Statistics added for lro/lso bytes, count for tx stop queue and
wake queue and skb alloc failure count.
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kernel default tx csum function (ethtool_op_get_tx_csum) doesn't show
correct csum status. It takes various FLAGS (NETIF_F_ALL_CSUM) in account
to show tx csum status, which driver doesn't set while disabling tx csum.
Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch cleans up some unused code from be_load_fw().
Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When PCI functions are virtuialized in applications by assigning PCI
functions to VM (PCI passthrough), the be2net driver in the VM sees a
different function number. So, use of PCI function number in any
calculation will break existing code. This patch takes care of it.
Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 6b03a53a (tcp: use limited socket backlog) added the possibility
of dropping frames when backlog queue is full.
Commit d218d111 (tcp: Generalized TTL Security Mechanism) added the
possibility of dropping frames when TTL is under a given limit.
This patch adds new SNMP MIB entries, named TCPBacklogDrop and
TCPMinTTLDrop, published in /proc/net/netstat in TcpExt: line
netstat -s | egrep "TCPBacklogDrop|TCPMinTTLDrop"
TCPBacklogDrop: 0
TCPMinTTLDrop: 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the "__must_check" tag to sk_add_backlog() so that any failure to
check and drop packets will be warned about.
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Thanks to Paul McKenny for pointing out that it is incorrect to use
synchronize_rcu_bh to ensure that pending callbacks have completed.
Instead we should use rcu_barrier_bh.
Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
As Paul McKenney correctly pointed out, __br_mdb_ip_get needs
to use the RCU list walking primitive in order to work correctly
on platforms where data-dependency ordering is not guaranteed.
Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
IPV6_PREFER_SRC_xxx definitions:
| #define IPV6_PREFER_SRC_TMP 0x0001
| #define IPV6_PREFER_SRC_PUBLIC 0x0002
| #define IPV6_PREFER_SRC_COA 0x0004
RT6_LOOKUP_F_xxx definitions:
| #define RT6_LOOKUP_F_SRCPREF_TMP 0x00000008
| #define RT6_LOOKUP_F_SRCPREF_PUBLIC 0x00000010
| #define RT6_LOOKUP_F_SRCPREF_COA 0x00000020
So, we can translate between these two groups by shift operation
instead of multiple 'if's.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If we were unable to detect a PHY on any of the MDIO bus id we tried instead of
bailing out with -ENODEV, assume the MAC is connected to a switch and use MDIO
bus 0. This unbreaks quite a lot of devices out there whose switch cannot be
detected.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Despite what the comment above CPMAC_SKB_SIZE says, the hardware also
needs to account for the FCS length in a received frame. This patch fix
the receiving of 802.1q frames which have 4 more bytes. While at it
unhardcode the definition and use the one from if_vlan.h.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update the CAN Maintainer responsibilities and add source paths.
Additional the SocketCAN core ML is not subscribers-only anymore.
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Acked-by: Wolfgang Grandegger <wg@grandegger.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This one removes trailing whitespace in pegasus.h and more importantly
adds new Pegasus compatible device.
Signed-off-by: Julian Brown <julian@codesourcery.com>
Signed-off-by: Petko Manolov <petkan@nucleusys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the call to kcalloc() fails then we should return -ENOMEM.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We test that "prot->rsk_prot" is non-null right before we dereference it
on this line.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
"skb" is alway non-null here, but even if it were null the check isn't
needed because dev_kfree_skb() can handle it.
This eliminates a smatch warning about dereferencing a variable before
checking that it is non-null.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We dereference "port" on the lines immediately before and immediately
after the test so port should hopefully never be null here.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix a race at the end of NAPI processing in ks8695_poll() function.
Signed-off-by:Figo.zhang <figo1802@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently s2io is dumping debug messages using the interface name
before it was allocated, showing a message like the following:
s2io: eth%d: Ring Mem PHY: 0x7ef80000
s2io: s2io_reset: Resetting XFrame card eth%d
This patch just fixes it, printing the pci bus information for
the card instead of the interface name.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
when receiving a particular type of NFS v2 UDP traffic, the hardware could
DMA some bad data and then hang, possibly corrupting memory.
