deliverable/linux.git
10 years agortnetlink: remove IFLA_BOND_SLAVE definition
Jiri Pirko [Fri, 24 Jan 2014 07:39:16 +0000 (08:39 +0100)] 
rtnetlink: remove IFLA_BOND_SLAVE definition

This is in net-next only, for couple of days. Not used anymore, and never
should have been. So just remove it and pretend it was never there.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agortnetlink: remove check for fill_slave_info in rtnl_have_link_slave_info
Jiri Pirko [Thu, 23 Jan 2014 18:19:21 +0000 (19:19 +0100)] 
rtnetlink: remove check for fill_slave_info in rtnl_have_link_slave_info

This check is not needed because the same check is done before
fill_slave_info is used in rtnl_link_slave_info_fill.
Also, by removing this check, kernel will fillup IFLA_INFO_SLAVE_KIND
even for slaves of masters which does not implement fill_slave_info.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'qlcnic'
David S. Miller [Fri, 24 Jan 2014 00:13:39 +0000 (16:13 -0800)] 
Merge branch 'qlcnic'

Himanshu Madhani says:

====================
qlcnic: Refactoring and enhancements for all adapters.

This patch series contains following patches.

o Enhanced debug data collection when we are in Tx-timeout situation.
o Enhanced MSI-x vector calculation for defualt load path as well
  as for TSS/RSS ring change path.
o Refactored interrupt coalescing code for all adapters.
o Refactored interrupt handling as well as cleanup of poll controller
  code patch for all adapters.
o changed rx_mac_learn type to boolean.

Please apply to net-next.
====================

Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: update version to 5.3.55
Himanshu Madhani [Thu, 23 Jan 2014 22:18:34 +0000 (17:18 -0500)] 
qlcnic: update version to 5.3.55

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Enhance logic to calculate msix vectors.
Himanshu Madhani [Thu, 23 Jan 2014 22:18:33 +0000 (17:18 -0500)] 
qlcnic: Enhance logic to calculate msix vectors.

o Refactored MSI-x vector calculation for All adapters.
  Decoupled logic in the code which was using same call to
  request MSI-x vectors in default driver load, as well as
  during set_channel() operation for TSS/RSS. This refactoring
  simplifies code for TSS/RSS code path as well as probe path
  of the driver load for all adapters.

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Refactor interrupt coalescing code for all adapters.
Himanshu Madhani [Thu, 23 Jan 2014 22:18:32 +0000 (17:18 -0500)] 
qlcnic: Refactor interrupt coalescing code for all adapters.

o Refactor configuration of interrupt coalescing parameters for
  all supported adapters.

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Update poll controller code path
Manish chopra [Thu, 23 Jan 2014 22:18:31 +0000 (17:18 -0500)] 
qlcnic: Update poll controller code path

Add support for MSI/MSI-X mode in poll controller routine.

Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Interrupt code cleanup
Manish chopra [Thu, 23 Jan 2014 22:18:30 +0000 (17:18 -0500)] 
qlcnic: Interrupt code cleanup

o Added hardware ops for interrupt enable/disable functions

Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Enhance Tx timeout debugging.
Himanshu Madhani [Thu, 23 Jan 2014 22:18:29 +0000 (17:18 -0500)] 
qlcnic: Enhance Tx timeout debugging.

o Dump each Tx queue details with all descriptors, queue indices
  and Tx queue stats to imporve data colletion in situations
  where Tx timeout occurs.

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqlcnic: Use bool for rx_mac_learn.
Sucheta Chakraborty [Thu, 23 Jan 2014 22:18:28 +0000 (17:18 -0500)] 
qlcnic: Use bool for rx_mac_learn.

o Use boolean type instead of u8.

Signed-off-by: Sucheta Chakraborty <sucheta.chakraborty@qlogic.com>
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: fix u64 division
Nikolay Aleksandrov [Thu, 23 Jan 2014 22:43:43 +0000 (23:43 +0100)] 
bonding: fix u64 division

After the option conversion downdelay and updelay divide a u64
and on a 32 bit this causes the following errors:
ERROR: "__udivdi3" [drivers/net/bonding/bonding.ko] undefined!
ERROR: "__umoddi3" [drivers/net/bonding/bonding.ko] undefined!

Fix it by using a normal int instead because newval->value is capped
at INT_MAX by the way the option is defined.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agortnetlink: add missing IFLA_BOND_AD_INFO_UNSPEC
Jiri Pirko [Thu, 23 Jan 2014 15:51:27 +0000 (16:51 +0100)] 
rtnetlink: add missing IFLA_BOND_AD_INFO_UNSPEC

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Use the correct maximum TX DMA ring size for SFC9100
Ben Hutchings [Thu, 23 Jan 2014 14:35:48 +0000 (14:35 +0000)] 
sfc: Use the correct maximum TX DMA ring size for SFC9100

As part of a workaround for a hardware erratum in the SFC9100 family
(SF bug 35388), the TX_DESC_UPD_DWORD register address is also used
for communicating with the event block, and only descriptor pointer
values < 2048 are valid.

If the TX DMA ring size is increased to 4096 descriptors (which the
firmware still allows) then we may write a descriptor pointer
value >= 2048, which has entirely different and undesirable effects!

Limit the TX DMA ring size correctly when this workaround is in
effect.

Fixes: 8127d661e77f ('sfc: Add support for Solarflare SFC9100 family')
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAdd Shradha Shah as the sfc driver maintainer.
Shradha Shah [Thu, 23 Jan 2014 13:19:55 +0000 (13:19 +0000)] 
Add Shradha Shah as the sfc driver maintainer.

I will be taking over the work from Ben Hutchings.

Signed-off-by: Shradha Shah <sshah@solarflare.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/vxlan: Share RX skb de-marking and checksum checks with ovs
Or Gerlitz [Thu, 23 Jan 2014 09:28:13 +0000 (11:28 +0200)] 
net/vxlan: Share RX skb de-marking and checksum checks with ovs

Make sure the practice set by commit 0afb166 "vxlan: Add capability
of Rx checksum offload for inner packet" is applied when the skb
goes through the portion of the RX code which is shared between
vxlan netdevices and ovs vxlan port instances.

Cc: Joseph Gasparakis <joseph.gasparakis@intel.com>
Cc: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotulip: cleanup by using ARRAY_SIZE()
Dan Carpenter [Thu, 23 Jan 2014 08:26:37 +0000 (11:26 +0300)] 
tulip: cleanup by using ARRAY_SIZE()

In this situation then ARRAY_SIZE() and sizeof() are the same, but we're
really dealing with array indexes and not byte offsets so ARRAY_SIZE()
is cleaner.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Grant Grundler <grundler@parisc-linux.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoip_tunnel: clear IPCB in ip_tunnel_xmit() in case dst_link_failure() is called
Duan Jiong [Thu, 23 Jan 2014 06:00:25 +0000 (14:00 +0800)] 
ip_tunnel: clear IPCB in ip_tunnel_xmit() in case dst_link_failure() is called

commit a622260254ee48("ip_tunnel: fix kernel panic with icmp_dest_unreach")
clear IPCB in ip_tunnel_xmit()  , or else skb->cb[] may contain garbage from
GSO segmentation layer.

But commit 0e6fbc5b6c621("ip_tunnels: extend iptunnel_xmit()") refactor codes,
and it clear IPCB behind the dst_link_failure().

So clear IPCB in ip_tunnel_xmit() just like commti a622260254ee48("ip_tunnel:
fix kernel panic with icmp_dest_unreach").

Signed-off-by: Duan Jiong <duanj.fnst@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/cxgb4: Don't retrieve stats during recovery
Gavin Shan [Thu, 23 Jan 2014 04:27:35 +0000 (12:27 +0800)] 
net/cxgb4: Don't retrieve stats during recovery

We possibly retrieve the adapter's statistics during EEH recovery
and that should be disallowed. Otherwise, it would possibly incur
replicate EEH error and EEH recovery is going to fail eventually.

