If there are no explicit metrics attached to a route, hook
fi->fib_info up to dst_default_metrics.
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit c6d14c8456 (net: Introduce for_each_netdev_rcu() iterator)
added a race in dev_seq_next().
The rcu_dereference() call should be done _before_ testing the end of
list, or we might return a wrong net_device if a concurrent thread
changes net_device list under us.
Note : discovered thanks to a sparse warning :
net/core/dev.c:3919:9: error: incompatible types in comparison expression
(different address spaces)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
pskb_expand_head() triggers a kmemcheck warning when copy of
skb_shared_info is done in pskb_expand_head()
This is because destructor_arg field is not necessarily initialized at
this point. Add kmemcheck_annotate_variable() call in __alloc_skb() to
instruct kmemcheck this is a normal situation.
Resolves bugzilla.kernel.org 27212
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=27212
Reported-by: Christian Casteyde <casteyde.christian@free.fr>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
I'm testing an API that uses IFLA_AF_SPEC attribute.
In the rtnetlink core , the set_link_af() member
of the rtnl_af_ops struct receives the nested attribute
(as I expected), but the validate_link_af() member
receives the parent attribute.
IMO, this patch fixes this.
Signed-off-by: Kurt Van Dijck <kurt.van.dijck@eia.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Routing metrics are now copy-on-write.
Initially a route entry points it's metrics at a read-only location.
If a routing table entry exists, it will point there. Else it will
point at the all zero metric place-holder called 'dst_default_metrics'.
The writeability state of the metrics is stored in the low bits of the
metrics pointer, we have two bits left to spare if we want to store
more states.
For the initial implementation, COW is implemented simply via kmalloc.
However future enhancements will change this to place the writable
metrics somewhere else, in order to increase sharing. Very likely
this "somewhere else" will be the inetpeer cache.
Note also that this means that metrics updates may transiently fail
if we cannot COW the metrics successfully.
But even by itself, this patch should decrease memory usage and
increase cache locality especially for routing workloads. In those
cases the read-only metric copies stay in place and never get written
to.
TCP workloads where metrics get updated, and those rare cases where
PMTU triggers occur, will take a very slight performance hit. But
that hit will be alleviated when the long-term writable metrics
move to a more sharable location.
Since the metrics storage went from a u32 array of RTAX_MAX entries to
what is essentially a pointer, some retooling of the dst_entry layout
was necessary.
Most importantly, we need to preserve the alignment of the reference
count so that it doesn't share cache lines with the read-mostly state,
as per Eric Dumazet's alignment assertion checks.
The only non-trivial bit here is the move of the 'flags' member into
the writeable cacheline. This is OK since we are always accessing the
flags around the same moment when we made a modification to the
reference count.
Signed-off-by: David S. Miller <davem@davemloft.net>
We spend lot of time clearing pages in pktgen.
(Or not clearing them on ipv6 and leaking kernel memory)
Since we dont modify them, we can use one zeroed page, and get
references on it. This page can use NUMA affinity as well.
Define pktgen_finalize_skb() helper, used both in ipv4 and ipv6
Results using skbs with one frag :
Before patch :
Result: OK: 608980458(c608978520+d1938) nsec, 1000000000
(100byte,1frags)
1642088pps 1313Mb/sec (1313670400bps) errors: 0
After patch :
Result: OK: 345285014(c345283891+d1123) nsec, 1000000000
(100byte,1frags)
2896158pps 2316Mb/sec (2316926400bps) errors: 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The group of a network device can be queried or changed from userspace
using sysfs.
For example, considering sysfs mounted in /sys, one can change the group
that interface lo belongs to:
echo 1 > /sys/class/net/lo/group
Signed-off-by: Vlad Dogaru <ddvlad@rosedu.org>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a conflict between commit b00916b1 and a77f5db3. This patch resolves
the conflict by clearing the heap allocation in ethtool_get_regs().