Disable the NFS parsing in this hardware, verified to fix the bug.
Originally reported and reproduced by RedHat's Neil Horman
CC: nhorman@tuxdriver.com
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The typhoon driver was incorrectly using smp_wmb() to order memory
accesses against IO to the NIC in a few instances. Use wmb() instead,
which is required to actually order between memory types.
Signed-off-by: David Dillow <dave@thedillows.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
On 03/04/2010 09:26 AM, Ben Hutchings wrote:
> On Thu, 2010-03-04 at 00:51 -0800, Jeff Kirsher wrote:
>> From: Jeff Garzik<jgarzik@redhat.com>
>>
>> This patch is an alternative approach for accessing string
>> counts, vs. the drvinfo indirect approach. This way the drvinfo
>> space doesn't run out, and we don't break ABI later.
> [...]
>> --- a/net/core/ethtool.c
>> +++ b/net/core/ethtool.c
>> @@ -214,6 +214,10 @@ static noinline int ethtool_get_drvinfo(struct net_device *dev, void __user *use
>> info.cmd = ETHTOOL_GDRVINFO;
>> ops->get_drvinfo(dev,&info);
>>
>> + /*
>> + * this method of obtaining string set info is deprecated;
>> + * consider using ETHTOOL_GSSET_INFO instead
>> + */
>
> This comment belongs on the interface (ethtool.h) not the
> implementation.
Debatable -- the current comment is located at the callsite of
ops->get_sset_count(), which is where an implementor might think to add
a new call. Not all the numeric fields in ethtool_drvinfo are obtained
from ->get_sset_count().
Hence the "some" in the attached patch to include/linux/ethtool.h,
addressing your comment.
> [...]
>> +static noinline int ethtool_get_sset_info(struct net_device *dev,
>> + void __user *useraddr)
>> +{
> [...]
>> + /* calculate size of return buffer */
>> + for (i = 0; i< 64; i++)
>> + if (sset_mask& (1ULL<< i))
>> + n_bits++;
> [...]
>
> We have a function for this:
>
> n_bits = hweight64(sset_mask);
Agreed.
I've attached a follow-up patch, which should enable my/Jeff's kernel
patch to be applied, followed by this one.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is an alternative approach for accessing string
counts, vs. the drvinfo indirect approach. This way the drvinfo
space doesn't run out, and we don't break ABI later.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: David Brown <davidb@quicinc.com>
Signed-off-by: Daniel Walker <dwalker@codeaurora.org>
Acked-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_add_backlog -> __sk_add_backlog
sk_add_backlog_limited -> sk_add_backlog
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make x25 adapt to the limited socket backlog change.
Cc: Andrew Hendry <andrew.hendry@gmail.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make tipc adapt to the limited socket backlog change.
Cc: Jon Maloy <jon.maloy@ericsson.com>
Cc: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make sctp adapt to the limited socket backlog change.
Cc: Vlad Yasevich <vladislav.yasevich@hp.com>
Cc: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make llc adapt to the limited socket backlog change.
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make udp adapt to the limited socket backlog change.
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: "Pekka Savola (ipv6)" <pekkas@netcore.fi>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make tcp adapt to the limited socket backlog change.
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: "Pekka Savola (ipv6)" <pekkas@netcore.fi>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We got system OOM while running some UDP netperf testing on the loopback
device. The case is multiple senders sent stream UDP packets to a single
receiver via loopback on local host. Of course, the receiver is not able
to handle all the packets in time. But we surprisingly found that these
packets were not discarded due to the receiver's sk->sk_rcvbuf limit.
Instead, they are kept queuing to sk->sk_backlog and finally ate up all
the memory. We believe this is a secure hole that a none privileged user
can crash the system.
The root cause for this problem is, when the receiver is doing
__release_sock() (i.e. after userspace recv, kernel udp_recvmsg ->
skb_free_datagram_locked -> release_sock), it moves skbs from backlog to
sk_receive_queue with the softirq enabled. In the above case, multiple
busy senders will almost make it an endless loop. The skbs in the
backlog end up eat all the system memory.