The patch reuses statistics lock and checks net_device is attached
before going to retrieve statistics, so that the problem can be
avoided.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/cxgb4: Avoid disabling PCI device for towice
Gavin Shan [Thu, 23 Jan 2014 04:27:34 +0000 (12:27 +0800)] 
net/cxgb4: Avoid disabling PCI device for towice

If we have EEH error happens to the adapter and we have to remove
it from the system for some reasons (e.g. more than 5 EEH errors
detected from the device in last hour), the adapter will be disabled
for towice separately by eeh_err_detected() and remove_one(), which
will incur following unexpected backtrace. The patch tries to avoid
it.

WARNING: at drivers/pci/pci.c:1431
CPU: 12 PID: 121 Comm: eehd Not tainted 3.13.0-rc7+ #1
task: c0000001823a3780 ti: c00000018240c000 task.ti: c00000018240c000
NIP: c0000000003c1e40 LR: c0000000003c1e3c CTR: 0000000001764c5c
REGS: c00000018240f470 TRAP: 0700   Not tainted  (3.13.0-rc7+)
MSR: 8000000000029032 <SF,EE,ME,IR,DR,RI>  CR: 28000024  XER: 00000004
CFAR: c000000000706528 SOFTE: 1
GPR00: c0000000003c1e3c c00000018240f6f0 c0000000010fe1f8 0000000000000035
GPR04: 0000000000000000 0000000000000000 00000000003ae509 0000000000000000
GPR08: 000000000000346f 0000000000000000 0000000000000000 0000000000003fef
GPR12: 0000000028000022 c00000000ec93000 c0000000000c11b0 c000000184ac3e40
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR24: 0000000000000000 c0000000009398d8 c00000000101f9c0 c0000001860ae000
GPR28: c000000182ba0000 00000000000001f0 c0000001860ae6f8 c0000001860ae000
NIP [c0000000003c1e40] .pci_disable_device+0xd0/0xf0
LR [c0000000003c1e3c] .pci_disable_device+0xcc/0xf0
Call Trace:
[c0000000003c1e3c] .pci_disable_device+0xcc/0xf0 (unreliable)
[d0000000073881c4] .remove_one+0x174/0x320 [cxgb4]
[c0000000003c57e0] .pci_device_remove+0x60/0x100
[c00000000046396c] .__device_release_driver+0x9c/0x120
[c000000000463a20] .device_release_driver+0x30/0x60
[c0000000003bcdb4] .pci_stop_bus_device+0x94/0xd0
[c0000000003bcf48] .pci_stop_and_remove_bus_device+0x18/0x30
[c00000000003f548] .pcibios_remove_pci_devices+0xa8/0x140
[c000000000035c00] .eeh_handle_normal_event+0xa0/0x3c0
[c000000000035f50] .eeh_handle_event+0x30/0x2b0
[c0000000000362c4] .eeh_event_handler+0xf4/0x1b0
[c0000000000c12b8] .kthread+0x108/0x130
[c00000000000a168] .ret_from_kernel_thread+0x5c/0x74

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodrivers: net: cpsw: enable promiscuous mode support
Mugunthan V N [Wed, 22 Jan 2014 18:33:12 +0000 (00:03 +0530)] 
drivers: net: cpsw: enable promiscuous mode support

Enable promiscuous mode support for CPSW.

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Correctly sync addresses from multiple sources to single device
Vlad Yasevich [Wed, 22 Jan 2014 17:54:15 +0000 (12:54 -0500)] 
net: Correctly sync addresses from multiple sources to single device

When we have multiple devices attempting to sync the same address
to a single destination, each device should be permitted to sync
it once.  To accomplish this, pass the 'sync_cnt' of the source
address when adding the addresss to the lower device.  'sync_cnt'
tracks how many time a given address has been succefully synced.
This way, we know that if the 'sync_cnt' passed in is 0, we should
sync this address.

Also, turn 'synced' member back into the counter as was originally
done in
   commit 4543fbefe6e06a9e40d9f2b28d688393a299f079.
   net: count hw_addr syncs so that unsync works properly.
It tracks how many time a given address has been added via a
'sync' operation.  For every successfull 'sync' the counter is
incremented, and for ever 'unsync', the counter is decremented.
This makes sure that the address will be properly removed from
the the lower device when all the upper devices have removed it.

Reported-by: Andrey Dmitrov <andrey.dmitrov@oktetlabs.ru>
CC: Andrey Dmitrov <andrey.dmitrov@oktetlabs.ru>
CC: Alexandra N. Kossovsky <Alexandra.Kossovsky@oktetlabs.ru>
CC: Konstantin Ushakov <Konstantin.Ushakov@oktetlabs.ru>
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/udp_offload: Handle static checker complaints
Shlomo Pongratz [Wed, 22 Jan 2014 13:23:29 +0000 (15:23 +0200)] 
net/udp_offload: Handle static checker complaints

Fixed few issues around using __rcu prefix and rcu_assign_pointer, also
fixed a warning print to use ntohs(port) and not htons(port).

net/ipv4/udp_offload.c:112:9: error: incompatible types in comparison expression (different address spaces)
net/ipv4/udp_offload.c:113:9: error: incompatible types in comparison expression (different address spaces)
net/ipv4/udp_offload.c:176:19: error: incompatible types in comparison expression (different address spaces)

Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotcp: metrics: Handle v6/v4-mapped sockets in tcp-metrics
Christoph Paasch [Wed, 22 Jan 2014 12:58:44 +0000 (13:58 +0100)] 
tcp: metrics: Handle v6/v4-mapped sockets in tcp-metrics

A socket may be v6/v4-mapped. In that case sk->sk_family is AF_INET6,
but the IP being used is actually an IPv4-address.
Current's tcp-metrics will thus represent it as an IPv6-address:

root@server:~# ip tcp_metrics
::ffff:10.1.1.2 age 22.920sec rtt 18750us rttvar 15000us cwnd 10
10.1.1.2 age 47.970sec rtt 16250us rttvar 10000us cwnd 10

This patch modifies the tcp-metrics so that they are able to handle the
v6/v4-mapped sockets correctly.

Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years ago6lowpan: add a license to 6lowpan_iphc module
Yann Droneaud [Wed, 22 Jan 2014 19:25:24 +0000 (20:25 +0100)] 
6lowpan: add a license to 6lowpan_iphc module

Since commit 8df8c56a5abc, 6lowpan_iphc is a module of its own.

Unfortunately, it lacks some infrastructure to behave like a
good kernel citizen:

  kernel: 6lowpan_iphc: module license 'unspecified' taints kernel.
  kernel: Disabling lock debugging due to kernel taint

This patch adds the basic MODULE_LICENSE(); with GPL license:
the code was copied from net/ieee802154/6lowpan.c which is GPL
and the module exports symbol with EXPORT_SYMBOL_GPL();.

Cc: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Cc: Alexander Aring <alex.aring@gmail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoteam: Don't allow team devices to change network namespaces.
Weilong Chen [Thu, 23 Jan 2014 03:07:22 +0000 (11:07 +0800)] 
team: Don't allow team devices to change network namespaces.

Like bonding, team as netdevice doesn't cross netns boundaries.

Team ports and team itself live in same netns.

Signed-off-by: Weilong Chen <chenweilong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: Don't allow bond devices to change network namespaces.
Weilong Chen [Wed, 22 Jan 2014 09:16:30 +0000 (17:16 +0800)] 
bonding: Don't allow bond devices to change network namespaces.

Like bridge, bonding as netdevice doesn't cross netns boundaries.

Bonding ports and bonding itself live in same netns.

Signed-off-by: Weilong Chen <chenweilong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agortnetlink: remove ndo_get_slave
Jiri Pirko [Wed, 22 Jan 2014 08:05:57 +0000 (09:05 +0100)] 
rtnetlink: remove ndo_get_slave

No longer used API bond-specific can be removed now. This is now handled
in a generic way in rtnl_link_ops.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert netlink to use slave data info api
Jiri Pirko [Wed, 22 Jan 2014 08:05:56 +0000 (09:05 +0100)] 
bonding: convert netlink to use slave data info api

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agortnetlink: provide api for getting and setting slave info
Jiri Pirko [Wed, 22 Jan 2014 08:05:55 +0000 (09:05 +0100)] 
rtnetlink: provide api for getting and setting slave info

Recent patch
bonding: add netlink attributes to slave link dev (1d3ee88ae0d6)

Introduced yet another device specific way to access slave information
over rtnetlink. There is one already there for bridge.