Cc: stable@kernel.org
Signed-off-by: Eugene Teo <eugeneteo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reduce printk() levels to KERN_INFO in netdev_fix_features() as this will
be used by ethtool and might spam dmesg unnecessarily.
This converts the function to use netdev_info() instead of plain printk().
As a side effect, bonding and bridge devices will now log dropped features
on every slave device change.
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Quoting Ben Hutchings: we presumably won't be defining features that
can only be enabled on 64-bit architectures.
Occurences found by `grep -r` on net/, drivers/net, include/
[ Move features and vlan_features next to each other in
struct netdev, as per Eric Dumazet's suggestion -DaveM ]
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow drivers for multiqueue hardware with flow filter tables to
accelerate RFS. The driver must:
1. Set net_device::rx_cpu_rmap to a cpu_rmap of the RX completion
IRQs (in queue order). This will provide a mapping from CPUs to the
queues for which completions are handled nearest to them.
2. Implement net_device_ops::ndo_rx_flow_steer. This operation adds
or replaces a filter steering the given flow to the given RX queue, if
possible.
3. Periodically remove filters for which rps_may_expire_flow() returns
true.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Suppose that several linear skbs of the same flow were received by GRO. They
were thus merged into one skb with a frag_list. Then a new skb of the same flow
arrives, but it is a paged skb with data starting in its frags[].
Before adding the skb to the frag_list skb_gro_receive() will of course adjust
the skb to throw away the headers. It correctly modifies the page_offset and
size of the frag, but it leaves incorrect information in the skb:
->data_len is not decreased at all.
->len is decreased only by headlen, as if no change were done to the frag.
Later in a receiving process this causes skb_copy_datagram_iovec() to return
-EFAULT and this is seen in userspace as the result of the recv() syscall.
In practice the bug can be reproduced with the sfc driver. By default the
driver uses an adaptive scheme when it switches between using
napi_gro_receive() (with skbs) and napi_gro_frags() (with pages). The bug is
reproduced when under rx load with enough successful GRO merging the driver
decides to switch from the former to the latter.
Manual control is also possible, so reproducing this is easy with netcat:
- on machine1 (with sfc): nc -l 12345 > /dev/null
- on machine2: nc machine1 12345 < /dev/zero
- on machine1:
echo 1 > /sys/module/sfc/parameters/rx_alloc_method # use skbs
echo 2 > /sys/module/sfc/parameters/rx_alloc_method # use pages
- See that nc has quit suddenly.
[v2: Modified by Eric Dumazet to avoid advancing skb->data past the end
and to use a temporary variable.]
Signed-off-by: Michal Schmidt <mschmidt@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 941666c2e3 "net: RCU conversion of dev_getbyhwaddr() and
arp_ioctl()" introduced a regression, reported by Jamie Heilman.
"arp -Ds 192.168.2.41 eth0 pub" triggered the ASSERT_RTNL() assert
in pneigh_lookup()
Removing RTNL requirement from arp_ioctl() was a mistake, just revert
that part.
Reported-by: Jamie Heilman <jamie@audible.transient.net>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rtnl_group_changelink() is invoked by rtnl_newlink() before the link
attributes have been validated. Additionally the group changes are
performed even if NLM_F_CREATE is specified and a new link is
created, while more reasonable semantics would be to set the group
value on the newly created link.
Fix both problems by moving the rtnl_group_changelink() invocation
down to the handling of non-existant links without NLM_F_CREATE()
and add a dev_set_group() call to rtnl_create_link().
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Vlad Dogaru <ddvlad@rosedu.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix some minor issues and sparse (__rcu) warnings
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch converts stab qdisc management to RCU, so that we can perform
the qdisc_calculate_pkt_len() call before getting qdisc lock.
This shortens the lock's held time in __dev_xmit_skb().