The issue is not only for UDP. Any protocols using socket backlog is
potentially affected. The patch adds limit for socket backlog so that
the backlog size cannot be expanded endlessly.
Reported-by: Alex Shi <alex.shi@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru
Cc: "Pekka Savola (ipv6)" <pekkas@netcore.fi>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Vlad Yasevich <vladislav.yasevich@hp.com>
Cc: Sridhar Samudrala <sri@us.ibm.com>
Cc: Jon Maloy <jon.maloy@ericsson.com>
Cc: Allan Stephens <allan.stephens@windriver.com>
Cc: Andrew Hendry <andrew.hendry@gmail.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
My previous patch (655ffee284) added locking in
a bad way. Because rndis_set_oid can sleep, there is need to prepare multicast
addresses into local buffer under netif_addr_lock first, then call
rndis_set_oid outside. This caused reorganizing of the whole function.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Reported-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
queue restart tasklets need to be stopped after napi handlers are stopped
since the latter can restart them. So stop them after stopping napi.
Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Starting with commit a3bc1f11e9 ("gianfar: Revive SKB
recycling") gianfar driver sooner or later stops transmitting any
packets on SMP machines.
start_xmit() prepares new skb for transmitting, generally it does
three things:
1. sets up all BDs (marks them ready to send), except the first one.
2. stores skb into tx_queue->tx_skbuff so that clean_tx_ring()
would cleanup it later.
3. sets up the first BD, i.e. marks it ready.
Here is what clean_tx_ring() does:
1. reads skbs from tx_queue->tx_skbuff
2. checks if the *last* BD is ready. If it's still ready [to send]
then it it isn't transmitted, so clean_tx_ring() returns.
Otherwise it actually cleanups BDs. All is OK.
Now, if there is just one BD, code flow:
- start_xmit(): stores skb into tx_skbuff. Note that the first BD
(which is also the last one) isn't marked as ready, yet.
- clean_tx_ring(): sees that skb is not null, *and* its lstatus
says that it is NOT ready (like if BD was sent), so it cleans
it up (bad!)
- start_xmit(): marks BD as ready [to send], but it's too late.
We can fix this simply by reordering lstatus/tx_skbuff writes.
Reported-by: Martyn Welch <martyn.welch@ge.com>
Bisected-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Tested-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Tested-by: Martyn Welch <martyn.welch@ge.com>
Cc: Sandeep Gopalpet <Sandeep.Kumar@freescale.com>
Cc: Stable <stable@vger.kernel.org> [2.6.33]
Signed-off-by: David S. Miller <davem@davemloft.net>
r8169 needs certain writes to be visible to other CPUs or the NIC before
touching the hardware, but was using smp_wmb() which is only required to
order cacheable memory access. Switch to wmb() which is required to
order both cacheable and non-cacheable memory.
Noticed by Catalin Marinas and Paul Mackerras.
Signed-off-by: David Dillow <dave@thedillows.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix TIPC to disallow sending to remote addresses prior to entering NET_MODE
user programs can oops the kernel by sending datagrams via AF_TIPC prior to
entering networked mode. The following backtrace has been observed:
ID: 13459 TASK: ffff810014640040 CPU: 0 COMMAND: "tipc-client"
[exception RIP: tipc_node_select_next_hop+90]
RIP: ffffffff8869d3c3 RSP: ffff81002d9a5ab8 RFLAGS: 00010202
RAX: 0000000000000001 RBX: 0000000000000001 RCX: 0000000000000001
RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000001001001
RBP: 0000000001001001 R8: 0074736575716552 R9: 0000000000000000
R10: ffff81003fbd0680 R11: 00000000000000c8 R12: 0000000000000008
R13: 0000000000000001 R14: 0000000000000001 R15: ffff810015c6ca00
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
RIP: 0000003cbd8d49a3 RSP: 00007fffc84e0be8 RFLAGS: 00010206
RAX: 000000000000002c RBX: ffffffff8005d116 RCX: 0000000000000000
RDX: 0000000000000008 RSI: 00007fffc84e0c00 RDI: 0000000000000003
RBP: 0000000000000000 R8: 00007fffc84e0c10 R9: 0000000000000010
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
R13: 00007fffc84e0d10 R14: 0000000000000000 R15: 00007fffc84e0c30
ORIG_RAX: 000000000000002c CS: 0033 SS: 002b
What happens is that, when the tipc module in inserted it enters a standalone
node mode in which communication to its own address is allowed <0.0.0> but not
to other addresses, since the appropriate data structures have not been
allocated yet (specifically the tipc_net pointer). There is nothing stopping a
client from trying to send such a message however, and if that happens, we
attempt to dereference tipc_net.zones while the pointer is still NULL, and
explode. The fix is pretty straightforward. Since these oopses all arise from
the dereference of global pointers prior to their assignment to allocated
values, and since these allocations are small (about 2k total), lets convert
these pointers to static arrays of the appropriate size. All the accesses to
these bits consider 0/NULL to be a non match when searching, so all the lookups
still work properly, and there is no longer a chance of a bad dererence
anywhere. As a bonus, this lets us eliminate the setup/teardown routines for
those pointers, and elimnates the need to preform any locking around them to
prevent access while their being allocated/freed.