This patch introduces generic way to do this, for getting and setting
info as well by extending link_ops. Later on, this new interface will
be used for bridge ports as well.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agortnetlink: put "BOND" into nl attribute names which are related to bonding
Jiri Pirko [Wed, 22 Jan 2014 08:05:54 +0000 (09:05 +0100)] 
rtnetlink: put "BOND" into nl attribute names which are related to bonding

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: change name of sysfs dir for bonding slaves
Jiri Pirko [Wed, 22 Jan 2014 08:05:53 +0000 (09:05 +0100)] 
bonding: change name of sysfs dir for bonding slaves

Allow user to identify easily what the attributes are related to. Change
the name of the group to "bonding_slave" to be similar to master which
is named "bonding".

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/neighbour: queue work on power efficient wq
viresh kumar [Wed, 22 Jan 2014 06:53:33 +0000 (12:23 +0530)] 
net/neighbour: queue work on power efficient wq

Workqueue used in neighbour layer have no real dependency of scheduling these on
the cpu which scheduled them.

On a idle system, it is observed that an idle cpu wakes up many times just to
service this work. It would be better if we can schedule it on a cpu which the
scheduler believes to be the most appropriate one.

This patch replaces normal workqueues with power efficient versions. This
doesn't change existing behavior of code unless CONFIG_WQ_POWER_EFFICIENT is
enabled.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/ipv4: queue work on power efficient wq
viresh kumar [Wed, 22 Jan 2014 06:53:32 +0000 (12:23 +0530)] 
net/ipv4: queue work on power efficient wq

Workqueue used in ipv4 layer have no real dependency of scheduling these on the
cpu which scheduled them.

On a idle system, it is observed that an idle cpu wakes up many times just to
service this work. It would be better if we can schedule it on a cpu which the
scheduler believes to be the most appropriate one.

This patch replaces normal workqueues with power efficient versions. This
doesn't change existing behavior of code unless CONFIG_WQ_POWER_EFFICIENT is
enabled.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipv6: enable anycast addresses as source addresses for datagrams
FX Le Bail [Wed, 22 Jan 2014 06:42:37 +0000 (07:42 +0100)] 
ipv6: enable anycast addresses as source addresses for datagrams

This change allows to consider an anycast address valid as source address
when given via an IPV6_PKTINFO or IPV6_2292PKTINFO ancillary data item.
So, when sending a datagram with ancillary data, the unicast and anycast
addresses are handled in the same way.

- Adds ipv6_chk_acast_addr_src() to check if an anycast address is link-local
  on given interface or is global.
- Uses it in ip6_datagram_send_ctl().

Signed-off-by: Francois-Xavier Le Bail <fx.lebail@yahoo.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: update comments of "struct msghdr" with the more accurate RFC3542 ones
FX Le Bail [Wed, 22 Jan 2014 04:39:46 +0000 (05:39 +0100)] 
net: update comments of "struct msghdr" with the more accurate RFC3542 ones

Signed-off-by: Francois-Xavier Le Bail <fx.lebail@yahoo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotuntap: Fix for a race in accessing numqueues
Dominic Curran [Wed, 22 Jan 2014 03:03:23 +0000 (03:03 +0000)] 
tuntap: Fix for a race in accessing numqueues

A patch for fixing a race between queue selection and changing queues
was introduced in commit 92bb73ea2("tuntap: fix a possible race between
queue selection and changing queues").

The fix was to prevent the driver from re-reading the tun->numqueues
more than once within tun_select_queue() using ACCESS_ONCE().

We have been experiancing 'Divide-by-zero' errors in tun_net_xmit()
since we moved from 3.6 to 3.10, and believe that they come from a
simular source where the value of tun->numqueues changes to zero
between the first and a subsequent read of tun->numqueues.

The fix is a simular use of ACCESS_ONCE(), as well as a multiply
instead of a divide in the if statement.

Signed-off-by: Dominic Curran <dominic.curran@citrix.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Maxim Krasnyansky <maxk@qti.qualcomm.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Max Krasnyansky <maxk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobridge: Remove unnecessary vlan_put_tag in br_handle_vlan
Toshiaki Makita [Wed, 22 Jan 2014 00:29:37 +0000 (09:29 +0900)] 
bridge: Remove unnecessary vlan_put_tag in br_handle_vlan

br_handle_vlan() pushes HW accelerated vlan tag into skbuff when outgoing
port is the bridge device.
This is unnecessary because __netif_receive_skb_core() can handle skbs
with HW accelerated vlan tag. In current implementation,
__netif_receive_skb_core() needs to extract the vlan tag embedded in skb
data. This could cause low network performance especially when receiving
frames at a high frame rate on the bridge device.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Acked-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotcp: metrics: Fix rcu-race when deleting multiple entries
Christoph Paasch [Tue, 21 Jan 2014 12:30:26 +0000 (13:30 +0100)] 
tcp: metrics: Fix rcu-race when deleting multiple entries

In bbf852b96ebdc6d1 I introduced the tmlist, which allows to delete
multiple entries from the cache that match a specified destination if no
source-IP is specified.

However, as the cache is an RCU-list, we should not create this tmlist, as
it will change the tcpm_next pointer of the element that will be deleted
and so a thread iterating over the cache's entries while holding the
RCU-lock might get "redirected" to this tmlist.

This patch fixes this, by reverting back to the old behavior prior to
bbf852b96ebdc6d1, which means that we simply change the tcpm_next
pointer of the previous element (pp) to jump over the one we are
deleting.
The difference is that we call kfree_rcu() directly on the cache entry,
which allows us to delete multiple entries from the list.

Fixes: bbf852b96ebdc6d1 (tcp: metrics: Delete all entries matching a certain destination)
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'bonding'
David S. Miller [Thu, 23 Jan 2014 01:45:39 +0000 (17:45 -0800)] 
Merge branch 'bonding'

Ding Tianhong says:

====================
bonding: fix primary problem for bonding

If the slave's name changed, and the bond params primary is exist,
the bond should deal with the situation in two ways:

1) If the slave was the primary slave yet, clean the primary slave
   and reselect active slave.
2) If the slave's new name is as same as bond primary, set the slave
   as primary slave and reselect active slave.

If the new primary is not matching any slave in the bond, the bond should
record it to params, clean the primary slave and select a new active slave.

Update bonding.txt for primary description.

v2.1->v1: Because there are too many indentions and useless verification, so rewrite
  the logic for updating the primary slave.
  Modify some comments for to clean the typos.

v3->v2.1: Veaceslav disagree the first patch and modify the logic for it
  (bonding: update the primary slave when changing slave's name)
  and resend it himself (bonding: handle slave's name change with primary_slave logic),
  so remove the first patch and send the last two patches.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: update bonding.txt for primary description
dingtianhong [Sat, 18 Jan 2014 08:28:57 +0000 (16:28 +0800)] 
bonding: update bonding.txt for primary description

Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: clean the primary slave if there is no slave matching new primary
dingtianhong [Sat, 18 Jan 2014 08:28:52 +0000 (16:28 +0800)] 
bonding: clean the primary slave if there is no slave matching new primary

If the new primay is not matching any slave in the bond, the bond should
record it to params, clean the primary slave and select a new active slave.

Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: vxlan: convert to act as a pernet subsystem
Daniel Borkmann [Wed, 22 Jan 2014 20:07:53 +0000 (21:07 +0100)] 
net: vxlan: convert to act as a pernet subsystem

As per suggestion from Eric W. Biederman, vxlan should be using
{un,}register_pernet_subsys() instead of {un,}register_pernet_device()
to ensure the vxlan_net structure is initialized before and cleaned
up after all network devices in a given network namespace i.e. when
dealing with network notifiers. This is similarly handeled already in
commit 91e2ff3528ac ("net: Teach vlans to cleanup as a pernet subsystem")
and, thus, improves upon fd27e0d44a89 ("net: vxlan: do not use vxlan_net
before checking event type"). Just as in 91e2ff3528ac, we do not need
to explicitly handle deletion of vxlan devices as network namespace
exit calls dellink on all remaining virtual devices, and
rtnl_link_unregister() calls dellink on all outstanding devices in that
network namespace, so we can entirely drop the pernet exit operation
as well. Moreover, on vxlan module exit, rcu_barrier() is called by
netns since commit 3a765edadb28 ("netns: Add an explicit rcu_barrier
to unregister_pernet_{device|subsys}"), so this may be omitted. Tested
with various scenarios and works well on my side.