This permits more qdiscs to get TCQ_F_CAN_BYPASS status, avoiding lot of
cache misses and so reducing latencies.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Octavian Purdila <opurdila@ixiacom.com>
Reviewed-by: Octavian Purdila <opurdila@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch provides a mechanism for lower layer devices to
steer traffic using skb->priority to tx queues. This allows
for hardware based QOS schemes to use the default qdisc without
incurring the penalties related to global state and the qdisc
lock. While reliably receiving skbs on the correct tx ring
to avoid head of line blocking resulting from shuffling in
the LLD. Finally, all the goodness from txq caching and xps/rps
can still be leveraged.
Many drivers and hardware exist with the ability to implement
QOS schemes in the hardware but currently these drivers tend
to rely on firmware to reroute specific traffic, a driver
specific select_queue or the queue_mapping action in the
qdisc.
By using select_queue for this drivers need to be updated for
each and every traffic type and we lose the goodness of much
of the upstream work. Firmware solutions are inherently
inflexible. And finally if admins are expected to build a
qdisc and filter rules to steer traffic this requires knowledge
of how the hardware is currently configured. The number of tx
queues and the queue offsets may change depending on resources.
Also this approach incurs all the overhead of a qdisc with filters.
With the mechanism in this patch users can set skb priority using
expected methods ie setsockopt() or the stack can set the priority
directly. Then the skb will be steered to the correct tx queues
aligned with hardware QOS traffic classes. In the normal case with
single traffic class and all queues in this class everything
works as is until the LLD enables multiple tcs.
To steer the skb we mask out the lower 4 bits of the priority
and allow the hardware to configure upto 15 distinct classes
of traffic. This is expected to be sufficient for most applications
at any rate it is more then the 8021Q spec designates and is
equal to the number of prio bands currently implemented in
the default qdisc.
This in conjunction with a userspace application such as
lldpad can be used to implement 8021Q transmission selection
algorithms one of these algorithms being the extended transmission
selection algorithm currently being used for DCB.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a rtnetlink request specifies a negative or zero ifindex and has no
interface name attribute, but has a group attribute, then the chenges
are made to all the interfaces belonging to the specified group.
Signed-off-by: Vlad Dogaru <ddvlad@rosedu.org>
Acked-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
Net devices can now be grouped, enabling simpler manipulation from
userspace. This patch adds a group field to the net_device structure, as
well as rtnetlink support to query and modify it.
Signed-off-by: Vlad Dogaru <ddvlad@rosedu.org>
Acked-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (41 commits)
sctp: user perfect name for Delayed SACK Timer option
net: fix can_checksum_protocol() arguments swap
Revert "netlink: test for all flags of the NLM_F_DUMP composite"
gianfar: Fix misleading indentation in startup_gfar()
net/irda/sh_irda: return to RX mode when TX error
net offloading: Do not mask out NETIF_F_HW_VLAN_TX for vlan.
USB CDC NCM: tx_fixup() race condition fix
ns83820: Avoid bad pointer deref in ns83820_init_one().
ipv6: Silence privacy extensions initialization
bnx2x: Update bnx2x version to 1.62.00-4
bnx2x: Fix AER setting for BCM57712
bnx2x: Fix BCM84823 LED behavior
bnx2x: Mark full duplex on some external PHYs
bnx2x: Fix BCM8073/BCM8727 microcode loading
bnx2x: LED fix for BCM8727 over BCM57712
bnx2x: Common init will be executed only once after POR
bnx2x: Swap BCM8073 PHY polarity if required
iwlwifi: fix valid chain reading from EEPROM
ath5k: fix locking in tx_complete_poll_work
ath9k_hw: do PA offset calibration only on longcal interval
...
commit 0363466866 (net offloading: Convert checksums to use
centrally computed features.) mistakenly swapped can_checksum_protocol()
arguments.
This broke IPv6 on bnx2 for instance, on NIC without TCPv6 checksum
offloads.
Reported-by: Hans de Bruin <jmdebruin@xmsnet.nl>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Packet filter (BPF) doesnt need to disable softirqs, being fully
re-entrant and lock-less.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In netif_skb_features() we return only the features that are valid for vlans
if we have a vlan packet. However, we should not mask out NETIF_F_HW_VLAN_TX
since it enables transmission of vlan tags and is obviously valid.