I've updated the tipc_net structure to behave this way to fix the exact reported
problem, and also fixed up the tipc_bearers and media_list arrays to fix an
obvious simmilar problem that arises from issuing tipc-config commands to
manipulate bearers/links prior to entering networked mode
I've tested this for a few hours by running the sanity tests and stress test
with the tipcutils suite, and nothing has fallen over. There have been a few
lockdep warnings, but those were there before, and can be addressed later, as
they didn't actually result in any deadlock.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Allan Stephens <allan.stephens@windriver.com>
CC: David S. Miller <davem@davemloft.net>
CC: tipc-discussion@lists.sourceforge.net
bearer.c | 37 ++++++-------------------------------
bearer.h | 2 +-
net.c | 25 ++++---------------------
3 files changed, 11 insertions(+), 53 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
ipgre_header() can be called with zero daddr when the gre device is
configured as multipoint tunnel and still has the NOARP flag set (which is
typically cleared by the userspace arp daemon). If the NOARP packets are
not dropped, ipgre_tunnel_xmit() will take rt->rt_gateway (= NBMA IP) and
use that for route look up (and may lead to bogus xfrm acquires).
The multicast address check is removed as sending to multicast group should
be ok. In fact, if gre device has a multicast address as destination
ipgre_header is always called with multicast address.
Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Decreases the odds wakee will suffer from frequent cache misses.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This solves a potential race problem during the cleanup process.
The issue is that addrconf_ifdown() needs to traverse address list,
but then drop lock to call the notifier. The version in -next
could get confused if add/delete happened during this window.
Original code (2.6.32 and earlier) was okay because all addresses
were always deleted.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
My recent change in net-next to retain permanent addresses caused regression.
Device refcount would not go to zero when device was unregistered because
left over anycast reference would hold ipv6 dev reference which would hold
device references...
The correct procedure is to call notify chain when address is no longer
available for use. When interface comes back DAD timer will notify
back that address is available.
Also, link local addresses should be purged when interface is brought
down. The address might be changed.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Router Solicitation timer races with device state changes
because it doesn't lock the device. Use local variable to avoid
one repeated dereference.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Timer code runs in bottom half, so there is no need for
using _bh form of locking. Also check if device is not ready
to avoid race with address that is no longer active.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Export rt2x00soc_probe from rt2x00soc as it is used in rt2800pci.
Otherwise loading rt2800pci gives "rt2800pci: Unknown symbol
rt2x00soc_probe".
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Handling HT configuration changes involved setting the channel
with the new HT parameters and then issuing a rate_update()
notification to the driver.
This behavior changed after the off-channel changes. Now, the channel
is not updated with the new HT params in enable_ht() - instead, it
is now done when the scan work terminates. This results in the driver
depending on stale information, defaulting to non-HT mode always.
Fix this by passing the new channel type to the driver.
Cc: stable@kernel.org
Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
br_multicast calls ip_send_check(), so it should depend on INET.
built-in:
br_multicast.c:(.text+0x88cf4): undefined reference to `ip_send_check'
or modular:
ERROR: "ip_send_check" [net/bridge/bridge.ko] undefined!
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>