Suggested-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosch_htb: let skb->priority refer to non-leaf class
Harry Mason [Fri, 17 Jan 2014 13:22:32 +0000 (13:22 +0000)] 
sch_htb: let skb->priority refer to non-leaf class

If the class in skb->priority is not a leaf, apply filters from the
selected class, not the qdisc. This lets netfilter or user space
partially classify the packet.

Signed-off-by: Harry Mason <harry.mason@smoothwall.net>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoaf_packet: Add Queue mapping mode to af_packet fanout operation
Neil Horman [Wed, 22 Jan 2014 21:01:44 +0000 (16:01 -0500)] 
af_packet: Add Queue mapping mode to af_packet fanout operation

This patch adds a queue mapping mode to the fanout operation of af_packet
sockets.  This allows user space af_packet users to better filter on flows
ingressing and egressing via a specific hardware queue, and avoids the potential
packet reordering that can occur when FANOUT_CPU is being used and irq affinity
varies.

Tested successfully by myself.  applies to net-next

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'bonding_option_api'
David S. Miller [Wed, 22 Jan 2014 23:39:18 +0000 (15:39 -0800)] 
Merge branch 'bonding_option_api'

Nikolay Aleksandrov says:

====================
bonding: introduce new option API

This patchset's goal is to introduce a new option API which should be used
to properly describe the bonding options with their mode dependcies and
requirements. With this patchset applied we get centralized option
manipulation, automatic RTNL acquire per option setting, automatic option
range checking, mode dependcy checking and other various flags which are
described in detail in patch 01's commit message and comments.
Also the parameter passing is changed to use a specialized structure which
is initialized to a value depending on the needs.
The main exported functions are:
 __bond_opt_set() - set an option (RTNL should be acquired prior)
 bond_opt_init(val|str) - init a bond_opt_value struct for value or string
                          parameter passing
 bond_opt_tryset_rtnl() - function which tries to acquire rtnl, mainly used
                          for sysfs
 bond_opt_parse - used to parse or check for valid values
 bond_opt_get - retrieve a pointer to bond_option struct for some option
 bond_opt_get_val - retrieve a pointer to a bond_opt_value struct for
                    some value

The same functions are used to set an option via sysfs and netlink, just
the parameter that's passed is usually initialized in a different way.
The converted options have multiple style fixes, there're some longer
lines but they looked either ugly or were strings/pr_warnings, if you
think some line would be better broken just let me know :-) there're
also a few sscanf false-positive warnings.
I decided to keep the "unsuppmodes" way of mode dep checking since it's
straight forward, if we make a more general way for checking dependencies
it'll be easy to change it.

Future plans for this work include:
 - Automatic sysfs generation from the bond_opts[].
 - Use of the API in bond_check_params() and thus cleaning it up (this has
   actually started, I'll take care of the rest in a separate patch)
 - Clean up all option-unrelated files of option definitions and functions

I've tried to leave as much documentation as possible, if there's anything
unclear please let me know. One more thing, I haven't moved all
option-related functions from bonding.h to the new bond_options.h, this
will be done in a separate patch, it's in my todo list.

This patchset has been tested by setting each converted option via sysfs
and netlink to a couple of wrong values, a couple of correct values and
some random values, also for the opts that have flags they have been
tested as well.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert slaves to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:40 +0000 (14:53 +0100)] 
bonding: convert slaves to use the new option API

This patch adds the necessary changes so slaves would use
the new bonding option API. Also move the option to its own set function
in bond_options.c and fix some style errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert lp_interval to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:39 +0000 (14:53 +0100)] 
bonding: convert lp_interval to use the new option API

This patch adds the necessary changes so lp_interval would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert resend_igmp to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:38 +0000 (14:53 +0100)] 
bonding: convert resend_igmp to use the new option API

This patch adds the necessary changes so resend_igmp would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert all_slaves_active to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:37 +0000 (14:53 +0100)] 
bonding: convert all_slaves_active to use the new option API

This patch adds the necessary changes so all_slaves_active would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert queue_id to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:36 +0000 (14:53 +0100)] 
bonding: convert queue_id to use the new option API

This patch adds the necessary changes so queue_id would use
the new bonding option API. Also move it to its own set function in
bond_options.c and fix some style errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert active_slave to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:35 +0000 (14:53 +0100)] 
bonding: convert active_slave to use the new option API

This patch adds the necessary changes so active_slave would use
the new bonding option API. Also some trivial/style fixes.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert use_carrier to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:34 +0000 (14:53 +0100)] 
bonding: convert use_carrier to use the new option API

This patch adds the necessary changes so use_carrier would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert primary_reselect to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:33 +0000 (14:53 +0100)] 
bonding: convert primary_reselect to use the new option API

This patch adds the necessary changes so primary_reselect would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert primary to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:32 +0000 (14:53 +0100)] 
bonding: convert primary to use the new option API

This patch adds the necessary changes so primary would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert miimon to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:31 +0000 (14:53 +0100)] 
bonding: convert miimon to use the new option API

This patch adds the necessary changes so miimon would use
the new bonding option API. The "default" definition has been removed as
it was 0.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert num_peer_notif to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:30 +0000 (14:53 +0100)] 
bonding: convert num_peer_notif to use the new option API

This patch adds the necessary changes so num_peer_notif would use
the new bonding option API.
When the auto-sysfs generation is done an alias should be added for
this option as there're currently 2 entries in sysfs for it.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert ad_select to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:29 +0000 (14:53 +0100)] 
bonding: convert ad_select to use the new option API

This patch adds the necessary changes so ad_select would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert min_links to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:28 +0000 (14:53 +0100)] 
bonding: convert min_links to use the new option API

This patch adds the necessary changes so min_links would use
the new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert lacp_rate to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:27 +0000 (14:53 +0100)] 
bonding: convert lacp_rate to use the new option API

This patch adds the necessary changes so lacp_rate would use
the new bonding option API. Also some trivial/style error fixes.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert updelay to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:26 +0000 (14:53 +0100)] 
bonding: convert updelay to use the new option API

This patch adds the necessary changes so updelay would use
the new bonding option API. Also some trivial style fixes.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert downdelay to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:25 +0000 (14:53 +0100)] 
bonding: convert downdelay to use the new option API

This patch adds the necessary changes so downdelay would use
the new bonding option API. Also some trivial style fixes.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert arp_ip_target to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:24 +0000 (14:53 +0100)] 
bonding: convert arp_ip_target to use the new option API

This patch adds the necessary changes so arp_ip_target would use
the new bonding option API. This option is an exception because of
the way it's currently implemented that's why its netlink code is
a bit different from the other options to keep the functionality as
before and at the same time to have a single set function.

This patch also fixes a few stylistic errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert arp_interval to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:23 +0000 (14:53 +0100)] 
bonding: convert arp_interval to use the new option API

This patch adds the necessary changes so arp_interval would use
the new bonding option API. The "default" definition has been removed as
it was 0.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert fail_over_mac to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:22 +0000 (14:53 +0100)] 
bonding: convert fail_over_mac to use the new option API

This patch adds the necessary changes so fail_over_mac would use
the new bonding option API. Also fixes a trivial copy/paste error in
bond_check_params where the wrong variable was used for the error msg.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert arp_all_targets to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:21 +0000 (14:53 +0100)] 
bonding: convert arp_all_targets to use the new option API

This patch adds the necessary changes so arp_all_targets would use the
new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert arp_validate to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:20 +0000 (14:53 +0100)] 
bonding: convert arp_validate to use the new option API

This patch adds the necessary changes so arp_validate would use the
new bonding option API. Also fix some trivial/style errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert xmit_hash_policy to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:19 +0000 (14:53 +0100)] 
bonding: convert xmit_hash_policy to use the new option API

This patch adds the necessary changes so xmit_hash_policy would use the
new bonding option API. Also fix some trivial/style errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert packets_per_slave to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:18 +0000 (14:53 +0100)] 
bonding: convert packets_per_slave to use the new option API