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (47 commits)
GRETH: resolve SMP issues and other problems
GRETH: handle frame error interrupts
GRETH: avoid writing bad speed/duplex when setting transfer mode
GRETH: fixed skb buffer memory leak on frame errors
GRETH: GBit transmit descriptor handling optimization
GRETH: fix opening/closing
GRETH: added raw AMBA vendor/device number to match against.
cassini: Fix build bustage on x86.
e1000e: consistent use of Rx/Tx vs. RX/TX/rx/tx in comments/logs
e1000e: update Copyright for 2011
e1000: Avoid unhandled IRQ
r8169: keep firmware in memory.
netdev: tilepro: Use is_unicast_ether_addr helper
etherdevice.h: Add is_unicast_ether_addr function
ks8695net: Use default implementation of ethtool_ops::get_link
ks8695net: Disable non-working ethtool operations
USB CDC NCM: Don't deref NULL in cdc_ncm_rx_fixup() and don't use uninitialized variable.
vxge: Remember to release firmware after upgrading firmware
netdev: bfin_mac: Remove is_multicast_ether_addr use in netdev_for_each_mc_addr
ipsec: update MAX_AH_AUTH_LEN to support sha512
...
After recent changes, (percpu stats on vlan/tunnels...), we dont need
anymore per struct netdev_queue tx_bytes/tx_packets/tx_dropped counters.
Only remaining users are ixgbe, sch_teql, gianfar & macvlan :
1) ixgbe can be converted to use existing tx_ring counters.
2) macvlan incremented txq->tx_dropped, it can use the
dev->stats.tx_dropped counter.
3) sch_teql : almost revert ab35cd4b8f (Use net_device internal stats)
Now we have ndo_get_stats64(), use it, even for "unsigned long"
fields (No need to bring back a struct net_device_stats)
4) gianfar adds a stats structure per tx queue to hold
tx_bytes/tx_packets
This removes a lockdep warning (and possible lockup) in rndis gadget,
calling dev_get_stats() from hard IRQ context.
Ref: http://www.spinics.net/lists/netdev/msg149202.html
Reported-by: Neil Jones <neiljay@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Alexander Duyck <alexander.h.duyck@intel.com>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Sandeep Gopalpet <sandeep.kumar@freescale.com>
CC: Michal Nazarewicz <mina86@mina86.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The IPv6 tproxy patches split IPv6 defragmentation off of conntrack, but
failed to update the #ifdef stanzas guarding the defragmentation related
fields and code in skbuff and conntrack related code in nf_defrag_ipv6.c.
This patch adds the required #ifdefs so that IPv6 tproxy can truly be used
without connection tracking.
Original report:
http://marc.info/?l=linux-netdev&m=129010118516341&w=2
Reported-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: KOVACS Krisztian <hidden@balabit.hu>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
HTB takes into account skb is segmented in stats updates.
Generalize this to all schedulers.
They should use qdisc_bstats_update() helper instead of manipulating
bstats.bytes and bstats.packets
Add bstats_update() helper too for classes that use
gnet_stats_basic_packed fields.
