulist <jjschlst@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

b43: Remove irqs_disabled() sanity checks

Remove all irqs_disabled() sanity checks, as they are not safe on
a RT-enabled kernel and will trigger bogus warnings.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

b43: Fix DMA mapping leakage

This fixes a DMA mapping leakage in the case where we reject a DMA
buffer because of its address.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Cc: Christian Casteyde <casteyde.christian@free.fr>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: fix __devexit_p points to __devexit functions

The iwlxxxx_pci_remove functions are not needed when drivers are not
compiled as modules - they can thus be discarded at kernel link time.
This is already captured by having them as __devexit_p in the pci_driver
struct - these are supposed to be pointers to __devexit functions, but was not.
This is now fixed.

This problem was reported by Toralf Forster when testing the compilation of
2.6.25-rc6.

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
CC: Toralf Forster <toralf.foerster@gmx.de>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

MAINTAINERS: update iwlwifi git url

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

arlan: fix warning when PROC_FS=n

drivers/net/wireless/arlan-proc.c:1216: warning: 'arlan_root_table' defined but not used

Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: mac start synchronization issue

This patch fixes a synchronization problem on the 4965 and 3945 with the
mac start callback routine.  The problem is that this function exits BEFORE the
'xxx_alive_start' has completed.  This can lead to a problem if a
subsequent MAC callback attempts to issue a firmware command.

Signed-off-by: Rick Farrington <rickdic@hotmail.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

wavelan_cs arm fix

Even when all fields are unsigned char, struct still might have
alignment > 1.  Does so on arm, unless you explicitly say that
it's packed...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

rt2x00: Add id for Corega CG-WLUSB2GPX

This adds the id for Corega CG-WLUSB2GPX.

Signed-off-by: Masakazu Mokuno <mokuno@sm.sony.co.jp>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: fix a typo in Kconfig message

Signed-off-by: Pascal Terjan <pterjan@mandriva.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

[IPSEC]: Fix inter address family IPsec tunnel handling.

Signed-off-by: Kazunori MIYAZAWA <kazunori@miyazawa.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NEIGH]: Fix race between pneigh deletion and ipv6's ndisc_recv_ns (v3).

Proxy neighbors do not have any reference counting, so any caller
of pneigh_lookup (unless it's a netlink triggered add/del routine)
should _not_ perform any actions on the found proxy entry.

There's one exception from this rule - the ipv6's ndisc_recv_ns()
uses found entry to check the flags for NTF_ROUTER.

This creates a race between the ndisc and pneigh_delete - after
the pneigh is returned to the caller, the nd_tbl.lock is dropped
and the deleting procedure may proceed.

One of the fixes would be to add a reference counting, but this
problem exists for ndisc only. Besides such a patch would be too
big for -rc4.

So I propose to introduce a __pneigh_lookup() which is supposed
to be called with the lock held and use it in ndisc code to check
the flags on alive pneigh entry.

Changes from v2:
As David noticed, Exported the __pneigh_lookup() to ipv6 module.
The checkpatch generates a warning on it, since the EXPORT_SYMBOL
does not follow the symbol itself, but in this file all the
exports come at the end, so I decided no to break this harmony.

Changes from v1:
Fixed comments from YOSHIFUJI - indentation of prototype in header
and the pndisc_check_router() name - and a compilation fix, pointed
by Daniel - the is_routed was (falsely) considered as uninitialized
by gcc.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

sch_htb: fix "too many events" situation

HTB is event driven algorithm and part of its work is to apply
scheduled events at proper times. It tried to defend itself from
livelock by processing only limited number of events per dequeue.
Because of faster computers some users already hit this hardcoded
limit.

This patch limits processing up to 2 jiffies (why not 1 jiffie ?
because it might stop prematurely when only fraction of jiffie
remains).

Signed-off-by: Martin Devera <devik@cdi.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>

connector: convert to single-threaded workqueue

From: Evgeniy Polyakov <johnpol@2ka.mipt.ru>

We don't need one cqueue thread for each CPU.  cqueue is used for
receiving userspace datagrams, which are very rare and thus will
happily live with a single queue.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: When proc_create() fails, do some error handling work and return -ENOMEM.

Signed-off-by: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SUNGEM]: Fix NAPI assertion failure.

As reported by Johannes Berg:

I started getting this warning with recent kernels:

[  773.908927] ------------[ cut here ]------------
[  773.908954] Badness at net/core/dev.c:2204
 ...

If we loop more than once in gem_poll(), we'll
use more than the real budget in our gem_rx()
calls, thus eventually trigger the caller's
assertions in net_rx_action().

Subtract "work_done" from "budget" for the second
arg to gem_rx() to fix the bug.

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

BNX2X: prevent ethtool from setting port type

On 10GBaseT boards setting the type to TP will cause the driver to try
to configure 1GBaseT.
Since there are currently no boards that support setting of the port
type, disable this for now.

Signed-off-by: Eliezer Tamir <eliezert@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[9P] net/9p/trans_fd.c: remove unused variable

The variable cb is initialized but never used otherwise.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T;
identifier i;
constant C;
@@

(
extern T i;
|
- T i;
  <+... when != i
- i = C;
  ...+>
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6] net/ipv6/ndisc.c: remove unused variable

The variable hlen is initialized but never used otherwise.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T;
identifier i;
constant C;
@@

(
extern T i;
|
- T i;
  <+... when != i
- i = C;
  ...+>
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: fix warning from rcu_assign_poinger

This gets rid of a warning caused by the test in rcu_assign_pointer.
I tried to fix rcu_assign_pointer, but that devolved into a long set
of discussions about doing it right that came to no real solution.
Since the test in rcu_assign_pointer for constant NULL would never
succeed in fib_trie, just open code instead.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Let skbs grow over a page on fast peers

While testing the virtio-net driver on KVM with TSO I noticed
that TSO performance with a 1500 MTU is significantly worse
compared to the performance of non-TSO with a 16436 MTU.  The
packet dump shows that most of the packets sent are smaller
than a page.

Looking at the code this actually is quite obvious as it always
stop extending the packet if it's the first packet yet to be
sent and if it's larger than the MSS.  Since each extension is
bound by the page size, this means that (given a 1500 MTU) we're
very unlikely to construct packets greater than a page, provided
that the receiver and the path is fast enough so that packets can
always be sent immediately.

The fix is also quite obvious.  The push calls inside the loop
is just an optimisation so that we don't end up doing all the
sending at the end of the loop.  Therefore there is no specific
reason why it has to do so at MSS boundaries.  For TSO, the
most natural extension of this optimisation is to do the pushing
once the skb exceeds the TSO size goal.

This is what the patch does and testing with KVM shows that the
TSO performance with a 1500 MTU easily surpasses that of a 16436
MTU and indeed the packet sizes sent are generally larger than
16436.

I don't see any obvious downsides for slower peers or connections,
but it would be prudent to test this extensively to ensure that
those cases don't regress.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DLCI]: Fix tiny race between module unload and sock_ioctl.

This is a narrow pedantry :) but the dlci_ioctl_hook check and call
should not be parted with the mutex lock.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>