This patch adds the necessary changes so packets_per_slave would use the
new bonding option API.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: convert mode setting to use the new option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:17 +0000 (14:53 +0100)] 
bonding: convert mode setting to use the new option API

This patch makes the bond's mode setting use the new option API and
adds support for dependency printing which relies on having an entry for
the mode option in the bond_opts[] array.
Also add the ability to print the mode name when mode dependency fails
and fix some trivial/style errors.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: add infrastructure for an option API
Nikolay Aleksandrov [Wed, 22 Jan 2014 13:53:16 +0000 (14:53 +0100)] 
bonding: add infrastructure for an option API

This patch adds the necessary basic infrastructure to support
centralized and unified option manipulation API for the bonding. The new
structure bond_option will be used to describe each option with its
dependencies on modes which will be checked automatically thus removing a
lot of duplicated code. Also automatic range checking is added for
some options. Currently the option setting function requires RTNL to
be acquired prior to calling it, since many options already rely on RTNL
it seemed like the best choice to protect all against common race
conditions.
In order to add an option the following steps need to be done:
1. Add an entry BOND_OPT_<option> to bond_options.h so it gets a unique id
   and a bit corresponding to the id
2. Add a bond_option entry to the bond_opts[] array in bond_options.c which
   describes the option, its dependencies and its manipulation function
3. Add code to export the option through sysfs and/or as a module parameter
   (the sysfs export will be made automatically in the future)

The options can have different flags set, currently the following are
supported:
BOND_OPTFLAG_NOSLAVES - require that the bond device has no slaves prior
                        to setting the option
BOND_OPTFLAG_IFDOWN - require that the bond device is down prior to
                      setting the option
BOND_OPTFLAG_RAWVAL - don't parse the value but return it raw for the
                      option to parse

There's a new value structure to describe different types of values
which can have the following flags:
BOND_VALFLAG_DEFAULT - marks the default option (permanent string alias
                       to this option is "default")
BOND_VALFLAG_MIN - the minimum value that this option can have
BOND_VALFLAG_MAX - the maximum value that this option can have

An example would be nice here, so if we have an option which can have
the values "off"(2), "special"(4, default) and supports a range, say
16 - 32, it should be defined as follows:
"off", 2,
"special", 4, BOND_VALFLAG_DEFAULT,
"rangemin", 16, BOND_VALFLAG_MIN,
"rangemax", 32, BOND_VALFLAG_MAX
So we have the valid intervals: [2, 2], [4, 4], [16, 32]
Also the valid strings: "off" = 2, "special" and "default" = 4
                        "rangemin" = 16, "rangemax" = 32

BOND_VALFLAG_(MIN|MAX) can be used to specify a valid range for an
option, if MIN is omitted then 0 is considered as a minimum. If an
exact match is found in the values[] table it will be returned,
otherwise the range is tried (if available).

The option parameter passing is done by using a special structure called
bond_opt_value which can take either a string or a value to parse. One
of the bond_opt_init(val|str) macros should be used depending on which
one does the user want to parse (string or value). Then a call to
__bond_opt_set should be done under RTNL.

Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'reciprocal'
David S. Miller [Wed, 22 Jan 2014 07:19:26 +0000 (23:19 -0800)] 
Merge branch 'reciprocal'

Hannes Frederic Sowa says:

====================
reciprocal_divide update

This patch is on top of aee636c4809fa5 ("bpf: do not use reciprocal
divide") from Eric that sits in net tree. It will not create a merge
conflict, but it depends on this one, so we suggest, if possible, to
merge net into net-next.

We are proposing this change with only small modifications from the
v2 version, namely updating the name of trim to reciprocal_scale
(as commented on by Ben Hutchings and Eric Dumazet, thanks!).

We thought about introducing the reciprocal_divide algorithm in
parallel to the one already used by the kernel but faced organizational
issues, leading us to the conclusion that it is best to just replace
the old one: We could not come up with names for the different
implementations and also with a way to describe the differences to
guide developers which one to choose in which situation. This is
because we cannot specify the correct semantics for the version
which is currently used by the kernel. Altough it seems to not be
causing problems in the kernel, we cannot surely say so in the
case of flex_array for the future. Current usage seems ok, but
future users could run into problems.

Changelog:

v1->v2:
 - changed name to prandom_u32_max in p1
 - changed name to trim in p2
 - reworked code in p3
v2->v3:
 - p1 and p3 stays unchanged, only small update in commit
   message in p3
 - changed name to reciprocal_scale in p2
 - fixed kernel doc format
v3->v4:
 - pseduo -> pseudo (thanks to Tilman Schmidt)
v4->v5:
 - fix pseduo -> pseudo for real now, sorry for the noise
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoreciprocal_divide: update/correction of the algorithm
Hannes Frederic Sowa [Wed, 22 Jan 2014 01:29:41 +0000 (02:29 +0100)] 
reciprocal_divide: update/correction of the algorithm

Jakub Zawadzki noticed that some divisions by reciprocal_divide()
were not correct [1][2], which he could also show with BPF code
after divisions are transformed into reciprocal_value() for runtime
invariance which can be passed to reciprocal_divide() later on;
reverse in BPF dump ended up with a different, off-by-one K in
some situations.

This has been fixed by Eric Dumazet in commit aee636c4809fa5
("bpf: do not use reciprocal divide"). This follow-up patch
improves reciprocal_value() and reciprocal_divide() to work in
all cases by using Granlund and Montgomery method, so that also
future use is safe and without any non-obvious side-effects.
Known problems with the old implementation were that division by 1
always returned 0 and some off-by-ones when the dividend and divisor
where very large. This seemed to not be problematic with its
current users, as far as we can tell. Eric Dumazet checked for
the slab usage, we cannot surely say so in the case of flex_array.
Still, in order to fix that, we propose an extension from the
original implementation from commit 6a2d7a955d8d resp. [3][4],
by using the algorithm proposed in "Division by Invariant Integers
Using Multiplication" [5], Torbjörn Granlund and Peter L.
Montgomery, that is, pseudocode for q = n/d where q, n, d is in
u32 universe:

1) Initialization:

  int l = ceil(log_2 d)
  uword m' = floor((1<<32)*((1<<l)-d)/d)+1
  int sh_1 = min(l,1)
  int sh_2 = max(l-1,0)

2) For q = n/d, all uword:

  uword t = (n*m')>>32
  q = (t+((n-t)>>sh_1))>>sh_2

The assembler implementation from Agner Fog [6] also helped a lot
while implementing. We have tested the implementation on x86_64,
ppc64, i686, s390x; on x86_64/haswell we're still half the latency
compared to normal divide.

Joint work with Daniel Borkmann.

  [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c
  [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c
  [3] https://gmplib.org/~tege/division-paper.pdf
  [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html
  [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556
  [6] http://www.agner.org/optimize/asmlib.zip

Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Austin S Hemmelgarn <ahferroin7@gmail.com>
Cc: linux-kernel@vger.kernel.org
Cc: Jesse Gross <jesse@nicira.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Andy Gospodarek <andy@greyhouse.net>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: introduce reciprocal_scale helper and convert users
Daniel Borkmann [Wed, 22 Jan 2014 01:29:40 +0000 (02:29 +0100)] 
net: introduce reciprocal_scale helper and convert users

As David Laight suggests, we shouldn't necessarily call this
reciprocal_divide() when users didn't requested a reciprocal_value();
lets keep the basic idea and call it reciprocal_scale(). More
background information on this topic can be found in [1].

Joint work with Hannes Frederic Sowa.

  [1] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html

Suggested-by: David Laight <david.laight@aculab.com>
Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agorandom32: add prandom_u32_max and convert open coded users
Daniel Borkmann [Wed, 22 Jan 2014 01:29:39 +0000 (02:29 +0100)] 
random32: add prandom_u32_max and convert open coded users

Many functions have open coded a function that returns a random
number in range [0,N-1]. Under the assumption that we have a PRNG
such as taus113 with being well distributed in [0, ~0U] space,
we can implement such a function as uword t = (n*m')>>32, where
m' is a random number obtained from PRNG, n the right open interval
border and t our resulting random number, with n,m',t in u32 universe.