Note : Right now, TCQ_F_CAN_BYPASS shortcurt can be taken only if no
stab is setup on qdisc.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Added alloc_netdev_mqs function which allows the number of transmit and
receive queues to be specified independenty. alloc_netdev_mq was
changed to a macro to call the new function. Also added
alloc_etherdev_mqs with same purpose.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to compute the features for other offloads (primarily
scatter/gather), we need to first check the ability of the NIC to
offload the checksum for the packet. Since we have already computed
this, we can directly use the result instead of figuring it out
again.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This switches skb_need_linearize() to use the features that have
been centrally computed. In doing so, this fixes a problem where
scatter/gather should not be used because the card does not support
checksum offloading on that type of packet. On device registration
we only check that some form of checksum offloading is available if
scatter/gatther is enabled but we must also check at transmission
time. Examples of this include IPv6 or vlan packets on a NIC that
only supports IPv4 offloading.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This switches dev_gso_segment() to use the device features computed
by the centralized routine. In doing so, it fixes a problem where
it would always use dev->features, instead of those appropriate
to the number of vlan tags if any are present.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that there is a single function that can compute the device
features relevant to a packet, we don't want to run it for each
offload. This converts netif_needs_gso() to take the features
of the device, rather than computing them itself.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
netif_get_vlan_features() is currently only used by netif_needs_gso(),
so it only concerns itself with GSO features. However, several other
places also should take into account the contents of the packet when
deciding whether to offload to hardware. This generalizes the function
to return features about all of the various forms of offloading. Since
offloads tend to be linked together, this avoids duplicating the logic
in each location (i.e. the scatter/gather code also needs the checksum
logic).
Suggested-by: Michał Mirosław <mirqus@gmail.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We currently only have software fallback for one type of checksum: the
TCP/UDP one's complement. This means that a protocol that uses hardware
offloading for a different type of checksum (FCoE, SCTP) must directly
check the device's features and do the right thing ahead of time. By
the time we get to dev_can_checksum(), we're only deciding whether to
apply the one algorithm in software or hardware. NETIF_F_HW_CSUM has the
same capabilities as the software version, so we should always use it if
present. The primary advantage of this is multiply tagged vlans can use
hardware checksumming.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix new kernel-doc notation warning in net/core/filter.c:
Warning(net/core/filter.c:172): No description found for parameter 'fentry'
Warning(net/core/filter.c:172): Excess function parameter 'filter' description in 'sk_run_filter'
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Due to NLM_F_DUMP is composed of two bits, NLM_F_ROOT | NLM_F_MATCH,
when doing "if (x & NLM_F_DUMP)", it tests for _either_ of the bits
being set. Because NLM_F_MATCH's value overlaps with NLM_F_EXCL,
non-dump requests with NLM_F_EXCL set are mistaken as dump requests.
Substitute the condition to test for _all_ bits being set.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
* 'for-2.6.38' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (33 commits)
usb: don't use flush_scheduled_work()
speedtch: don't abuse struct delayed_work
media/video: don't use flush_scheduled_work()
media/video: explicitly flush request_module work
ioc4: use static work_struct for ioc4_load_modules()
init: don't call flush_scheduled_work() from do_initcalls()
s390: don't use flush_scheduled_work()
rtc: don't use flush_scheduled_work()
mmc: update workqueue usages
mfd: update workqueue usages
dvb: don't use flush_scheduled_work()
leds-wm8350: don't use flush_scheduled_work()
mISDN: don't use flush_scheduled_work()
macintosh/ams: don't use flush_scheduled_work()
vmwgfx: don't use flush_scheduled_work()
tpm: don't use flush_scheduled_work()
sonypi: don't use flush_scheduled_work()
hvsi: don't use flush_scheduled_work()
xen: don't use flush_scheduled_work()
gdrom: don't use flush_scheduled_work()
...
Fixed up trivial conflict in drivers/media/video/bt8xx/bttv-input.c
as per Tejun.
Leonardo Chiquitto found poll() could block forever on tcp sockets and
Urgent data was received, if the event flag only contains POLLPRI.
He did a bisection and found commit 4938d7e023 (poll: avoid extra
wakeups in select/poll) was the source of the problem.
Problem is TCP sockets use standard sock_def_readable() function for
their sk_data_ready() handler, and sock_def_readable() doesnt signal
POLLPRI.
Only TCP is affected by the problem. Adding POLLPRI to the list of flags
might trigger unnecessary schedules, but URGENT handling is such a
seldom used feature this seems a good compromise.
Thanks a lot to Leonardo for providing the bisection result and a test
program as well.
Reference : http://www.spinics.net/lists/netdev/msg151793.html
Reported-and-bisected-by: Leonardo Chiquitto <leonardo.lists@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