Lets go with Joe and simply call it prandom_u32_max(), although
technically we have an right open interval endpoint, but that we
have documented. Other users can further be migrated to the new
prandom_u32_max() function later on; for now, we need to make sure
to migrate reciprocal_divide() users for the reciprocal_divide()
follow-up fixup since their function signatures are going to change.

Joint work with Hannes Frederic Sowa.

Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/mlx4_core: Remove unnecessary validation for port number
Moni Shoua [Tue, 21 Jan 2014 08:19:37 +0000 (10:19 +0200)] 
net/mlx4_core: Remove unnecessary validation for port number

This is a fix to a regression introduced by commit:
"982290a net/mlx4_core: Check port number for validity
before accessing data"

IPoIB could not attach to multicast group and we get this in dmesg:
[144214.145008] ib0: failed to attach to multicast group, ret = -22
[144214.145016] ib0: couldn't attach QP to multicast group ff12:401b:ffff:0000:0000:0000:ffff:ffff
[144214.145019] ib0: multicast join failed for ff12:401b:ffff:0000:0000:0000:ffff:ffff, status -22

The cause to the problem is because port is extracted from gid[5].
Which is only valid for Ethernet.
Removed this validation in mlx4_qp_attach_common(), which is accessed
from both Ethernet and IB flows.
Error flow for bad port value in Ethernet is already exists in that
function.

Signed-off-by: Moni Shoua <monis@mellanox.co.il>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobe2net: Fix be_vlan_add/rem_vid() routines
Somnath Kotur [Tue, 21 Jan 2014 10:20:55 +0000 (15:50 +0530)] 
be2net: Fix be_vlan_add/rem_vid() routines

The current logic to put interface into VLAN Promiscous mode is not correct.
We should increment "adapter->vlans_added" before calling be_vid_config().
Also removed some unwanted log messages.

Signed-off-by: Kalesh AP <kalesh.purayil@emulex.com>
Signed-off-by: Somnath Kotur <somnath.kotur@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Fix VF flr flow
Ariel Elior [Tue, 21 Jan 2014 08:31:20 +0000 (10:31 +0200)] 
bnx2x: Fix VF flr flow

When a VF originating from a given PF is flr-ed, that PF gets an interrupt
from the chip management and takes a part in the flr process.

This patch fixes several corner cases in which the driver performs its part
of the flr flow out-of-order, causing the FW to assert due to badly timed
messages received from the driver.

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Missing change from the ether_addr_copy() fixups.
David S. Miller [Wed, 22 Jan 2014 06:54:01 +0000 (22:54 -0800)] 
net: Missing change from the ether_addr_copy() fixups.

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: filter: let bpf_tell_extensions return SKF_AD_MAX
Daniel Borkmann [Mon, 20 Jan 2014 23:19:37 +0000 (00:19 +0100)] 
net: filter: let bpf_tell_extensions return SKF_AD_MAX

Michal Sekletar added in commit ea02f9411d9f ("net: introduce
SO_BPF_EXTENSIONS") a facility where user space can enquire
the BPF ancillary instruction set, which is imho a step into
the right direction for letting user space high-level to BPF
optimizers make an informed decision for possibly using these
extensions.

The original rationale was to return through a getsockopt(2)
a bitfield of which instructions are supported and which
are not, as of right now, we just return 0 to indicate a
base support for SKF_AD_PROTOCOL up to SKF_AD_PAY_OFFSET.
Limitations of this approach are that this API which we need
to maintain for a long time can only support a maximum of 32
extensions, and needs to be additionally maintained/updated
when each new extension that comes in.

I thought about this a bit more and what we can do here to
overcome this is to just return SKF_AD_MAX. Since we never
remove any extension since we cannot break user space and
always linearly increase SKF_AD_MAX on each newly added
extension, user space can make a decision on what extensions
are supported in the whole set of extensions and which aren't,
by just checking which of them from the whole set have an
offset < SKF_AD_MAX of the underlying kernel.

Since SKF_AD_MAX must be updated each time we add new ones,
we don't need to introduce an additional enum and got
maintenance for free. At some point in time when
SO_BPF_EXTENSIONS becomes ubiquitous for most kernels, then
an application can simply make use of this and easily be run
on newer or older underlying kernels without needing to be
recompiled, of course. Since that is for 3.14, it's not too
late to do this change.

Cc: Michal Sekletar <msekleta@redhat.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Michal Sekletar <msekleta@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Fix some fallout from the etner_addr_copy() changes.
David S. Miller [Wed, 22 Jan 2014 02:55:22 +0000 (18:55 -0800)] 
net: Fix some fallout from the etner_addr_copy() changes.

net/appletalk/aarp.c: In function â€˜__aarp_send_query’:
net/appletalk/aarp.c:137:2: error: implicit declaration of function â€˜ether_addr_copy’ [-Werror=implicit-function-declaration]
 ...
net/atm/lec.c: In function â€˜send_to_lecd’:
net/atm/lec.c:524:3: warning: passing argument 1 of â€˜ether_addr_copy’ from incompatible pointer type [enabled by default]
In file included from net/atm/lec.c:17:0:
include/linux/etherdevice.h:227:20: note: expected â€˜u8 *’ but argument is of type â€˜unsigned char (*)[6]’
 ...
net/caif/caif_usb.c: In function â€˜cfusbl_create’:
net/caif/caif_usb.c:108:2: error: implicit declaration of function â€˜ether_addr_copy’ [-Werror=implicit-function-declaration]

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'sctp'
David S. Miller [Wed, 22 Jan 2014 02:41:46 +0000 (18:41 -0800)] 
Merge branch 'sctp'

Wang Weidong says:

====================
sctp: remove some macro locking wrappers

In sctp.h we can find some macro locking wrappers. As Neil point out that:

"Its because in the origional implementation of the sctp protocol, there was a
user space test harness which built the kernel module for userspace execution to
cary our some unit testing on the code.  It did so by redefining some of those
locking macros to user space friendly code.  IIRC we haven't use those unit
tests in years, and so should be removing them, not adding them to other
locations."

So I remove them.
====================

Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Vlad Yasevich <vyasevich@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_bh_[un]lock_sock
wangweidong [Tue, 21 Jan 2014 07:44:12 +0000 (15:44 +0800)] 
sctp: remove macros sctp_bh_[un]lock_sock

Redefined bh_[un]lock_sock to sctp_bh[un]lock_sock for user
space friendly code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_{lock|release}_sock
wangweidong [Tue, 21 Jan 2014 07:44:11 +0000 (15:44 +0800)] 
sctp: remove macros sctp_{lock|release}_sock

Redefined {lock|release}_sock to sctp_{lock|release}_sock for user space friendly
code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_read_[un]lock
wangweidong [Tue, 21 Jan 2014 07:44:10 +0000 (15:44 +0800)] 
sctp: remove macros sctp_read_[un]lock

Redefined read_[un]lock to sctp_read_[un]lock for user space
friendly code which we haven't use in years, and the macros
we never used, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_write_[un]_lock
wangweidong [Tue, 21 Jan 2014 07:44:09 +0000 (15:44 +0800)] 
sctp: remove macros sctp_write_[un]_lock

Redefined write_[un]lock to sctp_write_[un]lock for user space
friendly code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_spin_[un]lock
wangweidong [Tue, 21 Jan 2014 07:44:08 +0000 (15:44 +0800)] 
sctp: remove macros sctp_spin_[un]lock

Redefined spin_[un]lock to sctp_spin_[un]lock for user space friendly
code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_local_bh_{disable|enable}
wangweidong [Tue, 21 Jan 2014 07:44:07 +0000 (15:44 +0800)] 
sctp: remove macros sctp_local_bh_{disable|enable}

Redefined local_bh_{disable|enable} to sctp_local_bh_{disable|enable}
for user space friendly code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosctp: remove macros sctp_spin_[un]lock_irqrestore
wangweidong [Tue, 21 Jan 2014 07:44:06 +0000 (15:44 +0800)] 
sctp: remove macros sctp_spin_[un]lock_irqrestore

Redefined spin_[un]lock_irqstore to sctp_spin_[un]lock_irqrestore for user
space friendly code which we haven't use in years, so removing them.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodsa: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:20 +0000 (09:52 -0800)] 
dsa: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agopktgen: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:19 +0000 (09:52 -0800)] 
pktgen: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:18 +0000 (09:52 -0800)] 
netpoll: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agocaif_usb: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:17 +0000 (09:52 -0800)] 
caif_usb: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoatm: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:16 +0000 (09:52 -0800)] 
atm: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoappletalk: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:15 +0000 (09:52 -0800)] 
appletalk: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Convert struct aarp_entry.hwaddr[6] to hwaddr[ETH_ALEN].

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years ago8021q: Use ether_addr_copy
Joe Perches [Mon, 20 Jan 2014 17:52:14 +0000 (09:52 -0800)] 
8021q: Use ether_addr_copy

Use ether_addr_copy instead of memcpy(a, b, ETH_ALEN) to
save some cycles on arm and powerpc.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'gro_udp_encap'
David S. Miller [Wed, 22 Jan 2014 02:05:12 +0000 (18:05 -0800)] 
Merge branch 'gro_udp_encap'

Or Gerlitz says:

====================
net: Add GRO support for UDP encapsulating protocols

This series adds GRO handlers for protocols that do UDP encapsulation, with the
intent of being able to coalesce packets which encapsulate packets belonging to
the same TCP session.

For GRO purposes, the destination UDP port takes the role of the ether type
field in the ethernet header or the next protocol in the IP header.

The UDP GRO handler will only attempt to coalesce packets whose destination
port is registered to have gro handler.

The patches done against net-next 75e4364f67 "net: stmmac: fix NULL pointer
dereference in stmmac_get_tx_hwtstamp"

Or.

v4 --> v5 changes:
  - followed Eric's directives to avoid using atomic get/put ops on the
    udp gro receive and complete callbacks and instead keep the rcu_read_lock
    when calling the next handler on the chain.

v3 --> v4 changes:

  - applied feedback from Tom on some micro-optimizations that save
    branches and goto directives in the udp gro logic

 - applied feedback from Eric on correct RCU programming for the
   add/remove flow of the upper protocols udp gro handlers

v2 --> v3 changes:

 - moved to use linked list to store the udp gro handlers, this solves the
   problem of consuming 512KB of memory for the handlers.

 - use a mark on the skb GRO CB data to disallow running the udp gro_receive twice
   on a packet, this solves the problem of udp encapsulated packets whose inner VM
   packet is udp and happen to carry a port which has registered offloads - and flush it.

 - invoke the udp offload protocol registration and de-registration from the vxlan driver
   in a sleepable context
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Add GRO support for vxlan traffic
Or Gerlitz [Mon, 20 Jan 2014 11:59:21 +0000 (13:59 +0200)] 
net: Add GRO support for vxlan traffic

Add GRO handlers for vxlann, by using the UDP GRO infrastructure.

For single TCP session that goes through vxlan tunneling I got nice
improvement from 6.8Gbs to 11.5Gbs

--> UDP/VXLAN GRO disabled
$ netperf  -H 192.168.52.147 -c -C

$ netperf -t TCP_STREAM -H 192.168.52.147 -c -C
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.52.147 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

 87380  65536  65536    10.00      6799.75   12.54    24.79    0.604   1.195

--> UDP/VXLAN GRO enabled

$ netperf -t TCP_STREAM -H 192.168.52.147 -c -C
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.52.147 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

 87380  65536  65536    10.00      11562.72   24.90    20.34    0.706   0.577

Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Export gro_find_by_type helpers
Or Gerlitz [Mon, 20 Jan 2014 11:59:20 +0000 (13:59 +0200)] 
net: Export gro_find_by_type helpers

Export the gro_find_receive/complete_by_type helpers to they can be invoked
by the gro callbacks of encapsulation protocols such as vxlan.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Add GRO support for UDP encapsulating protocols
Or Gerlitz [Mon, 20 Jan 2014 11:59:19 +0000 (13:59 +0200)] 
net: Add GRO support for UDP encapsulating protocols

Add GRO handlers for protocols that do UDP encapsulation, with the intent of
being able to coalesce packets which encapsulate packets belonging to
the same TCP session.

For GRO purposes, the destination UDP port takes the role of the ether type
field in the ethernet header or the next protocol in the IP header.

The UDP GRO handler will only attempt to coalesce packets whose destination
port is registered to have gro handler.

Use a mark on the skb GRO CB data to disallow (flush) running the udp gro receive
code twice on a packet. This solves the problem of udp encapsulated packets whose
inner VM packet is udp and happen to carry a port which has registered offloads.

Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agostmmac: Fix kernel crashes for jumbo frames
Vince Bridgers [Mon, 20 Jan 2014 11:39:01 +0000 (05:39 -0600)] 
stmmac: Fix kernel crashes for jumbo frames

These changes correct the following issues with jumbo frames on the
stmmac driver:

1) The Synopsys EMAC can be configured to support different FIFO
sizes at core configuration time. There's no way to query the
controller and know the FIFO size, so the driver needs to get this
information from the device tree in order to know how to correctly
handle MTU changes and setting up dma buffers. The default
max-frame-size is as currently used, which is the size of a jumbo
frame.

2) The driver was enabling Jumbo frames by default, but was not allocating
dma buffers of sufficient size to handle the maximum possible packet
size that could be received. This led to memory corruption since DMAs were
occurring beyond the extent of the allocated receive buffers for certain types
of network traffic.

kernel BUG at net/core/skbuff.c:126!
Internal error: Oops - BUG: 0 [#1] SMP ARM
Modules linked in:
CPU: 0 PID: 563 Comm: sockperf Not tainted 3.13.0-rc6-01523-gf7111b9 #31
task: ef35e580 ti: ef252000 task.ti: ef252000
PC is at skb_panic+0x60/0x64
LR is at skb_panic+0x60/0x64
pc : [<c03c7c3c>]    lr : [<c03c7c3c>]    psr: 60000113
sp : ef253c18  ip : 60000113  fp : 00000000
r10: ef3a5400  r9 : 00000ebc  r8 : ef3a546c
r7 : ee59f000  r6 : ee59f084  r5 : ee59ff40  r4 : ee59f140
r3 : 000003e2  r2 : 00000007  r1 : c0b9c420  r0 : 0000007d
Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment user
Control: 10c5387d  Table: 2e8ac04a  DAC: 00000015
Process sockperf (pid: 563, stack limit = 0xef252248)
Stack: (0xef253c18 to 0xef254000)
3c00:                                                       00000ebc ee59f000
3c20: ee59f084 ee59ff40 ee59f140 c04a9cd8 ee8c50c0 00000ebc ee59ff40 00000000
3c40: ee59f140 c02d0ef0 00000056 ef1eda80 ee8c50c0 00000ebc 22bbef29 c0318f8c
3c60: 00000056 ef3a547c ffe2c716 c02c9c90 c0ba1298 ef3a5838 ef3a5838 ef3a5400
3c80: 000020c0 ee573840 000055cb ef3f2050 c053f0e0 c0319214 22b9b085 22d92813
3ca0: 00001c80 004b8e00 ef3a5400 ee573840 ef3f2064 22d92813 ef3f2064 000055cb
3cc0: ef3f2050 c031a19c ef252000 00000000 00000000 c0561bc0 00000000 ff00ffff
3ce0: c05621c0 ef3a5400 ef3f2064 ee573840 00000020 ef3f2064 000055cb ef3f2050
3d00: c053f0e0 c031cad0 c053e740 00000e60 00000000 00000000 ee573840 ef3a5400
3d20: ef0a6e00 00000000 ef3f2064 c032507c 00010000 00000020 c0561bc0 c0561bc0
3d40: ee599850 c032799c 00000000 ee573840 c055a380 ef3a5400 00000000 ef3f2064
3d60: ef3f2050 c032799c 0101c7c0 2b6755cb c059a280 c030e4d8 000055cb ffffffff
3d80: ee574fc0 c055a380 ee574000 ee573840 00002b67 ee573840 c03fe9c4 c053fa68
3da0: c055a380 00001f6f 00000000 ee573840 c053f0e0 c0304fdc ef0a6e01 ef3f2050
3dc0: ee573858 ef031000 ee573840 c03055d8 c0ba0c40 ef000f40 00100100 c053f0dc
3de0: c053ffdc c053f0f0 00000008 00000000 ef031000 c02da948 00001140 00000000
3e00: c0563c78 ef253e5f 00000020 ee573840 00000020 c053f0f0 ef313400 ee573840
3e20: c053f0e0 00000000 00000000 c05380c0 ef313400 00001000 00000015 c02df280
3e40: ee574000 ef001e00 00000000 00001080 00000042 005cd980 ef031500 ef031500
3e60: 00000000 c02df824 ef031500 c053e390 c0541084 f00b1e00 c05925e8 c02df864
3e80: 00001f5c ef031440 c053e390 c0278524 00000002 00000000 c0b9eb48 c02df280
3ea0: ee8c7180 00000100 c0542ca8 00000015 00000040 ef031500 ef031500 ef031500
3ec0: c027803c ef252000 00000040 000000ec c05380c0 c0b9eb40 c0b9eb48 c02df940
3ee0: ef060780 ffffa4dd c0564a9c c056343c 002e80a8 00000080 ef031500 00000001
3f00: c053808c ef252000 fffec100 00000003 00000004 002e80a8 0000000c c00258f0
3f20: 002e80a8 c005e704 00000005 00000100 c05634d0 c0538080 c05333e0 00000000
3f40: 0000000a c0565580 c05380c0 ffffa4dc c05434f4 00400100 00000004 c0534cd4
3f60: 00000098 00000000 fffec100 002e80a8 00000004 002e80a8 002a20e0 c0025da8
3f80: c0534cd4 c000f020 fffec10c c053ea60 ef253fb0 c0008530 0000ffe2 b6ef67f4
3fa0: 40000010 ffffffff 00000124 c0012f3c 0000ffe2 002e80f0 0000ffe2 00004000
3fc0: becb6338 becb6334 00000004 00000124 002e80a8 00000004 002e80a8 002a20e0
3fe0: becb6300 becb62f4 002773bb b6ef67f4 40000010 ffffffff 00000000 00000000
[<c03c7c3c>] (skb_panic+0x60/0x64) from [<c02d0ef0>] (skb_put+0x4c/0x50)
[<c02d0ef0>] (skb_put+0x4c/0x50) from [<c0318f8c>] (tcp_collapse+0x314/0x3ec)
[<c0318f8c>] (tcp_collapse+0x314/0x3ec) from [<c0319214>]
(tcp_try_rmem_schedule+0x1b0/0x3c4)
[<c0319214>] (tcp_try_rmem_schedule+0x1b0/0x3c4) from [<c031a19c>]
(tcp_data_queue+0x480/0xe6c)
[<c031a19c>] (tcp_data_queue+0x480/0xe6c) from [<c031cad0>]
(tcp_rcv_established+0x180/0x62c)
[<c031cad0>] (tcp_rcv_established+0x180/0x62c) from [<c032507c>]
(tcp_v4_do_rcv+0x13c/0x31c)
[<c032507c>] (tcp_v4_do_rcv+0x13c/0x31c) from [<c032799c>]
(tcp_v4_rcv+0x718/0x73c)
[<c032799c>] (tcp_v4_rcv+0x718/0x73c) from [<c0304fdc>]
(ip_local_deliver+0x98/0x274)
[<c0304fdc>] (ip_local_deliver+0x98/0x274) from [<c03055d8>]
(ip_rcv+0x420/0x758)
[<c03055d8>] (ip_rcv+0x420/0x758) from [<c02da948>]
(__netif_receive_skb_core+0x44c/0x5bc)
[<c02da948>] (__netif_receive_skb_core+0x44c/0x5bc) from [<c02df280>]
(netif_receive_skb+0x48/0xb4)
[<c02df280>] (netif_receive_skb+0x48/0xb4) from [<c02df824>]
(napi_gro_flush+0x70/0x94)
[<c02df824>] (napi_gro_flush+0x70/0x94) from [<c02df864>]
(napi_complete+0x1c/0x34)
[<c02df864>] (napi_complete+0x1c/0x34) from [<c0278524>]
(stmmac_poll+0x4e8/0x5c8)
[<c0278524>] (stmmac_poll+0x4e8/0x5c8) from [<c02df940>]
(net_rx_action+0xc4/0x1e4)
[<c02df940>] (net_rx_action+0xc4/0x1e4) from [<c00258f0>]
(__do_softirq+0x12c/0x2e8)
[<c00258f0>] (__do_softirq+0x12c/0x2e8) from [<c0025da8>] (irq_exit+0x78/0xac)
[<c0025da8>] (irq_exit+0x78/0xac) from [<c000f020>] (handle_IRQ+0x44/0x90)
[<c000f020>] (handle_IRQ+0x44/0x90) from [<c0008530>]
(gic_handle_irq+0x2c/0x5c)
[<c0008530>] (gic_handle_irq+0x2c/0x5c) from [<c0012f3c>]
(__irq_usr+0x3c/0x60)

3) The driver was setting the dma buffer size after allocating dma buffers,
which caused a system panic when changing the MTU.

BUG: Bad page state in process ifconfig  pfn:2e850
page:c0b72a00 count:0 mapcount:0 mapping:  (null) index:0x0
page flags: 0x200(arch_1)
Modules linked in:
CPU: 0 PID: 566 Comm: ifconfig Not tainted 3.13.0-rc6-01523-gf7111b9 #29
[<c001547c>] (unwind_backtrace+0x0/0xf8) from [<c00122dc>]
(show_stack+0x10/0x14)
[<c00122dc>] (show_stack+0x10/0x14) from [<c03c793c>] (dump_stack+0x70/0x88)
[<c03c793c>] (dump_stack+0x70/0x88) from [<c00b2620>] (bad_page+0xc8/0x118)
[<c00b2620>] (bad_page+0xc8/0x118) from [<c00b302c>]
(get_page_from_freelist+0x744/0x870)
[<c00b302c>] (get_page_from_freelist+0x744/0x870) from [<c00b40f4>]
(__alloc_pages_nodemask+0x118/0x86c)
[<c00b40f4>] (__alloc_pages_nodemask+0x118/0x86c) from [<c00b4858>]
(__get_free_pages+0x10/0x54)
[<c00b4858>] (__get_free_pages+0x10/0x54) from [<c00cba1c>]
(kmalloc_order_trace+0x24/0xa0)
[<c00cba1c>] (kmalloc_order_trace+0x24/0xa0) from [<c02d199c>]
(__kmalloc_reserve.isra.21+0x24/0x70)
[<c02d199c>] (__kmalloc_reserve.isra.21+0x24/0x70) from [<c02d240c>]
(__alloc_skb+0x68/0x13c)
[<c02d240c>] (__alloc_skb+0x68/0x13c) from [<c02d3930>]
(__netdev_alloc_skb+0x3c/0xe8)
[<c02d3930>] (__netdev_alloc_skb+0x3c/0xe8) from [<c0279378>]
(stmmac_open+0x63c/0x1024)
[<c0279378>] (stmmac_open+0x63c/0x1024) from [<c02e18cc>]
(__dev_open+0xa0/0xfc)
[<c02e18cc>] (__dev_open+0xa0/0xfc) from [<c02e1b40>]
(__dev_change_flags+0x94/0x158)
[<c02e1b40>] (__dev_change_flags+0x94/0x158) from [<c02e1c24>]
(dev_change_flags+0x18/0x48)
[<c02e1c24>] (dev_change_flags+0x18/0x48) from [<c0337bc0>]
(devinet_ioctl+0x638/0x700)
[<c0337bc0>] (devinet_ioctl+0x638/0x700) from [<c02c7aec>]
(sock_ioctl+0x64/0x290)
[<c02c7aec>] (sock_ioctl+0x64/0x290) from [<c0100890>]
(do_vfs_ioctl+0x78/0x5b8)
[<c0100890>] (do_vfs_ioctl+0x78/0x5b8) from [<c0100e0c>] (SyS_ioctl+0x3c/0x5c)
[<c0100e0c>] (SyS_ioctl+0x3c/0x5c) from [<c000e760>]

The fixes have been verified using reproducible, automated testing.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This page took 0.056288 seconds and 5 git commands to generate.